diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,120027 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.4,
+  "eval_steps": 2000,
+  "global_step": 20000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 4e-05,
+      "grad_norm": 456.0,
+      "learning_rate": 1.18e-05,
+      "loss": 85.4554,
+      "loss/crossentropy": 9.650346755981445,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 8.066818237304688,
+      "step": 2
+    },
+    {
+      "epoch": 8e-05,
+      "grad_norm": 416.0,
+      "learning_rate": 1.3600000000000002e-05,
+      "loss": 84.3418,
+      "loss/crossentropy": 9.544375896453857,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 7.628942489624023,
+      "step": 4
+    },
+    {
+      "epoch": 0.00012,
+      "grad_norm": 466.0,
+      "learning_rate": 1.54e-05,
+      "loss": 87.2187,
+      "loss/crossentropy": 9.569977283477783,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 7.7909746170043945,
+      "step": 6
+    },
+    {
+      "epoch": 0.00016,
+      "grad_norm": 247.0,
+      "learning_rate": 1.72e-05,
+      "loss": 82.5078,
+      "loss/crossentropy": 9.06786823272705,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 7.3673131465911865,
+      "step": 8
+    },
+    {
+      "epoch": 0.0002,
+      "grad_norm": 179.0,
+      "learning_rate": 1.9e-05,
+      "loss": 78.2757,
+      "loss/crossentropy": 8.918366432189941,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 6.912693023681641,
+      "step": 10
+    },
+    {
+      "epoch": 0.00024,
+      "grad_norm": 148.0,
+      "learning_rate": 2.0800000000000004e-05,
+      "loss": 74.4248,
+      "loss/crossentropy": 8.443636417388916,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 6.567321538925171,
+      "step": 12
+    },
+    {
+      "epoch": 0.00028,
+      "grad_norm": 131.0,
+      "learning_rate": 2.2600000000000004e-05,
+      "loss": 73.0003,
+      "loss/crossentropy": 8.428278923034668,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 6.706400156021118,
+      "step": 14
+    },
+    {
+      "epoch": 0.00032,
+      "grad_norm": 181.0,
+      "grad_norm_var": 16279.8625,
+      "learning_rate": 2.4400000000000004e-05,
+      "loss": 70.0047,
+      "loss/crossentropy": 8.216889381408691,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 6.056080102920532,
+      "step": 16
+    },
+    {
+      "epoch": 0.00036,
+      "grad_norm": 90.5,
+      "grad_norm_var": 14154.148958333333,
+      "learning_rate": 2.6200000000000003e-05,
+      "loss": 69.9766,
+      "loss/crossentropy": 8.191599607467651,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 6.429446697235107,
+      "step": 18
+    },
+    {
+      "epoch": 0.0004,
+      "grad_norm": 52.25,
+      "grad_norm_var": 12194.27890625,
+      "learning_rate": 2.8000000000000003e-05,
+      "loss": 64.3807,
+      "loss/crossentropy": 7.506032228469849,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 5.794633388519287,
+      "step": 20
+    },
+    {
+      "epoch": 0.00044,
+      "grad_norm": 39.25,
+      "grad_norm_var": 6249.4875,
+      "learning_rate": 2.9800000000000006e-05,
+      "loss": 61.2802,
+      "loss/crossentropy": 7.152851343154907,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 5.261489152908325,
+      "step": 22
+    },
+    {
+      "epoch": 0.00048,
+      "grad_norm": 57.0,
+      "grad_norm_var": 4626.8875,
+      "learning_rate": 3.16e-05,
+      "loss": 58.3454,
+      "loss/crossentropy": 6.956738471984863,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 5.020895004272461,
+      "step": 24
+    },
+    {
+      "epoch": 0.00052,
+      "grad_norm": 86.0,
+      "grad_norm_var": 4244.565625,
+      "learning_rate": 3.3400000000000005e-05,
+      "loss": 54.2703,
+      "loss/crossentropy": 6.686542987823486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 4.801911354064941,
+      "step": 26
+    },
+    {
+      "epoch": 0.00056,
+      "grad_norm": 110.5,
+      "grad_norm_var": 3868.5875,
+      "learning_rate": 3.520000000000001e-05,
+      "loss": 51.7343,
+      "loss/crossentropy": 6.4867262840271,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 4.5746169090271,
+      "step": 28
+    },
+    {
+      "epoch": 0.0006,
+      "grad_norm": 50.0,
+      "grad_norm_var": 3953.82890625,
+      "learning_rate": 3.7e-05,
+      "loss": 49.6807,
+      "loss/crossentropy": 6.364065408706665,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 4.230688810348511,
+      "step": 30
+    },
+    {
+      "epoch": 0.00064,
+      "grad_norm": 68.0,
+      "grad_norm_var": 3157.4958333333334,
+      "learning_rate": 3.88e-05,
+      "loss": 44.7112,
+      "loss/crossentropy": 5.731794834136963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 3.6969637870788574,
+      "step": 32
+    },
+    {
+      "epoch": 0.00068,
+      "grad_norm": 50.75,
+      "grad_norm_var": 500.59973958333336,
+      "learning_rate": 4.0600000000000004e-05,
+      "loss": 42.2923,
+      "loss/crossentropy": 5.553718328475952,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 3.7271878719329834,
+      "step": 34
+    },
+    {
+      "epoch": 0.00072,
+      "grad_norm": 55.5,
+      "grad_norm_var": 342.2122395833333,
+      "learning_rate": 4.240000000000001e-05,
+      "loss": 37.7465,
+      "loss/crossentropy": 5.023651361465454,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 3.2670111656188965,
+      "step": 36
+    },
+    {
+      "epoch": 0.00076,
+      "grad_norm": 75.5,
+      "grad_norm_var": 283.37395833333335,
+      "learning_rate": 4.420000000000001e-05,
+      "loss": 35.1313,
+      "loss/crossentropy": 4.921839237213135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 2.950661063194275,
+      "step": 38
+    },
+    {
+      "epoch": 0.0008,
+      "grad_norm": 44.5,
+      "grad_norm_var": 299.1958333333333,
+      "learning_rate": 4.600000000000001e-05,
+      "loss": 32.3316,
+      "loss/crossentropy": 4.782621145248413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 2.7473502159118652,
+      "step": 40
+    },
+    {
+      "epoch": 0.00084,
+      "grad_norm": 36.5,
+      "grad_norm_var": 361.1372395833333,
+      "learning_rate": 4.78e-05,
+      "loss": 28.4104,
+      "loss/crossentropy": 3.8754972219467163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 2.410745143890381,
+      "step": 42
+    },
+    {
+      "epoch": 0.00088,
+      "grad_norm": 36.25,
+      "grad_norm_var": 225.80598958333334,
+      "learning_rate": 4.96e-05,
+      "loss": 26.1806,
+      "loss/crossentropy": 3.9885865449905396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 2.1414425373077393,
+      "step": 44
+    },
+    {
+      "epoch": 0.00092,
+      "grad_norm": 46.0,
+      "grad_norm_var": 249.475,
+      "learning_rate": 5.14e-05,
+      "loss": 24.4012,
+      "loss/crossentropy": 3.750515580177307,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 2.063339352607727,
+      "step": 46
+    },
+    {
+      "epoch": 0.00096,
+      "grad_norm": 20.875,
+      "grad_norm_var": 315.8291015625,
+      "learning_rate": 5.3200000000000006e-05,
+      "loss": 22.822,
+      "loss/crossentropy": 3.7912577390670776,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 1.9196046590805054,
+      "step": 48
+    },
+    {
+      "epoch": 0.001,
+      "grad_norm": 35.5,
+      "grad_norm_var": 279.284375,
+      "learning_rate": 5.500000000000001e-05,
+      "loss": 21.036,
+      "loss/crossentropy": 3.777758002281189,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 1.7479944229125977,
+      "step": 50
+    },
+    {
+      "epoch": 0.00104,
+      "grad_norm": 20.75,
+      "grad_norm_var": 306.15149739583336,
+      "learning_rate": 5.680000000000001e-05,
+      "loss": 20.3608,
+      "loss/crossentropy": 3.5903185606002808,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 1.5533717274665833,
+      "step": 52
+    },
+    {
+      "epoch": 0.00108,
+      "grad_norm": 43.75,
+      "grad_norm_var": 213.07395833333334,
+      "learning_rate": 5.860000000000001e-05,
+      "loss": 18.813,
+      "loss/crossentropy": 3.691780686378479,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 1.4755533933639526,
+      "step": 54
+    },
+    {
+      "epoch": 0.00112,
+      "grad_norm": 21.25,
+      "grad_norm_var": 70.690625,
+      "learning_rate": 6.040000000000001e-05,
+      "loss": 19.1421,
+      "loss/crossentropy": 3.557003617286682,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 1.5198156833648682,
+      "step": 56
+    },
+    {
+      "epoch": 0.00116,
+      "grad_norm": 21.5,
+      "grad_norm_var": 76.30390625,
+      "learning_rate": 6.220000000000001e-05,
+      "loss": 17.2705,
+      "loss/crossentropy": 3.2730292081832886,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 1.4131997227668762,
+      "step": 58
+    },
+    {
+      "epoch": 0.0012,
+      "grad_norm": 19.875,
+      "grad_norm_var": 77.11399739583334,
+      "learning_rate": 6.400000000000001e-05,
+      "loss": 16.4712,
+      "loss/crossentropy": 3.419156074523926,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 1.3277101516723633,
+      "step": 60
+    },
+    {
+      "epoch": 0.00124,
+      "grad_norm": 25.75,
+      "grad_norm_var": 48.60520833333333,
+      "learning_rate": 6.58e-05,
+      "loss": 16.6219,
+      "loss/crossentropy": 2.973878502845764,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 1.3438007831573486,
+      "step": 62
+    },
+    {
+      "epoch": 0.00128,
+      "grad_norm": 34.5,
+      "grad_norm_var": 53.18020833333333,
+      "learning_rate": 6.76e-05,
+      "loss": 15.0929,
+      "loss/crossentropy": 2.892021059989929,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 1.1624282002449036,
+      "step": 64
+    },
+    {
+      "epoch": 0.00132,
+      "grad_norm": 15.4375,
+      "grad_norm_var": 51.195947265625,
+      "learning_rate": 6.94e-05,
+      "loss": 15.1967,
+      "loss/crossentropy": 2.954660177230835,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 1.116301715373993,
+      "step": 66
+    },
+    {
+      "epoch": 0.00136,
+      "grad_norm": 32.0,
+      "grad_norm_var": 51.064306640625,
+      "learning_rate": 7.120000000000001e-05,
+      "loss": 14.9397,
+      "loss/crossentropy": 3.2686156034469604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 1.1978037357330322,
+      "step": 68
+    },
+    {
+      "epoch": 0.0014,
+      "grad_norm": 29.25,
+      "grad_norm_var": 32.794270833333336,
+      "learning_rate": 7.3e-05,
+      "loss": 14.4846,
+      "loss/crossentropy": 2.7956581115722656,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 1.1973016262054443,
+      "step": 70
+    },
+    {
+      "epoch": 0.00144,
+      "grad_norm": 15.25,
+      "grad_norm_var": 37.80149739583333,
+      "learning_rate": 7.48e-05,
+      "loss": 14.1296,
+      "loss/crossentropy": 3.08966863155365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 1.151496708393097,
+      "step": 72
+    },
+    {
+      "epoch": 0.00148,
+      "grad_norm": 18.625,
+      "grad_norm_var": 41.88984375,
+      "learning_rate": 7.66e-05,
+      "loss": 13.6812,
+      "loss/crossentropy": 2.949987292289734,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.9787414371967316,
+      "step": 74
+    },
+    {
+      "epoch": 0.00152,
+      "grad_norm": 15.1875,
+      "grad_norm_var": 49.143489583333334,
+      "learning_rate": 7.840000000000001e-05,
+      "loss": 12.8901,
+      "loss/crossentropy": 3.1161292791366577,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 1.0640113949775696,
+      "step": 76
+    },
+    {
+      "epoch": 0.00156,
+      "grad_norm": 22.125,
+      "grad_norm_var": 47.63170572916667,
+      "learning_rate": 8.020000000000001e-05,
+      "loss": 13.157,
+      "loss/crossentropy": 3.3661664724349976,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 1.1353825330734253,
+      "step": 78
+    },
+    {
+      "epoch": 0.0016,
+      "grad_norm": 16.25,
+      "grad_norm_var": 35.33274739583333,
+      "learning_rate": 8.200000000000001e-05,
+      "loss": 12.9372,
+      "loss/crossentropy": 2.927241563796997,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.9984134435653687,
+      "step": 80
+    },
+    {
+      "epoch": 0.00164,
+      "grad_norm": 13.625,
+      "grad_norm_var": 37.53951822916667,
+      "learning_rate": 8.38e-05,
+      "loss": 12.0477,
+      "loss/crossentropy": 3.1273285150527954,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.9241160154342651,
+      "step": 82
+    },
+    {
+      "epoch": 0.00168,
+      "grad_norm": 19.125,
+      "grad_norm_var": 19.602718098958334,
+      "learning_rate": 8.560000000000001e-05,
+      "loss": 11.9084,
+      "loss/crossentropy": 2.737278938293457,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.8822851181030273,
+      "step": 84
+    },
+    {
+      "epoch": 0.00172,
+      "grad_norm": 13.0625,
+      "grad_norm_var": 11.107014973958334,
+      "learning_rate": 8.740000000000001e-05,
+      "loss": 11.8594,
+      "loss/crossentropy": 2.4452388286590576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.8693483769893646,
+      "step": 86
+    },
+    {
+      "epoch": 0.00176,
+      "grad_norm": 18.375,
+      "grad_norm_var": 11.328369140625,
+      "learning_rate": 8.92e-05,
+      "loss": 11.5058,
+      "loss/crossentropy": 2.89771831035614,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.8749278783798218,
+      "step": 88
+    },
+    {
+      "epoch": 0.0018,
+      "grad_norm": 12.1875,
+      "grad_norm_var": 13.811572265625,
+      "learning_rate": 9.1e-05,
+      "loss": 11.9281,
+      "loss/crossentropy": 3.0173208713531494,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.8714744746685028,
+      "step": 90
+    },
+    {
+      "epoch": 0.00184,
+      "grad_norm": 20.375,
+      "grad_norm_var": 16.114306640625,
+      "learning_rate": 9.28e-05,
+      "loss": 11.4244,
+      "loss/crossentropy": 2.588515043258667,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.9651442170143127,
+      "step": 92
+    },
+    {
+      "epoch": 0.00188,
+      "grad_norm": 13.375,
+      "grad_norm_var": 12.784309895833333,
+      "learning_rate": 9.46e-05,
+      "loss": 11.6092,
+      "loss/crossentropy": 2.8116774559020996,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.9143906235694885,
+      "step": 94
+    },
+    {
+      "epoch": 0.00192,
+      "grad_norm": 16.375,
+      "grad_norm_var": 13.225764973958333,
+      "learning_rate": 9.64e-05,
+      "loss": 10.6723,
+      "loss/crossentropy": 2.8734441995620728,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.814399927854538,
+      "step": 96
+    },
+    {
+      "epoch": 0.00196,
+      "grad_norm": 10.3125,
+      "grad_norm_var": 17.228238932291667,
+      "learning_rate": 9.82e-05,
+      "loss": 10.6577,
+      "loss/crossentropy": 2.664194703102112,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.7905783653259277,
+      "step": 98
+    },
+    {
+      "epoch": 0.002,
+      "grad_norm": 12.3125,
+      "grad_norm_var": 17.113785807291666,
+      "learning_rate": 0.0001,
+      "loss": 10.6847,
+      "loss/crossentropy": 2.4851003885269165,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.769305944442749,
+      "step": 100
+    },
+    {
+      "epoch": 0.00204,
+      "grad_norm": 10.75,
+      "grad_norm_var": 13.9140625,
+      "learning_rate": 0.0001,
+      "loss": 10.8119,
+      "loss/crossentropy": 2.2757182121276855,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.8241511881351471,
+      "step": 102
+    },
+    {
+      "epoch": 0.00208,
+      "grad_norm": 11.375,
+      "grad_norm_var": 13.615738932291666,
+      "learning_rate": 0.0001,
+      "loss": 10.6244,
+      "loss/crossentropy": 2.7211785316467285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.8518709540367126,
+      "step": 104
+    },
+    {
+      "epoch": 0.00212,
+      "grad_norm": 11.625,
+      "grad_norm_var": 14.163395182291667,
+      "learning_rate": 0.0001,
+      "loss": 10.5629,
+      "loss/crossentropy": 2.387019991874695,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.7568954229354858,
+      "step": 106
+    },
+    {
+      "epoch": 0.00216,
+      "grad_norm": 9.9375,
+      "grad_norm_var": 10.788525390625,
+      "learning_rate": 0.0001,
+      "loss": 10.1364,
+      "loss/crossentropy": 2.5363346338272095,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.7648341059684753,
+      "step": 108
+    },
+    {
+      "epoch": 0.0022,
+      "grad_norm": 20.0,
+      "grad_norm_var": 14.898893229166667,
+      "learning_rate": 0.0001,
+      "loss": 10.8773,
+      "loss/crossentropy": 2.8450236320495605,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.737193763256073,
+      "step": 110
+    },
+    {
+      "epoch": 0.00224,
+      "grad_norm": 10.25,
+      "grad_norm_var": 15.364518229166666,
+      "learning_rate": 0.0001,
+      "loss": 9.4554,
+      "loss/crossentropy": 2.4827451705932617,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6924614012241364,
+      "step": 112
+    },
+    {
+      "epoch": 0.00228,
+      "grad_norm": 10.5625,
+      "grad_norm_var": 7.461832682291667,
+      "learning_rate": 0.0001,
+      "loss": 9.9651,
+      "loss/crossentropy": 2.093318462371826,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.7038464546203613,
+      "step": 114
+    },
+    {
+      "epoch": 0.00232,
+      "grad_norm": 9.6875,
+      "grad_norm_var": 7.597330729166667,
+      "learning_rate": 0.0001,
+      "loss": 10.0297,
+      "loss/crossentropy": 2.5149790048599243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6980189085006714,
+      "step": 116
+    },
+    {
+      "epoch": 0.00236,
+      "grad_norm": 11.625,
+      "grad_norm_var": 6.672509765625,
+      "learning_rate": 0.0001,
+      "loss": 9.8176,
+      "loss/crossentropy": 2.6022276878356934,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.7005251348018646,
+      "step": 118
+    },
+    {
+      "epoch": 0.0024,
+      "grad_norm": 7.625,
+      "grad_norm_var": 7.298160807291667,
+      "learning_rate": 0.0001,
+      "loss": 9.5658,
+      "loss/crossentropy": 2.6836462020874023,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.7237774729728699,
+      "step": 120
+    },
+    {
+      "epoch": 0.00244,
+      "grad_norm": 9.5625,
+      "grad_norm_var": 7.402018229166667,
+      "learning_rate": 0.0001,
+      "loss": 9.7376,
+      "loss/crossentropy": 2.6823805570602417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.7570162117481232,
+      "step": 122
+    },
+    {
+      "epoch": 0.00248,
+      "grad_norm": 11.25,
+      "grad_norm_var": 7.391259765625,
+      "learning_rate": 0.0001,
+      "loss": 9.4713,
+      "loss/crossentropy": 2.6233514547348022,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.7183247208595276,
+      "step": 124
+    },
+    {
+      "epoch": 0.00252,
+      "grad_norm": 9.9375,
+      "grad_norm_var": 1.0839680989583333,
+      "learning_rate": 0.0001,
+      "loss": 9.2243,
+      "loss/crossentropy": 2.331676959991455,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6982125043869019,
+      "step": 126
+    },
+    {
+      "epoch": 0.00256,
+      "grad_norm": 9.1875,
+      "grad_norm_var": 0.9687337239583333,
+      "learning_rate": 0.0001,
+      "loss": 9.4777,
+      "loss/crossentropy": 2.429046392440796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.696417510509491,
+      "step": 128
+    },
+    {
+      "epoch": 0.0026,
+      "grad_norm": 14.8125,
+      "grad_norm_var": 2.448291015625,
+      "learning_rate": 0.0001,
+      "loss": 9.9024,
+      "loss/crossentropy": 2.5262571573257446,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.716008871793747,
+      "step": 130
+    },
+    {
+      "epoch": 0.00264,
+      "grad_norm": 8.8125,
+      "grad_norm_var": 2.7044270833333335,
+      "learning_rate": 0.0001,
+      "loss": 9.2836,
+      "loss/crossentropy": 2.187526524066925,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.666053056716919,
+      "step": 132
+    },
+    {
+      "epoch": 0.00268,
+      "grad_norm": 8.9375,
+      "grad_norm_var": 3.285270182291667,
+      "learning_rate": 0.0001,
+      "loss": 9.8338,
+      "loss/crossentropy": 2.4199057817459106,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6709816455841064,
+      "step": 134
+    },
+    {
+      "epoch": 0.00272,
+      "grad_norm": 9.6875,
+      "grad_norm_var": 3.4072916666666666,
+      "learning_rate": 0.0001,
+      "loss": 9.4225,
+      "loss/crossentropy": 2.1963008642196655,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5963725447654724,
+      "step": 136
+    },
+    {
+      "epoch": 0.00276,
+      "grad_norm": 8.0625,
+      "grad_norm_var": 3.556884765625,
+      "learning_rate": 0.0001,
+      "loss": 9.44,
+      "loss/crossentropy": 2.5878132581710815,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6362220048904419,
+      "step": 138
+    },
+    {
+      "epoch": 0.0028,
+      "grad_norm": 9.25,
+      "grad_norm_var": 3.4852701822916665,
+      "learning_rate": 0.0001,
+      "loss": 9.4314,
+      "loss/crossentropy": 2.7800480127334595,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6819457113742828,
+      "step": 140
+    },
+    {
+      "epoch": 0.00284,
+      "grad_norm": 8.4375,
+      "grad_norm_var": 3.838997395833333,
+      "learning_rate": 0.0001,
+      "loss": 9.1047,
+      "loss/crossentropy": 2.5055110454559326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.674308180809021,
+      "step": 142
+    },
+    {
+      "epoch": 0.00288,
+      "grad_norm": 8.25,
+      "grad_norm_var": 4.078499348958333,
+      "learning_rate": 0.0001,
+      "loss": 9.1578,
+      "loss/crossentropy": 2.8532944917678833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.7083481848239899,
+      "step": 144
+    },
+    {
+      "epoch": 0.00292,
+      "grad_norm": 7.8125,
+      "grad_norm_var": 2.2759765625,
+      "learning_rate": 0.0001,
+      "loss": 8.8023,
+      "loss/crossentropy": 2.442527174949646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6510869562625885,
+      "step": 146
+    },
+    {
+      "epoch": 0.00296,
+      "grad_norm": 10.9375,
+      "grad_norm_var": 2.8544881184895834,
+      "learning_rate": 0.0001,
+      "loss": 8.7238,
+      "loss/crossentropy": 2.516597867012024,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6362285614013672,
+      "step": 148
+    },
+    {
+      "epoch": 0.003,
+      "grad_norm": 7.40625,
+      "grad_norm_var": 1.8615885416666667,
+      "learning_rate": 0.0001,
+      "loss": 8.5543,
+      "loss/crossentropy": 2.8672900199890137,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6459421515464783,
+      "step": 150
+    },
+    {
+      "epoch": 0.00304,
+      "grad_norm": 8.1875,
+      "grad_norm_var": 1.7195963541666666,
+      "learning_rate": 0.0001,
+      "loss": 8.6403,
+      "loss/crossentropy": 2.2042795419692993,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5720505118370056,
+      "step": 152
+    },
+    {
+      "epoch": 0.00308,
+      "grad_norm": 7.8125,
+      "grad_norm_var": 1.876025390625,
+      "learning_rate": 0.0001,
+      "loss": 8.768,
+      "loss/crossentropy": 2.225563883781433,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6328675150871277,
+      "step": 154
+    },
+    {
+      "epoch": 0.00312,
+      "grad_norm": 8.0625,
+      "grad_norm_var": 1.5925618489583333,
+      "learning_rate": 0.0001,
+      "loss": 8.5743,
+      "loss/crossentropy": 2.3541462421417236,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5927431881427765,
+      "step": 156
+    },
+    {
+      "epoch": 0.00316,
+      "grad_norm": 7.34375,
+      "grad_norm_var": 1.834619140625,
+      "learning_rate": 0.0001,
+      "loss": 8.7329,
+      "loss/crossentropy": 2.4685616493225098,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6299368739128113,
+      "step": 158
+    },
+    {
+      "epoch": 0.0032,
+      "grad_norm": 9.25,
+      "grad_norm_var": 1.4429646809895833,
+      "learning_rate": 0.0001,
+      "loss": 8.4796,
+      "loss/crossentropy": 2.4637919664382935,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6084515154361725,
+      "step": 160
+    },
+    {
+      "epoch": 0.00324,
+      "grad_norm": 9.0625,
+      "grad_norm_var": 1.3692545572916666,
+      "learning_rate": 0.0001,
+      "loss": 9.0446,
+      "loss/crossentropy": 2.598397374153137,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.574739396572113,
+      "step": 162
+    },
+    {
+      "epoch": 0.00328,
+      "grad_norm": 7.15625,
+      "grad_norm_var": 0.9171834309895833,
+      "learning_rate": 0.0001,
+      "loss": 8.1508,
+      "loss/crossentropy": 2.5183030366897583,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5985860526561737,
+      "step": 164
+    },
+    {
+      "epoch": 0.00332,
+      "grad_norm": 9.125,
+      "grad_norm_var": 0.9571614583333333,
+      "learning_rate": 0.0001,
+      "loss": 8.4296,
+      "loss/crossentropy": 2.252183437347412,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5627379417419434,
+      "step": 166
+    },
+    {
+      "epoch": 0.00336,
+      "grad_norm": 7.875,
+      "grad_norm_var": 0.70533447265625,
+      "learning_rate": 0.0001,
+      "loss": 8.4549,
+      "loss/crossentropy": 2.5720516443252563,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5945309698581696,
+      "step": 168
+    },
+    {
+      "epoch": 0.0034,
+      "grad_norm": 9.1875,
+      "grad_norm_var": 0.8844034830729167,
+      "learning_rate": 0.0001,
+      "loss": 8.6096,
+      "loss/crossentropy": 2.3004332184791565,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5401738286018372,
+      "step": 170
+    },
+    {
+      "epoch": 0.00344,
+      "grad_norm": 7.71875,
+      "grad_norm_var": 0.948681640625,
+      "learning_rate": 0.0001,
+      "loss": 8.5484,
+      "loss/crossentropy": 2.689734935760498,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6384358406066895,
+      "step": 172
+    },
+    {
+      "epoch": 0.00348,
+      "grad_norm": 6.84375,
+      "grad_norm_var": 0.9418253580729167,
+      "learning_rate": 0.0001,
+      "loss": 8.1888,
+      "loss/crossentropy": 1.944397747516632,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5828913450241089,
+      "step": 174
+    },
+    {
+      "epoch": 0.00352,
+      "grad_norm": 6.53125,
+      "grad_norm_var": 0.8817342122395834,
+      "learning_rate": 0.0001,
+      "loss": 8.0577,
+      "loss/crossentropy": 2.8166507482528687,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5916908979415894,
+      "step": 176
+    },
+    {
+      "epoch": 0.00356,
+      "grad_norm": 7.09375,
+      "grad_norm_var": 0.7456990559895833,
+      "learning_rate": 0.0001,
+      "loss": 8.7701,
+      "loss/crossentropy": 2.3208402395248413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5968939661979675,
+      "step": 178
+    },
+    {
+      "epoch": 0.0036,
+      "grad_norm": 7.25,
+      "grad_norm_var": 0.7425130208333334,
+      "learning_rate": 0.0001,
+      "loss": 8.2615,
+      "loss/crossentropy": 2.817763566970825,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5800873041152954,
+      "step": 180
+    },
+    {
+      "epoch": 0.00364,
+      "grad_norm": 6.78125,
+      "grad_norm_var": 0.6126912434895834,
+      "learning_rate": 0.0001,
+      "loss": 8.3053,
+      "loss/crossentropy": 2.250023365020752,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5279016494750977,
+      "step": 182
+    },
+    {
+      "epoch": 0.00368,
+      "grad_norm": 6.09375,
+      "grad_norm_var": 0.6917805989583333,
+      "learning_rate": 0.0001,
+      "loss": 7.7974,
+      "loss/crossentropy": 2.1337096095085144,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5393811166286469,
+      "step": 184
+    },
+    {
+      "epoch": 0.00372,
+      "grad_norm": 6.34375,
+      "grad_norm_var": 0.5962198893229167,
+      "learning_rate": 0.0001,
+      "loss": 7.7258,
+      "loss/crossentropy": 2.6338934898376465,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6444519460201263,
+      "step": 186
+    },
+    {
+      "epoch": 0.00376,
+      "grad_norm": 10.4375,
+      "grad_norm_var": 1.2786458333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.0762,
+      "loss/crossentropy": 2.66677463054657,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6012931764125824,
+      "step": 188
+    },
+    {
+      "epoch": 0.0038,
+      "grad_norm": 6.375,
+      "grad_norm_var": 1.6642537434895834,
+      "learning_rate": 0.0001,
+      "loss": 8.3177,
+      "loss/crossentropy": 2.3731196522712708,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6299596428871155,
+      "step": 190
+    },
+    {
+      "epoch": 0.00384,
+      "grad_norm": 7.4375,
+      "grad_norm_var": 1.7719889322916667,
+      "learning_rate": 0.0001,
+      "loss": 8.214,
+      "loss/crossentropy": 2.411492705345154,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5328834354877472,
+      "step": 192
+    },
+    {
+      "epoch": 0.00388,
+      "grad_norm": 7.15625,
+      "grad_norm_var": 1.9489420572916667,
+      "learning_rate": 0.0001,
+      "loss": 7.9763,
+      "loss/crossentropy": 2.2402734756469727,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.539243072271347,
+      "step": 194
+    },
+    {
+      "epoch": 0.00392,
+      "grad_norm": 7.53125,
+      "grad_norm_var": 1.895556640625,
+      "learning_rate": 0.0001,
+      "loss": 7.9292,
+      "loss/crossentropy": 2.3250681161880493,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5374342203140259,
+      "step": 196
+    },
+    {
+      "epoch": 0.00396,
+      "grad_norm": 7.75,
+      "grad_norm_var": 1.8979451497395834,
+      "learning_rate": 0.0001,
+      "loss": 7.9201,
+      "loss/crossentropy": 2.42138135433197,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5410921573638916,
+      "step": 198
+    },
+    {
+      "epoch": 0.004,
+      "grad_norm": 6.78125,
+      "grad_norm_var": 1.8954386393229166,
+      "learning_rate": 0.0001,
+      "loss": 7.7597,
+      "loss/crossentropy": 2.1954251527786255,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.49367184937000275,
+      "step": 200
+    },
+    {
+      "epoch": 0.00404,
+      "grad_norm": 6.75,
+      "grad_norm_var": 1.6848307291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.9033,
+      "loss/crossentropy": 2.81479811668396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5567455887794495,
+      "step": 202
+    },
+    {
+      "epoch": 0.00408,
+      "grad_norm": 6.5,
+      "grad_norm_var": 1.08228759765625,
+      "learning_rate": 0.0001,
+      "loss": 7.9812,
+      "loss/crossentropy": 2.611761450767517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.544409453868866,
+      "step": 204
+    },
+    {
+      "epoch": 0.00412,
+      "grad_norm": 6.125,
+      "grad_norm_var": 0.650244140625,
+      "learning_rate": 0.0001,
+      "loss": 7.7921,
+      "loss/crossentropy": 2.1369245052337646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5181446373462677,
+      "step": 206
+    },
+    {
+      "epoch": 0.00416,
+      "grad_norm": 7.1875,
+      "grad_norm_var": 0.30745035807291665,
+      "learning_rate": 0.0001,
+      "loss": 8.3375,
+      "loss/crossentropy": 2.435856580734253,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5506252646446228,
+      "step": 208
+    },
+    {
+      "epoch": 0.0042,
+      "grad_norm": 6.21875,
+      "grad_norm_var": 0.26848958333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.7599,
+      "loss/crossentropy": 2.2404768466949463,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.46975645422935486,
+      "step": 210
+    },
+    {
+      "epoch": 0.00424,
+      "grad_norm": 6.71875,
+      "grad_norm_var": 0.19713541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.7083,
+      "loss/crossentropy": 2.4866777658462524,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5217231214046478,
+      "step": 212
+    },
+    {
+      "epoch": 0.00428,
+      "grad_norm": 6.15625,
+      "grad_norm_var": 0.121484375,
+      "learning_rate": 0.0001,
+      "loss": 7.6519,
+      "loss/crossentropy": 2.074867010116577,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4683506190776825,
+      "step": 214
+    },
+    {
+      "epoch": 0.00432,
+      "grad_norm": 7.59375,
+      "grad_norm_var": 0.21679280598958334,
+      "learning_rate": 0.0001,
+      "loss": 7.6062,
+      "loss/crossentropy": 2.2040151357650757,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5235520601272583,
+      "step": 216
+    },
+    {
+      "epoch": 0.00436,
+      "grad_norm": 7.25,
+      "grad_norm_var": 0.25690104166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.886,
+      "loss/crossentropy": 2.174479365348816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5243187248706818,
+      "step": 218
+    },
+    {
+      "epoch": 0.0044,
+      "grad_norm": 6.28125,
+      "grad_norm_var": 0.26523030598958336,
+      "learning_rate": 0.0001,
+      "loss": 7.7535,
+      "loss/crossentropy": 2.5678584575653076,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5818615555763245,
+      "step": 220
+    },
+    {
+      "epoch": 0.00444,
+      "grad_norm": 6.0625,
+      "grad_norm_var": 0.2814412434895833,
+      "learning_rate": 0.0001,
+      "loss": 7.859,
+      "loss/crossentropy": 2.4551891088485718,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5400412082672119,
+      "step": 222
+    },
+    {
+      "epoch": 0.00448,
+      "grad_norm": 6.15625,
+      "grad_norm_var": 0.25310872395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.7341,
+      "loss/crossentropy": 2.0638335943222046,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.49279752373695374,
+      "step": 224
+    },
+    {
+      "epoch": 0.00452,
+      "grad_norm": 6.6875,
+      "grad_norm_var": 0.38235677083333336,
+      "learning_rate": 0.0001,
+      "loss": 8.1064,
+      "loss/crossentropy": 2.553247332572937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5566797852516174,
+      "step": 226
+    },
+    {
+      "epoch": 0.00456,
+      "grad_norm": 6.0625,
+      "grad_norm_var": 0.39451497395833335,
+      "learning_rate": 0.0001,
+      "loss": 7.7812,
+      "loss/crossentropy": 2.544332265853882,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5118530094623566,
+      "step": 228
+    },
+    {
+      "epoch": 0.0046,
+      "grad_norm": 5.875,
+      "grad_norm_var": 0.4054036458333333,
+      "learning_rate": 0.0001,
+      "loss": 7.064,
+      "loss/crossentropy": 2.191234052181244,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5157675743103027,
+      "step": 230
+    },
+    {
+      "epoch": 0.00464,
+      "grad_norm": 5.4375,
+      "grad_norm_var": 0.360791015625,
+      "learning_rate": 0.0001,
+      "loss": 7.6611,
+      "loss/crossentropy": 2.3671151399612427,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5233491659164429,
+      "step": 232
+    },
+    {
+      "epoch": 0.00468,
+      "grad_norm": 7.125,
+      "grad_norm_var": 0.4180826822916667,
+      "learning_rate": 0.0001,
+      "loss": 7.4509,
+      "loss/crossentropy": 2.3003920316696167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5037694871425629,
+      "step": 234
+    },
+    {
+      "epoch": 0.00472,
+      "grad_norm": 5.625,
+      "grad_norm_var": 0.4305338541666667,
+      "learning_rate": 0.0001,
+      "loss": 7.8236,
+      "loss/crossentropy": 2.4672670364379883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5187652707099915,
+      "step": 236
+    },
+    {
+      "epoch": 0.00476,
+      "grad_norm": 6.0625,
+      "grad_norm_var": 0.4493326822916667,
+      "learning_rate": 0.0001,
+      "loss": 7.3246,
+      "loss/crossentropy": 2.179289937019348,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5111505687236786,
+      "step": 238
+    },
+    {
+      "epoch": 0.0048,
+      "grad_norm": 6.34375,
+      "grad_norm_var": 0.5123697916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.6064,
+      "loss/crossentropy": 2.2424627542495728,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5187103897333145,
+      "step": 240
+    },
+    {
+      "epoch": 0.00484,
+      "grad_norm": 5.96875,
+      "grad_norm_var": 0.2986979166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.8108,
+      "loss/crossentropy": 2.8024520874023438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5700482130050659,
+      "step": 242
+    },
+    {
+      "epoch": 0.00488,
+      "grad_norm": 6.25,
+      "grad_norm_var": 0.4554524739583333,
+      "learning_rate": 0.0001,
+      "loss": 7.6644,
+      "loss/crossentropy": 2.3653491735458374,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5232449471950531,
+      "step": 244
+    },
+    {
+      "epoch": 0.00492,
+      "grad_norm": 6.90625,
+      "grad_norm_var": 0.48513997395833336,
+      "learning_rate": 0.0001,
+      "loss": 7.5028,
+      "loss/crossentropy": 2.6201778650283813,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5684538185596466,
+      "step": 246
+    },
+    {
+      "epoch": 0.00496,
+      "grad_norm": 7.3125,
+      "grad_norm_var": 0.49654541015625,
+      "learning_rate": 0.0001,
+      "loss": 7.6631,
+      "loss/crossentropy": 2.2811471819877625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5543638169765472,
+      "step": 248
+    },
+    {
+      "epoch": 0.005,
+      "grad_norm": 5.625,
+      "grad_norm_var": 0.48544514973958336,
+      "learning_rate": 0.0001,
+      "loss": 7.6914,
+      "loss/crossentropy": 2.4381459951400757,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5594777166843414,
+      "step": 250
+    },
+    {
+      "epoch": 0.00504,
+      "grad_norm": 9.375,
+      "grad_norm_var": 1.04000244140625,
+      "learning_rate": 0.0001,
+      "loss": 7.7346,
+      "loss/crossentropy": 2.435782313346863,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5159732699394226,
+      "step": 252
+    },
+    {
+      "epoch": 0.00508,
+      "grad_norm": 5.46875,
+      "grad_norm_var": 1.0892862955729166,
+      "learning_rate": 0.0001,
+      "loss": 8.0282,
+      "loss/crossentropy": 2.7867215871810913,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.551640123128891,
+      "step": 254
+    },
+    {
+      "epoch": 0.00512,
+      "grad_norm": 5.46875,
+      "grad_norm_var": 1.2432902018229166,
+      "learning_rate": 0.0001,
+      "loss": 7.0744,
+      "loss/crossentropy": 1.9328945875167847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.45153285562992096,
+      "step": 256
+    },
+    {
+      "epoch": 0.00516,
+      "grad_norm": 6.21875,
+      "grad_norm_var": 1.2460286458333334,
+      "learning_rate": 0.0001,
+      "loss": 7.292,
+      "loss/crossentropy": 2.552613139152527,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5215992629528046,
+      "step": 258
+    },
+    {
+      "epoch": 0.0052,
+      "grad_norm": 5.40625,
+      "grad_norm_var": 1.1848958333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.4992,
+      "loss/crossentropy": 2.3720492124557495,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5024219453334808,
+      "step": 260
+    },
+    {
+      "epoch": 0.00524,
+      "grad_norm": 8.1875,
+      "grad_norm_var": 1.4898274739583333,
+      "learning_rate": 0.0001,
+      "loss": 7.4676,
+      "loss/crossentropy": 2.465815782546997,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5383751839399338,
+      "step": 262
+    },
+    {
+      "epoch": 0.00528,
+      "grad_norm": 6.0625,
+      "grad_norm_var": 1.5113240559895833,
+      "learning_rate": 0.0001,
+      "loss": 7.3163,
+      "loss/crossentropy": 2.2791935205459595,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.534159854054451,
+      "step": 264
+    },
+    {
+      "epoch": 0.00532,
+      "grad_norm": 6.28125,
+      "grad_norm_var": 1.3855305989583333,
+      "learning_rate": 0.0001,
+      "loss": 7.9279,
+      "loss/crossentropy": 2.48906409740448,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5344790518283844,
+      "step": 266
+    },
+    {
+      "epoch": 0.00536,
+      "grad_norm": 4.90625,
+      "grad_norm_var": 1.0180826822916667,
+      "learning_rate": 0.0001,
+      "loss": 7.3178,
+      "loss/crossentropy": 2.0858306884765625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.47646892070770264,
+      "step": 268
+    },
+    {
+      "epoch": 0.0054,
+      "grad_norm": 8.375,
+      "grad_norm_var": 1.18258056640625,
+      "learning_rate": 0.0001,
+      "loss": 7.383,
+      "loss/crossentropy": 2.159322738647461,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5543113648891449,
+      "step": 270
+    },
+    {
+      "epoch": 0.00544,
+      "grad_norm": 4.9375,
+      "grad_norm_var": 1.204931640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1635,
+      "loss/crossentropy": 2.249913454055786,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4896702766418457,
+      "step": 272
+    },
+    {
+      "epoch": 0.00548,
+      "grad_norm": 8.125,
+      "grad_norm_var": 1.388916015625,
+      "learning_rate": 0.0001,
+      "loss": 7.3565,
+      "loss/crossentropy": 1.998712420463562,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4770403504371643,
+      "step": 274
+    },
+    {
+      "epoch": 0.00552,
+      "grad_norm": 5.53125,
+      "grad_norm_var": 1.9266764322916667,
+      "learning_rate": 0.0001,
+      "loss": 7.6522,
+      "loss/crossentropy": 2.391260862350464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5259605348110199,
+      "step": 276
+    },
+    {
+      "epoch": 0.00556,
+      "grad_norm": 6.1875,
+      "grad_norm_var": 1.6884765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2935,
+      "loss/crossentropy": 2.523361325263977,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4724537879228592,
+      "step": 278
+    },
+    {
+      "epoch": 0.0056,
+      "grad_norm": 5.34375,
+      "grad_norm_var": 1.7386555989583334,
+      "learning_rate": 0.0001,
+      "loss": 7.3505,
+      "loss/crossentropy": 2.281963586807251,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.49724647402763367,
+      "step": 280
+    },
+    {
+      "epoch": 0.00564,
+      "grad_norm": 5.0625,
+      "grad_norm_var": 1.8446451822916667,
+      "learning_rate": 0.0001,
+      "loss": 7.1079,
+      "loss/crossentropy": 2.2403814792633057,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.45384156703948975,
+      "step": 282
+    },
+    {
+      "epoch": 0.00568,
+      "grad_norm": 5.6875,
+      "grad_norm_var": 1.670166015625,
+      "learning_rate": 0.0001,
+      "loss": 7.4318,
+      "loss/crossentropy": 2.2687963247299194,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.46648281812667847,
+      "step": 284
+    },
+    {
+      "epoch": 0.00572,
+      "grad_norm": 5.75,
+      "grad_norm_var": 1.38541259765625,
+      "learning_rate": 0.0001,
+      "loss": 7.3035,
+      "loss/crossentropy": 2.3336217403411865,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4792183041572571,
+      "step": 286
+    },
+    {
+      "epoch": 0.00576,
+      "grad_norm": 6.4375,
+      "grad_norm_var": 1.292431640625,
+      "learning_rate": 0.0001,
+      "loss": 7.0031,
+      "loss/crossentropy": 2.4006571769714355,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.473391056060791,
+      "step": 288
+    },
+    {
+      "epoch": 0.0058,
+      "grad_norm": 5.09375,
+      "grad_norm_var": 1.18671875,
+      "learning_rate": 0.0001,
+      "loss": 6.8037,
+      "loss/crossentropy": 2.0306124687194824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4722501188516617,
+      "step": 290
+    },
+    {
+      "epoch": 0.00584,
+      "grad_norm": 5.34375,
+      "grad_norm_var": 0.38865559895833335,
+      "learning_rate": 0.0001,
+      "loss": 7.1378,
+      "loss/crossentropy": 2.412277102470398,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.48340730369091034,
+      "step": 292
+    },
+    {
+      "epoch": 0.00588,
+      "grad_norm": 6.21875,
+      "grad_norm_var": 0.3952433268229167,
+      "learning_rate": 0.0001,
+      "loss": 7.3589,
+      "loss/crossentropy": 2.3195769786834717,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5074818134307861,
+      "step": 294
+    },
+    {
+      "epoch": 0.00592,
+      "grad_norm": 6.71875,
+      "grad_norm_var": 0.2867024739583333,
+      "learning_rate": 0.0001,
+      "loss": 7.2826,
+      "loss/crossentropy": 2.3265275955200195,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5351093411445618,
+      "step": 296
+    },
+    {
+      "epoch": 0.00596,
+      "grad_norm": 5.625,
+      "grad_norm_var": 0.26171468098958334,
+      "learning_rate": 0.0001,
+      "loss": 7.0106,
+      "loss/crossentropy": 2.210574746131897,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.45469868183135986,
+      "step": 298
+    },
+    {
+      "epoch": 0.006,
+      "grad_norm": 6.3125,
+      "grad_norm_var": 0.3034138997395833,
+      "learning_rate": 0.0001,
+      "loss": 7.4741,
+      "loss/crossentropy": 2.347964644432068,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.45269395411014557,
+      "step": 300
+    },
+    {
+      "epoch": 0.00604,
+      "grad_norm": 5.0,
+      "grad_norm_var": 0.34308268229166666,
+      "learning_rate": 0.0001,
+      "loss": 6.8901,
+      "loss/crossentropy": 2.197494626045227,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.45010584592819214,
+      "step": 302
+    },
+    {
+      "epoch": 0.00608,
+      "grad_norm": 5.8125,
+      "grad_norm_var": 0.32224934895833335,
+      "learning_rate": 0.0001,
+      "loss": 6.909,
+      "loss/crossentropy": 2.3295196890830994,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.48989084362983704,
+      "step": 304
+    },
+    {
+      "epoch": 0.00612,
+      "grad_norm": 6.3125,
+      "grad_norm_var": 0.26347249348958335,
+      "learning_rate": 0.0001,
+      "loss": 7.6606,
+      "loss/crossentropy": 2.5208678245544434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.46991507709026337,
+      "step": 306
+    },
+    {
+      "epoch": 0.00616,
+      "grad_norm": 5.40625,
+      "grad_norm_var": 0.22849934895833332,
+      "learning_rate": 0.0001,
+      "loss": 7.2881,
+      "loss/crossentropy": 2.6091307401657104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5090319812297821,
+      "step": 308
+    },
+    {
+      "epoch": 0.0062,
+      "grad_norm": 5.4375,
+      "grad_norm_var": 0.3083984375,
+      "learning_rate": 0.0001,
+      "loss": 7.0888,
+      "loss/crossentropy": 2.4142966270446777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.45959460735321045,
+      "step": 310
+    },
+    {
+      "epoch": 0.00624,
+      "grad_norm": 6.0625,
+      "grad_norm_var": 0.25006103515625,
+      "learning_rate": 0.0001,
+      "loss": 7.3054,
+      "loss/crossentropy": 2.3062673807144165,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4381408095359802,
+      "step": 312
+    },
+    {
+      "epoch": 0.00628,
+      "grad_norm": 4.875,
+      "grad_norm_var": 0.29498697916666666,
+      "learning_rate": 0.0001,
+      "loss": 6.5202,
+      "loss/crossentropy": 2.1124885082244873,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4392775595188141,
+      "step": 314
+    },
+    {
+      "epoch": 0.00632,
+      "grad_norm": 5.09375,
+      "grad_norm_var": 0.3001302083333333,
+      "learning_rate": 0.0001,
+      "loss": 6.3297,
+      "loss/crossentropy": 2.0250568985939026,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4133095294237137,
+      "step": 316
+    },
+    {
+      "epoch": 0.00636,
+      "grad_norm": 5.625,
+      "grad_norm_var": 0.31021728515625,
+      "learning_rate": 0.0001,
+      "loss": 6.9903,
+      "loss/crossentropy": 2.4011316299438477,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.46330246329307556,
+      "step": 318
+    },
+    {
+      "epoch": 0.0064,
+      "grad_norm": 5.65625,
+      "grad_norm_var": 0.30305582682291665,
+      "learning_rate": 0.0001,
+      "loss": 7.2114,
+      "loss/crossentropy": 2.487559676170349,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.47237157821655273,
+      "step": 320
+    },
+    {
+      "epoch": 0.00644,
+      "grad_norm": 5.1875,
+      "grad_norm_var": 0.2775349934895833,
+      "learning_rate": 0.0001,
+      "loss": 6.5935,
+      "loss/crossentropy": 1.999566912651062,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4165455400943756,
+      "step": 322
+    },
+    {
+      "epoch": 0.00648,
+      "grad_norm": 6.03125,
+      "grad_norm_var": 0.27496337890625,
+      "learning_rate": 0.0001,
+      "loss": 7.0573,
+      "loss/crossentropy": 2.545841693878174,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4725654572248459,
+      "step": 324
+    },
+    {
+      "epoch": 0.00652,
+      "grad_norm": 5.0625,
+      "grad_norm_var": 0.2528483072916667,
+      "learning_rate": 0.0001,
+      "loss": 7.2351,
+      "loss/crossentropy": 2.119086444377899,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4649319499731064,
+      "step": 326
+    },
+    {
+      "epoch": 0.00656,
+      "grad_norm": 5.21875,
+      "grad_norm_var": 0.221728515625,
+      "learning_rate": 0.0001,
+      "loss": 6.8367,
+      "loss/crossentropy": 2.365525245666504,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5121739506721497,
+      "step": 328
+    },
+    {
+      "epoch": 0.0066,
+      "grad_norm": 5.25,
+      "grad_norm_var": 0.17467041015625,
+      "learning_rate": 0.0001,
+      "loss": 6.8384,
+      "loss/crossentropy": 2.2604740858078003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4704676419496536,
+      "step": 330
+    },
+    {
+      "epoch": 0.00664,
+      "grad_norm": 6.15625,
+      "grad_norm_var": 0.18977864583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.5125,
+      "loss/crossentropy": 2.4891955852508545,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5196183770895004,
+      "step": 332
+    },
+    {
+      "epoch": 0.00668,
+      "grad_norm": 5.46875,
+      "grad_norm_var": 0.1767578125,
+      "learning_rate": 0.0001,
+      "loss": 7.3139,
+      "loss/crossentropy": 2.430082321166992,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.47671228647232056,
+      "step": 334
+    },
+    {
+      "epoch": 0.00672,
+      "grad_norm": 5.53125,
+      "grad_norm_var": 0.187353515625,
+      "learning_rate": 0.0001,
+      "loss": 6.6969,
+      "loss/crossentropy": 2.2450510263442993,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5124029517173767,
+      "step": 336
+    },
+    {
+      "epoch": 0.00676,
+      "grad_norm": 5.375,
+      "grad_norm_var": 0.18251546223958334,
+      "learning_rate": 0.0001,
+      "loss": 6.8537,
+      "loss/crossentropy": 2.225212812423706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4358299970626831,
+      "step": 338
+    },
+    {
+      "epoch": 0.0068,
+      "grad_norm": 6.71875,
+      "grad_norm_var": 26.47734375,
+      "learning_rate": 0.0001,
+      "loss": 6.8775,
+      "loss/crossentropy": 2.320846140384674,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.42757023870944977,
+      "step": 340
+    },
+    {
+      "epoch": 0.00684,
+      "grad_norm": 4.84375,
+      "grad_norm_var": 26.739453125,
+      "learning_rate": 0.0001,
+      "loss": 6.7394,
+      "loss/crossentropy": 2.419093132019043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4633703678846359,
+      "step": 342
+    },
+    {
+      "epoch": 0.00688,
+      "grad_norm": 5.3125,
+      "grad_norm_var": 26.632405598958332,
+      "learning_rate": 0.0001,
+      "loss": 6.7304,
+      "loss/crossentropy": 1.939517080783844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4387335330247879,
+      "step": 344
+    },
+    {
+      "epoch": 0.00692,
+      "grad_norm": 5.75,
+      "grad_norm_var": 26.504410807291666,
+      "learning_rate": 0.0001,
+      "loss": 7.0914,
+      "loss/crossentropy": 2.695888638496399,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.48031529784202576,
+      "step": 346
+    },
+    {
+      "epoch": 0.00696,
+      "grad_norm": 6.625,
+      "grad_norm_var": 26.468094889322916,
+      "learning_rate": 0.0001,
+      "loss": 6.8381,
+      "loss/crossentropy": 2.245330333709717,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5032568573951721,
+      "step": 348
+    },
+    {
+      "epoch": 0.007,
+      "grad_norm": 4.28125,
+      "grad_norm_var": 26.707421875,
+      "learning_rate": 0.0001,
+      "loss": 6.4774,
+      "loss/crossentropy": 1.9668607115745544,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4390462785959244,
+      "step": 350
+    },
+    {
+      "epoch": 0.00704,
+      "grad_norm": 5.25,
+      "grad_norm_var": 26.92008056640625,
+      "learning_rate": 0.0001,
+      "loss": 6.7795,
+      "loss/crossentropy": 2.4035123586654663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.43734824657440186,
+      "step": 352
+    },
+    {
+      "epoch": 0.00708,
+      "grad_norm": 6.5,
+      "grad_norm_var": 26.851460774739582,
+      "learning_rate": 0.0001,
+      "loss": 7.3932,
+      "loss/crossentropy": 2.4636529684066772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5208825469017029,
+      "step": 354
+    },
+    {
+      "epoch": 0.00712,
+      "grad_norm": 6.4375,
+      "grad_norm_var": 0.5743123372395833,
+      "learning_rate": 0.0001,
+      "loss": 6.8482,
+      "loss/crossentropy": 2.085066556930542,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3971068561077118,
+      "step": 356
+    },
+    {
+      "epoch": 0.00716,
+      "grad_norm": 4.59375,
+      "grad_norm_var": 0.57847900390625,
+      "learning_rate": 0.0001,
+      "loss": 6.8962,
+      "loss/crossentropy": 2.194266200065613,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.41747787594795227,
+      "step": 358
+    },
+    {
+      "epoch": 0.0072,
+      "grad_norm": 4.59375,
+      "grad_norm_var": 0.649462890625,
+      "learning_rate": 0.0001,
+      "loss": 6.7381,
+      "loss/crossentropy": 2.4678618907928467,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4633233994245529,
+      "step": 360
+    },
+    {
+      "epoch": 0.00724,
+      "grad_norm": 5.59375,
+      "grad_norm_var": 0.6476847330729166,
+      "learning_rate": 0.0001,
+      "loss": 6.5748,
+      "loss/crossentropy": 2.362962484359741,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4360974431037903,
+      "step": 362
+    },
+    {
+      "epoch": 0.00728,
+      "grad_norm": 5.6875,
+      "grad_norm_var": 0.5360514322916666,
+      "learning_rate": 0.0001,
+      "loss": 7.3497,
+      "loss/crossentropy": 2.3162096738815308,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4641396403312683,
+      "step": 364
+    },
+    {
+      "epoch": 0.00732,
+      "grad_norm": 6.0625,
+      "grad_norm_var": 0.4325358072916667,
+      "learning_rate": 0.0001,
+      "loss": 7.0856,
+      "loss/crossentropy": 2.279396176338196,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4331911951303482,
+      "step": 366
+    },
+    {
+      "epoch": 0.00736,
+      "grad_norm": 4.96875,
+      "grad_norm_var": 0.377587890625,
+      "learning_rate": 0.0001,
+      "loss": 6.8288,
+      "loss/crossentropy": 2.333961606025696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.43770918250083923,
+      "step": 368
+    },
+    {
+      "epoch": 0.0074,
+      "grad_norm": 5.1875,
+      "grad_norm_var": 0.27955322265625,
+      "learning_rate": 0.0001,
+      "loss": 6.9245,
+      "loss/crossentropy": 2.130259871482849,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4134685546159744,
+      "step": 370
+    },
+    {
+      "epoch": 0.00744,
+      "grad_norm": 5.59375,
+      "grad_norm_var": 0.19478759765625,
+      "learning_rate": 0.0001,
+      "loss": 6.4884,
+      "loss/crossentropy": 2.3000282049179077,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.46207693219184875,
+      "step": 372
+    },
+    {
+      "epoch": 0.00748,
+      "grad_norm": 5.59375,
+      "grad_norm_var": 0.20399983723958334,
+      "learning_rate": 0.0001,
+      "loss": 7.3714,
+      "loss/crossentropy": 2.687412142753601,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.46891947090625763,
+      "step": 374
+    },
+    {
+      "epoch": 0.00752,
+      "grad_norm": 4.28125,
+      "grad_norm_var": 0.21962483723958334,
+      "learning_rate": 0.0001,
+      "loss": 6.4194,
+      "loss/crossentropy": 2.2366563081741333,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.42876073718070984,
+      "step": 376
+    },
+    {
+      "epoch": 0.00756,
+      "grad_norm": 5.71875,
+      "grad_norm_var": 0.23108317057291666,
+      "learning_rate": 0.0001,
+      "loss": 7.0141,
+      "loss/crossentropy": 2.5960274934768677,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4519210159778595,
+      "step": 378
+    },
+    {
+      "epoch": 0.0076,
+      "grad_norm": 5.71875,
+      "grad_norm_var": 0.23435872395833332,
+      "learning_rate": 0.0001,
+      "loss": 6.9654,
+      "loss/crossentropy": 2.4690704345703125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5009289383888245,
+      "step": 380
+    },
+    {
+      "epoch": 0.00764,
+      "grad_norm": 5.25,
+      "grad_norm_var": 0.17952067057291668,
+      "learning_rate": 0.0001,
+      "loss": 6.6068,
+      "loss/crossentropy": 2.188890814781189,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4912077784538269,
+      "step": 382
+    },
+    {
+      "epoch": 0.00768,
+      "grad_norm": 5.15625,
+      "grad_norm_var": 0.19055582682291666,
+      "learning_rate": 0.0001,
+      "loss": 6.5789,
+      "loss/crossentropy": 2.2374125719070435,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4437579810619354,
+      "step": 384
+    },
+    {
+      "epoch": 0.00772,
+      "grad_norm": 5.1875,
+      "grad_norm_var": 0.19308268229166667,
+      "learning_rate": 0.0001,
+      "loss": 6.8081,
+      "loss/crossentropy": 2.101546287536621,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.39717453718185425,
+      "step": 386
+    },
+    {
+      "epoch": 0.00776,
+      "grad_norm": 5.1875,
+      "grad_norm_var": 0.182666015625,
+      "learning_rate": 0.0001,
+      "loss": 6.5688,
+      "loss/crossentropy": 2.2907408475875854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4657403528690338,
+      "step": 388
+    },
+    {
+      "epoch": 0.0078,
+      "grad_norm": 4.53125,
+      "grad_norm_var": 0.18203125,
+      "learning_rate": 0.0001,
+      "loss": 6.4664,
+      "loss/crossentropy": 1.9909976720809937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.39810416102409363,
+      "step": 390
+    },
+    {
+      "epoch": 0.00784,
+      "grad_norm": 6.28125,
+      "grad_norm_var": 0.23746337890625,
+      "learning_rate": 0.0001,
+      "loss": 6.6015,
+      "loss/crossentropy": 2.109456777572632,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4226441979408264,
+      "step": 392
+    },
+    {
+      "epoch": 0.00788,
+      "grad_norm": 6.03125,
+      "grad_norm_var": 0.26549072265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2592,
+      "loss/crossentropy": 2.330615997314453,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5082881152629852,
+      "step": 394
+    },
+    {
+      "epoch": 0.00792,
+      "grad_norm": 5.90625,
+      "grad_norm_var": 0.49332275390625,
+      "learning_rate": 0.0001,
+      "loss": 7.301,
+      "loss/crossentropy": 2.37632155418396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5454596877098083,
+      "step": 396
+    },
+    {
+      "epoch": 0.00796,
+      "grad_norm": 4.9375,
+      "grad_norm_var": 0.5669921875,
+      "learning_rate": 0.0001,
+      "loss": 6.6057,
+      "loss/crossentropy": 2.0303866863250732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3780263066291809,
+      "step": 398
+    },
+    {
+      "epoch": 0.008,
+      "grad_norm": 4.1875,
+      "grad_norm_var": 0.6671712239583333,
+      "learning_rate": 0.0001,
+      "loss": 6.8129,
+      "loss/crossentropy": 2.077945590019226,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4111042767763138,
+      "step": 400
+    },
+    {
+      "epoch": 0.00804,
+      "grad_norm": 5.5625,
+      "grad_norm_var": 0.6867146809895833,
+      "learning_rate": 0.0001,
+      "loss": 6.919,
+      "loss/crossentropy": 2.3042391538619995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.44224822521209717,
+      "step": 402
+    },
+    {
+      "epoch": 0.00808,
+      "grad_norm": 4.3125,
+      "grad_norm_var": 0.74537353515625,
+      "learning_rate": 0.0001,
+      "loss": 6.4785,
+      "loss/crossentropy": 2.15978467464447,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.44845885038375854,
+      "step": 404
+    },
+    {
+      "epoch": 0.00812,
+      "grad_norm": 4.90625,
+      "grad_norm_var": 0.69576416015625,
+      "learning_rate": 0.0001,
+      "loss": 6.3875,
+      "loss/crossentropy": 2.4571259021759033,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4629521369934082,
+      "step": 406
+    },
+    {
+      "epoch": 0.00816,
+      "grad_norm": 5.40625,
+      "grad_norm_var": 0.6654947916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1165,
+      "loss/crossentropy": 2.653234601020813,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5556869208812714,
+      "step": 408
+    },
+    {
+      "epoch": 0.0082,
+      "grad_norm": 4.53125,
+      "grad_norm_var": 0.66138916015625,
+      "learning_rate": 0.0001,
+      "loss": 6.6643,
+      "loss/crossentropy": 1.9738067388534546,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.42570993304252625,
+      "step": 410
+    },
+    {
+      "epoch": 0.00824,
+      "grad_norm": 4.0625,
+      "grad_norm_var": 0.46920166015625,
+      "learning_rate": 0.0001,
+      "loss": 6.2205,
+      "loss/crossentropy": 2.093988060951233,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40544557571411133,
+      "step": 412
+    },
+    {
+      "epoch": 0.00828,
+      "grad_norm": 4.34375,
+      "grad_norm_var": 0.36213785807291665,
+      "learning_rate": 0.0001,
+      "loss": 6.6356,
+      "loss/crossentropy": 2.4798851013183594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4204765260219574,
+      "step": 414
+    },
+    {
+      "epoch": 0.00832,
+      "grad_norm": 5.625,
+      "grad_norm_var": 0.3986287434895833,
+      "learning_rate": 0.0001,
+      "loss": 6.5601,
+      "loss/crossentropy": 2.4342020750045776,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5148662775754929,
+      "step": 416
+    },
+    {
+      "epoch": 0.00836,
+      "grad_norm": 5.40625,
+      "grad_norm_var": 0.3985514322916667,
+      "learning_rate": 0.0001,
+      "loss": 6.7757,
+      "loss/crossentropy": 2.3637804985046387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.45475171506404877,
+      "step": 418
+    },
+    {
+      "epoch": 0.0084,
+      "grad_norm": 4.09375,
+      "grad_norm_var": 0.39586181640625,
+      "learning_rate": 0.0001,
+      "loss": 6.6923,
+      "loss/crossentropy": 2.4066261053085327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.45181581377983093,
+      "step": 420
+    },
+    {
+      "epoch": 0.00844,
+      "grad_norm": 4.0,
+      "grad_norm_var": 0.43176676432291666,
+      "learning_rate": 0.0001,
+      "loss": 6.2428,
+      "loss/crossentropy": 2.1273797750473022,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3976929485797882,
+      "step": 422
+    },
+    {
+      "epoch": 0.00848,
+      "grad_norm": 4.90625,
+      "grad_norm_var": 0.26330973307291666,
+      "learning_rate": 0.0001,
+      "loss": 6.6524,
+      "loss/crossentropy": 2.4227113723754883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.46154454350471497,
+      "step": 424
+    },
+    {
+      "epoch": 0.00852,
+      "grad_norm": 4.65625,
+      "grad_norm_var": 0.261181640625,
+      "learning_rate": 0.0001,
+      "loss": 6.6558,
+      "loss/crossentropy": 2.3502479791641235,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.43580910563468933,
+      "step": 426
+    },
+    {
+      "epoch": 0.00856,
+      "grad_norm": 5.3125,
+      "grad_norm_var": 0.26236572265625,
+      "learning_rate": 0.0001,
+      "loss": 6.903,
+      "loss/crossentropy": 2.5034282207489014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4730219095945358,
+      "step": 428
+    },
+    {
+      "epoch": 0.0086,
+      "grad_norm": 4.40625,
+      "grad_norm_var": 0.2557576497395833,
+      "learning_rate": 0.0001,
+      "loss": 6.2148,
+      "loss/crossentropy": 2.0902098417282104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.409458264708519,
+      "step": 430
+    },
+    {
+      "epoch": 0.00864,
+      "grad_norm": 4.9375,
+      "grad_norm_var": 0.19420166015625,
+      "learning_rate": 0.0001,
+      "loss": 6.4634,
+      "loss/crossentropy": 2.2203429341316223,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.41066519916057587,
+      "step": 432
+    },
+    {
+      "epoch": 0.00868,
+      "grad_norm": 5.03125,
+      "grad_norm_var": 0.14855143229166667,
+      "learning_rate": 0.0001,
+      "loss": 6.6943,
+      "loss/crossentropy": 2.568304419517517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.48846572637557983,
+      "step": 434
+    },
+    {
+      "epoch": 0.00872,
+      "grad_norm": 5.15625,
+      "grad_norm_var": 0.13489176432291666,
+      "learning_rate": 0.0001,
+      "loss": 6.4829,
+      "loss/crossentropy": 2.359646439552307,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.43610572814941406,
+      "step": 436
+    },
+    {
+      "epoch": 0.00876,
+      "grad_norm": 5.03125,
+      "grad_norm_var": 0.08870035807291667,
+      "learning_rate": 0.0001,
+      "loss": 6.6119,
+      "loss/crossentropy": 2.2751121520996094,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.45837563276290894,
+      "step": 438
+    },
+    {
+      "epoch": 0.0088,
+      "grad_norm": 4.125,
+      "grad_norm_var": 0.13665364583333334,
+      "learning_rate": 0.0001,
+      "loss": 6.5338,
+      "loss/crossentropy": 2.334506392478943,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4196038395166397,
+      "step": 440
+    },
+    {
+      "epoch": 0.00884,
+      "grad_norm": 4.1875,
+      "grad_norm_var": 0.15891927083333332,
+      "learning_rate": 0.0001,
+      "loss": 6.2206,
+      "loss/crossentropy": 1.9731069803237915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.41661541163921356,
+      "step": 442
+    },
+    {
+      "epoch": 0.00888,
+      "grad_norm": 5.03125,
+      "grad_norm_var": 0.13339436848958333,
+      "learning_rate": 0.0001,
+      "loss": 6.3377,
+      "loss/crossentropy": 2.319058418273926,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4656156301498413,
+      "step": 444
+    },
+    {
+      "epoch": 0.00892,
+      "grad_norm": 4.125,
+      "grad_norm_var": 0.15013020833333332,
+      "learning_rate": 0.0001,
+      "loss": 6.5345,
+      "loss/crossentropy": 2.309122085571289,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.41400712728500366,
+      "step": 446
+    },
+    {
+      "epoch": 0.00896,
+      "grad_norm": 4.53125,
+      "grad_norm_var": 0.14468994140625,
+      "learning_rate": 0.0001,
+      "loss": 6.5385,
+      "loss/crossentropy": 1.867617905139923,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.39080700278282166,
+      "step": 448
+    },
+    {
+      "epoch": 0.009,
+      "grad_norm": 5.21875,
+      "grad_norm_var": 0.79830322265625,
+      "learning_rate": 0.0001,
+      "loss": 6.799,
+      "loss/crossentropy": 2.205033838748932,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4887084364891052,
+      "step": 450
+    },
+    {
+      "epoch": 0.00904,
+      "grad_norm": 4.9375,
+      "grad_norm_var": 0.826171875,
+      "learning_rate": 0.0001,
+      "loss": 6.6476,
+      "loss/crossentropy": 2.3054174184799194,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4140937328338623,
+      "step": 452
+    },
+    {
+      "epoch": 0.00908,
+      "grad_norm": 4.53125,
+      "grad_norm_var": 0.8572224934895833,
+      "learning_rate": 0.0001,
+      "loss": 6.7036,
+      "loss/crossentropy": 2.1358219981193542,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3849467635154724,
+      "step": 454
+    },
+    {
+      "epoch": 0.00912,
+      "grad_norm": 4.9375,
+      "grad_norm_var": 0.8132120768229166,
+      "learning_rate": 0.0001,
+      "loss": 6.4024,
+      "loss/crossentropy": 1.9811997413635254,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.42263032495975494,
+      "step": 456
+    },
+    {
+      "epoch": 0.00916,
+      "grad_norm": 4.8125,
+      "grad_norm_var": 0.7787109375,
+      "learning_rate": 0.0001,
+      "loss": 6.8381,
+      "loss/crossentropy": 2.319555103778839,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4539293050765991,
+      "step": 458
+    },
+    {
+      "epoch": 0.0092,
+      "grad_norm": 4.6875,
+      "grad_norm_var": 0.7744425455729167,
+      "learning_rate": 0.0001,
+      "loss": 6.72,
+      "loss/crossentropy": 2.4030569791793823,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.41373930871486664,
+      "step": 460
+    },
+    {
+      "epoch": 0.00924,
+      "grad_norm": 4.8125,
+      "grad_norm_var": 0.7218587239583333,
+      "learning_rate": 0.0001,
+      "loss": 6.7376,
+      "loss/crossentropy": 2.4479328393936157,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.447835311293602,
+      "step": 462
+    },
+    {
+      "epoch": 0.00928,
+      "grad_norm": 5.5,
+      "grad_norm_var": 0.7289021809895834,
+      "learning_rate": 0.0001,
+      "loss": 7.0562,
+      "loss/crossentropy": 2.056324601173401,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4374549984931946,
+      "step": 464
+    },
+    {
+      "epoch": 0.00932,
+      "grad_norm": 4.28125,
+      "grad_norm_var": 0.19256184895833334,
+      "learning_rate": 0.0001,
+      "loss": 6.6994,
+      "loss/crossentropy": 2.4104079008102417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4148041307926178,
+      "step": 466
+    },
+    {
+      "epoch": 0.00936,
+      "grad_norm": 4.75,
+      "grad_norm_var": 0.13527018229166668,
+      "learning_rate": 0.0001,
+      "loss": 7.0758,
+      "loss/crossentropy": 2.734652876853943,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.511719822883606,
+      "step": 468
+    },
+    {
+      "epoch": 0.0094,
+      "grad_norm": 5.0,
+      "grad_norm_var": 0.13118082682291668,
+      "learning_rate": 0.0001,
+      "loss": 6.4772,
+      "loss/crossentropy": 2.1748571395874023,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3852091133594513,
+      "step": 470
+    },
+    {
+      "epoch": 0.00944,
+      "grad_norm": 4.625,
+      "grad_norm_var": 0.13912353515625,
+      "learning_rate": 0.0001,
+      "loss": 6.7127,
+      "loss/crossentropy": 2.476449966430664,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4194856435060501,
+      "step": 472
+    },
+    {
+      "epoch": 0.00948,
+      "grad_norm": 4.40625,
+      "grad_norm_var": 0.16887613932291667,
+      "learning_rate": 0.0001,
+      "loss": 6.494,
+      "loss/crossentropy": 2.6383973360061646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.44083887338638306,
+      "step": 474
+    },
+    {
+      "epoch": 0.00952,
+      "grad_norm": 4.375,
+      "grad_norm_var": 0.18435872395833333,
+      "learning_rate": 0.0001,
+      "loss": 6.3184,
+      "loss/crossentropy": 2.3149259090423584,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3863615244626999,
+      "step": 476
+    },
+    {
+      "epoch": 0.00956,
+      "grad_norm": 5.0625,
+      "grad_norm_var": 0.18388264973958332,
+      "learning_rate": 0.0001,
+      "loss": 6.6217,
+      "loss/crossentropy": 2.3096635341644287,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40995509922504425,
+      "step": 478
+    },
+    {
+      "epoch": 0.0096,
+      "grad_norm": 4.34375,
+      "grad_norm_var": 0.10846354166666666,
+      "learning_rate": 0.0001,
+      "loss": 6.4849,
+      "loss/crossentropy": 2.6495853662490845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.412080317735672,
+      "step": 480
+    },
+    {
+      "epoch": 0.00964,
+      "grad_norm": 5.21875,
+      "grad_norm_var": 0.13948160807291668,
+      "learning_rate": 0.0001,
+      "loss": 6.9293,
+      "loss/crossentropy": 2.445754885673523,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.45622071623802185,
+      "step": 482
+    },
+    {
+      "epoch": 0.00968,
+      "grad_norm": 4.40625,
+      "grad_norm_var": 0.13661702473958334,
+      "learning_rate": 0.0001,
+      "loss": 6.4119,
+      "loss/crossentropy": 2.418110966682434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40424875915050507,
+      "step": 484
+    },
+    {
+      "epoch": 0.00972,
+      "grad_norm": 5.0625,
+      "grad_norm_var": 0.15959879557291667,
+      "learning_rate": 0.0001,
+      "loss": 6.6356,
+      "loss/crossentropy": 1.9564435482025146,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.45834848284721375,
+      "step": 486
+    },
+    {
+      "epoch": 0.00976,
+      "grad_norm": 5.375,
+      "grad_norm_var": 0.16946207682291667,
+      "learning_rate": 0.0001,
+      "loss": 6.7056,
+      "loss/crossentropy": 2.3772581815719604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4418800473213196,
+      "step": 488
+    },
+    {
+      "epoch": 0.0098,
+      "grad_norm": 4.0625,
+      "grad_norm_var": 0.208056640625,
+      "learning_rate": 0.0001,
+      "loss": 6.3239,
+      "loss/crossentropy": 1.9526153802871704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3661540001630783,
+      "step": 490
+    },
+    {
+      "epoch": 0.00984,
+      "grad_norm": 4.875,
+      "grad_norm_var": 0.19999593098958332,
+      "learning_rate": 0.0001,
+      "loss": 6.7642,
+      "loss/crossentropy": 2.40561842918396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.44370119273662567,
+      "step": 492
+    },
+    {
+      "epoch": 0.00988,
+      "grad_norm": 4.53125,
+      "grad_norm_var": 0.1943359375,
+      "learning_rate": 0.0001,
+      "loss": 6.6475,
+      "loss/crossentropy": 2.4316108226776123,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4437306672334671,
+      "step": 494
+    },
+    {
+      "epoch": 0.00992,
+      "grad_norm": 4.53125,
+      "grad_norm_var": 0.25276285807291665,
+      "learning_rate": 0.0001,
+      "loss": 6.4095,
+      "loss/crossentropy": 2.2919591665267944,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4170967787504196,
+      "step": 496
+    },
+    {
+      "epoch": 0.00996,
+      "grad_norm": 5.03125,
+      "grad_norm_var": 0.21116129557291666,
+      "learning_rate": 0.0001,
+      "loss": 6.6291,
+      "loss/crossentropy": 2.571357250213623,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4073399156332016,
+      "step": 498
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.375,
+      "grad_norm_var": 0.21330973307291667,
+      "learning_rate": 0.0001,
+      "loss": 6.2903,
+      "loss/crossentropy": 2.389290928840637,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.41799379885196686,
+      "step": 500
+    },
+    {
+      "epoch": 0.01004,
+      "grad_norm": 4.90625,
+      "grad_norm_var": 0.20245768229166666,
+      "learning_rate": 0.0001,
+      "loss": 6.4319,
+      "loss/crossentropy": 2.0904359221458435,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3914492577314377,
+      "step": 502
+    },
+    {
+      "epoch": 0.01008,
+      "grad_norm": 3.625,
+      "grad_norm_var": 0.218212890625,
+      "learning_rate": 0.0001,
+      "loss": 6.5581,
+      "loss/crossentropy": 2.5435129404067993,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4660491645336151,
+      "step": 504
+    },
+    {
+      "epoch": 0.01012,
+      "grad_norm": 4.75,
+      "grad_norm_var": 0.21458333333333332,
+      "learning_rate": 0.0001,
+      "loss": 6.3665,
+      "loss/crossentropy": 2.039083242416382,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.38188865780830383,
+      "step": 506
+    },
+    {
+      "epoch": 0.01016,
+      "grad_norm": 5.21875,
+      "grad_norm_var": 0.361181640625,
+      "learning_rate": 0.0001,
+      "loss": 6.4862,
+      "loss/crossentropy": 2.056805729866028,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4257620573043823,
+      "step": 508
+    },
+    {
+      "epoch": 0.0102,
+      "grad_norm": 4.03125,
+      "grad_norm_var": 0.38592122395833334,
+      "learning_rate": 0.0001,
+      "loss": 6.4822,
+      "loss/crossentropy": 2.6178410053253174,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.43898941576480865,
+      "step": 510
+    },
+    {
+      "epoch": 0.01024,
+      "grad_norm": 4.59375,
+      "grad_norm_var": 0.3405558268229167,
+      "learning_rate": 0.0001,
+      "loss": 6.4638,
+      "loss/crossentropy": 2.232435703277588,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37769296765327454,
+      "step": 512
+    },
+    {
+      "epoch": 0.01028,
+      "grad_norm": 4.40625,
+      "grad_norm_var": 0.33033447265625,
+      "learning_rate": 0.0001,
+      "loss": 6.378,
+      "loss/crossentropy": 1.8679735660552979,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3805970698595047,
+      "step": 514
+    },
+    {
+      "epoch": 0.01032,
+      "grad_norm": 4.3125,
+      "grad_norm_var": 0.33175455729166664,
+      "learning_rate": 0.0001,
+      "loss": 6.7453,
+      "loss/crossentropy": 2.6537472009658813,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4450981914997101,
+      "step": 516
+    },
+    {
+      "epoch": 0.01036,
+      "grad_norm": 4.03125,
+      "grad_norm_var": 0.34915262858072915,
+      "learning_rate": 0.0001,
+      "loss": 6.2745,
+      "loss/crossentropy": 2.5199841260910034,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4748214781284332,
+      "step": 518
+    },
+    {
+      "epoch": 0.0104,
+      "grad_norm": 5.5,
+      "grad_norm_var": 0.3482004801432292,
+      "learning_rate": 0.0001,
+      "loss": 6.6212,
+      "loss/crossentropy": 2.6603333950042725,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4144483357667923,
+      "step": 520
+    },
+    {
+      "epoch": 0.01044,
+      "grad_norm": 4.28125,
+      "grad_norm_var": 0.3241119384765625,
+      "learning_rate": 0.0001,
+      "loss": 6.4073,
+      "loss/crossentropy": 2.284039616584778,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.42801250517368317,
+      "step": 522
+    },
+    {
+      "epoch": 0.01048,
+      "grad_norm": 3.78125,
+      "grad_norm_var": 0.1894683837890625,
+      "learning_rate": 0.0001,
+      "loss": 6.2153,
+      "loss/crossentropy": 2.473629951477051,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4107852131128311,
+      "step": 524
+    },
+    {
+      "epoch": 0.01052,
+      "grad_norm": 4.5,
+      "grad_norm_var": 0.1827789306640625,
+      "learning_rate": 0.0001,
+      "loss": 6.6746,
+      "loss/crossentropy": 2.1443774700164795,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3606857359409332,
+      "step": 526
+    },
+    {
+      "epoch": 0.01056,
+      "grad_norm": 3.96875,
+      "grad_norm_var": 0.18889058430989583,
+      "learning_rate": 0.0001,
+      "loss": 5.8493,
+      "loss/crossentropy": 1.8425135016441345,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37770508229732513,
+      "step": 528
+    },
+    {
+      "epoch": 0.0106,
+      "grad_norm": 3.859375,
+      "grad_norm_var": 0.179052734375,
+      "learning_rate": 0.0001,
+      "loss": 6.3319,
+      "loss/crossentropy": 2.3705164194107056,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4354119151830673,
+      "step": 530
+    },
+    {
+      "epoch": 0.01064,
+      "grad_norm": 4.53125,
+      "grad_norm_var": 0.16344401041666667,
+      "learning_rate": 0.0001,
+      "loss": 6.4823,
+      "loss/crossentropy": 2.1141316294670105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34213581681251526,
+      "step": 532
+    },
+    {
+      "epoch": 0.01068,
+      "grad_norm": 4.3125,
+      "grad_norm_var": 0.15579325358072918,
+      "learning_rate": 0.0001,
+      "loss": 6.0454,
+      "loss/crossentropy": 2.14878511428833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3756616413593292,
+      "step": 534
+    },
+    {
+      "epoch": 0.01072,
+      "grad_norm": 4.34375,
+      "grad_norm_var": 0.052155558268229166,
+      "learning_rate": 0.0001,
+      "loss": 6.4363,
+      "loss/crossentropy": 2.2513046264648438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.416190966963768,
+      "step": 536
+    },
+    {
+      "epoch": 0.01076,
+      "grad_norm": 3.921875,
+      "grad_norm_var": 0.049169921875,
+      "learning_rate": 0.0001,
+      "loss": 6.2674,
+      "loss/crossentropy": 2.1337047815322876,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3839789927005768,
+      "step": 538
+    },
+    {
+      "epoch": 0.0108,
+      "grad_norm": 5.8125,
+      "grad_norm_var": 0.20488993326822916,
+      "learning_rate": 0.0001,
+      "loss": 6.0559,
+      "loss/crossentropy": 2.2019962072372437,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3806414008140564,
+      "step": 540
+    },
+    {
+      "epoch": 0.01084,
+      "grad_norm": 4.125,
+      "grad_norm_var": 0.2133697509765625,
+      "learning_rate": 0.0001,
+      "loss": 5.8434,
+      "loss/crossentropy": 2.113224983215332,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3797851800918579,
+      "step": 542
+    },
+    {
+      "epoch": 0.01088,
+      "grad_norm": 4.0,
+      "grad_norm_var": 0.21507059733072917,
+      "learning_rate": 0.0001,
+      "loss": 6.5131,
+      "loss/crossentropy": 2.461037516593933,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.41689516603946686,
+      "step": 544
+    },
+    {
+      "epoch": 0.01092,
+      "grad_norm": 4.625,
+      "grad_norm_var": 0.24849853515625,
+      "learning_rate": 0.0001,
+      "loss": 6.4191,
+      "loss/crossentropy": 2.277098298072815,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4108494818210602,
+      "step": 546
+    },
+    {
+      "epoch": 0.01096,
+      "grad_norm": 4.34375,
+      "grad_norm_var": 0.48818359375,
+      "learning_rate": 0.0001,
+      "loss": 6.3436,
+      "loss/crossentropy": 2.007936477661133,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36796192824840546,
+      "step": 548
+    },
+    {
+      "epoch": 0.011,
+      "grad_norm": 4.625,
+      "grad_norm_var": 0.5631795247395833,
+      "learning_rate": 0.0001,
+      "loss": 6.5793,
+      "loss/crossentropy": 2.197320520877838,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.38722094893455505,
+      "step": 550
+    },
+    {
+      "epoch": 0.01104,
+      "grad_norm": 4.6875,
+      "grad_norm_var": 0.5419230143229167,
+      "learning_rate": 0.0001,
+      "loss": 6.3185,
+      "loss/crossentropy": 2.225432515144348,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3928917348384857,
+      "step": 552
+    },
+    {
+      "epoch": 0.01108,
+      "grad_norm": 4.28125,
+      "grad_norm_var": 0.5455067952473959,
+      "learning_rate": 0.0001,
+      "loss": 5.9686,
+      "loss/crossentropy": 2.5253326892852783,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.41547106206417084,
+      "step": 554
+    },
+    {
+      "epoch": 0.01112,
+      "grad_norm": 4.15625,
+      "grad_norm_var": 0.44996337890625,
+      "learning_rate": 0.0001,
+      "loss": 6.2158,
+      "loss/crossentropy": 2.488635540008545,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4523312896490097,
+      "step": 556
+    },
+    {
+      "epoch": 0.01116,
+      "grad_norm": 4.03125,
+      "grad_norm_var": 0.46119384765625,
+      "learning_rate": 0.0001,
+      "loss": 6.3632,
+      "loss/crossentropy": 2.395568609237671,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40695953369140625,
+      "step": 558
+    },
+    {
+      "epoch": 0.0112,
+      "grad_norm": 4.28125,
+      "grad_norm_var": 0.4615234375,
+      "learning_rate": 0.0001,
+      "loss": 5.998,
+      "loss/crossentropy": 2.383823275566101,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4482497274875641,
+      "step": 560
+    },
+    {
+      "epoch": 0.01124,
+      "grad_norm": 4.46875,
+      "grad_norm_var": 0.46280008951822915,
+      "learning_rate": 0.0001,
+      "loss": 5.968,
+      "loss/crossentropy": 2.0261693000793457,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36933301389217377,
+      "step": 562
+    },
+    {
+      "epoch": 0.01128,
+      "grad_norm": 4.65625,
+      "grad_norm_var": 0.2221588134765625,
+      "learning_rate": 0.0001,
+      "loss": 6.1467,
+      "loss/crossentropy": 2.131769895553589,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.341948002576828,
+      "step": 564
+    },
+    {
+      "epoch": 0.01132,
+      "grad_norm": 4.8125,
+      "grad_norm_var": 0.0982574462890625,
+      "learning_rate": 0.0001,
+      "loss": 6.5288,
+      "loss/crossentropy": 2.3899158239364624,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.417646586894989,
+      "step": 566
+    },
+    {
+      "epoch": 0.01136,
+      "grad_norm": 4.21875,
+      "grad_norm_var": 0.07683817545572917,
+      "learning_rate": 0.0001,
+      "loss": 6.6198,
+      "loss/crossentropy": 2.3139528036117554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.39010919630527496,
+      "step": 568
+    },
+    {
+      "epoch": 0.0114,
+      "grad_norm": 4.3125,
+      "grad_norm_var": 0.06782124837239584,
+      "learning_rate": 0.0001,
+      "loss": 6.1209,
+      "loss/crossentropy": 1.966201364994049,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3554569333791733,
+      "step": 570
+    },
+    {
+      "epoch": 0.01144,
+      "grad_norm": 4.21875,
+      "grad_norm_var": 0.07773335774739583,
+      "learning_rate": 0.0001,
+      "loss": 6.1746,
+      "loss/crossentropy": 2.2325466871261597,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4008040726184845,
+      "step": 572
+    },
+    {
+      "epoch": 0.01148,
+      "grad_norm": 4.96875,
+      "grad_norm_var": 0.09848531087239583,
+      "learning_rate": 0.0001,
+      "loss": 6.121,
+      "loss/crossentropy": 1.7670194506645203,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3362845778465271,
+      "step": 574
+    },
+    {
+      "epoch": 0.01152,
+      "grad_norm": 4.125,
+      "grad_norm_var": 0.10274149576822916,
+      "learning_rate": 0.0001,
+      "loss": 6.3956,
+      "loss/crossentropy": 2.332284092903137,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4190225303173065,
+      "step": 576
+    },
+    {
+      "epoch": 0.01156,
+      "grad_norm": 4.71875,
+      "grad_norm_var": 0.10549723307291667,
+      "learning_rate": 0.0001,
+      "loss": 6.4367,
+      "loss/crossentropy": 2.265386700630188,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4109695851802826,
+      "step": 578
+    },
+    {
+      "epoch": 0.0116,
+      "grad_norm": 5.15625,
+      "grad_norm_var": 0.13811442057291667,
+      "learning_rate": 0.0001,
+      "loss": 6.4475,
+      "loss/crossentropy": 2.265889286994934,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.42450472712516785,
+      "step": 580
+    },
+    {
+      "epoch": 0.01164,
+      "grad_norm": 4.5625,
+      "grad_norm_var": 0.163134765625,
+      "learning_rate": 0.0001,
+      "loss": 6.1483,
+      "loss/crossentropy": 1.867847204208374,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3627365529537201,
+      "step": 582
+    },
+    {
+      "epoch": 0.01168,
+      "grad_norm": 3.796875,
+      "grad_norm_var": 0.23681538899739582,
+      "learning_rate": 0.0001,
+      "loss": 6.5909,
+      "loss/crossentropy": 2.5827555656433105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5194894820451736,
+      "step": 584
+    },
+    {
+      "epoch": 0.01172,
+      "grad_norm": 4.5625,
+      "grad_norm_var": 0.23188374837239584,
+      "learning_rate": 0.0001,
+      "loss": 6.6319,
+      "loss/crossentropy": 2.26031893491745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.412450835108757,
+      "step": 586
+    },
+    {
+      "epoch": 0.01176,
+      "grad_norm": 5.53125,
+      "grad_norm_var": 0.2718739827473958,
+      "learning_rate": 0.0001,
+      "loss": 6.5782,
+      "loss/crossentropy": 2.1885104179382324,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.41636165976524353,
+      "step": 588
+    },
+    {
+      "epoch": 0.0118,
+      "grad_norm": 4.53125,
+      "grad_norm_var": 0.26240132649739584,
+      "learning_rate": 0.0001,
+      "loss": 6.5247,
+      "loss/crossentropy": 2.682767391204834,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4279082715511322,
+      "step": 590
+    },
+    {
+      "epoch": 0.01184,
+      "grad_norm": 4.3125,
+      "grad_norm_var": 0.24807027180989583,
+      "learning_rate": 0.0001,
+      "loss": 6.5862,
+      "loss/crossentropy": 2.185304641723633,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4266812950372696,
+      "step": 592
+    },
+    {
+      "epoch": 0.01188,
+      "grad_norm": 4.15625,
+      "grad_norm_var": 0.2589752197265625,
+      "learning_rate": 0.0001,
+      "loss": 6.2494,
+      "loss/crossentropy": 2.383716344833374,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.38570962846279144,
+      "step": 594
+    },
+    {
+      "epoch": 0.01192,
+      "grad_norm": 4.3125,
+      "grad_norm_var": 0.2337066650390625,
+      "learning_rate": 0.0001,
+      "loss": 6.3066,
+      "loss/crossentropy": 2.2963398694992065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.43451687693595886,
+      "step": 596
+    },
+    {
+      "epoch": 0.01196,
+      "grad_norm": 4.96875,
+      "grad_norm_var": 0.2128570556640625,
+      "learning_rate": 0.0001,
+      "loss": 6.3368,
+      "loss/crossentropy": 2.2728757858276367,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4280686676502228,
+      "step": 598
+    },
+    {
+      "epoch": 0.012,
+      "grad_norm": 4.34375,
+      "grad_norm_var": 0.141650390625,
+      "learning_rate": 0.0001,
+      "loss": 6.1147,
+      "loss/crossentropy": 2.3486615419387817,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37224848568439484,
+      "step": 600
+    },
+    {
+      "epoch": 0.01204,
+      "grad_norm": 3.765625,
+      "grad_norm_var": 0.17512613932291668,
+      "learning_rate": 0.0001,
+      "loss": 6.4017,
+      "loss/crossentropy": 2.3265292644500732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4079990088939667,
+      "step": 602
+    },
+    {
+      "epoch": 0.01208,
+      "grad_norm": 3.84375,
+      "grad_norm_var": 0.08847554524739583,
+      "learning_rate": 0.0001,
+      "loss": 6.0861,
+      "loss/crossentropy": 2.45253849029541,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.42630523443222046,
+      "step": 604
+    },
+    {
+      "epoch": 0.01212,
+      "grad_norm": 4.03125,
+      "grad_norm_var": 0.08964742024739583,
+      "learning_rate": 0.0001,
+      "loss": 6.4101,
+      "loss/crossentropy": 2.4417446851730347,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40342913568019867,
+      "step": 606
+    },
+    {
+      "epoch": 0.01216,
+      "grad_norm": 4.25,
+      "grad_norm_var": 0.08886617024739583,
+      "learning_rate": 0.0001,
+      "loss": 6.2513,
+      "loss/crossentropy": 2.1483529210090637,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3846609443426132,
+      "step": 608
+    },
+    {
+      "epoch": 0.0122,
+      "grad_norm": 4.1875,
+      "grad_norm_var": 0.09990132649739583,
+      "learning_rate": 0.0001,
+      "loss": 6.5479,
+      "loss/crossentropy": 2.481536865234375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.440000057220459,
+      "step": 610
+    },
+    {
+      "epoch": 0.01224,
+      "grad_norm": 4.15625,
+      "grad_norm_var": 0.09464518229166667,
+      "learning_rate": 0.0001,
+      "loss": 6.4986,
+      "loss/crossentropy": 2.4472655057907104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.41242682933807373,
+      "step": 612
+    },
+    {
+      "epoch": 0.01228,
+      "grad_norm": 4.09375,
+      "grad_norm_var": 0.07828369140625,
+      "learning_rate": 0.0001,
+      "loss": 6.4348,
+      "loss/crossentropy": 2.3511135578155518,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.38262398540973663,
+      "step": 614
+    },
+    {
+      "epoch": 0.01232,
+      "grad_norm": 3.71875,
+      "grad_norm_var": 0.085791015625,
+      "learning_rate": 0.0001,
+      "loss": 6.4821,
+      "loss/crossentropy": 2.5090683698654175,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.44584622979164124,
+      "step": 616
+    },
+    {
+      "epoch": 0.01236,
+      "grad_norm": 4.75,
+      "grad_norm_var": 0.1117340087890625,
+      "learning_rate": 0.0001,
+      "loss": 6.0395,
+      "loss/crossentropy": 2.166012942790985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40271493792533875,
+      "step": 618
+    },
+    {
+      "epoch": 0.0124,
+      "grad_norm": 5.09375,
+      "grad_norm_var": 0.15511067708333334,
+      "learning_rate": 0.0001,
+      "loss": 6.4644,
+      "loss/crossentropy": 2.583309531211853,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.43380285799503326,
+      "step": 620
+    },
+    {
+      "epoch": 0.01244,
+      "grad_norm": 3.984375,
+      "grad_norm_var": 0.19877827962239583,
+      "learning_rate": 0.0001,
+      "loss": 6.3289,
+      "loss/crossentropy": 2.125720262527466,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40389589965343475,
+      "step": 622
+    },
+    {
+      "epoch": 0.01248,
+      "grad_norm": 6.34375,
+      "grad_norm_var": 0.4984283447265625,
+      "learning_rate": 0.0001,
+      "loss": 5.9651,
+      "loss/crossentropy": 1.7034094333648682,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31466029584407806,
+      "step": 624
+    },
+    {
+      "epoch": 0.01252,
+      "grad_norm": 4.40625,
+      "grad_norm_var": 0.4901041666666667,
+      "learning_rate": 0.0001,
+      "loss": 5.9451,
+      "loss/crossentropy": 2.163281202316284,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.38622182607650757,
+      "step": 626
+    },
+    {
+      "epoch": 0.01256,
+      "grad_norm": 3.875,
+      "grad_norm_var": 0.4982818603515625,
+      "learning_rate": 0.0001,
+      "loss": 5.8931,
+      "loss/crossentropy": 1.7754453420639038,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33256995677948,
+      "step": 628
+    },
+    {
+      "epoch": 0.0126,
+      "grad_norm": 4.09375,
+      "grad_norm_var": 0.4894683837890625,
+      "learning_rate": 0.0001,
+      "loss": 6.8606,
+      "loss/crossentropy": 2.273309350013733,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5319447964429855,
+      "step": 630
+    },
+    {
+      "epoch": 0.01264,
+      "grad_norm": 3.828125,
+      "grad_norm_var": 0.480126953125,
+      "learning_rate": 0.0001,
+      "loss": 6.2103,
+      "loss/crossentropy": 2.397401988506317,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4255771040916443,
+      "step": 632
+    },
+    {
+      "epoch": 0.01268,
+      "grad_norm": 5.28125,
+      "grad_norm_var": 0.521875,
+      "learning_rate": 0.0001,
+      "loss": 6.4179,
+      "loss/crossentropy": 2.3898611068725586,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.46314406394958496,
+      "step": 634
+    },
+    {
+      "epoch": 0.01272,
+      "grad_norm": 7.40625,
+      "grad_norm_var": 1.0609212239583334,
+      "learning_rate": 0.0001,
+      "loss": 6.5634,
+      "loss/crossentropy": 2.3740471601486206,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.48123428225517273,
+      "step": 636
+    },
+    {
+      "epoch": 0.01276,
+      "grad_norm": 3.90625,
+      "grad_norm_var": 1.0519846598307292,
+      "learning_rate": 0.0001,
+      "loss": 6.1136,
+      "loss/crossentropy": 2.236217498779297,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.39564159512519836,
+      "step": 638
+    },
+    {
+      "epoch": 0.0128,
+      "grad_norm": 4.21875,
+      "grad_norm_var": 0.7972157796223959,
+      "learning_rate": 0.0001,
+      "loss": 6.3134,
+      "loss/crossentropy": 2.4049174785614014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.41312308609485626,
+      "step": 640
+    },
+    {
+      "epoch": 0.01284,
+      "grad_norm": 4.0,
+      "grad_norm_var": 0.80142822265625,
+      "learning_rate": 0.0001,
+      "loss": 6.2726,
+      "loss/crossentropy": 2.173800766468048,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3983110189437866,
+      "step": 642
+    },
+    {
+      "epoch": 0.01288,
+      "grad_norm": 4.65625,
+      "grad_norm_var": 0.80084228515625,
+      "learning_rate": 0.0001,
+      "loss": 6.2779,
+      "loss/crossentropy": 2.2124537229537964,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.38358184695243835,
+      "step": 644
+    },
+    {
+      "epoch": 0.01292,
+      "grad_norm": 3.90625,
+      "grad_norm_var": 0.8355377197265625,
+      "learning_rate": 0.0001,
+      "loss": 6.1209,
+      "loss/crossentropy": 2.4939264059066772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3805152475833893,
+      "step": 646
+    },
+    {
+      "epoch": 0.01296,
+      "grad_norm": 4.25,
+      "grad_norm_var": 0.8250935872395834,
+      "learning_rate": 0.0001,
+      "loss": 6.0714,
+      "loss/crossentropy": 2.4526472091674805,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3958089202642441,
+      "step": 648
+    },
+    {
+      "epoch": 0.013,
+      "grad_norm": 4.25,
+      "grad_norm_var": 0.7562978108723958,
+      "learning_rate": 0.0001,
+      "loss": 6.3648,
+      "loss/crossentropy": 2.5171029567718506,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.446771502494812,
+      "step": 650
+    },
+    {
+      "epoch": 0.01304,
+      "grad_norm": 4.125,
+      "grad_norm_var": 0.1185943603515625,
+      "learning_rate": 0.0001,
+      "loss": 6.1656,
+      "loss/crossentropy": 2.270598888397217,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.38787929713726044,
+      "step": 652
+    },
+    {
+      "epoch": 0.01308,
+      "grad_norm": 4.15625,
+      "grad_norm_var": 0.11404520670572917,
+      "learning_rate": 0.0001,
+      "loss": 5.7739,
+      "loss/crossentropy": 1.8847617506980896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3258303850889206,
+      "step": 654
+    },
+    {
+      "epoch": 0.01312,
+      "grad_norm": 3.9375,
+      "grad_norm_var": 0.11286519368489584,
+      "learning_rate": 0.0001,
+      "loss": 6.042,
+      "loss/crossentropy": 2.2471452951431274,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.38874460756778717,
+      "step": 656
+    },
+    {
+      "epoch": 0.01316,
+      "grad_norm": 4.125,
+      "grad_norm_var": 0.10369364420572917,
+      "learning_rate": 0.0001,
+      "loss": 6.4383,
+      "loss/crossentropy": 2.3776252269744873,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40553848445415497,
+      "step": 658
+    },
+    {
+      "epoch": 0.0132,
+      "grad_norm": 3.84375,
+      "grad_norm_var": 0.06461588541666667,
+      "learning_rate": 0.0001,
+      "loss": 5.5389,
+      "loss/crossentropy": 2.291012167930603,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3630271404981613,
+      "step": 660
+    },
+    {
+      "epoch": 0.01324,
+      "grad_norm": 3.90625,
+      "grad_norm_var": 0.06116536458333333,
+      "learning_rate": 0.0001,
+      "loss": 6.298,
+      "loss/crossentropy": 2.2029112577438354,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37700483202934265,
+      "step": 662
+    },
+    {
+      "epoch": 0.01328,
+      "grad_norm": 3.984375,
+      "grad_norm_var": 0.0598541259765625,
+      "learning_rate": 0.0001,
+      "loss": 6.4093,
+      "loss/crossentropy": 2.571411967277527,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5071892440319061,
+      "step": 664
+    },
+    {
+      "epoch": 0.01332,
+      "grad_norm": 3.484375,
+      "grad_norm_var": 0.045947265625,
+      "learning_rate": 0.0001,
+      "loss": 5.6839,
+      "loss/crossentropy": 2.148792862892151,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3591457009315491,
+      "step": 666
+    },
+    {
+      "epoch": 0.01336,
+      "grad_norm": 4.09375,
+      "grad_norm_var": 0.04755452473958333,
+      "learning_rate": 0.0001,
+      "loss": 6.4444,
+      "loss/crossentropy": 2.5091140270233154,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3882133811712265,
+      "step": 668
+    },
+    {
+      "epoch": 0.0134,
+      "grad_norm": 4.09375,
+      "grad_norm_var": 0.049332682291666666,
+      "learning_rate": 0.0001,
+      "loss": 6.15,
+      "loss/crossentropy": 2.4669524431228638,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3941466957330704,
+      "step": 670
+    },
+    {
+      "epoch": 0.01344,
+      "grad_norm": 3.9375,
+      "grad_norm_var": 0.06620686848958333,
+      "learning_rate": 0.0001,
+      "loss": 6.5358,
+      "loss/crossentropy": 2.4111422300338745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3973373472690582,
+      "step": 672
+    },
+    {
+      "epoch": 0.01348,
+      "grad_norm": 4.1875,
+      "grad_norm_var": 0.06495768229166667,
+      "learning_rate": 0.0001,
+      "loss": 5.765,
+      "loss/crossentropy": 2.1109864115715027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36189010739326477,
+      "step": 674
+    },
+    {
+      "epoch": 0.01352,
+      "grad_norm": 3.578125,
+      "grad_norm_var": 0.0744140625,
+      "learning_rate": 0.0001,
+      "loss": 6.0289,
+      "loss/crossentropy": 2.069494664669037,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3562029302120209,
+      "step": 676
+    },
+    {
+      "epoch": 0.01356,
+      "grad_norm": 4.0625,
+      "grad_norm_var": 0.07224934895833333,
+      "learning_rate": 0.0001,
+      "loss": 6.4526,
+      "loss/crossentropy": 2.1924527883529663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.38948506116867065,
+      "step": 678
+    },
+    {
+      "epoch": 0.0136,
+      "grad_norm": 3.6875,
+      "grad_norm_var": 0.082421875,
+      "learning_rate": 0.0001,
+      "loss": 5.7311,
+      "loss/crossentropy": 2.1603400707244873,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40414859354496,
+      "step": 680
+    },
+    {
+      "epoch": 0.01364,
+      "grad_norm": 3.71875,
+      "grad_norm_var": 0.07177632649739583,
+      "learning_rate": 0.0001,
+      "loss": 6.2459,
+      "loss/crossentropy": 2.515262722969055,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4363710880279541,
+      "step": 682
+    },
+    {
+      "epoch": 0.01368,
+      "grad_norm": 4.09375,
+      "grad_norm_var": 0.07869364420572916,
+      "learning_rate": 0.0001,
+      "loss": 6.1174,
+      "loss/crossentropy": 2.3615161180496216,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35220713913440704,
+      "step": 684
+    },
+    {
+      "epoch": 0.01372,
+      "grad_norm": 6.03125,
+      "grad_norm_var": 0.3293690999348958,
+      "learning_rate": 0.0001,
+      "loss": 6.1941,
+      "loss/crossentropy": 1.920493245124817,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35293935239315033,
+      "step": 686
+    },
+    {
+      "epoch": 0.01376,
+      "grad_norm": 4.03125,
+      "grad_norm_var": 0.31579488118489585,
+      "learning_rate": 0.0001,
+      "loss": 6.4064,
+      "loss/crossentropy": 2.493665337562561,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4243907481431961,
+      "step": 688
+    },
+    {
+      "epoch": 0.0138,
+      "grad_norm": 4.03125,
+      "grad_norm_var": 0.3451324462890625,
+      "learning_rate": 0.0001,
+      "loss": 6.1519,
+      "loss/crossentropy": 2.1182271242141724,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.39624081552028656,
+      "step": 690
+    },
+    {
+      "epoch": 0.01384,
+      "grad_norm": 4.09375,
+      "grad_norm_var": 0.32034098307291664,
+      "learning_rate": 0.0001,
+      "loss": 6.2457,
+      "loss/crossentropy": 2.146227180957794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3491668850183487,
+      "step": 692
+    },
+    {
+      "epoch": 0.01388,
+      "grad_norm": 3.734375,
+      "grad_norm_var": 0.3376261393229167,
+      "learning_rate": 0.0001,
+      "loss": 5.7331,
+      "loss/crossentropy": 1.8458876609802246,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34608474373817444,
+      "step": 694
+    },
+    {
+      "epoch": 0.01392,
+      "grad_norm": 4.03125,
+      "grad_norm_var": 0.3294911702473958,
+      "learning_rate": 0.0001,
+      "loss": 6.3661,
+      "loss/crossentropy": 2.270371675491333,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33937984704971313,
+      "step": 696
+    },
+    {
+      "epoch": 0.01396,
+      "grad_norm": 3.875,
+      "grad_norm_var": 0.32066141764322914,
+      "learning_rate": 0.0001,
+      "loss": 5.8264,
+      "loss/crossentropy": 2.040702223777771,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3821101486682892,
+      "step": 698
+    },
+    {
+      "epoch": 0.014,
+      "grad_norm": 4.09375,
+      "grad_norm_var": 0.3197428385416667,
+      "learning_rate": 0.0001,
+      "loss": 6.1216,
+      "loss/crossentropy": 2.1711822152137756,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35383065044879913,
+      "step": 700
+    },
+    {
+      "epoch": 0.01404,
+      "grad_norm": 3.890625,
+      "grad_norm_var": 0.0597076416015625,
+      "learning_rate": 0.0001,
+      "loss": 5.8384,
+      "loss/crossentropy": 2.3292651176452637,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37580642104148865,
+      "step": 702
+    },
+    {
+      "epoch": 0.01408,
+      "grad_norm": 4.34375,
+      "grad_norm_var": 0.06575520833333333,
+      "learning_rate": 0.0001,
+      "loss": 6.0789,
+      "loss/crossentropy": 2.243735432624817,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3997037708759308,
+      "step": 704
+    },
+    {
+      "epoch": 0.01412,
+      "grad_norm": 3.921875,
+      "grad_norm_var": 0.0524566650390625,
+      "learning_rate": 0.0001,
+      "loss": 5.9987,
+      "loss/crossentropy": 1.9908145666122437,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33391132950782776,
+      "step": 706
+    },
+    {
+      "epoch": 0.01416,
+      "grad_norm": 4.375,
+      "grad_norm_var": 0.06243082682291667,
+      "learning_rate": 0.0001,
+      "loss": 5.536,
+      "loss/crossentropy": 2.280096471309662,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3899015784263611,
+      "step": 708
+    },
+    {
+      "epoch": 0.0142,
+      "grad_norm": 3.6875,
+      "grad_norm_var": 0.12727762858072916,
+      "learning_rate": 0.0001,
+      "loss": 5.9373,
+      "loss/crossentropy": 2.0714810490608215,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3797626197338104,
+      "step": 710
+    },
+    {
+      "epoch": 0.01424,
+      "grad_norm": 3.875,
+      "grad_norm_var": 0.14480692545572918,
+      "learning_rate": 0.0001,
+      "loss": 6.2466,
+      "loss/crossentropy": 2.250674605369568,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3948116898536682,
+      "step": 712
+    },
+    {
+      "epoch": 0.01428,
+      "grad_norm": 3.65625,
+      "grad_norm_var": 0.15719401041666667,
+      "learning_rate": 0.0001,
+      "loss": 5.8891,
+      "loss/crossentropy": 2.0895228385925293,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34471337497234344,
+      "step": 714
+    },
+    {
+      "epoch": 0.01432,
+      "grad_norm": 4.71875,
+      "grad_norm_var": 0.1826812744140625,
+      "learning_rate": 0.0001,
+      "loss": 6.3748,
+      "loss/crossentropy": 2.337170124053955,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4226381927728653,
+      "step": 716
+    },
+    {
+      "epoch": 0.01436,
+      "grad_norm": 4.03125,
+      "grad_norm_var": 0.174462890625,
+      "learning_rate": 0.0001,
+      "loss": 6.3996,
+      "loss/crossentropy": 2.337436556816101,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3738469183444977,
+      "step": 718
+    },
+    {
+      "epoch": 0.0144,
+      "grad_norm": 4.15625,
+      "grad_norm_var": 0.1669921875,
+      "learning_rate": 0.0001,
+      "loss": 6.0278,
+      "loss/crossentropy": 1.9506489634513855,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37431904673576355,
+      "step": 720
+    },
+    {
+      "epoch": 0.01444,
+      "grad_norm": 3.71875,
+      "grad_norm_var": 0.17283528645833332,
+      "learning_rate": 0.0001,
+      "loss": 5.8083,
+      "loss/crossentropy": 2.253044009208679,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36500048637390137,
+      "step": 722
+    },
+    {
+      "epoch": 0.01448,
+      "grad_norm": 5.15625,
+      "grad_norm_var": 0.23413798014322917,
+      "learning_rate": 0.0001,
+      "loss": 6.2307,
+      "loss/crossentropy": 1.9980219006538391,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34674490988254547,
+      "step": 724
+    },
+    {
+      "epoch": 0.01452,
+      "grad_norm": 3.609375,
+      "grad_norm_var": 0.1903472900390625,
+      "learning_rate": 0.0001,
+      "loss": 5.8283,
+      "loss/crossentropy": 1.894662618637085,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34921175241470337,
+      "step": 726
+    },
+    {
+      "epoch": 0.01456,
+      "grad_norm": 3.609375,
+      "grad_norm_var": 0.18310546875,
+      "learning_rate": 0.0001,
+      "loss": 6.1117,
+      "loss/crossentropy": 2.304685115814209,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36858032643795013,
+      "step": 728
+    },
+    {
+      "epoch": 0.0146,
+      "grad_norm": 3.6875,
+      "grad_norm_var": 0.17696940104166667,
+      "learning_rate": 0.0001,
+      "loss": 6.1041,
+      "loss/crossentropy": 1.9020891189575195,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3854057639837265,
+      "step": 730
+    },
+    {
+      "epoch": 0.01464,
+      "grad_norm": 3.4375,
+      "grad_norm_var": 0.15485026041666666,
+      "learning_rate": 0.0001,
+      "loss": 5.5327,
+      "loss/crossentropy": 1.707019329071045,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2961048036813736,
+      "step": 732
+    },
+    {
+      "epoch": 0.01468,
+      "grad_norm": 3.828125,
+      "grad_norm_var": 0.15344136555989582,
+      "learning_rate": 0.0001,
+      "loss": 6.0543,
+      "loss/crossentropy": 2.423463463783264,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3739102631807327,
+      "step": 734
+    },
+    {
+      "epoch": 0.01472,
+      "grad_norm": 4.0625,
+      "grad_norm_var": 0.15335286458333333,
+      "learning_rate": 0.0001,
+      "loss": 6.2855,
+      "loss/crossentropy": 2.0490055680274963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4027387350797653,
+      "step": 736
+    },
+    {
+      "epoch": 0.01476,
+      "grad_norm": 3.984375,
+      "grad_norm_var": 0.15038960774739582,
+      "learning_rate": 0.0001,
+      "loss": 6.1236,
+      "loss/crossentropy": 2.3712635040283203,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3602859079837799,
+      "step": 738
+    },
+    {
+      "epoch": 0.0148,
+      "grad_norm": 3.75,
+      "grad_norm_var": 0.058649698893229164,
+      "learning_rate": 0.0001,
+      "loss": 6.2938,
+      "loss/crossentropy": 2.306379556655884,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.38698625564575195,
+      "step": 740
+    },
+    {
+      "epoch": 0.01484,
+      "grad_norm": 3.984375,
+      "grad_norm_var": 0.055939737955729166,
+      "learning_rate": 0.0001,
+      "loss": 6.3983,
+      "loss/crossentropy": 2.6846178770065308,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3881431221961975,
+      "step": 742
+    },
+    {
+      "epoch": 0.01488,
+      "grad_norm": 3.9375,
+      "grad_norm_var": 0.058690388997395836,
+      "learning_rate": 0.0001,
+      "loss": 5.7984,
+      "loss/crossentropy": 2.0555977821350098,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36078818142414093,
+      "step": 744
+    },
+    {
+      "epoch": 0.01492,
+      "grad_norm": 4.09375,
+      "grad_norm_var": 0.059325154622395834,
+      "learning_rate": 0.0001,
+      "loss": 5.7834,
+      "loss/crossentropy": 2.0597304701805115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34516778588294983,
+      "step": 746
+    },
+    {
+      "epoch": 0.01496,
+      "grad_norm": 3.5625,
+      "grad_norm_var": 0.049479166666666664,
+      "learning_rate": 0.0001,
+      "loss": 5.8911,
+      "loss/crossentropy": 1.9607329964637756,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33818933367729187,
+      "step": 748
+    },
+    {
+      "epoch": 0.015,
+      "grad_norm": 3.546875,
+      "grad_norm_var": 0.057291666666666664,
+      "learning_rate": 0.0001,
+      "loss": 5.8484,
+      "loss/crossentropy": 1.7854246497154236,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3463115990161896,
+      "step": 750
+    },
+    {
+      "epoch": 0.01504,
+      "grad_norm": 3.703125,
+      "grad_norm_var": 0.0590240478515625,
+      "learning_rate": 0.0001,
+      "loss": 5.7788,
+      "loss/crossentropy": 1.845078468322754,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3714388310909271,
+      "step": 752
+    },
+    {
+      "epoch": 0.01508,
+      "grad_norm": 3.9375,
+      "grad_norm_var": 0.05771077473958333,
+      "learning_rate": 0.0001,
+      "loss": 6.1677,
+      "loss/crossentropy": 2.3951027393341064,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3723580837249756,
+      "step": 754
+    },
+    {
+      "epoch": 0.01512,
+      "grad_norm": 3.6875,
+      "grad_norm_var": 0.032486979166666666,
+      "learning_rate": 0.0001,
+      "loss": 6.219,
+      "loss/crossentropy": 2.500870108604431,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.42450079321861267,
+      "step": 756
+    },
+    {
+      "epoch": 0.01516,
+      "grad_norm": 3.75,
+      "grad_norm_var": 0.030060831705729166,
+      "learning_rate": 0.0001,
+      "loss": 5.886,
+      "loss/crossentropy": 2.1584274768829346,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3798908591270447,
+      "step": 758
+    },
+    {
+      "epoch": 0.0152,
+      "grad_norm": 3.5625,
+      "grad_norm_var": 0.028251139322916667,
+      "learning_rate": 0.0001,
+      "loss": 5.7937,
+      "loss/crossentropy": 2.3126983642578125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33131279051303864,
+      "step": 760
+    },
+    {
+      "epoch": 0.01524,
+      "grad_norm": 3.5,
+      "grad_norm_var": 0.020246378580729165,
+      "learning_rate": 0.0001,
+      "loss": 5.9182,
+      "loss/crossentropy": 2.349764347076416,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3910932093858719,
+      "step": 762
+    },
+    {
+      "epoch": 0.01528,
+      "grad_norm": 3.640625,
+      "grad_norm_var": 0.0224273681640625,
+      "learning_rate": 0.0001,
+      "loss": 6.0472,
+      "loss/crossentropy": 2.2232795357704163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3483322113752365,
+      "step": 764
+    },
+    {
+      "epoch": 0.01532,
+      "grad_norm": 3.65625,
+      "grad_norm_var": 0.0207427978515625,
+      "learning_rate": 0.0001,
+      "loss": 6.0312,
+      "loss/crossentropy": 2.2273412942886353,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.38545119762420654,
+      "step": 766
+    },
+    {
+      "epoch": 0.01536,
+      "grad_norm": 4.125,
+      "grad_norm_var": 0.03355712890625,
+      "learning_rate": 0.0001,
+      "loss": 6.0523,
+      "loss/crossentropy": 2.5879149436950684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.38191574811935425,
+      "step": 768
+    },
+    {
+      "epoch": 0.0154,
+      "grad_norm": 3.953125,
+      "grad_norm_var": 0.034032185872395836,
+      "learning_rate": 0.0001,
+      "loss": 6.027,
+      "loss/crossentropy": 2.3305420875549316,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3704134076833725,
+      "step": 770
+    },
+    {
+      "epoch": 0.01544,
+      "grad_norm": 3.59375,
+      "grad_norm_var": 0.03241780598958333,
+      "learning_rate": 0.0001,
+      "loss": 6.121,
+      "loss/crossentropy": 2.0433666706085205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34387587010860443,
+      "step": 772
+    },
+    {
+      "epoch": 0.01548,
+      "grad_norm": 3.609375,
+      "grad_norm_var": 0.03337300618489583,
+      "learning_rate": 0.0001,
+      "loss": 5.5837,
+      "loss/crossentropy": 2.1127337217330933,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3484109789133072,
+      "step": 774
+    },
+    {
+      "epoch": 0.01552,
+      "grad_norm": 3.859375,
+      "grad_norm_var": 0.029002888997395834,
+      "learning_rate": 0.0001,
+      "loss": 6.0028,
+      "loss/crossentropy": 2.1637459993362427,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3757011145353317,
+      "step": 776
+    },
+    {
+      "epoch": 0.01556,
+      "grad_norm": 3.734375,
+      "grad_norm_var": 0.024446614583333335,
+      "learning_rate": 0.0001,
+      "loss": 5.9904,
+      "loss/crossentropy": 2.4118471145629883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3797076344490051,
+      "step": 778
+    },
+    {
+      "epoch": 0.0156,
+      "grad_norm": 3.828125,
+      "grad_norm_var": 0.022191365559895832,
+      "learning_rate": 0.0001,
+      "loss": 6.2649,
+      "loss/crossentropy": 1.9410768151283264,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3254295587539673,
+      "step": 780
+    },
+    {
+      "epoch": 0.01564,
+      "grad_norm": 3.609375,
+      "grad_norm_var": 0.023558553059895834,
+      "learning_rate": 0.0001,
+      "loss": 5.9008,
+      "loss/crossentropy": 2.1669737100601196,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3583529591560364,
+      "step": 782
+    },
+    {
+      "epoch": 0.01568,
+      "grad_norm": 3.5625,
+      "grad_norm_var": 0.017606608072916665,
+      "learning_rate": 0.0001,
+      "loss": 5.9868,
+      "loss/crossentropy": 2.217113733291626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3503105044364929,
+      "step": 784
+    },
+    {
+      "epoch": 0.01572,
+      "grad_norm": 3.84375,
+      "grad_norm_var": 0.017154947916666666,
+      "learning_rate": 0.0001,
+      "loss": 6.0695,
+      "loss/crossentropy": 2.588438868522644,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3941201716661453,
+      "step": 786
+    },
+    {
+      "epoch": 0.01576,
+      "grad_norm": 3.546875,
+      "grad_norm_var": 0.018310546875,
+      "learning_rate": 0.0001,
+      "loss": 5.9436,
+      "loss/crossentropy": 2.3925808668136597,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3755808621644974,
+      "step": 788
+    },
+    {
+      "epoch": 0.0158,
+      "grad_norm": 3.625,
+      "grad_norm_var": 0.018684895833333333,
+      "learning_rate": 0.0001,
+      "loss": 5.7254,
+      "loss/crossentropy": 1.9568504691123962,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3865346759557724,
+      "step": 790
+    },
+    {
+      "epoch": 0.01584,
+      "grad_norm": 3.96875,
+      "grad_norm_var": 0.024104817708333334,
+      "learning_rate": 0.0001,
+      "loss": 6.0174,
+      "loss/crossentropy": 2.336462616920471,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36937348544597626,
+      "step": 792
+    },
+    {
+      "epoch": 0.01588,
+      "grad_norm": 3.421875,
+      "grad_norm_var": 0.03623046875,
+      "learning_rate": 0.0001,
+      "loss": 5.7742,
+      "loss/crossentropy": 2.1867082715034485,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35803911089897156,
+      "step": 794
+    },
+    {
+      "epoch": 0.01592,
+      "grad_norm": 3.765625,
+      "grad_norm_var": 0.03877665201822917,
+      "learning_rate": 0.0001,
+      "loss": 6.2825,
+      "loss/crossentropy": 2.070562243461609,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.42781224846839905,
+      "step": 796
+    },
+    {
+      "epoch": 0.01596,
+      "grad_norm": 4.09375,
+      "grad_norm_var": 0.044384765625,
+      "learning_rate": 0.0001,
+      "loss": 6.401,
+      "loss/crossentropy": 2.160820960998535,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34495553374290466,
+      "step": 798
+    },
+    {
+      "epoch": 0.016,
+      "grad_norm": 4.0,
+      "grad_norm_var": 0.045426432291666666,
+      "learning_rate": 0.0001,
+      "loss": 5.9628,
+      "loss/crossentropy": 2.3424230813980103,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4417698383331299,
+      "step": 800
+    },
+    {
+      "epoch": 0.01604,
+      "grad_norm": 4.59375,
+      "grad_norm_var": 0.08385009765625,
+      "learning_rate": 0.0001,
+      "loss": 6.1984,
+      "loss/crossentropy": 2.090175747871399,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35883304476737976,
+      "step": 802
+    },
+    {
+      "epoch": 0.01608,
+      "grad_norm": 3.90625,
+      "grad_norm_var": 0.0759429931640625,
+      "learning_rate": 0.0001,
+      "loss": 6.2044,
+      "loss/crossentropy": 2.460660457611084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36723683774471283,
+      "step": 804
+    },
+    {
+      "epoch": 0.01612,
+      "grad_norm": 3.78125,
+      "grad_norm_var": 0.0783203125,
+      "learning_rate": 0.0001,
+      "loss": 5.8788,
+      "loss/crossentropy": 2.2680885791778564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3925183415412903,
+      "step": 806
+    },
+    {
+      "epoch": 0.01616,
+      "grad_norm": 3.796875,
+      "grad_norm_var": 0.10422261555989583,
+      "learning_rate": 0.0001,
+      "loss": 6.1179,
+      "loss/crossentropy": 2.272566020488739,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3425859659910202,
+      "step": 808
+    },
+    {
+      "epoch": 0.0162,
+      "grad_norm": 3.546875,
+      "grad_norm_var": 0.10061442057291667,
+      "learning_rate": 0.0001,
+      "loss": 5.8933,
+      "loss/crossentropy": 2.2417107820510864,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3634066879749298,
+      "step": 810
+    },
+    {
+      "epoch": 0.01624,
+      "grad_norm": 4.09375,
+      "grad_norm_var": 0.10075581868489583,
+      "learning_rate": 0.0001,
+      "loss": 5.9907,
+      "loss/crossentropy": 2.2117987275123596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3669978231191635,
+      "step": 812
+    },
+    {
+      "epoch": 0.01628,
+      "grad_norm": 4.53125,
+      "grad_norm_var": 0.12078348795572917,
+      "learning_rate": 0.0001,
+      "loss": 6.1767,
+      "loss/crossentropy": 2.3471380472183228,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.39207911491394043,
+      "step": 814
+    },
+    {
+      "epoch": 0.01632,
+      "grad_norm": 4.1875,
+      "grad_norm_var": 0.12200113932291666,
+      "learning_rate": 0.0001,
+      "loss": 6.005,
+      "loss/crossentropy": 2.1516740322113037,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3394138962030411,
+      "step": 816
+    },
+    {
+      "epoch": 0.01636,
+      "grad_norm": 3.765625,
+      "grad_norm_var": 0.10528055826822917,
+      "learning_rate": 0.0001,
+      "loss": 6.0827,
+      "loss/crossentropy": 2.5085272789001465,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.39268872141838074,
+      "step": 818
+    },
+    {
+      "epoch": 0.0164,
+      "grad_norm": 3.515625,
+      "grad_norm_var": 0.1198883056640625,
+      "learning_rate": 0.0001,
+      "loss": 6.0363,
+      "loss/crossentropy": 2.3051916360855103,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3798936903476715,
+      "step": 820
+    },
+    {
+      "epoch": 0.01644,
+      "grad_norm": 3.5,
+      "grad_norm_var": 0.14094136555989584,
+      "learning_rate": 0.0001,
+      "loss": 5.4403,
+      "loss/crossentropy": 2.1685640811920166,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3666132390499115,
+      "step": 822
+    },
+    {
+      "epoch": 0.01648,
+      "grad_norm": 3.53125,
+      "grad_norm_var": 0.109521484375,
+      "learning_rate": 0.0001,
+      "loss": 5.6981,
+      "loss/crossentropy": 2.3374987840652466,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32505205273628235,
+      "step": 824
+    },
+    {
+      "epoch": 0.01652,
+      "grad_norm": 5.4375,
+      "grad_norm_var": 0.280859375,
+      "learning_rate": 0.0001,
+      "loss": 6.1428,
+      "loss/crossentropy": 2.6427528858184814,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4296632409095764,
+      "step": 826
+    },
+    {
+      "epoch": 0.01656,
+      "grad_norm": 4.0,
+      "grad_norm_var": 0.28609619140625,
+      "learning_rate": 0.0001,
+      "loss": 5.8304,
+      "loss/crossentropy": 2.1534847617149353,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37605202198028564,
+      "step": 828
+    },
+    {
+      "epoch": 0.0166,
+      "grad_norm": 4.28125,
+      "grad_norm_var": 0.2637278238932292,
+      "learning_rate": 0.0001,
+      "loss": 6.2115,
+      "loss/crossentropy": 1.9642478227615356,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.43740569055080414,
+      "step": 830
+    },
+    {
+      "epoch": 0.01664,
+      "grad_norm": 4.59375,
+      "grad_norm_var": 3.3863433837890624,
+      "learning_rate": 0.0001,
+      "loss": 6.5306,
+      "loss/crossentropy": 2.283148407936096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3901669532060623,
+      "step": 832
+    },
+    {
+      "epoch": 0.01668,
+      "grad_norm": 3.46875,
+      "grad_norm_var": 3.382255045572917,
+      "learning_rate": 0.0001,
+      "loss": 6.0831,
+      "loss/crossentropy": 2.418351888656616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4011431038379669,
+      "step": 834
+    },
+    {
+      "epoch": 0.01672,
+      "grad_norm": 4.03125,
+      "grad_norm_var": 3.322565714518229,
+      "learning_rate": 0.0001,
+      "loss": 6.1852,
+      "loss/crossentropy": 2.40928852558136,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40596309304237366,
+      "step": 836
+    },
+    {
+      "epoch": 0.01676,
+      "grad_norm": 3.34375,
+      "grad_norm_var": 3.3059234619140625,
+      "learning_rate": 0.0001,
+      "loss": 5.791,
+      "loss/crossentropy": 2.4211392402648926,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3640473484992981,
+      "step": 838
+    },
+    {
+      "epoch": 0.0168,
+      "grad_norm": 3.484375,
+      "grad_norm_var": 3.2890625,
+      "learning_rate": 0.0001,
+      "loss": 5.6553,
+      "loss/crossentropy": 2.047215461730957,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3346950262784958,
+      "step": 840
+    },
+    {
+      "epoch": 0.01684,
+      "grad_norm": 3.578125,
+      "grad_norm_var": 3.28623046875,
+      "learning_rate": 0.0001,
+      "loss": 5.7093,
+      "loss/crossentropy": 2.020021378993988,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3442958742380142,
+      "step": 842
+    },
+    {
+      "epoch": 0.01688,
+      "grad_norm": 3.96875,
+      "grad_norm_var": 3.27164306640625,
+      "learning_rate": 0.0001,
+      "loss": 6.1484,
+      "loss/crossentropy": 2.1585127115249634,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3768642693758011,
+      "step": 844
+    },
+    {
+      "epoch": 0.01692,
+      "grad_norm": 3.84375,
+      "grad_norm_var": 3.298802693684896,
+      "learning_rate": 0.0001,
+      "loss": 5.8512,
+      "loss/crossentropy": 2.2717286348342896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3644135594367981,
+      "step": 846
+    },
+    {
+      "epoch": 0.01696,
+      "grad_norm": 3.828125,
+      "grad_norm_var": 0.06539713541666667,
+      "learning_rate": 0.0001,
+      "loss": 5.9203,
+      "loss/crossentropy": 2.2683321237564087,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3795373737812042,
+      "step": 848
+    },
+    {
+      "epoch": 0.017,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.07066141764322917,
+      "learning_rate": 0.0001,
+      "loss": 5.7235,
+      "loss/crossentropy": 2.1189464330673218,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35512739419937134,
+      "step": 850
+    },
+    {
+      "epoch": 0.01704,
+      "grad_norm": 3.65625,
+      "grad_norm_var": 0.060384114583333336,
+      "learning_rate": 0.0001,
+      "loss": 5.6536,
+      "loss/crossentropy": 2.260777235031128,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35152101516723633,
+      "step": 852
+    },
+    {
+      "epoch": 0.01708,
+      "grad_norm": 3.671875,
+      "grad_norm_var": 0.06897379557291666,
+      "learning_rate": 0.0001,
+      "loss": 5.93,
+      "loss/crossentropy": 2.323577642440796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32252687215805054,
+      "step": 854
+    },
+    {
+      "epoch": 0.01712,
+      "grad_norm": 3.71875,
+      "grad_norm_var": 0.07024637858072917,
+      "learning_rate": 0.0001,
+      "loss": 6.1006,
+      "loss/crossentropy": 2.5965300798416138,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.43394728004932404,
+      "step": 856
+    },
+    {
+      "epoch": 0.01716,
+      "grad_norm": 3.140625,
+      "grad_norm_var": 0.08170572916666667,
+      "learning_rate": 0.0001,
+      "loss": 5.8612,
+      "loss/crossentropy": 2.078580856323242,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3437638282775879,
+      "step": 858
+    },
+    {
+      "epoch": 0.0172,
+      "grad_norm": 3.359375,
+      "grad_norm_var": 0.08163960774739583,
+      "learning_rate": 0.0001,
+      "loss": 5.8394,
+      "loss/crossentropy": 2.425456404685974,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3522993326187134,
+      "step": 860
+    },
+    {
+      "epoch": 0.01724,
+      "grad_norm": 3.515625,
+      "grad_norm_var": 0.08478190104166666,
+      "learning_rate": 0.0001,
+      "loss": 6.0154,
+      "loss/crossentropy": 2.2830835580825806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37188920378685,
+      "step": 862
+    },
+    {
+      "epoch": 0.01728,
+      "grad_norm": 3.8125,
+      "grad_norm_var": 0.06896870930989583,
+      "learning_rate": 0.0001,
+      "loss": 5.9086,
+      "loss/crossentropy": 2.090674340724945,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32989686727523804,
+      "step": 864
+    },
+    {
+      "epoch": 0.01732,
+      "grad_norm": 3.8125,
+      "grad_norm_var": 0.0698394775390625,
+      "learning_rate": 0.0001,
+      "loss": 5.9617,
+      "loss/crossentropy": 2.304458498954773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37342821061611176,
+      "step": 866
+    },
+    {
+      "epoch": 0.01736,
+      "grad_norm": 3.421875,
+      "grad_norm_var": 0.0715484619140625,
+      "learning_rate": 0.0001,
+      "loss": 5.8593,
+      "loss/crossentropy": 2.6545844078063965,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3627774566411972,
+      "step": 868
+    },
+    {
+      "epoch": 0.0174,
+      "grad_norm": 3.3125,
+      "grad_norm_var": 0.059789021809895836,
+      "learning_rate": 0.0001,
+      "loss": 5.6956,
+      "loss/crossentropy": 1.9977945685386658,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.318715900182724,
+      "step": 870
+    },
+    {
+      "epoch": 0.01744,
+      "grad_norm": 3.203125,
+      "grad_norm_var": 0.08033854166666667,
+      "learning_rate": 0.0001,
+      "loss": 5.7408,
+      "loss/crossentropy": 1.9226595759391785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3187277615070343,
+      "step": 872
+    },
+    {
+      "epoch": 0.01748,
+      "grad_norm": 3.6875,
+      "grad_norm_var": 0.0694732666015625,
+      "learning_rate": 0.0001,
+      "loss": 5.9863,
+      "loss/crossentropy": 2.323302686214447,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36724327504634857,
+      "step": 874
+    },
+    {
+      "epoch": 0.01752,
+      "grad_norm": 3.5625,
+      "grad_norm_var": 0.07043355305989583,
+      "learning_rate": 0.0001,
+      "loss": 5.9913,
+      "loss/crossentropy": 2.254343032836914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3444042354822159,
+      "step": 876
+    },
+    {
+      "epoch": 0.01756,
+      "grad_norm": 3.296875,
+      "grad_norm_var": 0.0774078369140625,
+      "learning_rate": 0.0001,
+      "loss": 5.5038,
+      "loss/crossentropy": 2.0819836854934692,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3477388769388199,
+      "step": 878
+    },
+    {
+      "epoch": 0.0176,
+      "grad_norm": 3.703125,
+      "grad_norm_var": 0.10198160807291666,
+      "learning_rate": 0.0001,
+      "loss": 5.9947,
+      "loss/crossentropy": 2.377693295478821,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36597058176994324,
+      "step": 880
+    },
+    {
+      "epoch": 0.01764,
+      "grad_norm": 4.21875,
+      "grad_norm_var": 0.11789449055989583,
+      "learning_rate": 0.0001,
+      "loss": 6.3011,
+      "loss/crossentropy": 2.5598798990249634,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4082639515399933,
+      "step": 882
+    },
+    {
+      "epoch": 0.01768,
+      "grad_norm": 3.6875,
+      "grad_norm_var": 0.1294830322265625,
+      "learning_rate": 0.0001,
+      "loss": 5.9966,
+      "loss/crossentropy": 2.2847843170166016,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3531750440597534,
+      "step": 884
+    },
+    {
+      "epoch": 0.01772,
+      "grad_norm": 3.390625,
+      "grad_norm_var": 0.12878316243489582,
+      "learning_rate": 0.0001,
+      "loss": 5.6685,
+      "loss/crossentropy": 1.8283140063285828,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.336555078625679,
+      "step": 886
+    },
+    {
+      "epoch": 0.01776,
+      "grad_norm": 3.65625,
+      "grad_norm_var": 0.09971415201822917,
+      "learning_rate": 0.0001,
+      "loss": 5.9507,
+      "loss/crossentropy": 2.1001436710357666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3380406051874161,
+      "step": 888
+    },
+    {
+      "epoch": 0.0178,
+      "grad_norm": 4.21875,
+      "grad_norm_var": 0.11876627604166666,
+      "learning_rate": 0.0001,
+      "loss": 5.7552,
+      "loss/crossentropy": 2.0079030990600586,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35362809896469116,
+      "step": 890
+    },
+    {
+      "epoch": 0.01784,
+      "grad_norm": 4.46875,
+      "grad_norm_var": 0.15650634765625,
+      "learning_rate": 0.0001,
+      "loss": 5.7416,
+      "loss/crossentropy": 2.176286220550537,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34369874000549316,
+      "step": 892
+    },
+    {
+      "epoch": 0.01788,
+      "grad_norm": 3.984375,
+      "grad_norm_var": 0.13087565104166668,
+      "learning_rate": 0.0001,
+      "loss": 5.9236,
+      "loss/crossentropy": 2.17675244808197,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3495863378047943,
+      "step": 894
+    },
+    {
+      "epoch": 0.01792,
+      "grad_norm": 4.0,
+      "grad_norm_var": 0.12189127604166666,
+      "learning_rate": 0.0001,
+      "loss": 5.9109,
+      "loss/crossentropy": 2.312318801879883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40353919565677643,
+      "step": 896
+    },
+    {
+      "epoch": 0.01796,
+      "grad_norm": 3.375,
+      "grad_norm_var": 0.14010009765625,
+      "learning_rate": 0.0001,
+      "loss": 6.071,
+      "loss/crossentropy": 2.28923499584198,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3770768642425537,
+      "step": 898
+    },
+    {
+      "epoch": 0.018,
+      "grad_norm": 3.546875,
+      "grad_norm_var": 0.1447662353515625,
+      "learning_rate": 0.0001,
+      "loss": 6.0275,
+      "loss/crossentropy": 2.2060720920562744,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35431359708309174,
+      "step": 900
+    },
+    {
+      "epoch": 0.01804,
+      "grad_norm": 3.296875,
+      "grad_norm_var": 0.17552083333333332,
+      "learning_rate": 0.0001,
+      "loss": 5.4052,
+      "loss/crossentropy": 2.0325206518173218,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3037416934967041,
+      "step": 902
+    },
+    {
+      "epoch": 0.01808,
+      "grad_norm": 3.546875,
+      "grad_norm_var": 0.17635091145833334,
+      "learning_rate": 0.0001,
+      "loss": 5.9148,
+      "loss/crossentropy": 2.1943042278289795,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3509814143180847,
+      "step": 904
+    },
+    {
+      "epoch": 0.01812,
+      "grad_norm": 3.4375,
+      "grad_norm_var": 0.1696197509765625,
+      "learning_rate": 0.0001,
+      "loss": 5.563,
+      "loss/crossentropy": 1.9589214324951172,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31621459126472473,
+      "step": 906
+    },
+    {
+      "epoch": 0.01816,
+      "grad_norm": 3.375,
+      "grad_norm_var": 0.12841389973958334,
+      "learning_rate": 0.0001,
+      "loss": 5.6511,
+      "loss/crossentropy": 2.329489588737488,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3837278485298157,
+      "step": 908
+    },
+    {
+      "epoch": 0.0182,
+      "grad_norm": 3.5625,
+      "grad_norm_var": 0.09648030598958333,
+      "learning_rate": 0.0001,
+      "loss": 5.8082,
+      "loss/crossentropy": 2.1757726669311523,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3659580200910568,
+      "step": 910
+    },
+    {
+      "epoch": 0.01824,
+      "grad_norm": 3.75,
+      "grad_norm_var": 0.08772786458333333,
+      "learning_rate": 0.0001,
+      "loss": 5.7372,
+      "loss/crossentropy": 2.1498661041259766,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3474857211112976,
+      "step": 912
+    },
+    {
+      "epoch": 0.01828,
+      "grad_norm": 15.8125,
+      "grad_norm_var": 9.504325358072917,
+      "learning_rate": 0.0001,
+      "loss": 5.9297,
+      "loss/crossentropy": 2.4722740650177,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.505307987332344,
+      "step": 914
+    },
+    {
+      "epoch": 0.01832,
+      "grad_norm": 9.0,
+      "grad_norm_var": 10.75227762858073,
+      "learning_rate": 0.0001,
+      "loss": 5.6296,
+      "loss/crossentropy": 1.855428695678711,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31347331404685974,
+      "step": 916
+    },
+    {
+      "epoch": 0.01836,
+      "grad_norm": 3.75,
+      "grad_norm_var": 10.50523173014323,
+      "learning_rate": 0.0001,
+      "loss": 5.9769,
+      "loss/crossentropy": 2.326256275177002,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3996751010417938,
+      "step": 918
+    },
+    {
+      "epoch": 0.0184,
+      "grad_norm": 3.546875,
+      "grad_norm_var": 10.518257649739583,
+      "learning_rate": 0.0001,
+      "loss": 5.8444,
+      "loss/crossentropy": 2.3712844848632812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37993232905864716,
+      "step": 920
+    },
+    {
+      "epoch": 0.01844,
+      "grad_norm": 3.5,
+      "grad_norm_var": 10.657861328125,
+      "learning_rate": 0.0001,
+      "loss": 5.5577,
+      "loss/crossentropy": 2.0161430835723877,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3590858578681946,
+      "step": 922
+    },
+    {
+      "epoch": 0.01848,
+      "grad_norm": 5.40625,
+      "grad_norm_var": 10.520926920572917,
+      "learning_rate": 0.0001,
+      "loss": 5.6675,
+      "loss/crossentropy": 2.2401121258735657,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33380126953125,
+      "step": 924
+    },
+    {
+      "epoch": 0.01852,
+      "grad_norm": 3.328125,
+      "grad_norm_var": 10.598356119791667,
+      "learning_rate": 0.0001,
+      "loss": 5.9331,
+      "loss/crossentropy": 2.2354423999786377,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34832654893398285,
+      "step": 926
+    },
+    {
+      "epoch": 0.01856,
+      "grad_norm": 3.46875,
+      "grad_norm_var": 10.647850545247396,
+      "learning_rate": 0.0001,
+      "loss": 5.5212,
+      "loss/crossentropy": 2.158566176891327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3664778769016266,
+      "step": 928
+    },
+    {
+      "epoch": 0.0186,
+      "grad_norm": 3.703125,
+      "grad_norm_var": 2.1230377197265624,
+      "learning_rate": 0.0001,
+      "loss": 5.9146,
+      "loss/crossentropy": 2.270231008529663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35586032271385193,
+      "step": 930
+    },
+    {
+      "epoch": 0.01864,
+      "grad_norm": 3.609375,
+      "grad_norm_var": 0.3744303385416667,
+      "learning_rate": 0.0001,
+      "loss": 6.0013,
+      "loss/crossentropy": 2.233540892601013,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3789799362421036,
+      "step": 932
+    },
+    {
+      "epoch": 0.01868,
+      "grad_norm": 3.65625,
+      "grad_norm_var": 0.27898763020833334,
+      "learning_rate": 0.0001,
+      "loss": 5.5019,
+      "loss/crossentropy": 1.9381731152534485,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28776855766773224,
+      "step": 934
+    },
+    {
+      "epoch": 0.01872,
+      "grad_norm": 3.34375,
+      "grad_norm_var": 0.28227437337239586,
+      "learning_rate": 0.0001,
+      "loss": 5.6759,
+      "loss/crossentropy": 2.4225244522094727,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.355392187833786,
+      "step": 936
+    },
+    {
+      "epoch": 0.01876,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.28084208170572916,
+      "learning_rate": 0.0001,
+      "loss": 5.911,
+      "loss/crossentropy": 2.58090603351593,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3640855699777603,
+      "step": 938
+    },
+    {
+      "epoch": 0.0188,
+      "grad_norm": 3.5625,
+      "grad_norm_var": 0.03673502604166667,
+      "learning_rate": 0.0001,
+      "loss": 6.0623,
+      "loss/crossentropy": 2.436452269554138,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3797999918460846,
+      "step": 940
+    },
+    {
+      "epoch": 0.01884,
+      "grad_norm": 3.3125,
+      "grad_norm_var": 0.03680013020833333,
+      "learning_rate": 0.0001,
+      "loss": 5.7428,
+      "loss/crossentropy": 2.0378769636154175,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3278265744447708,
+      "step": 942
+    },
+    {
+      "epoch": 0.01888,
+      "grad_norm": 3.546875,
+      "grad_norm_var": 0.03860270182291667,
+      "learning_rate": 0.0001,
+      "loss": 5.6211,
+      "loss/crossentropy": 2.1212962865829468,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34586282074451447,
+      "step": 944
+    },
+    {
+      "epoch": 0.01892,
+      "grad_norm": 3.515625,
+      "grad_norm_var": 0.040087890625,
+      "learning_rate": 0.0001,
+      "loss": 5.6695,
+      "loss/crossentropy": 2.1884353160858154,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40608666837215424,
+      "step": 946
+    },
+    {
+      "epoch": 0.01896,
+      "grad_norm": 3.671875,
+      "grad_norm_var": 0.046858723958333334,
+      "learning_rate": 0.0001,
+      "loss": 5.6684,
+      "loss/crossentropy": 2.2093260288238525,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3519093841314316,
+      "step": 948
+    },
+    {
+      "epoch": 0.019,
+      "grad_norm": 3.328125,
+      "grad_norm_var": 0.03243815104166667,
+      "learning_rate": 0.0001,
+      "loss": 6.0842,
+      "loss/crossentropy": 2.4246588945388794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37247334420681,
+      "step": 950
+    },
+    {
+      "epoch": 0.01904,
+      "grad_norm": 3.671875,
+      "grad_norm_var": 0.031022135416666666,
+      "learning_rate": 0.0001,
+      "loss": 5.6116,
+      "loss/crossentropy": 1.932490050792694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32985979318618774,
+      "step": 952
+    },
+    {
+      "epoch": 0.01908,
+      "grad_norm": 3.90625,
+      "grad_norm_var": 0.7084920247395833,
+      "learning_rate": 0.0001,
+      "loss": 5.7393,
+      "loss/crossentropy": 2.4439035654067993,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.38164034485816956,
+      "step": 954
+    },
+    {
+      "epoch": 0.01912,
+      "grad_norm": 3.296875,
+      "grad_norm_var": 0.72437744140625,
+      "learning_rate": 0.0001,
+      "loss": 5.6255,
+      "loss/crossentropy": 1.8876591920852661,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3267661929130554,
+      "step": 956
+    },
+    {
+      "epoch": 0.01916,
+      "grad_norm": 3.578125,
+      "grad_norm_var": 0.6990193684895833,
+      "learning_rate": 0.0001,
+      "loss": 5.7367,
+      "loss/crossentropy": 2.284990072250366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34810060262680054,
+      "step": 958
+    },
+    {
+      "epoch": 0.0192,
+      "grad_norm": 3.53125,
+      "grad_norm_var": 0.6961008707682291,
+      "learning_rate": 0.0001,
+      "loss": 5.888,
+      "loss/crossentropy": 2.333263397216797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.42767176032066345,
+      "step": 960
+    },
+    {
+      "epoch": 0.01924,
+      "grad_norm": 3.296875,
+      "grad_norm_var": 0.7323527018229167,
+      "learning_rate": 0.0001,
+      "loss": 5.6021,
+      "loss/crossentropy": 2.2526148557662964,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3284706473350525,
+      "step": 962
+    },
+    {
+      "epoch": 0.01928,
+      "grad_norm": 4.25,
+      "grad_norm_var": 0.7586252848307292,
+      "learning_rate": 0.0001,
+      "loss": 5.5479,
+      "loss/crossentropy": 2.1782984137535095,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35957905650138855,
+      "step": 964
+    },
+    {
+      "epoch": 0.01932,
+      "grad_norm": 4.28125,
+      "grad_norm_var": 0.75205078125,
+      "learning_rate": 0.0001,
+      "loss": 6.2783,
+      "loss/crossentropy": 2.292098045349121,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35862940549850464,
+      "step": 966
+    },
+    {
+      "epoch": 0.01936,
+      "grad_norm": 3.546875,
+      "grad_norm_var": 0.7503214518229167,
+      "learning_rate": 0.0001,
+      "loss": 5.9252,
+      "loss/crossentropy": 2.102781653404236,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3527261018753052,
+      "step": 968
+    },
+    {
+      "epoch": 0.0194,
+      "grad_norm": 3.453125,
+      "grad_norm_var": 0.11111653645833333,
+      "learning_rate": 0.0001,
+      "loss": 5.8891,
+      "loss/crossentropy": 2.223380208015442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3608042299747467,
+      "step": 970
+    },
+    {
+      "epoch": 0.01944,
+      "grad_norm": 3.4375,
+      "grad_norm_var": 0.10501200358072917,
+      "learning_rate": 0.0001,
+      "loss": 5.3348,
+      "loss/crossentropy": 2.0684096813201904,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31996411085128784,
+      "step": 972
+    },
+    {
+      "epoch": 0.01948,
+      "grad_norm": 4.15625,
+      "grad_norm_var": 0.126708984375,
+      "learning_rate": 0.0001,
+      "loss": 5.6642,
+      "loss/crossentropy": 2.2011090517044067,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34404022991657257,
+      "step": 974
+    },
+    {
+      "epoch": 0.01952,
+      "grad_norm": 3.140625,
+      "grad_norm_var": 0.14937235514322916,
+      "learning_rate": 0.0001,
+      "loss": 5.5033,
+      "loss/crossentropy": 2.027641534805298,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31718479096889496,
+      "step": 976
+    },
+    {
+      "epoch": 0.01956,
+      "grad_norm": 3.484375,
+      "grad_norm_var": 0.13909403483072916,
+      "learning_rate": 0.0001,
+      "loss": 5.7294,
+      "loss/crossentropy": 2.311842203140259,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3904002010822296,
+      "step": 978
+    },
+    {
+      "epoch": 0.0196,
+      "grad_norm": 3.859375,
+      "grad_norm_var": 0.11236572265625,
+      "learning_rate": 0.0001,
+      "loss": 5.4402,
+      "loss/crossentropy": 2.3605271577835083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.38789358735084534,
+      "step": 980
+    },
+    {
+      "epoch": 0.01964,
+      "grad_norm": 3.765625,
+      "grad_norm_var": 0.095166015625,
+      "learning_rate": 0.0001,
+      "loss": 6.1094,
+      "loss/crossentropy": 2.1687097549438477,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3534909188747406,
+      "step": 982
+    },
+    {
+      "epoch": 0.01968,
+      "grad_norm": 3.6875,
+      "grad_norm_var": 0.0962554931640625,
+      "learning_rate": 0.0001,
+      "loss": 5.6656,
+      "loss/crossentropy": 2.194393038749695,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34674669802188873,
+      "step": 984
+    },
+    {
+      "epoch": 0.01972,
+      "grad_norm": 3.359375,
+      "grad_norm_var": 0.09602762858072916,
+      "learning_rate": 0.0001,
+      "loss": 5.5513,
+      "loss/crossentropy": 2.1355903148651123,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3180558532476425,
+      "step": 986
+    },
+    {
+      "epoch": 0.01976,
+      "grad_norm": 3.390625,
+      "grad_norm_var": 0.08559468587239584,
+      "learning_rate": 0.0001,
+      "loss": 5.9431,
+      "loss/crossentropy": 2.2688111066818237,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.365144744515419,
+      "step": 988
+    },
+    {
+      "epoch": 0.0198,
+      "grad_norm": 3.65625,
+      "grad_norm_var": 0.06303609212239583,
+      "learning_rate": 0.0001,
+      "loss": 5.5117,
+      "loss/crossentropy": 2.423216700553894,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34281550347805023,
+      "step": 990
+    },
+    {
+      "epoch": 0.01984,
+      "grad_norm": 3.375,
+      "grad_norm_var": 0.0419586181640625,
+      "learning_rate": 0.0001,
+      "loss": 5.4698,
+      "loss/crossentropy": 1.9360128045082092,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3343205749988556,
+      "step": 992
+    },
+    {
+      "epoch": 0.01988,
+      "grad_norm": 3.421875,
+      "grad_norm_var": 0.04468994140625,
+      "learning_rate": 0.0001,
+      "loss": 5.8128,
+      "loss/crossentropy": 2.181576132774353,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35567884147167206,
+      "step": 994
+    },
+    {
+      "epoch": 0.01992,
+      "grad_norm": 3.578125,
+      "grad_norm_var": 0.037398274739583334,
+      "learning_rate": 0.0001,
+      "loss": 5.9295,
+      "loss/crossentropy": 2.166663408279419,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34105008840560913,
+      "step": 996
+    },
+    {
+      "epoch": 0.01996,
+      "grad_norm": 3.203125,
+      "grad_norm_var": 0.027408854166666666,
+      "learning_rate": 0.0001,
+      "loss": 5.5579,
+      "loss/crossentropy": 2.285332202911377,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35812389850616455,
+      "step": 998
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.0,
+      "grad_norm_var": 35.396484375,
+      "learning_rate": 0.0001,
+      "loss": 6.3986,
+      "loss/crossentropy": 2.088365077972412,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3717931807041168,
+      "step": 1000
+    },
+    {
+      "epoch": 0.02004,
+      "grad_norm": 4.25,
+      "grad_norm_var": 35.223714192708336,
+      "learning_rate": 0.0001,
+      "loss": 6.1327,
+      "loss/crossentropy": 2.4051828384399414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.420933797955513,
+      "step": 1002
+    },
+    {
+      "epoch": 0.02008,
+      "grad_norm": 3.265625,
+      "grad_norm_var": 35.27108968098958,
+      "learning_rate": 0.0001,
+      "loss": 5.6789,
+      "loss/crossentropy": 2.3092572689056396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37571050226688385,
+      "step": 1004
+    },
+    {
+      "epoch": 0.02012,
+      "grad_norm": 3.640625,
+      "grad_norm_var": 35.29562072753906,
+      "learning_rate": 0.0001,
+      "loss": 5.6972,
+      "loss/crossentropy": 2.147248387336731,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29974566400051117,
+      "step": 1006
+    },
+    {
+      "epoch": 0.02016,
+      "grad_norm": 3.984375,
+      "grad_norm_var": 35.141299438476565,
+      "learning_rate": 0.0001,
+      "loss": 5.988,
+      "loss/crossentropy": 2.3385868668556213,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4254964739084244,
+      "step": 1008
+    },
+    {
+      "epoch": 0.0202,
+      "grad_norm": 3.625,
+      "grad_norm_var": 35.15125223795573,
+      "learning_rate": 0.0001,
+      "loss": 5.7553,
+      "loss/crossentropy": 2.142681658267975,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3764440715312958,
+      "step": 1010
+    },
+    {
+      "epoch": 0.02024,
+      "grad_norm": 3.34375,
+      "grad_norm_var": 35.18043619791667,
+      "learning_rate": 0.0001,
+      "loss": 5.5947,
+      "loss/crossentropy": 2.241790771484375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3385400176048279,
+      "step": 1012
+    },
+    {
+      "epoch": 0.02028,
+      "grad_norm": 3.859375,
+      "grad_norm_var": 34.836360677083334,
+      "learning_rate": 0.0001,
+      "loss": 6.3228,
+      "loss/crossentropy": 2.1563867330551147,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3954617381095886,
+      "step": 1014
+    },
+    {
+      "epoch": 0.02032,
+      "grad_norm": 3.484375,
+      "grad_norm_var": 0.09921468098958333,
+      "learning_rate": 0.0001,
+      "loss": 5.4613,
+      "loss/crossentropy": 1.9462800025939941,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32294341921806335,
+      "step": 1016
+    },
+    {
+      "epoch": 0.02036,
+      "grad_norm": 3.53125,
+      "grad_norm_var": 0.08026936848958334,
+      "learning_rate": 0.0001,
+      "loss": 5.4993,
+      "loss/crossentropy": 1.83676278591156,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3142092078924179,
+      "step": 1018
+    },
+    {
+      "epoch": 0.0204,
+      "grad_norm": 3.84375,
+      "grad_norm_var": 0.09038798014322917,
+      "learning_rate": 0.0001,
+      "loss": 5.8174,
+      "loss/crossentropy": 1.951962649822235,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3889008015394211,
+      "step": 1020
+    },
+    {
+      "epoch": 0.02044,
+      "grad_norm": 3.578125,
+      "grad_norm_var": 0.09976806640625,
+      "learning_rate": 0.0001,
+      "loss": 5.7173,
+      "loss/crossentropy": 2.299771785736084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3214885741472244,
+      "step": 1022
+    },
+    {
+      "epoch": 0.02048,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.09135640462239583,
+      "learning_rate": 0.0001,
+      "loss": 5.2291,
+      "loss/crossentropy": 1.9117569327354431,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32827669382095337,
+      "step": 1024
+    },
+    {
+      "epoch": 0.02052,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.096533203125,
+      "learning_rate": 0.0001,
+      "loss": 5.7974,
+      "loss/crossentropy": 2.484488010406494,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34361426532268524,
+      "step": 1026
+    },
+    {
+      "epoch": 0.02056,
+      "grad_norm": 3.59375,
+      "grad_norm_var": 0.09973551432291666,
+      "learning_rate": 0.0001,
+      "loss": 5.7044,
+      "loss/crossentropy": 2.3155311346054077,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3734404444694519,
+      "step": 1028
+    },
+    {
+      "epoch": 0.0206,
+      "grad_norm": 3.125,
+      "grad_norm_var": 0.04983317057291667,
+      "learning_rate": 0.0001,
+      "loss": 5.4202,
+      "loss/crossentropy": 2.081188380718231,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3250262886285782,
+      "step": 1030
+    },
+    {
+      "epoch": 0.02064,
+      "grad_norm": 3.09375,
+      "grad_norm_var": 0.0574127197265625,
+      "learning_rate": 0.0001,
+      "loss": 5.5885,
+      "loss/crossentropy": 2.044768512248993,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34609031677246094,
+      "step": 1032
+    },
+    {
+      "epoch": 0.02068,
+      "grad_norm": 4.0,
+      "grad_norm_var": 0.0875640869140625,
+      "learning_rate": 0.0001,
+      "loss": 6.078,
+      "loss/crossentropy": 2.0666560530662537,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.365878626704216,
+      "step": 1034
+    },
+    {
+      "epoch": 0.02072,
+      "grad_norm": 3.5625,
+      "grad_norm_var": 0.08336588541666666,
+      "learning_rate": 0.0001,
+      "loss": 5.9891,
+      "loss/crossentropy": 2.2933902740478516,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3754771202802658,
+      "step": 1036
+    },
+    {
+      "epoch": 0.02076,
+      "grad_norm": 3.28125,
+      "grad_norm_var": 0.08640848795572917,
+      "learning_rate": 0.0001,
+      "loss": 5.8105,
+      "loss/crossentropy": 2.28829288482666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3676798492670059,
+      "step": 1038
+    },
+    {
+      "epoch": 0.0208,
+      "grad_norm": 3.5,
+      "grad_norm_var": 0.08378499348958333,
+      "learning_rate": 0.0001,
+      "loss": 5.9305,
+      "loss/crossentropy": 2.5891193151474,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40094244480133057,
+      "step": 1040
+    },
+    {
+      "epoch": 0.02084,
+      "grad_norm": 3.234375,
+      "grad_norm_var": 0.08056233723958334,
+      "learning_rate": 0.0001,
+      "loss": 5.8579,
+      "loss/crossentropy": 2.238967180252075,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3226759731769562,
+      "step": 1042
+    },
+    {
+      "epoch": 0.02088,
+      "grad_norm": 3.296875,
+      "grad_norm_var": 0.07796223958333333,
+      "learning_rate": 0.0001,
+      "loss": 5.2959,
+      "loss/crossentropy": 2.0116711258888245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3511478453874588,
+      "step": 1044
+    },
+    {
+      "epoch": 0.02092,
+      "grad_norm": 3.515625,
+      "grad_norm_var": 0.07344462076822916,
+      "learning_rate": 0.0001,
+      "loss": 5.4885,
+      "loss/crossentropy": 2.4924051761627197,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3375450670719147,
+      "step": 1046
+    },
+    {
+      "epoch": 0.02096,
+      "grad_norm": 3.234375,
+      "grad_norm_var": 0.059235636393229166,
+      "learning_rate": 0.0001,
+      "loss": 5.6838,
+      "loss/crossentropy": 2.138728439807892,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34410610795021057,
+      "step": 1048
+    },
+    {
+      "epoch": 0.021,
+      "grad_norm": 3.515625,
+      "grad_norm_var": 0.046122233072916664,
+      "learning_rate": 0.0001,
+      "loss": 5.7371,
+      "loss/crossentropy": 2.3748635053634644,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3580169975757599,
+      "step": 1050
+    },
+    {
+      "epoch": 0.02104,
+      "grad_norm": 3.125,
+      "grad_norm_var": 0.044310506184895834,
+      "learning_rate": 0.0001,
+      "loss": 5.427,
+      "loss/crossentropy": 2.061552882194519,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31069953739643097,
+      "step": 1052
+    },
+    {
+      "epoch": 0.02108,
+      "grad_norm": 3.515625,
+      "grad_norm_var": 0.03769124348958333,
+      "learning_rate": 0.0001,
+      "loss": 5.3469,
+      "loss/crossentropy": 2.299555718898773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31592319905757904,
+      "step": 1054
+    },
+    {
+      "epoch": 0.02112,
+      "grad_norm": 3.53125,
+      "grad_norm_var": 0.04571940104166667,
+      "learning_rate": 0.0001,
+      "loss": 6.1254,
+      "loss/crossentropy": 2.4866377115249634,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37435297667980194,
+      "step": 1056
+    },
+    {
+      "epoch": 0.02116,
+      "grad_norm": 3.203125,
+      "grad_norm_var": 0.047684733072916666,
+      "learning_rate": 0.0001,
+      "loss": 5.5676,
+      "loss/crossentropy": 1.8185054063796997,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2784232199192047,
+      "step": 1058
+    },
+    {
+      "epoch": 0.0212,
+      "grad_norm": 3.3125,
+      "grad_norm_var": 0.04840087890625,
+      "learning_rate": 0.0001,
+      "loss": 5.6184,
+      "loss/crossentropy": 2.215538501739502,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3471361994743347,
+      "step": 1060
+    },
+    {
+      "epoch": 0.02124,
+      "grad_norm": 3.078125,
+      "grad_norm_var": 0.057616170247395834,
+      "learning_rate": 0.0001,
+      "loss": 5.7615,
+      "loss/crossentropy": 2.6912894248962402,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35911867022514343,
+      "step": 1062
+    },
+    {
+      "epoch": 0.02128,
+      "grad_norm": 3.125,
+      "grad_norm_var": 0.06670633951822917,
+      "learning_rate": 0.0001,
+      "loss": 5.3436,
+      "loss/crossentropy": 1.9757090210914612,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29868973791599274,
+      "step": 1064
+    },
+    {
+      "epoch": 0.02132,
+      "grad_norm": 3.21875,
+      "grad_norm_var": 0.052294921875,
+      "learning_rate": 0.0001,
+      "loss": 5.5334,
+      "loss/crossentropy": 2.3396666049957275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3561312407255173,
+      "step": 1066
+    },
+    {
+      "epoch": 0.02136,
+      "grad_norm": 3.359375,
+      "grad_norm_var": 0.0477935791015625,
+      "learning_rate": 0.0001,
+      "loss": 5.7564,
+      "loss/crossentropy": 2.3498982191085815,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3842166066169739,
+      "step": 1068
+    },
+    {
+      "epoch": 0.0214,
+      "grad_norm": 3.234375,
+      "grad_norm_var": 0.03717041015625,
+      "learning_rate": 0.0001,
+      "loss": 5.8936,
+      "loss/crossentropy": 2.037585139274597,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3429017663002014,
+      "step": 1070
+    },
+    {
+      "epoch": 0.02144,
+      "grad_norm": 3.484375,
+      "grad_norm_var": 0.0216949462890625,
+      "learning_rate": 0.0001,
+      "loss": 5.6938,
+      "loss/crossentropy": 2.4804376363754272,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35338981449604034,
+      "step": 1072
+    },
+    {
+      "epoch": 0.02148,
+      "grad_norm": 3.375,
+      "grad_norm_var": 0.027448527018229165,
+      "learning_rate": 0.0001,
+      "loss": 5.8146,
+      "loss/crossentropy": 2.5210201740264893,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3769296407699585,
+      "step": 1074
+    },
+    {
+      "epoch": 0.02152,
+      "grad_norm": 3.40625,
+      "grad_norm_var": 0.03369852701822917,
+      "learning_rate": 0.0001,
+      "loss": 5.7854,
+      "loss/crossentropy": 2.1258187294006348,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3353133201599121,
+      "step": 1076
+    },
+    {
+      "epoch": 0.02156,
+      "grad_norm": 3.328125,
+      "grad_norm_var": 0.026676432291666666,
+      "learning_rate": 0.0001,
+      "loss": 5.5436,
+      "loss/crossentropy": 2.2107938528060913,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34667155146598816,
+      "step": 1078
+    },
+    {
+      "epoch": 0.0216,
+      "grad_norm": 3.171875,
+      "grad_norm_var": 0.0194244384765625,
+      "learning_rate": 0.0001,
+      "loss": 5.7639,
+      "loss/crossentropy": 1.9614633321762085,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2958581745624542,
+      "step": 1080
+    },
+    {
+      "epoch": 0.02164,
+      "grad_norm": 3.515625,
+      "grad_norm_var": 0.0194976806640625,
+      "learning_rate": 0.0001,
+      "loss": 5.8219,
+      "loss/crossentropy": 2.1403249502182007,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3220343589782715,
+      "step": 1082
+    },
+    {
+      "epoch": 0.02168,
+      "grad_norm": 3.28125,
+      "grad_norm_var": 0.022493489583333335,
+      "learning_rate": 0.0001,
+      "loss": 5.6037,
+      "loss/crossentropy": 1.6533048152923584,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2814648747444153,
+      "step": 1084
+    },
+    {
+      "epoch": 0.02172,
+      "grad_norm": 3.59375,
+      "grad_norm_var": 0.0232086181640625,
+      "learning_rate": 0.0001,
+      "loss": 5.7731,
+      "loss/crossentropy": 2.69880211353302,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3835880607366562,
+      "step": 1086
+    },
+    {
+      "epoch": 0.02176,
+      "grad_norm": 3.5,
+      "grad_norm_var": 0.0240631103515625,
+      "learning_rate": 0.0001,
+      "loss": 5.8119,
+      "loss/crossentropy": 2.214504837989807,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31686101853847504,
+      "step": 1088
+    },
+    {
+      "epoch": 0.0218,
+      "grad_norm": 3.765625,
+      "grad_norm_var": 0.027164713541666666,
+      "learning_rate": 0.0001,
+      "loss": 6.0376,
+      "loss/crossentropy": 2.2377456426620483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3682183176279068,
+      "step": 1090
+    },
+    {
+      "epoch": 0.02184,
+      "grad_norm": 3.359375,
+      "grad_norm_var": 0.027799479166666665,
+      "learning_rate": 0.0001,
+      "loss": 5.7387,
+      "loss/crossentropy": 2.0977545976638794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32313986122608185,
+      "step": 1092
+    },
+    {
+      "epoch": 0.02188,
+      "grad_norm": 3.125,
+      "grad_norm_var": 0.0371490478515625,
+      "learning_rate": 0.0001,
+      "loss": 5.6407,
+      "loss/crossentropy": 2.1717870235443115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3619600385427475,
+      "step": 1094
+    },
+    {
+      "epoch": 0.02192,
+      "grad_norm": 3.546875,
+      "grad_norm_var": 0.03674723307291667,
+      "learning_rate": 0.0001,
+      "loss": 5.554,
+      "loss/crossentropy": 2.2805471420288086,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3328556418418884,
+      "step": 1096
+    },
+    {
+      "epoch": 0.02196,
+      "grad_norm": 3.71875,
+      "grad_norm_var": 0.04625244140625,
+      "learning_rate": 0.0001,
+      "loss": 5.8009,
+      "loss/crossentropy": 2.034846782684326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3610256612300873,
+      "step": 1098
+    },
+    {
+      "epoch": 0.022,
+      "grad_norm": 3.359375,
+      "grad_norm_var": 0.04612528483072917,
+      "learning_rate": 0.0001,
+      "loss": 5.6121,
+      "loss/crossentropy": 2.0208348631858826,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3311958611011505,
+      "step": 1100
+    },
+    {
+      "epoch": 0.02204,
+      "grad_norm": 3.59375,
+      "grad_norm_var": 0.04737040201822917,
+      "learning_rate": 0.0001,
+      "loss": 5.7135,
+      "loss/crossentropy": 2.1020554900169373,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3638540059328079,
+      "step": 1102
+    },
+    {
+      "epoch": 0.02208,
+      "grad_norm": 3.46875,
+      "grad_norm_var": 0.046019490559895834,
+      "learning_rate": 0.0001,
+      "loss": 5.5099,
+      "loss/crossentropy": 2.28346848487854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3434429168701172,
+      "step": 1104
+    },
+    {
+      "epoch": 0.02212,
+      "grad_norm": 3.21875,
+      "grad_norm_var": 0.1152740478515625,
+      "learning_rate": 0.0001,
+      "loss": 5.6866,
+      "loss/crossentropy": 2.103623867034912,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34000229835510254,
+      "step": 1106
+    },
+    {
+      "epoch": 0.02216,
+      "grad_norm": 4.1875,
+      "grad_norm_var": 0.14846089680989583,
+      "learning_rate": 0.0001,
+      "loss": 5.5196,
+      "loss/crossentropy": 2.205894947052002,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3432965874671936,
+      "step": 1108
+    },
+    {
+      "epoch": 0.0222,
+      "grad_norm": 3.109375,
+      "grad_norm_var": 0.16467183430989582,
+      "learning_rate": 0.0001,
+      "loss": 5.6166,
+      "loss/crossentropy": 2.395035982131958,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3834904432296753,
+      "step": 1110
+    },
+    {
+      "epoch": 0.02224,
+      "grad_norm": 3.296875,
+      "grad_norm_var": 0.16505533854166668,
+      "learning_rate": 0.0001,
+      "loss": 5.9042,
+      "loss/crossentropy": 2.3755353689193726,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33256760239601135,
+      "step": 1112
+    },
+    {
+      "epoch": 0.02228,
+      "grad_norm": 3.28125,
+      "grad_norm_var": 0.16402079264322916,
+      "learning_rate": 0.0001,
+      "loss": 6.0135,
+      "loss/crossentropy": 2.6754449605941772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3524845540523529,
+      "step": 1114
+    },
+    {
+      "epoch": 0.02232,
+      "grad_norm": 3.515625,
+      "grad_norm_var": 0.15730692545572916,
+      "learning_rate": 0.0001,
+      "loss": 5.6448,
+      "loss/crossentropy": 2.2398552894592285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.343609020113945,
+      "step": 1116
+    },
+    {
+      "epoch": 0.02236,
+      "grad_norm": 3.3125,
+      "grad_norm_var": 0.16770833333333332,
+      "learning_rate": 0.0001,
+      "loss": 5.3075,
+      "loss/crossentropy": 2.399322271347046,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34514427185058594,
+      "step": 1118
+    },
+    {
+      "epoch": 0.0224,
+      "grad_norm": 3.3125,
+      "grad_norm_var": 0.17229410807291667,
+      "learning_rate": 0.0001,
+      "loss": 5.8886,
+      "loss/crossentropy": 2.4002050161361694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35784730315208435,
+      "step": 1120
+    },
+    {
+      "epoch": 0.02244,
+      "grad_norm": 3.21875,
+      "grad_norm_var": 0.10064188639322917,
+      "learning_rate": 0.0001,
+      "loss": 5.6239,
+      "loss/crossentropy": 2.293683171272278,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37052902579307556,
+      "step": 1122
+    },
+    {
+      "epoch": 0.02248,
+      "grad_norm": 3.234375,
+      "grad_norm_var": 0.0532379150390625,
+      "learning_rate": 0.0001,
+      "loss": 5.8244,
+      "loss/crossentropy": 2.2177391052246094,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3482564836740494,
+      "step": 1124
+    },
+    {
+      "epoch": 0.02252,
+      "grad_norm": 4.1875,
+      "grad_norm_var": 0.0637603759765625,
+      "learning_rate": 0.0001,
+      "loss": 5.8959,
+      "loss/crossentropy": 2.288211703300476,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3401540517807007,
+      "step": 1126
+    },
+    {
+      "epoch": 0.02256,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.0933258056640625,
+      "learning_rate": 0.0001,
+      "loss": 5.5054,
+      "loss/crossentropy": 2.1786144971847534,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.331302285194397,
+      "step": 1128
+    },
+    {
+      "epoch": 0.0226,
+      "grad_norm": 3.5,
+      "grad_norm_var": 0.09163004557291667,
+      "learning_rate": 0.0001,
+      "loss": 5.7604,
+      "loss/crossentropy": 2.0390175580978394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32953669130802155,
+      "step": 1130
+    },
+    {
+      "epoch": 0.02264,
+      "grad_norm": 3.5,
+      "grad_norm_var": 0.0921295166015625,
+      "learning_rate": 0.0001,
+      "loss": 5.6952,
+      "loss/crossentropy": 2.188543677330017,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3585694134235382,
+      "step": 1132
+    },
+    {
+      "epoch": 0.02268,
+      "grad_norm": 3.296875,
+      "grad_norm_var": 0.08284098307291667,
+      "learning_rate": 0.0001,
+      "loss": 5.732,
+      "loss/crossentropy": 2.0731694102287292,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3121813088655472,
+      "step": 1134
+    },
+    {
+      "epoch": 0.02272,
+      "grad_norm": 3.390625,
+      "grad_norm_var": 0.0794097900390625,
+      "learning_rate": 0.0001,
+      "loss": 5.6306,
+      "loss/crossentropy": 2.144552707672119,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3303966820240021,
+      "step": 1136
+    },
+    {
+      "epoch": 0.02276,
+      "grad_norm": 3.890625,
+      "grad_norm_var": 0.09900614420572916,
+      "learning_rate": 0.0001,
+      "loss": 5.3794,
+      "loss/crossentropy": 2.1617428064346313,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3420899361371994,
+      "step": 1138
+    },
+    {
+      "epoch": 0.0228,
+      "grad_norm": 3.109375,
+      "grad_norm_var": 0.12813212076822916,
+      "learning_rate": 0.0001,
+      "loss": 5.2834,
+      "loss/crossentropy": 1.8098865747451782,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28376901149749756,
+      "step": 1140
+    },
+    {
+      "epoch": 0.02284,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.09795633951822917,
+      "learning_rate": 0.0001,
+      "loss": 5.3911,
+      "loss/crossentropy": 2.133797824382782,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3587050139904022,
+      "step": 1142
+    },
+    {
+      "epoch": 0.02288,
+      "grad_norm": 4.96875,
+      "grad_norm_var": 0.24221903483072918,
+      "learning_rate": 0.0001,
+      "loss": 5.8787,
+      "loss/crossentropy": 2.378090739250183,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4244185537099838,
+      "step": 1144
+    },
+    {
+      "epoch": 0.02292,
+      "grad_norm": 3.03125,
+      "grad_norm_var": 0.2736887613932292,
+      "learning_rate": 0.0001,
+      "loss": 5.6692,
+      "loss/crossentropy": 2.4442414045333862,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3569464683532715,
+      "step": 1146
+    },
+    {
+      "epoch": 0.02296,
+      "grad_norm": 3.671875,
+      "grad_norm_var": 0.29189453125,
+      "learning_rate": 0.0001,
+      "loss": 5.5468,
+      "loss/crossentropy": 2.6446973085403442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36859095096588135,
+      "step": 1148
+    },
+    {
+      "epoch": 0.023,
+      "grad_norm": 4.4375,
+      "grad_norm_var": 0.3506988525390625,
+      "learning_rate": 0.0001,
+      "loss": 5.9727,
+      "loss/crossentropy": 2.4100207090377808,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4082309305667877,
+      "step": 1150
+    },
+    {
+      "epoch": 0.02304,
+      "grad_norm": 3.25,
+      "grad_norm_var": 0.3552317301432292,
+      "learning_rate": 0.0001,
+      "loss": 5.3537,
+      "loss/crossentropy": 2.1472485661506653,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3299275189638138,
+      "step": 1152
+    },
+    {
+      "epoch": 0.02308,
+      "grad_norm": 3.21875,
+      "grad_norm_var": 0.3376373291015625,
+      "learning_rate": 0.0001,
+      "loss": 5.4393,
+      "loss/crossentropy": 2.1891872882843018,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.356732040643692,
+      "step": 1154
+    },
+    {
+      "epoch": 0.02312,
+      "grad_norm": 3.71875,
+      "grad_norm_var": 0.30201416015625,
+      "learning_rate": 0.0001,
+      "loss": 6.0049,
+      "loss/crossentropy": 2.1432933807373047,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.360423281788826,
+      "step": 1156
+    },
+    {
+      "epoch": 0.02316,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.2923248291015625,
+      "learning_rate": 0.0001,
+      "loss": 5.5541,
+      "loss/crossentropy": 2.176819324493408,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.340317040681839,
+      "step": 1158
+    },
+    {
+      "epoch": 0.0232,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.17407938639322917,
+      "learning_rate": 0.0001,
+      "loss": 5.2312,
+      "loss/crossentropy": 2.325207471847534,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3309635668992996,
+      "step": 1160
+    },
+    {
+      "epoch": 0.02324,
+      "grad_norm": 3.515625,
+      "grad_norm_var": 0.14537353515625,
+      "learning_rate": 0.0001,
+      "loss": 5.7322,
+      "loss/crossentropy": 2.2536743879318237,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31422293186187744,
+      "step": 1162
+    },
+    {
+      "epoch": 0.02328,
+      "grad_norm": 3.328125,
+      "grad_norm_var": 0.12657877604166667,
+      "learning_rate": 0.0001,
+      "loss": 5.4354,
+      "loss/crossentropy": 2.180476427078247,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2956361174583435,
+      "step": 1164
+    },
+    {
+      "epoch": 0.02332,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.0527008056640625,
+      "learning_rate": 0.0001,
+      "loss": 5.6067,
+      "loss/crossentropy": 1.995088815689087,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2936599552631378,
+      "step": 1166
+    },
+    {
+      "epoch": 0.02336,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.05378316243489583,
+      "learning_rate": 0.0001,
+      "loss": 5.7304,
+      "loss/crossentropy": 2.2555994987487793,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3630402684211731,
+      "step": 1168
+    },
+    {
+      "epoch": 0.0234,
+      "grad_norm": 3.296875,
+      "grad_norm_var": 0.05856119791666667,
+      "learning_rate": 0.0001,
+      "loss": 5.1613,
+      "loss/crossentropy": 2.0441415905952454,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2960120141506195,
+      "step": 1170
+    },
+    {
+      "epoch": 0.02344,
+      "grad_norm": 3.25,
+      "grad_norm_var": 0.0204498291015625,
+      "learning_rate": 0.0001,
+      "loss": 5.2238,
+      "loss/crossentropy": 1.8090497255325317,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29750876128673553,
+      "step": 1172
+    },
+    {
+      "epoch": 0.02348,
+      "grad_norm": 3.234375,
+      "grad_norm_var": 0.0281158447265625,
+      "learning_rate": 0.0001,
+      "loss": 5.2506,
+      "loss/crossentropy": 2.2780312299728394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3376633822917938,
+      "step": 1174
+    },
+    {
+      "epoch": 0.02352,
+      "grad_norm": 3.4375,
+      "grad_norm_var": 0.02506103515625,
+      "learning_rate": 0.0001,
+      "loss": 5.53,
+      "loss/crossentropy": 1.8047232627868652,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29132279753685,
+      "step": 1176
+    },
+    {
+      "epoch": 0.02356,
+      "grad_norm": 3.265625,
+      "grad_norm_var": 0.020417277018229166,
+      "learning_rate": 0.0001,
+      "loss": 5.4149,
+      "loss/crossentropy": 2.203469753265381,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3124798536300659,
+      "step": 1178
+    },
+    {
+      "epoch": 0.0236,
+      "grad_norm": 3.34375,
+      "grad_norm_var": 0.019775390625,
+      "learning_rate": 0.0001,
+      "loss": 5.4481,
+      "loss/crossentropy": 2.078102231025696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.331977054476738,
+      "step": 1180
+    },
+    {
+      "epoch": 0.02364,
+      "grad_norm": 3.765625,
+      "grad_norm_var": 0.0368316650390625,
+      "learning_rate": 0.0001,
+      "loss": 5.93,
+      "loss/crossentropy": 2.2701854705810547,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3202142268419266,
+      "step": 1182
+    },
+    {
+      "epoch": 0.02368,
+      "grad_norm": 3.671875,
+      "grad_norm_var": 0.047240193684895834,
+      "learning_rate": 0.0001,
+      "loss": 5.6414,
+      "loss/crossentropy": 1.930423617362976,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3192301094532013,
+      "step": 1184
+    },
+    {
+      "epoch": 0.02372,
+      "grad_norm": 3.546875,
+      "grad_norm_var": 0.06542867024739583,
+      "learning_rate": 0.0001,
+      "loss": 6.0688,
+      "loss/crossentropy": 2.291581869125366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3759836256504059,
+      "step": 1186
+    },
+    {
+      "epoch": 0.02376,
+      "grad_norm": 3.375,
+      "grad_norm_var": 0.06529541015625,
+      "learning_rate": 0.0001,
+      "loss": 5.6689,
+      "loss/crossentropy": 1.9887789487838745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31285202503204346,
+      "step": 1188
+    },
+    {
+      "epoch": 0.0238,
+      "grad_norm": 3.109375,
+      "grad_norm_var": 0.054671223958333334,
+      "learning_rate": 0.0001,
+      "loss": 5.7457,
+      "loss/crossentropy": 2.290129065513611,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3461494445800781,
+      "step": 1190
+    },
+    {
+      "epoch": 0.02384,
+      "grad_norm": 3.6875,
+      "grad_norm_var": 0.0604888916015625,
+      "learning_rate": 0.0001,
+      "loss": 6.288,
+      "loss/crossentropy": 2.3252567052841187,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3817393332719803,
+      "step": 1192
+    },
+    {
+      "epoch": 0.02388,
+      "grad_norm": 3.21875,
+      "grad_norm_var": 0.06230367024739583,
+      "learning_rate": 0.0001,
+      "loss": 5.4498,
+      "loss/crossentropy": 1.736217737197876,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28928153216838837,
+      "step": 1194
+    },
+    {
+      "epoch": 0.02392,
+      "grad_norm": 3.09375,
+      "grad_norm_var": 0.06886393229166667,
+      "learning_rate": 0.0001,
+      "loss": 5.6234,
+      "loss/crossentropy": 2.112093210220337,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3196643739938736,
+      "step": 1196
+    },
+    {
+      "epoch": 0.02396,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.07258707682291667,
+      "learning_rate": 0.0001,
+      "loss": 5.3552,
+      "loss/crossentropy": 2.3334370851516724,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3539857119321823,
+      "step": 1198
+    },
+    {
+      "epoch": 0.024,
+      "grad_norm": 3.484375,
+      "grad_norm_var": 0.09695536295572917,
+      "learning_rate": 0.0001,
+      "loss": 5.3293,
+      "loss/crossentropy": 1.7393567562103271,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3211805671453476,
+      "step": 1200
+    },
+    {
+      "epoch": 0.02404,
+      "grad_norm": 3.4375,
+      "grad_norm_var": 0.07935791015625,
+      "learning_rate": 0.0001,
+      "loss": 5.6292,
+      "loss/crossentropy": 2.314788579940796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37599092721939087,
+      "step": 1202
+    },
+    {
+      "epoch": 0.02408,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.076318359375,
+      "learning_rate": 0.0001,
+      "loss": 5.435,
+      "loss/crossentropy": 2.2820088863372803,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33955833315849304,
+      "step": 1204
+    },
+    {
+      "epoch": 0.02412,
+      "grad_norm": 3.46875,
+      "grad_norm_var": 0.07183837890625,
+      "learning_rate": 0.0001,
+      "loss": 5.87,
+      "loss/crossentropy": 2.012476146221161,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3048281967639923,
+      "step": 1206
+    },
+    {
+      "epoch": 0.02416,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.07356669108072916,
+      "learning_rate": 0.0001,
+      "loss": 5.6751,
+      "loss/crossentropy": 2.315858483314514,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32628118991851807,
+      "step": 1208
+    },
+    {
+      "epoch": 0.0242,
+      "grad_norm": 3.359375,
+      "grad_norm_var": 0.14773763020833333,
+      "learning_rate": 0.0001,
+      "loss": 5.6813,
+      "loss/crossentropy": 2.3260581493377686,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.352965384721756,
+      "step": 1210
+    },
+    {
+      "epoch": 0.02424,
+      "grad_norm": 4.0625,
+      "grad_norm_var": 0.1748687744140625,
+      "learning_rate": 0.0001,
+      "loss": 5.689,
+      "loss/crossentropy": 2.199007749557495,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3089260905981064,
+      "step": 1212
+    },
+    {
+      "epoch": 0.02428,
+      "grad_norm": 3.703125,
+      "grad_norm_var": 1.4942047119140625,
+      "learning_rate": 0.0001,
+      "loss": 5.798,
+      "loss/crossentropy": 2.30281138420105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3629491478204727,
+      "step": 1214
+    },
+    {
+      "epoch": 0.02432,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 1.5125640869140624,
+      "learning_rate": 0.0001,
+      "loss": 5.4927,
+      "loss/crossentropy": 2.238295316696167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3436104357242584,
+      "step": 1216
+    },
+    {
+      "epoch": 0.02436,
+      "grad_norm": 3.9375,
+      "grad_norm_var": 1.501488240559896,
+      "learning_rate": 0.0001,
+      "loss": 5.5763,
+      "loss/crossentropy": 2.52456271648407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40382860600948334,
+      "step": 1218
+    },
+    {
+      "epoch": 0.0244,
+      "grad_norm": 3.46875,
+      "grad_norm_var": 1.47197265625,
+      "learning_rate": 0.0001,
+      "loss": 5.8796,
+      "loss/crossentropy": 2.4516665935516357,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.378818541765213,
+      "step": 1220
+    },
+    {
+      "epoch": 0.02444,
+      "grad_norm": 3.046875,
+      "grad_norm_var": 1.485480753580729,
+      "learning_rate": 0.0001,
+      "loss": 5.5438,
+      "loss/crossentropy": 2.593857169151306,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3743816912174225,
+      "step": 1222
+    },
+    {
+      "epoch": 0.02448,
+      "grad_norm": 3.375,
+      "grad_norm_var": 1.4396799723307292,
+      "learning_rate": 0.0001,
+      "loss": 5.2807,
+      "loss/crossentropy": 1.873874843120575,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2743247449398041,
+      "step": 1224
+    },
+    {
+      "epoch": 0.02452,
+      "grad_norm": 3.265625,
+      "grad_norm_var": 1.4642862955729166,
+      "learning_rate": 0.0001,
+      "loss": 5.6455,
+      "loss/crossentropy": 2.1173152923583984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.350399985909462,
+      "step": 1226
+    },
+    {
+      "epoch": 0.02456,
+      "grad_norm": 3.5,
+      "grad_norm_var": 1.4721018473307292,
+      "learning_rate": 0.0001,
+      "loss": 5.7318,
+      "loss/crossentropy": 2.3650271892547607,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3241463750600815,
+      "step": 1228
+    },
+    {
+      "epoch": 0.0246,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.08748270670572916,
+      "learning_rate": 0.0001,
+      "loss": 5.5229,
+      "loss/crossentropy": 2.180622935295105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31293927133083344,
+      "step": 1230
+    },
+    {
+      "epoch": 0.02464,
+      "grad_norm": 3.546875,
+      "grad_norm_var": 0.08329976399739583,
+      "learning_rate": 0.0001,
+      "loss": 5.7443,
+      "loss/crossentropy": 2.230265259742737,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3405953049659729,
+      "step": 1232
+    },
+    {
+      "epoch": 0.02468,
+      "grad_norm": 3.6875,
+      "grad_norm_var": 0.07164713541666666,
+      "learning_rate": 0.0001,
+      "loss": 5.5518,
+      "loss/crossentropy": 2.050285518169403,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32737791538238525,
+      "step": 1234
+    },
+    {
+      "epoch": 0.02472,
+      "grad_norm": 3.140625,
+      "grad_norm_var": 0.07604166666666666,
+      "learning_rate": 0.0001,
+      "loss": 5.4412,
+      "loss/crossentropy": 2.2357693910598755,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34366659820079803,
+      "step": 1236
+    },
+    {
+      "epoch": 0.02476,
+      "grad_norm": 3.09375,
+      "grad_norm_var": 0.09081624348958334,
+      "learning_rate": 0.0001,
+      "loss": 5.6066,
+      "loss/crossentropy": 2.308778762817383,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33825138211250305,
+      "step": 1238
+    },
+    {
+      "epoch": 0.0248,
+      "grad_norm": 3.390625,
+      "grad_norm_var": 0.0979888916015625,
+      "learning_rate": 0.0001,
+      "loss": 5.419,
+      "loss/crossentropy": 2.016503393650055,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30135589838027954,
+      "step": 1240
+    },
+    {
+      "epoch": 0.02484,
+      "grad_norm": 3.359375,
+      "grad_norm_var": 0.0870025634765625,
+      "learning_rate": 0.0001,
+      "loss": 5.2494,
+      "loss/crossentropy": 1.8450073599815369,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34388674795627594,
+      "step": 1242
+    },
+    {
+      "epoch": 0.02488,
+      "grad_norm": 3.515625,
+      "grad_norm_var": 0.05263671875,
+      "learning_rate": 0.0001,
+      "loss": 5.7092,
+      "loss/crossentropy": 2.486730694770813,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35796378552913666,
+      "step": 1244
+    },
+    {
+      "epoch": 0.02492,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.0688385009765625,
+      "learning_rate": 0.0001,
+      "loss": 5.2565,
+      "loss/crossentropy": 2.010675370693207,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31364670395851135,
+      "step": 1246
+    },
+    {
+      "epoch": 0.02496,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.07418619791666667,
+      "learning_rate": 0.0001,
+      "loss": 5.366,
+      "loss/crossentropy": 2.128747880458832,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3106560483574867,
+      "step": 1248
+    },
+    {
+      "epoch": 0.025,
+      "grad_norm": 3.359375,
+      "grad_norm_var": 0.06523335774739583,
+      "learning_rate": 0.0001,
+      "loss": 5.8005,
+      "loss/crossentropy": 2.4563735723495483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33769866824150085,
+      "step": 1250
+    },
+    {
+      "epoch": 0.02504,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.06453450520833333,
+      "learning_rate": 0.0001,
+      "loss": 5.372,
+      "loss/crossentropy": 2.3785592317581177,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36167748272418976,
+      "step": 1252
+    },
+    {
+      "epoch": 0.02508,
+      "grad_norm": 3.21875,
+      "grad_norm_var": 0.029069010416666666,
+      "learning_rate": 0.0001,
+      "loss": 5.5506,
+      "loss/crossentropy": 2.2642308473587036,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3315223157405853,
+      "step": 1254
+    },
+    {
+      "epoch": 0.02512,
+      "grad_norm": 3.46875,
+      "grad_norm_var": 0.028416951497395832,
+      "learning_rate": 0.0001,
+      "loss": 5.4608,
+      "loss/crossentropy": 2.2246991395950317,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33083613216876984,
+      "step": 1256
+    },
+    {
+      "epoch": 0.02516,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.0290435791015625,
+      "learning_rate": 0.0001,
+      "loss": 5.762,
+      "loss/crossentropy": 2.129785656929016,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32672610878944397,
+      "step": 1258
+    },
+    {
+      "epoch": 0.0252,
+      "grad_norm": 3.125,
+      "grad_norm_var": 0.0243804931640625,
+      "learning_rate": 0.0001,
+      "loss": 5.7297,
+      "loss/crossentropy": 2.0835453271865845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3504233658313751,
+      "step": 1260
+    },
+    {
+      "epoch": 0.02524,
+      "grad_norm": 3.703125,
+      "grad_norm_var": 0.024637858072916668,
+      "learning_rate": 0.0001,
+      "loss": 5.6707,
+      "loss/crossentropy": 2.443893313407898,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3636191487312317,
+      "step": 1262
+    },
+    {
+      "epoch": 0.02528,
+      "grad_norm": 3.21875,
+      "grad_norm_var": 0.02197265625,
+      "learning_rate": 0.0001,
+      "loss": 5.6107,
+      "loss/crossentropy": 2.367433190345764,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35734108090400696,
+      "step": 1264
+    },
+    {
+      "epoch": 0.02532,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.02662353515625,
+      "learning_rate": 0.0001,
+      "loss": 5.049,
+      "loss/crossentropy": 1.8102024793624878,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28676700592041016,
+      "step": 1266
+    },
+    {
+      "epoch": 0.02536,
+      "grad_norm": 3.0,
+      "grad_norm_var": 0.03738606770833333,
+      "learning_rate": 0.0001,
+      "loss": 5.3916,
+      "loss/crossentropy": 2.54524827003479,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3380637466907501,
+      "step": 1268
+    },
+    {
+      "epoch": 0.0254,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.040913899739583336,
+      "learning_rate": 0.0001,
+      "loss": 5.6667,
+      "loss/crossentropy": 2.4164276123046875,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36949749290943146,
+      "step": 1270
+    },
+    {
+      "epoch": 0.02544,
+      "grad_norm": 3.515625,
+      "grad_norm_var": 0.043745930989583334,
+      "learning_rate": 0.0001,
+      "loss": 5.4235,
+      "loss/crossentropy": 2.4335602521896362,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3608807325363159,
+      "step": 1272
+    },
+    {
+      "epoch": 0.02548,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.0422760009765625,
+      "learning_rate": 0.0001,
+      "loss": 5.5858,
+      "loss/crossentropy": 2.2711308002471924,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3631722033023834,
+      "step": 1274
+    },
+    {
+      "epoch": 0.02552,
+      "grad_norm": 3.140625,
+      "grad_norm_var": 0.0430816650390625,
+      "learning_rate": 0.0001,
+      "loss": 5.181,
+      "loss/crossentropy": 2.378043293952942,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35628968477249146,
+      "step": 1276
+    },
+    {
+      "epoch": 0.02556,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.02431640625,
+      "learning_rate": 0.0001,
+      "loss": 5.5721,
+      "loss/crossentropy": 1.8950039148330688,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3105602264404297,
+      "step": 1278
+    },
+    {
+      "epoch": 0.0256,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.026659138997395835,
+      "learning_rate": 0.0001,
+      "loss": 5.4649,
+      "loss/crossentropy": 1.8309656977653503,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27458132803440094,
+      "step": 1280
+    },
+    {
+      "epoch": 0.02564,
+      "grad_norm": 3.40625,
+      "grad_norm_var": 0.034195963541666666,
+      "learning_rate": 0.0001,
+      "loss": 5.993,
+      "loss/crossentropy": 2.3949296474456787,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37693680822849274,
+      "step": 1282
+    },
+    {
+      "epoch": 0.02568,
+      "grad_norm": 3.140625,
+      "grad_norm_var": 0.026725260416666667,
+      "learning_rate": 0.0001,
+      "loss": 5.6157,
+      "loss/crossentropy": 2.497879147529602,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36076007783412933,
+      "step": 1284
+    },
+    {
+      "epoch": 0.02572,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.027339680989583334,
+      "learning_rate": 0.0001,
+      "loss": 5.354,
+      "loss/crossentropy": 2.108432352542877,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3115152269601822,
+      "step": 1286
+    },
+    {
+      "epoch": 0.02576,
+      "grad_norm": 3.171875,
+      "grad_norm_var": 0.0201568603515625,
+      "learning_rate": 0.0001,
+      "loss": 5.5424,
+      "loss/crossentropy": 2.079313635826111,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31575673818588257,
+      "step": 1288
+    },
+    {
+      "epoch": 0.0258,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.020099894205729166,
+      "learning_rate": 0.0001,
+      "loss": 5.2594,
+      "loss/crossentropy": 2.3390332460403442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3293873071670532,
+      "step": 1290
+    },
+    {
+      "epoch": 0.02584,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.019612630208333332,
+      "learning_rate": 0.0001,
+      "loss": 5.2895,
+      "loss/crossentropy": 2.180980920791626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30600421130657196,
+      "step": 1292
+    },
+    {
+      "epoch": 0.02588,
+      "grad_norm": 2.9375,
+      "grad_norm_var": 0.022858683268229166,
+      "learning_rate": 0.0001,
+      "loss": 5.2784,
+      "loss/crossentropy": 2.0647836327552795,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3114248663187027,
+      "step": 1294
+    },
+    {
+      "epoch": 0.02592,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.02958984375,
+      "learning_rate": 0.0001,
+      "loss": 5.5063,
+      "loss/crossentropy": 1.931971788406372,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32757866382598877,
+      "step": 1296
+    },
+    {
+      "epoch": 0.02596,
+      "grad_norm": 3.09375,
+      "grad_norm_var": 0.0202301025390625,
+      "learning_rate": 0.0001,
+      "loss": 5.6389,
+      "loss/crossentropy": 2.1180718541145325,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3272206783294678,
+      "step": 1298
+    },
+    {
+      "epoch": 0.026,
+      "grad_norm": 3.4375,
+      "grad_norm_var": 0.024616495768229166,
+      "learning_rate": 0.0001,
+      "loss": 5.5069,
+      "loss/crossentropy": 1.8535473346710205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3100028932094574,
+      "step": 1300
+    },
+    {
+      "epoch": 0.02604,
+      "grad_norm": 4.5,
+      "grad_norm_var": 0.16031901041666666,
+      "learning_rate": 0.0001,
+      "loss": 5.574,
+      "loss/crossentropy": 1.9625197052955627,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3074956685304642,
+      "step": 1302
+    },
+    {
+      "epoch": 0.02608,
+      "grad_norm": 3.25,
+      "grad_norm_var": 0.1673736572265625,
+      "learning_rate": 0.0001,
+      "loss": 5.3764,
+      "loss/crossentropy": 2.248521149158478,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33484284579753876,
+      "step": 1304
+    },
+    {
+      "epoch": 0.02612,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.1750640869140625,
+      "learning_rate": 0.0001,
+      "loss": 5.8318,
+      "loss/crossentropy": 2.6374051570892334,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3604440838098526,
+      "step": 1306
+    },
+    {
+      "epoch": 0.02616,
+      "grad_norm": 4.0625,
+      "grad_norm_var": 0.22333984375,
+      "learning_rate": 0.0001,
+      "loss": 5.5151,
+      "loss/crossentropy": 2.3213003873825073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3659953773021698,
+      "step": 1308
+    },
+    {
+      "epoch": 0.0262,
+      "grad_norm": 3.734375,
+      "grad_norm_var": 0.21988525390625,
+      "learning_rate": 0.0001,
+      "loss": 5.9334,
+      "loss/crossentropy": 2.3527311086654663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3638792932033539,
+      "step": 1310
+    },
+    {
+      "epoch": 0.02624,
+      "grad_norm": 3.140625,
+      "grad_norm_var": 0.20551656087239584,
+      "learning_rate": 0.0001,
+      "loss": 5.4442,
+      "loss/crossentropy": 1.6319801807403564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29070258140563965,
+      "step": 1312
+    },
+    {
+      "epoch": 0.02628,
+      "grad_norm": 3.125,
+      "grad_norm_var": 0.215966796875,
+      "learning_rate": 0.0001,
+      "loss": 5.1864,
+      "loss/crossentropy": 1.986265480518341,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30515219271183014,
+      "step": 1314
+    },
+    {
+      "epoch": 0.02632,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.23336181640625,
+      "learning_rate": 0.0001,
+      "loss": 5.5808,
+      "loss/crossentropy": 2.237283766269684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32876546680927277,
+      "step": 1316
+    },
+    {
+      "epoch": 0.02636,
+      "grad_norm": 3.859375,
+      "grad_norm_var": 0.13362223307291668,
+      "learning_rate": 0.0001,
+      "loss": 5.4868,
+      "loss/crossentropy": 2.215467691421509,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34449851512908936,
+      "step": 1318
+    },
+    {
+      "epoch": 0.0264,
+      "grad_norm": 3.03125,
+      "grad_norm_var": 0.12567952473958333,
+      "learning_rate": 0.0001,
+      "loss": 5.7427,
+      "loss/crossentropy": 2.264952063560486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.387825608253479,
+      "step": 1320
+    },
+    {
+      "epoch": 0.02644,
+      "grad_norm": 3.125,
+      "grad_norm_var": 0.12009989420572917,
+      "learning_rate": 0.0001,
+      "loss": 5.5614,
+      "loss/crossentropy": 2.0678945779800415,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3228907287120819,
+      "step": 1322
+    },
+    {
+      "epoch": 0.02648,
+      "grad_norm": 4.40625,
+      "grad_norm_var": 0.212841796875,
+      "learning_rate": 0.0001,
+      "loss": 5.6259,
+      "loss/crossentropy": 2.1414765119552612,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31144315004348755,
+      "step": 1324
+    },
+    {
+      "epoch": 0.02652,
+      "grad_norm": 4.03125,
+      "grad_norm_var": 0.23813374837239584,
+      "learning_rate": 0.0001,
+      "loss": 5.8487,
+      "loss/crossentropy": 2.3898890018463135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3721010088920593,
+      "step": 1326
+    },
+    {
+      "epoch": 0.02656,
+      "grad_norm": 3.421875,
+      "grad_norm_var": 0.23483784993489584,
+      "learning_rate": 0.0001,
+      "loss": 5.5417,
+      "loss/crossentropy": 2.404030442237854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36486634612083435,
+      "step": 1328
+    },
+    {
+      "epoch": 0.0266,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.22004801432291668,
+      "learning_rate": 0.0001,
+      "loss": 5.7535,
+      "loss/crossentropy": 2.1436617970466614,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3091920465230942,
+      "step": 1330
+    },
+    {
+      "epoch": 0.02664,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.2074615478515625,
+      "learning_rate": 0.0001,
+      "loss": 5.2832,
+      "loss/crossentropy": 2.106055796146393,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3344803601503372,
+      "step": 1332
+    },
+    {
+      "epoch": 0.02668,
+      "grad_norm": 3.640625,
+      "grad_norm_var": 0.19583333333333333,
+      "learning_rate": 0.0001,
+      "loss": 6.0091,
+      "loss/crossentropy": 2.465435266494751,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34129244089126587,
+      "step": 1334
+    },
+    {
+      "epoch": 0.02672,
+      "grad_norm": 3.21875,
+      "grad_norm_var": 0.18728841145833333,
+      "learning_rate": 0.0001,
+      "loss": 5.5569,
+      "loss/crossentropy": 1.9109253883361816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30202071368694305,
+      "step": 1336
+    },
+    {
+      "epoch": 0.02676,
+      "grad_norm": 3.0,
+      "grad_norm_var": 0.1890045166015625,
+      "learning_rate": 0.0001,
+      "loss": 5.7093,
+      "loss/crossentropy": 2.267784833908081,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32984335720539093,
+      "step": 1338
+    },
+    {
+      "epoch": 0.0268,
+      "grad_norm": 3.0625,
+      "grad_norm_var": 0.07266337076822917,
+      "learning_rate": 0.0001,
+      "loss": 5.4432,
+      "loss/crossentropy": 2.6131194829940796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3797858655452728,
+      "step": 1340
+    },
+    {
+      "epoch": 0.02684,
+      "grad_norm": 3.171875,
+      "grad_norm_var": 0.031891886393229166,
+      "learning_rate": 0.0001,
+      "loss": 5.7456,
+      "loss/crossentropy": 2.214663505554199,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.340796560049057,
+      "step": 1342
+    },
+    {
+      "epoch": 0.02688,
+      "grad_norm": 3.0,
+      "grad_norm_var": 0.03245340983072917,
+      "learning_rate": 0.0001,
+      "loss": 5.4205,
+      "loss/crossentropy": 2.0236783027648926,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3101559728384018,
+      "step": 1344
+    },
+    {
+      "epoch": 0.02692,
+      "grad_norm": 3.078125,
+      "grad_norm_var": 0.030565388997395835,
+      "learning_rate": 0.0001,
+      "loss": 5.2671,
+      "loss/crossentropy": 2.3260135650634766,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.339004784822464,
+      "step": 1346
+    },
+    {
+      "epoch": 0.02696,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.0317047119140625,
+      "learning_rate": 0.0001,
+      "loss": 5.6529,
+      "loss/crossentropy": 2.177807927131653,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3171197772026062,
+      "step": 1348
+    },
+    {
+      "epoch": 0.027,
+      "grad_norm": 3.375,
+      "grad_norm_var": 0.0202056884765625,
+      "learning_rate": 0.0001,
+      "loss": 5.4682,
+      "loss/crossentropy": 2.351730227470398,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34749266505241394,
+      "step": 1350
+    },
+    {
+      "epoch": 0.02704,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.017867024739583334,
+      "learning_rate": 0.0001,
+      "loss": 5.3692,
+      "loss/crossentropy": 2.2959564924240112,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3270048499107361,
+      "step": 1352
+    },
+    {
+      "epoch": 0.02708,
+      "grad_norm": 3.03125,
+      "grad_norm_var": 0.016890462239583334,
+      "learning_rate": 0.0001,
+      "loss": 5.5243,
+      "loss/crossentropy": 2.399070382118225,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32505376636981964,
+      "step": 1354
+    },
+    {
+      "epoch": 0.02712,
+      "grad_norm": 3.046875,
+      "grad_norm_var": 0.01812744140625,
+      "learning_rate": 0.0001,
+      "loss": 5.3306,
+      "loss/crossentropy": 1.9084061980247498,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3073619455099106,
+      "step": 1356
+    },
+    {
+      "epoch": 0.02716,
+      "grad_norm": 3.078125,
+      "grad_norm_var": 0.01441650390625,
+      "learning_rate": 0.0001,
+      "loss": 5.3942,
+      "loss/crossentropy": 2.1204254627227783,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2885961979627609,
+      "step": 1358
+    },
+    {
+      "epoch": 0.0272,
+      "grad_norm": 3.28125,
+      "grad_norm_var": 0.014842732747395834,
+      "learning_rate": 0.0001,
+      "loss": 5.7313,
+      "loss/crossentropy": 2.0167239904403687,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3026747703552246,
+      "step": 1360
+    },
+    {
+      "epoch": 0.02724,
+      "grad_norm": 3.171875,
+      "grad_norm_var": 0.014742024739583333,
+      "learning_rate": 0.0001,
+      "loss": 5.3987,
+      "loss/crossentropy": 1.9588357210159302,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2976878881454468,
+      "step": 1362
+    },
+    {
+      "epoch": 0.02728,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.019449869791666668,
+      "learning_rate": 0.0001,
+      "loss": 5.1658,
+      "loss/crossentropy": 1.8169561624526978,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2874959260225296,
+      "step": 1364
+    },
+    {
+      "epoch": 0.02732,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.014354451497395834,
+      "learning_rate": 0.0001,
+      "loss": 5.5128,
+      "loss/crossentropy": 2.258527398109436,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34180621802806854,
+      "step": 1366
+    },
+    {
+      "epoch": 0.02736,
+      "grad_norm": 3.09375,
+      "grad_norm_var": 0.017284138997395834,
+      "learning_rate": 0.0001,
+      "loss": 5.4374,
+      "loss/crossentropy": 2.1959571838378906,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3124794214963913,
+      "step": 1368
+    },
+    {
+      "epoch": 0.0274,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.0198150634765625,
+      "learning_rate": 0.0001,
+      "loss": 5.2299,
+      "loss/crossentropy": 2.1830934286117554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3260872811079025,
+      "step": 1370
+    },
+    {
+      "epoch": 0.02744,
+      "grad_norm": 3.203125,
+      "grad_norm_var": 0.02017822265625,
+      "learning_rate": 0.0001,
+      "loss": 5.6831,
+      "loss/crossentropy": 2.411653518676758,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3575899302959442,
+      "step": 1372
+    },
+    {
+      "epoch": 0.02748,
+      "grad_norm": 3.140625,
+      "grad_norm_var": 0.021512858072916665,
+      "learning_rate": 0.0001,
+      "loss": 5.3919,
+      "loss/crossentropy": 2.1585222482681274,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30734311044216156,
+      "step": 1374
+    },
+    {
+      "epoch": 0.02752,
+      "grad_norm": 2.90625,
+      "grad_norm_var": 0.023824055989583332,
+      "learning_rate": 0.0001,
+      "loss": 5.37,
+      "loss/crossentropy": 2.3621217012405396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3373589664697647,
+      "step": 1376
+    },
+    {
+      "epoch": 0.02756,
+      "grad_norm": 3.109375,
+      "grad_norm_var": 0.023160807291666665,
+      "learning_rate": 0.0001,
+      "loss": 5.4712,
+      "loss/crossentropy": 2.1203317046165466,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28425413370132446,
+      "step": 1378
+    },
+    {
+      "epoch": 0.0276,
+      "grad_norm": 3.296875,
+      "grad_norm_var": 0.0223297119140625,
+      "learning_rate": 0.0001,
+      "loss": 5.5438,
+      "loss/crossentropy": 2.2568705081939697,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.296497106552124,
+      "step": 1380
+    },
+    {
+      "epoch": 0.02764,
+      "grad_norm": 3.5625,
+      "grad_norm_var": 0.03665364583333333,
+      "learning_rate": 0.0001,
+      "loss": 5.5613,
+      "loss/crossentropy": 2.260026216506958,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34534430503845215,
+      "step": 1382
+    },
+    {
+      "epoch": 0.02768,
+      "grad_norm": 5.5625,
+      "grad_norm_var": 0.4100494384765625,
+      "learning_rate": 0.0001,
+      "loss": 5.6217,
+      "loss/crossentropy": 1.9400787949562073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32853779196739197,
+      "step": 1384
+    },
+    {
+      "epoch": 0.02772,
+      "grad_norm": 3.390625,
+      "grad_norm_var": 0.39205322265625,
+      "learning_rate": 0.0001,
+      "loss": 5.2009,
+      "loss/crossentropy": 2.168904423713684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3421178460121155,
+      "step": 1386
+    },
+    {
+      "epoch": 0.02776,
+      "grad_norm": 2.90625,
+      "grad_norm_var": 0.405419921875,
+      "learning_rate": 0.0001,
+      "loss": 5.3992,
+      "loss/crossentropy": 2.3474777936935425,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31982940435409546,
+      "step": 1388
+    },
+    {
+      "epoch": 0.0278,
+      "grad_norm": 3.140625,
+      "grad_norm_var": 0.40615234375,
+      "learning_rate": 0.0001,
+      "loss": 5.5791,
+      "loss/crossentropy": 2.3416868448257446,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3400905281305313,
+      "step": 1390
+    },
+    {
+      "epoch": 0.02784,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.4003570556640625,
+      "learning_rate": 0.0001,
+      "loss": 5.4413,
+      "loss/crossentropy": 2.299672842025757,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3143058717250824,
+      "step": 1392
+    },
+    {
+      "epoch": 0.02788,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.3945058186848958,
+      "learning_rate": 0.0001,
+      "loss": 5.6231,
+      "loss/crossentropy": 2.3058812618255615,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33790935575962067,
+      "step": 1394
+    },
+    {
+      "epoch": 0.02792,
+      "grad_norm": 3.03125,
+      "grad_norm_var": 0.3947987874348958,
+      "learning_rate": 0.0001,
+      "loss": 5.324,
+      "loss/crossentropy": 2.2137999534606934,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33156970143318176,
+      "step": 1396
+    },
+    {
+      "epoch": 0.02796,
+      "grad_norm": 4.1875,
+      "grad_norm_var": 0.4427571614583333,
+      "learning_rate": 0.0001,
+      "loss": 5.4567,
+      "loss/crossentropy": 2.04184353351593,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32263143360614777,
+      "step": 1398
+    },
+    {
+      "epoch": 0.028,
+      "grad_norm": 3.265625,
+      "grad_norm_var": 0.09589436848958334,
+      "learning_rate": 0.0001,
+      "loss": 5.339,
+      "loss/crossentropy": 2.0377472639083862,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3136487454175949,
+      "step": 1400
+    },
+    {
+      "epoch": 0.02804,
+      "grad_norm": 3.171875,
+      "grad_norm_var": 0.09228413899739583,
+      "learning_rate": 0.0001,
+      "loss": 5.5838,
+      "loss/crossentropy": 2.5366055965423584,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3476262539625168,
+      "step": 1402
+    },
+    {
+      "epoch": 0.02808,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.09063212076822917,
+      "learning_rate": 0.0001,
+      "loss": 5.2447,
+      "loss/crossentropy": 2.088012456893921,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31111815571784973,
+      "step": 1404
+    },
+    {
+      "epoch": 0.02812,
+      "grad_norm": 3.53125,
+      "grad_norm_var": 0.10530598958333333,
+      "learning_rate": 0.0001,
+      "loss": 5.7316,
+      "loss/crossentropy": 2.1750329732894897,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3138856291770935,
+      "step": 1406
+    },
+    {
+      "epoch": 0.02816,
+      "grad_norm": 3.625,
+      "grad_norm_var": 0.09811909993489583,
+      "learning_rate": 0.0001,
+      "loss": 5.4708,
+      "loss/crossentropy": 1.977031648159027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3179774433374405,
+      "step": 1408
+    },
+    {
+      "epoch": 0.0282,
+      "grad_norm": 3.296875,
+      "grad_norm_var": 0.12561848958333333,
+      "learning_rate": 0.0001,
+      "loss": 5.3815,
+      "loss/crossentropy": 2.0594210028648376,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32200056314468384,
+      "step": 1410
+    },
+    {
+      "epoch": 0.02824,
+      "grad_norm": 3.59375,
+      "grad_norm_var": 0.11237691243489584,
+      "learning_rate": 0.0001,
+      "loss": 5.443,
+      "loss/crossentropy": 2.3887473344802856,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3303599953651428,
+      "step": 1412
+    },
+    {
+      "epoch": 0.02828,
+      "grad_norm": 3.453125,
+      "grad_norm_var": 0.06500244140625,
+      "learning_rate": 0.0001,
+      "loss": 5.5507,
+      "loss/crossentropy": 2.188898801803589,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3378629684448242,
+      "step": 1414
+    },
+    {
+      "epoch": 0.02832,
+      "grad_norm": 3.171875,
+      "grad_norm_var": 0.06982014973958334,
+      "learning_rate": 0.0001,
+      "loss": 5.458,
+      "loss/crossentropy": 1.981561303138733,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2814937084913254,
+      "step": 1416
+    },
+    {
+      "epoch": 0.02836,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.0756011962890625,
+      "learning_rate": 0.0001,
+      "loss": 5.271,
+      "loss/crossentropy": 2.2141716480255127,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3429889380931854,
+      "step": 1418
+    },
+    {
+      "epoch": 0.0284,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.0818267822265625,
+      "learning_rate": 0.0001,
+      "loss": 5.048,
+      "loss/crossentropy": 2.0720977783203125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30601558089256287,
+      "step": 1420
+    },
+    {
+      "epoch": 0.02844,
+      "grad_norm": 3.21875,
+      "grad_norm_var": 0.07434488932291666,
+      "learning_rate": 0.0001,
+      "loss": 5.6407,
+      "loss/crossentropy": 2.0516344904899597,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3182393014431,
+      "step": 1422
+    },
+    {
+      "epoch": 0.02848,
+      "grad_norm": 3.28125,
+      "grad_norm_var": 0.070166015625,
+      "learning_rate": 0.0001,
+      "loss": 5.6532,
+      "loss/crossentropy": 2.161284327507019,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3383013904094696,
+      "step": 1424
+    },
+    {
+      "epoch": 0.02852,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.03163960774739583,
+      "learning_rate": 0.0001,
+      "loss": 5.1594,
+      "loss/crossentropy": 1.9955796599388123,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3169983923435211,
+      "step": 1426
+    },
+    {
+      "epoch": 0.02856,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.023421223958333334,
+      "learning_rate": 0.0001,
+      "loss": 5.3951,
+      "loss/crossentropy": 2.1046979427337646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30594320595264435,
+      "step": 1428
+    },
+    {
+      "epoch": 0.0286,
+      "grad_norm": 2.90625,
+      "grad_norm_var": 0.020231119791666665,
+      "learning_rate": 0.0001,
+      "loss": 5.364,
+      "loss/crossentropy": 1.9611601829528809,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3039780706167221,
+      "step": 1430
+    },
+    {
+      "epoch": 0.02864,
+      "grad_norm": 2.9375,
+      "grad_norm_var": 0.021610514322916666,
+      "learning_rate": 0.0001,
+      "loss": 5.8823,
+      "loss/crossentropy": 2.256209373474121,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3315662145614624,
+      "step": 1432
+    },
+    {
+      "epoch": 0.02868,
+      "grad_norm": 2.9375,
+      "grad_norm_var": 0.030594889322916666,
+      "learning_rate": 0.0001,
+      "loss": 5.2821,
+      "loss/crossentropy": 2.150561034679413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33046241104602814,
+      "step": 1434
+    },
+    {
+      "epoch": 0.02872,
+      "grad_norm": 3.078125,
+      "grad_norm_var": 0.024898274739583334,
+      "learning_rate": 0.0001,
+      "loss": 5.3609,
+      "loss/crossentropy": 2.279554605484009,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3621693551540375,
+      "step": 1436
+    },
+    {
+      "epoch": 0.02876,
+      "grad_norm": 3.015625,
+      "grad_norm_var": 0.0259918212890625,
+      "learning_rate": 0.0001,
+      "loss": 5.4323,
+      "loss/crossentropy": 2.0183790922164917,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33361808955669403,
+      "step": 1438
+    },
+    {
+      "epoch": 0.0288,
+      "grad_norm": 4.15625,
+      "grad_norm_var": 0.7352203369140625,
+      "learning_rate": 0.0001,
+      "loss": 5.4781,
+      "loss/crossentropy": 1.964252531528473,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31638333201408386,
+      "step": 1440
+    },
+    {
+      "epoch": 0.02884,
+      "grad_norm": 3.046875,
+      "grad_norm_var": 0.73902587890625,
+      "learning_rate": 0.0001,
+      "loss": 5.6907,
+      "loss/crossentropy": 1.9271634817123413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3408525586128235,
+      "step": 1442
+    },
+    {
+      "epoch": 0.02888,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.7411417643229167,
+      "learning_rate": 0.0001,
+      "loss": 5.4869,
+      "loss/crossentropy": 2.4400887489318848,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33455249667167664,
+      "step": 1444
+    },
+    {
+      "epoch": 0.02892,
+      "grad_norm": 3.078125,
+      "grad_norm_var": 0.7270904541015625,
+      "learning_rate": 0.0001,
+      "loss": 5.6179,
+      "loss/crossentropy": 2.465711832046509,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3528379648923874,
+      "step": 1446
+    },
+    {
+      "epoch": 0.02896,
+      "grad_norm": 3.328125,
+      "grad_norm_var": 0.7147288004557292,
+      "learning_rate": 0.0001,
+      "loss": 5.5125,
+      "loss/crossentropy": 2.269619941711426,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32898105680942535,
+      "step": 1448
+    },
+    {
+      "epoch": 0.029,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.68287353515625,
+      "learning_rate": 0.0001,
+      "loss": 5.5775,
+      "loss/crossentropy": 2.0418076515197754,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3170912265777588,
+      "step": 1450
+    },
+    {
+      "epoch": 0.02904,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.6841756184895833,
+      "learning_rate": 0.0001,
+      "loss": 5.5964,
+      "loss/crossentropy": 2.291188359260559,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35694004595279694,
+      "step": 1452
+    },
+    {
+      "epoch": 0.02908,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.7253163655598959,
+      "learning_rate": 0.0001,
+      "loss": 5.1608,
+      "loss/crossentropy": 2.497802972793579,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3271156847476959,
+      "step": 1454
+    },
+    {
+      "epoch": 0.02912,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.0334625244140625,
+      "learning_rate": 0.0001,
+      "loss": 5.1527,
+      "loss/crossentropy": 2.161794900894165,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32509492337703705,
+      "step": 1456
+    },
+    {
+      "epoch": 0.02916,
+      "grad_norm": 3.3125,
+      "grad_norm_var": 0.03470052083333333,
+      "learning_rate": 0.0001,
+      "loss": 5.7939,
+      "loss/crossentropy": 2.632015347480774,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36461199820041656,
+      "step": 1458
+    },
+    {
+      "epoch": 0.0292,
+      "grad_norm": 3.03125,
+      "grad_norm_var": 0.03372395833333333,
+      "learning_rate": 0.0001,
+      "loss": 5.6233,
+      "loss/crossentropy": 2.2478749752044678,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3311367332935333,
+      "step": 1460
+    },
+    {
+      "epoch": 0.02924,
+      "grad_norm": 4.03125,
+      "grad_norm_var": 0.09263916015625,
+      "learning_rate": 0.0001,
+      "loss": 5.2528,
+      "loss/crossentropy": 1.8451723456382751,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31802159547805786,
+      "step": 1462
+    },
+    {
+      "epoch": 0.02928,
+      "grad_norm": 3.71875,
+      "grad_norm_var": 0.11772359212239583,
+      "learning_rate": 0.0001,
+      "loss": 5.6469,
+      "loss/crossentropy": 2.6684207916259766,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35483625531196594,
+      "step": 1464
+    },
+    {
+      "epoch": 0.02932,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.11543680826822916,
+      "learning_rate": 0.0001,
+      "loss": 5.2866,
+      "loss/crossentropy": 2.497614622116089,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3422156721353531,
+      "step": 1466
+    },
+    {
+      "epoch": 0.02936,
+      "grad_norm": 4.25,
+      "grad_norm_var": 0.19032796223958334,
+      "learning_rate": 0.0001,
+      "loss": 5.7424,
+      "loss/crossentropy": 2.750740170478821,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.377558171749115,
+      "step": 1468
+    },
+    {
+      "epoch": 0.0294,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.17550455729166667,
+      "learning_rate": 0.0001,
+      "loss": 5.1741,
+      "loss/crossentropy": 1.9961607456207275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3166217654943466,
+      "step": 1470
+    },
+    {
+      "epoch": 0.02944,
+      "grad_norm": 3.359375,
+      "grad_norm_var": 0.1655914306640625,
+      "learning_rate": 0.0001,
+      "loss": 5.3114,
+      "loss/crossentropy": 2.0874768495559692,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2912164479494095,
+      "step": 1472
+    },
+    {
+      "epoch": 0.02948,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.174072265625,
+      "learning_rate": 0.0001,
+      "loss": 5.2102,
+      "loss/crossentropy": 2.112182080745697,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29954925179481506,
+      "step": 1474
+    },
+    {
+      "epoch": 0.02952,
+      "grad_norm": 3.6875,
+      "grad_norm_var": 0.4074371337890625,
+      "learning_rate": 0.0001,
+      "loss": 5.7839,
+      "loss/crossentropy": 2.1319644451141357,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3002544492483139,
+      "step": 1476
+    },
+    {
+      "epoch": 0.02956,
+      "grad_norm": 3.0625,
+      "grad_norm_var": 0.37743733723958334,
+      "learning_rate": 0.0001,
+      "loss": 5.8305,
+      "loss/crossentropy": 2.3029643297195435,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3262677788734436,
+      "step": 1478
+    },
+    {
+      "epoch": 0.0296,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.3826243082682292,
+      "learning_rate": 0.0001,
+      "loss": 5.6803,
+      "loss/crossentropy": 2.8598941564559937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.38816460967063904,
+      "step": 1480
+    },
+    {
+      "epoch": 0.02964,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.37844950358072915,
+      "learning_rate": 0.0001,
+      "loss": 5.2177,
+      "loss/crossentropy": 2.134063720703125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3128468096256256,
+      "step": 1482
+    },
+    {
+      "epoch": 0.02968,
+      "grad_norm": 3.125,
+      "grad_norm_var": 0.31961263020833336,
+      "learning_rate": 0.0001,
+      "loss": 5.5234,
+      "loss/crossentropy": 2.481287717819214,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3209179639816284,
+      "step": 1484
+    },
+    {
+      "epoch": 0.02972,
+      "grad_norm": 3.3125,
+      "grad_norm_var": 0.30549723307291665,
+      "learning_rate": 0.0001,
+      "loss": 5.3571,
+      "loss/crossentropy": 2.0571895837783813,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3121480941772461,
+      "step": 1486
+    },
+    {
+      "epoch": 0.02976,
+      "grad_norm": 3.203125,
+      "grad_norm_var": 0.30614827473958334,
+      "learning_rate": 0.0001,
+      "loss": 5.2725,
+      "loss/crossentropy": 2.1930073499679565,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3031492233276367,
+      "step": 1488
+    },
+    {
+      "epoch": 0.0298,
+      "grad_norm": 3.046875,
+      "grad_norm_var": 0.30426025390625,
+      "learning_rate": 0.0001,
+      "loss": 5.3256,
+      "loss/crossentropy": 2.403902530670166,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32206277549266815,
+      "step": 1490
+    },
+    {
+      "epoch": 0.02984,
+      "grad_norm": 2.9375,
+      "grad_norm_var": 0.0538482666015625,
+      "learning_rate": 0.0001,
+      "loss": 5.2859,
+      "loss/crossentropy": 2.131115198135376,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31651052832603455,
+      "step": 1492
+    },
+    {
+      "epoch": 0.02988,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.05826416015625,
+      "learning_rate": 0.0001,
+      "loss": 5.1232,
+      "loss/crossentropy": 2.024750769138336,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3142934888601303,
+      "step": 1494
+    },
+    {
+      "epoch": 0.02992,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.06279195149739583,
+      "learning_rate": 0.0001,
+      "loss": 5.3804,
+      "loss/crossentropy": 1.9606707692146301,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28610387444496155,
+      "step": 1496
+    },
+    {
+      "epoch": 0.02996,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.062474568684895836,
+      "learning_rate": 0.0001,
+      "loss": 5.5731,
+      "loss/crossentropy": 2.31516432762146,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32875190675258636,
+      "step": 1498
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.359375,
+      "grad_norm_var": 0.0711822509765625,
+      "learning_rate": 0.0001,
+      "loss": 5.7626,
+      "loss/crossentropy": 2.295942783355713,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3379499167203903,
+      "step": 1500
+    },
+    {
+      "epoch": 0.03004,
+      "grad_norm": 3.21875,
+      "grad_norm_var": 0.06961263020833333,
+      "learning_rate": 0.0001,
+      "loss": 5.5997,
+      "loss/crossentropy": 2.1859577894210815,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32143887877464294,
+      "step": 1502
+    },
+    {
+      "epoch": 0.03008,
+      "grad_norm": 3.28125,
+      "grad_norm_var": 0.07023824055989583,
+      "learning_rate": 0.0001,
+      "loss": 5.4159,
+      "loss/crossentropy": 2.1852502822875977,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3582807630300522,
+      "step": 1504
+    },
+    {
+      "epoch": 0.03012,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.07810770670572917,
+      "learning_rate": 0.0001,
+      "loss": 5.3011,
+      "loss/crossentropy": 2.174897611141205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33067959547042847,
+      "step": 1506
+    },
+    {
+      "epoch": 0.03016,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.04057515462239583,
+      "learning_rate": 0.0001,
+      "loss": 5.2775,
+      "loss/crossentropy": 2.24343740940094,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34563779830932617,
+      "step": 1508
+    },
+    {
+      "epoch": 0.0302,
+      "grad_norm": 2.90625,
+      "grad_norm_var": 0.043473307291666666,
+      "learning_rate": 0.0001,
+      "loss": 5.317,
+      "loss/crossentropy": 1.9828822612762451,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2887475937604904,
+      "step": 1510
+    },
+    {
+      "epoch": 0.03024,
+      "grad_norm": 3.0,
+      "grad_norm_var": 0.03655192057291667,
+      "learning_rate": 0.0001,
+      "loss": 5.3172,
+      "loss/crossentropy": 2.2110393047332764,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32290786504745483,
+      "step": 1512
+    },
+    {
+      "epoch": 0.03028,
+      "grad_norm": 2.90625,
+      "grad_norm_var": 0.04527587890625,
+      "learning_rate": 0.0001,
+      "loss": 5.2598,
+      "loss/crossentropy": 2.3797603845596313,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3262799382209778,
+      "step": 1514
+    },
+    {
+      "epoch": 0.03032,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.0281158447265625,
+      "learning_rate": 0.0001,
+      "loss": 5.4107,
+      "loss/crossentropy": 2.0183085799217224,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2915680408477783,
+      "step": 1516
+    },
+    {
+      "epoch": 0.03036,
+      "grad_norm": 3.78125,
+      "grad_norm_var": 0.9888631184895833,
+      "learning_rate": 0.0001,
+      "loss": 5.4632,
+      "loss/crossentropy": 1.875212013721466,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30001460015773773,
+      "step": 1518
+    },
+    {
+      "epoch": 0.0304,
+      "grad_norm": 3.0625,
+      "grad_norm_var": 0.9916300455729167,
+      "learning_rate": 0.0001,
+      "loss": 5.4406,
+      "loss/crossentropy": 2.1000564098358154,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.354188472032547,
+      "step": 1520
+    },
+    {
+      "epoch": 0.03044,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.980126953125,
+      "learning_rate": 0.0001,
+      "loss": 5.4837,
+      "loss/crossentropy": 2.071643114089966,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.333335280418396,
+      "step": 1522
+    },
+    {
+      "epoch": 0.03048,
+      "grad_norm": 3.203125,
+      "grad_norm_var": 0.9749664306640625,
+      "learning_rate": 0.0001,
+      "loss": 5.2716,
+      "loss/crossentropy": 2.4253947734832764,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31365686655044556,
+      "step": 1524
+    },
+    {
+      "epoch": 0.03052,
+      "grad_norm": 3.0625,
+      "grad_norm_var": 0.9605133056640625,
+      "learning_rate": 0.0001,
+      "loss": 5.1601,
+      "loss/crossentropy": 1.967090904712677,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30677899718284607,
+      "step": 1526
+    },
+    {
+      "epoch": 0.03056,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.954443359375,
+      "learning_rate": 0.0001,
+      "loss": 5.0971,
+      "loss/crossentropy": 2.112701952457428,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3029911667108536,
+      "step": 1528
+    },
+    {
+      "epoch": 0.0306,
+      "grad_norm": 4.5625,
+      "grad_norm_var": 1.0084269205729166,
+      "learning_rate": 0.0001,
+      "loss": 5.6836,
+      "loss/crossentropy": 2.5657063722610474,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3484763503074646,
+      "step": 1530
+    },
+    {
+      "epoch": 0.03064,
+      "grad_norm": 3.234375,
+      "grad_norm_var": 1.110497029622396,
+      "learning_rate": 0.0001,
+      "loss": 5.3888,
+      "loss/crossentropy": 2.1214585304260254,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3243858218193054,
+      "step": 1532
+    },
+    {
+      "epoch": 0.03068,
+      "grad_norm": 3.234375,
+      "grad_norm_var": 0.35461832682291666,
+      "learning_rate": 0.0001,
+      "loss": 5.4662,
+      "loss/crossentropy": 2.427902936935425,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3428474962711334,
+      "step": 1534
+    },
+    {
+      "epoch": 0.03072,
+      "grad_norm": 3.078125,
+      "grad_norm_var": 0.36741434733072914,
+      "learning_rate": 0.0001,
+      "loss": 5.2956,
+      "loss/crossentropy": 1.975690484046936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3136949688196182,
+      "step": 1536
+    },
+    {
+      "epoch": 0.03076,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.38342692057291666,
+      "learning_rate": 0.0001,
+      "loss": 5.1233,
+      "loss/crossentropy": 2.295845150947571,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31703390181064606,
+      "step": 1538
+    },
+    {
+      "epoch": 0.0308,
+      "grad_norm": 3.171875,
+      "grad_norm_var": 0.37892252604166665,
+      "learning_rate": 0.0001,
+      "loss": 5.8286,
+      "loss/crossentropy": 2.117497444152832,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32601243257522583,
+      "step": 1540
+    },
+    {
+      "epoch": 0.03084,
+      "grad_norm": 3.140625,
+      "grad_norm_var": 0.49339192708333335,
+      "learning_rate": 0.0001,
+      "loss": 5.4065,
+      "loss/crossentropy": 2.3824862241744995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3332519829273224,
+      "step": 1542
+    },
+    {
+      "epoch": 0.03088,
+      "grad_norm": 3.140625,
+      "grad_norm_var": 0.4940582275390625,
+      "learning_rate": 0.0001,
+      "loss": 5.0672,
+      "loss/crossentropy": 1.9037857055664062,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3143462985754013,
+      "step": 1544
+    },
+    {
+      "epoch": 0.03092,
+      "grad_norm": 3.09375,
+      "grad_norm_var": 0.4064737955729167,
+      "learning_rate": 0.0001,
+      "loss": 5.2328,
+      "loss/crossentropy": 1.9191133379936218,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29443541169166565,
+      "step": 1546
+    },
+    {
+      "epoch": 0.03096,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.28172200520833335,
+      "learning_rate": 0.0001,
+      "loss": 5.3402,
+      "loss/crossentropy": 2.216462254524231,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31411711871623993,
+      "step": 1548
+    },
+    {
+      "epoch": 0.031,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.227685546875,
+      "learning_rate": 0.0001,
+      "loss": 5.0455,
+      "loss/crossentropy": 1.8064754605293274,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2667318135499954,
+      "step": 1550
+    },
+    {
+      "epoch": 0.03104,
+      "grad_norm": 3.453125,
+      "grad_norm_var": 0.46340738932291664,
+      "learning_rate": 0.0001,
+      "loss": 5.5672,
+      "loss/crossentropy": 2.488176465034485,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3422655761241913,
+      "step": 1552
+    },
+    {
+      "epoch": 0.03108,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.4529205322265625,
+      "learning_rate": 0.0001,
+      "loss": 5.4404,
+      "loss/crossentropy": 2.5670164823532104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3462950587272644,
+      "step": 1554
+    },
+    {
+      "epoch": 0.03112,
+      "grad_norm": 3.25,
+      "grad_norm_var": 0.46499735514322915,
+      "learning_rate": 0.0001,
+      "loss": 5.0295,
+      "loss/crossentropy": 2.0630581378936768,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29241877794265747,
+      "step": 1556
+    },
+    {
+      "epoch": 0.03116,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.30891011555989584,
+      "learning_rate": 0.0001,
+      "loss": 5.4751,
+      "loss/crossentropy": 2.685954213142395,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37143297493457794,
+      "step": 1558
+    },
+    {
+      "epoch": 0.0312,
+      "grad_norm": 3.421875,
+      "grad_norm_var": 0.311669921875,
+      "learning_rate": 0.0001,
+      "loss": 5.6995,
+      "loss/crossentropy": 1.9786988496780396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40276341140270233,
+      "step": 1560
+    },
+    {
+      "epoch": 0.03124,
+      "grad_norm": 3.0,
+      "grad_norm_var": 0.31383056640625,
+      "learning_rate": 0.0001,
+      "loss": 5.6695,
+      "loss/crossentropy": 2.1484411358833313,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3264364004135132,
+      "step": 1562
+    },
+    {
+      "epoch": 0.03128,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.3103424072265625,
+      "learning_rate": 0.0001,
+      "loss": 5.3015,
+      "loss/crossentropy": 2.1411852836608887,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3156583160161972,
+      "step": 1564
+    },
+    {
+      "epoch": 0.03132,
+      "grad_norm": 2.90625,
+      "grad_norm_var": 0.31302083333333336,
+      "learning_rate": 0.0001,
+      "loss": 5.2911,
+      "loss/crossentropy": 2.1509006023406982,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3066476732492447,
+      "step": 1566
+    },
+    {
+      "epoch": 0.03136,
+      "grad_norm": 2.90625,
+      "grad_norm_var": 0.0416412353515625,
+      "learning_rate": 0.0001,
+      "loss": 5.1904,
+      "loss/crossentropy": 1.7540676593780518,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2705047130584717,
+      "step": 1568
+    },
+    {
+      "epoch": 0.0314,
+      "grad_norm": 3.046875,
+      "grad_norm_var": 0.0431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.9637,
+      "loss/crossentropy": 2.2091184854507446,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2920738309621811,
+      "step": 1570
+    },
+    {
+      "epoch": 0.03144,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.045735677083333336,
+      "learning_rate": 0.0001,
+      "loss": 5.6048,
+      "loss/crossentropy": 1.8405091762542725,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2981649935245514,
+      "step": 1572
+    },
+    {
+      "epoch": 0.03148,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.0420806884765625,
+      "learning_rate": 0.0001,
+      "loss": 5.4258,
+      "loss/crossentropy": 2.2292457818984985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30654460191726685,
+      "step": 1574
+    },
+    {
+      "epoch": 0.03152,
+      "grad_norm": 4.21875,
+      "grad_norm_var": 0.12868550618489583,
+      "learning_rate": 0.0001,
+      "loss": 5.8439,
+      "loss/crossentropy": 2.653907895088196,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35303865373134613,
+      "step": 1576
+    },
+    {
+      "epoch": 0.03156,
+      "grad_norm": 3.234375,
+      "grad_norm_var": 0.13888346354166667,
+      "learning_rate": 0.0001,
+      "loss": 5.6145,
+      "loss/crossentropy": 2.6799376010894775,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37383997440338135,
+      "step": 1578
+    },
+    {
+      "epoch": 0.0316,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.14431864420572918,
+      "learning_rate": 0.0001,
+      "loss": 5.3321,
+      "loss/crossentropy": 2.3438535928726196,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30706922709941864,
+      "step": 1580
+    },
+    {
+      "epoch": 0.03164,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.14254150390625,
+      "learning_rate": 0.0001,
+      "loss": 5.2139,
+      "loss/crossentropy": 2.1885964274406433,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30707718431949615,
+      "step": 1582
+    },
+    {
+      "epoch": 0.03168,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.14042561848958332,
+      "learning_rate": 0.0001,
+      "loss": 5.1661,
+      "loss/crossentropy": 2.0471584796905518,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31723180413246155,
+      "step": 1584
+    },
+    {
+      "epoch": 0.03172,
+      "grad_norm": 3.0,
+      "grad_norm_var": 0.1363433837890625,
+      "learning_rate": 0.0001,
+      "loss": 5.227,
+      "loss/crossentropy": 2.135176420211792,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29839280247688293,
+      "step": 1586
+    },
+    {
+      "epoch": 0.03176,
+      "grad_norm": 3.3125,
+      "grad_norm_var": 0.13482666015625,
+      "learning_rate": 0.0001,
+      "loss": 5.6412,
+      "loss/crossentropy": 2.4444775581359863,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33574284613132477,
+      "step": 1588
+    },
+    {
+      "epoch": 0.0318,
+      "grad_norm": 3.140625,
+      "grad_norm_var": 0.12195536295572916,
+      "learning_rate": 0.0001,
+      "loss": 5.4289,
+      "loss/crossentropy": 2.1725653409957886,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.310588076710701,
+      "step": 1590
+    },
+    {
+      "epoch": 0.03184,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.04572652180989583,
+      "learning_rate": 0.0001,
+      "loss": 5.3727,
+      "loss/crossentropy": 2.3610929250717163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32158929109573364,
+      "step": 1592
+    },
+    {
+      "epoch": 0.03188,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.0502105712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.8794,
+      "loss/crossentropy": 1.9271156787872314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28283432126045227,
+      "step": 1594
+    },
+    {
+      "epoch": 0.03192,
+      "grad_norm": 3.203125,
+      "grad_norm_var": 0.04372456868489583,
+      "learning_rate": 0.0001,
+      "loss": 5.3912,
+      "loss/crossentropy": 2.4196890592575073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3463610112667084,
+      "step": 1596
+    },
+    {
+      "epoch": 0.03196,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.0500152587890625,
+      "learning_rate": 0.0001,
+      "loss": 5.1524,
+      "loss/crossentropy": 2.207236647605896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33829018473625183,
+      "step": 1598
+    },
+    {
+      "epoch": 0.032,
+      "grad_norm": 3.0625,
+      "grad_norm_var": 0.04101155598958333,
+      "learning_rate": 0.0001,
+      "loss": 5.4724,
+      "loss/crossentropy": 2.3757678270339966,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3342677056789398,
+      "step": 1600
+    },
+    {
+      "epoch": 0.03204,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.04512430826822917,
+      "learning_rate": 0.0001,
+      "loss": 5.1763,
+      "loss/crossentropy": 2.2605016231536865,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3062159866094589,
+      "step": 1602
+    },
+    {
+      "epoch": 0.03208,
+      "grad_norm": 3.359375,
+      "grad_norm_var": 0.04794921875,
+      "learning_rate": 0.0001,
+      "loss": 5.7139,
+      "loss/crossentropy": 2.4536768198013306,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36668023467063904,
+      "step": 1604
+    },
+    {
+      "epoch": 0.03212,
+      "grad_norm": 3.09375,
+      "grad_norm_var": 0.0523834228515625,
+      "learning_rate": 0.0001,
+      "loss": 5.137,
+      "loss/crossentropy": 1.9870036244392395,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2974477708339691,
+      "step": 1606
+    },
+    {
+      "epoch": 0.03216,
+      "grad_norm": 3.015625,
+      "grad_norm_var": 0.039876302083333336,
+      "learning_rate": 0.0001,
+      "loss": 5.3926,
+      "loss/crossentropy": 2.1852506399154663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3327452540397644,
+      "step": 1608
+    },
+    {
+      "epoch": 0.0322,
+      "grad_norm": 3.140625,
+      "grad_norm_var": 0.03128255208333333,
+      "learning_rate": 0.0001,
+      "loss": 5.4964,
+      "loss/crossentropy": 2.2226197719573975,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3110218793153763,
+      "step": 1610
+    },
+    {
+      "epoch": 0.03224,
+      "grad_norm": 3.09375,
+      "grad_norm_var": 0.029523722330729165,
+      "learning_rate": 0.0001,
+      "loss": 5.398,
+      "loss/crossentropy": 1.8255922198295593,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28216174244880676,
+      "step": 1612
+    },
+    {
+      "epoch": 0.03228,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.023200480143229167,
+      "learning_rate": 0.0001,
+      "loss": 5.2777,
+      "loss/crossentropy": 1.9663920998573303,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3095496743917465,
+      "step": 1614
+    },
+    {
+      "epoch": 0.03232,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.023908487955729165,
+      "learning_rate": 0.0001,
+      "loss": 5.1578,
+      "loss/crossentropy": 2.2089942693710327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3307983875274658,
+      "step": 1616
+    },
+    {
+      "epoch": 0.03236,
+      "grad_norm": 3.03125,
+      "grad_norm_var": 0.01793212890625,
+      "learning_rate": 0.0001,
+      "loss": 5.3331,
+      "loss/crossentropy": 2.261039137840271,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.314799427986145,
+      "step": 1618
+    },
+    {
+      "epoch": 0.0324,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.015555826822916667,
+      "learning_rate": 0.0001,
+      "loss": 5.1674,
+      "loss/crossentropy": 2.350824236869812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3515756279230118,
+      "step": 1620
+    },
+    {
+      "epoch": 0.03244,
+      "grad_norm": 2.9375,
+      "grad_norm_var": 0.010749308268229167,
+      "learning_rate": 0.0001,
+      "loss": 5.4853,
+      "loss/crossentropy": 2.2964736223220825,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3059935122728348,
+      "step": 1622
+    },
+    {
+      "epoch": 0.03248,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.016597493489583334,
+      "learning_rate": 0.0001,
+      "loss": 5.2503,
+      "loss/crossentropy": 2.36459481716156,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33763329684734344,
+      "step": 1624
+    },
+    {
+      "epoch": 0.03252,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.0126617431640625,
+      "learning_rate": 0.0001,
+      "loss": 5.2242,
+      "loss/crossentropy": 2.165920853614807,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31318019330501556,
+      "step": 1626
+    },
+    {
+      "epoch": 0.03256,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.010628255208333333,
+      "learning_rate": 0.0001,
+      "loss": 5.2544,
+      "loss/crossentropy": 2.326790690422058,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3044355511665344,
+      "step": 1628
+    },
+    {
+      "epoch": 0.0326,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.01129150390625,
+      "learning_rate": 0.0001,
+      "loss": 5.3369,
+      "loss/crossentropy": 1.8848688006401062,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.306812584400177,
+      "step": 1630
+    },
+    {
+      "epoch": 0.03264,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.0134185791015625,
+      "learning_rate": 0.0001,
+      "loss": 5.351,
+      "loss/crossentropy": 2.045863091945648,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3091462701559067,
+      "step": 1632
+    },
+    {
+      "epoch": 0.03268,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.01148681640625,
+      "learning_rate": 0.0001,
+      "loss": 4.9742,
+      "loss/crossentropy": 2.0707273483276367,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3038959503173828,
+      "step": 1634
+    },
+    {
+      "epoch": 0.03272,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.011237589518229167,
+      "learning_rate": 0.0001,
+      "loss": 5.2076,
+      "loss/crossentropy": 2.0786932706832886,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29513464868068695,
+      "step": 1636
+    },
+    {
+      "epoch": 0.03276,
+      "grad_norm": 3.203125,
+      "grad_norm_var": 0.01900634765625,
+      "learning_rate": 0.0001,
+      "loss": 5.4237,
+      "loss/crossentropy": 2.1527108550071716,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32102274894714355,
+      "step": 1638
+    },
+    {
+      "epoch": 0.0328,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.021361287434895834,
+      "learning_rate": 0.0001,
+      "loss": 5.126,
+      "loss/crossentropy": 2.0383081436157227,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3253529220819473,
+      "step": 1640
+    },
+    {
+      "epoch": 0.03284,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.021809895833333332,
+      "learning_rate": 0.0001,
+      "loss": 5.0775,
+      "loss/crossentropy": 2.2801902294158936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32200203835964203,
+      "step": 1642
+    },
+    {
+      "epoch": 0.03288,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.022184244791666665,
+      "learning_rate": 0.0001,
+      "loss": 5.3035,
+      "loss/crossentropy": 2.164717435836792,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2997436225414276,
+      "step": 1644
+    },
+    {
+      "epoch": 0.03292,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.025191243489583334,
+      "learning_rate": 0.0001,
+      "loss": 5.5166,
+      "loss/crossentropy": 2.389414072036743,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32998231053352356,
+      "step": 1646
+    },
+    {
+      "epoch": 0.03296,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.023949178059895833,
+      "learning_rate": 0.0001,
+      "loss": 5.3225,
+      "loss/crossentropy": 2.09418523311615,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33829881250858307,
+      "step": 1648
+    },
+    {
+      "epoch": 0.033,
+      "grad_norm": 3.109375,
+      "grad_norm_var": 0.021800740559895834,
+      "learning_rate": 0.0001,
+      "loss": 5.261,
+      "loss/crossentropy": 2.324030637741089,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30337512493133545,
+      "step": 1650
+    },
+    {
+      "epoch": 0.03304,
+      "grad_norm": 3.109375,
+      "grad_norm_var": 0.020992024739583334,
+      "learning_rate": 0.0001,
+      "loss": 5.4162,
+      "loss/crossentropy": 1.8635556101799011,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27077023684978485,
+      "step": 1652
+    },
+    {
+      "epoch": 0.03308,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.0215972900390625,
+      "learning_rate": 0.0001,
+      "loss": 5.2371,
+      "loss/crossentropy": 2.1776190996170044,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30546560883522034,
+      "step": 1654
+    },
+    {
+      "epoch": 0.03312,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.021240234375,
+      "learning_rate": 0.0001,
+      "loss": 5.1721,
+      "loss/crossentropy": 2.1003682613372803,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3304767310619354,
+      "step": 1656
+    },
+    {
+      "epoch": 0.03316,
+      "grad_norm": 3.203125,
+      "grad_norm_var": 0.02584228515625,
+      "learning_rate": 0.0001,
+      "loss": 5.8029,
+      "loss/crossentropy": 2.4331823587417603,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3653264045715332,
+      "step": 1658
+    },
+    {
+      "epoch": 0.0332,
+      "grad_norm": 3.125,
+      "grad_norm_var": 0.1084381103515625,
+      "learning_rate": 0.0001,
+      "loss": 5.7157,
+      "loss/crossentropy": 1.9939777851104736,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28948865830898285,
+      "step": 1660
+    },
+    {
+      "epoch": 0.03324,
+      "grad_norm": 2.9375,
+      "grad_norm_var": 0.10896809895833333,
+      "learning_rate": 0.0001,
+      "loss": 4.933,
+      "loss/crossentropy": 1.9093859791755676,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3007328063249588,
+      "step": 1662
+    },
+    {
+      "epoch": 0.03328,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.11998697916666666,
+      "learning_rate": 0.0001,
+      "loss": 5.1861,
+      "loss/crossentropy": 2.2847355604171753,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2975463569164276,
+      "step": 1664
+    },
+    {
+      "epoch": 0.03332,
+      "grad_norm": 3.046875,
+      "grad_norm_var": 0.120166015625,
+      "learning_rate": 0.0001,
+      "loss": 5.3623,
+      "loss/crossentropy": 2.0081310868263245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28596948087215424,
+      "step": 1666
+    },
+    {
+      "epoch": 0.03336,
+      "grad_norm": 3.125,
+      "grad_norm_var": 0.11974283854166666,
+      "learning_rate": 0.0001,
+      "loss": 5.2338,
+      "loss/crossentropy": 2.189584493637085,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30838510394096375,
+      "step": 1668
+    },
+    {
+      "epoch": 0.0334,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.12802327473958333,
+      "learning_rate": 0.0001,
+      "loss": 5.2749,
+      "loss/crossentropy": 2.204169988632202,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3091500401496887,
+      "step": 1670
+    },
+    {
+      "epoch": 0.03344,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.1338775634765625,
+      "learning_rate": 0.0001,
+      "loss": 5.2755,
+      "loss/crossentropy": 2.195966899394989,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30603383481502533,
+      "step": 1672
+    },
+    {
+      "epoch": 0.03348,
+      "grad_norm": 3.0,
+      "grad_norm_var": 0.1251953125,
+      "learning_rate": 0.0001,
+      "loss": 5.8768,
+      "loss/crossentropy": 2.5402153730392456,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3657945841550827,
+      "step": 1674
+    },
+    {
+      "epoch": 0.03352,
+      "grad_norm": 3.09375,
+      "grad_norm_var": 0.0277252197265625,
+      "learning_rate": 0.0001,
+      "loss": 5.5387,
+      "loss/crossentropy": 2.1721729040145874,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29539716243743896,
+      "step": 1676
+    },
+    {
+      "epoch": 0.03356,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.027253214518229166,
+      "learning_rate": 0.0001,
+      "loss": 5.2355,
+      "loss/crossentropy": 2.0591543912887573,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29050062596797943,
+      "step": 1678
+    },
+    {
+      "epoch": 0.0336,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.025846354166666665,
+      "learning_rate": 0.0001,
+      "loss": 5.4895,
+      "loss/crossentropy": 2.1396639347076416,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3261077404022217,
+      "step": 1680
+    },
+    {
+      "epoch": 0.03364,
+      "grad_norm": 3.015625,
+      "grad_norm_var": 0.024689737955729166,
+      "learning_rate": 0.0001,
+      "loss": 5.2094,
+      "loss/crossentropy": 1.9553123712539673,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3204822689294815,
+      "step": 1682
+    },
+    {
+      "epoch": 0.03368,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.023746744791666666,
+      "learning_rate": 0.0001,
+      "loss": 5.3417,
+      "loss/crossentropy": 2.4139883518218994,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32602658867836,
+      "step": 1684
+    },
+    {
+      "epoch": 0.03372,
+      "grad_norm": 2.90625,
+      "grad_norm_var": 0.019624837239583335,
+      "learning_rate": 0.0001,
+      "loss": 5.3479,
+      "loss/crossentropy": 1.9849293231964111,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32610756158828735,
+      "step": 1686
+    },
+    {
+      "epoch": 0.03376,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.008622233072916667,
+      "learning_rate": 0.0001,
+      "loss": 5.6218,
+      "loss/crossentropy": 2.698970675468445,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3749641329050064,
+      "step": 1688
+    },
+    {
+      "epoch": 0.0338,
+      "grad_norm": 3.0625,
+      "grad_norm_var": 0.01285400390625,
+      "learning_rate": 0.0001,
+      "loss": 5.7617,
+      "loss/crossentropy": 2.715620517730713,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35715436935424805,
+      "step": 1690
+    },
+    {
+      "epoch": 0.03384,
+      "grad_norm": 3.0,
+      "grad_norm_var": 0.01201171875,
+      "learning_rate": 0.0001,
+      "loss": 5.5073,
+      "loss/crossentropy": 2.7213666439056396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34585659205913544,
+      "step": 1692
+    },
+    {
+      "epoch": 0.03388,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.016044108072916667,
+      "learning_rate": 0.0001,
+      "loss": 5.2674,
+      "loss/crossentropy": 2.277606725692749,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3140410780906677,
+      "step": 1694
+    },
+    {
+      "epoch": 0.03392,
+      "grad_norm": 3.765625,
+      "grad_norm_var": 0.05364176432291667,
+      "learning_rate": 0.0001,
+      "loss": 5.2633,
+      "loss/crossentropy": 2.332197904586792,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30997559428215027,
+      "step": 1696
+    },
+    {
+      "epoch": 0.03396,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.05464579264322917,
+      "learning_rate": 0.0001,
+      "loss": 5.4323,
+      "loss/crossentropy": 2.4413230419158936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3075388967990875,
+      "step": 1698
+    },
+    {
+      "epoch": 0.034,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.0567535400390625,
+      "learning_rate": 0.0001,
+      "loss": 5.1099,
+      "loss/crossentropy": 2.2601696252822876,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3050367534160614,
+      "step": 1700
+    },
+    {
+      "epoch": 0.03404,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.06572977701822917,
+      "learning_rate": 0.0001,
+      "loss": 4.9925,
+      "loss/crossentropy": 2.2910414934158325,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3248990923166275,
+      "step": 1702
+    },
+    {
+      "epoch": 0.03408,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 0.06843159993489584,
+      "learning_rate": 0.0001,
+      "loss": 5.3814,
+      "loss/crossentropy": 1.9898682832717896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27573561668395996,
+      "step": 1704
+    },
+    {
+      "epoch": 0.03412,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.06398824055989584,
+      "learning_rate": 0.0001,
+      "loss": 5.1506,
+      "loss/crossentropy": 2.1234602332115173,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30724021792411804,
+      "step": 1706
+    },
+    {
+      "epoch": 0.03416,
+      "grad_norm": 3.390625,
+      "grad_norm_var": 0.0764801025390625,
+      "learning_rate": 0.0001,
+      "loss": 5.5433,
+      "loss/crossentropy": 2.34807026386261,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32687780261039734,
+      "step": 1708
+    },
+    {
+      "epoch": 0.0342,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.07669169108072917,
+      "learning_rate": 0.0001,
+      "loss": 5.1249,
+      "loss/crossentropy": 2.17264860868454,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3086177706718445,
+      "step": 1710
+    },
+    {
+      "epoch": 0.03424,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.0303131103515625,
+      "learning_rate": 0.0001,
+      "loss": 5.3613,
+      "loss/crossentropy": 2.1094497442245483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29483039677143097,
+      "step": 1712
+    },
+    {
+      "epoch": 0.03428,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.03284505208333333,
+      "learning_rate": 0.0001,
+      "loss": 5.256,
+      "loss/crossentropy": 2.2379074692726135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2799292802810669,
+      "step": 1714
+    },
+    {
+      "epoch": 0.03432,
+      "grad_norm": 3.078125,
+      "grad_norm_var": 0.21389567057291667,
+      "learning_rate": 0.0001,
+      "loss": 5.5834,
+      "loss/crossentropy": 2.4616905450820923,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33721986413002014,
+      "step": 1716
+    },
+    {
+      "epoch": 0.03436,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.2042877197265625,
+      "learning_rate": 0.0001,
+      "loss": 5.4415,
+      "loss/crossentropy": 2.383226990699768,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3242805302143097,
+      "step": 1718
+    },
+    {
+      "epoch": 0.0344,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.19931538899739584,
+      "learning_rate": 0.0001,
+      "loss": 5.5514,
+      "loss/crossentropy": 2.495948314666748,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3131762146949768,
+      "step": 1720
+    },
+    {
+      "epoch": 0.03444,
+      "grad_norm": 3.109375,
+      "grad_norm_var": 0.19524739583333334,
+      "learning_rate": 0.0001,
+      "loss": 5.6767,
+      "loss/crossentropy": 2.1921653747558594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.320631667971611,
+      "step": 1722
+    },
+    {
+      "epoch": 0.03448,
+      "grad_norm": 3.40625,
+      "grad_norm_var": 0.20221354166666666,
+      "learning_rate": 0.0001,
+      "loss": 5.2,
+      "loss/crossentropy": 2.0795475840568542,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29602357745170593,
+      "step": 1724
+    },
+    {
+      "epoch": 0.03452,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.201171875,
+      "learning_rate": 0.0001,
+      "loss": 4.7489,
+      "loss/crossentropy": 1.911207377910614,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.287412166595459,
+      "step": 1726
+    },
+    {
+      "epoch": 0.03456,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.20129801432291666,
+      "learning_rate": 0.0001,
+      "loss": 5.155,
+      "loss/crossentropy": 2.0638214349746704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32240423560142517,
+      "step": 1728
+    },
+    {
+      "epoch": 0.0346,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.20066630045572917,
+      "learning_rate": 0.0001,
+      "loss": 5.3464,
+      "loss/crossentropy": 2.355573534965515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3161381185054779,
+      "step": 1730
+    },
+    {
+      "epoch": 0.03464,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.04185282389322917,
+      "learning_rate": 0.0001,
+      "loss": 5.1039,
+      "loss/crossentropy": 2.2227123975753784,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31909704208374023,
+      "step": 1732
+    },
+    {
+      "epoch": 0.03468,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.03954671223958333,
+      "learning_rate": 0.0001,
+      "loss": 5.2249,
+      "loss/crossentropy": 2.203602910041809,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3238847255706787,
+      "step": 1734
+    },
+    {
+      "epoch": 0.03472,
+      "grad_norm": 3.34375,
+      "grad_norm_var": 0.07062174479166666,
+      "learning_rate": 0.0001,
+      "loss": 5.0623,
+      "loss/crossentropy": 2.2696332335472107,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3174774497747421,
+      "step": 1736
+    },
+    {
+      "epoch": 0.03476,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.07385660807291666,
+      "learning_rate": 0.0001,
+      "loss": 5.0469,
+      "loss/crossentropy": 1.8124465942382812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2645348161458969,
+      "step": 1738
+    },
+    {
+      "epoch": 0.0348,
+      "grad_norm": 3.015625,
+      "grad_norm_var": 0.0578765869140625,
+      "learning_rate": 0.0001,
+      "loss": 5.589,
+      "loss/crossentropy": 2.1951464414596558,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31606370210647583,
+      "step": 1740
+    },
+    {
+      "epoch": 0.03484,
+      "grad_norm": 3.109375,
+      "grad_norm_var": 0.05624593098958333,
+      "learning_rate": 0.0001,
+      "loss": 5.4867,
+      "loss/crossentropy": 2.4413230419158936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33688417077064514,
+      "step": 1742
+    },
+    {
+      "epoch": 0.03488,
+      "grad_norm": 3.09375,
+      "grad_norm_var": 0.05705973307291667,
+      "learning_rate": 0.0001,
+      "loss": 5.7898,
+      "loss/crossentropy": 2.1357219219207764,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3476516157388687,
+      "step": 1744
+    },
+    {
+      "epoch": 0.03492,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.0551666259765625,
+      "learning_rate": 0.0001,
+      "loss": 5.4465,
+      "loss/crossentropy": 2.1557592153549194,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3080083876848221,
+      "step": 1746
+    },
+    {
+      "epoch": 0.03496,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.0512603759765625,
+      "learning_rate": 0.0001,
+      "loss": 5.4949,
+      "loss/crossentropy": 2.3549705743789673,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33814217150211334,
+      "step": 1748
+    },
+    {
+      "epoch": 0.035,
+      "grad_norm": 3.109375,
+      "grad_norm_var": 0.04348958333333333,
+      "learning_rate": 0.0001,
+      "loss": 5.5881,
+      "loss/crossentropy": 2.382844924926758,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31944599747657776,
+      "step": 1750
+    },
+    {
+      "epoch": 0.03504,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.02138671875,
+      "learning_rate": 0.0001,
+      "loss": 5.2937,
+      "loss/crossentropy": 2.3312920331954956,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3346693813800812,
+      "step": 1752
+    },
+    {
+      "epoch": 0.03508,
+      "grad_norm": 3.015625,
+      "grad_norm_var": 0.019677734375,
+      "learning_rate": 0.0001,
+      "loss": 5.1981,
+      "loss/crossentropy": 2.1921491026878357,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31133508682250977,
+      "step": 1754
+    },
+    {
+      "epoch": 0.03512,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.024494425455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.8796,
+      "loss/crossentropy": 2.0229611992836,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26832816004753113,
+      "step": 1756
+    },
+    {
+      "epoch": 0.03516,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.025764973958333333,
+      "learning_rate": 0.0001,
+      "loss": 5.0322,
+      "loss/crossentropy": 1.8138108849525452,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2576165944337845,
+      "step": 1758
+    },
+    {
+      "epoch": 0.0352,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.0232574462890625,
+      "learning_rate": 0.0001,
+      "loss": 5.1276,
+      "loss/crossentropy": 2.0019126534461975,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28094957768917084,
+      "step": 1760
+    },
+    {
+      "epoch": 0.03524,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.022554524739583335,
+      "learning_rate": 0.0001,
+      "loss": 5.1776,
+      "loss/crossentropy": 2.400240898132324,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30936548113822937,
+      "step": 1762
+    },
+    {
+      "epoch": 0.03528,
+      "grad_norm": 3.328125,
+      "grad_norm_var": 0.03495992024739583,
+      "learning_rate": 0.0001,
+      "loss": 4.9401,
+      "loss/crossentropy": 2.019958734512329,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27741560339927673,
+      "step": 1764
+    },
+    {
+      "epoch": 0.03532,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.025032552083333333,
+      "learning_rate": 0.0001,
+      "loss": 5.0912,
+      "loss/crossentropy": 1.9099596738815308,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28769225627183914,
+      "step": 1766
+    },
+    {
+      "epoch": 0.03536,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 0.024344889322916667,
+      "learning_rate": 0.0001,
+      "loss": 5.42,
+      "loss/crossentropy": 2.2508283853530884,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2945093661546707,
+      "step": 1768
+    },
+    {
+      "epoch": 0.0354,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.0222320556640625,
+      "learning_rate": 0.0001,
+      "loss": 5.2516,
+      "loss/crossentropy": 1.9332409501075745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26304441690444946,
+      "step": 1770
+    },
+    {
+      "epoch": 0.03544,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.022557576497395832,
+      "learning_rate": 0.0001,
+      "loss": 5.1896,
+      "loss/crossentropy": 2.143627643585205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28407415747642517,
+      "step": 1772
+    },
+    {
+      "epoch": 0.03548,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.029683430989583332,
+      "learning_rate": 0.0001,
+      "loss": 5.4763,
+      "loss/crossentropy": 2.32085120677948,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3176119029521942,
+      "step": 1774
+    },
+    {
+      "epoch": 0.03552,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.028539021809895832,
+      "learning_rate": 0.0001,
+      "loss": 5.1717,
+      "loss/crossentropy": 1.7893801927566528,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27407801151275635,
+      "step": 1776
+    },
+    {
+      "epoch": 0.03556,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.027228800455729167,
+      "learning_rate": 0.0001,
+      "loss": 5.42,
+      "loss/crossentropy": 2.206292986869812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3071517199277878,
+      "step": 1778
+    },
+    {
+      "epoch": 0.0356,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.013263956705729166,
+      "learning_rate": 0.0001,
+      "loss": 5.1879,
+      "loss/crossentropy": 2.1285043954849243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3164139539003372,
+      "step": 1780
+    },
+    {
+      "epoch": 0.03564,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.018635050455729166,
+      "learning_rate": 0.0001,
+      "loss": 5.2826,
+      "loss/crossentropy": 2.16570383310318,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3052050769329071,
+      "step": 1782
+    },
+    {
+      "epoch": 0.03568,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.019950358072916667,
+      "learning_rate": 0.0001,
+      "loss": 5.0921,
+      "loss/crossentropy": 1.9799941778182983,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2655785381793976,
+      "step": 1784
+    },
+    {
+      "epoch": 0.03572,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.0209625244140625,
+      "learning_rate": 0.0001,
+      "loss": 5.2468,
+      "loss/crossentropy": 1.9801498651504517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2824363112449646,
+      "step": 1786
+    },
+    {
+      "epoch": 0.03576,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.020947265625,
+      "learning_rate": 0.0001,
+      "loss": 5.0131,
+      "loss/crossentropy": 1.5805786848068237,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25234321504831314,
+      "step": 1788
+    },
+    {
+      "epoch": 0.0358,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.021484375,
+      "learning_rate": 0.0001,
+      "loss": 5.296,
+      "loss/crossentropy": 2.2434048652648926,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2896551638841629,
+      "step": 1790
+    },
+    {
+      "epoch": 0.03584,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.022411092122395834,
+      "learning_rate": 0.0001,
+      "loss": 5.0179,
+      "loss/crossentropy": 1.9738762378692627,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2959328889846802,
+      "step": 1792
+    },
+    {
+      "epoch": 0.03588,
+      "grad_norm": 3.0625,
+      "grad_norm_var": 0.0272857666015625,
+      "learning_rate": 0.0001,
+      "loss": 5.2317,
+      "loss/crossentropy": 2.222583770751953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29915711283683777,
+      "step": 1794
+    },
+    {
+      "epoch": 0.03592,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.032942708333333334,
+      "learning_rate": 0.0001,
+      "loss": 5.4192,
+      "loss/crossentropy": 2.188909649848938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.294817179441452,
+      "step": 1796
+    },
+    {
+      "epoch": 0.03596,
+      "grad_norm": 3.015625,
+      "grad_norm_var": 0.030907185872395833,
+      "learning_rate": 0.0001,
+      "loss": 5.6303,
+      "loss/crossentropy": 2.4745373725891113,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32838208973407745,
+      "step": 1798
+    },
+    {
+      "epoch": 0.036,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 0.028595987955729166,
+      "learning_rate": 0.0001,
+      "loss": 5.3466,
+      "loss/crossentropy": 1.9314215779304504,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29102426767349243,
+      "step": 1800
+    },
+    {
+      "epoch": 0.03604,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.026276652018229166,
+      "learning_rate": 0.0001,
+      "loss": 4.9782,
+      "loss/crossentropy": 2.0099900364875793,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30170293152332306,
+      "step": 1802
+    },
+    {
+      "epoch": 0.03608,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.024388631184895832,
+      "learning_rate": 0.0001,
+      "loss": 5.2506,
+      "loss/crossentropy": 2.1273564100265503,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3019937574863434,
+      "step": 1804
+    },
+    {
+      "epoch": 0.03612,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.016341145833333334,
+      "learning_rate": 0.0001,
+      "loss": 5.1003,
+      "loss/crossentropy": 2.065160095691681,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2817380279302597,
+      "step": 1806
+    },
+    {
+      "epoch": 0.03616,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.018561808268229167,
+      "learning_rate": 0.0001,
+      "loss": 5.1569,
+      "loss/crossentropy": 2.262821078300476,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33245067298412323,
+      "step": 1808
+    },
+    {
+      "epoch": 0.0362,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.020003255208333334,
+      "learning_rate": 0.0001,
+      "loss": 4.8323,
+      "loss/crossentropy": 2.164163827896118,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26008155941963196,
+      "step": 1810
+    },
+    {
+      "epoch": 0.03624,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.012398274739583333,
+      "learning_rate": 0.0001,
+      "loss": 5.2517,
+      "loss/crossentropy": 2.147629737854004,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3073730617761612,
+      "step": 1812
+    },
+    {
+      "epoch": 0.03628,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.023900349934895832,
+      "learning_rate": 0.0001,
+      "loss": 5.1255,
+      "loss/crossentropy": 2.04233980178833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2981158718466759,
+      "step": 1814
+    },
+    {
+      "epoch": 0.03632,
+      "grad_norm": 3.0,
+      "grad_norm_var": 0.026439412434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.9465,
+      "loss/crossentropy": 2.264409065246582,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32124973833560944,
+      "step": 1816
+    },
+    {
+      "epoch": 0.03636,
+      "grad_norm": 3.109375,
+      "grad_norm_var": 0.033610026041666664,
+      "learning_rate": 0.0001,
+      "loss": 5.4324,
+      "loss/crossentropy": 2.092079997062683,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3318764269351959,
+      "step": 1818
+    },
+    {
+      "epoch": 0.0364,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.03850504557291667,
+      "learning_rate": 0.0001,
+      "loss": 5.2766,
+      "loss/crossentropy": 2.1007314324378967,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2930498272180557,
+      "step": 1820
+    },
+    {
+      "epoch": 0.03644,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.03697001139322917,
+      "learning_rate": 0.0001,
+      "loss": 5.129,
+      "loss/crossentropy": 2.2377375960350037,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30473431944847107,
+      "step": 1822
+    },
+    {
+      "epoch": 0.03648,
+      "grad_norm": 3.296875,
+      "grad_norm_var": 0.04684244791666667,
+      "learning_rate": 0.0001,
+      "loss": 5.4209,
+      "loss/crossentropy": 2.0965787172317505,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30526305735111237,
+      "step": 1824
+    },
+    {
+      "epoch": 0.03652,
+      "grad_norm": 3.078125,
+      "grad_norm_var": 0.04755859375,
+      "learning_rate": 0.0001,
+      "loss": 5.0359,
+      "loss/crossentropy": 2.208711862564087,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2727830111980438,
+      "step": 1826
+    },
+    {
+      "epoch": 0.03656,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.050633748372395836,
+      "learning_rate": 0.0001,
+      "loss": 5.1692,
+      "loss/crossentropy": 2.1706738471984863,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3054092824459076,
+      "step": 1828
+    },
+    {
+      "epoch": 0.0366,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.037262980143229166,
+      "learning_rate": 0.0001,
+      "loss": 5.2311,
+      "loss/crossentropy": 2.0891621112823486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3111531287431717,
+      "step": 1830
+    },
+    {
+      "epoch": 0.03664,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.03560791015625,
+      "learning_rate": 0.0001,
+      "loss": 5.0537,
+      "loss/crossentropy": 2.0721256732940674,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30227208137512207,
+      "step": 1832
+    },
+    {
+      "epoch": 0.03668,
+      "grad_norm": 3.015625,
+      "grad_norm_var": 0.03023681640625,
+      "learning_rate": 0.0001,
+      "loss": 5.1267,
+      "loss/crossentropy": 2.2015734910964966,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3185647875070572,
+      "step": 1834
+    },
+    {
+      "epoch": 0.03672,
+      "grad_norm": 3.03125,
+      "grad_norm_var": 0.026493326822916666,
+      "learning_rate": 0.0001,
+      "loss": 5.1973,
+      "loss/crossentropy": 2.1985132694244385,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3032165467739105,
+      "step": 1836
+    },
+    {
+      "epoch": 0.03676,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 0.026334635416666665,
+      "learning_rate": 0.0001,
+      "loss": 5.2041,
+      "loss/crossentropy": 2.170067548751831,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3295029550790787,
+      "step": 1838
+    },
+    {
+      "epoch": 0.0368,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.014615885416666667,
+      "learning_rate": 0.0001,
+      "loss": 5.0049,
+      "loss/crossentropy": 2.113592267036438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2775915116071701,
+      "step": 1840
+    },
+    {
+      "epoch": 0.03684,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.0117828369140625,
+      "learning_rate": 0.0001,
+      "loss": 5.4221,
+      "loss/crossentropy": 2.1905024647712708,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33297547698020935,
+      "step": 1842
+    },
+    {
+      "epoch": 0.03688,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.012919108072916666,
+      "learning_rate": 0.0001,
+      "loss": 5.1859,
+      "loss/crossentropy": 2.252369999885559,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29375749826431274,
+      "step": 1844
+    },
+    {
+      "epoch": 0.03692,
+      "grad_norm": 3.015625,
+      "grad_norm_var": 0.020182291666666668,
+      "learning_rate": 0.0001,
+      "loss": 4.8942,
+      "loss/crossentropy": 1.7526759505271912,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2707225978374481,
+      "step": 1846
+    },
+    {
+      "epoch": 0.03696,
+      "grad_norm": 3.21875,
+      "grad_norm_var": 0.028880818684895834,
+      "learning_rate": 0.0001,
+      "loss": 5.6529,
+      "loss/crossentropy": 2.592544913291931,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3520146906375885,
+      "step": 1848
+    },
+    {
+      "epoch": 0.037,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.0278961181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.9816,
+      "loss/crossentropy": 1.8699345588684082,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2852860391139984,
+      "step": 1850
+    },
+    {
+      "epoch": 0.03704,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.03209228515625,
+      "learning_rate": 0.0001,
+      "loss": 5.1326,
+      "loss/crossentropy": 2.2219313383102417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3210798054933548,
+      "step": 1852
+    },
+    {
+      "epoch": 0.03708,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.0327056884765625,
+      "learning_rate": 0.0001,
+      "loss": 5.2301,
+      "loss/crossentropy": 1.9926818013191223,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26928839832544327,
+      "step": 1854
+    },
+    {
+      "epoch": 0.03712,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.032515462239583334,
+      "learning_rate": 0.0001,
+      "loss": 5.0372,
+      "loss/crossentropy": 2.019917130470276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28046920895576477,
+      "step": 1856
+    },
+    {
+      "epoch": 0.03716,
+      "grad_norm": 4.84375,
+      "grad_norm_var": 0.2762858072916667,
+      "learning_rate": 0.0001,
+      "loss": 5.6297,
+      "loss/crossentropy": 2.2585690021514893,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30598941445350647,
+      "step": 1858
+    },
+    {
+      "epoch": 0.0372,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.27327067057291665,
+      "learning_rate": 0.0001,
+      "loss": 5.3508,
+      "loss/crossentropy": 2.298324942588806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3494870364665985,
+      "step": 1860
+    },
+    {
+      "epoch": 0.03724,
+      "grad_norm": 3.125,
+      "grad_norm_var": 0.25745035807291666,
+      "learning_rate": 0.0001,
+      "loss": 5.54,
+      "loss/crossentropy": 2.430496573448181,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3227449208498001,
+      "step": 1862
+    },
+    {
+      "epoch": 0.03728,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.26183980305989585,
+      "learning_rate": 0.0001,
+      "loss": 5.1876,
+      "loss/crossentropy": 2.090576171875,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2790074646472931,
+      "step": 1864
+    },
+    {
+      "epoch": 0.03732,
+      "grad_norm": 3.0625,
+      "grad_norm_var": 0.25806884765625,
+      "learning_rate": 0.0001,
+      "loss": 5.1799,
+      "loss/crossentropy": 2.2794109582901,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3135898858308792,
+      "step": 1866
+    },
+    {
+      "epoch": 0.03736,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.2597076416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.9333,
+      "loss/crossentropy": 2.2433481216430664,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27558377385139465,
+      "step": 1868
+    },
+    {
+      "epoch": 0.0374,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.26806233723958334,
+      "learning_rate": 0.0001,
+      "loss": 5.292,
+      "loss/crossentropy": 2.3111730813980103,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3172578364610672,
+      "step": 1870
+    },
+    {
+      "epoch": 0.03744,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.270654296875,
+      "learning_rate": 0.0001,
+      "loss": 5.2364,
+      "loss/crossentropy": 2.0028095841407776,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35644619166851044,
+      "step": 1872
+    },
+    {
+      "epoch": 0.03748,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.04501546223958333,
+      "learning_rate": 0.0001,
+      "loss": 5.2183,
+      "loss/crossentropy": 2.347644329071045,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3119680881500244,
+      "step": 1874
+    },
+    {
+      "epoch": 0.03752,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.03474833170572917,
+      "learning_rate": 0.0001,
+      "loss": 5.0787,
+      "loss/crossentropy": 2.118954062461853,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28975334763526917,
+      "step": 1876
+    },
+    {
+      "epoch": 0.03756,
+      "grad_norm": 3.0625,
+      "grad_norm_var": 0.033568318684895834,
+      "learning_rate": 0.0001,
+      "loss": 5.6649,
+      "loss/crossentropy": 2.4376548528671265,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3358190506696701,
+      "step": 1878
+    },
+    {
+      "epoch": 0.0376,
+      "grad_norm": 3.28125,
+      "grad_norm_var": 0.042724609375,
+      "learning_rate": 0.0001,
+      "loss": 5.4924,
+      "loss/crossentropy": 2.5907636880874634,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3234570771455765,
+      "step": 1880
+    },
+    {
+      "epoch": 0.03764,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.04352925618489583,
+      "learning_rate": 0.0001,
+      "loss": 5.0298,
+      "loss/crossentropy": 1.574956238269806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26118964701890945,
+      "step": 1882
+    },
+    {
+      "epoch": 0.03768,
+      "grad_norm": 3.203125,
+      "grad_norm_var": 0.04599609375,
+      "learning_rate": 0.0001,
+      "loss": 5.3339,
+      "loss/crossentropy": 2.3571736812591553,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.308853879570961,
+      "step": 1884
+    },
+    {
+      "epoch": 0.03772,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.03877665201822917,
+      "learning_rate": 0.0001,
+      "loss": 5.0807,
+      "loss/crossentropy": 2.1266958117485046,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.307782918214798,
+      "step": 1886
+    },
+    {
+      "epoch": 0.03776,
+      "grad_norm": 3.0,
+      "grad_norm_var": 0.03876546223958333,
+      "learning_rate": 0.0001,
+      "loss": 5.5402,
+      "loss/crossentropy": 2.3529077768325806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32287272810935974,
+      "step": 1888
+    },
+    {
+      "epoch": 0.0378,
+      "grad_norm": 3.015625,
+      "grad_norm_var": 0.03905843098958333,
+      "learning_rate": 0.0001,
+      "loss": 5.4555,
+      "loss/crossentropy": 2.277345299720764,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32319171726703644,
+      "step": 1890
+    },
+    {
+      "epoch": 0.03784,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.035008748372395836,
+      "learning_rate": 0.0001,
+      "loss": 5.01,
+      "loss/crossentropy": 2.102940857410431,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.292771652340889,
+      "step": 1892
+    },
+    {
+      "epoch": 0.03788,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.03998921712239583,
+      "learning_rate": 0.0001,
+      "loss": 5.0784,
+      "loss/crossentropy": 1.9744033813476562,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2844446450471878,
+      "step": 1894
+    },
+    {
+      "epoch": 0.03792,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.03250325520833333,
+      "learning_rate": 0.0001,
+      "loss": 4.9859,
+      "loss/crossentropy": 1.8222022652626038,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28611525893211365,
+      "step": 1896
+    },
+    {
+      "epoch": 0.03796,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.03135477701822917,
+      "learning_rate": 0.0001,
+      "loss": 5.0704,
+      "loss/crossentropy": 2.1963966488838196,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2988738566637039,
+      "step": 1898
+    },
+    {
+      "epoch": 0.038,
+      "grad_norm": 3.046875,
+      "grad_norm_var": 0.025927734375,
+      "learning_rate": 0.0001,
+      "loss": 5.1155,
+      "loss/crossentropy": 1.9982789754867554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.312205046415329,
+      "step": 1900
+    },
+    {
+      "epoch": 0.03804,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.020612589518229165,
+      "learning_rate": 0.0001,
+      "loss": 5.2097,
+      "loss/crossentropy": 2.1999258995056152,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29916079342365265,
+      "step": 1902
+    },
+    {
+      "epoch": 0.03808,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.016999308268229166,
+      "learning_rate": 0.0001,
+      "loss": 5.2233,
+      "loss/crossentropy": 2.0725532174110413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3093830645084381,
+      "step": 1904
+    },
+    {
+      "epoch": 0.03812,
+      "grad_norm": 3.03125,
+      "grad_norm_var": 0.027079264322916668,
+      "learning_rate": 0.0001,
+      "loss": 5.7014,
+      "loss/crossentropy": 2.2504276037216187,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.332836389541626,
+      "step": 1906
+    },
+    {
+      "epoch": 0.03816,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.026927693684895834,
+      "learning_rate": 0.0001,
+      "loss": 5.3413,
+      "loss/crossentropy": 2.1570577025413513,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3210139572620392,
+      "step": 1908
+    },
+    {
+      "epoch": 0.0382,
+      "grad_norm": 3.59375,
+      "grad_norm_var": 0.05137430826822917,
+      "learning_rate": 0.0001,
+      "loss": 5.6215,
+      "loss/crossentropy": 2.0739041566848755,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3091907352209091,
+      "step": 1910
+    },
+    {
+      "epoch": 0.03824,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.0502838134765625,
+      "learning_rate": 0.0001,
+      "loss": 5.0772,
+      "loss/crossentropy": 2.0542168021202087,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2791624963283539,
+      "step": 1912
+    },
+    {
+      "epoch": 0.03828,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.0561431884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.9386,
+      "loss/crossentropy": 1.9705287218093872,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25483617186546326,
+      "step": 1914
+    },
+    {
+      "epoch": 0.03832,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.05819905598958333,
+      "learning_rate": 0.0001,
+      "loss": 5.054,
+      "loss/crossentropy": 2.0234111547470093,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3110152333974838,
+      "step": 1916
+    },
+    {
+      "epoch": 0.03836,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.06004130045572917,
+      "learning_rate": 0.0001,
+      "loss": 5.2723,
+      "loss/crossentropy": 2.010735273361206,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2943515181541443,
+      "step": 1918
+    },
+    {
+      "epoch": 0.0384,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.06575419108072916,
+      "learning_rate": 0.0001,
+      "loss": 4.9471,
+      "loss/crossentropy": 2.1912686824798584,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27914971113204956,
+      "step": 1920
+    },
+    {
+      "epoch": 0.03844,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.0709869384765625,
+      "learning_rate": 0.0001,
+      "loss": 5.085,
+      "loss/crossentropy": 1.9889940023422241,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26166442036628723,
+      "step": 1922
+    },
+    {
+      "epoch": 0.03848,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.07668863932291667,
+      "learning_rate": 0.0001,
+      "loss": 5.3534,
+      "loss/crossentropy": 2.154898941516876,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31905338168144226,
+      "step": 1924
+    },
+    {
+      "epoch": 0.03852,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.03997395833333333,
+      "learning_rate": 0.0001,
+      "loss": 5.0847,
+      "loss/crossentropy": 2.44269061088562,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28736811876296997,
+      "step": 1926
+    },
+    {
+      "epoch": 0.03856,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.04108784993489583,
+      "learning_rate": 0.0001,
+      "loss": 5.0585,
+      "loss/crossentropy": 1.6790328621864319,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32607489824295044,
+      "step": 1928
+    },
+    {
+      "epoch": 0.0386,
+      "grad_norm": 3.046875,
+      "grad_norm_var": 0.05115559895833333,
+      "learning_rate": 0.0001,
+      "loss": 5.336,
+      "loss/crossentropy": 2.0223641991615295,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2871186435222626,
+      "step": 1930
+    },
+    {
+      "epoch": 0.03864,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.05756734212239583,
+      "learning_rate": 0.0001,
+      "loss": 5.549,
+      "loss/crossentropy": 2.451051712036133,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33973076939582825,
+      "step": 1932
+    },
+    {
+      "epoch": 0.03868,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.052262369791666666,
+      "learning_rate": 0.0001,
+      "loss": 5.4403,
+      "loss/crossentropy": 2.2884862422943115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29729554057121277,
+      "step": 1934
+    },
+    {
+      "epoch": 0.03872,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 0.04889322916666667,
+      "learning_rate": 0.0001,
+      "loss": 5.6203,
+      "loss/crossentropy": 2.113345444202423,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36945630609989166,
+      "step": 1936
+    },
+    {
+      "epoch": 0.03876,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.03752848307291667,
+      "learning_rate": 0.0001,
+      "loss": 5.1845,
+      "loss/crossentropy": 2.139409840106964,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2899101823568344,
+      "step": 1938
+    },
+    {
+      "epoch": 0.0388,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.03622639973958333,
+      "learning_rate": 0.0001,
+      "loss": 5.5397,
+      "loss/crossentropy": 2.1029305458068848,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29830390214920044,
+      "step": 1940
+    },
+    {
+      "epoch": 0.03884,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.03816630045572917,
+      "learning_rate": 0.0001,
+      "loss": 5.1383,
+      "loss/crossentropy": 1.7736502885818481,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27648696303367615,
+      "step": 1942
+    },
+    {
+      "epoch": 0.03888,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.03942057291666667,
+      "learning_rate": 0.0001,
+      "loss": 5.0885,
+      "loss/crossentropy": 2.0281469225883484,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30329641699790955,
+      "step": 1944
+    },
+    {
+      "epoch": 0.03892,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 0.03528544108072917,
+      "learning_rate": 0.0001,
+      "loss": 5.3265,
+      "loss/crossentropy": 2.404891610145569,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3563212752342224,
+      "step": 1946
+    },
+    {
+      "epoch": 0.03896,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.03572489420572917,
+      "learning_rate": 0.0001,
+      "loss": 5.0657,
+      "loss/crossentropy": 2.2187922596931458,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2999056503176689,
+      "step": 1948
+    },
+    {
+      "epoch": 0.039,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.03566792805989583,
+      "learning_rate": 0.0001,
+      "loss": 5.0499,
+      "loss/crossentropy": 2.3901994228363037,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3288661539554596,
+      "step": 1950
+    },
+    {
+      "epoch": 0.03904,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.04069722493489583,
+      "learning_rate": 0.0001,
+      "loss": 5.252,
+      "loss/crossentropy": 2.274617314338684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2928028404712677,
+      "step": 1952
+    },
+    {
+      "epoch": 0.03908,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.03430582682291667,
+      "learning_rate": 0.0001,
+      "loss": 5.4463,
+      "loss/crossentropy": 2.2478950023651123,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3027127981185913,
+      "step": 1954
+    },
+    {
+      "epoch": 0.03912,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.030321248372395835,
+      "learning_rate": 0.0001,
+      "loss": 5.2707,
+      "loss/crossentropy": 2.0634876489639282,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29374830424785614,
+      "step": 1956
+    },
+    {
+      "epoch": 0.03916,
+      "grad_norm": 2.90625,
+      "grad_norm_var": 0.0287994384765625,
+      "learning_rate": 0.0001,
+      "loss": 5.3441,
+      "loss/crossentropy": 2.171326994895935,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2995557487010956,
+      "step": 1958
+    },
+    {
+      "epoch": 0.0392,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.040266927083333334,
+      "learning_rate": 0.0001,
+      "loss": 4.9937,
+      "loss/crossentropy": 2.142563223838806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27376608550548553,
+      "step": 1960
+    },
+    {
+      "epoch": 0.03924,
+      "grad_norm": 5.34375,
+      "grad_norm_var": 0.42568359375,
+      "learning_rate": 0.0001,
+      "loss": 5.4145,
+      "loss/crossentropy": 2.3189245462417603,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37109237909317017,
+      "step": 1962
+    },
+    {
+      "epoch": 0.03928,
+      "grad_norm": 3.203125,
+      "grad_norm_var": 0.40812886555989586,
+      "learning_rate": 0.0001,
+      "loss": 5.1826,
+      "loss/crossentropy": 2.1483139991760254,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32499393820762634,
+      "step": 1964
+    },
+    {
+      "epoch": 0.03932,
+      "grad_norm": 3.046875,
+      "grad_norm_var": 0.404736328125,
+      "learning_rate": 0.0001,
+      "loss": 5.445,
+      "loss/crossentropy": 2.2916383743286133,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29367291927337646,
+      "step": 1966
+    },
+    {
+      "epoch": 0.03936,
+      "grad_norm": 2.9375,
+      "grad_norm_var": 0.3999582926432292,
+      "learning_rate": 0.0001,
+      "loss": 5.1663,
+      "loss/crossentropy": 2.4217371940612793,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31309331953525543,
+      "step": 1968
+    },
+    {
+      "epoch": 0.0394,
+      "grad_norm": 2.9375,
+      "grad_norm_var": 0.40103759765625,
+      "learning_rate": 0.0001,
+      "loss": 5.2057,
+      "loss/crossentropy": 1.9491975903511047,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29921913146972656,
+      "step": 1970
+    },
+    {
+      "epoch": 0.03944,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.39126688639322915,
+      "learning_rate": 0.0001,
+      "loss": 5.1788,
+      "loss/crossentropy": 2.144432306289673,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29162150621414185,
+      "step": 1972
+    },
+    {
+      "epoch": 0.03948,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.40748291015625,
+      "learning_rate": 0.0001,
+      "loss": 5.1336,
+      "loss/crossentropy": 1.9492529034614563,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2875414192676544,
+      "step": 1974
+    },
+    {
+      "epoch": 0.03952,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.4002593994140625,
+      "learning_rate": 0.0001,
+      "loss": 5.053,
+      "loss/crossentropy": 1.9269813895225525,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2755381464958191,
+      "step": 1976
+    },
+    {
+      "epoch": 0.03956,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.03798726399739583,
+      "learning_rate": 0.0001,
+      "loss": 4.8879,
+      "loss/crossentropy": 2.074360489845276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2987503558397293,
+      "step": 1978
+    },
+    {
+      "epoch": 0.0396,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.02086181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.8834,
+      "loss/crossentropy": 2.257633090019226,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2992274910211563,
+      "step": 1980
+    },
+    {
+      "epoch": 0.03964,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.01802978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.9533,
+      "loss/crossentropy": 1.8207083940505981,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2752758115530014,
+      "step": 1982
+    },
+    {
+      "epoch": 0.03968,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.015623982747395833,
+      "learning_rate": 0.0001,
+      "loss": 5.343,
+      "loss/crossentropy": 2.105292797088623,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3131226450204849,
+      "step": 1984
+    },
+    {
+      "epoch": 0.03972,
+      "grad_norm": 3.484375,
+      "grad_norm_var": 0.05191650390625,
+      "learning_rate": 0.0001,
+      "loss": 5.4785,
+      "loss/crossentropy": 2.1191373467445374,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3175530731678009,
+      "step": 1986
+    },
+    {
+      "epoch": 0.03976,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.051878865559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.9236,
+      "loss/crossentropy": 2.2214397192001343,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3091724067926407,
+      "step": 1988
+    },
+    {
+      "epoch": 0.0398,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.05133056640625,
+      "learning_rate": 0.0001,
+      "loss": 5.0031,
+      "loss/crossentropy": 1.7347424626350403,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2825440764427185,
+      "step": 1990
+    },
+    {
+      "epoch": 0.03984,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 0.0500396728515625,
+      "learning_rate": 0.0001,
+      "loss": 5.0951,
+      "loss/crossentropy": 2.1566559076309204,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2961925268173218,
+      "step": 1992
+    },
+    {
+      "epoch": 0.03988,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.06396077473958334,
+      "learning_rate": 0.0001,
+      "loss": 4.9195,
+      "loss/crossentropy": 2.2129205465316772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29672613739967346,
+      "step": 1994
+    },
+    {
+      "epoch": 0.03992,
+      "grad_norm": 3.3125,
+      "grad_norm_var": 0.07595113118489584,
+      "learning_rate": 0.0001,
+      "loss": 5.7534,
+      "loss/crossentropy": 2.4702744483947754,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3707122802734375,
+      "step": 1996
+    },
+    {
+      "epoch": 0.03996,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.06731669108072917,
+      "learning_rate": 0.0001,
+      "loss": 5.0296,
+      "loss/crossentropy": 2.0463536977767944,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3124735355377197,
+      "step": 1998
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.07281494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.959,
+      "loss/crossentropy": 2.1550235748291016,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32177163660526276,
+      "step": 2000
+    },
+    {
+      "epoch": 0.04004,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.059798177083333334,
+      "learning_rate": 0.0001,
+      "loss": 5.2078,
+      "loss/crossentropy": 2.1312190890312195,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29013920575380325,
+      "step": 2002
+    },
+    {
+      "epoch": 0.04008,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.06082356770833333,
+      "learning_rate": 0.0001,
+      "loss": 5.0086,
+      "loss/crossentropy": 1.8546085357666016,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2619543671607971,
+      "step": 2004
+    },
+    {
+      "epoch": 0.04012,
+      "grad_norm": 3.015625,
+      "grad_norm_var": 0.0561920166015625,
+      "learning_rate": 0.0001,
+      "loss": 5.3416,
+      "loss/crossentropy": 2.262398660182953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28742220997810364,
+      "step": 2006
+    },
+    {
+      "epoch": 0.04016,
+      "grad_norm": 3.109375,
+      "grad_norm_var": 0.05734049479166667,
+      "learning_rate": 0.0001,
+      "loss": 5.4315,
+      "loss/crossentropy": 2.156043767929077,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30488699674606323,
+      "step": 2008
+    },
+    {
+      "epoch": 0.0402,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.04290364583333333,
+      "learning_rate": 0.0001,
+      "loss": 4.8761,
+      "loss/crossentropy": 1.925516963005066,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2795708477497101,
+      "step": 2010
+    },
+    {
+      "epoch": 0.04024,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.031281534830729166,
+      "learning_rate": 0.0001,
+      "loss": 5.0729,
+      "loss/crossentropy": 1.947714388370514,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27192793786525726,
+      "step": 2012
+    },
+    {
+      "epoch": 0.04028,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.038939412434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.6214,
+      "loss/crossentropy": 1.9584010243415833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2835424840450287,
+      "step": 2014
+    },
+    {
+      "epoch": 0.04032,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.04108784993489583,
+      "learning_rate": 0.0001,
+      "loss": 5.1974,
+      "loss/crossentropy": 2.461808919906616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3104698956012726,
+      "step": 2016
+    },
+    {
+      "epoch": 0.04036,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.021903483072916667,
+      "learning_rate": 0.0001,
+      "loss": 5.286,
+      "loss/crossentropy": 2.094545900821686,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30010756850242615,
+      "step": 2018
+    },
+    {
+      "epoch": 0.0404,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.021923828125,
+      "learning_rate": 0.0001,
+      "loss": 5.1274,
+      "loss/crossentropy": 2.353589177131653,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29775144159793854,
+      "step": 2020
+    },
+    {
+      "epoch": 0.04044,
+      "grad_norm": 3.375,
+      "grad_norm_var": 10.55537821451823,
+      "learning_rate": 0.0001,
+      "loss": 5.3359,
+      "loss/crossentropy": 2.4468252658843994,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34534354507923126,
+      "step": 2022
+    },
+    {
+      "epoch": 0.04048,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 10.529678344726562,
+      "learning_rate": 0.0001,
+      "loss": 5.5028,
+      "loss/crossentropy": 2.2037755250930786,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.303710475564003,
+      "step": 2024
+    },
+    {
+      "epoch": 0.04052,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 10.546240234375,
+      "learning_rate": 0.0001,
+      "loss": 4.9229,
+      "loss/crossentropy": 1.9658318161964417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2829178273677826,
+      "step": 2026
+    },
+    {
+      "epoch": 0.04056,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 10.555729166666667,
+      "learning_rate": 0.0001,
+      "loss": 4.8996,
+      "loss/crossentropy": 2.118351697921753,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29469528794288635,
+      "step": 2028
+    },
+    {
+      "epoch": 0.0406,
+      "grad_norm": 2.875,
+      "grad_norm_var": 10.507957967122396,
+      "learning_rate": 0.0001,
+      "loss": 5.5817,
+      "loss/crossentropy": 2.172826111316681,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35036201775074005,
+      "step": 2030
+    },
+    {
+      "epoch": 0.04064,
+      "grad_norm": 3.953125,
+      "grad_norm_var": 10.432124837239583,
+      "learning_rate": 0.0001,
+      "loss": 5.4846,
+      "loss/crossentropy": 2.185975730419159,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3397497236728668,
+      "step": 2032
+    },
+    {
+      "epoch": 0.04068,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 10.411026000976562,
+      "learning_rate": 0.0001,
+      "loss": 5.0145,
+      "loss/crossentropy": 2.043874442577362,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3034070134162903,
+      "step": 2034
+    },
+    {
+      "epoch": 0.04072,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 10.380106608072916,
+      "learning_rate": 0.0001,
+      "loss": 5.3958,
+      "loss/crossentropy": 2.3315287828445435,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32198067009449005,
+      "step": 2036
+    },
+    {
+      "epoch": 0.04076,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.1112213134765625,
+      "learning_rate": 0.0001,
+      "loss": 5.0266,
+      "loss/crossentropy": 2.05656898021698,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30919137597084045,
+      "step": 2038
+    },
+    {
+      "epoch": 0.0408,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.10188395182291667,
+      "learning_rate": 0.0001,
+      "loss": 5.3205,
+      "loss/crossentropy": 2.2451635599136353,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3412973880767822,
+      "step": 2040
+    },
+    {
+      "epoch": 0.04084,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.10715738932291667,
+      "learning_rate": 0.0001,
+      "loss": 5.1734,
+      "loss/crossentropy": 2.527924060821533,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3178148865699768,
+      "step": 2042
+    },
+    {
+      "epoch": 0.04088,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.10305582682291667,
+      "learning_rate": 0.0001,
+      "loss": 4.8441,
+      "loss/crossentropy": 2.03126460313797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29549214243888855,
+      "step": 2044
+    },
+    {
+      "epoch": 0.04092,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.11038004557291667,
+      "learning_rate": 0.0001,
+      "loss": 5.0134,
+      "loss/crossentropy": 2.029997706413269,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27938202023506165,
+      "step": 2046
+    },
+    {
+      "epoch": 0.04096,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.03211263020833333,
+      "learning_rate": 0.0001,
+      "loss": 4.9321,
+      "loss/crossentropy": 1.764098048210144,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27712512016296387,
+      "step": 2048
+    },
+    {
+      "epoch": 0.041,
+      "grad_norm": 3.046875,
+      "grad_norm_var": 0.0356597900390625,
+      "learning_rate": 0.0001,
+      "loss": 5.435,
+      "loss/crossentropy": 2.605324864387512,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3540754020214081,
+      "step": 2050
+    },
+    {
+      "epoch": 0.04104,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.0349761962890625,
+      "learning_rate": 0.0001,
+      "loss": 5.0207,
+      "loss/crossentropy": 1.9333613514900208,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29670488089323044,
+      "step": 2052
+    },
+    {
+      "epoch": 0.04108,
+      "grad_norm": 3.078125,
+      "grad_norm_var": 0.040022786458333334,
+      "learning_rate": 0.0001,
+      "loss": 5.0056,
+      "loss/crossentropy": 1.7876797914505005,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2579014301300049,
+      "step": 2054
+    },
+    {
+      "epoch": 0.04112,
+      "grad_norm": 3.125,
+      "grad_norm_var": 0.0417877197265625,
+      "learning_rate": 0.0001,
+      "loss": 5.1401,
+      "loss/crossentropy": 2.0947588682174683,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3334304690361023,
+      "step": 2056
+    },
+    {
+      "epoch": 0.04116,
+      "grad_norm": 3.0,
+      "grad_norm_var": 0.035965983072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.9009,
+      "loss/crossentropy": 2.1838767528533936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.306907519698143,
+      "step": 2058
+    },
+    {
+      "epoch": 0.0412,
+      "grad_norm": 3.015625,
+      "grad_norm_var": 0.0341461181640625,
+      "learning_rate": 0.0001,
+      "loss": 5.3724,
+      "loss/crossentropy": 2.2180997133255005,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.314627081155777,
+      "step": 2060
+    },
+    {
+      "epoch": 0.04124,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.025419108072916665,
+      "learning_rate": 0.0001,
+      "loss": 4.8362,
+      "loss/crossentropy": 1.914646863937378,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2848198413848877,
+      "step": 2062
+    },
+    {
+      "epoch": 0.04128,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.0262603759765625,
+      "learning_rate": 0.0001,
+      "loss": 5.4538,
+      "loss/crossentropy": 2.42458713054657,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31268230080604553,
+      "step": 2064
+    },
+    {
+      "epoch": 0.04132,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.02457275390625,
+      "learning_rate": 0.0001,
+      "loss": 5.2497,
+      "loss/crossentropy": 2.23202121257782,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30988384783267975,
+      "step": 2066
+    },
+    {
+      "epoch": 0.04136,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.0284820556640625,
+      "learning_rate": 0.0001,
+      "loss": 5.0416,
+      "loss/crossentropy": 2.0225483179092407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26953594386577606,
+      "step": 2068
+    },
+    {
+      "epoch": 0.0414,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.028669230143229165,
+      "learning_rate": 0.0001,
+      "loss": 4.8259,
+      "loss/crossentropy": 1.8593338131904602,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27706706523895264,
+      "step": 2070
+    },
+    {
+      "epoch": 0.04144,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.023387654622395834,
+      "learning_rate": 0.0001,
+      "loss": 4.9949,
+      "loss/crossentropy": 2.373727560043335,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3000074476003647,
+      "step": 2072
+    },
+    {
+      "epoch": 0.04148,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.022704060872395834,
+      "learning_rate": 0.0001,
+      "loss": 4.9438,
+      "loss/crossentropy": 1.959564983844757,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26365046203136444,
+      "step": 2074
+    },
+    {
+      "epoch": 0.04152,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.014159138997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.983,
+      "loss/crossentropy": 2.0590676069259644,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2816864550113678,
+      "step": 2076
+    },
+    {
+      "epoch": 0.04156,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.017121378580729166,
+      "learning_rate": 0.0001,
+      "loss": 5.2049,
+      "loss/crossentropy": 2.147680163383484,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29019051790237427,
+      "step": 2078
+    },
+    {
+      "epoch": 0.0416,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.020198567708333334,
+      "learning_rate": 0.0001,
+      "loss": 5.6103,
+      "loss/crossentropy": 2.212267220020294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2804659754037857,
+      "step": 2080
+    },
+    {
+      "epoch": 0.04164,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.020406087239583332,
+      "learning_rate": 0.0001,
+      "loss": 5.0489,
+      "loss/crossentropy": 2.144743025302887,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2946523129940033,
+      "step": 2082
+    },
+    {
+      "epoch": 0.04168,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.019172159830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.9311,
+      "loss/crossentropy": 2.3702481985092163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2923210561275482,
+      "step": 2084
+    },
+    {
+      "epoch": 0.04172,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.019684855143229166,
+      "learning_rate": 0.0001,
+      "loss": 5.2291,
+      "loss/crossentropy": 1.9512975811958313,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28704553842544556,
+      "step": 2086
+    },
+    {
+      "epoch": 0.04176,
+      "grad_norm": 3.25,
+      "grad_norm_var": 0.25388895670572914,
+      "learning_rate": 0.0001,
+      "loss": 4.9572,
+      "loss/crossentropy": 2.180745005607605,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28361976146698,
+      "step": 2088
+    },
+    {
+      "epoch": 0.0418,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.24712626139322916,
+      "learning_rate": 0.0001,
+      "loss": 4.8579,
+      "loss/crossentropy": 1.9768275022506714,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28042787313461304,
+      "step": 2090
+    },
+    {
+      "epoch": 0.04184,
+      "grad_norm": 2.90625,
+      "grad_norm_var": 0.24544270833333334,
+      "learning_rate": 0.0001,
+      "loss": 5.2602,
+      "loss/crossentropy": 2.148472547531128,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30099035799503326,
+      "step": 2092
+    },
+    {
+      "epoch": 0.04188,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.24763081868489584,
+      "learning_rate": 0.0001,
+      "loss": 5.0152,
+      "loss/crossentropy": 2.1698715686798096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3140450567007065,
+      "step": 2094
+    },
+    {
+      "epoch": 0.04192,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.24172770182291667,
+      "learning_rate": 0.0001,
+      "loss": 4.8679,
+      "loss/crossentropy": 2.1142334938049316,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2847675681114197,
+      "step": 2096
+    },
+    {
+      "epoch": 0.04196,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.2395660400390625,
+      "learning_rate": 0.0001,
+      "loss": 5.2185,
+      "loss/crossentropy": 2.1908479928970337,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28702451288700104,
+      "step": 2098
+    },
+    {
+      "epoch": 0.042,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.23321024576822916,
+      "learning_rate": 0.0001,
+      "loss": 5.0212,
+      "loss/crossentropy": 2.0519612431526184,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29224735498428345,
+      "step": 2100
+    },
+    {
+      "epoch": 0.04204,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.2412994384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.871,
+      "loss/crossentropy": 1.9304961562156677,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28785137832164764,
+      "step": 2102
+    },
+    {
+      "epoch": 0.04208,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.0134185791015625,
+      "learning_rate": 0.0001,
+      "loss": 5.2462,
+      "loss/crossentropy": 2.297300934791565,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.302143856883049,
+      "step": 2104
+    },
+    {
+      "epoch": 0.04212,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.019782511393229167,
+      "learning_rate": 0.0001,
+      "loss": 5.0491,
+      "loss/crossentropy": 2.2764381170272827,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28773219883441925,
+      "step": 2106
+    },
+    {
+      "epoch": 0.04216,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.019391886393229165,
+      "learning_rate": 0.0001,
+      "loss": 5.0563,
+      "loss/crossentropy": 2.141321837902069,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3108212947845459,
+      "step": 2108
+    },
+    {
+      "epoch": 0.0422,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.0185699462890625,
+      "learning_rate": 0.0001,
+      "loss": 5.0362,
+      "loss/crossentropy": 1.9619495272636414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2938811331987381,
+      "step": 2110
+    },
+    {
+      "epoch": 0.04224,
+      "grad_norm": 3.015625,
+      "grad_norm_var": 0.026325480143229166,
+      "learning_rate": 0.0001,
+      "loss": 5.4496,
+      "loss/crossentropy": 1.9741051197052002,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28835102915763855,
+      "step": 2112
+    },
+    {
+      "epoch": 0.04228,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.029255167643229166,
+      "learning_rate": 0.0001,
+      "loss": 4.9303,
+      "loss/crossentropy": 1.9510936737060547,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2839510589838028,
+      "step": 2114
+    },
+    {
+      "epoch": 0.04232,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.026753743489583332,
+      "learning_rate": 0.0001,
+      "loss": 5.2446,
+      "loss/crossentropy": 2.0201885104179382,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30627067387104034,
+      "step": 2116
+    },
+    {
+      "epoch": 0.04236,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.022493489583333335,
+      "learning_rate": 0.0001,
+      "loss": 5.1411,
+      "loss/crossentropy": 2.4522262811660767,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.308579683303833,
+      "step": 2118
+    },
+    {
+      "epoch": 0.0424,
+      "grad_norm": 3.09375,
+      "grad_norm_var": 0.04345296223958333,
+      "learning_rate": 0.0001,
+      "loss": 5.5535,
+      "loss/crossentropy": 1.9289590120315552,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29346515238285065,
+      "step": 2120
+    },
+    {
+      "epoch": 0.04244,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.03437398274739583,
+      "learning_rate": 0.0001,
+      "loss": 5.0588,
+      "loss/crossentropy": 2.2020061016082764,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2992282509803772,
+      "step": 2122
+    },
+    {
+      "epoch": 0.04248,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.031494140625,
+      "learning_rate": 0.0001,
+      "loss": 5.2466,
+      "loss/crossentropy": 2.180301785469055,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28323256969451904,
+      "step": 2124
+    },
+    {
+      "epoch": 0.04252,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.03435872395833333,
+      "learning_rate": 0.0001,
+      "loss": 4.9061,
+      "loss/crossentropy": 2.1250513792037964,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.272533118724823,
+      "step": 2126
+    },
+    {
+      "epoch": 0.04256,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.037873331705729166,
+      "learning_rate": 0.0001,
+      "loss": 5.4375,
+      "loss/crossentropy": 2.3509981632232666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3719516545534134,
+      "step": 2128
+    },
+    {
+      "epoch": 0.0426,
+      "grad_norm": 3.078125,
+      "grad_norm_var": 0.03508707682291667,
+      "learning_rate": 0.0001,
+      "loss": 5.3067,
+      "loss/crossentropy": 2.135426163673401,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3582882583141327,
+      "step": 2130
+    },
+    {
+      "epoch": 0.04264,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.0398590087890625,
+      "learning_rate": 0.0001,
+      "loss": 5.2406,
+      "loss/crossentropy": 2.316452383995056,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30162203311920166,
+      "step": 2132
+    },
+    {
+      "epoch": 0.04268,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.03846028645833333,
+      "learning_rate": 0.0001,
+      "loss": 5.0372,
+      "loss/crossentropy": 2.0325432419776917,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.287256121635437,
+      "step": 2134
+    },
+    {
+      "epoch": 0.04272,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.0236236572265625,
+      "learning_rate": 0.0001,
+      "loss": 5.1985,
+      "loss/crossentropy": 2.070056974887848,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2552843391895294,
+      "step": 2136
+    },
+    {
+      "epoch": 0.04276,
+      "grad_norm": 3.078125,
+      "grad_norm_var": 0.029157511393229165,
+      "learning_rate": 0.0001,
+      "loss": 5.0623,
+      "loss/crossentropy": 1.7005944848060608,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24984879791736603,
+      "step": 2138
+    },
+    {
+      "epoch": 0.0428,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.03203125,
+      "learning_rate": 0.0001,
+      "loss": 5.0862,
+      "loss/crossentropy": 1.6700931787490845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2572527676820755,
+      "step": 2140
+    },
+    {
+      "epoch": 0.04284,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.03125,
+      "learning_rate": 0.0001,
+      "loss": 5.0186,
+      "loss/crossentropy": 2.3074774742126465,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31137382984161377,
+      "step": 2142
+    },
+    {
+      "epoch": 0.04288,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.0229400634765625,
+      "learning_rate": 0.0001,
+      "loss": 5.1973,
+      "loss/crossentropy": 2.103408098220825,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30157215893268585,
+      "step": 2144
+    },
+    {
+      "epoch": 0.04292,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.020979817708333334,
+      "learning_rate": 0.0001,
+      "loss": 4.8206,
+      "loss/crossentropy": 1.8602584600448608,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28300634026527405,
+      "step": 2146
+    },
+    {
+      "epoch": 0.04296,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.019188435872395833,
+      "learning_rate": 0.0001,
+      "loss": 5.0525,
+      "loss/crossentropy": 2.337582588195801,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28086431324481964,
+      "step": 2148
+    },
+    {
+      "epoch": 0.043,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.026439412434895834,
+      "learning_rate": 0.0001,
+      "loss": 5.2405,
+      "loss/crossentropy": 2.2635254859924316,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3113311231136322,
+      "step": 2150
+    },
+    {
+      "epoch": 0.04304,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.0277984619140625,
+      "learning_rate": 0.0001,
+      "loss": 5.187,
+      "loss/crossentropy": 2.3971948623657227,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32285284996032715,
+      "step": 2152
+    },
+    {
+      "epoch": 0.04308,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.021891276041666668,
+      "learning_rate": 0.0001,
+      "loss": 5.1438,
+      "loss/crossentropy": 1.8900776505470276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28163351118564606,
+      "step": 2154
+    },
+    {
+      "epoch": 0.04312,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.020536295572916665,
+      "learning_rate": 0.0001,
+      "loss": 4.9774,
+      "loss/crossentropy": 1.908443808555603,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2901918590068817,
+      "step": 2156
+    },
+    {
+      "epoch": 0.04316,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.020409138997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.7808,
+      "loss/crossentropy": 1.8003268837928772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2762569487094879,
+      "step": 2158
+    },
+    {
+      "epoch": 0.0432,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.0221832275390625,
+      "learning_rate": 0.0001,
+      "loss": 5.0477,
+      "loss/crossentropy": 1.996739387512207,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24626458436250687,
+      "step": 2160
+    },
+    {
+      "epoch": 0.04324,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.018880208333333332,
+      "learning_rate": 0.0001,
+      "loss": 5.1737,
+      "loss/crossentropy": 2.0175461173057556,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31059183180332184,
+      "step": 2162
+    },
+    {
+      "epoch": 0.04328,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.019266764322916668,
+      "learning_rate": 0.0001,
+      "loss": 5.0448,
+      "loss/crossentropy": 2.0009909868240356,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2835536003112793,
+      "step": 2164
+    },
+    {
+      "epoch": 0.04332,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.017967732747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.9035,
+      "loss/crossentropy": 1.9848785400390625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26657119393348694,
+      "step": 2166
+    },
+    {
+      "epoch": 0.04336,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.01207275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.9108,
+      "loss/crossentropy": 2.076065957546234,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2801993191242218,
+      "step": 2168
+    },
+    {
+      "epoch": 0.0434,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.011617024739583334,
+      "learning_rate": 0.0001,
+      "loss": 5.1425,
+      "loss/crossentropy": 2.1208528876304626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3152369260787964,
+      "step": 2170
+    },
+    {
+      "epoch": 0.04344,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.011400349934895833,
+      "learning_rate": 0.0001,
+      "loss": 5.0608,
+      "loss/crossentropy": 2.1971306204795837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31046128273010254,
+      "step": 2172
+    },
+    {
+      "epoch": 0.04348,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.013158162434895834,
+      "learning_rate": 0.0001,
+      "loss": 5.2445,
+      "loss/crossentropy": 2.275176525115967,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3090529441833496,
+      "step": 2174
+    },
+    {
+      "epoch": 0.04352,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.01197509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.9366,
+      "loss/crossentropy": 2.1574501395225525,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.280165433883667,
+      "step": 2176
+    },
+    {
+      "epoch": 0.04356,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.012548828125,
+      "learning_rate": 0.0001,
+      "loss": 5.2338,
+      "loss/crossentropy": 2.4236754179000854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3061629384756088,
+      "step": 2178
+    },
+    {
+      "epoch": 0.0436,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.012809244791666667,
+      "learning_rate": 0.0001,
+      "loss": 5.1707,
+      "loss/crossentropy": 2.1282758712768555,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3050261586904526,
+      "step": 2180
+    },
+    {
+      "epoch": 0.04364,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.008561197916666667,
+      "learning_rate": 0.0001,
+      "loss": 5.4629,
+      "loss/crossentropy": 2.4244707822799683,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33464157581329346,
+      "step": 2182
+    },
+    {
+      "epoch": 0.04368,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.010904947916666666,
+      "learning_rate": 0.0001,
+      "loss": 5.3891,
+      "loss/crossentropy": 2.289917469024658,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3181813210248947,
+      "step": 2184
+    },
+    {
+      "epoch": 0.04372,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 0.01031494140625,
+      "learning_rate": 0.0001,
+      "loss": 5.4207,
+      "loss/crossentropy": 2.1540024280548096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.298043891787529,
+      "step": 2186
+    },
+    {
+      "epoch": 0.04376,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.0145660400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.8595,
+      "loss/crossentropy": 1.6615915298461914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23485098034143448,
+      "step": 2188
+    },
+    {
+      "epoch": 0.0438,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.01357421875,
+      "learning_rate": 0.0001,
+      "loss": 5.0595,
+      "loss/crossentropy": 2.352560341358185,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2965056747198105,
+      "step": 2190
+    },
+    {
+      "epoch": 0.04384,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.0183746337890625,
+      "learning_rate": 0.0001,
+      "loss": 5.1463,
+      "loss/crossentropy": 2.0864007472991943,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2859686613082886,
+      "step": 2192
+    },
+    {
+      "epoch": 0.04388,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.016600545247395834,
+      "learning_rate": 0.0001,
+      "loss": 5.2599,
+      "loss/crossentropy": 1.8934992551803589,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24994677305221558,
+      "step": 2194
+    },
+    {
+      "epoch": 0.04392,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.02476806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.9976,
+      "loss/crossentropy": 2.2395824193954468,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2719078063964844,
+      "step": 2196
+    },
+    {
+      "epoch": 0.04396,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.02515869140625,
+      "learning_rate": 0.0001,
+      "loss": 5.2631,
+      "loss/crossentropy": 2.089230954647064,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2759791761636734,
+      "step": 2198
+    },
+    {
+      "epoch": 0.044,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.023567708333333333,
+      "learning_rate": 0.0001,
+      "loss": 5.298,
+      "loss/crossentropy": 2.2770241498947144,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31005042791366577,
+      "step": 2200
+    },
+    {
+      "epoch": 0.04404,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.028938802083333333,
+      "learning_rate": 0.0001,
+      "loss": 4.6842,
+      "loss/crossentropy": 2.067028760910034,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30712655186653137,
+      "step": 2202
+    },
+    {
+      "epoch": 0.04408,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.026395670572916665,
+      "learning_rate": 0.0001,
+      "loss": 5.0557,
+      "loss/crossentropy": 2.4397774934768677,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3508765548467636,
+      "step": 2204
+    },
+    {
+      "epoch": 0.04412,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 0.028880818684895834,
+      "learning_rate": 0.0001,
+      "loss": 4.966,
+      "loss/crossentropy": 1.8136217594146729,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2495090439915657,
+      "step": 2206
+    },
+    {
+      "epoch": 0.04416,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.022945149739583334,
+      "learning_rate": 0.0001,
+      "loss": 5.11,
+      "loss/crossentropy": 2.4620203971862793,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31419822573661804,
+      "step": 2208
+    },
+    {
+      "epoch": 0.0442,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.020042928059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.9756,
+      "loss/crossentropy": 1.8817986249923706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2705220878124237,
+      "step": 2210
+    },
+    {
+      "epoch": 0.04424,
+      "grad_norm": 3.125,
+      "grad_norm_var": 0.02769775390625,
+      "learning_rate": 0.0001,
+      "loss": 5.0069,
+      "loss/crossentropy": 1.9593598246574402,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.262384794652462,
+      "step": 2212
+    },
+    {
+      "epoch": 0.04428,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.03303120930989583,
+      "learning_rate": 0.0001,
+      "loss": 4.9133,
+      "loss/crossentropy": 2.1003851294517517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28965799510478973,
+      "step": 2214
+    },
+    {
+      "epoch": 0.04432,
+      "grad_norm": 3.46875,
+      "grad_norm_var": 0.07888081868489584,
+      "learning_rate": 0.0001,
+      "loss": 5.3243,
+      "loss/crossentropy": 2.23227858543396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3120953291654587,
+      "step": 2216
+    },
+    {
+      "epoch": 0.04436,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.0726226806640625,
+      "learning_rate": 0.0001,
+      "loss": 5.0901,
+      "loss/crossentropy": 1.8880399465560913,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2740217447280884,
+      "step": 2218
+    },
+    {
+      "epoch": 0.0444,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.07111714680989584,
+      "learning_rate": 0.0001,
+      "loss": 4.9444,
+      "loss/crossentropy": 2.132355511188507,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27536119520664215,
+      "step": 2220
+    },
+    {
+      "epoch": 0.04444,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.07419331868489583,
+      "learning_rate": 0.0001,
+      "loss": 4.7325,
+      "loss/crossentropy": 1.831633746623993,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26966987550258636,
+      "step": 2222
+    },
+    {
+      "epoch": 0.04448,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.08056538899739583,
+      "learning_rate": 0.0001,
+      "loss": 5.004,
+      "loss/crossentropy": 2.066656529903412,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27303647994995117,
+      "step": 2224
+    },
+    {
+      "epoch": 0.04452,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.0809234619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.9265,
+      "loss/crossentropy": 2.1416667699813843,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2781240791082382,
+      "step": 2226
+    },
+    {
+      "epoch": 0.04456,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.07366536458333334,
+      "learning_rate": 0.0001,
+      "loss": 5.0701,
+      "loss/crossentropy": 1.7953566908836365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27528999745845795,
+      "step": 2228
+    },
+    {
+      "epoch": 0.0446,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.06728515625,
+      "learning_rate": 0.0001,
+      "loss": 5.0182,
+      "loss/crossentropy": 2.1580333709716797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2998732179403305,
+      "step": 2230
+    },
+    {
+      "epoch": 0.04464,
+      "grad_norm": 2.9375,
+      "grad_norm_var": 0.034326171875,
+      "learning_rate": 0.0001,
+      "loss": 5.3619,
+      "loss/crossentropy": 2.1685701608657837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3457205891609192,
+      "step": 2232
+    },
+    {
+      "epoch": 0.04468,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.03345438639322917,
+      "learning_rate": 0.0001,
+      "loss": 4.7602,
+      "loss/crossentropy": 1.9424286484718323,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2728651314973831,
+      "step": 2234
+    },
+    {
+      "epoch": 0.04472,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.029736328125,
+      "learning_rate": 0.0001,
+      "loss": 5.2351,
+      "loss/crossentropy": 2.2802772521972656,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30141082406044006,
+      "step": 2236
+    },
+    {
+      "epoch": 0.04476,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.03277587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.8906,
+      "loss/crossentropy": 1.9490987062454224,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2573640048503876,
+      "step": 2238
+    },
+    {
+      "epoch": 0.0448,
+      "grad_norm": 3.171875,
+      "grad_norm_var": 0.037679036458333336,
+      "learning_rate": 0.0001,
+      "loss": 5.2112,
+      "loss/crossentropy": 1.993924081325531,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25628305971622467,
+      "step": 2240
+    },
+    {
+      "epoch": 0.04484,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.0359283447265625,
+      "learning_rate": 0.0001,
+      "loss": 5.268,
+      "loss/crossentropy": 2.5151875019073486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.326447993516922,
+      "step": 2242
+    },
+    {
+      "epoch": 0.04488,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.04641520182291667,
+      "learning_rate": 0.0001,
+      "loss": 5.0191,
+      "loss/crossentropy": 2.5175565481185913,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3036232739686966,
+      "step": 2244
+    },
+    {
+      "epoch": 0.04492,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.05388997395833333,
+      "learning_rate": 0.0001,
+      "loss": 4.8489,
+      "loss/crossentropy": 2.020721971988678,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28087201714515686,
+      "step": 2246
+    },
+    {
+      "epoch": 0.04496,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.045703125,
+      "learning_rate": 0.0001,
+      "loss": 5.6809,
+      "loss/crossentropy": 2.4800511598587036,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3100287467241287,
+      "step": 2248
+    },
+    {
+      "epoch": 0.045,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.0449127197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.9055,
+      "loss/crossentropy": 1.826172411441803,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2782330811023712,
+      "step": 2250
+    },
+    {
+      "epoch": 0.04504,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.04533589680989583,
+      "learning_rate": 0.0001,
+      "loss": 5.133,
+      "loss/crossentropy": 2.256316304206848,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3070906698703766,
+      "step": 2252
+    },
+    {
+      "epoch": 0.04508,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.041402180989583336,
+      "learning_rate": 0.0001,
+      "loss": 5.173,
+      "loss/crossentropy": 1.9046601057052612,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2894355356693268,
+      "step": 2254
+    },
+    {
+      "epoch": 0.04512,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.03288472493489583,
+      "learning_rate": 0.0001,
+      "loss": 5.0311,
+      "loss/crossentropy": 1.8359373211860657,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2705196440219879,
+      "step": 2256
+    },
+    {
+      "epoch": 0.04516,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.0349273681640625,
+      "learning_rate": 0.0001,
+      "loss": 4.717,
+      "loss/crossentropy": 2.096512258052826,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26635921001434326,
+      "step": 2258
+    },
+    {
+      "epoch": 0.0452,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.031371053059895834,
+      "learning_rate": 0.0001,
+      "loss": 5.6736,
+      "loss/crossentropy": 2.4621278047561646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3513137400150299,
+      "step": 2260
+    },
+    {
+      "epoch": 0.04524,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.020442708333333334,
+      "learning_rate": 0.0001,
+      "loss": 5.1413,
+      "loss/crossentropy": 1.8345229029655457,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2414976954460144,
+      "step": 2262
+    },
+    {
+      "epoch": 0.04528,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.026432291666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.9746,
+      "loss/crossentropy": 2.24505877494812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31794628500938416,
+      "step": 2264
+    },
+    {
+      "epoch": 0.04532,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.0285552978515625,
+      "learning_rate": 0.0001,
+      "loss": 5.1691,
+      "loss/crossentropy": 2.2141382694244385,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2925301343202591,
+      "step": 2266
+    },
+    {
+      "epoch": 0.04536,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.03681233723958333,
+      "learning_rate": 0.0001,
+      "loss": 5.0559,
+      "loss/crossentropy": 2.1515613794326782,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2843547910451889,
+      "step": 2268
+    },
+    {
+      "epoch": 0.0454,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.03422749837239583,
+      "learning_rate": 0.0001,
+      "loss": 4.6899,
+      "loss/crossentropy": 2.1234883666038513,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2981575280427933,
+      "step": 2270
+    },
+    {
+      "epoch": 0.04544,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.03806864420572917,
+      "learning_rate": 0.0001,
+      "loss": 4.9871,
+      "loss/crossentropy": 2.1212490797042847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.276496559381485,
+      "step": 2272
+    },
+    {
+      "epoch": 0.04548,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.03574930826822917,
+      "learning_rate": 0.0001,
+      "loss": 5.2925,
+      "loss/crossentropy": 2.4330636262893677,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2815839499235153,
+      "step": 2274
+    },
+    {
+      "epoch": 0.04552,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.028238932291666668,
+      "learning_rate": 0.0001,
+      "loss": 5.2874,
+      "loss/crossentropy": 2.110591411590576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2832919806241989,
+      "step": 2276
+    },
+    {
+      "epoch": 0.04556,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.20754801432291667,
+      "learning_rate": 0.0001,
+      "loss": 5.0303,
+      "loss/crossentropy": 2.231989800930023,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28183089196681976,
+      "step": 2278
+    },
+    {
+      "epoch": 0.0456,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.194873046875,
+      "learning_rate": 0.0001,
+      "loss": 5.3746,
+      "loss/crossentropy": 2.1275558471679688,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29279497265815735,
+      "step": 2280
+    },
+    {
+      "epoch": 0.04564,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.1891510009765625,
+      "learning_rate": 0.0001,
+      "loss": 5.2023,
+      "loss/crossentropy": 1.7988306283950806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2625636160373688,
+      "step": 2282
+    },
+    {
+      "epoch": 0.04568,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.18694254557291667,
+      "learning_rate": 0.0001,
+      "loss": 5.2017,
+      "loss/crossentropy": 2.3405990600585938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30855217576026917,
+      "step": 2284
+    },
+    {
+      "epoch": 0.04572,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.18612874348958333,
+      "learning_rate": 0.0001,
+      "loss": 5.4582,
+      "loss/crossentropy": 2.2062121629714966,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30755001306533813,
+      "step": 2286
+    },
+    {
+      "epoch": 0.04576,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.18968098958333332,
+      "learning_rate": 0.0001,
+      "loss": 4.8984,
+      "loss/crossentropy": 1.9439310431480408,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26844222843647003,
+      "step": 2288
+    },
+    {
+      "epoch": 0.0458,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.19010416666666666,
+      "learning_rate": 0.0001,
+      "loss": 5.2097,
+      "loss/crossentropy": 2.3106162548065186,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30821681022644043,
+      "step": 2290
+    },
+    {
+      "epoch": 0.04584,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.19246317545572916,
+      "learning_rate": 0.0001,
+      "loss": 5.1401,
+      "loss/crossentropy": 2.3809561729431152,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3307010233402252,
+      "step": 2292
+    },
+    {
+      "epoch": 0.04588,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.01103515625,
+      "learning_rate": 0.0001,
+      "loss": 5.4066,
+      "loss/crossentropy": 2.209702253341675,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2795914113521576,
+      "step": 2294
+    },
+    {
+      "epoch": 0.04592,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.0158111572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.8772,
+      "loss/crossentropy": 2.4084372520446777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31208573281764984,
+      "step": 2296
+    },
+    {
+      "epoch": 0.04596,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.0146392822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.7757,
+      "loss/crossentropy": 1.9384723901748657,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2625032365322113,
+      "step": 2298
+    },
+    {
+      "epoch": 0.046,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.013704427083333333,
+      "learning_rate": 0.0001,
+      "loss": 5.2455,
+      "loss/crossentropy": 2.150592088699341,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.278359517455101,
+      "step": 2300
+    },
+    {
+      "epoch": 0.04604,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.01256103515625,
+      "learning_rate": 0.0001,
+      "loss": 5.0788,
+      "loss/crossentropy": 1.8317970037460327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2927433103322983,
+      "step": 2302
+    },
+    {
+      "epoch": 0.04608,
+      "grad_norm": 3.09375,
+      "grad_norm_var": 0.029564412434895833,
+      "learning_rate": 0.0001,
+      "loss": 5.091,
+      "loss/crossentropy": 2.323367118835449,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2943577915430069,
+      "step": 2304
+    },
+    {
+      "epoch": 0.04612,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.03943583170572917,
+      "learning_rate": 0.0001,
+      "loss": 5.5301,
+      "loss/crossentropy": 2.369907855987549,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2990037202835083,
+      "step": 2306
+    },
+    {
+      "epoch": 0.04616,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.04127197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.7508,
+      "loss/crossentropy": 1.691443145275116,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27280642092227936,
+      "step": 2308
+    },
+    {
+      "epoch": 0.0462,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.0464508056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.9413,
+      "loss/crossentropy": 2.2883838415145874,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3064710944890976,
+      "step": 2310
+    },
+    {
+      "epoch": 0.04624,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.035380045572916664,
+      "learning_rate": 0.0001,
+      "loss": 5.4165,
+      "loss/crossentropy": 2.2042444944381714,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3241504430770874,
+      "step": 2312
+    },
+    {
+      "epoch": 0.04628,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.048680623372395836,
+      "learning_rate": 0.0001,
+      "loss": 4.6657,
+      "loss/crossentropy": 1.977793574333191,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2775610163807869,
+      "step": 2314
+    },
+    {
+      "epoch": 0.04632,
+      "grad_norm": 24.875,
+      "grad_norm_var": 30.570881144205728,
+      "learning_rate": 0.0001,
+      "loss": 5.8585,
+      "loss/crossentropy": 2.034530758857727,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27170561254024506,
+      "step": 2316
+    },
+    {
+      "epoch": 0.04636,
+      "grad_norm": 2.875,
+      "grad_norm_var": 30.404881795247395,
+      "learning_rate": 0.0001,
+      "loss": 5.1565,
+      "loss/crossentropy": 2.439123511314392,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2999258190393448,
+      "step": 2318
+    },
+    {
+      "epoch": 0.0464,
+      "grad_norm": 2.5,
+      "grad_norm_var": 30.530557250976564,
+      "learning_rate": 0.0001,
+      "loss": 4.8907,
+      "loss/crossentropy": 2.2600624561309814,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2952795475721359,
+      "step": 2320
+    },
+    {
+      "epoch": 0.04644,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 30.544131469726562,
+      "learning_rate": 0.0001,
+      "loss": 5.0521,
+      "loss/crossentropy": 2.144679367542267,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28713342547416687,
+      "step": 2322
+    },
+    {
+      "epoch": 0.04648,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 30.469155883789064,
+      "learning_rate": 0.0001,
+      "loss": 5.5054,
+      "loss/crossentropy": 2.34474778175354,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.292633980512619,
+      "step": 2324
+    },
+    {
+      "epoch": 0.04652,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 30.491536458333332,
+      "learning_rate": 0.0001,
+      "loss": 4.5903,
+      "loss/crossentropy": 1.96743243932724,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.265610933303833,
+      "step": 2326
+    },
+    {
+      "epoch": 0.04656,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 30.524051920572916,
+      "learning_rate": 0.0001,
+      "loss": 5.2353,
+      "loss/crossentropy": 2.3895785808563232,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3287513107061386,
+      "step": 2328
+    },
+    {
+      "epoch": 0.0466,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 30.446451822916668,
+      "learning_rate": 0.0001,
+      "loss": 4.9017,
+      "loss/crossentropy": 2.0607098937034607,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30049796402454376,
+      "step": 2330
+    },
+    {
+      "epoch": 0.04664,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.08322652180989583,
+      "learning_rate": 0.0001,
+      "loss": 5.1698,
+      "loss/crossentropy": 2.162124752998352,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2983020693063736,
+      "step": 2332
+    },
+    {
+      "epoch": 0.04668,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 0.05139567057291667,
+      "learning_rate": 0.0001,
+      "loss": 5.2702,
+      "loss/crossentropy": 2.34970760345459,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3322293907403946,
+      "step": 2334
+    },
+    {
+      "epoch": 0.04672,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.04527587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.9763,
+      "loss/crossentropy": 1.9286972284317017,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2752893418073654,
+      "step": 2336
+    },
+    {
+      "epoch": 0.04676,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.04159749348958333,
+      "learning_rate": 0.0001,
+      "loss": 5.1677,
+      "loss/crossentropy": 2.182044267654419,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30808278918266296,
+      "step": 2338
+    },
+    {
+      "epoch": 0.0468,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.03916727701822917,
+      "learning_rate": 0.0001,
+      "loss": 5.28,
+      "loss/crossentropy": 2.0002610087394714,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2944178581237793,
+      "step": 2340
+    },
+    {
+      "epoch": 0.04684,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.034830729166666664,
+      "learning_rate": 0.0001,
+      "loss": 4.984,
+      "loss/crossentropy": 2.0721842646598816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2769011855125427,
+      "step": 2342
+    },
+    {
+      "epoch": 0.04688,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.03178609212239583,
+      "learning_rate": 0.0001,
+      "loss": 5.0911,
+      "loss/crossentropy": 1.9710460305213928,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28619086742401123,
+      "step": 2344
+    },
+    {
+      "epoch": 0.04692,
+      "grad_norm": 3.28125,
+      "grad_norm_var": 1.54605712890625,
+      "learning_rate": 0.0001,
+      "loss": 5.5092,
+      "loss/crossentropy": 2.0506762266159058,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31705181300640106,
+      "step": 2346
+    },
+    {
+      "epoch": 0.04696,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 1.5601064046223958,
+      "learning_rate": 0.0001,
+      "loss": 5.0045,
+      "loss/crossentropy": 2.0095930695533752,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2831149846315384,
+      "step": 2348
+    },
+    {
+      "epoch": 0.047,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 1.573631795247396,
+      "learning_rate": 0.0001,
+      "loss": 5.19,
+      "loss/crossentropy": 2.023163616657257,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2813292294740677,
+      "step": 2350
+    },
+    {
+      "epoch": 0.04704,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 1.561424763997396,
+      "learning_rate": 0.0001,
+      "loss": 5.2907,
+      "loss/crossentropy": 2.230435371398926,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30891451239585876,
+      "step": 2352
+    },
+    {
+      "epoch": 0.04708,
+      "grad_norm": 3.0625,
+      "grad_norm_var": 1.5700154622395834,
+      "learning_rate": 0.0001,
+      "loss": 4.9261,
+      "loss/crossentropy": 2.1553521156311035,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29839709401130676,
+      "step": 2354
+    },
+    {
+      "epoch": 0.04712,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 1.5770792643229166,
+      "learning_rate": 0.0001,
+      "loss": 5.2553,
+      "loss/crossentropy": 2.175648272037506,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2972148358821869,
+      "step": 2356
+    },
+    {
+      "epoch": 0.04716,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 1.5980377197265625,
+      "learning_rate": 0.0001,
+      "loss": 5.0436,
+      "loss/crossentropy": 2.3852503299713135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3166612535715103,
+      "step": 2358
+    },
+    {
+      "epoch": 0.0472,
+      "grad_norm": 2.75,
+      "grad_norm_var": 1.5826456705729166,
+      "learning_rate": 0.0001,
+      "loss": 5.1827,
+      "loss/crossentropy": 2.1905999183654785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29487256705760956,
+      "step": 2360
+    },
+    {
+      "epoch": 0.04724,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.0413482666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.9295,
+      "loss/crossentropy": 1.9224759340286255,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27821336686611176,
+      "step": 2362
+    },
+    {
+      "epoch": 0.04728,
+      "grad_norm": 3.109375,
+      "grad_norm_var": 0.04057515462239583,
+      "learning_rate": 0.0001,
+      "loss": 5.1078,
+      "loss/crossentropy": 2.5025261640548706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3345927745103836,
+      "step": 2364
+    },
+    {
+      "epoch": 0.04732,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.04006245930989583,
+      "learning_rate": 0.0001,
+      "loss": 5.0502,
+      "loss/crossentropy": 2.2385451793670654,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27095621824264526,
+      "step": 2366
+    },
+    {
+      "epoch": 0.04736,
+      "grad_norm": 4.5,
+      "grad_norm_var": 0.22924702962239582,
+      "learning_rate": 0.0001,
+      "loss": 5.2761,
+      "loss/crossentropy": 2.0266553163528442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2856762409210205,
+      "step": 2368
+    },
+    {
+      "epoch": 0.0474,
+      "grad_norm": 3.359375,
+      "grad_norm_var": 0.24001363118489583,
+      "learning_rate": 0.0001,
+      "loss": 5.4918,
+      "loss/crossentropy": 2.5139355659484863,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3186872750520706,
+      "step": 2370
+    },
+    {
+      "epoch": 0.04744,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.24321187337239583,
+      "learning_rate": 0.0001,
+      "loss": 5.1302,
+      "loss/crossentropy": 2.066476881504059,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2889470160007477,
+      "step": 2372
+    },
+    {
+      "epoch": 0.04748,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.23772786458333334,
+      "learning_rate": 0.0001,
+      "loss": 5.0031,
+      "loss/crossentropy": 2.0537307262420654,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31795741617679596,
+      "step": 2374
+    },
+    {
+      "epoch": 0.04752,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.24492085774739583,
+      "learning_rate": 0.0001,
+      "loss": 4.9922,
+      "loss/crossentropy": 1.9254986643791199,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25411880016326904,
+      "step": 2376
+    },
+    {
+      "epoch": 0.04756,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.23855692545572918,
+      "learning_rate": 0.0001,
+      "loss": 5.0284,
+      "loss/crossentropy": 2.221043348312378,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28534361720085144,
+      "step": 2378
+    },
+    {
+      "epoch": 0.0476,
+      "grad_norm": 2.90625,
+      "grad_norm_var": 0.228271484375,
+      "learning_rate": 0.0001,
+      "loss": 5.2106,
+      "loss/crossentropy": 2.3516281843185425,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3128499984741211,
+      "step": 2380
+    },
+    {
+      "epoch": 0.04764,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.22349853515625,
+      "learning_rate": 0.0001,
+      "loss": 5.6266,
+      "loss/crossentropy": 2.2144338488578796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3071902245283127,
+      "step": 2382
+    },
+    {
+      "epoch": 0.04768,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.0567291259765625,
+      "learning_rate": 0.0001,
+      "loss": 5.2429,
+      "loss/crossentropy": 2.3324203491210938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.322970449924469,
+      "step": 2384
+    },
+    {
+      "epoch": 0.04772,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.03332926432291667,
+      "learning_rate": 0.0001,
+      "loss": 4.7732,
+      "loss/crossentropy": 2.08349871635437,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2774003893136978,
+      "step": 2386
+    },
+    {
+      "epoch": 0.04776,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.03902587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.8585,
+      "loss/crossentropy": 1.9565780758857727,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29790589213371277,
+      "step": 2388
+    },
+    {
+      "epoch": 0.0478,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.022782389322916666,
+      "learning_rate": 0.0001,
+      "loss": 5.0176,
+      "loss/crossentropy": 2.261398434638977,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3143853694200516,
+      "step": 2390
+    },
+    {
+      "epoch": 0.04784,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.024494425455729166,
+      "learning_rate": 0.0001,
+      "loss": 5.0688,
+      "loss/crossentropy": 1.9077460169792175,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25433091819286346,
+      "step": 2392
+    },
+    {
+      "epoch": 0.04788,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.0240142822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.9531,
+      "loss/crossentropy": 1.9948468208312988,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2598777562379837,
+      "step": 2394
+    },
+    {
+      "epoch": 0.04792,
+      "grad_norm": 2.9375,
+      "grad_norm_var": 0.45455729166666664,
+      "learning_rate": 0.0001,
+      "loss": 5.0972,
+      "loss/crossentropy": 2.1177526116371155,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2775426208972931,
+      "step": 2396
+    },
+    {
+      "epoch": 0.04796,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.4634348551432292,
+      "learning_rate": 0.0001,
+      "loss": 4.8571,
+      "loss/crossentropy": 2.1756062507629395,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33678852021694183,
+      "step": 2398
+    },
+    {
+      "epoch": 0.048,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.4576171875,
+      "learning_rate": 0.0001,
+      "loss": 5.2617,
+      "loss/crossentropy": 2.0923725366592407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3255026638507843,
+      "step": 2400
+    },
+    {
+      "epoch": 0.04804,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.4471181233723958,
+      "learning_rate": 0.0001,
+      "loss": 4.9421,
+      "loss/crossentropy": 1.9236682653427124,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2642124071717262,
+      "step": 2402
+    },
+    {
+      "epoch": 0.04808,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.43757222493489584,
+      "learning_rate": 0.0001,
+      "loss": 5.0604,
+      "loss/crossentropy": 2.1742242574691772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30006398260593414,
+      "step": 2404
+    },
+    {
+      "epoch": 0.04812,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.43835347493489585,
+      "learning_rate": 0.0001,
+      "loss": 4.7077,
+      "loss/crossentropy": 1.7445701956748962,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25791122019290924,
+      "step": 2406
+    },
+    {
+      "epoch": 0.04816,
+      "grad_norm": 4.34375,
+      "grad_norm_var": 0.5614735921223958,
+      "learning_rate": 0.0001,
+      "loss": 5.1289,
+      "loss/crossentropy": 1.8616467714309692,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2532489001750946,
+      "step": 2408
+    },
+    {
+      "epoch": 0.0482,
+      "grad_norm": 3.203125,
+      "grad_norm_var": 0.5608723958333334,
+      "learning_rate": 0.0001,
+      "loss": 5.0486,
+      "loss/crossentropy": 1.9146783351898193,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2659924626350403,
+      "step": 2410
+    },
+    {
+      "epoch": 0.04824,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.21818745930989583,
+      "learning_rate": 0.0001,
+      "loss": 5.0972,
+      "loss/crossentropy": 2.179564118385315,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29064056277275085,
+      "step": 2412
+    },
+    {
+      "epoch": 0.04828,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.21357421875,
+      "learning_rate": 0.0001,
+      "loss": 4.9578,
+      "loss/crossentropy": 2.04409658908844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27891072630882263,
+      "step": 2414
+    },
+    {
+      "epoch": 0.04832,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.22568257649739584,
+      "learning_rate": 0.0001,
+      "loss": 4.8833,
+      "loss/crossentropy": 2.590337038040161,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3158426731824875,
+      "step": 2416
+    },
+    {
+      "epoch": 0.04836,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.23155924479166667,
+      "learning_rate": 0.0001,
+      "loss": 4.6919,
+      "loss/crossentropy": 1.8753941059112549,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2669401317834854,
+      "step": 2418
+    },
+    {
+      "epoch": 0.0484,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.23361714680989584,
+      "learning_rate": 0.0001,
+      "loss": 5.0231,
+      "loss/crossentropy": 2.1412659287452698,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2839512526988983,
+      "step": 2420
+    },
+    {
+      "epoch": 0.04844,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.23371988932291668,
+      "learning_rate": 0.0001,
+      "loss": 5.1187,
+      "loss/crossentropy": 2.545991063117981,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3331379294395447,
+      "step": 2422
+    },
+    {
+      "epoch": 0.04848,
+      "grad_norm": 3.25,
+      "grad_norm_var": 0.089599609375,
+      "learning_rate": 0.0001,
+      "loss": 5.1246,
+      "loss/crossentropy": 2.12838077545166,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3043065369129181,
+      "step": 2424
+    },
+    {
+      "epoch": 0.04852,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.03660380045572917,
+      "learning_rate": 0.0001,
+      "loss": 5.3821,
+      "loss/crossentropy": 2.1983221769332886,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2846619784832001,
+      "step": 2426
+    },
+    {
+      "epoch": 0.04856,
+      "grad_norm": 3.0625,
+      "grad_norm_var": 0.03819986979166667,
+      "learning_rate": 0.0001,
+      "loss": 5.1044,
+      "loss/crossentropy": 2.241136312484741,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30275705456733704,
+      "step": 2428
+    },
+    {
+      "epoch": 0.0486,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 0.03831278483072917,
+      "learning_rate": 0.0001,
+      "loss": 5.203,
+      "loss/crossentropy": 2.097459554672241,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2986721396446228,
+      "step": 2430
+    },
+    {
+      "epoch": 0.04864,
+      "grad_norm": 4.4375,
+      "grad_norm_var": 0.20159403483072916,
+      "learning_rate": 0.0001,
+      "loss": 5.1158,
+      "loss/crossentropy": 2.333081007003784,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2975248098373413,
+      "step": 2432
+    },
+    {
+      "epoch": 0.04868,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.19758707682291668,
+      "learning_rate": 0.0001,
+      "loss": 4.9183,
+      "loss/crossentropy": 2.3171510696411133,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28452740609645844,
+      "step": 2434
+    },
+    {
+      "epoch": 0.04872,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.2141754150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.8853,
+      "loss/crossentropy": 1.8334497213363647,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25271379947662354,
+      "step": 2436
+    },
+    {
+      "epoch": 0.04876,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.20258687337239584,
+      "learning_rate": 0.0001,
+      "loss": 5.3809,
+      "loss/crossentropy": 2.2712661027908325,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30827929079532623,
+      "step": 2438
+    },
+    {
+      "epoch": 0.0488,
+      "grad_norm": 3.3125,
+      "grad_norm_var": 0.20465087890625,
+      "learning_rate": 0.0001,
+      "loss": 5.0863,
+      "loss/crossentropy": 2.160263180732727,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3078690320253372,
+      "step": 2440
+    },
+    {
+      "epoch": 0.04884,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.20429585774739584,
+      "learning_rate": 0.0001,
+      "loss": 5.2224,
+      "loss/crossentropy": 2.071319878101349,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2733730524778366,
+      "step": 2442
+    },
+    {
+      "epoch": 0.04888,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.203076171875,
+      "learning_rate": 0.0001,
+      "loss": 5.1476,
+      "loss/crossentropy": 2.0742560029029846,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2835587412118912,
+      "step": 2444
+    },
+    {
+      "epoch": 0.04892,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.24807535807291667,
+      "learning_rate": 0.0001,
+      "loss": 5.0211,
+      "loss/crossentropy": 1.8836837410926819,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3089100867509842,
+      "step": 2446
+    },
+    {
+      "epoch": 0.04896,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.177099609375,
+      "learning_rate": 0.0001,
+      "loss": 4.9537,
+      "loss/crossentropy": 1.8539315462112427,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2697101980447769,
+      "step": 2448
+    },
+    {
+      "epoch": 0.049,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.17392578125,
+      "learning_rate": 0.0001,
+      "loss": 5.1907,
+      "loss/crossentropy": 2.219490647315979,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2898380011320114,
+      "step": 2450
+    },
+    {
+      "epoch": 0.04904,
+      "grad_norm": 3.1875,
+      "grad_norm_var": 0.13853251139322917,
+      "learning_rate": 0.0001,
+      "loss": 5.468,
+      "loss/crossentropy": 2.328765392303467,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4013051837682724,
+      "step": 2452
+    },
+    {
+      "epoch": 0.04908,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.14879557291666667,
+      "learning_rate": 0.0001,
+      "loss": 4.8967,
+      "loss/crossentropy": 1.9204192161560059,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25281261652708054,
+      "step": 2454
+    },
+    {
+      "epoch": 0.04912,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.15563151041666667,
+      "learning_rate": 0.0001,
+      "loss": 5.0935,
+      "loss/crossentropy": 2.377043604850769,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2928486764431,
+      "step": 2456
+    },
+    {
+      "epoch": 0.04916,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.15930989583333333,
+      "learning_rate": 0.0001,
+      "loss": 5.4528,
+      "loss/crossentropy": 2.4364209175109863,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3111976683139801,
+      "step": 2458
+    },
+    {
+      "epoch": 0.0492,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.15985921223958333,
+      "learning_rate": 0.0001,
+      "loss": 5.1357,
+      "loss/crossentropy": 2.3738330602645874,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3019126206636429,
+      "step": 2460
+    },
+    {
+      "epoch": 0.04924,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.1141265869140625,
+      "learning_rate": 0.0001,
+      "loss": 5.2876,
+      "loss/crossentropy": 2.4575772285461426,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3428986072540283,
+      "step": 2462
+    },
+    {
+      "epoch": 0.04928,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.028888956705729166,
+      "learning_rate": 0.0001,
+      "loss": 4.6505,
+      "loss/crossentropy": 1.9849627017974854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27611130475997925,
+      "step": 2464
+    },
+    {
+      "epoch": 0.04932,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.030402628580729167,
+      "learning_rate": 0.0001,
+      "loss": 4.8482,
+      "loss/crossentropy": 1.9617170691490173,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.257377490401268,
+      "step": 2466
+    },
+    {
+      "epoch": 0.04936,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.01871337890625,
+      "learning_rate": 0.0001,
+      "loss": 5.1739,
+      "loss/crossentropy": 2.4031273126602173,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28843145072460175,
+      "step": 2468
+    },
+    {
+      "epoch": 0.0494,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.0165435791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.8452,
+      "loss/crossentropy": 1.7263792753219604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25591571629047394,
+      "step": 2470
+    },
+    {
+      "epoch": 0.04944,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.0191802978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.7154,
+      "loss/crossentropy": 2.106898784637451,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.280623197555542,
+      "step": 2472
+    },
+    {
+      "epoch": 0.04948,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.018358357747395835,
+      "learning_rate": 0.0001,
+      "loss": 5.0092,
+      "loss/crossentropy": 2.325208902359009,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27794161438941956,
+      "step": 2474
+    },
+    {
+      "epoch": 0.04952,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.013736979166666666,
+      "learning_rate": 0.0001,
+      "loss": 5.1128,
+      "loss/crossentropy": 2.367414712905884,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31346653401851654,
+      "step": 2476
+    },
+    {
+      "epoch": 0.04956,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.013792928059895833,
+      "learning_rate": 0.0001,
+      "loss": 5.2611,
+      "loss/crossentropy": 2.191115140914917,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30560287833213806,
+      "step": 2478
+    },
+    {
+      "epoch": 0.0496,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.012430826822916666,
+      "learning_rate": 0.0001,
+      "loss": 4.9993,
+      "loss/crossentropy": 2.2559624314308167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3029457628726959,
+      "step": 2480
+    },
+    {
+      "epoch": 0.04964,
+      "grad_norm": 3.390625,
+      "grad_norm_var": 0.04160868326822917,
+      "learning_rate": 0.0001,
+      "loss": 5.125,
+      "loss/crossentropy": 2.0088382363319397,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2923784404993057,
+      "step": 2482
+    },
+    {
+      "epoch": 0.04968,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.03943583170572917,
+      "learning_rate": 0.0001,
+      "loss": 4.8858,
+      "loss/crossentropy": 1.8445329070091248,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2587483897805214,
+      "step": 2484
+    },
+    {
+      "epoch": 0.04972,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.03534749348958333,
+      "learning_rate": 0.0001,
+      "loss": 4.7918,
+      "loss/crossentropy": 2.015140950679779,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27480585873126984,
+      "step": 2486
+    },
+    {
+      "epoch": 0.04976,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.030833943684895834,
+      "learning_rate": 0.0001,
+      "loss": 5.1959,
+      "loss/crossentropy": 1.918801188468933,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28786011040210724,
+      "step": 2488
+    },
+    {
+      "epoch": 0.0498,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.030269368489583334,
+      "learning_rate": 0.0001,
+      "loss": 5.0108,
+      "loss/crossentropy": 1.9899121522903442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25822708010673523,
+      "step": 2490
+    },
+    {
+      "epoch": 0.04984,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.037206013997395836,
+      "learning_rate": 0.0001,
+      "loss": 4.7259,
+      "loss/crossentropy": 2.3775535821914673,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31345370411872864,
+      "step": 2492
+    },
+    {
+      "epoch": 0.04988,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.03791402180989583,
+      "learning_rate": 0.0001,
+      "loss": 4.9496,
+      "loss/crossentropy": 2.0874632596969604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2836592495441437,
+      "step": 2494
+    },
+    {
+      "epoch": 0.04992,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.046019490559895834,
+      "learning_rate": 0.0001,
+      "loss": 5.2404,
+      "loss/crossentropy": 2.226976454257965,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27826404571533203,
+      "step": 2496
+    },
+    {
+      "epoch": 0.04996,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.023900349934895832,
+      "learning_rate": 0.0001,
+      "loss": 5.1503,
+      "loss/crossentropy": 2.4569294452667236,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31981319189071655,
+      "step": 2498
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.025581868489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.922,
+      "loss/crossentropy": 2.1134212017059326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28260529041290283,
+      "step": 2500
+    },
+    {
+      "epoch": 0.05004,
+      "grad_norm": 3.0,
+      "grad_norm_var": 0.029195149739583332,
+      "learning_rate": 0.0001,
+      "loss": 5.3555,
+      "loss/crossentropy": 2.2914888858795166,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3182682394981384,
+      "step": 2502
+    },
+    {
+      "epoch": 0.05008,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.030078125,
+      "learning_rate": 0.0001,
+      "loss": 4.9644,
+      "loss/crossentropy": 2.3261003494262695,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30558250844478607,
+      "step": 2504
+    },
+    {
+      "epoch": 0.05012,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.030598958333333332,
+      "learning_rate": 0.0001,
+      "loss": 4.9517,
+      "loss/crossentropy": 2.351989507675171,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2934701144695282,
+      "step": 2506
+    },
+    {
+      "epoch": 0.05016,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.025031534830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.71,
+      "loss/crossentropy": 1.8742690086364746,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26305729895830154,
+      "step": 2508
+    },
+    {
+      "epoch": 0.0502,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.0247222900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.8214,
+      "loss/crossentropy": 2.1668856143951416,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2728075534105301,
+      "step": 2510
+    },
+    {
+      "epoch": 0.05024,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.0137115478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.7171,
+      "loss/crossentropy": 1.773424208164215,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26808495819568634,
+      "step": 2512
+    },
+    {
+      "epoch": 0.05028,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.013109334309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.9224,
+      "loss/crossentropy": 1.7541643977165222,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2600491940975189,
+      "step": 2514
+    },
+    {
+      "epoch": 0.05032,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.011324055989583333,
+      "learning_rate": 0.0001,
+      "loss": 5.129,
+      "loss/crossentropy": 1.9693496227264404,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2844541072845459,
+      "step": 2516
+    },
+    {
+      "epoch": 0.05036,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.009065755208333333,
+      "learning_rate": 0.0001,
+      "loss": 4.9202,
+      "loss/crossentropy": 1.7539461851119995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2542608380317688,
+      "step": 2518
+    },
+    {
+      "epoch": 0.0504,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.017878214518229168,
+      "learning_rate": 0.0001,
+      "loss": 5.0686,
+      "loss/crossentropy": 2.155138313770294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3023010194301605,
+      "step": 2520
+    },
+    {
+      "epoch": 0.05044,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.018619791666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.9674,
+      "loss/crossentropy": 2.069350838661194,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26573850214481354,
+      "step": 2522
+    },
+    {
+      "epoch": 0.05048,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.022101847330729167,
+      "learning_rate": 0.0001,
+      "loss": 5.0295,
+      "loss/crossentropy": 2.1864534616470337,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27845603227615356,
+      "step": 2524
+    },
+    {
+      "epoch": 0.05052,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.023509724934895834,
+      "learning_rate": 0.0001,
+      "loss": 5.1173,
+      "loss/crossentropy": 2.2462135553359985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29221346974372864,
+      "step": 2526
+    },
+    {
+      "epoch": 0.05056,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.030492146809895832,
+      "learning_rate": 0.0001,
+      "loss": 5.0027,
+      "loss/crossentropy": 2.043266773223877,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25857964158058167,
+      "step": 2528
+    },
+    {
+      "epoch": 0.0506,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.0299957275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.9672,
+      "loss/crossentropy": 1.8892702460289001,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2879898101091385,
+      "step": 2530
+    },
+    {
+      "epoch": 0.05064,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.03238525390625,
+      "learning_rate": 0.0001,
+      "loss": 4.5332,
+      "loss/crossentropy": 1.9220558404922485,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26652154326438904,
+      "step": 2532
+    },
+    {
+      "epoch": 0.05068,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.03430989583333333,
+      "learning_rate": 0.0001,
+      "loss": 4.4176,
+      "loss/crossentropy": 1.7282914519309998,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2367371916770935,
+      "step": 2534
+    },
+    {
+      "epoch": 0.05072,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.021092732747395832,
+      "learning_rate": 0.0001,
+      "loss": 4.9589,
+      "loss/crossentropy": 1.729803204536438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25624871999025345,
+      "step": 2536
+    },
+    {
+      "epoch": 0.05076,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.020873006184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.6952,
+      "loss/crossentropy": 2.0921449661254883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2855897545814514,
+      "step": 2538
+    },
+    {
+      "epoch": 0.0508,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.015404256184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.8967,
+      "loss/crossentropy": 2.0569751858711243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2829667925834656,
+      "step": 2540
+    },
+    {
+      "epoch": 0.05084,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.015973917643229165,
+      "learning_rate": 0.0001,
+      "loss": 5.2438,
+      "loss/crossentropy": 1.983904242515564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30939212441444397,
+      "step": 2542
+    },
+    {
+      "epoch": 0.05088,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.011864217122395833,
+      "learning_rate": 0.0001,
+      "loss": 4.9968,
+      "loss/crossentropy": 2.2631462812423706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27867285907268524,
+      "step": 2544
+    },
+    {
+      "epoch": 0.05092,
+      "grad_norm": 3.078125,
+      "grad_norm_var": 0.023856608072916667,
+      "learning_rate": 0.0001,
+      "loss": 4.9621,
+      "loss/crossentropy": 1.9918989539146423,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27703428268432617,
+      "step": 2546
+    },
+    {
+      "epoch": 0.05096,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.021110026041666667,
+      "learning_rate": 0.0001,
+      "loss": 5.2041,
+      "loss/crossentropy": 2.1356931924819946,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27815964818000793,
+      "step": 2548
+    },
+    {
+      "epoch": 0.051,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.019071451822916665,
+      "learning_rate": 0.0001,
+      "loss": 4.6514,
+      "loss/crossentropy": 2.172751545906067,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2971910834312439,
+      "step": 2550
+    },
+    {
+      "epoch": 0.05104,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.019527180989583334,
+      "learning_rate": 0.0001,
+      "loss": 5.0178,
+      "loss/crossentropy": 2.0799094438552856,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30459292232990265,
+      "step": 2552
+    },
+    {
+      "epoch": 0.05108,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.0211090087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.8235,
+      "loss/crossentropy": 1.7769129872322083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2526697665452957,
+      "step": 2554
+    },
+    {
+      "epoch": 0.05112,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.024494425455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.8457,
+      "loss/crossentropy": 2.044790804386139,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2821648418903351,
+      "step": 2556
+    },
+    {
+      "epoch": 0.05116,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.0287109375,
+      "learning_rate": 0.0001,
+      "loss": 5.5336,
+      "loss/crossentropy": 2.3708614110946655,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30810464918613434,
+      "step": 2558
+    },
+    {
+      "epoch": 0.0512,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.028709920247395833,
+      "learning_rate": 0.0001,
+      "loss": 5.2385,
+      "loss/crossentropy": 2.2216718196868896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2919304668903351,
+      "step": 2560
+    },
+    {
+      "epoch": 0.05124,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.019466145833333334,
+      "learning_rate": 0.0001,
+      "loss": 5.28,
+      "loss/crossentropy": 2.4692097902297974,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30620162189006805,
+      "step": 2562
+    },
+    {
+      "epoch": 0.05128,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.026936848958333332,
+      "learning_rate": 0.0001,
+      "loss": 4.9476,
+      "loss/crossentropy": 2.0491623282432556,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25617313385009766,
+      "step": 2564
+    },
+    {
+      "epoch": 0.05132,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.023368326822916667,
+      "learning_rate": 0.0001,
+      "loss": 4.958,
+      "loss/crossentropy": 1.8305597305297852,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25237561762332916,
+      "step": 2566
+    },
+    {
+      "epoch": 0.05136,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.0223297119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.9853,
+      "loss/crossentropy": 1.9471853971481323,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2712964415550232,
+      "step": 2568
+    },
+    {
+      "epoch": 0.0514,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.0198638916015625,
+      "learning_rate": 0.0001,
+      "loss": 5.0932,
+      "loss/crossentropy": 2.575412631034851,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3106851130723953,
+      "step": 2570
+    },
+    {
+      "epoch": 0.05144,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.020686848958333334,
+      "learning_rate": 0.0001,
+      "loss": 4.6755,
+      "loss/crossentropy": 2.0210241079330444,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.269680991768837,
+      "step": 2572
+    },
+    {
+      "epoch": 0.05148,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.02076416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.6308,
+      "loss/crossentropy": 1.9054389595985413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24933087825775146,
+      "step": 2574
+    },
+    {
+      "epoch": 0.05152,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.021284993489583334,
+      "learning_rate": 0.0001,
+      "loss": 4.9682,
+      "loss/crossentropy": 2.142069697380066,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3162301778793335,
+      "step": 2576
+    },
+    {
+      "epoch": 0.05156,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.0197662353515625,
+      "learning_rate": 0.0001,
+      "loss": 5.018,
+      "loss/crossentropy": 1.9952309727668762,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2560836151242256,
+      "step": 2578
+    },
+    {
+      "epoch": 0.0516,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.016405232747395835,
+      "learning_rate": 0.0001,
+      "loss": 4.8889,
+      "loss/crossentropy": 2.0579317212104797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26214616745710373,
+      "step": 2580
+    },
+    {
+      "epoch": 0.05164,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.016927083333333332,
+      "learning_rate": 0.0001,
+      "loss": 4.6306,
+      "loss/crossentropy": 2.076499104499817,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26244185864925385,
+      "step": 2582
+    },
+    {
+      "epoch": 0.05168,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 0.026146443684895833,
+      "learning_rate": 0.0001,
+      "loss": 4.9393,
+      "loss/crossentropy": 2.2277501821517944,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32310059666633606,
+      "step": 2584
+    },
+    {
+      "epoch": 0.05172,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.03141988118489583,
+      "learning_rate": 0.0001,
+      "loss": 5.0929,
+      "loss/crossentropy": 2.101436138153076,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26044395565986633,
+      "step": 2586
+    },
+    {
+      "epoch": 0.05176,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.030939737955729168,
+      "learning_rate": 0.0001,
+      "loss": 5.045,
+      "loss/crossentropy": 2.2617305517196655,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2768043726682663,
+      "step": 2588
+    },
+    {
+      "epoch": 0.0518,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.0349029541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.989,
+      "loss/crossentropy": 2.2669495940208435,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2941794842481613,
+      "step": 2590
+    },
+    {
+      "epoch": 0.05184,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.03871968587239583,
+      "learning_rate": 0.0001,
+      "loss": 4.7676,
+      "loss/crossentropy": 1.8102391958236694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24766983091831207,
+      "step": 2592
+    },
+    {
+      "epoch": 0.05188,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.04006754557291667,
+      "learning_rate": 0.0001,
+      "loss": 5.0022,
+      "loss/crossentropy": 2.1426219940185547,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2808763086795807,
+      "step": 2594
+    },
+    {
+      "epoch": 0.05192,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.0368804931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.8649,
+      "loss/crossentropy": 2.0731321573257446,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28768520057201385,
+      "step": 2596
+    },
+    {
+      "epoch": 0.05196,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.03619384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.8019,
+      "loss/crossentropy": 1.8902159333229065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27209727466106415,
+      "step": 2598
+    },
+    {
+      "epoch": 0.052,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.030855305989583335,
+      "learning_rate": 0.0001,
+      "loss": 4.837,
+      "loss/crossentropy": 2.1860097646713257,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3191404938697815,
+      "step": 2600
+    },
+    {
+      "epoch": 0.05204,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.027074178059895832,
+      "learning_rate": 0.0001,
+      "loss": 4.8471,
+      "loss/crossentropy": 2.1010658740997314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2924908995628357,
+      "step": 2602
+    },
+    {
+      "epoch": 0.05208,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.0285064697265625,
+      "learning_rate": 0.0001,
+      "loss": 5.1725,
+      "loss/crossentropy": 2.0668399930000305,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2691944092512131,
+      "step": 2604
+    },
+    {
+      "epoch": 0.05212,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.023824055989583332,
+      "learning_rate": 0.0001,
+      "loss": 5.0115,
+      "loss/crossentropy": 2.310541272163391,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31448885798454285,
+      "step": 2606
+    },
+    {
+      "epoch": 0.05216,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.0222808837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.7731,
+      "loss/crossentropy": 2.023577332496643,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28528447449207306,
+      "step": 2608
+    },
+    {
+      "epoch": 0.0522,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.0217193603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.8408,
+      "loss/crossentropy": 2.0232901573181152,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25540125370025635,
+      "step": 2610
+    },
+    {
+      "epoch": 0.05224,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.019554646809895833,
+      "learning_rate": 0.0001,
+      "loss": 4.8374,
+      "loss/crossentropy": 2.147561550140381,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2914447784423828,
+      "step": 2612
+    },
+    {
+      "epoch": 0.05228,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.0193359375,
+      "learning_rate": 0.0001,
+      "loss": 4.8044,
+      "loss/crossentropy": 1.9136184453964233,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27879445254802704,
+      "step": 2614
+    },
+    {
+      "epoch": 0.05232,
+      "grad_norm": 3.03125,
+      "grad_norm_var": 0.0606842041015625,
+      "learning_rate": 0.0001,
+      "loss": 5.1245,
+      "loss/crossentropy": 2.1118494272232056,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2990800142288208,
+      "step": 2616
+    },
+    {
+      "epoch": 0.05236,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.0599761962890625,
+      "learning_rate": 0.0001,
+      "loss": 5.2189,
+      "loss/crossentropy": 1.9564325213432312,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2805679142475128,
+      "step": 2618
+    },
+    {
+      "epoch": 0.0524,
+      "grad_norm": 3.046875,
+      "grad_norm_var": 0.06641337076822916,
+      "learning_rate": 0.0001,
+      "loss": 4.9511,
+      "loss/crossentropy": 2.0683051347732544,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2906472980976105,
+      "step": 2620
+    },
+    {
+      "epoch": 0.05244,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.06653645833333334,
+      "learning_rate": 0.0001,
+      "loss": 5.058,
+      "loss/crossentropy": 2.0510823130607605,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28148986399173737,
+      "step": 2622
+    },
+    {
+      "epoch": 0.05248,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.06256103515625,
+      "learning_rate": 0.0001,
+      "loss": 5.2113,
+      "loss/crossentropy": 2.2972904443740845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32902073860168457,
+      "step": 2624
+    },
+    {
+      "epoch": 0.05252,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 0.0582427978515625,
+      "learning_rate": 0.0001,
+      "loss": 5.0939,
+      "loss/crossentropy": 1.9502179026603699,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2844041585922241,
+      "step": 2626
+    },
+    {
+      "epoch": 0.05256,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.05607808430989583,
+      "learning_rate": 0.0001,
+      "loss": 5.1078,
+      "loss/crossentropy": 2.1577298045158386,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2673380598425865,
+      "step": 2628
+    },
+    {
+      "epoch": 0.0526,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.05271708170572917,
+      "learning_rate": 0.0001,
+      "loss": 5.0514,
+      "loss/crossentropy": 2.1707664132118225,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2697260081768036,
+      "step": 2630
+    },
+    {
+      "epoch": 0.05264,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.017606608072916665,
+      "learning_rate": 0.0001,
+      "loss": 4.9617,
+      "loss/crossentropy": 2.0975311398506165,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27033862471580505,
+      "step": 2632
+    },
+    {
+      "epoch": 0.05268,
+      "grad_norm": 3.25,
+      "grad_norm_var": 0.033869425455729164,
+      "learning_rate": 0.0001,
+      "loss": 5.1841,
+      "loss/crossentropy": 2.197197914123535,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28717951476573944,
+      "step": 2634
+    },
+    {
+      "epoch": 0.05272,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.03570556640625,
+      "learning_rate": 0.0001,
+      "loss": 5.2558,
+      "loss/crossentropy": 2.2898266315460205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31359314918518066,
+      "step": 2636
+    },
+    {
+      "epoch": 0.05276,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.0347564697265625,
+      "learning_rate": 0.0001,
+      "loss": 5.4849,
+      "loss/crossentropy": 2.525710701942444,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3199215829372406,
+      "step": 2638
+    },
+    {
+      "epoch": 0.0528,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.03349609375,
+      "learning_rate": 0.0001,
+      "loss": 5.0495,
+      "loss/crossentropy": 2.2799761295318604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28715676069259644,
+      "step": 2640
+    },
+    {
+      "epoch": 0.05284,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.032373046875,
+      "learning_rate": 0.0001,
+      "loss": 4.9837,
+      "loss/crossentropy": 1.8681190013885498,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25435833632946014,
+      "step": 2642
+    },
+    {
+      "epoch": 0.05288,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.03730061848958333,
+      "learning_rate": 0.0001,
+      "loss": 4.807,
+      "loss/crossentropy": 1.9067540168762207,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25444281101226807,
+      "step": 2644
+    },
+    {
+      "epoch": 0.05292,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.03732096354166667,
+      "learning_rate": 0.0001,
+      "loss": 5.0326,
+      "loss/crossentropy": 2.370971202850342,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31173495948314667,
+      "step": 2646
+    },
+    {
+      "epoch": 0.05296,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.03752848307291667,
+      "learning_rate": 0.0001,
+      "loss": 5.1148,
+      "loss/crossentropy": 2.1829749941825867,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29970400035381317,
+      "step": 2648
+    },
+    {
+      "epoch": 0.053,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.020052083333333335,
+      "learning_rate": 0.0001,
+      "loss": 5.1086,
+      "loss/crossentropy": 2.0818498134613037,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27712464332580566,
+      "step": 2650
+    },
+    {
+      "epoch": 0.05304,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.010789998372395833,
+      "learning_rate": 0.0001,
+      "loss": 5.1071,
+      "loss/crossentropy": 2.2080377340316772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27552157640457153,
+      "step": 2652
+    },
+    {
+      "epoch": 0.05308,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.01109619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.9685,
+      "loss/crossentropy": 1.7115904092788696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23266373574733734,
+      "step": 2654
+    },
+    {
+      "epoch": 0.05312,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.009598795572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.8654,
+      "loss/crossentropy": 2.1736810207366943,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29696571826934814,
+      "step": 2656
+    },
+    {
+      "epoch": 0.05316,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.012886555989583333,
+      "learning_rate": 0.0001,
+      "loss": 5.0834,
+      "loss/crossentropy": 2.2366485595703125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3062315583229065,
+      "step": 2658
+    },
+    {
+      "epoch": 0.0532,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.018748982747395834,
+      "learning_rate": 0.0001,
+      "loss": 5.0888,
+      "loss/crossentropy": 1.9835070371627808,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35663366317749023,
+      "step": 2660
+    },
+    {
+      "epoch": 0.05324,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.026688639322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.6725,
+      "loss/crossentropy": 2.148723840713501,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2687191218137741,
+      "step": 2662
+    },
+    {
+      "epoch": 0.05328,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.02789306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.9191,
+      "loss/crossentropy": 2.2642128467559814,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2738381028175354,
+      "step": 2664
+    },
+    {
+      "epoch": 0.05332,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.028483072916666668,
+      "learning_rate": 0.0001,
+      "loss": 4.8209,
+      "loss/crossentropy": 1.839052438735962,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2566695362329483,
+      "step": 2666
+    },
+    {
+      "epoch": 0.05336,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.026927693684895834,
+      "learning_rate": 0.0001,
+      "loss": 4.8771,
+      "loss/crossentropy": 2.083684980869293,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2632910907268524,
+      "step": 2668
+    },
+    {
+      "epoch": 0.0534,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.028434244791666667,
+      "learning_rate": 0.0001,
+      "loss": 4.8585,
+      "loss/crossentropy": 2.2125936150550842,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31363190710544586,
+      "step": 2670
+    },
+    {
+      "epoch": 0.05344,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.029157511393229165,
+      "learning_rate": 0.0001,
+      "loss": 5.0354,
+      "loss/crossentropy": 2.2075263261795044,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2850564122200012,
+      "step": 2672
+    },
+    {
+      "epoch": 0.05348,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.026656087239583334,
+      "learning_rate": 0.0001,
+      "loss": 5.0145,
+      "loss/crossentropy": 2.0876463651657104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2972792685031891,
+      "step": 2674
+    },
+    {
+      "epoch": 0.05352,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.020406087239583332,
+      "learning_rate": 0.0001,
+      "loss": 5.2755,
+      "loss/crossentropy": 2.4033172130584717,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.304116889834404,
+      "step": 2676
+    },
+    {
+      "epoch": 0.05356,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.014501953125,
+      "learning_rate": 0.0001,
+      "loss": 5.0921,
+      "loss/crossentropy": 2.30586314201355,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3264722675085068,
+      "step": 2678
+    },
+    {
+      "epoch": 0.0536,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.015152994791666667,
+      "learning_rate": 0.0001,
+      "loss": 5.0941,
+      "loss/crossentropy": 2.2175174951553345,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2661540359258652,
+      "step": 2680
+    },
+    {
+      "epoch": 0.05364,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.022606404622395833,
+      "learning_rate": 0.0001,
+      "loss": 5.1162,
+      "loss/crossentropy": 2.0583900213241577,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26995618641376495,
+      "step": 2682
+    },
+    {
+      "epoch": 0.05368,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.020173136393229166,
+      "learning_rate": 0.0001,
+      "loss": 4.9357,
+      "loss/crossentropy": 2.310309052467346,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29642508924007416,
+      "step": 2684
+    },
+    {
+      "epoch": 0.05372,
+      "grad_norm": 3.15625,
+      "grad_norm_var": 0.03332926432291667,
+      "learning_rate": 0.0001,
+      "loss": 4.7945,
+      "loss/crossentropy": 2.134134352207184,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2619224041700363,
+      "step": 2686
+    },
+    {
+      "epoch": 0.05376,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.03306884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.8411,
+      "loss/crossentropy": 1.930562138557434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27108173072338104,
+      "step": 2688
+    },
+    {
+      "epoch": 0.0538,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.03455403645833333,
+      "learning_rate": 0.0001,
+      "loss": 4.9591,
+      "loss/crossentropy": 2.3414769172668457,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.270420178771019,
+      "step": 2690
+    },
+    {
+      "epoch": 0.05384,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.0356842041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.83,
+      "loss/crossentropy": 2.1726938486099243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2710433751344681,
+      "step": 2692
+    },
+    {
+      "epoch": 0.05388,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.0361968994140625,
+      "learning_rate": 0.0001,
+      "loss": 5.0163,
+      "loss/crossentropy": 2.220117926597595,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3212582617998123,
+      "step": 2694
+    },
+    {
+      "epoch": 0.05392,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.03297119140625,
+      "learning_rate": 0.0001,
+      "loss": 5.1845,
+      "loss/crossentropy": 2.2557668685913086,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3096832036972046,
+      "step": 2696
+    },
+    {
+      "epoch": 0.05396,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.0246246337890625,
+      "learning_rate": 0.0001,
+      "loss": 5.0399,
+      "loss/crossentropy": 1.94975346326828,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2530350238084793,
+      "step": 2698
+    },
+    {
+      "epoch": 0.054,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.024169921875,
+      "learning_rate": 0.0001,
+      "loss": 5.19,
+      "loss/crossentropy": 2.4622775316238403,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3135389983654022,
+      "step": 2700
+    },
+    {
+      "epoch": 0.05404,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.0110748291015625,
+      "learning_rate": 0.0001,
+      "loss": 5.2005,
+      "loss/crossentropy": 2.5367363691329956,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30742934346199036,
+      "step": 2702
+    },
+    {
+      "epoch": 0.05408,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.014631144205729167,
+      "learning_rate": 0.0001,
+      "loss": 5.146,
+      "loss/crossentropy": 2.5733184814453125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33410580456256866,
+      "step": 2704
+    },
+    {
+      "epoch": 0.05412,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.011725870768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.8888,
+      "loss/crossentropy": 1.9339997172355652,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28760699927806854,
+      "step": 2706
+    },
+    {
+      "epoch": 0.05416,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.010724894205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.8719,
+      "loss/crossentropy": 1.8515672087669373,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23648252338171005,
+      "step": 2708
+    },
+    {
+      "epoch": 0.0542,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.014420572916666667,
+      "learning_rate": 0.0001,
+      "loss": 4.6598,
+      "loss/crossentropy": 2.0973429083824158,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2605738639831543,
+      "step": 2710
+    },
+    {
+      "epoch": 0.05424,
+      "grad_norm": 2.90625,
+      "grad_norm_var": 0.021219889322916668,
+      "learning_rate": 0.0001,
+      "loss": 5.2795,
+      "loss/crossentropy": 2.406570076942444,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2992263287305832,
+      "step": 2712
+    },
+    {
+      "epoch": 0.05428,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.024958292643229168,
+      "learning_rate": 0.0001,
+      "loss": 4.8591,
+      "loss/crossentropy": 2.040315330028534,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27726222574710846,
+      "step": 2714
+    },
+    {
+      "epoch": 0.05432,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.025178019205729166,
+      "learning_rate": 0.0001,
+      "loss": 4.7879,
+      "loss/crossentropy": 2.250051259994507,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2764698565006256,
+      "step": 2716
+    },
+    {
+      "epoch": 0.05436,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.026985677083333333,
+      "learning_rate": 0.0001,
+      "loss": 4.8813,
+      "loss/crossentropy": 2.25112247467041,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3101722151041031,
+      "step": 2718
+    },
+    {
+      "epoch": 0.0544,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.028709920247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.7242,
+      "loss/crossentropy": 2.261968731880188,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2805032432079315,
+      "step": 2720
+    },
+    {
+      "epoch": 0.05444,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.030078125,
+      "learning_rate": 0.0001,
+      "loss": 5.0449,
+      "loss/crossentropy": 2.376634955406189,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25400323420763016,
+      "step": 2722
+    },
+    {
+      "epoch": 0.05448,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.03351949055989583,
+      "learning_rate": 0.0001,
+      "loss": 5.2325,
+      "loss/crossentropy": 2.61246657371521,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3170415759086609,
+      "step": 2724
+    },
+    {
+      "epoch": 0.05452,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.0296051025390625,
+      "learning_rate": 0.0001,
+      "loss": 5.0433,
+      "loss/crossentropy": 2.3982752561569214,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2730572074651718,
+      "step": 2726
+    },
+    {
+      "epoch": 0.05456,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.0207916259765625,
+      "learning_rate": 0.0001,
+      "loss": 4.8836,
+      "loss/crossentropy": 1.9890516996383667,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26453813910484314,
+      "step": 2728
+    },
+    {
+      "epoch": 0.0546,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.016559855143229166,
+      "learning_rate": 0.0001,
+      "loss": 4.7252,
+      "loss/crossentropy": 2.1825047731399536,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28369753062725067,
+      "step": 2730
+    },
+    {
+      "epoch": 0.05464,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.015360514322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.9445,
+      "loss/crossentropy": 1.9745987057685852,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2662041634321213,
+      "step": 2732
+    },
+    {
+      "epoch": 0.05468,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.020783487955729166,
+      "learning_rate": 0.0001,
+      "loss": 4.7591,
+      "loss/crossentropy": 2.2962852716445923,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2986691743135452,
+      "step": 2734
+    },
+    {
+      "epoch": 0.05472,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.018342081705729166,
+      "learning_rate": 0.0001,
+      "loss": 4.9712,
+      "loss/crossentropy": 2.0517550110816956,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27293455600738525,
+      "step": 2736
+    },
+    {
+      "epoch": 0.05476,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.021434529622395834,
+      "learning_rate": 0.0001,
+      "loss": 5.4052,
+      "loss/crossentropy": 2.327734112739563,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3019224554300308,
+      "step": 2738
+    },
+    {
+      "epoch": 0.0548,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.018680826822916666,
+      "learning_rate": 0.0001,
+      "loss": 4.7002,
+      "loss/crossentropy": 2.236689567565918,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2816064953804016,
+      "step": 2740
+    },
+    {
+      "epoch": 0.05484,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.020894368489583332,
+      "learning_rate": 0.0001,
+      "loss": 5.1321,
+      "loss/crossentropy": 2.0209690928459167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2734442874789238,
+      "step": 2742
+    },
+    {
+      "epoch": 0.05488,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.026276652018229166,
+      "learning_rate": 0.0001,
+      "loss": 4.9841,
+      "loss/crossentropy": 2.2264864444732666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28523482382297516,
+      "step": 2744
+    },
+    {
+      "epoch": 0.05492,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.024738566080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.8661,
+      "loss/crossentropy": 1.9897980690002441,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26623376458883286,
+      "step": 2746
+    },
+    {
+      "epoch": 0.05496,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.024442545572916665,
+      "learning_rate": 0.0001,
+      "loss": 4.9299,
+      "loss/crossentropy": 2.0583779215812683,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2803298681974411,
+      "step": 2748
+    },
+    {
+      "epoch": 0.055,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.022001139322916665,
+      "learning_rate": 0.0001,
+      "loss": 4.9818,
+      "loss/crossentropy": 1.8448269367218018,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2606130689382553,
+      "step": 2750
+    },
+    {
+      "epoch": 0.05504,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.02301025390625,
+      "learning_rate": 0.0001,
+      "loss": 5.0885,
+      "loss/crossentropy": 2.294836401939392,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29403699934482574,
+      "step": 2752
+    },
+    {
+      "epoch": 0.05508,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.02808837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4637,
+      "loss/crossentropy": 2.2402058839797974,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28176650404930115,
+      "step": 2754
+    },
+    {
+      "epoch": 0.05512,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.0298980712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.7805,
+      "loss/crossentropy": 1.7882421612739563,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25275079905986786,
+      "step": 2756
+    },
+    {
+      "epoch": 0.05516,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.02919921875,
+      "learning_rate": 0.0001,
+      "loss": 5.1277,
+      "loss/crossentropy": 2.4185458421707153,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30663780868053436,
+      "step": 2758
+    },
+    {
+      "epoch": 0.0552,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.0318511962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.9259,
+      "loss/crossentropy": 2.2588642835617065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26785464584827423,
+      "step": 2760
+    },
+    {
+      "epoch": 0.05524,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.034012858072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.8359,
+      "loss/crossentropy": 2.145754337310791,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2710302472114563,
+      "step": 2762
+    },
+    {
+      "epoch": 0.05528,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.03648681640625,
+      "learning_rate": 0.0001,
+      "loss": 4.739,
+      "loss/crossentropy": 2.3627569675445557,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2926081120967865,
+      "step": 2764
+    },
+    {
+      "epoch": 0.05532,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.03632405598958333,
+      "learning_rate": 0.0001,
+      "loss": 4.8512,
+      "loss/crossentropy": 1.988103210926056,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24308288842439651,
+      "step": 2766
+    },
+    {
+      "epoch": 0.05536,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.040445963541666664,
+      "learning_rate": 0.0001,
+      "loss": 5.1075,
+      "loss/crossentropy": 2.2497689723968506,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.313574954867363,
+      "step": 2768
+    },
+    {
+      "epoch": 0.0554,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.030492146809895832,
+      "learning_rate": 0.0001,
+      "loss": 4.8978,
+      "loss/crossentropy": 2.2603683471679688,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27997657656669617,
+      "step": 2770
+    },
+    {
+      "epoch": 0.05544,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.030614217122395832,
+      "learning_rate": 0.0001,
+      "loss": 4.9485,
+      "loss/crossentropy": 2.2585065364837646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2818940281867981,
+      "step": 2772
+    },
+    {
+      "epoch": 0.05548,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.027730305989583332,
+      "learning_rate": 0.0001,
+      "loss": 5.2222,
+      "loss/crossentropy": 2.1413429975509644,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2950669527053833,
+      "step": 2774
+    },
+    {
+      "epoch": 0.05552,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.0171875,
+      "learning_rate": 0.0001,
+      "loss": 5.1567,
+      "loss/crossentropy": 1.9994583129882812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3946940451860428,
+      "step": 2776
+    },
+    {
+      "epoch": 0.05556,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.014762369791666667,
+      "learning_rate": 0.0001,
+      "loss": 5.2877,
+      "loss/crossentropy": 2.423824667930603,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30817069113254547,
+      "step": 2778
+    },
+    {
+      "epoch": 0.0556,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.013719685872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.6657,
+      "loss/crossentropy": 1.8579126000404358,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24312064796686172,
+      "step": 2780
+    },
+    {
+      "epoch": 0.05564,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.018485514322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.7653,
+      "loss/crossentropy": 2.3444939851760864,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28466545045375824,
+      "step": 2782
+    },
+    {
+      "epoch": 0.05568,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.03806864420572917,
+      "learning_rate": 0.0001,
+      "loss": 5.1187,
+      "loss/crossentropy": 2.221144914627075,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2694521099328995,
+      "step": 2784
+    },
+    {
+      "epoch": 0.05572,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.0376861572265625,
+      "learning_rate": 0.0001,
+      "loss": 5.0401,
+      "loss/crossentropy": 1.9372909665107727,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2698594778776169,
+      "step": 2786
+    },
+    {
+      "epoch": 0.05576,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.038834635416666666,
+      "learning_rate": 0.0001,
+      "loss": 4.834,
+      "loss/crossentropy": 2.129204750061035,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27333614230155945,
+      "step": 2788
+    },
+    {
+      "epoch": 0.0558,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.04045817057291667,
+      "learning_rate": 0.0001,
+      "loss": 4.8462,
+      "loss/crossentropy": 1.6917370557785034,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24290545284748077,
+      "step": 2790
+    },
+    {
+      "epoch": 0.05584,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.04096577962239583,
+      "learning_rate": 0.0001,
+      "loss": 4.6942,
+      "loss/crossentropy": 1.7883749604225159,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2554834187030792,
+      "step": 2792
+    },
+    {
+      "epoch": 0.05588,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.04560445149739583,
+      "learning_rate": 0.0001,
+      "loss": 4.6835,
+      "loss/crossentropy": 1.867136001586914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2667583078145981,
+      "step": 2794
+    },
+    {
+      "epoch": 0.05592,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.04143473307291667,
+      "learning_rate": 0.0001,
+      "loss": 4.9686,
+      "loss/crossentropy": 2.032800853252411,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2580890506505966,
+      "step": 2796
+    },
+    {
+      "epoch": 0.05596,
+      "grad_norm": 3.03125,
+      "grad_norm_var": 0.04389546712239583,
+      "learning_rate": 0.0001,
+      "loss": 5.018,
+      "loss/crossentropy": 1.9867863655090332,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2694792151451111,
+      "step": 2798
+    },
+    {
+      "epoch": 0.056,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.026786295572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.9137,
+      "loss/crossentropy": 2.1026757955551147,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2664051130414009,
+      "step": 2800
+    },
+    {
+      "epoch": 0.05604,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.0259918212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.9883,
+      "loss/crossentropy": 2.0649160742759705,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2473129704594612,
+      "step": 2802
+    },
+    {
+      "epoch": 0.05608,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.030451456705729168,
+      "learning_rate": 0.0001,
+      "loss": 4.8959,
+      "loss/crossentropy": 2.2110248804092407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27131715416908264,
+      "step": 2804
+    },
+    {
+      "epoch": 0.05612,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.032933553059895836,
+      "learning_rate": 0.0001,
+      "loss": 4.9603,
+      "loss/crossentropy": 2.3483060598373413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27071496844291687,
+      "step": 2806
+    },
+    {
+      "epoch": 0.05616,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.03277587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.9273,
+      "loss/crossentropy": 1.9061944484710693,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2550426125526428,
+      "step": 2808
+    },
+    {
+      "epoch": 0.0562,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.0278472900390625,
+      "learning_rate": 0.0001,
+      "loss": 5.1734,
+      "loss/crossentropy": 2.3073103427886963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30051568150520325,
+      "step": 2810
+    },
+    {
+      "epoch": 0.05624,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.027106730143229167,
+      "learning_rate": 0.0001,
+      "loss": 5.0268,
+      "loss/crossentropy": 2.393891453742981,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29226459562778473,
+      "step": 2812
+    },
+    {
+      "epoch": 0.05628,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.018578084309895833,
+      "learning_rate": 0.0001,
+      "loss": 5.0839,
+      "loss/crossentropy": 2.3082761764526367,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29539754986763,
+      "step": 2814
+    },
+    {
+      "epoch": 0.05632,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.021776326497395835,
+      "learning_rate": 0.0001,
+      "loss": 4.8973,
+      "loss/crossentropy": 2.7815494537353516,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33258646726608276,
+      "step": 2816
+    },
+    {
+      "epoch": 0.05636,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.031538899739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.7095,
+      "loss/crossentropy": 2.077984571456909,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.261022225022316,
+      "step": 2818
+    },
+    {
+      "epoch": 0.0564,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.0251617431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.6389,
+      "loss/crossentropy": 2.0524688363075256,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2936979830265045,
+      "step": 2820
+    },
+    {
+      "epoch": 0.05644,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.018212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.9657,
+      "loss/crossentropy": 1.8323140740394592,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26426824927330017,
+      "step": 2822
+    },
+    {
+      "epoch": 0.05648,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.0201171875,
+      "learning_rate": 0.0001,
+      "loss": 4.9513,
+      "loss/crossentropy": 2.2290462255477905,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28852027654647827,
+      "step": 2824
+    },
+    {
+      "epoch": 0.05652,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.016600545247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.7828,
+      "loss/crossentropy": 1.8788678050041199,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24312467128038406,
+      "step": 2826
+    },
+    {
+      "epoch": 0.05656,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.01646728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.7882,
+      "loss/crossentropy": 1.9402090311050415,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25435876101255417,
+      "step": 2828
+    },
+    {
+      "epoch": 0.0566,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.0144683837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5823,
+      "loss/crossentropy": 2.4833847284317017,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27234241366386414,
+      "step": 2830
+    },
+    {
+      "epoch": 0.05664,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.0154296875,
+      "learning_rate": 0.0001,
+      "loss": 5.0521,
+      "loss/crossentropy": 2.351561665534973,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28028184175491333,
+      "step": 2832
+    },
+    {
+      "epoch": 0.05668,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.019722493489583333,
+      "learning_rate": 0.0001,
+      "loss": 5.0688,
+      "loss/crossentropy": 2.0372352600097656,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26005299389362335,
+      "step": 2834
+    },
+    {
+      "epoch": 0.05672,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.021744791666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.4282,
+      "loss/crossentropy": 1.6522082090377808,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26683834940195084,
+      "step": 2836
+    },
+    {
+      "epoch": 0.05676,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.023778279622395832,
+      "learning_rate": 0.0001,
+      "loss": 4.7165,
+      "loss/crossentropy": 1.8886643052101135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23464814573526382,
+      "step": 2838
+    },
+    {
+      "epoch": 0.0568,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.023119099934895835,
+      "learning_rate": 0.0001,
+      "loss": 4.9477,
+      "loss/crossentropy": 2.2161459922790527,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29381541907787323,
+      "step": 2840
+    },
+    {
+      "epoch": 0.05684,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.024779256184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.7408,
+      "loss/crossentropy": 2.1542043685913086,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2753119468688965,
+      "step": 2842
+    },
+    {
+      "epoch": 0.05688,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.026090494791666665,
+      "learning_rate": 0.0001,
+      "loss": 5.1288,
+      "loss/crossentropy": 2.5605628490448,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.329460546374321,
+      "step": 2844
+    },
+    {
+      "epoch": 0.05692,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.023224894205729166,
+      "learning_rate": 0.0001,
+      "loss": 4.9955,
+      "loss/crossentropy": 2.1000319719314575,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.273986279964447,
+      "step": 2846
+    },
+    {
+      "epoch": 0.05696,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.023558553059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.6813,
+      "loss/crossentropy": 2.1036760210990906,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26293954253196716,
+      "step": 2848
+    },
+    {
+      "epoch": 0.057,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.014286295572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.9799,
+      "loss/crossentropy": 2.2130608558654785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2977828085422516,
+      "step": 2850
+    },
+    {
+      "epoch": 0.05704,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.01461181640625,
+      "learning_rate": 0.0001,
+      "loss": 5.1761,
+      "loss/crossentropy": 2.1878823041915894,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27322643995285034,
+      "step": 2852
+    },
+    {
+      "epoch": 0.05708,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.012547810872395834,
+      "learning_rate": 0.0001,
+      "loss": 4.9448,
+      "loss/crossentropy": 2.1559258699417114,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32625503838062286,
+      "step": 2854
+    },
+    {
+      "epoch": 0.05712,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.015848795572916668,
+      "learning_rate": 0.0001,
+      "loss": 4.4611,
+      "loss/crossentropy": 2.0818992257118225,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2677721679210663,
+      "step": 2856
+    },
+    {
+      "epoch": 0.05716,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.016243489583333333,
+      "learning_rate": 0.0001,
+      "loss": 5.2755,
+      "loss/crossentropy": 2.544907331466675,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30028045177459717,
+      "step": 2858
+    },
+    {
+      "epoch": 0.0572,
+      "grad_norm": 3.203125,
+      "grad_norm_var": 0.04491780598958333,
+      "learning_rate": 0.0001,
+      "loss": 5.2855,
+      "loss/crossentropy": 2.5932188034057617,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30431173741817474,
+      "step": 2860
+    },
+    {
+      "epoch": 0.05724,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.046219889322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.9799,
+      "loss/crossentropy": 2.1007986068725586,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2989690601825714,
+      "step": 2862
+    },
+    {
+      "epoch": 0.05728,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.044169108072916664,
+      "learning_rate": 0.0001,
+      "loss": 4.7706,
+      "loss/crossentropy": 2.0102875232696533,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26936179399490356,
+      "step": 2864
+    },
+    {
+      "epoch": 0.05732,
+      "grad_norm": 3.109375,
+      "grad_norm_var": 0.056559244791666664,
+      "learning_rate": 0.0001,
+      "loss": 5.2099,
+      "loss/crossentropy": 2.3457159996032715,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29194609820842743,
+      "step": 2866
+    },
+    {
+      "epoch": 0.05736,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.055582682291666664,
+      "learning_rate": 0.0001,
+      "loss": 4.8768,
+      "loss/crossentropy": 2.559054732322693,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3018783777952194,
+      "step": 2868
+    },
+    {
+      "epoch": 0.0574,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.05831705729166667,
+      "learning_rate": 0.0001,
+      "loss": 4.9863,
+      "loss/crossentropy": 2.0641059279441833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2789234220981598,
+      "step": 2870
+    },
+    {
+      "epoch": 0.05744,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.04683837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.9222,
+      "loss/crossentropy": 2.0819749236106873,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26825186610221863,
+      "step": 2872
+    },
+    {
+      "epoch": 0.05748,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.046263631184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.9586,
+      "loss/crossentropy": 2.1114020347595215,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2613547742366791,
+      "step": 2874
+    },
+    {
+      "epoch": 0.05752,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.02916259765625,
+      "learning_rate": 0.0001,
+      "loss": 4.9474,
+      "loss/crossentropy": 2.174915075302124,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29503974318504333,
+      "step": 2876
+    },
+    {
+      "epoch": 0.05756,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.0265289306640625,
+      "learning_rate": 0.0001,
+      "loss": 5.0541,
+      "loss/crossentropy": 2.4342113733291626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30623678863048553,
+      "step": 2878
+    },
+    {
+      "epoch": 0.0576,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.024918619791666666,
+      "learning_rate": 0.0001,
+      "loss": 5.0316,
+      "loss/crossentropy": 2.0518307089805603,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28422991931438446,
+      "step": 2880
+    },
+    {
+      "epoch": 0.05764,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.011921183268229166,
+      "learning_rate": 0.0001,
+      "loss": 5.0806,
+      "loss/crossentropy": 2.5378612279891968,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29730531573295593,
+      "step": 2882
+    },
+    {
+      "epoch": 0.05768,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.014972941080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.8685,
+      "loss/crossentropy": 2.2670027017593384,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26347288489341736,
+      "step": 2884
+    },
+    {
+      "epoch": 0.05772,
+      "grad_norm": 3.671875,
+      "grad_norm_var": 0.07822265625,
+      "learning_rate": 0.0001,
+      "loss": 5.0385,
+      "loss/crossentropy": 2.351823568344116,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3020637035369873,
+      "step": 2886
+    },
+    {
+      "epoch": 0.05776,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.0792633056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.9699,
+      "loss/crossentropy": 2.190839111804962,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2639586254954338,
+      "step": 2888
+    },
+    {
+      "epoch": 0.0578,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.08765869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.6882,
+      "loss/crossentropy": 2.148400902748108,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27762140333652496,
+      "step": 2890
+    },
+    {
+      "epoch": 0.05784,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.09696858723958333,
+      "learning_rate": 0.0001,
+      "loss": 5.1033,
+      "loss/crossentropy": 2.01130074262619,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2678648605942726,
+      "step": 2892
+    },
+    {
+      "epoch": 0.05788,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.09986572265625,
+      "learning_rate": 0.0001,
+      "loss": 5.0829,
+      "loss/crossentropy": 2.2239269018173218,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2971164286136627,
+      "step": 2894
+    },
+    {
+      "epoch": 0.05792,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.10974934895833334,
+      "learning_rate": 0.0001,
+      "loss": 4.48,
+      "loss/crossentropy": 1.9573850631713867,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.241651751101017,
+      "step": 2896
+    },
+    {
+      "epoch": 0.05796,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.11169331868489583,
+      "learning_rate": 0.0001,
+      "loss": 4.8806,
+      "loss/crossentropy": 1.9522782564163208,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2685594707727432,
+      "step": 2898
+    },
+    {
+      "epoch": 0.058,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.10886128743489583,
+      "learning_rate": 0.0001,
+      "loss": 4.9335,
+      "loss/crossentropy": 2.1742069721221924,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27867090702056885,
+      "step": 2900
+    },
+    {
+      "epoch": 0.05804,
+      "grad_norm": 5.15625,
+      "grad_norm_var": 0.44882405598958336,
+      "learning_rate": 0.0001,
+      "loss": 4.9164,
+      "loss/crossentropy": 2.201782703399658,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2731190174818039,
+      "step": 2902
+    },
+    {
+      "epoch": 0.05808,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.4452301025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.9787,
+      "loss/crossentropy": 2.159709095954895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2940017879009247,
+      "step": 2904
+    },
+    {
+      "epoch": 0.05812,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.4297271728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.9192,
+      "loss/crossentropy": 2.1989885568618774,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2776087671518326,
+      "step": 2906
+    },
+    {
+      "epoch": 0.05816,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.42688395182291666,
+      "learning_rate": 0.0001,
+      "loss": 4.8851,
+      "loss/crossentropy": 1.9634575247764587,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27917809784412384,
+      "step": 2908
+    },
+    {
+      "epoch": 0.0582,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.43651936848958334,
+      "learning_rate": 0.0001,
+      "loss": 4.7776,
+      "loss/crossentropy": 2.1553479433059692,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28876082599163055,
+      "step": 2910
+    },
+    {
+      "epoch": 0.05824,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.42451883951822916,
+      "learning_rate": 0.0001,
+      "loss": 4.881,
+      "loss/crossentropy": 2.1035598516464233,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2851293236017227,
+      "step": 2912
+    },
+    {
+      "epoch": 0.05828,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 0.42097880045572916,
+      "learning_rate": 0.0001,
+      "loss": 5.2733,
+      "loss/crossentropy": 2.17076575756073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28258734941482544,
+      "step": 2914
+    },
+    {
+      "epoch": 0.05832,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.42477213541666664,
+      "learning_rate": 0.0001,
+      "loss": 4.9147,
+      "loss/crossentropy": 2.2914711236953735,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29451698064804077,
+      "step": 2916
+    },
+    {
+      "epoch": 0.05836,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.021410115559895835,
+      "learning_rate": 0.0001,
+      "loss": 5.1395,
+      "loss/crossentropy": 2.554638981819153,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30992935597896576,
+      "step": 2918
+    },
+    {
+      "epoch": 0.0584,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.05735575358072917,
+      "learning_rate": 0.0001,
+      "loss": 4.6262,
+      "loss/crossentropy": 1.808376431465149,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22574126720428467,
+      "step": 2920
+    },
+    {
+      "epoch": 0.05844,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.05660400390625,
+      "learning_rate": 0.0001,
+      "loss": 5.1636,
+      "loss/crossentropy": 2.1471784114837646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27325738966464996,
+      "step": 2922
+    },
+    {
+      "epoch": 0.05848,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.054182942708333334,
+      "learning_rate": 0.0001,
+      "loss": 5.071,
+      "loss/crossentropy": 2.2175731658935547,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3013547956943512,
+      "step": 2924
+    },
+    {
+      "epoch": 0.05852,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.04949544270833333,
+      "learning_rate": 0.0001,
+      "loss": 5.467,
+      "loss/crossentropy": 2.369373917579651,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29249751567840576,
+      "step": 2926
+    },
+    {
+      "epoch": 0.05856,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.054915364583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.8778,
+      "loss/crossentropy": 2.1758522987365723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28846102952957153,
+      "step": 2928
+    },
+    {
+      "epoch": 0.0586,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.05413411458333333,
+      "learning_rate": 0.0001,
+      "loss": 4.929,
+      "loss/crossentropy": 2.46218478679657,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2824682295322418,
+      "step": 2930
+    },
+    {
+      "epoch": 0.05864,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.0606597900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.5557,
+      "loss/crossentropy": 2.058937907218933,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26767465472221375,
+      "step": 2932
+    },
+    {
+      "epoch": 0.05868,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.06109619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.9103,
+      "loss/crossentropy": 2.312442421913147,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2952658236026764,
+      "step": 2934
+    },
+    {
+      "epoch": 0.05872,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.026911417643229168,
+      "learning_rate": 0.0001,
+      "loss": 5.1452,
+      "loss/crossentropy": 2.18839955329895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31788623332977295,
+      "step": 2936
+    },
+    {
+      "epoch": 0.05876,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.025537109375,
+      "learning_rate": 0.0001,
+      "loss": 5.1358,
+      "loss/crossentropy": 2.1330811977386475,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28829698264598846,
+      "step": 2938
+    },
+    {
+      "epoch": 0.0588,
+      "grad_norm": 2.9375,
+      "grad_norm_var": 0.0294586181640625,
+      "learning_rate": 0.0001,
+      "loss": 5.1071,
+      "loss/crossentropy": 2.124038338661194,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2708826810121536,
+      "step": 2940
+    },
+    {
+      "epoch": 0.05884,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.024312337239583332,
+      "learning_rate": 0.0001,
+      "loss": 4.8479,
+      "loss/crossentropy": 2.1934449076652527,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27314090728759766,
+      "step": 2942
+    },
+    {
+      "epoch": 0.05888,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.024535115559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.9904,
+      "loss/crossentropy": 1.967636525630951,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23761005699634552,
+      "step": 2944
+    },
+    {
+      "epoch": 0.05892,
+      "grad_norm": 3.015625,
+      "grad_norm_var": 0.034821573893229166,
+      "learning_rate": 0.0001,
+      "loss": 4.9214,
+      "loss/crossentropy": 2.2380826473236084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3402934819459915,
+      "step": 2946
+    },
+    {
+      "epoch": 0.05896,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.0299468994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.7702,
+      "loss/crossentropy": 2.227039933204651,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27934837341308594,
+      "step": 2948
+    },
+    {
+      "epoch": 0.059,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.03258056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.789,
+      "loss/crossentropy": 1.999170958995819,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.275806725025177,
+      "step": 2950
+    },
+    {
+      "epoch": 0.05904,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.03241780598958333,
+      "learning_rate": 0.0001,
+      "loss": 4.5754,
+      "loss/crossentropy": 1.8843520879745483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26203446090221405,
+      "step": 2952
+    },
+    {
+      "epoch": 0.05908,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.033426920572916664,
+      "learning_rate": 0.0001,
+      "loss": 4.5485,
+      "loss/crossentropy": 2.0742241740226746,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28777699172496796,
+      "step": 2954
+    },
+    {
+      "epoch": 0.05912,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.028206380208333333,
+      "learning_rate": 0.0001,
+      "loss": 4.9676,
+      "loss/crossentropy": 2.2272751331329346,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27761510014533997,
+      "step": 2956
+    },
+    {
+      "epoch": 0.05916,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.029878743489583335,
+      "learning_rate": 0.0001,
+      "loss": 4.6329,
+      "loss/crossentropy": 2.2758638858795166,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28144824504852295,
+      "step": 2958
+    },
+    {
+      "epoch": 0.0592,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.060530598958333334,
+      "learning_rate": 0.0001,
+      "loss": 4.9543,
+      "loss/crossentropy": 2.245158016681671,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3030128926038742,
+      "step": 2960
+    },
+    {
+      "epoch": 0.05924,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.053498331705729166,
+      "learning_rate": 0.0001,
+      "loss": 4.567,
+      "loss/crossentropy": 2.3337708711624146,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29228493571281433,
+      "step": 2962
+    },
+    {
+      "epoch": 0.05928,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.05788472493489583,
+      "learning_rate": 0.0001,
+      "loss": 4.6477,
+      "loss/crossentropy": 2.3765406608581543,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2715871036052704,
+      "step": 2964
+    },
+    {
+      "epoch": 0.05932,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.0562652587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.8649,
+      "loss/crossentropy": 2.090362787246704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2505037933588028,
+      "step": 2966
+    },
+    {
+      "epoch": 0.05936,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.056396484375,
+      "learning_rate": 0.0001,
+      "loss": 4.8669,
+      "loss/crossentropy": 2.011539399623871,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27219071984291077,
+      "step": 2968
+    },
+    {
+      "epoch": 0.0594,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.05693359375,
+      "learning_rate": 0.0001,
+      "loss": 5.2081,
+      "loss/crossentropy": 2.1791563034057617,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2796429842710495,
+      "step": 2970
+    },
+    {
+      "epoch": 0.05944,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.05869852701822917,
+      "learning_rate": 0.0001,
+      "loss": 5.2689,
+      "loss/crossentropy": 2.4645297527313232,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3106658458709717,
+      "step": 2972
+    },
+    {
+      "epoch": 0.05948,
+      "grad_norm": 7.0,
+      "grad_norm_var": 1.25670166015625,
+      "learning_rate": 0.0001,
+      "loss": 5.0715,
+      "loss/crossentropy": 2.2050880193710327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26860976219177246,
+      "step": 2974
+    },
+    {
+      "epoch": 0.05952,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 1.236034138997396,
+      "learning_rate": 0.0001,
+      "loss": 4.5815,
+      "loss/crossentropy": 2.0141921639442444,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27980539202690125,
+      "step": 2976
+    },
+    {
+      "epoch": 0.05956,
+      "grad_norm": 4.4375,
+      "grad_norm_var": 1.3665924072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.8067,
+      "loss/crossentropy": 1.9399088025093079,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26379524916410446,
+      "step": 2978
+    },
+    {
+      "epoch": 0.0596,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 1.3243886311848958,
+      "learning_rate": 0.0001,
+      "loss": 5.1743,
+      "loss/crossentropy": 2.5415157079696655,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.291620597243309,
+      "step": 2980
+    },
+    {
+      "epoch": 0.05964,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 1.3170562744140626,
+      "learning_rate": 0.0001,
+      "loss": 5.0695,
+      "loss/crossentropy": 2.1215542554855347,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26189403235912323,
+      "step": 2982
+    },
+    {
+      "epoch": 0.05968,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 1.32301025390625,
+      "learning_rate": 0.0001,
+      "loss": 5.0049,
+      "loss/crossentropy": 1.7749422788619995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27546317875385284,
+      "step": 2984
+    },
+    {
+      "epoch": 0.05972,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 1.3371490478515624,
+      "learning_rate": 0.0001,
+      "loss": 4.8331,
+      "loss/crossentropy": 2.3383208513259888,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2790983319282532,
+      "step": 2986
+    },
+    {
+      "epoch": 0.05976,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 1.35230712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.8656,
+      "loss/crossentropy": 2.3688780069351196,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2752939611673355,
+      "step": 2988
+    },
+    {
+      "epoch": 0.0598,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.22967020670572916,
+      "learning_rate": 0.0001,
+      "loss": 4.7299,
+      "loss/crossentropy": 2.261468529701233,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2725592106580734,
+      "step": 2990
+    },
+    {
+      "epoch": 0.05984,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.2226226806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.7417,
+      "loss/crossentropy": 2.0632028579711914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2617062032222748,
+      "step": 2992
+    },
+    {
+      "epoch": 0.05988,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.022151692708333334,
+      "learning_rate": 0.0001,
+      "loss": 4.4282,
+      "loss/crossentropy": 2.1409813165664673,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2657589614391327,
+      "step": 2994
+    },
+    {
+      "epoch": 0.05992,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.0258941650390625,
+      "learning_rate": 0.0001,
+      "loss": 5.1816,
+      "loss/crossentropy": 2.431404948234558,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3087555915117264,
+      "step": 2996
+    },
+    {
+      "epoch": 0.05996,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.029743448893229166,
+      "learning_rate": 0.0001,
+      "loss": 4.8253,
+      "loss/crossentropy": 2.165238618850708,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29382775723934174,
+      "step": 2998
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.034566243489583336,
+      "learning_rate": 0.0001,
+      "loss": 4.6328,
+      "loss/crossentropy": 1.9987847208976746,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24466252326965332,
+      "step": 3000
+    },
+    {
+      "epoch": 0.06004,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.04084370930989583,
+      "learning_rate": 0.0001,
+      "loss": 4.5474,
+      "loss/crossentropy": 2.1153565049171448,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2435970976948738,
+      "step": 3002
+    },
+    {
+      "epoch": 0.06008,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.046402994791666666,
+      "learning_rate": 0.0001,
+      "loss": 4.7765,
+      "loss/crossentropy": 1.8660866618156433,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24413185566663742,
+      "step": 3004
+    },
+    {
+      "epoch": 0.06012,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.04309794108072917,
+      "learning_rate": 0.0001,
+      "loss": 4.7476,
+      "loss/crossentropy": 2.1816678047180176,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28010787069797516,
+      "step": 3006
+    },
+    {
+      "epoch": 0.06016,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.040827433268229164,
+      "learning_rate": 0.0001,
+      "loss": 4.5995,
+      "loss/crossentropy": 2.1673622131347656,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2659531533718109,
+      "step": 3008
+    },
+    {
+      "epoch": 0.0602,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.040913899739583336,
+      "learning_rate": 0.0001,
+      "loss": 5.2466,
+      "loss/crossentropy": 2.277818202972412,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2929365783929825,
+      "step": 3010
+    },
+    {
+      "epoch": 0.06024,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.0284576416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.8904,
+      "loss/crossentropy": 2.294926404953003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2753874659538269,
+      "step": 3012
+    },
+    {
+      "epoch": 0.06028,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.028173828125,
+      "learning_rate": 0.0001,
+      "loss": 4.735,
+      "loss/crossentropy": 2.0222257375717163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27295801043510437,
+      "step": 3014
+    },
+    {
+      "epoch": 0.06032,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.023628743489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.818,
+      "loss/crossentropy": 2.43736469745636,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2934436649084091,
+      "step": 3016
+    },
+    {
+      "epoch": 0.06036,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.020426432291666668,
+      "learning_rate": 0.0001,
+      "loss": 5.0053,
+      "loss/crossentropy": 2.027767241001129,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26353244483470917,
+      "step": 3018
+    },
+    {
+      "epoch": 0.0604,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.015034993489583334,
+      "learning_rate": 0.0001,
+      "loss": 4.9059,
+      "loss/crossentropy": 2.150290071964264,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27290941774845123,
+      "step": 3020
+    },
+    {
+      "epoch": 0.06044,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.014697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.8307,
+      "loss/crossentropy": 1.8709848523139954,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25272610783576965,
+      "step": 3022
+    },
+    {
+      "epoch": 0.06048,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.015949503580729166,
+      "learning_rate": 0.0001,
+      "loss": 5.0166,
+      "loss/crossentropy": 2.2170883417129517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29401010274887085,
+      "step": 3024
+    },
+    {
+      "epoch": 0.06052,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.0109375,
+      "learning_rate": 0.0001,
+      "loss": 4.6221,
+      "loss/crossentropy": 1.728318691253662,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24246910959482193,
+      "step": 3026
+    },
+    {
+      "epoch": 0.06056,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.0095123291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.5597,
+      "loss/crossentropy": 1.8453290462493896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24357211589813232,
+      "step": 3028
+    },
+    {
+      "epoch": 0.0606,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.010456339518229166,
+      "learning_rate": 0.0001,
+      "loss": 4.6617,
+      "loss/crossentropy": 2.273196220397949,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29239149391651154,
+      "step": 3030
+    },
+    {
+      "epoch": 0.06064,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.012165323893229166,
+      "learning_rate": 0.0001,
+      "loss": 4.9464,
+      "loss/crossentropy": 2.031871259212494,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2604901194572449,
+      "step": 3032
+    },
+    {
+      "epoch": 0.06068,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.008915201822916666,
+      "learning_rate": 0.0001,
+      "loss": 5.1618,
+      "loss/crossentropy": 2.1781771183013916,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2853371948003769,
+      "step": 3034
+    },
+    {
+      "epoch": 0.06072,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.006590779622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.816,
+      "loss/crossentropy": 1.9578353762626648,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23352904617786407,
+      "step": 3036
+    },
+    {
+      "epoch": 0.06076,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.012516276041666666,
+      "learning_rate": 0.0001,
+      "loss": 5.3012,
+      "loss/crossentropy": 2.470985770225525,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28890977799892426,
+      "step": 3038
+    },
+    {
+      "epoch": 0.0608,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.013361612955729166,
+      "learning_rate": 0.0001,
+      "loss": 4.9243,
+      "loss/crossentropy": 2.056231141090393,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2963344603776932,
+      "step": 3040
+    },
+    {
+      "epoch": 0.06084,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.013646443684895834,
+      "learning_rate": 0.0001,
+      "loss": 4.8612,
+      "loss/crossentropy": 2.000037968158722,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27152783423662186,
+      "step": 3042
+    },
+    {
+      "epoch": 0.06088,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.0111480712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.8576,
+      "loss/crossentropy": 2.323809027671814,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29995349049568176,
+      "step": 3044
+    },
+    {
+      "epoch": 0.06092,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.010407511393229167,
+      "learning_rate": 0.0001,
+      "loss": 5.0094,
+      "loss/crossentropy": 2.06933856010437,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26831966638565063,
+      "step": 3046
+    },
+    {
+      "epoch": 0.06096,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.008463541666666666,
+      "learning_rate": 0.0001,
+      "loss": 5.042,
+      "loss/crossentropy": 2.1903880834579468,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2833031117916107,
+      "step": 3048
+    },
+    {
+      "epoch": 0.061,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.008893839518229167,
+      "learning_rate": 0.0001,
+      "loss": 4.9557,
+      "loss/crossentropy": 1.910680890083313,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2708408683538437,
+      "step": 3050
+    },
+    {
+      "epoch": 0.06104,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.0083404541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.8902,
+      "loss/crossentropy": 2.5203059911727905,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2981104403734207,
+      "step": 3052
+    },
+    {
+      "epoch": 0.06108,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.00611572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.68,
+      "loss/crossentropy": 1.7918179035186768,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2462354451417923,
+      "step": 3054
+    },
+    {
+      "epoch": 0.06112,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.006441243489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.9908,
+      "loss/crossentropy": 2.185121774673462,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3046490252017975,
+      "step": 3056
+    },
+    {
+      "epoch": 0.06116,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.01064453125,
+      "learning_rate": 0.0001,
+      "loss": 4.8041,
+      "loss/crossentropy": 1.859390914440155,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2523205131292343,
+      "step": 3058
+    },
+    {
+      "epoch": 0.0612,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.011930338541666667,
+      "learning_rate": 0.0001,
+      "loss": 4.7388,
+      "loss/crossentropy": 1.9202255606651306,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26971636712551117,
+      "step": 3060
+    },
+    {
+      "epoch": 0.06124,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.011481730143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.7421,
+      "loss/crossentropy": 2.059127449989319,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25721821188926697,
+      "step": 3062
+    },
+    {
+      "epoch": 0.06128,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.0154205322265625,
+      "learning_rate": 0.0001,
+      "loss": 5.1392,
+      "loss/crossentropy": 2.3587669134140015,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30162671208381653,
+      "step": 3064
+    },
+    {
+      "epoch": 0.06132,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.018050130208333334,
+      "learning_rate": 0.0001,
+      "loss": 5.2748,
+      "loss/crossentropy": 2.239704966545105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2713697552680969,
+      "step": 3066
+    },
+    {
+      "epoch": 0.06136,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.02076416015625,
+      "learning_rate": 0.0001,
+      "loss": 5.0534,
+      "loss/crossentropy": 2.230944514274597,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27418144047260284,
+      "step": 3068
+    },
+    {
+      "epoch": 0.0614,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.020563761393229168,
+      "learning_rate": 0.0001,
+      "loss": 4.9145,
+      "loss/crossentropy": 2.196586310863495,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24946682900190353,
+      "step": 3070
+    },
+    {
+      "epoch": 0.06144,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.019873046875,
+      "learning_rate": 0.0001,
+      "loss": 5.058,
+      "loss/crossentropy": 2.115275800228119,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2668849602341652,
+      "step": 3072
+    },
+    {
+      "epoch": 0.06148,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.014286295572916666,
+      "learning_rate": 0.0001,
+      "loss": 5.1311,
+      "loss/crossentropy": 2.0227994322776794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2587117701768875,
+      "step": 3074
+    },
+    {
+      "epoch": 0.06152,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.016022745768229166,
+      "learning_rate": 0.0001,
+      "loss": 4.711,
+      "loss/crossentropy": 2.231368660926819,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2621329501271248,
+      "step": 3076
+    },
+    {
+      "epoch": 0.06156,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.0190093994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.6707,
+      "loss/crossentropy": 2.469847083091736,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2758704572916031,
+      "step": 3078
+    },
+    {
+      "epoch": 0.0616,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.01666259765625,
+      "learning_rate": 0.0001,
+      "loss": 4.5753,
+      "loss/crossentropy": 2.257850766181946,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28196004033088684,
+      "step": 3080
+    },
+    {
+      "epoch": 0.06164,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.014969889322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.8783,
+      "loss/crossentropy": 2.52754008769989,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.301498681306839,
+      "step": 3082
+    },
+    {
+      "epoch": 0.06168,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.012279256184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.6595,
+      "loss/crossentropy": 2.120129644870758,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30073782801628113,
+      "step": 3084
+    },
+    {
+      "epoch": 0.06172,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.0231353759765625,
+      "learning_rate": 0.0001,
+      "loss": 5.017,
+      "loss/crossentropy": 2.1483632922172546,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29400819540023804,
+      "step": 3086
+    },
+    {
+      "epoch": 0.06176,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.022297159830729166,
+      "learning_rate": 0.0001,
+      "loss": 5.2447,
+      "loss/crossentropy": 2.5588048696517944,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3099561035633087,
+      "step": 3088
+    },
+    {
+      "epoch": 0.0618,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.019852701822916666,
+      "learning_rate": 0.0001,
+      "loss": 4.5507,
+      "loss/crossentropy": 2.338167190551758,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2638262137770653,
+      "step": 3090
+    },
+    {
+      "epoch": 0.06184,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.019603474934895834,
+      "learning_rate": 0.0001,
+      "loss": 5.0471,
+      "loss/crossentropy": 2.61361825466156,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29420773684978485,
+      "step": 3092
+    },
+    {
+      "epoch": 0.06188,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.019108072916666666,
+      "learning_rate": 0.0001,
+      "loss": 4.6931,
+      "loss/crossentropy": 2.1078842878341675,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27283959090709686,
+      "step": 3094
+    },
+    {
+      "epoch": 0.06192,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.023958333333333335,
+      "learning_rate": 0.0001,
+      "loss": 4.463,
+      "loss/crossentropy": 2.1501123905181885,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3052368611097336,
+      "step": 3096
+    },
+    {
+      "epoch": 0.06196,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.026656087239583334,
+      "learning_rate": 0.0001,
+      "loss": 5.0482,
+      "loss/crossentropy": 2.3622714281082153,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2851791977882385,
+      "step": 3098
+    },
+    {
+      "epoch": 0.062,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.05203450520833333,
+      "learning_rate": 0.0001,
+      "loss": 4.3506,
+      "loss/crossentropy": 1.9757406115531921,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24451570957899094,
+      "step": 3100
+    },
+    {
+      "epoch": 0.06204,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.07011311848958333,
+      "learning_rate": 0.0001,
+      "loss": 5.0409,
+      "loss/crossentropy": 2.1915838718414307,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2749434858560562,
+      "step": 3102
+    },
+    {
+      "epoch": 0.06208,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.07214253743489583,
+      "learning_rate": 0.0001,
+      "loss": 4.8408,
+      "loss/crossentropy": 2.130649447441101,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2590179592370987,
+      "step": 3104
+    },
+    {
+      "epoch": 0.06212,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.06317952473958334,
+      "learning_rate": 0.0001,
+      "loss": 4.9083,
+      "loss/crossentropy": 2.4478825330734253,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30854369699954987,
+      "step": 3106
+    },
+    {
+      "epoch": 0.06216,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.06524149576822917,
+      "learning_rate": 0.0001,
+      "loss": 4.8249,
+      "loss/crossentropy": 1.7108858227729797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22502756118774414,
+      "step": 3108
+    },
+    {
+      "epoch": 0.0622,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.06788736979166667,
+      "learning_rate": 0.0001,
+      "loss": 4.5355,
+      "loss/crossentropy": 1.79804128408432,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24767793715000153,
+      "step": 3110
+    },
+    {
+      "epoch": 0.06224,
+      "grad_norm": 3.28125,
+      "grad_norm_var": 0.07480061848958333,
+      "learning_rate": 0.0001,
+      "loss": 4.8779,
+      "loss/crossentropy": 1.899846076965332,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23476070165634155,
+      "step": 3112
+    },
+    {
+      "epoch": 0.06228,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.0790679931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.7506,
+      "loss/crossentropy": 1.8458155393600464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24417708814144135,
+      "step": 3114
+    },
+    {
+      "epoch": 0.06232,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.06979166666666667,
+      "learning_rate": 0.0001,
+      "loss": 4.814,
+      "loss/crossentropy": 1.947302520275116,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2566901594400406,
+      "step": 3116
+    },
+    {
+      "epoch": 0.06236,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.046923828125,
+      "learning_rate": 0.0001,
+      "loss": 4.8424,
+      "loss/crossentropy": 2.039161205291748,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26955537497997284,
+      "step": 3118
+    },
+    {
+      "epoch": 0.0624,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.04488016764322917,
+      "learning_rate": 0.0001,
+      "loss": 5.1656,
+      "loss/crossentropy": 2.1528661251068115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.291052982211113,
+      "step": 3120
+    },
+    {
+      "epoch": 0.06244,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.04644775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.7941,
+      "loss/crossentropy": 2.4113996028900146,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2962254136800766,
+      "step": 3122
+    },
+    {
+      "epoch": 0.06248,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.044722493489583334,
+      "learning_rate": 0.0001,
+      "loss": 4.8463,
+      "loss/crossentropy": 2.0718825459480286,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24676042795181274,
+      "step": 3124
+    },
+    {
+      "epoch": 0.06252,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.04442952473958333,
+      "learning_rate": 0.0001,
+      "loss": 4.9723,
+      "loss/crossentropy": 2.433539032936096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.263284370303154,
+      "step": 3126
+    },
+    {
+      "epoch": 0.06256,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.0087799072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.7688,
+      "loss/crossentropy": 2.047150671482086,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2655043303966522,
+      "step": 3128
+    },
+    {
+      "epoch": 0.0626,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.008219401041666666,
+      "learning_rate": 0.0001,
+      "loss": 4.9257,
+      "loss/crossentropy": 1.8897674679756165,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2314881831407547,
+      "step": 3130
+    },
+    {
+      "epoch": 0.06264,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.00797119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.8053,
+      "loss/crossentropy": 2.1436809301376343,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26931750774383545,
+      "step": 3132
+    },
+    {
+      "epoch": 0.06268,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.0081695556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.67,
+      "loss/crossentropy": 1.9773973226547241,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26880529522895813,
+      "step": 3134
+    },
+    {
+      "epoch": 0.06272,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.021903483072916667,
+      "learning_rate": 0.0001,
+      "loss": 5.1236,
+      "loss/crossentropy": 2.208973228931427,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29979653656482697,
+      "step": 3136
+    },
+    {
+      "epoch": 0.06276,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.026676432291666666,
+      "learning_rate": 0.0001,
+      "loss": 5.0619,
+      "loss/crossentropy": 2.64120090007782,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29249200224876404,
+      "step": 3138
+    },
+    {
+      "epoch": 0.0628,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.0306549072265625,
+      "learning_rate": 0.0001,
+      "loss": 5.0152,
+      "loss/crossentropy": 2.293095588684082,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2689897269010544,
+      "step": 3140
+    },
+    {
+      "epoch": 0.06284,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.028758748372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.71,
+      "loss/crossentropy": 2.455227494239807,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3202812373638153,
+      "step": 3142
+    },
+    {
+      "epoch": 0.06288,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.0253814697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.9622,
+      "loss/crossentropy": 1.9667487740516663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2702512592077255,
+      "step": 3144
+    },
+    {
+      "epoch": 0.06292,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.024637858072916668,
+      "learning_rate": 0.0001,
+      "loss": 4.7756,
+      "loss/crossentropy": 1.8181490898132324,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2310035452246666,
+      "step": 3146
+    },
+    {
+      "epoch": 0.06296,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.0373687744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.9556,
+      "loss/crossentropy": 1.9985284805297852,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2944463640451431,
+      "step": 3148
+    },
+    {
+      "epoch": 0.063,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.04142964680989583,
+      "learning_rate": 0.0001,
+      "loss": 5.0316,
+      "loss/crossentropy": 2.3850419521331787,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29605095088481903,
+      "step": 3150
+    },
+    {
+      "epoch": 0.06304,
+      "grad_norm": 4.125,
+      "grad_norm_var": 0.17021077473958332,
+      "learning_rate": 0.0001,
+      "loss": 5.2062,
+      "loss/crossentropy": 2.3815460205078125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30247962474823,
+      "step": 3152
+    },
+    {
+      "epoch": 0.06308,
+      "grad_norm": 3.421875,
+      "grad_norm_var": 0.19840087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5851,
+      "loss/crossentropy": 1.995104193687439,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2506560683250427,
+      "step": 3154
+    },
+    {
+      "epoch": 0.06312,
+      "grad_norm": 3.328125,
+      "grad_norm_var": 0.2305084228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5885,
+      "loss/crossentropy": 2.2037696838378906,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2716705799102783,
+      "step": 3156
+    },
+    {
+      "epoch": 0.06316,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.23944905598958333,
+      "learning_rate": 0.0001,
+      "loss": 4.7155,
+      "loss/crossentropy": 2.067265272140503,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25495442003011703,
+      "step": 3158
+    },
+    {
+      "epoch": 0.0632,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.23763020833333334,
+      "learning_rate": 0.0001,
+      "loss": 5.1219,
+      "loss/crossentropy": 2.227355480194092,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27113111317157745,
+      "step": 3160
+    },
+    {
+      "epoch": 0.06324,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.23921610514322916,
+      "learning_rate": 0.0001,
+      "loss": 4.6777,
+      "loss/crossentropy": 1.9077441096305847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2630976662039757,
+      "step": 3162
+    },
+    {
+      "epoch": 0.06328,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.23515523274739583,
+      "learning_rate": 0.0001,
+      "loss": 5.0084,
+      "loss/crossentropy": 2.1737552881240845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28365984559059143,
+      "step": 3164
+    },
+    {
+      "epoch": 0.06332,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.22542317708333334,
+      "learning_rate": 0.0001,
+      "loss": 4.9822,
+      "loss/crossentropy": 1.9357402920722961,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27245980501174927,
+      "step": 3166
+    },
+    {
+      "epoch": 0.06336,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.0995269775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.7681,
+      "loss/crossentropy": 1.8484191298484802,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2933400571346283,
+      "step": 3168
+    },
+    {
+      "epoch": 0.0634,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.057738240559895834,
+      "learning_rate": 0.0001,
+      "loss": 5.0719,
+      "loss/crossentropy": 2.3085306882858276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2926720678806305,
+      "step": 3170
+    },
+    {
+      "epoch": 0.06344,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.040380859375,
+      "learning_rate": 0.0001,
+      "loss": 5.1464,
+      "loss/crossentropy": 2.080895185470581,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2638559564948082,
+      "step": 3172
+    },
+    {
+      "epoch": 0.06348,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.03287353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.6857,
+      "loss/crossentropy": 2.035506248474121,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2692716419696808,
+      "step": 3174
+    },
+    {
+      "epoch": 0.06352,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.035252888997395836,
+      "learning_rate": 0.0001,
+      "loss": 4.9479,
+      "loss/crossentropy": 2.1712740659713745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26579485833644867,
+      "step": 3176
+    },
+    {
+      "epoch": 0.06356,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.04252827962239583,
+      "learning_rate": 0.0001,
+      "loss": 5.1654,
+      "loss/crossentropy": 2.2256147861480713,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27605514228343964,
+      "step": 3178
+    },
+    {
+      "epoch": 0.0636,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.046507771809895834,
+      "learning_rate": 0.0001,
+      "loss": 4.6489,
+      "loss/crossentropy": 2.0917118191719055,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2560323476791382,
+      "step": 3180
+    },
+    {
+      "epoch": 0.06364,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.045491536458333336,
+      "learning_rate": 0.0001,
+      "loss": 4.9833,
+      "loss/crossentropy": 2.3109938502311707,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27040477097034454,
+      "step": 3182
+    },
+    {
+      "epoch": 0.06368,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.03984273274739583,
+      "learning_rate": 0.0001,
+      "loss": 5.1441,
+      "loss/crossentropy": 2.3505672812461853,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29180124402046204,
+      "step": 3184
+    },
+    {
+      "epoch": 0.06372,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.04014383951822917,
+      "learning_rate": 0.0001,
+      "loss": 5.0224,
+      "loss/crossentropy": 2.2471169233322144,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2676118314266205,
+      "step": 3186
+    },
+    {
+      "epoch": 0.06376,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.0264312744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.8247,
+      "loss/crossentropy": 2.3312125205993652,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29243919253349304,
+      "step": 3188
+    },
+    {
+      "epoch": 0.0638,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.02681884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.9374,
+      "loss/crossentropy": 2.1709930896759033,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2786664366722107,
+      "step": 3190
+    },
+    {
+      "epoch": 0.06384,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.02476806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.7391,
+      "loss/crossentropy": 1.8477665185928345,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26132869720458984,
+      "step": 3192
+    },
+    {
+      "epoch": 0.06388,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.01724853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.8167,
+      "loss/crossentropy": 2.2102121114730835,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26027651876211166,
+      "step": 3194
+    },
+    {
+      "epoch": 0.06392,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.0168365478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.6381,
+      "loss/crossentropy": 2.0011088252067566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25629256665706635,
+      "step": 3196
+    },
+    {
+      "epoch": 0.06396,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.01685791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.9215,
+      "loss/crossentropy": 2.319412350654602,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28315384685993195,
+      "step": 3198
+    },
+    {
+      "epoch": 0.064,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.0145172119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.7936,
+      "loss/crossentropy": 2.192026138305664,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.264492504298687,
+      "step": 3200
+    },
+    {
+      "epoch": 0.06404,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.017024739583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.8178,
+      "loss/crossentropy": 2.1745734214782715,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29369065165519714,
+      "step": 3202
+    },
+    {
+      "epoch": 0.06408,
+      "grad_norm": 3.109375,
+      "grad_norm_var": 0.042313639322916666,
+      "learning_rate": 0.0001,
+      "loss": 5.577,
+      "loss/crossentropy": 2.5039013624191284,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2835303544998169,
+      "step": 3204
+    },
+    {
+      "epoch": 0.06412,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.04075419108072917,
+      "learning_rate": 0.0001,
+      "loss": 5.0308,
+      "loss/crossentropy": 2.4255706071853638,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3014884740114212,
+      "step": 3206
+    },
+    {
+      "epoch": 0.06416,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.03873291015625,
+      "learning_rate": 0.0001,
+      "loss": 5.002,
+      "loss/crossentropy": 2.266150116920471,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2698349952697754,
+      "step": 3208
+    },
+    {
+      "epoch": 0.0642,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.03805338541666667,
+      "learning_rate": 0.0001,
+      "loss": 5.0173,
+      "loss/crossentropy": 2.2042946815490723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28830648958683014,
+      "step": 3210
+    },
+    {
+      "epoch": 0.06424,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.03578999837239583,
+      "learning_rate": 0.0001,
+      "loss": 5.0292,
+      "loss/crossentropy": 2.4034690856933594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2834039032459259,
+      "step": 3212
+    },
+    {
+      "epoch": 0.06428,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.0444000244140625,
+      "learning_rate": 0.0001,
+      "loss": 4.6655,
+      "loss/crossentropy": 2.1347755193710327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26405099034309387,
+      "step": 3214
+    },
+    {
+      "epoch": 0.06432,
+      "grad_norm": 3.34375,
+      "grad_norm_var": 0.07415262858072917,
+      "learning_rate": 0.0001,
+      "loss": 5.0862,
+      "loss/crossentropy": 2.018262207508087,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27144598215818405,
+      "step": 3216
+    },
+    {
+      "epoch": 0.06436,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.076953125,
+      "learning_rate": 0.0001,
+      "loss": 4.3969,
+      "loss/crossentropy": 2.0254003405570984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24976836144924164,
+      "step": 3218
+    },
+    {
+      "epoch": 0.0644,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.06026102701822917,
+      "learning_rate": 0.0001,
+      "loss": 4.6779,
+      "loss/crossentropy": 2.1952659487724304,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2786559462547302,
+      "step": 3220
+    },
+    {
+      "epoch": 0.06444,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.06024983723958333,
+      "learning_rate": 0.0001,
+      "loss": 4.8828,
+      "loss/crossentropy": 2.1383036375045776,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2570757120847702,
+      "step": 3222
+    },
+    {
+      "epoch": 0.06448,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.05915425618489583,
+      "learning_rate": 0.0001,
+      "loss": 4.8248,
+      "loss/crossentropy": 2.267812967300415,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2774328589439392,
+      "step": 3224
+    },
+    {
+      "epoch": 0.06452,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.0587066650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.9389,
+      "loss/crossentropy": 1.906205415725708,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2528962790966034,
+      "step": 3226
+    },
+    {
+      "epoch": 0.06456,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.05895182291666667,
+      "learning_rate": 0.0001,
+      "loss": 4.9823,
+      "loss/crossentropy": 2.1628336906433105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27087944746017456,
+      "step": 3228
+    },
+    {
+      "epoch": 0.0646,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.05807291666666667,
+      "learning_rate": 0.0001,
+      "loss": 4.5893,
+      "loss/crossentropy": 1.8845162391662598,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2241554707288742,
+      "step": 3230
+    },
+    {
+      "epoch": 0.06464,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.019237263997395834,
+      "learning_rate": 0.0001,
+      "loss": 4.9143,
+      "loss/crossentropy": 2.4157146215438843,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2770439088344574,
+      "step": 3232
+    },
+    {
+      "epoch": 0.06468,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.021825154622395832,
+      "learning_rate": 0.0001,
+      "loss": 4.832,
+      "loss/crossentropy": 1.9979816675186157,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23696578294038773,
+      "step": 3234
+    },
+    {
+      "epoch": 0.06472,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.016364542643229167,
+      "learning_rate": 0.0001,
+      "loss": 5.1123,
+      "loss/crossentropy": 2.1790190935134888,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.252600260078907,
+      "step": 3236
+    },
+    {
+      "epoch": 0.06476,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.016624959309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.5943,
+      "loss/crossentropy": 2.1612111926078796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27929094433784485,
+      "step": 3238
+    },
+    {
+      "epoch": 0.0648,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.08124898274739584,
+      "learning_rate": 0.0001,
+      "loss": 4.8293,
+      "loss/crossentropy": 2.261025071144104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3034510314464569,
+      "step": 3240
+    },
+    {
+      "epoch": 0.06484,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.081689453125,
+      "learning_rate": 0.0001,
+      "loss": 4.8243,
+      "loss/crossentropy": 2.02247554063797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2655429244041443,
+      "step": 3242
+    },
+    {
+      "epoch": 0.06488,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.08289286295572916,
+      "learning_rate": 0.0001,
+      "loss": 4.9428,
+      "loss/crossentropy": 2.495269775390625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32837189733982086,
+      "step": 3244
+    },
+    {
+      "epoch": 0.06492,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.07550455729166666,
+      "learning_rate": 0.0001,
+      "loss": 4.9928,
+      "loss/crossentropy": 2.365166425704956,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27000221610069275,
+      "step": 3246
+    },
+    {
+      "epoch": 0.06496,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.0746246337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.837,
+      "loss/crossentropy": 1.8728906512260437,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2711311876773834,
+      "step": 3248
+    },
+    {
+      "epoch": 0.065,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.07427978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.7916,
+      "loss/crossentropy": 1.916531264781952,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23036544770002365,
+      "step": 3250
+    },
+    {
+      "epoch": 0.06504,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.07463277180989583,
+      "learning_rate": 0.0001,
+      "loss": 4.9399,
+      "loss/crossentropy": 1.996503233909607,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24317056685686111,
+      "step": 3252
+    },
+    {
+      "epoch": 0.06508,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.07366536458333334,
+      "learning_rate": 0.0001,
+      "loss": 4.8746,
+      "loss/crossentropy": 2.101921260356903,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2775810658931732,
+      "step": 3254
+    },
+    {
+      "epoch": 0.06512,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.005125935872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.7526,
+      "loss/crossentropy": 1.9286046028137207,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23964455723762512,
+      "step": 3256
+    },
+    {
+      "epoch": 0.06516,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.005464680989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.8216,
+      "loss/crossentropy": 2.224915862083435,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27516382932662964,
+      "step": 3258
+    },
+    {
+      "epoch": 0.0652,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.0133697509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.6553,
+      "loss/crossentropy": 2.259337306022644,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27185751497745514,
+      "step": 3260
+    },
+    {
+      "epoch": 0.06524,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.013117472330729166,
+      "learning_rate": 0.0001,
+      "loss": 4.7095,
+      "loss/crossentropy": 2.1081044673919678,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23740330338478088,
+      "step": 3262
+    },
+    {
+      "epoch": 0.06528,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.0131744384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.6375,
+      "loss/crossentropy": 1.9032058119773865,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2508246824145317,
+      "step": 3264
+    },
+    {
+      "epoch": 0.06532,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.013353474934895833,
+      "learning_rate": 0.0001,
+      "loss": 4.7165,
+      "loss/crossentropy": 2.0773792266845703,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2797544598579407,
+      "step": 3266
+    },
+    {
+      "epoch": 0.06536,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.013509114583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.9136,
+      "loss/crossentropy": 2.1591526865959167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2540442571043968,
+      "step": 3268
+    },
+    {
+      "epoch": 0.0654,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.014351399739583333,
+      "learning_rate": 0.0001,
+      "loss": 4.6248,
+      "loss/crossentropy": 1.7735809683799744,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24107928574085236,
+      "step": 3270
+    },
+    {
+      "epoch": 0.06544,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.014647420247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.7568,
+      "loss/crossentropy": 2.069553792476654,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26356737315654755,
+      "step": 3272
+    },
+    {
+      "epoch": 0.06548,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.014623006184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.7274,
+      "loss/crossentropy": 1.9688642024993896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29453104734420776,
+      "step": 3274
+    },
+    {
+      "epoch": 0.06552,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.012043253580729166,
+      "learning_rate": 0.0001,
+      "loss": 4.4264,
+      "loss/crossentropy": 1.785252034664154,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23213861882686615,
+      "step": 3276
+    },
+    {
+      "epoch": 0.06556,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.013255818684895834,
+      "learning_rate": 0.0001,
+      "loss": 4.779,
+      "loss/crossentropy": 2.155774712562561,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25758183747529984,
+      "step": 3278
+    },
+    {
+      "epoch": 0.0656,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.016161092122395835,
+      "learning_rate": 0.0001,
+      "loss": 5.0453,
+      "loss/crossentropy": 2.0403348803520203,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25007252395153046,
+      "step": 3280
+    },
+    {
+      "epoch": 0.06564,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.015208943684895834,
+      "learning_rate": 0.0001,
+      "loss": 4.7161,
+      "loss/crossentropy": 1.9963608384132385,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2608266994357109,
+      "step": 3282
+    },
+    {
+      "epoch": 0.06568,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.019364420572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.9275,
+      "loss/crossentropy": 2.2021098732948303,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28361018002033234,
+      "step": 3284
+    },
+    {
+      "epoch": 0.06572,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.01812744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.6533,
+      "loss/crossentropy": 2.0363903641700745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2641526162624359,
+      "step": 3286
+    },
+    {
+      "epoch": 0.06576,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.020466105143229166,
+      "learning_rate": 0.0001,
+      "loss": 4.6884,
+      "loss/crossentropy": 2.1052531003952026,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24344927817583084,
+      "step": 3288
+    },
+    {
+      "epoch": 0.0658,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.02027587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.9246,
+      "loss/crossentropy": 2.0664029717445374,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25679293274879456,
+      "step": 3290
+    },
+    {
+      "epoch": 0.06584,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.013678995768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.7802,
+      "loss/crossentropy": 2.168351709842682,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2759328931570053,
+      "step": 3292
+    },
+    {
+      "epoch": 0.06588,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.013703409830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.8481,
+      "loss/crossentropy": 1.9305949211120605,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26165173947811127,
+      "step": 3294
+    },
+    {
+      "epoch": 0.06592,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.017476399739583332,
+      "learning_rate": 0.0001,
+      "loss": 4.6942,
+      "loss/crossentropy": 1.9257155060768127,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24499841034412384,
+      "step": 3296
+    },
+    {
+      "epoch": 0.06596,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.017513020833333334,
+      "learning_rate": 0.0001,
+      "loss": 4.8941,
+      "loss/crossentropy": 2.1406426429748535,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26676414906978607,
+      "step": 3298
+    },
+    {
+      "epoch": 0.066,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.016422526041666666,
+      "learning_rate": 0.0001,
+      "loss": 4.8293,
+      "loss/crossentropy": 2.099781036376953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2916935384273529,
+      "step": 3300
+    },
+    {
+      "epoch": 0.06604,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.023094685872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.9504,
+      "loss/crossentropy": 2.1077913641929626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2667912393808365,
+      "step": 3302
+    },
+    {
+      "epoch": 0.06608,
+      "grad_norm": 3.90625,
+      "grad_norm_var": 0.1297271728515625,
+      "learning_rate": 0.0001,
+      "loss": 5.2952,
+      "loss/crossentropy": 2.2583223581314087,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2907385528087616,
+      "step": 3304
+    },
+    {
+      "epoch": 0.06612,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.13405659993489583,
+      "learning_rate": 0.0001,
+      "loss": 4.5813,
+      "loss/crossentropy": 1.847477912902832,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2272188812494278,
+      "step": 3306
+    },
+    {
+      "epoch": 0.06616,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.1392486572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.6502,
+      "loss/crossentropy": 1.7610225677490234,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23973071575164795,
+      "step": 3308
+    },
+    {
+      "epoch": 0.0662,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.1464019775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.7123,
+      "loss/crossentropy": 1.977916419506073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25456882268190384,
+      "step": 3310
+    },
+    {
+      "epoch": 0.06624,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.14095052083333334,
+      "learning_rate": 0.0001,
+      "loss": 4.7353,
+      "loss/crossentropy": 2.2196428775787354,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2563377171754837,
+      "step": 3312
+    },
+    {
+      "epoch": 0.06628,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.14537760416666667,
+      "learning_rate": 0.0001,
+      "loss": 4.6907,
+      "loss/crossentropy": 2.0861470699310303,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27121224999427795,
+      "step": 3314
+    },
+    {
+      "epoch": 0.06632,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.14287821451822916,
+      "learning_rate": 0.0001,
+      "loss": 4.6484,
+      "loss/crossentropy": 1.9716283679008484,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.277963787317276,
+      "step": 3316
+    },
+    {
+      "epoch": 0.06636,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.1375,
+      "learning_rate": 0.0001,
+      "loss": 4.7985,
+      "loss/crossentropy": 2.5604729652404785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31031742691993713,
+      "step": 3318
+    },
+    {
+      "epoch": 0.0664,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.022337849934895834,
+      "learning_rate": 0.0001,
+      "loss": 5.4329,
+      "loss/crossentropy": 2.2672252655029297,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29225634038448334,
+      "step": 3320
+    },
+    {
+      "epoch": 0.06644,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.020015462239583334,
+      "learning_rate": 0.0001,
+      "loss": 5.0557,
+      "loss/crossentropy": 2.0992931723594666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2648598402738571,
+      "step": 3322
+    },
+    {
+      "epoch": 0.06648,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.014827473958333334,
+      "learning_rate": 0.0001,
+      "loss": 5.0895,
+      "loss/crossentropy": 2.208917260169983,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2998420298099518,
+      "step": 3324
+    },
+    {
+      "epoch": 0.06652,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.012791951497395834,
+      "learning_rate": 0.0001,
+      "loss": 4.8834,
+      "loss/crossentropy": 2.290796995162964,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2638905793428421,
+      "step": 3326
+    },
+    {
+      "epoch": 0.06656,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.012214152018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.8521,
+      "loss/crossentropy": 2.4156445264816284,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28743064403533936,
+      "step": 3328
+    },
+    {
+      "epoch": 0.0666,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.012132771809895833,
+      "learning_rate": 0.0001,
+      "loss": 5.2205,
+      "loss/crossentropy": 2.4604904651641846,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27585768699645996,
+      "step": 3330
+    },
+    {
+      "epoch": 0.06664,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.013818359375,
+      "learning_rate": 0.0001,
+      "loss": 4.8296,
+      "loss/crossentropy": 1.8613844513893127,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27038969844579697,
+      "step": 3332
+    },
+    {
+      "epoch": 0.06668,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.014046223958333333,
+      "learning_rate": 0.0001,
+      "loss": 4.8777,
+      "loss/crossentropy": 2.232776403427124,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2686954140663147,
+      "step": 3334
+    },
+    {
+      "epoch": 0.06672,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.0120758056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.9536,
+      "loss/crossentropy": 2.070033550262451,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26176171004772186,
+      "step": 3336
+    },
+    {
+      "epoch": 0.06676,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.01187744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.8616,
+      "loss/crossentropy": 2.38937509059906,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2539200633764267,
+      "step": 3338
+    },
+    {
+      "epoch": 0.0668,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.010933430989583333,
+      "learning_rate": 0.0001,
+      "loss": 5.1104,
+      "loss/crossentropy": 2.272845983505249,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2856733053922653,
+      "step": 3340
+    },
+    {
+      "epoch": 0.06684,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.01181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.7935,
+      "loss/crossentropy": 2.427489161491394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30572691559791565,
+      "step": 3342
+    },
+    {
+      "epoch": 0.06688,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.009407552083333333,
+      "learning_rate": 0.0001,
+      "loss": 4.9801,
+      "loss/crossentropy": 2.4701327085494995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28469331562519073,
+      "step": 3344
+    },
+    {
+      "epoch": 0.06692,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.006917317708333333,
+      "learning_rate": 0.0001,
+      "loss": 4.815,
+      "loss/crossentropy": 2.0733558535575867,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.292859822511673,
+      "step": 3346
+    },
+    {
+      "epoch": 0.06696,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.0061187744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.8379,
+      "loss/crossentropy": 2.301755905151367,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3174070864915848,
+      "step": 3348
+    },
+    {
+      "epoch": 0.067,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.012580362955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3359,
+      "loss/crossentropy": 2.039419114589691,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24543144553899765,
+      "step": 3350
+    },
+    {
+      "epoch": 0.06704,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.0144683837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.7938,
+      "loss/crossentropy": 1.9247611165046692,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2595005929470062,
+      "step": 3352
+    },
+    {
+      "epoch": 0.06708,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.01630859375,
+      "learning_rate": 0.0001,
+      "loss": 4.8294,
+      "loss/crossentropy": 2.224974751472473,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26880575716495514,
+      "step": 3354
+    },
+    {
+      "epoch": 0.06712,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.0195220947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.8922,
+      "loss/crossentropy": 2.2601993083953857,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.279633030295372,
+      "step": 3356
+    },
+    {
+      "epoch": 0.06716,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.017085774739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.5911,
+      "loss/crossentropy": 1.8156417608261108,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22118539363145828,
+      "step": 3358
+    },
+    {
+      "epoch": 0.0672,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.020140584309895834,
+      "learning_rate": 0.0001,
+      "loss": 5.2032,
+      "loss/crossentropy": 2.207859516143799,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2771998345851898,
+      "step": 3360
+    },
+    {
+      "epoch": 0.06724,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.02017822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.9352,
+      "loss/crossentropy": 1.9886083602905273,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2545766308903694,
+      "step": 3362
+    },
+    {
+      "epoch": 0.06728,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.025777180989583332,
+      "learning_rate": 0.0001,
+      "loss": 4.4323,
+      "loss/crossentropy": 1.7046592235565186,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2207612618803978,
+      "step": 3364
+    },
+    {
+      "epoch": 0.06732,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.021122233072916666,
+      "learning_rate": 0.0001,
+      "loss": 5.0209,
+      "loss/crossentropy": 2.0283663868904114,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2743644416332245,
+      "step": 3366
+    },
+    {
+      "epoch": 0.06736,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.019498697916666665,
+      "learning_rate": 0.0001,
+      "loss": 4.4788,
+      "loss/crossentropy": 1.975772500038147,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2505730241537094,
+      "step": 3368
+    },
+    {
+      "epoch": 0.0674,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.019950358072916667,
+      "learning_rate": 0.0001,
+      "loss": 4.6863,
+      "loss/crossentropy": 1.9021872282028198,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25430944561958313,
+      "step": 3370
+    },
+    {
+      "epoch": 0.06744,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.0182037353515625,
+      "learning_rate": 0.0001,
+      "loss": 5.1859,
+      "loss/crossentropy": 2.3888463973999023,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27896443009376526,
+      "step": 3372
+    },
+    {
+      "epoch": 0.06748,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.02086181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.6041,
+      "loss/crossentropy": 1.7844690680503845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2536455988883972,
+      "step": 3374
+    },
+    {
+      "epoch": 0.06752,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.021442667643229166,
+      "learning_rate": 0.0001,
+      "loss": 4.4538,
+      "loss/crossentropy": 1.919598639011383,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26129309833049774,
+      "step": 3376
+    },
+    {
+      "epoch": 0.06756,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.022835286458333333,
+      "learning_rate": 0.0001,
+      "loss": 5.0568,
+      "loss/crossentropy": 2.2292110919952393,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25612247735261917,
+      "step": 3378
+    },
+    {
+      "epoch": 0.0676,
+      "grad_norm": 3.4375,
+      "grad_norm_var": 0.0779296875,
+      "learning_rate": 0.0001,
+      "loss": 4.8674,
+      "loss/crossentropy": 2.2235841751098633,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2607487738132477,
+      "step": 3380
+    },
+    {
+      "epoch": 0.06764,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.08089192708333333,
+      "learning_rate": 0.0001,
+      "loss": 4.5872,
+      "loss/crossentropy": 2.1375235319137573,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2344205006957054,
+      "step": 3382
+    },
+    {
+      "epoch": 0.06768,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.08501688639322917,
+      "learning_rate": 0.0001,
+      "loss": 4.6076,
+      "loss/crossentropy": 2.020237445831299,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2550392523407936,
+      "step": 3384
+    },
+    {
+      "epoch": 0.06772,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.07911783854166667,
+      "learning_rate": 0.0001,
+      "loss": 4.6939,
+      "loss/crossentropy": 2.138959765434265,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2742607295513153,
+      "step": 3386
+    },
+    {
+      "epoch": 0.06776,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.07864481608072917,
+      "learning_rate": 0.0001,
+      "loss": 4.9065,
+      "loss/crossentropy": 2.521559953689575,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2892928719520569,
+      "step": 3388
+    },
+    {
+      "epoch": 0.0678,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.07815348307291667,
+      "learning_rate": 0.0001,
+      "loss": 5.049,
+      "loss/crossentropy": 2.169550120830536,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27750439941883087,
+      "step": 3390
+    },
+    {
+      "epoch": 0.06784,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.07464090983072917,
+      "learning_rate": 0.0001,
+      "loss": 4.7401,
+      "loss/crossentropy": 1.975584864616394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2557516545057297,
+      "step": 3392
+    },
+    {
+      "epoch": 0.06788,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.07617899576822916,
+      "learning_rate": 0.0001,
+      "loss": 4.9207,
+      "loss/crossentropy": 2.5837322473526,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3444886952638626,
+      "step": 3394
+    },
+    {
+      "epoch": 0.06792,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.022294108072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.743,
+      "loss/crossentropy": 1.963110864162445,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24120041728019714,
+      "step": 3396
+    },
+    {
+      "epoch": 0.06796,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.0211090087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5759,
+      "loss/crossentropy": 2.381603956222534,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26319295167922974,
+      "step": 3398
+    },
+    {
+      "epoch": 0.068,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.023078409830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.5642,
+      "loss/crossentropy": 2.054026961326599,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24540280550718307,
+      "step": 3400
+    },
+    {
+      "epoch": 0.06804,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.0187652587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3661,
+      "loss/crossentropy": 2.047453820705414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2524164840579033,
+      "step": 3402
+    },
+    {
+      "epoch": 0.06808,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.018317667643229167,
+      "learning_rate": 0.0001,
+      "loss": 4.7839,
+      "loss/crossentropy": 1.8616933226585388,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23905867338180542,
+      "step": 3404
+    },
+    {
+      "epoch": 0.06812,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.016422526041666666,
+      "learning_rate": 0.0001,
+      "loss": 4.6932,
+      "loss/crossentropy": 2.4304568767547607,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2802550047636032,
+      "step": 3406
+    },
+    {
+      "epoch": 0.06816,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.01318359375,
+      "learning_rate": 0.0001,
+      "loss": 4.7005,
+      "loss/crossentropy": 1.8178748488426208,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24266308546066284,
+      "step": 3408
+    },
+    {
+      "epoch": 0.0682,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.0116363525390625,
+      "learning_rate": 0.0001,
+      "loss": 5.0152,
+      "loss/crossentropy": 2.025859773159027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.282450333237648,
+      "step": 3410
+    },
+    {
+      "epoch": 0.06824,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.010383097330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.737,
+      "loss/crossentropy": 2.032994568347931,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26047058403491974,
+      "step": 3412
+    },
+    {
+      "epoch": 0.06828,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.007840983072916667,
+      "learning_rate": 0.0001,
+      "loss": 4.6428,
+      "loss/crossentropy": 2.2468607425689697,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2814445495605469,
+      "step": 3414
+    },
+    {
+      "epoch": 0.06832,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.03629150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.6854,
+      "loss/crossentropy": 2.2534161806106567,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3231179416179657,
+      "step": 3416
+    },
+    {
+      "epoch": 0.06836,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.03443603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.9599,
+      "loss/crossentropy": 2.1678181886672974,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2840191572904587,
+      "step": 3418
+    },
+    {
+      "epoch": 0.0684,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.03453776041666667,
+      "learning_rate": 0.0001,
+      "loss": 4.8667,
+      "loss/crossentropy": 2.053459882736206,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.271483838558197,
+      "step": 3420
+    },
+    {
+      "epoch": 0.06844,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.03243815104166667,
+      "learning_rate": 0.0001,
+      "loss": 4.714,
+      "loss/crossentropy": 2.2278919219970703,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2781473994255066,
+      "step": 3422
+    },
+    {
+      "epoch": 0.06848,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.0384674072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.5262,
+      "loss/crossentropy": 1.9582479000091553,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24171434342861176,
+      "step": 3424
+    },
+    {
+      "epoch": 0.06852,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.04121805826822917,
+      "learning_rate": 0.0001,
+      "loss": 4.4734,
+      "loss/crossentropy": 1.808964192867279,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22458729147911072,
+      "step": 3426
+    },
+    {
+      "epoch": 0.06856,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.04296468098958333,
+      "learning_rate": 0.0001,
+      "loss": 4.464,
+      "loss/crossentropy": 1.9225260019302368,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2508438527584076,
+      "step": 3428
+    },
+    {
+      "epoch": 0.0686,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.044465128580729166,
+      "learning_rate": 0.0001,
+      "loss": 4.9647,
+      "loss/crossentropy": 2.2446881532669067,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26919613778591156,
+      "step": 3430
+    },
+    {
+      "epoch": 0.06864,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.017899576822916666,
+      "learning_rate": 0.0001,
+      "loss": 4.6865,
+      "loss/crossentropy": 2.2047033309936523,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2581355720758438,
+      "step": 3432
+    },
+    {
+      "epoch": 0.06868,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.015729777018229165,
+      "learning_rate": 0.0001,
+      "loss": 4.7139,
+      "loss/crossentropy": 2.1223543882369995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25814756751060486,
+      "step": 3434
+    },
+    {
+      "epoch": 0.06872,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.016813151041666665,
+      "learning_rate": 0.0001,
+      "loss": 4.6771,
+      "loss/crossentropy": 2.1641053557395935,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25222497433423996,
+      "step": 3436
+    },
+    {
+      "epoch": 0.06876,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.0189605712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.6972,
+      "loss/crossentropy": 1.9569795727729797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2681007981300354,
+      "step": 3438
+    },
+    {
+      "epoch": 0.0688,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.013216145833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.8214,
+      "loss/crossentropy": 2.212220251560211,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28191742300987244,
+      "step": 3440
+    },
+    {
+      "epoch": 0.06884,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.011546834309895834,
+      "learning_rate": 0.0001,
+      "loss": 4.4162,
+      "loss/crossentropy": 1.9564262628555298,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2560906559228897,
+      "step": 3442
+    },
+    {
+      "epoch": 0.06888,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.012043253580729166,
+      "learning_rate": 0.0001,
+      "loss": 4.6837,
+      "loss/crossentropy": 1.8553346395492554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24626825004816055,
+      "step": 3444
+    },
+    {
+      "epoch": 0.06892,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.012007649739583333,
+      "learning_rate": 0.0001,
+      "loss": 4.7914,
+      "loss/crossentropy": 1.9803723692893982,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2712126225233078,
+      "step": 3446
+    },
+    {
+      "epoch": 0.06896,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.010423787434895833,
+      "learning_rate": 0.0001,
+      "loss": 4.857,
+      "loss/crossentropy": 1.9914751648902893,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2565220594406128,
+      "step": 3448
+    },
+    {
+      "epoch": 0.069,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.010758463541666667,
+      "learning_rate": 0.0001,
+      "loss": 4.8013,
+      "loss/crossentropy": 2.2114094495773315,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28365010023117065,
+      "step": 3450
+    },
+    {
+      "epoch": 0.06904,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.009577433268229166,
+      "learning_rate": 0.0001,
+      "loss": 5.0322,
+      "loss/crossentropy": 2.4366514682769775,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30788426101207733,
+      "step": 3452
+    },
+    {
+      "epoch": 0.06908,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.010595703125,
+      "learning_rate": 0.0001,
+      "loss": 4.7902,
+      "loss/crossentropy": 2.304569959640503,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2835633158683777,
+      "step": 3454
+    },
+    {
+      "epoch": 0.06912,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.027318318684895832,
+      "learning_rate": 0.0001,
+      "loss": 5.151,
+      "loss/crossentropy": 2.2518080472946167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.278719499707222,
+      "step": 3456
+    },
+    {
+      "epoch": 0.06916,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.027904256184895834,
+      "learning_rate": 0.0001,
+      "loss": 5.1688,
+      "loss/crossentropy": 2.333768129348755,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28694969415664673,
+      "step": 3458
+    },
+    {
+      "epoch": 0.0692,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.0260650634765625,
+      "learning_rate": 0.0001,
+      "loss": 5.1421,
+      "loss/crossentropy": 2.3534432649612427,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3005864769220352,
+      "step": 3460
+    },
+    {
+      "epoch": 0.06924,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.022981770833333335,
+      "learning_rate": 0.0001,
+      "loss": 4.9502,
+      "loss/crossentropy": 2.0703017711639404,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25969168543815613,
+      "step": 3462
+    },
+    {
+      "epoch": 0.06928,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.023502604166666666,
+      "learning_rate": 0.0001,
+      "loss": 5.0258,
+      "loss/crossentropy": 2.167420506477356,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32322582602500916,
+      "step": 3464
+    },
+    {
+      "epoch": 0.06932,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.021222941080729165,
+      "learning_rate": 0.0001,
+      "loss": 4.5342,
+      "loss/crossentropy": 2.0845181941986084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26804475486278534,
+      "step": 3466
+    },
+    {
+      "epoch": 0.06936,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.021219889322916668,
+      "learning_rate": 0.0001,
+      "loss": 5.0459,
+      "loss/crossentropy": 2.4165114164352417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2866530567407608,
+      "step": 3468
+    },
+    {
+      "epoch": 0.0694,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.022359212239583332,
+      "learning_rate": 0.0001,
+      "loss": 5.0938,
+      "loss/crossentropy": 2.4152863025665283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30185502767562866,
+      "step": 3470
+    },
+    {
+      "epoch": 0.06944,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.014924112955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.6362,
+      "loss/crossentropy": 2.1878501176834106,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2736133933067322,
+      "step": 3472
+    },
+    {
+      "epoch": 0.06948,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.015262858072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.812,
+      "loss/crossentropy": 2.055173695087433,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2609568238258362,
+      "step": 3474
+    },
+    {
+      "epoch": 0.06952,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.015550740559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.8201,
+      "loss/crossentropy": 2.050000250339508,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2713121324777603,
+      "step": 3476
+    },
+    {
+      "epoch": 0.06956,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.019169108072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.9916,
+      "loss/crossentropy": 2.227464199066162,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2710970640182495,
+      "step": 3478
+    },
+    {
+      "epoch": 0.0696,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.016828409830729165,
+      "learning_rate": 0.0001,
+      "loss": 4.7435,
+      "loss/crossentropy": 2.096015691757202,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29556676745414734,
+      "step": 3480
+    },
+    {
+      "epoch": 0.06964,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.028804524739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.6738,
+      "loss/crossentropy": 1.9252901673316956,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2976933419704437,
+      "step": 3482
+    },
+    {
+      "epoch": 0.06968,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.03385009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.7679,
+      "loss/crossentropy": 2.258090019226074,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27399829030036926,
+      "step": 3484
+    },
+    {
+      "epoch": 0.06972,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.03192952473958333,
+      "learning_rate": 0.0001,
+      "loss": 4.7614,
+      "loss/crossentropy": 2.2776867151260376,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29224833846092224,
+      "step": 3486
+    },
+    {
+      "epoch": 0.06976,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.031819661458333336,
+      "learning_rate": 0.0001,
+      "loss": 5.1628,
+      "loss/crossentropy": 2.097061276435852,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26103661954402924,
+      "step": 3488
+    },
+    {
+      "epoch": 0.0698,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.03264567057291667,
+      "learning_rate": 0.0001,
+      "loss": 4.7364,
+      "loss/crossentropy": 2.206419885158539,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2775641232728958,
+      "step": 3490
+    },
+    {
+      "epoch": 0.06984,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.035563151041666664,
+      "learning_rate": 0.0001,
+      "loss": 4.7216,
+      "loss/crossentropy": 1.962704062461853,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2421455979347229,
+      "step": 3492
+    },
+    {
+      "epoch": 0.06988,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.032698567708333334,
+      "learning_rate": 0.0001,
+      "loss": 4.7358,
+      "loss/crossentropy": 2.223302483558655,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2709382176399231,
+      "step": 3494
+    },
+    {
+      "epoch": 0.06992,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.034764607747395836,
+      "learning_rate": 0.0001,
+      "loss": 4.8563,
+      "loss/crossentropy": 2.259950876235962,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29514479637145996,
+      "step": 3496
+    },
+    {
+      "epoch": 0.06996,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.021141560872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.6147,
+      "loss/crossentropy": 2.2337416410446167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2585422098636627,
+      "step": 3498
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.022663370768229166,
+      "learning_rate": 0.0001,
+      "loss": 4.9278,
+      "loss/crossentropy": 2.131904423236847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32493171095848083,
+      "step": 3500
+    },
+    {
+      "epoch": 0.07004,
+      "grad_norm": 3.109375,
+      "grad_norm_var": 0.04553629557291667,
+      "learning_rate": 0.0001,
+      "loss": 4.9285,
+      "loss/crossentropy": 2.461912155151367,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2917899489402771,
+      "step": 3502
+    },
+    {
+      "epoch": 0.07008,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.04366861979166667,
+      "learning_rate": 0.0001,
+      "loss": 4.6466,
+      "loss/crossentropy": 2.05659943819046,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24786780774593353,
+      "step": 3504
+    },
+    {
+      "epoch": 0.07012,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.0412261962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.7282,
+      "loss/crossentropy": 2.3972705602645874,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29012058675289154,
+      "step": 3506
+    },
+    {
+      "epoch": 0.07016,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.03945210774739583,
+      "learning_rate": 0.0001,
+      "loss": 4.6561,
+      "loss/crossentropy": 1.8465647101402283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26938022673130035,
+      "step": 3508
+    },
+    {
+      "epoch": 0.0702,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.0434722900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.9007,
+      "loss/crossentropy": 2.2447493076324463,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27260421216487885,
+      "step": 3510
+    },
+    {
+      "epoch": 0.07024,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.042464192708333334,
+      "learning_rate": 0.0001,
+      "loss": 4.7774,
+      "loss/crossentropy": 2.4258209466934204,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.285652831196785,
+      "step": 3512
+    },
+    {
+      "epoch": 0.07028,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.04248046875,
+      "learning_rate": 0.0001,
+      "loss": 4.7353,
+      "loss/crossentropy": 2.1033068895339966,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2628294378519058,
+      "step": 3514
+    },
+    {
+      "epoch": 0.07032,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.03819071451822917,
+      "learning_rate": 0.0001,
+      "loss": 4.8036,
+      "loss/crossentropy": 2.2740964889526367,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27228833734989166,
+      "step": 3516
+    },
+    {
+      "epoch": 0.07036,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.015104166666666667,
+      "learning_rate": 0.0001,
+      "loss": 4.7699,
+      "loss/crossentropy": 2.2315655946731567,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26613348722457886,
+      "step": 3518
+    },
+    {
+      "epoch": 0.0704,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.014720662434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.6038,
+      "loss/crossentropy": 1.9001839756965637,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.248238705098629,
+      "step": 3520
+    },
+    {
+      "epoch": 0.07044,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.013655598958333333,
+      "learning_rate": 0.0001,
+      "loss": 4.7034,
+      "loss/crossentropy": 2.0940937399864197,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28850243985652924,
+      "step": 3522
+    },
+    {
+      "epoch": 0.07048,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.014338175455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.9187,
+      "loss/crossentropy": 1.9088054299354553,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2577759325504303,
+      "step": 3524
+    },
+    {
+      "epoch": 0.07052,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.01031494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.7852,
+      "loss/crossentropy": 2.1965672969818115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2696071192622185,
+      "step": 3526
+    },
+    {
+      "epoch": 0.07056,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.010770670572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.7657,
+      "loss/crossentropy": 2.245758891105652,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2646239101886749,
+      "step": 3528
+    },
+    {
+      "epoch": 0.0706,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.013895670572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.7794,
+      "loss/crossentropy": 2.180204927921295,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2582162171602249,
+      "step": 3530
+    },
+    {
+      "epoch": 0.07064,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.019562784830729166,
+      "learning_rate": 0.0001,
+      "loss": 5.04,
+      "loss/crossentropy": 2.193474531173706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40557755529880524,
+      "step": 3532
+    },
+    {
+      "epoch": 0.07068,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.021361287434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.9562,
+      "loss/crossentropy": 2.2667607069015503,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28741554915905,
+      "step": 3534
+    },
+    {
+      "epoch": 0.07072,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.022093709309895834,
+      "learning_rate": 0.0001,
+      "loss": 4.8215,
+      "loss/crossentropy": 1.9890388250350952,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2612537443637848,
+      "step": 3536
+    },
+    {
+      "epoch": 0.07076,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.02392578125,
+      "learning_rate": 0.0001,
+      "loss": 4.7544,
+      "loss/crossentropy": 1.9390615820884705,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23259633034467697,
+      "step": 3538
+    },
+    {
+      "epoch": 0.0708,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.023193359375,
+      "learning_rate": 0.0001,
+      "loss": 4.8227,
+      "loss/crossentropy": 2.3050389289855957,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2738967537879944,
+      "step": 3540
+    },
+    {
+      "epoch": 0.07084,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.025862630208333334,
+      "learning_rate": 0.0001,
+      "loss": 4.4323,
+      "loss/crossentropy": 2.3832077980041504,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29863911867141724,
+      "step": 3542
+    },
+    {
+      "epoch": 0.07088,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.02822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.7101,
+      "loss/crossentropy": 1.8186699748039246,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22366883605718613,
+      "step": 3544
+    },
+    {
+      "epoch": 0.07092,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.024235026041666666,
+      "learning_rate": 0.0001,
+      "loss": 4.8151,
+      "loss/crossentropy": 2.2650288343429565,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2733971029520035,
+      "step": 3546
+    },
+    {
+      "epoch": 0.07096,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.0191070556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.7514,
+      "loss/crossentropy": 2.432945966720581,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.269818976521492,
+      "step": 3548
+    },
+    {
+      "epoch": 0.071,
+      "grad_norm": 4.15625,
+      "grad_norm_var": 0.19250386555989582,
+      "learning_rate": 0.0001,
+      "loss": 4.9461,
+      "loss/crossentropy": 2.021497666835785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2488800287246704,
+      "step": 3550
+    },
+    {
+      "epoch": 0.07104,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.2001129150390625,
+      "learning_rate": 0.0001,
+      "loss": 5.0341,
+      "loss/crossentropy": 2.0840535163879395,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2780974507331848,
+      "step": 3552
+    },
+    {
+      "epoch": 0.07108,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.19724833170572917,
+      "learning_rate": 0.0001,
+      "loss": 4.8123,
+      "loss/crossentropy": 2.352238416671753,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2830911874771118,
+      "step": 3554
+    },
+    {
+      "epoch": 0.07112,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.19795633951822916,
+      "learning_rate": 0.0001,
+      "loss": 4.9632,
+      "loss/crossentropy": 2.395397186279297,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2861028015613556,
+      "step": 3556
+    },
+    {
+      "epoch": 0.07116,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.19630533854166668,
+      "learning_rate": 0.0001,
+      "loss": 5.0203,
+      "loss/crossentropy": 2.6454248428344727,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28888577222824097,
+      "step": 3558
+    },
+    {
+      "epoch": 0.0712,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.19394429524739584,
+      "learning_rate": 0.0001,
+      "loss": 4.8228,
+      "loss/crossentropy": 2.21127188205719,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2676347494125366,
+      "step": 3560
+    },
+    {
+      "epoch": 0.07124,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.19245503743489584,
+      "learning_rate": 0.0001,
+      "loss": 4.7502,
+      "loss/crossentropy": 2.0187097787857056,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24967321753501892,
+      "step": 3562
+    },
+    {
+      "epoch": 0.07128,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.19409077962239582,
+      "learning_rate": 0.0001,
+      "loss": 4.9548,
+      "loss/crossentropy": 2.1822619438171387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24893341958522797,
+      "step": 3564
+    },
+    {
+      "epoch": 0.07132,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.03178609212239583,
+      "learning_rate": 0.0001,
+      "loss": 5.0912,
+      "loss/crossentropy": 2.486607313156128,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29665203392505646,
+      "step": 3566
+    },
+    {
+      "epoch": 0.07136,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.020048014322916665,
+      "learning_rate": 0.0001,
+      "loss": 5.0494,
+      "loss/crossentropy": 2.2631163597106934,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.268096499145031,
+      "step": 3568
+    },
+    {
+      "epoch": 0.0714,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.02760009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.7668,
+      "loss/crossentropy": 2.3393882513046265,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29145532846450806,
+      "step": 3570
+    },
+    {
+      "epoch": 0.07144,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.03980712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.8281,
+      "loss/crossentropy": 2.007299244403839,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30512700974941254,
+      "step": 3572
+    },
+    {
+      "epoch": 0.07148,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.0369049072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.7891,
+      "loss/crossentropy": 1.9879329800605774,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22510841488838196,
+      "step": 3574
+    },
+    {
+      "epoch": 0.07152,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.03443603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.9688,
+      "loss/crossentropy": 2.387833833694458,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3007010221481323,
+      "step": 3576
+    },
+    {
+      "epoch": 0.07156,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.03664957682291667,
+      "learning_rate": 0.0001,
+      "loss": 4.7975,
+      "loss/crossentropy": 2.3306411504745483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.269440695643425,
+      "step": 3578
+    },
+    {
+      "epoch": 0.0716,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.03462626139322917,
+      "learning_rate": 0.0001,
+      "loss": 4.5348,
+      "loss/crossentropy": 1.8359156847000122,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23353691399097443,
+      "step": 3580
+    },
+    {
+      "epoch": 0.07164,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.033080037434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.9226,
+      "loss/crossentropy": 2.257680654525757,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2764490246772766,
+      "step": 3582
+    },
+    {
+      "epoch": 0.07168,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.03310445149739583,
+      "learning_rate": 0.0001,
+      "loss": 4.6434,
+      "loss/crossentropy": 2.589483857154846,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28172188997268677,
+      "step": 3584
+    },
+    {
+      "epoch": 0.07172,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.04182840983072917,
+      "learning_rate": 0.0001,
+      "loss": 4.5434,
+      "loss/crossentropy": 1.8202016949653625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22330023348331451,
+      "step": 3586
+    },
+    {
+      "epoch": 0.07176,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.028609212239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.8344,
+      "loss/crossentropy": 2.0311816334724426,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2622206509113312,
+      "step": 3588
+    },
+    {
+      "epoch": 0.0718,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.028400675455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.5158,
+      "loss/crossentropy": 1.7991753220558167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23093865811824799,
+      "step": 3590
+    },
+    {
+      "epoch": 0.07184,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.030924479166666668,
+      "learning_rate": 0.0001,
+      "loss": 4.3148,
+      "loss/crossentropy": 1.6141473054885864,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21914401650428772,
+      "step": 3592
+    },
+    {
+      "epoch": 0.07188,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.0289703369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.8007,
+      "loss/crossentropy": 2.3337208032608032,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2612123265862465,
+      "step": 3594
+    },
+    {
+      "epoch": 0.07192,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0322662353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5811,
+      "loss/crossentropy": 2.191028594970703,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2598741352558136,
+      "step": 3596
+    },
+    {
+      "epoch": 0.07196,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.03178609212239583,
+      "learning_rate": 0.0001,
+      "loss": 4.9562,
+      "loss/crossentropy": 2.0293691158294678,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27143266797065735,
+      "step": 3598
+    },
+    {
+      "epoch": 0.072,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.03437398274739583,
+      "learning_rate": 0.0001,
+      "loss": 5.1335,
+      "loss/crossentropy": 2.1257725954055786,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33300966024398804,
+      "step": 3600
+    },
+    {
+      "epoch": 0.07204,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.0171875,
+      "learning_rate": 0.0001,
+      "loss": 4.8223,
+      "loss/crossentropy": 2.296278953552246,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28724825382232666,
+      "step": 3602
+    },
+    {
+      "epoch": 0.07208,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.016731770833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.8925,
+      "loss/crossentropy": 2.258358597755432,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26510895788669586,
+      "step": 3604
+    },
+    {
+      "epoch": 0.07212,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.0170318603515625,
+      "learning_rate": 0.0001,
+      "loss": 5.0383,
+      "loss/crossentropy": 2.0454649925231934,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2725224494934082,
+      "step": 3606
+    },
+    {
+      "epoch": 0.07216,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.015250651041666667,
+      "learning_rate": 0.0001,
+      "loss": 4.6582,
+      "loss/crossentropy": 2.1844204664230347,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2729914039373398,
+      "step": 3608
+    },
+    {
+      "epoch": 0.0722,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.015608723958333333,
+      "learning_rate": 0.0001,
+      "loss": 4.4613,
+      "loss/crossentropy": 1.8897106647491455,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26307350397109985,
+      "step": 3610
+    },
+    {
+      "epoch": 0.07224,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.013377888997395834,
+      "learning_rate": 0.0001,
+      "loss": 4.5695,
+      "loss/crossentropy": 1.9441962838172913,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24454529583454132,
+      "step": 3612
+    },
+    {
+      "epoch": 0.07228,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.063623046875,
+      "learning_rate": 0.0001,
+      "loss": 4.7654,
+      "loss/crossentropy": 2.10969078540802,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2900787442922592,
+      "step": 3614
+    },
+    {
+      "epoch": 0.07232,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.06301676432291667,
+      "learning_rate": 0.0001,
+      "loss": 4.5195,
+      "loss/crossentropy": 2.1384644508361816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2722831964492798,
+      "step": 3616
+    },
+    {
+      "epoch": 0.07236,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.06412760416666667,
+      "learning_rate": 0.0001,
+      "loss": 4.4995,
+      "loss/crossentropy": 2.0648157596588135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24860350787639618,
+      "step": 3618
+    },
+    {
+      "epoch": 0.0724,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.06441650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.7863,
+      "loss/crossentropy": 2.2188034057617188,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2476331740617752,
+      "step": 3620
+    },
+    {
+      "epoch": 0.07244,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.06516927083333333,
+      "learning_rate": 0.0001,
+      "loss": 4.792,
+      "loss/crossentropy": 2.1361395120620728,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24962469190359116,
+      "step": 3622
+    },
+    {
+      "epoch": 0.07248,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.06457417805989583,
+      "learning_rate": 0.0001,
+      "loss": 4.739,
+      "loss/crossentropy": 2.2646392583847046,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.260420486330986,
+      "step": 3624
+    },
+    {
+      "epoch": 0.07252,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.0630523681640625,
+      "learning_rate": 0.0001,
+      "loss": 4.6937,
+      "loss/crossentropy": 2.2822424173355103,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2585812509059906,
+      "step": 3626
+    },
+    {
+      "epoch": 0.07256,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.06852925618489583,
+      "learning_rate": 0.0001,
+      "loss": 4.2868,
+      "loss/crossentropy": 1.8197516798973083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24321961402893066,
+      "step": 3628
+    },
+    {
+      "epoch": 0.0726,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.007649739583333333,
+      "learning_rate": 0.0001,
+      "loss": 4.4835,
+      "loss/crossentropy": 2.1475032567977905,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2687895894050598,
+      "step": 3630
+    },
+    {
+      "epoch": 0.07264,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.006538899739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.2353,
+      "loss/crossentropy": 1.9497992992401123,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24942665547132492,
+      "step": 3632
+    },
+    {
+      "epoch": 0.07268,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.007059733072916667,
+      "learning_rate": 0.0001,
+      "loss": 4.7455,
+      "loss/crossentropy": 1.8786492347717285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23713821917772293,
+      "step": 3634
+    },
+    {
+      "epoch": 0.07272,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.00738525390625,
+      "learning_rate": 0.0001,
+      "loss": 5.089,
+      "loss/crossentropy": 2.474532127380371,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3133770227432251,
+      "step": 3636
+    },
+    {
+      "epoch": 0.07276,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.0073394775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.912,
+      "loss/crossentropy": 2.1231455206871033,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2547219544649124,
+      "step": 3638
+    },
+    {
+      "epoch": 0.0728,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.009007771809895834,
+      "learning_rate": 0.0001,
+      "loss": 4.4727,
+      "loss/crossentropy": 2.0511630177497864,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26947300136089325,
+      "step": 3640
+    },
+    {
+      "epoch": 0.07284,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.01060791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.7332,
+      "loss/crossentropy": 1.7076187133789062,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21764510869979858,
+      "step": 3642
+    },
+    {
+      "epoch": 0.07288,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.009723917643229166,
+      "learning_rate": 0.0001,
+      "loss": 4.8396,
+      "loss/crossentropy": 2.069926142692566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26851218193769455,
+      "step": 3644
+    },
+    {
+      "epoch": 0.07292,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.010054524739583333,
+      "learning_rate": 0.0001,
+      "loss": 5.0802,
+      "loss/crossentropy": 2.1369277238845825,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27878354489803314,
+      "step": 3646
+    },
+    {
+      "epoch": 0.07296,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.008072916666666667,
+      "learning_rate": 0.0001,
+      "loss": 4.3659,
+      "loss/crossentropy": 2.095974624156952,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25285808742046356,
+      "step": 3648
+    },
+    {
+      "epoch": 0.073,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.011888631184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.9301,
+      "loss/crossentropy": 2.240627646446228,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24894578754901886,
+      "step": 3650
+    },
+    {
+      "epoch": 0.07304,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.013899739583333333,
+      "learning_rate": 0.0001,
+      "loss": 4.6588,
+      "loss/crossentropy": 2.2270851135253906,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26908691227436066,
+      "step": 3652
+    },
+    {
+      "epoch": 0.07308,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.017039998372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.7385,
+      "loss/crossentropy": 2.2684017419815063,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28442947566509247,
+      "step": 3654
+    },
+    {
+      "epoch": 0.07312,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.07785542805989583,
+      "learning_rate": 0.0001,
+      "loss": 4.7585,
+      "loss/crossentropy": 2.0922030806541443,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2621888816356659,
+      "step": 3656
+    },
+    {
+      "epoch": 0.07316,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.08323567708333333,
+      "learning_rate": 0.0001,
+      "loss": 4.7425,
+      "loss/crossentropy": 2.0134947896003723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2717447876930237,
+      "step": 3658
+    },
+    {
+      "epoch": 0.0732,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.08206278483072917,
+      "learning_rate": 0.0001,
+      "loss": 4.5573,
+      "loss/crossentropy": 1.9246947765350342,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25095127522945404,
+      "step": 3660
+    },
+    {
+      "epoch": 0.07324,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.08561197916666667,
+      "learning_rate": 0.0001,
+      "loss": 4.8616,
+      "loss/crossentropy": 2.0655113458633423,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24166538566350937,
+      "step": 3662
+    },
+    {
+      "epoch": 0.07328,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.08198954264322916,
+      "learning_rate": 0.0001,
+      "loss": 4.9137,
+      "loss/crossentropy": 2.2706735730171204,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26484307646751404,
+      "step": 3664
+    },
+    {
+      "epoch": 0.07332,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.0883453369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2911,
+      "loss/crossentropy": 1.7969809770584106,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23407897353172302,
+      "step": 3666
+    },
+    {
+      "epoch": 0.07336,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.09109700520833333,
+      "learning_rate": 0.0001,
+      "loss": 4.7081,
+      "loss/crossentropy": 2.0398870706558228,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2502745985984802,
+      "step": 3668
+    },
+    {
+      "epoch": 0.0734,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.09381103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.8738,
+      "loss/crossentropy": 2.090283453464508,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2936585247516632,
+      "step": 3670
+    },
+    {
+      "epoch": 0.07344,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.035008748372395836,
+      "learning_rate": 0.0001,
+      "loss": 4.8305,
+      "loss/crossentropy": 2.286925792694092,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27951307594776154,
+      "step": 3672
+    },
+    {
+      "epoch": 0.07348,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.032059733072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.4885,
+      "loss/crossentropy": 2.0264610052108765,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24908316135406494,
+      "step": 3674
+    },
+    {
+      "epoch": 0.07352,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.0372467041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.9239,
+      "loss/crossentropy": 2.1947755217552185,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2865990996360779,
+      "step": 3676
+    },
+    {
+      "epoch": 0.07356,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.0422515869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.8784,
+      "loss/crossentropy": 2.0050416588783264,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23362033069133759,
+      "step": 3678
+    },
+    {
+      "epoch": 0.0736,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.041112263997395836,
+      "learning_rate": 0.0001,
+      "loss": 4.7423,
+      "loss/crossentropy": 1.8935424089431763,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2532464489340782,
+      "step": 3680
+    },
+    {
+      "epoch": 0.07364,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.0350250244140625,
+      "learning_rate": 0.0001,
+      "loss": 4.7389,
+      "loss/crossentropy": 2.0181053280830383,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2544455900788307,
+      "step": 3682
+    },
+    {
+      "epoch": 0.07368,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.03144124348958333,
+      "learning_rate": 0.0001,
+      "loss": 4.7099,
+      "loss/crossentropy": 2.1172796487808228,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27609144151210785,
+      "step": 3684
+    },
+    {
+      "epoch": 0.07372,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.023737589518229168,
+      "learning_rate": 0.0001,
+      "loss": 4.7185,
+      "loss/crossentropy": 2.3926355838775635,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28892992436885834,
+      "step": 3686
+    },
+    {
+      "epoch": 0.07376,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.026883951822916665,
+      "learning_rate": 0.0001,
+      "loss": 4.9086,
+      "loss/crossentropy": 2.2512835264205933,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26166096329689026,
+      "step": 3688
+    },
+    {
+      "epoch": 0.0738,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.0264312744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.6311,
+      "loss/crossentropy": 2.0656538009643555,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25399819016456604,
+      "step": 3690
+    },
+    {
+      "epoch": 0.07384,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.0222808837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.9565,
+      "loss/crossentropy": 2.454928994178772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2964669317007065,
+      "step": 3692
+    },
+    {
+      "epoch": 0.07388,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.009859212239583333,
+      "learning_rate": 0.0001,
+      "loss": 4.5703,
+      "loss/crossentropy": 1.988040804862976,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27345800399780273,
+      "step": 3694
+    },
+    {
+      "epoch": 0.07392,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.013752237955729166,
+      "learning_rate": 0.0001,
+      "loss": 4.9418,
+      "loss/crossentropy": 1.910742998123169,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24037255346775055,
+      "step": 3696
+    },
+    {
+      "epoch": 0.07396,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.013700358072916667,
+      "learning_rate": 0.0001,
+      "loss": 4.6541,
+      "loss/crossentropy": 2.19545578956604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2778017520904541,
+      "step": 3698
+    },
+    {
+      "epoch": 0.074,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.0127838134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3839,
+      "loss/crossentropy": 2.436691403388977,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26254843175411224,
+      "step": 3700
+    },
+    {
+      "epoch": 0.07404,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.012398274739583333,
+      "learning_rate": 0.0001,
+      "loss": 4.833,
+      "loss/crossentropy": 2.7458308935165405,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27952516078948975,
+      "step": 3702
+    },
+    {
+      "epoch": 0.07408,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.014876302083333333,
+      "learning_rate": 0.0001,
+      "loss": 4.8308,
+      "loss/crossentropy": 2.2321633100509644,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2633504122495651,
+      "step": 3704
+    },
+    {
+      "epoch": 0.07412,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.013801066080729167,
+      "learning_rate": 0.0001,
+      "loss": 4.8159,
+      "loss/crossentropy": 1.9883576035499573,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25319087505340576,
+      "step": 3706
+    },
+    {
+      "epoch": 0.07416,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.0147857666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.5546,
+      "loss/crossentropy": 1.7647870182991028,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21732009947299957,
+      "step": 3708
+    },
+    {
+      "epoch": 0.0742,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.046793619791666664,
+      "learning_rate": 0.0001,
+      "loss": 4.9271,
+      "loss/crossentropy": 2.1113381385803223,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25888970494270325,
+      "step": 3710
+    },
+    {
+      "epoch": 0.07424,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.04690348307291667,
+      "learning_rate": 0.0001,
+      "loss": 4.5878,
+      "loss/crossentropy": 1.975549578666687,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23899925500154495,
+      "step": 3712
+    },
+    {
+      "epoch": 0.07428,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.04664306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.9262,
+      "loss/crossentropy": 2.0562495589256287,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31428879499435425,
+      "step": 3714
+    },
+    {
+      "epoch": 0.07432,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.04951883951822917,
+      "learning_rate": 0.0001,
+      "loss": 4.3719,
+      "loss/crossentropy": 2.114805221557617,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25172004848718643,
+      "step": 3716
+    },
+    {
+      "epoch": 0.07436,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.052783203125,
+      "learning_rate": 0.0001,
+      "loss": 4.6032,
+      "loss/crossentropy": 2.1865739822387695,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.252632200717926,
+      "step": 3718
+    },
+    {
+      "epoch": 0.0744,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.04739176432291667,
+      "learning_rate": 0.0001,
+      "loss": 4.8493,
+      "loss/crossentropy": 2.2550876140594482,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27425335347652435,
+      "step": 3720
+    },
+    {
+      "epoch": 0.07444,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.047379557291666666,
+      "learning_rate": 0.0001,
+      "loss": 4.9072,
+      "loss/crossentropy": 2.293414354324341,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26979324221611023,
+      "step": 3722
+    },
+    {
+      "epoch": 0.07448,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.043745930989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.4962,
+      "loss/crossentropy": 2.014510452747345,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25709769129753113,
+      "step": 3724
+    },
+    {
+      "epoch": 0.07452,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.015746053059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.6485,
+      "loss/crossentropy": 2.0332603454589844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2585446834564209,
+      "step": 3726
+    },
+    {
+      "epoch": 0.07456,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.0165191650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.7268,
+      "loss/crossentropy": 1.9425334930419922,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22921039909124374,
+      "step": 3728
+    },
+    {
+      "epoch": 0.0746,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.019090779622395835,
+      "learning_rate": 0.0001,
+      "loss": 4.9306,
+      "loss/crossentropy": 2.1233898997306824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26014500856399536,
+      "step": 3730
+    },
+    {
+      "epoch": 0.07464,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.023566691080729167,
+      "learning_rate": 0.0001,
+      "loss": 4.9958,
+      "loss/crossentropy": 2.3929240703582764,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2791339308023453,
+      "step": 3732
+    },
+    {
+      "epoch": 0.07468,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.021370442708333333,
+      "learning_rate": 0.0001,
+      "loss": 4.6072,
+      "loss/crossentropy": 2.163137674331665,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26793332397937775,
+      "step": 3734
+    },
+    {
+      "epoch": 0.07472,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.024738566080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.723,
+      "loss/crossentropy": 2.1300129294395447,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2533607929944992,
+      "step": 3736
+    },
+    {
+      "epoch": 0.07476,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.028055826822916668,
+      "learning_rate": 0.0001,
+      "loss": 4.7232,
+      "loss/crossentropy": 1.9808942675590515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25632843375205994,
+      "step": 3738
+    },
+    {
+      "epoch": 0.0748,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.03328348795572917,
+      "learning_rate": 0.0001,
+      "loss": 4.8219,
+      "loss/crossentropy": 2.161437451839447,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2529585212469101,
+      "step": 3740
+    },
+    {
+      "epoch": 0.07484,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.03369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.6541,
+      "loss/crossentropy": 1.852737545967102,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2333993762731552,
+      "step": 3742
+    },
+    {
+      "epoch": 0.07488,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.03439127604166667,
+      "learning_rate": 0.0001,
+      "loss": 4.4355,
+      "loss/crossentropy": 1.664733350276947,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20309723168611526,
+      "step": 3744
+    },
+    {
+      "epoch": 0.07492,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.03276265462239583,
+      "learning_rate": 0.0001,
+      "loss": 4.4554,
+      "loss/crossentropy": 2.1815799474716187,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2470541000366211,
+      "step": 3746
+    },
+    {
+      "epoch": 0.07496,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.024657185872395834,
+      "learning_rate": 0.0001,
+      "loss": 4.7423,
+      "loss/crossentropy": 1.9546562433242798,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2685271203517914,
+      "step": 3748
+    },
+    {
+      "epoch": 0.075,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.024738566080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.5171,
+      "loss/crossentropy": 1.920817255973816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2558089941740036,
+      "step": 3750
+    },
+    {
+      "epoch": 0.07504,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.024982706705729166,
+      "learning_rate": 0.0001,
+      "loss": 4.4795,
+      "loss/crossentropy": 1.7570490837097168,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2265816479921341,
+      "step": 3752
+    },
+    {
+      "epoch": 0.07508,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 0.203564453125,
+      "learning_rate": 0.0001,
+      "loss": 4.756,
+      "loss/crossentropy": 2.1815105676651,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2536233216524124,
+      "step": 3754
+    },
+    {
+      "epoch": 0.07512,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.19650777180989584,
+      "learning_rate": 0.0001,
+      "loss": 4.7178,
+      "loss/crossentropy": 2.1385504603385925,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2445325404405594,
+      "step": 3756
+    },
+    {
+      "epoch": 0.07516,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.19650777180989584,
+      "learning_rate": 0.0001,
+      "loss": 4.6449,
+      "loss/crossentropy": 1.7325092554092407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2419673353433609,
+      "step": 3758
+    },
+    {
+      "epoch": 0.0752,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.19572652180989583,
+      "learning_rate": 0.0001,
+      "loss": 4.6096,
+      "loss/crossentropy": 2.358627676963806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2713842839002609,
+      "step": 3760
+    },
+    {
+      "epoch": 0.07524,
+      "grad_norm": 3.078125,
+      "grad_norm_var": 0.20392252604166666,
+      "learning_rate": 0.0001,
+      "loss": 5.1712,
+      "loss/crossentropy": 2.048672080039978,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3007172644138336,
+      "step": 3762
+    },
+    {
+      "epoch": 0.07528,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.19885660807291666,
+      "learning_rate": 0.0001,
+      "loss": 4.8473,
+      "loss/crossentropy": 2.2967183589935303,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28548331558704376,
+      "step": 3764
+    },
+    {
+      "epoch": 0.07532,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.19228108723958334,
+      "learning_rate": 0.0001,
+      "loss": 4.7541,
+      "loss/crossentropy": 2.1280438899993896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2825516611337662,
+      "step": 3766
+    },
+    {
+      "epoch": 0.07536,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.19146728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.8404,
+      "loss/crossentropy": 2.5528002977371216,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28681764006614685,
+      "step": 3768
+    },
+    {
+      "epoch": 0.0754,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.0549468994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.729,
+      "loss/crossentropy": 2.235885262489319,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.259520560503006,
+      "step": 3770
+    },
+    {
+      "epoch": 0.07544,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.05718994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.4705,
+      "loss/crossentropy": 1.8836966753005981,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2360890954732895,
+      "step": 3772
+    },
+    {
+      "epoch": 0.07548,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.056538899739583336,
+      "learning_rate": 0.0001,
+      "loss": 4.9291,
+      "loss/crossentropy": 2.3396376371383667,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30392636358737946,
+      "step": 3774
+    },
+    {
+      "epoch": 0.07552,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.051301066080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.7518,
+      "loss/crossentropy": 2.4024877548217773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2858506590127945,
+      "step": 3776
+    },
+    {
+      "epoch": 0.07556,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4073,
+      "loss/crossentropy": 2.138229727745056,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25458595901727676,
+      "step": 3778
+    },
+    {
+      "epoch": 0.0756,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.0359375,
+      "learning_rate": 0.0001,
+      "loss": 4.3882,
+      "loss/crossentropy": 1.8413254618644714,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23022352159023285,
+      "step": 3780
+    },
+    {
+      "epoch": 0.07564,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.034764607747395836,
+      "learning_rate": 0.0001,
+      "loss": 4.613,
+      "loss/crossentropy": 1.8554572463035583,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25644390285015106,
+      "step": 3782
+    },
+    {
+      "epoch": 0.07568,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.018452962239583332,
+      "learning_rate": 0.0001,
+      "loss": 4.7013,
+      "loss/crossentropy": 2.0096731781959534,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25370142608880997,
+      "step": 3784
+    },
+    {
+      "epoch": 0.07572,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.020018513997395834,
+      "learning_rate": 0.0001,
+      "loss": 4.9317,
+      "loss/crossentropy": 1.7932413220405579,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2426912784576416,
+      "step": 3786
+    },
+    {
+      "epoch": 0.07576,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0263336181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1599,
+      "loss/crossentropy": 2.0372042655944824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24203844368457794,
+      "step": 3788
+    },
+    {
+      "epoch": 0.0758,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.021773274739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.3627,
+      "loss/crossentropy": 1.8986076712608337,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24614746868610382,
+      "step": 3790
+    },
+    {
+      "epoch": 0.07584,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.022135416666666668,
+      "learning_rate": 0.0001,
+      "loss": 4.563,
+      "loss/crossentropy": 1.8080393075942993,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23206676542758942,
+      "step": 3792
+    },
+    {
+      "epoch": 0.07588,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.015067545572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.6857,
+      "loss/crossentropy": 1.7578041553497314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21686340868473053,
+      "step": 3794
+    },
+    {
+      "epoch": 0.07592,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.015087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4938,
+      "loss/crossentropy": 2.0115376710891724,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2651440501213074,
+      "step": 3796
+    },
+    {
+      "epoch": 0.07596,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.015803019205729168,
+      "learning_rate": 0.0001,
+      "loss": 4.598,
+      "loss/crossentropy": 2.028555393218994,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26161982119083405,
+      "step": 3798
+    },
+    {
+      "epoch": 0.076,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.016429646809895834,
+      "learning_rate": 0.0001,
+      "loss": 4.8315,
+      "loss/crossentropy": 2.158663272857666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28110067546367645,
+      "step": 3800
+    },
+    {
+      "epoch": 0.07604,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.019652303059895834,
+      "learning_rate": 0.0001,
+      "loss": 5.2376,
+      "loss/crossentropy": 2.2959556579589844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28477419912815094,
+      "step": 3802
+    },
+    {
+      "epoch": 0.07608,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.014997355143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.5289,
+      "loss/crossentropy": 2.149766206741333,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25216003507375717,
+      "step": 3804
+    },
+    {
+      "epoch": 0.07612,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.01416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.7496,
+      "loss/crossentropy": 1.9866302609443665,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26193149387836456,
+      "step": 3806
+    },
+    {
+      "epoch": 0.07616,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.0122955322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.9968,
+      "loss/crossentropy": 2.4230403900146484,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27897247672080994,
+      "step": 3808
+    },
+    {
+      "epoch": 0.0762,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.012613932291666666,
+      "learning_rate": 0.0001,
+      "loss": 4.9133,
+      "loss/crossentropy": 2.2995522022247314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27623192965984344,
+      "step": 3810
+    },
+    {
+      "epoch": 0.07624,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.013923136393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.6632,
+      "loss/crossentropy": 2.167468547821045,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26121728122234344,
+      "step": 3812
+    },
+    {
+      "epoch": 0.07628,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.013353474934895833,
+      "learning_rate": 0.0001,
+      "loss": 4.8435,
+      "loss/crossentropy": 2.3259944915771484,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29026439785957336,
+      "step": 3814
+    },
+    {
+      "epoch": 0.07632,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.041825358072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.8704,
+      "loss/crossentropy": 2.18080472946167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2550061345100403,
+      "step": 3816
+    },
+    {
+      "epoch": 0.07636,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.03871968587239583,
+      "learning_rate": 0.0001,
+      "loss": 4.5681,
+      "loss/crossentropy": 2.1185330748558044,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25014883279800415,
+      "step": 3818
+    },
+    {
+      "epoch": 0.0764,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.039383951822916666,
+      "learning_rate": 0.0001,
+      "loss": 4.5776,
+      "loss/crossentropy": 1.9028193354606628,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22502654790878296,
+      "step": 3820
+    },
+    {
+      "epoch": 0.07644,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.039713541666666664,
+      "learning_rate": 0.0001,
+      "loss": 5.0188,
+      "loss/crossentropy": 2.266402840614319,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24728389829397202,
+      "step": 3822
+    },
+    {
+      "epoch": 0.07648,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.04279683430989583,
+      "learning_rate": 0.0001,
+      "loss": 4.7036,
+      "loss/crossentropy": 2.0918440222740173,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2669295519590378,
+      "step": 3824
+    },
+    {
+      "epoch": 0.07652,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.04456380208333333,
+      "learning_rate": 0.0001,
+      "loss": 4.1446,
+      "loss/crossentropy": 1.6120481491088867,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21910656243562698,
+      "step": 3826
+    },
+    {
+      "epoch": 0.07656,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.044331868489583336,
+      "learning_rate": 0.0001,
+      "loss": 4.6352,
+      "loss/crossentropy": 2.2294809818267822,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2720891535282135,
+      "step": 3828
+    },
+    {
+      "epoch": 0.0766,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.0524322509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.873,
+      "loss/crossentropy": 2.2588730454444885,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2613115608692169,
+      "step": 3830
+    },
+    {
+      "epoch": 0.07664,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.02584228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5881,
+      "loss/crossentropy": 2.2658292055130005,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2729395925998688,
+      "step": 3832
+    },
+    {
+      "epoch": 0.07668,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.02662353515625,
+      "learning_rate": 0.0001,
+      "loss": 5.1787,
+      "loss/crossentropy": 2.314574718475342,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28539060056209564,
+      "step": 3834
+    },
+    {
+      "epoch": 0.07672,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.024738566080729166,
+      "learning_rate": 0.0001,
+      "loss": 5.0388,
+      "loss/crossentropy": 2.4684417247772217,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29843954741954803,
+      "step": 3836
+    },
+    {
+      "epoch": 0.07676,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.025944010416666666,
+      "learning_rate": 0.0001,
+      "loss": 4.5976,
+      "loss/crossentropy": 2.528733253479004,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2696636766195297,
+      "step": 3838
+    },
+    {
+      "epoch": 0.0768,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.025340779622395834,
+      "learning_rate": 0.0001,
+      "loss": 4.6449,
+      "loss/crossentropy": 2.203901529312134,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26598773896694183,
+      "step": 3840
+    },
+    {
+      "epoch": 0.07684,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.021907552083333334,
+      "learning_rate": 0.0001,
+      "loss": 4.7322,
+      "loss/crossentropy": 1.9192892909049988,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25631098449230194,
+      "step": 3842
+    },
+    {
+      "epoch": 0.07688,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.022077433268229165,
+      "learning_rate": 0.0001,
+      "loss": 4.6372,
+      "loss/crossentropy": 1.8314838409423828,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22842589765787125,
+      "step": 3844
+    },
+    {
+      "epoch": 0.07692,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.013395182291666667,
+      "learning_rate": 0.0001,
+      "loss": 4.6023,
+      "loss/crossentropy": 2.2416744232177734,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24890189617872238,
+      "step": 3846
+    },
+    {
+      "epoch": 0.07696,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.012580362955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.8454,
+      "loss/crossentropy": 2.034749209880829,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29580502212047577,
+      "step": 3848
+    },
+    {
+      "epoch": 0.077,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.0117095947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.5923,
+      "loss/crossentropy": 1.9982805848121643,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2618003487586975,
+      "step": 3850
+    },
+    {
+      "epoch": 0.07704,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.0127105712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.7704,
+      "loss/crossentropy": 2.065816104412079,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2571987137198448,
+      "step": 3852
+    },
+    {
+      "epoch": 0.07708,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.0100982666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.7493,
+      "loss/crossentropy": 1.933334231376648,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24824900180101395,
+      "step": 3854
+    },
+    {
+      "epoch": 0.07712,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.010514322916666667,
+      "learning_rate": 0.0001,
+      "loss": 4.5805,
+      "loss/crossentropy": 1.9197405576705933,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2559673935174942,
+      "step": 3856
+    },
+    {
+      "epoch": 0.07716,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.010252888997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.429,
+      "loss/crossentropy": 2.307250142097473,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2613677680492401,
+      "step": 3858
+    },
+    {
+      "epoch": 0.0772,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.01011962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4494,
+      "loss/crossentropy": 2.1120635271072388,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24830741435289383,
+      "step": 3860
+    },
+    {
+      "epoch": 0.07724,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.039013671875,
+      "learning_rate": 0.0001,
+      "loss": 4.8585,
+      "loss/crossentropy": 2.404169201850891,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2960711419582367,
+      "step": 3862
+    },
+    {
+      "epoch": 0.07728,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.038899739583333336,
+      "learning_rate": 0.0001,
+      "loss": 5.0257,
+      "loss/crossentropy": 2.2490307688713074,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2569812461733818,
+      "step": 3864
+    },
+    {
+      "epoch": 0.07732,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.0386871337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.9086,
+      "loss/crossentropy": 2.0773178339004517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2505042105913162,
+      "step": 3866
+    },
+    {
+      "epoch": 0.07736,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.038899739583333336,
+      "learning_rate": 0.0001,
+      "loss": 4.4621,
+      "loss/crossentropy": 1.83626389503479,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22876176983118057,
+      "step": 3868
+    },
+    {
+      "epoch": 0.0774,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.03911031087239583,
+      "learning_rate": 0.0001,
+      "loss": 4.5298,
+      "loss/crossentropy": 1.8159971833229065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22787011414766312,
+      "step": 3870
+    },
+    {
+      "epoch": 0.07744,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.0378326416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.5995,
+      "loss/crossentropy": 2.0361026525497437,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23802263289690018,
+      "step": 3872
+    },
+    {
+      "epoch": 0.07748,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.04433492024739583,
+      "learning_rate": 0.0001,
+      "loss": 4.9506,
+      "loss/crossentropy": 2.2464375495910645,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26598919928073883,
+      "step": 3874
+    },
+    {
+      "epoch": 0.07752,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.0408355712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.9416,
+      "loss/crossentropy": 2.163287401199341,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29068198800086975,
+      "step": 3876
+    },
+    {
+      "epoch": 0.07756,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.22014567057291667,
+      "learning_rate": 0.0001,
+      "loss": 4.9478,
+      "loss/crossentropy": 2.1638875007629395,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2609352171421051,
+      "step": 3878
+    },
+    {
+      "epoch": 0.0776,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.21923421223958334,
+      "learning_rate": 0.0001,
+      "loss": 4.6247,
+      "loss/crossentropy": 1.9216270446777344,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25062238425016403,
+      "step": 3880
+    },
+    {
+      "epoch": 0.07764,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.21585184733072918,
+      "learning_rate": 0.0001,
+      "loss": 5.0786,
+      "loss/crossentropy": 2.036958694458008,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2897229939699173,
+      "step": 3882
+    },
+    {
+      "epoch": 0.07768,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.2094635009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.6062,
+      "loss/crossentropy": 2.1493492126464844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.258526012301445,
+      "step": 3884
+    },
+    {
+      "epoch": 0.07772,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.20414937337239583,
+      "learning_rate": 0.0001,
+      "loss": 4.468,
+      "loss/crossentropy": 2.0496288537979126,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2386137992143631,
+      "step": 3886
+    },
+    {
+      "epoch": 0.07776,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.20829671223958332,
+      "learning_rate": 0.0001,
+      "loss": 4.2402,
+      "loss/crossentropy": 1.5763422846794128,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2227308303117752,
+      "step": 3888
+    },
+    {
+      "epoch": 0.0778,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.20852864583333333,
+      "learning_rate": 0.0001,
+      "loss": 5.0582,
+      "loss/crossentropy": 2.5032416582107544,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3047761619091034,
+      "step": 3890
+    },
+    {
+      "epoch": 0.07784,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.21199544270833334,
+      "learning_rate": 0.0001,
+      "loss": 4.7694,
+      "loss/crossentropy": 2.3609601259231567,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27848154306411743,
+      "step": 3892
+    },
+    {
+      "epoch": 0.07788,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.0202056884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.776,
+      "loss/crossentropy": 2.188078999519348,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2570301741361618,
+      "step": 3894
+    },
+    {
+      "epoch": 0.07792,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.018733723958333334,
+      "learning_rate": 0.0001,
+      "loss": 4.8374,
+      "loss/crossentropy": 1.9860637784004211,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25376833230257034,
+      "step": 3896
+    },
+    {
+      "epoch": 0.07796,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.010399373372395833,
+      "learning_rate": 0.0001,
+      "loss": 4.404,
+      "loss/crossentropy": 2.0886037945747375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24766233563423157,
+      "step": 3898
+    },
+    {
+      "epoch": 0.078,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.009845987955729166,
+      "learning_rate": 0.0001,
+      "loss": 4.6833,
+      "loss/crossentropy": 2.373010039329529,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27916407585144043,
+      "step": 3900
+    },
+    {
+      "epoch": 0.07804,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.009837849934895834,
+      "learning_rate": 0.0001,
+      "loss": 4.6295,
+      "loss/crossentropy": 1.6733890771865845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2211885154247284,
+      "step": 3902
+    },
+    {
+      "epoch": 0.07808,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.010184733072916667,
+      "learning_rate": 0.0001,
+      "loss": 4.6588,
+      "loss/crossentropy": 2.0506675243377686,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27091431617736816,
+      "step": 3904
+    },
+    {
+      "epoch": 0.07812,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.008784993489583334,
+      "learning_rate": 0.0001,
+      "loss": 4.712,
+      "loss/crossentropy": 2.3200724124908447,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26983049511909485,
+      "step": 3906
+    },
+    {
+      "epoch": 0.07816,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.0164459228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.7171,
+      "loss/crossentropy": 1.928814709186554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2423655241727829,
+      "step": 3908
+    },
+    {
+      "epoch": 0.0782,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.016389973958333335,
+      "learning_rate": 0.0001,
+      "loss": 4.6944,
+      "loss/crossentropy": 2.007555842399597,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2444767728447914,
+      "step": 3910
+    },
+    {
+      "epoch": 0.07824,
+      "grad_norm": 2.875,
+      "grad_norm_var": 0.027469889322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.7955,
+      "loss/crossentropy": 2.2054057121276855,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2849784791469574,
+      "step": 3912
+    },
+    {
+      "epoch": 0.07828,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.03986002604166667,
+      "learning_rate": 0.0001,
+      "loss": 4.6885,
+      "loss/crossentropy": 2.331532597541809,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2820790112018585,
+      "step": 3914
+    },
+    {
+      "epoch": 0.07832,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.036519368489583336,
+      "learning_rate": 0.0001,
+      "loss": 4.6967,
+      "loss/crossentropy": 2.142041563987732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25799722969532013,
+      "step": 3916
+    },
+    {
+      "epoch": 0.07836,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.0394683837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5415,
+      "loss/crossentropy": 2.010735809803009,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24202881753444672,
+      "step": 3918
+    },
+    {
+      "epoch": 0.0784,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.03970947265625,
+      "learning_rate": 0.0001,
+      "loss": 5.037,
+      "loss/crossentropy": 2.382808804512024,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27038049697875977,
+      "step": 3920
+    },
+    {
+      "epoch": 0.07844,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.0372955322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.7708,
+      "loss/crossentropy": 2.099658191204071,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2557126358151436,
+      "step": 3922
+    },
+    {
+      "epoch": 0.07848,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.03968098958333333,
+      "learning_rate": 0.0001,
+      "loss": 4.3775,
+      "loss/crossentropy": 1.7840275764465332,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22102414071559906,
+      "step": 3924
+    },
+    {
+      "epoch": 0.07852,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.039159138997395836,
+      "learning_rate": 0.0001,
+      "loss": 4.6637,
+      "loss/crossentropy": 1.8730725049972534,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23219943791627884,
+      "step": 3926
+    },
+    {
+      "epoch": 0.07856,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.034520467122395836,
+      "learning_rate": 0.0001,
+      "loss": 4.4432,
+      "loss/crossentropy": 1.9218623638153076,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23333143442869186,
+      "step": 3928
+    },
+    {
+      "epoch": 0.0786,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.0152740478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.8944,
+      "loss/crossentropy": 1.9885727763175964,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2596626430749893,
+      "step": 3930
+    },
+    {
+      "epoch": 0.07864,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.014989217122395834,
+      "learning_rate": 0.0001,
+      "loss": 4.9358,
+      "loss/crossentropy": 2.397018015384674,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2790713906288147,
+      "step": 3932
+    },
+    {
+      "epoch": 0.07868,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.014383951822916666,
+      "learning_rate": 0.0001,
+      "loss": 4.3062,
+      "loss/crossentropy": 1.7345170378684998,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23668452352285385,
+      "step": 3934
+    },
+    {
+      "epoch": 0.07872,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.013309733072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.6173,
+      "loss/crossentropy": 1.8630162477493286,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2371346428990364,
+      "step": 3936
+    },
+    {
+      "epoch": 0.07876,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.010367838541666667,
+      "learning_rate": 0.0001,
+      "loss": 4.5774,
+      "loss/crossentropy": 2.0738128423690796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24340355396270752,
+      "step": 3938
+    },
+    {
+      "epoch": 0.0788,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.010835774739583333,
+      "learning_rate": 0.0001,
+      "loss": 5.0027,
+      "loss/crossentropy": 2.2932467460632324,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2527567446231842,
+      "step": 3940
+    },
+    {
+      "epoch": 0.07884,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.009956868489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.6785,
+      "loss/crossentropy": 2.0000113248825073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23461253196001053,
+      "step": 3942
+    },
+    {
+      "epoch": 0.07888,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.008430989583333333,
+      "learning_rate": 0.0001,
+      "loss": 4.7307,
+      "loss/crossentropy": 2.0753955841064453,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2535083442926407,
+      "step": 3944
+    },
+    {
+      "epoch": 0.07892,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.008153279622395834,
+      "learning_rate": 0.0001,
+      "loss": 4.6906,
+      "loss/crossentropy": 2.167261242866516,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24734552949666977,
+      "step": 3946
+    },
+    {
+      "epoch": 0.07896,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.0108795166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.7347,
+      "loss/crossentropy": 1.9755831956863403,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24861737340688705,
+      "step": 3948
+    },
+    {
+      "epoch": 0.079,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.009261067708333333,
+      "learning_rate": 0.0001,
+      "loss": 4.6813,
+      "loss/crossentropy": 2.195417881011963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2518697530031204,
+      "step": 3950
+    },
+    {
+      "epoch": 0.07904,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.01920166015625,
+      "learning_rate": 0.0001,
+      "loss": 5.0172,
+      "loss/crossentropy": 2.5771371126174927,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28423887491226196,
+      "step": 3952
+    },
+    {
+      "epoch": 0.07908,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.017870076497395835,
+      "learning_rate": 0.0001,
+      "loss": 4.7071,
+      "loss/crossentropy": 1.683276355266571,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2075405865907669,
+      "step": 3954
+    },
+    {
+      "epoch": 0.07912,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.018993123372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.7128,
+      "loss/crossentropy": 2.2983756065368652,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28010235726833344,
+      "step": 3956
+    },
+    {
+      "epoch": 0.07916,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.020361328125,
+      "learning_rate": 0.0001,
+      "loss": 4.7896,
+      "loss/crossentropy": 2.3263272047042847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28625819087028503,
+      "step": 3958
+    },
+    {
+      "epoch": 0.0792,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.0204742431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.5201,
+      "loss/crossentropy": 1.9820871353149414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2594129145145416,
+      "step": 3960
+    },
+    {
+      "epoch": 0.07924,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.025634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4754,
+      "loss/crossentropy": 1.8991515636444092,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23694587498903275,
+      "step": 3962
+    },
+    {
+      "epoch": 0.07928,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.03843994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.9865,
+      "loss/crossentropy": 2.485508918762207,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2957809865474701,
+      "step": 3964
+    },
+    {
+      "epoch": 0.07932,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.037262980143229166,
+      "learning_rate": 0.0001,
+      "loss": 4.8871,
+      "loss/crossentropy": 2.156081974506378,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27474651485681534,
+      "step": 3966
+    },
+    {
+      "epoch": 0.07936,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.028544108072916668,
+      "learning_rate": 0.0001,
+      "loss": 4.8694,
+      "loss/crossentropy": 2.0370571613311768,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2732074484229088,
+      "step": 3968
+    },
+    {
+      "epoch": 0.0794,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.028251139322916667,
+      "learning_rate": 0.0001,
+      "loss": 5.062,
+      "loss/crossentropy": 2.3039989471435547,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31054411828517914,
+      "step": 3970
+    },
+    {
+      "epoch": 0.07944,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.028641764322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.7875,
+      "loss/crossentropy": 2.2280107736587524,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2711277902126312,
+      "step": 3972
+    },
+    {
+      "epoch": 0.07948,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.02935791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.8992,
+      "loss/crossentropy": 2.0609869956970215,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24142058193683624,
+      "step": 3974
+    },
+    {
+      "epoch": 0.07952,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.03277587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1391,
+      "loss/crossentropy": 2.0541876554489136,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24537865817546844,
+      "step": 3976
+    },
+    {
+      "epoch": 0.07956,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.6102701822916666,
+      "learning_rate": 0.0001,
+      "loss": 4.9591,
+      "loss/crossentropy": 2.344236969947815,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2770863175392151,
+      "step": 3978
+    },
+    {
+      "epoch": 0.0796,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.65465087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.7232,
+      "loss/crossentropy": 1.7899338603019714,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23951984196901321,
+      "step": 3980
+    },
+    {
+      "epoch": 0.07964,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.6591471354166667,
+      "learning_rate": 0.0001,
+      "loss": 4.6771,
+      "loss/crossentropy": 2.3253976106643677,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2691802680492401,
+      "step": 3982
+    },
+    {
+      "epoch": 0.07968,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.65885009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.8223,
+      "loss/crossentropy": 2.14141583442688,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2603989467024803,
+      "step": 3984
+    },
+    {
+      "epoch": 0.07972,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.6512196858723959,
+      "learning_rate": 0.0001,
+      "loss": 4.9059,
+      "loss/crossentropy": 2.262465476989746,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.291474387049675,
+      "step": 3986
+    },
+    {
+      "epoch": 0.07976,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.66064453125,
+      "learning_rate": 0.0001,
+      "loss": 4.5626,
+      "loss/crossentropy": 2.1835561990737915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23615659773349762,
+      "step": 3988
+    },
+    {
+      "epoch": 0.0798,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.6716145833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.8482,
+      "loss/crossentropy": 2.020140767097473,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2593151703476906,
+      "step": 3990
+    },
+    {
+      "epoch": 0.07984,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.6518513997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.7986,
+      "loss/crossentropy": 2.277661681175232,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2721617519855499,
+      "step": 3992
+    },
+    {
+      "epoch": 0.07988,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.09846089680989584,
+      "learning_rate": 0.0001,
+      "loss": 4.5113,
+      "loss/crossentropy": 2.1883193254470825,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2610222101211548,
+      "step": 3994
+    },
+    {
+      "epoch": 0.07992,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.01754150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.5987,
+      "loss/crossentropy": 2.152850031852722,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27418845891952515,
+      "step": 3996
+    },
+    {
+      "epoch": 0.07996,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.017513020833333334,
+      "learning_rate": 0.0001,
+      "loss": 4.8214,
+      "loss/crossentropy": 2.3313716650009155,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27529503405094147,
+      "step": 3998
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.020319620768229168,
+      "learning_rate": 0.0001,
+      "loss": 5.086,
+      "loss/crossentropy": 2.250498414039612,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26368021965026855,
+      "step": 4000
+    },
+    {
+      "epoch": 0.08004,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.011188761393229166,
+      "learning_rate": 0.0001,
+      "loss": 4.8005,
+      "loss/crossentropy": 2.322459101676941,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2868216335773468,
+      "step": 4002
+    },
+    {
+      "epoch": 0.08008,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.0142730712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.8693,
+      "loss/crossentropy": 1.9340506792068481,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2316983863711357,
+      "step": 4004
+    },
+    {
+      "epoch": 0.08012,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.0150054931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.7395,
+      "loss/crossentropy": 1.8635645508766174,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22694773972034454,
+      "step": 4006
+    },
+    {
+      "epoch": 0.08016,
+      "grad_norm": 10.375,
+      "grad_norm_var": 3.9615631103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.8916,
+      "loss/crossentropy": 1.9252317547798157,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24893560260534286,
+      "step": 4008
+    },
+    {
+      "epoch": 0.0802,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 3.9093424479166665,
+      "learning_rate": 0.0001,
+      "loss": 5.2636,
+      "loss/crossentropy": 2.1964328289031982,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2567693591117859,
+      "step": 4010
+    },
+    {
+      "epoch": 0.08024,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 3.8960113525390625,
+      "learning_rate": 0.0001,
+      "loss": 4.9054,
+      "loss/crossentropy": 2.296012043952942,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27360107749700546,
+      "step": 4012
+    },
+    {
+      "epoch": 0.08028,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 3.9009724934895833,
+      "learning_rate": 0.0001,
+      "loss": 4.8894,
+      "loss/crossentropy": 2.360015869140625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27541905641555786,
+      "step": 4014
+    },
+    {
+      "epoch": 0.08032,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 3.906591796875,
+      "learning_rate": 0.0001,
+      "loss": 4.8865,
+      "loss/crossentropy": 2.36370050907135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27881547808647156,
+      "step": 4016
+    },
+    {
+      "epoch": 0.08036,
+      "grad_norm": 2.25,
+      "grad_norm_var": 3.9068593343098956,
+      "learning_rate": 0.0001,
+      "loss": 4.6461,
+      "loss/crossentropy": 1.8704780340194702,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2512069493532181,
+      "step": 4018
+    },
+    {
+      "epoch": 0.0804,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 3.9156483968098956,
+      "learning_rate": 0.0001,
+      "loss": 4.733,
+      "loss/crossentropy": 2.1989234685897827,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2755106985569,
+      "step": 4020
+    },
+    {
+      "epoch": 0.08044,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 3.9420237223307293,
+      "learning_rate": 0.0001,
+      "loss": 4.3471,
+      "loss/crossentropy": 1.9905433058738708,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2579897418618202,
+      "step": 4022
+    },
+    {
+      "epoch": 0.08048,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.08837788899739583,
+      "learning_rate": 0.0001,
+      "loss": 4.9025,
+      "loss/crossentropy": 2.270000696182251,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2693728804588318,
+      "step": 4024
+    },
+    {
+      "epoch": 0.08052,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.08504130045572916,
+      "learning_rate": 0.0001,
+      "loss": 4.7569,
+      "loss/crossentropy": 2.178301692008972,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2617819905281067,
+      "step": 4026
+    },
+    {
+      "epoch": 0.08056,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.08346354166666667,
+      "learning_rate": 0.0001,
+      "loss": 4.687,
+      "loss/crossentropy": 2.518654465675354,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29282887279987335,
+      "step": 4028
+    },
+    {
+      "epoch": 0.0806,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.08859049479166667,
+      "learning_rate": 0.0001,
+      "loss": 4.4825,
+      "loss/crossentropy": 1.9181422591209412,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23210199177265167,
+      "step": 4030
+    },
+    {
+      "epoch": 0.08064,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.08816630045572917,
+      "learning_rate": 0.0001,
+      "loss": 4.6407,
+      "loss/crossentropy": 2.343222141265869,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28908614814281464,
+      "step": 4032
+    },
+    {
+      "epoch": 0.08068,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.0302886962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.6879,
+      "loss/crossentropy": 2.0816845893859863,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2571762502193451,
+      "step": 4034
+    },
+    {
+      "epoch": 0.08072,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.029878743489583335,
+      "learning_rate": 0.0001,
+      "loss": 4.3324,
+      "loss/crossentropy": 1.9679544568061829,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2316160574555397,
+      "step": 4036
+    },
+    {
+      "epoch": 0.08076,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.029195149739583332,
+      "learning_rate": 0.0001,
+      "loss": 4.7335,
+      "loss/crossentropy": 2.1553682684898376,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25725623965263367,
+      "step": 4038
+    },
+    {
+      "epoch": 0.0808,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.004964192708333333,
+      "learning_rate": 0.0001,
+      "loss": 4.6892,
+      "loss/crossentropy": 2.0269790291786194,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26298412680625916,
+      "step": 4040
+    },
+    {
+      "epoch": 0.08084,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.0060943603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.6686,
+      "loss/crossentropy": 1.984773874282837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24288517236709595,
+      "step": 4042
+    },
+    {
+      "epoch": 0.08088,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.0051910400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.9282,
+      "loss/crossentropy": 2.178356111049652,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2684163451194763,
+      "step": 4044
+    },
+    {
+      "epoch": 0.08092,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.0053670247395833336,
+      "learning_rate": 0.0001,
+      "loss": 4.8191,
+      "loss/crossentropy": 2.235984683036804,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26377667486667633,
+      "step": 4046
+    },
+    {
+      "epoch": 0.08096,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.005826822916666667,
+      "learning_rate": 0.0001,
+      "loss": 4.7026,
+      "loss/crossentropy": 2.085321545600891,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23403701931238174,
+      "step": 4048
+    },
+    {
+      "epoch": 0.081,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.0052642822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.9932,
+      "loss/crossentropy": 2.419228672981262,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27236658334732056,
+      "step": 4050
+    },
+    {
+      "epoch": 0.08104,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.0054972330729166664,
+      "learning_rate": 0.0001,
+      "loss": 4.6105,
+      "loss/crossentropy": 2.153620958328247,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28071053326129913,
+      "step": 4052
+    },
+    {
+      "epoch": 0.08108,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.012743123372395833,
+      "learning_rate": 0.0001,
+      "loss": 4.8775,
+      "loss/crossentropy": 2.1466477513313293,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27491800487041473,
+      "step": 4054
+    },
+    {
+      "epoch": 0.08112,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.013932291666666667,
+      "learning_rate": 0.0001,
+      "loss": 4.3707,
+      "loss/crossentropy": 2.2020710706710815,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25149518996477127,
+      "step": 4056
+    },
+    {
+      "epoch": 0.08116,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.013623046875,
+      "learning_rate": 0.0001,
+      "loss": 4.8458,
+      "loss/crossentropy": 2.264205574989319,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27612583339214325,
+      "step": 4058
+    },
+    {
+      "epoch": 0.0812,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.016434733072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.8644,
+      "loss/crossentropy": 2.269905209541321,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2692303955554962,
+      "step": 4060
+    },
+    {
+      "epoch": 0.08124,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.018505859375,
+      "learning_rate": 0.0001,
+      "loss": 4.5057,
+      "loss/crossentropy": 1.920631766319275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22550494968891144,
+      "step": 4062
+    },
+    {
+      "epoch": 0.08128,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.019749959309895832,
+      "learning_rate": 0.0001,
+      "loss": 5.0796,
+      "loss/crossentropy": 2.307617664337158,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22719035297632217,
+      "step": 4064
+    },
+    {
+      "epoch": 0.08132,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.022102864583333333,
+      "learning_rate": 0.0001,
+      "loss": 4.6167,
+      "loss/crossentropy": 2.113444685935974,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.240354023873806,
+      "step": 4066
+    },
+    {
+      "epoch": 0.08136,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.02232666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.9152,
+      "loss/crossentropy": 2.4516230821609497,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27590544521808624,
+      "step": 4068
+    },
+    {
+      "epoch": 0.0814,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.08772379557291667,
+      "learning_rate": 0.0001,
+      "loss": 4.5976,
+      "loss/crossentropy": 1.8287339806556702,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22503511607646942,
+      "step": 4070
+    },
+    {
+      "epoch": 0.08144,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.08479715983072916,
+      "learning_rate": 0.0001,
+      "loss": 5.1623,
+      "loss/crossentropy": 2.3468997478485107,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2773839682340622,
+      "step": 4072
+    },
+    {
+      "epoch": 0.08148,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.08782145182291666,
+      "learning_rate": 0.0001,
+      "loss": 4.5413,
+      "loss/crossentropy": 2.1307512521743774,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24247504770755768,
+      "step": 4074
+    },
+    {
+      "epoch": 0.08152,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.08982645670572917,
+      "learning_rate": 0.0001,
+      "loss": 4.7447,
+      "loss/crossentropy": 2.248755097389221,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2696859538555145,
+      "step": 4076
+    },
+    {
+      "epoch": 0.08156,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.09045817057291666,
+      "learning_rate": 0.0001,
+      "loss": 4.2948,
+      "loss/crossentropy": 2.0233980417251587,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2370016872882843,
+      "step": 4078
+    },
+    {
+      "epoch": 0.0816,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0923828125,
+      "learning_rate": 0.0001,
+      "loss": 4.432,
+      "loss/crossentropy": 1.9536627531051636,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23242933303117752,
+      "step": 4080
+    },
+    {
+      "epoch": 0.08164,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.0889801025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.5037,
+      "loss/crossentropy": 1.9631904363632202,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2337196245789528,
+      "step": 4082
+    },
+    {
+      "epoch": 0.08168,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.08935139973958334,
+      "learning_rate": 0.0001,
+      "loss": 4.7406,
+      "loss/crossentropy": 2.193789482116699,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2687358558177948,
+      "step": 4084
+    },
+    {
+      "epoch": 0.08172,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.020210774739583333,
+      "learning_rate": 0.0001,
+      "loss": 4.6454,
+      "loss/crossentropy": 2.308240056037903,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2679348289966583,
+      "step": 4086
+    },
+    {
+      "epoch": 0.08176,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.021903483072916667,
+      "learning_rate": 0.0001,
+      "loss": 4.9126,
+      "loss/crossentropy": 2.343047261238098,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3056950569152832,
+      "step": 4088
+    },
+    {
+      "epoch": 0.0818,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.022526041666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.7315,
+      "loss/crossentropy": 1.9583085179328918,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23911744356155396,
+      "step": 4090
+    },
+    {
+      "epoch": 0.08184,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.021198527018229166,
+      "learning_rate": 0.0001,
+      "loss": 4.6559,
+      "loss/crossentropy": 2.341569185256958,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26053962111473083,
+      "step": 4092
+    },
+    {
+      "epoch": 0.08188,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.034601847330729164,
+      "learning_rate": 0.0001,
+      "loss": 4.5773,
+      "loss/crossentropy": 2.068669080734253,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25753986835479736,
+      "step": 4094
+    },
+    {
+      "epoch": 0.08192,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.04309488932291667,
+      "learning_rate": 0.0001,
+      "loss": 5.0253,
+      "loss/crossentropy": 2.1461241841316223,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2628704681992531,
+      "step": 4096
+    },
+    {
+      "epoch": 0.08196,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.04810791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.5587,
+      "loss/crossentropy": 2.0718055963516235,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24923217296600342,
+      "step": 4098
+    },
+    {
+      "epoch": 0.082,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.0503082275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.379,
+      "loss/crossentropy": 1.9812004566192627,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2316955253481865,
+      "step": 4100
+    },
+    {
+      "epoch": 0.08204,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.0465240478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.7909,
+      "loss/crossentropy": 2.2669100761413574,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2806926518678665,
+      "step": 4102
+    },
+    {
+      "epoch": 0.08208,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.043196614583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.7502,
+      "loss/crossentropy": 2.0620261430740356,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2610047310590744,
+      "step": 4104
+    },
+    {
+      "epoch": 0.08212,
+      "grad_norm": 3.046875,
+      "grad_norm_var": 0.06297098795572917,
+      "learning_rate": 0.0001,
+      "loss": 4.6672,
+      "loss/crossentropy": 2.249971866607666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2783341556787491,
+      "step": 4106
+    },
+    {
+      "epoch": 0.08216,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.06320699055989583,
+      "learning_rate": 0.0001,
+      "loss": 4.834,
+      "loss/crossentropy": 2.1064823865890503,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27534550428390503,
+      "step": 4108
+    },
+    {
+      "epoch": 0.0822,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.05406901041666667,
+      "learning_rate": 0.0001,
+      "loss": 4.3944,
+      "loss/crossentropy": 1.886509656906128,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2315160632133484,
+      "step": 4110
+    },
+    {
+      "epoch": 0.08224,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.039290364583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.2969,
+      "loss/crossentropy": 1.6429635286331177,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20593100041151047,
+      "step": 4112
+    },
+    {
+      "epoch": 0.08228,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.0370025634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4581,
+      "loss/crossentropy": 2.3236618041992188,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2674623131752014,
+      "step": 4114
+    },
+    {
+      "epoch": 0.08232,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.03658447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.9734,
+      "loss/crossentropy": 2.1479567885398865,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25003983825445175,
+      "step": 4116
+    },
+    {
+      "epoch": 0.08236,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.03611551920572917,
+      "learning_rate": 0.0001,
+      "loss": 5.0477,
+      "loss/crossentropy": 2.140569031238556,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24322029948234558,
+      "step": 4118
+    },
+    {
+      "epoch": 0.0824,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.03762613932291667,
+      "learning_rate": 0.0001,
+      "loss": 4.6061,
+      "loss/crossentropy": 2.126375436782837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27629759907722473,
+      "step": 4120
+    },
+    {
+      "epoch": 0.08244,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.015843709309895832,
+      "learning_rate": 0.0001,
+      "loss": 4.9143,
+      "loss/crossentropy": 2.3699214458465576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2692546397447586,
+      "step": 4122
+    },
+    {
+      "epoch": 0.08248,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.010904947916666666,
+      "learning_rate": 0.0001,
+      "loss": 4.5672,
+      "loss/crossentropy": 2.013331353664398,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24442073702812195,
+      "step": 4124
+    },
+    {
+      "epoch": 0.08252,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.010152180989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.4841,
+      "loss/crossentropy": 2.1869460344314575,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.250150203704834,
+      "step": 4126
+    },
+    {
+      "epoch": 0.08256,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0127838134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4564,
+      "loss/crossentropy": 2.2725884914398193,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2810261696577072,
+      "step": 4128
+    },
+    {
+      "epoch": 0.0826,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.013688151041666667,
+      "learning_rate": 0.0001,
+      "loss": 4.7311,
+      "loss/crossentropy": 1.9190022945404053,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2693525403738022,
+      "step": 4130
+    },
+    {
+      "epoch": 0.08264,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.016112263997395834,
+      "learning_rate": 0.0001,
+      "loss": 4.7967,
+      "loss/crossentropy": 2.5477795600891113,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27040669322013855,
+      "step": 4132
+    },
+    {
+      "epoch": 0.08268,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.0113677978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.7617,
+      "loss/crossentropy": 2.231198728084564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28101250529289246,
+      "step": 4134
+    },
+    {
+      "epoch": 0.08272,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.011693318684895834,
+      "learning_rate": 0.0001,
+      "loss": 4.6334,
+      "loss/crossentropy": 2.17776882648468,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24743208289146423,
+      "step": 4136
+    },
+    {
+      "epoch": 0.08276,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.011799112955729166,
+      "learning_rate": 0.0001,
+      "loss": 5.0233,
+      "loss/crossentropy": 2.418373703956604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2954525351524353,
+      "step": 4138
+    },
+    {
+      "epoch": 0.0828,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.010969034830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.5564,
+      "loss/crossentropy": 2.054605543613434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23863950371742249,
+      "step": 4140
+    },
+    {
+      "epoch": 0.08284,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.011872355143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.8983,
+      "loss/crossentropy": 2.054013967514038,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29851172864437103,
+      "step": 4142
+    },
+    {
+      "epoch": 0.08288,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.008610026041666666,
+      "learning_rate": 0.0001,
+      "loss": 4.7425,
+      "loss/crossentropy": 2.193961024284363,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2563214898109436,
+      "step": 4144
+    },
+    {
+      "epoch": 0.08292,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.008382161458333334,
+      "learning_rate": 0.0001,
+      "loss": 4.7995,
+      "loss/crossentropy": 2.460008382797241,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26705074310302734,
+      "step": 4146
+    },
+    {
+      "epoch": 0.08296,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.005060831705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.894,
+      "loss/crossentropy": 2.508321523666382,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26217761635780334,
+      "step": 4148
+    },
+    {
+      "epoch": 0.083,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.0065582275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.6103,
+      "loss/crossentropy": 1.8445284366607666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23396535962820053,
+      "step": 4150
+    },
+    {
+      "epoch": 0.08304,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.005952962239583333,
+      "learning_rate": 0.0001,
+      "loss": 4.8048,
+      "loss/crossentropy": 2.433600902557373,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2825329154729843,
+      "step": 4152
+    },
+    {
+      "epoch": 0.08308,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.0053212483723958336,
+      "learning_rate": 0.0001,
+      "loss": 4.8632,
+      "loss/crossentropy": 2.386221170425415,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2941686511039734,
+      "step": 4154
+    },
+    {
+      "epoch": 0.08312,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.0052154541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.7486,
+      "loss/crossentropy": 1.9578949809074402,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24851053953170776,
+      "step": 4156
+    },
+    {
+      "epoch": 0.08316,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.004150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.6443,
+      "loss/crossentropy": 2.0034408569335938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25524984300136566,
+      "step": 4158
+    },
+    {
+      "epoch": 0.0832,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.0069488525390625,
+      "learning_rate": 0.0001,
+      "loss": 4.6856,
+      "loss/crossentropy": 2.3342589139938354,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31023962795734406,
+      "step": 4160
+    },
+    {
+      "epoch": 0.08324,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.042867024739583336,
+      "learning_rate": 0.0001,
+      "loss": 4.6762,
+      "loss/crossentropy": 2.3941839933395386,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2641746401786804,
+      "step": 4162
+    },
+    {
+      "epoch": 0.08328,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.04168294270833333,
+      "learning_rate": 0.0001,
+      "loss": 4.7149,
+      "loss/crossentropy": 2.371219038963318,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24197939038276672,
+      "step": 4164
+    },
+    {
+      "epoch": 0.08332,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.04442952473958333,
+      "learning_rate": 0.0001,
+      "loss": 4.7949,
+      "loss/crossentropy": 2.133378028869629,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25449906289577484,
+      "step": 4166
+    },
+    {
+      "epoch": 0.08336,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.04496968587239583,
+      "learning_rate": 0.0001,
+      "loss": 4.5974,
+      "loss/crossentropy": 1.7460412979125977,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2157151699066162,
+      "step": 4168
+    },
+    {
+      "epoch": 0.0834,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.046873982747395834,
+      "learning_rate": 0.0001,
+      "loss": 4.7234,
+      "loss/crossentropy": 2.215083122253418,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2823774367570877,
+      "step": 4170
+    },
+    {
+      "epoch": 0.08344,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.04820048014322917,
+      "learning_rate": 0.0001,
+      "loss": 4.5189,
+      "loss/crossentropy": 2.0528377890586853,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2551003098487854,
+      "step": 4172
+    },
+    {
+      "epoch": 0.08348,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.05054423014322917,
+      "learning_rate": 0.0001,
+      "loss": 4.4008,
+      "loss/crossentropy": 1.7953855395317078,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23185917735099792,
+      "step": 4174
+    },
+    {
+      "epoch": 0.08352,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.058934529622395836,
+      "learning_rate": 0.0001,
+      "loss": 4.4879,
+      "loss/crossentropy": 2.0794734954833984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23386041820049286,
+      "step": 4176
+    },
+    {
+      "epoch": 0.08356,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.031086222330729166,
+      "learning_rate": 0.0001,
+      "loss": 4.3802,
+      "loss/crossentropy": 2.1685845851898193,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26074862480163574,
+      "step": 4178
+    },
+    {
+      "epoch": 0.0836,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.031412760416666664,
+      "learning_rate": 0.0001,
+      "loss": 4.5507,
+      "loss/crossentropy": 2.1495825052261353,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2689145505428314,
+      "step": 4180
+    },
+    {
+      "epoch": 0.08364,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.024967447916666666,
+      "learning_rate": 0.0001,
+      "loss": 4.5258,
+      "loss/crossentropy": 2.043331503868103,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2498578578233719,
+      "step": 4182
+    },
+    {
+      "epoch": 0.08368,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.07787984212239583,
+      "learning_rate": 0.0001,
+      "loss": 5.0198,
+      "loss/crossentropy": 2.04026997089386,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2873340845108032,
+      "step": 4184
+    },
+    {
+      "epoch": 0.08372,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.09128316243489583,
+      "learning_rate": 0.0001,
+      "loss": 4.9537,
+      "loss/crossentropy": 2.4653968811035156,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2911294251680374,
+      "step": 4186
+    },
+    {
+      "epoch": 0.08376,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.09215494791666666,
+      "learning_rate": 0.0001,
+      "loss": 4.6589,
+      "loss/crossentropy": 2.2960848808288574,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23984040319919586,
+      "step": 4188
+    },
+    {
+      "epoch": 0.0838,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.09599202473958333,
+      "learning_rate": 0.0001,
+      "loss": 4.4324,
+      "loss/crossentropy": 2.011807084083557,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2523125037550926,
+      "step": 4190
+    },
+    {
+      "epoch": 0.08384,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.0886871337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.8437,
+      "loss/crossentropy": 2.0016889572143555,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2291206791996956,
+      "step": 4192
+    },
+    {
+      "epoch": 0.08388,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.09378255208333333,
+      "learning_rate": 0.0001,
+      "loss": 4.3604,
+      "loss/crossentropy": 1.97197824716568,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2595779076218605,
+      "step": 4194
+    },
+    {
+      "epoch": 0.08392,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.09763895670572917,
+      "learning_rate": 0.0001,
+      "loss": 4.5823,
+      "loss/crossentropy": 2.2910103797912598,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24870596826076508,
+      "step": 4196
+    },
+    {
+      "epoch": 0.08396,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.1001617431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.562,
+      "loss/crossentropy": 2.1453208923339844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25001347810029984,
+      "step": 4198
+    },
+    {
+      "epoch": 0.084,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.0398834228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.8835,
+      "loss/crossentropy": 2.1935043334960938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26807525753974915,
+      "step": 4200
+    },
+    {
+      "epoch": 0.08404,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.5912,
+      "loss/crossentropy": 2.039341926574707,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2460380420088768,
+      "step": 4202
+    },
+    {
+      "epoch": 0.08408,
+      "grad_norm": 3.09375,
+      "grad_norm_var": 0.0478179931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.8243,
+      "loss/crossentropy": 2.4660122394561768,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3042101263999939,
+      "step": 4204
+    },
+    {
+      "epoch": 0.08412,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.08088785807291667,
+      "learning_rate": 0.0001,
+      "loss": 4.8635,
+      "loss/crossentropy": 1.9346272349357605,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25924334675073624,
+      "step": 4206
+    },
+    {
+      "epoch": 0.08416,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.08311258951822917,
+      "learning_rate": 0.0001,
+      "loss": 4.5152,
+      "loss/crossentropy": 2.0120063424110413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24327433109283447,
+      "step": 4208
+    },
+    {
+      "epoch": 0.0842,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.0852935791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.6148,
+      "loss/crossentropy": 2.2359931468963623,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25374244898557663,
+      "step": 4210
+    },
+    {
+      "epoch": 0.08424,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.08185221354166666,
+      "learning_rate": 0.0001,
+      "loss": 4.5751,
+      "loss/crossentropy": 2.0038134455680847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22315364331007004,
+      "step": 4212
+    },
+    {
+      "epoch": 0.08428,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.08567301432291667,
+      "learning_rate": 0.0001,
+      "loss": 4.721,
+      "loss/crossentropy": 2.2041471004486084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2655494213104248,
+      "step": 4214
+    },
+    {
+      "epoch": 0.08432,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.08399149576822916,
+      "learning_rate": 0.0001,
+      "loss": 4.8091,
+      "loss/crossentropy": 2.344551682472229,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2743126451969147,
+      "step": 4216
+    },
+    {
+      "epoch": 0.08436,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.08025614420572917,
+      "learning_rate": 0.0001,
+      "loss": 4.7162,
+      "loss/crossentropy": 1.9694250226020813,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24425261467695236,
+      "step": 4218
+    },
+    {
+      "epoch": 0.0844,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.05671284993489583,
+      "learning_rate": 0.0001,
+      "loss": 4.8526,
+      "loss/crossentropy": 2.164921760559082,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.272469699382782,
+      "step": 4220
+    },
+    {
+      "epoch": 0.08444,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.02642822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.5513,
+      "loss/crossentropy": 1.944575309753418,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23848393559455872,
+      "step": 4222
+    },
+    {
+      "epoch": 0.08448,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.025386555989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.7416,
+      "loss/crossentropy": 2.278227686882019,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2558315545320511,
+      "step": 4224
+    },
+    {
+      "epoch": 0.08452,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.01802978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.7318,
+      "loss/crossentropy": 2.035117268562317,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2585935667157173,
+      "step": 4226
+    },
+    {
+      "epoch": 0.08456,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.020466105143229166,
+      "learning_rate": 0.0001,
+      "loss": 4.5674,
+      "loss/crossentropy": 2.0172035694122314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23832131922245026,
+      "step": 4228
+    },
+    {
+      "epoch": 0.0846,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.011165364583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.8113,
+      "loss/crossentropy": 2.0574535727500916,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23829226195812225,
+      "step": 4230
+    },
+    {
+      "epoch": 0.08464,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.010773722330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.6776,
+      "loss/crossentropy": 2.5003366470336914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27905476093292236,
+      "step": 4232
+    },
+    {
+      "epoch": 0.08468,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.012043253580729166,
+      "learning_rate": 0.0001,
+      "loss": 4.9137,
+      "loss/crossentropy": 2.207367777824402,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27426937222480774,
+      "step": 4234
+    },
+    {
+      "epoch": 0.08472,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.01627197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.716,
+      "loss/crossentropy": 2.240189790725708,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26208513230085373,
+      "step": 4236
+    },
+    {
+      "epoch": 0.08476,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.016600545247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4569,
+      "loss/crossentropy": 2.1357412338256836,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24108420312404633,
+      "step": 4238
+    },
+    {
+      "epoch": 0.0848,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.013004557291666666,
+      "learning_rate": 0.0001,
+      "loss": 4.7249,
+      "loss/crossentropy": 2.1073816418647766,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25213342159986496,
+      "step": 4240
+    },
+    {
+      "epoch": 0.08484,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.013102213541666666,
+      "learning_rate": 0.0001,
+      "loss": 4.9558,
+      "loss/crossentropy": 2.158124566078186,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2505848854780197,
+      "step": 4242
+    },
+    {
+      "epoch": 0.08488,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.011555989583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.7743,
+      "loss/crossentropy": 2.253539562225342,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2675466388463974,
+      "step": 4244
+    },
+    {
+      "epoch": 0.08492,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.01396484375,
+      "learning_rate": 0.0001,
+      "loss": 4.272,
+      "loss/crossentropy": 1.7170023918151855,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20157357305288315,
+      "step": 4246
+    },
+    {
+      "epoch": 0.08496,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.0188385009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4479,
+      "loss/crossentropy": 2.082640767097473,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2600134015083313,
+      "step": 4248
+    },
+    {
+      "epoch": 0.085,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.0165679931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4176,
+      "loss/crossentropy": 2.044301390647888,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23220707476139069,
+      "step": 4250
+    },
+    {
+      "epoch": 0.08504,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.0168121337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.648,
+      "loss/crossentropy": 2.293405532836914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28027066588401794,
+      "step": 4252
+    },
+    {
+      "epoch": 0.08508,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.016649373372395835,
+      "learning_rate": 0.0001,
+      "loss": 4.55,
+      "loss/crossentropy": 2.2604206800460815,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25545646995306015,
+      "step": 4254
+    },
+    {
+      "epoch": 0.08512,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.023949178059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.6258,
+      "loss/crossentropy": 2.118361234664917,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24799348413944244,
+      "step": 4256
+    },
+    {
+      "epoch": 0.08516,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.022847493489583332,
+      "learning_rate": 0.0001,
+      "loss": 4.6751,
+      "loss/crossentropy": 1.9369969964027405,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24676478654146194,
+      "step": 4258
+    },
+    {
+      "epoch": 0.0852,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.021728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5197,
+      "loss/crossentropy": 2.075170874595642,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21742676943540573,
+      "step": 4260
+    },
+    {
+      "epoch": 0.08524,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.018973795572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.4112,
+      "loss/crossentropy": 2.056099236011505,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2490842342376709,
+      "step": 4262
+    },
+    {
+      "epoch": 0.08528,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.014436848958333333,
+      "learning_rate": 0.0001,
+      "loss": 4.6169,
+      "loss/crossentropy": 2.2279993891716003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24964337795972824,
+      "step": 4264
+    },
+    {
+      "epoch": 0.08532,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.011393229166666666,
+      "learning_rate": 0.0001,
+      "loss": 4.6686,
+      "loss/crossentropy": 2.1645933389663696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24576786905527115,
+      "step": 4266
+    },
+    {
+      "epoch": 0.08536,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.010887654622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.4458,
+      "loss/crossentropy": 1.9033920764923096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23393237590789795,
+      "step": 4268
+    },
+    {
+      "epoch": 0.0854,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.011865234375,
+      "learning_rate": 0.0001,
+      "loss": 4.4022,
+      "loss/crossentropy": 2.153634190559387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2496839165687561,
+      "step": 4270
+    },
+    {
+      "epoch": 0.08544,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.0045206705729166664,
+      "learning_rate": 0.0001,
+      "loss": 4.4781,
+      "loss/crossentropy": 1.9188589453697205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23544569313526154,
+      "step": 4272
+    },
+    {
+      "epoch": 0.08548,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.004264322916666666,
+      "learning_rate": 0.0001,
+      "loss": 4.704,
+      "loss/crossentropy": 2.4337977170944214,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2951700836420059,
+      "step": 4274
+    },
+    {
+      "epoch": 0.08552,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.003902180989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.7051,
+      "loss/crossentropy": 1.9108383059501648,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24459081888198853,
+      "step": 4276
+    },
+    {
+      "epoch": 0.08556,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.003123982747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3751,
+      "loss/crossentropy": 1.6632736921310425,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22495487332344055,
+      "step": 4278
+    },
+    {
+      "epoch": 0.0856,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.021312459309895834,
+      "learning_rate": 0.0001,
+      "loss": 4.8144,
+      "loss/crossentropy": 2.519997477531433,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2657178193330765,
+      "step": 4280
+    },
+    {
+      "epoch": 0.08564,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.025275675455729167,
+      "learning_rate": 0.0001,
+      "loss": 4.6837,
+      "loss/crossentropy": 2.150822162628174,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30192580074071884,
+      "step": 4282
+    },
+    {
+      "epoch": 0.08568,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.023763020833333332,
+      "learning_rate": 0.0001,
+      "loss": 4.7411,
+      "loss/crossentropy": 1.9970062971115112,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24695321917533875,
+      "step": 4284
+    },
+    {
+      "epoch": 0.08572,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.024372355143229166,
+      "learning_rate": 0.0001,
+      "loss": 4.5361,
+      "loss/crossentropy": 2.3136098384857178,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25082121044397354,
+      "step": 4286
+    },
+    {
+      "epoch": 0.08576,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.025992838541666667,
+      "learning_rate": 0.0001,
+      "loss": 4.9171,
+      "loss/crossentropy": 2.112035870552063,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27019062638282776,
+      "step": 4288
+    },
+    {
+      "epoch": 0.0858,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.025992838541666667,
+      "learning_rate": 0.0001,
+      "loss": 4.4985,
+      "loss/crossentropy": 2.068653643131256,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24297921359539032,
+      "step": 4290
+    },
+    {
+      "epoch": 0.08584,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.025520833333333333,
+      "learning_rate": 0.0001,
+      "loss": 4.5182,
+      "loss/crossentropy": 1.9013578295707703,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23058265447616577,
+      "step": 4292
+    },
+    {
+      "epoch": 0.08588,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0294830322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.6825,
+      "loss/crossentropy": 2.149984359741211,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25223904848098755,
+      "step": 4294
+    },
+    {
+      "epoch": 0.08592,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.014469401041666666,
+      "learning_rate": 0.0001,
+      "loss": 4.4109,
+      "loss/crossentropy": 1.894010066986084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2323223054409027,
+      "step": 4296
+    },
+    {
+      "epoch": 0.08596,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.010123697916666667,
+      "learning_rate": 0.0001,
+      "loss": 4.7653,
+      "loss/crossentropy": 2.3351621627807617,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27194930613040924,
+      "step": 4298
+    },
+    {
+      "epoch": 0.086,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.0097320556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.741,
+      "loss/crossentropy": 2.224352180957794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24884501099586487,
+      "step": 4300
+    },
+    {
+      "epoch": 0.08604,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.008854166666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.6592,
+      "loss/crossentropy": 1.908318042755127,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24747492372989655,
+      "step": 4302
+    },
+    {
+      "epoch": 0.08608,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.00758056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.8566,
+      "loss/crossentropy": 2.1990396976470947,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27217794954776764,
+      "step": 4304
+    },
+    {
+      "epoch": 0.08612,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.0098297119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.6432,
+      "loss/crossentropy": 2.3146010637283325,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26875850558280945,
+      "step": 4306
+    },
+    {
+      "epoch": 0.08616,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.015208943684895834,
+      "learning_rate": 0.0001,
+      "loss": 4.8254,
+      "loss/crossentropy": 2.2507941722869873,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27052539587020874,
+      "step": 4308
+    },
+    {
+      "epoch": 0.0862,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.013199869791666667,
+      "learning_rate": 0.0001,
+      "loss": 4.4067,
+      "loss/crossentropy": 1.9077125787734985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22987178713083267,
+      "step": 4310
+    },
+    {
+      "epoch": 0.08624,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.01353759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4822,
+      "loss/crossentropy": 1.951395332813263,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2392890453338623,
+      "step": 4312
+    },
+    {
+      "epoch": 0.08628,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.033854166666666664,
+      "learning_rate": 0.0001,
+      "loss": 4.5371,
+      "loss/crossentropy": 1.9426860213279724,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24374966323375702,
+      "step": 4314
+    },
+    {
+      "epoch": 0.08632,
+      "grad_norm": 3.390625,
+      "grad_norm_var": 0.09062398274739583,
+      "learning_rate": 0.0001,
+      "loss": 5.253,
+      "loss/crossentropy": 2.2508288621902466,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3862452507019043,
+      "step": 4316
+    },
+    {
+      "epoch": 0.08636,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.09058329264322916,
+      "learning_rate": 0.0001,
+      "loss": 4.5288,
+      "loss/crossentropy": 2.1161463260650635,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2557579278945923,
+      "step": 4318
+    },
+    {
+      "epoch": 0.0864,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.09374593098958334,
+      "learning_rate": 0.0001,
+      "loss": 5.1146,
+      "loss/crossentropy": 2.2570544481277466,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31447017192840576,
+      "step": 4320
+    },
+    {
+      "epoch": 0.08644,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.08740234375,
+      "learning_rate": 0.0001,
+      "loss": 4.9724,
+      "loss/crossentropy": 2.3211100101470947,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26107798516750336,
+      "step": 4322
+    },
+    {
+      "epoch": 0.08648,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.09231669108072917,
+      "learning_rate": 0.0001,
+      "loss": 4.5236,
+      "loss/crossentropy": 2.1451058387756348,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23633568733930588,
+      "step": 4324
+    },
+    {
+      "epoch": 0.08652,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.09463602701822917,
+      "learning_rate": 0.0001,
+      "loss": 4.5828,
+      "loss/crossentropy": 1.9880141615867615,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23688867688179016,
+      "step": 4326
+    },
+    {
+      "epoch": 0.08656,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.09724934895833333,
+      "learning_rate": 0.0001,
+      "loss": 4.7098,
+      "loss/crossentropy": 2.021056890487671,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2543798238039017,
+      "step": 4328
+    },
+    {
+      "epoch": 0.0866,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.0814453125,
+      "learning_rate": 0.0001,
+      "loss": 4.6439,
+      "loss/crossentropy": 2.1323755979537964,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27410852909088135,
+      "step": 4330
+    },
+    {
+      "epoch": 0.08664,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.014940388997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.6945,
+      "loss/crossentropy": 1.9674875736236572,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24818265438079834,
+      "step": 4332
+    },
+    {
+      "epoch": 0.08668,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.01490478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1911,
+      "loss/crossentropy": 2.0466583967208862,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2567761391401291,
+      "step": 4334
+    },
+    {
+      "epoch": 0.08672,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.004423014322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.5937,
+      "loss/crossentropy": 2.138857126235962,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26430967450141907,
+      "step": 4336
+    },
+    {
+      "epoch": 0.08676,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.003902180989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.7168,
+      "loss/crossentropy": 2.164841413497925,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24389629065990448,
+      "step": 4338
+    },
+    {
+      "epoch": 0.0868,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.005322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.6017,
+      "loss/crossentropy": 2.2220189571380615,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24361558258533478,
+      "step": 4340
+    },
+    {
+      "epoch": 0.08684,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.004792277018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.3088,
+      "loss/crossentropy": 1.7106285095214844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21558403968811035,
+      "step": 4342
+    },
+    {
+      "epoch": 0.08688,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0053670247395833336,
+      "learning_rate": 0.0001,
+      "loss": 4.1647,
+      "loss/crossentropy": 1.9200173020362854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2502904310822487,
+      "step": 4344
+    },
+    {
+      "epoch": 0.08692,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.020796712239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.6817,
+      "loss/crossentropy": 1.883722960948944,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24818243086338043,
+      "step": 4346
+    },
+    {
+      "epoch": 0.08696,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.026146443684895833,
+      "learning_rate": 0.0001,
+      "loss": 4.7509,
+      "loss/crossentropy": 2.2069878578186035,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2518744319677353,
+      "step": 4348
+    },
+    {
+      "epoch": 0.087,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.026656087239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.7235,
+      "loss/crossentropy": 2.2158325910568237,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2551049590110779,
+      "step": 4350
+    },
+    {
+      "epoch": 0.08704,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.02730712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.8406,
+      "loss/crossentropy": 2.0580105781555176,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2742728739976883,
+      "step": 4352
+    },
+    {
+      "epoch": 0.08708,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.02906494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.5267,
+      "loss/crossentropy": 2.190276265144348,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.274506613612175,
+      "step": 4354
+    },
+    {
+      "epoch": 0.08712,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.027242024739583332,
+      "learning_rate": 0.0001,
+      "loss": 4.7499,
+      "loss/crossentropy": 2.2595328092575073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25295622646808624,
+      "step": 4356
+    },
+    {
+      "epoch": 0.08716,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.027391560872395835,
+      "learning_rate": 0.0001,
+      "loss": 5.1247,
+      "loss/crossentropy": 2.322342872619629,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2490948587656021,
+      "step": 4358
+    },
+    {
+      "epoch": 0.0872,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.024665323893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.7045,
+      "loss/crossentropy": 2.108223795890808,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25404803454875946,
+      "step": 4360
+    },
+    {
+      "epoch": 0.08724,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.013825480143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.5332,
+      "loss/crossentropy": 2.1363136768341064,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2679157853126526,
+      "step": 4362
+    },
+    {
+      "epoch": 0.08728,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.0076171875,
+      "learning_rate": 0.0001,
+      "loss": 4.6418,
+      "loss/crossentropy": 2.3207738399505615,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2725762128829956,
+      "step": 4364
+    },
+    {
+      "epoch": 0.08732,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.004813639322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.7431,
+      "loss/crossentropy": 2.3179128170013428,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27275949716567993,
+      "step": 4366
+    },
+    {
+      "epoch": 0.08736,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.005182902018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.7355,
+      "loss/crossentropy": 2.2130206823349,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2787477523088455,
+      "step": 4368
+    },
+    {
+      "epoch": 0.0874,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.004548136393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.6193,
+      "loss/crossentropy": 2.3350926637649536,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26483266800642014,
+      "step": 4370
+    },
+    {
+      "epoch": 0.08744,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.005736287434895833,
+      "learning_rate": 0.0001,
+      "loss": 4.7329,
+      "loss/crossentropy": 1.9162638187408447,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23748356848955154,
+      "step": 4372
+    },
+    {
+      "epoch": 0.08748,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.006322224934895833,
+      "learning_rate": 0.0001,
+      "loss": 4.7045,
+      "loss/crossentropy": 2.2708429098129272,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2709425985813141,
+      "step": 4374
+    },
+    {
+      "epoch": 0.08752,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.011604817708333333,
+      "learning_rate": 0.0001,
+      "loss": 4.3481,
+      "loss/crossentropy": 1.7216318845748901,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22498781234025955,
+      "step": 4376
+    },
+    {
+      "epoch": 0.08756,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.011937459309895834,
+      "learning_rate": 0.0001,
+      "loss": 4.4261,
+      "loss/crossentropy": 2.144331693649292,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25400668382644653,
+      "step": 4378
+    },
+    {
+      "epoch": 0.0876,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.011750284830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.5617,
+      "loss/crossentropy": 2.305369734764099,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29137127101421356,
+      "step": 4380
+    },
+    {
+      "epoch": 0.08764,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.018115234375,
+      "learning_rate": 0.0001,
+      "loss": 4.6861,
+      "loss/crossentropy": 2.1156765818595886,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24203064292669296,
+      "step": 4382
+    },
+    {
+      "epoch": 0.08768,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.016630045572916665,
+      "learning_rate": 0.0001,
+      "loss": 4.4544,
+      "loss/crossentropy": 1.9081769585609436,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22396781295537949,
+      "step": 4384
+    },
+    {
+      "epoch": 0.08772,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.017769368489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.704,
+      "loss/crossentropy": 2.0938609838485718,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27330371737480164,
+      "step": 4386
+    },
+    {
+      "epoch": 0.08776,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.017748006184895835,
+      "learning_rate": 0.0001,
+      "loss": 4.1179,
+      "loss/crossentropy": 1.9685207605361938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2287725731730461,
+      "step": 4388
+    },
+    {
+      "epoch": 0.0878,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.016109212239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.5233,
+      "loss/crossentropy": 1.7536925673484802,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2345883920788765,
+      "step": 4390
+    },
+    {
+      "epoch": 0.08784,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.029150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3897,
+      "loss/crossentropy": 2.145567834377289,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26126645505428314,
+      "step": 4392
+    },
+    {
+      "epoch": 0.08788,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.028348795572916665,
+      "learning_rate": 0.0001,
+      "loss": 4.4677,
+      "loss/crossentropy": 2.2211687564849854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3332909345626831,
+      "step": 4394
+    },
+    {
+      "epoch": 0.08792,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.028539021809895832,
+      "learning_rate": 0.0001,
+      "loss": 4.5545,
+      "loss/crossentropy": 1.919084072113037,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2480143904685974,
+      "step": 4396
+    },
+    {
+      "epoch": 0.08796,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.025121053059895832,
+      "learning_rate": 0.0001,
+      "loss": 4.6408,
+      "loss/crossentropy": 2.1769548654556274,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.269208699464798,
+      "step": 4398
+    },
+    {
+      "epoch": 0.088,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.028706868489583332,
+      "learning_rate": 0.0001,
+      "loss": 4.7372,
+      "loss/crossentropy": 1.8480086922645569,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2432136833667755,
+      "step": 4400
+    },
+    {
+      "epoch": 0.08804,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.026838175455729165,
+      "learning_rate": 0.0001,
+      "loss": 4.7303,
+      "loss/crossentropy": 2.1948903799057007,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28328536450862885,
+      "step": 4402
+    },
+    {
+      "epoch": 0.08808,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.022554524739583335,
+      "learning_rate": 0.0001,
+      "loss": 4.6929,
+      "loss/crossentropy": 2.163570761680603,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.282375693321228,
+      "step": 4404
+    },
+    {
+      "epoch": 0.08812,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.022359212239583332,
+      "learning_rate": 0.0001,
+      "loss": 4.6882,
+      "loss/crossentropy": 2.3737696409225464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26703887432813644,
+      "step": 4406
+    },
+    {
+      "epoch": 0.08816,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.015192667643229166,
+      "learning_rate": 0.0001,
+      "loss": 4.6959,
+      "loss/crossentropy": 2.2449779510498047,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2744368612766266,
+      "step": 4408
+    },
+    {
+      "epoch": 0.0882,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.015348307291666667,
+      "learning_rate": 0.0001,
+      "loss": 4.6602,
+      "loss/crossentropy": 2.1167399287223816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28688907623291016,
+      "step": 4410
+    },
+    {
+      "epoch": 0.08824,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.017118326822916665,
+      "learning_rate": 0.0001,
+      "loss": 4.3777,
+      "loss/crossentropy": 2.2249929904937744,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24534232914447784,
+      "step": 4412
+    },
+    {
+      "epoch": 0.08828,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.019261678059895832,
+      "learning_rate": 0.0001,
+      "loss": 4.7013,
+      "loss/crossentropy": 2.172566771507263,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2783561646938324,
+      "step": 4414
+    },
+    {
+      "epoch": 0.08832,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.018684895833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.3536,
+      "loss/crossentropy": 2.0709031224250793,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2419627606868744,
+      "step": 4416
+    },
+    {
+      "epoch": 0.08836,
+      "grad_norm": 3.359375,
+      "grad_norm_var": 0.0735260009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.8378,
+      "loss/crossentropy": 2.2390655279159546,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27328142523765564,
+      "step": 4418
+    },
+    {
+      "epoch": 0.0884,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.07681884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.6879,
+      "loss/crossentropy": 2.061118960380554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23206621408462524,
+      "step": 4420
+    },
+    {
+      "epoch": 0.08844,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.09147847493489583,
+      "learning_rate": 0.0001,
+      "loss": 4.3028,
+      "loss/crossentropy": 1.4919558763504028,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17881463468074799,
+      "step": 4422
+    },
+    {
+      "epoch": 0.08848,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.08662109375,
+      "learning_rate": 0.0001,
+      "loss": 4.7705,
+      "loss/crossentropy": 2.267301082611084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27718164026737213,
+      "step": 4424
+    },
+    {
+      "epoch": 0.08852,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.0862457275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.635,
+      "loss/crossentropy": 1.9008094668388367,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22865734994411469,
+      "step": 4426
+    },
+    {
+      "epoch": 0.08856,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.090966796875,
+      "learning_rate": 0.0001,
+      "loss": 4.385,
+      "loss/crossentropy": 1.8788060545921326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23521529138088226,
+      "step": 4428
+    },
+    {
+      "epoch": 0.0886,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.08844401041666666,
+      "learning_rate": 0.0001,
+      "loss": 4.6703,
+      "loss/crossentropy": 2.0775802731513977,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3402235209941864,
+      "step": 4430
+    },
+    {
+      "epoch": 0.08864,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.08642171223958334,
+      "learning_rate": 0.0001,
+      "loss": 4.5722,
+      "loss/crossentropy": 2.0950201749801636,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2665044367313385,
+      "step": 4432
+    },
+    {
+      "epoch": 0.08868,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.026167805989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.9235,
+      "loss/crossentropy": 2.328200340270996,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24911007285118103,
+      "step": 4434
+    },
+    {
+      "epoch": 0.08872,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.023273722330729166,
+      "learning_rate": 0.0001,
+      "loss": 4.7462,
+      "loss/crossentropy": 2.1840142011642456,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2855416387319565,
+      "step": 4436
+    },
+    {
+      "epoch": 0.08876,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.015559895833333334,
+      "learning_rate": 0.0001,
+      "loss": 4.1399,
+      "loss/crossentropy": 1.6167555451393127,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21646380424499512,
+      "step": 4438
+    },
+    {
+      "epoch": 0.0888,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.015543619791666666,
+      "learning_rate": 0.0001,
+      "loss": 4.5274,
+      "loss/crossentropy": 1.9902858138084412,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.237789124250412,
+      "step": 4440
+    },
+    {
+      "epoch": 0.08884,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.016357421875,
+      "learning_rate": 0.0001,
+      "loss": 4.6682,
+      "loss/crossentropy": 2.4779287576675415,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2726414203643799,
+      "step": 4442
+    },
+    {
+      "epoch": 0.08888,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.0116119384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.7415,
+      "loss/crossentropy": 2.0914896726608276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22887279838323593,
+      "step": 4444
+    },
+    {
+      "epoch": 0.08892,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.010570271809895834,
+      "learning_rate": 0.0001,
+      "loss": 4.498,
+      "loss/crossentropy": 2.0526055693626404,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2509382963180542,
+      "step": 4446
+    },
+    {
+      "epoch": 0.08896,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.010347493489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.4306,
+      "loss/crossentropy": 1.9779084920883179,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2320951297879219,
+      "step": 4448
+    },
+    {
+      "epoch": 0.089,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.010399373372395833,
+      "learning_rate": 0.0001,
+      "loss": 4.7725,
+      "loss/crossentropy": 2.2081698179244995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29023079574108124,
+      "step": 4450
+    },
+    {
+      "epoch": 0.08904,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.0074127197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.542,
+      "loss/crossentropy": 1.834806501865387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23430980741977692,
+      "step": 4452
+    },
+    {
+      "epoch": 0.08908,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.006884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.7087,
+      "loss/crossentropy": 2.4750468730926514,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27945323288440704,
+      "step": 4454
+    },
+    {
+      "epoch": 0.08912,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.007255045572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.6822,
+      "loss/crossentropy": 2.1766942739486694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28881968557834625,
+      "step": 4456
+    },
+    {
+      "epoch": 0.08916,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.009065755208333333,
+      "learning_rate": 0.0001,
+      "loss": 4.9204,
+      "loss/crossentropy": 2.265195846557617,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2616717368364334,
+      "step": 4458
+    },
+    {
+      "epoch": 0.0892,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.011473592122395833,
+      "learning_rate": 0.0001,
+      "loss": 4.5747,
+      "loss/crossentropy": 2.2438716888427734,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2630545049905777,
+      "step": 4460
+    },
+    {
+      "epoch": 0.08924,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.012572224934895833,
+      "learning_rate": 0.0001,
+      "loss": 4.3404,
+      "loss/crossentropy": 2.060324013233185,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23639824986457825,
+      "step": 4462
+    },
+    {
+      "epoch": 0.08928,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.0331207275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.733,
+      "loss/crossentropy": 1.8830525279045105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21016474813222885,
+      "step": 4464
+    },
+    {
+      "epoch": 0.08932,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.03435872395833333,
+      "learning_rate": 0.0001,
+      "loss": 4.7571,
+      "loss/crossentropy": 2.3420846462249756,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28108468651771545,
+      "step": 4466
+    },
+    {
+      "epoch": 0.08936,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.0337310791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.6851,
+      "loss/crossentropy": 1.9140342473983765,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22587041556835175,
+      "step": 4468
+    },
+    {
+      "epoch": 0.0894,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.032613118489583336,
+      "learning_rate": 0.0001,
+      "loss": 4.79,
+      "loss/crossentropy": 1.9753122925758362,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24506878852844238,
+      "step": 4470
+    },
+    {
+      "epoch": 0.08944,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.032389322916666664,
+      "learning_rate": 0.0001,
+      "loss": 4.7769,
+      "loss/crossentropy": 2.0735195875167847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2763114273548126,
+      "step": 4472
+    },
+    {
+      "epoch": 0.08948,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.034403483072916664,
+      "learning_rate": 0.0001,
+      "loss": 4.3593,
+      "loss/crossentropy": 1.7784460186958313,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22829323261976242,
+      "step": 4474
+    },
+    {
+      "epoch": 0.08952,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.051041666666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.9451,
+      "loss/crossentropy": 2.1188095808029175,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27410270273685455,
+      "step": 4476
+    },
+    {
+      "epoch": 0.08956,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.051634724934895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3392,
+      "loss/crossentropy": 2.320235252380371,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27514997124671936,
+      "step": 4478
+    },
+    {
+      "epoch": 0.0896,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.03766988118489583,
+      "learning_rate": 0.0001,
+      "loss": 4.8345,
+      "loss/crossentropy": 2.3023080825805664,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2516366094350815,
+      "step": 4480
+    },
+    {
+      "epoch": 0.08964,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.039567057291666666,
+      "learning_rate": 0.0001,
+      "loss": 4.6698,
+      "loss/crossentropy": 1.9677563905715942,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2362382560968399,
+      "step": 4482
+    },
+    {
+      "epoch": 0.08968,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.03951416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.9159,
+      "loss/crossentropy": 2.2005198001861572,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2760336175560951,
+      "step": 4484
+    },
+    {
+      "epoch": 0.08972,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0478912353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4418,
+      "loss/crossentropy": 1.9799931049346924,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25132423639297485,
+      "step": 4486
+    },
+    {
+      "epoch": 0.08976,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.10212300618489584,
+      "learning_rate": 0.0001,
+      "loss": 4.9643,
+      "loss/crossentropy": 2.205371141433716,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2695635259151459,
+      "step": 4488
+    },
+    {
+      "epoch": 0.0898,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.09692281087239583,
+      "learning_rate": 0.0001,
+      "loss": 4.9695,
+      "loss/crossentropy": 2.3500062227249146,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2736963629722595,
+      "step": 4490
+    },
+    {
+      "epoch": 0.08984,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.07867431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.9083,
+      "loss/crossentropy": 2.386352837085724,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2472890019416809,
+      "step": 4492
+    },
+    {
+      "epoch": 0.08988,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.07701416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3598,
+      "loss/crossentropy": 1.9863982200622559,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2288871705532074,
+      "step": 4494
+    },
+    {
+      "epoch": 0.08992,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.07757059733072917,
+      "learning_rate": 0.0001,
+      "loss": 4.6157,
+      "loss/crossentropy": 2.4088594913482666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24304527044296265,
+      "step": 4496
+    },
+    {
+      "epoch": 0.08996,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.07517801920572917,
+      "learning_rate": 0.0001,
+      "loss": 4.7457,
+      "loss/crossentropy": 2.1663140058517456,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2681227922439575,
+      "step": 4498
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0810455322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2199,
+      "loss/crossentropy": 1.9233570098876953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23169977217912674,
+      "step": 4500
+    },
+    {
+      "epoch": 0.09004,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.07579752604166666,
+      "learning_rate": 0.0001,
+      "loss": 4.4677,
+      "loss/crossentropy": 2.2940425872802734,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24809539318084717,
+      "step": 4502
+    },
+    {
+      "epoch": 0.09008,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.015771484375,
+      "learning_rate": 0.0001,
+      "loss": 4.5333,
+      "loss/crossentropy": 2.1874176263809204,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26286616921424866,
+      "step": 4504
+    },
+    {
+      "epoch": 0.09012,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.015868123372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.5029,
+      "loss/crossentropy": 2.190543472766876,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26766398549079895,
+      "step": 4506
+    },
+    {
+      "epoch": 0.09016,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.010823567708333334,
+      "learning_rate": 0.0001,
+      "loss": 4.464,
+      "loss/crossentropy": 2.132491707801819,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2514026165008545,
+      "step": 4508
+    },
+    {
+      "epoch": 0.0902,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.0107818603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5032,
+      "loss/crossentropy": 2.1492353677749634,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25663119554519653,
+      "step": 4510
+    },
+    {
+      "epoch": 0.09024,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.011454264322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.6166,
+      "loss/crossentropy": 2.2471927404403687,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24812395125627518,
+      "step": 4512
+    },
+    {
+      "epoch": 0.09028,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.007013956705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.5651,
+      "loss/crossentropy": 2.1944304704666138,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26567137241363525,
+      "step": 4514
+    },
+    {
+      "epoch": 0.09032,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.006883748372395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3054,
+      "loss/crossentropy": 1.7537739872932434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21784386038780212,
+      "step": 4516
+    },
+    {
+      "epoch": 0.09036,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.006981404622395834,
+      "learning_rate": 0.0001,
+      "loss": 4.6961,
+      "loss/crossentropy": 2.133580207824707,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27865441143512726,
+      "step": 4518
+    },
+    {
+      "epoch": 0.0904,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.008723958333333334,
+      "learning_rate": 0.0001,
+      "loss": 4.5486,
+      "loss/crossentropy": 2.0858315229415894,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26390860974788666,
+      "step": 4520
+    },
+    {
+      "epoch": 0.09044,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.008101399739583333,
+      "learning_rate": 0.0001,
+      "loss": 4.6833,
+      "loss/crossentropy": 1.9084516763687134,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23013630509376526,
+      "step": 4522
+    },
+    {
+      "epoch": 0.09048,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.00859375,
+      "learning_rate": 0.0001,
+      "loss": 4.5817,
+      "loss/crossentropy": 2.2123712301254272,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2739466577768326,
+      "step": 4524
+    },
+    {
+      "epoch": 0.09052,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.009691365559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.9898,
+      "loss/crossentropy": 2.3532934188842773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25509175658226013,
+      "step": 4526
+    },
+    {
+      "epoch": 0.09056,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.008137003580729166,
+      "learning_rate": 0.0001,
+      "loss": 4.7786,
+      "loss/crossentropy": 2.1543048620224,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23639082163572311,
+      "step": 4528
+    },
+    {
+      "epoch": 0.0906,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.0109527587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.647,
+      "loss/crossentropy": 2.1322286128997803,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23874164372682571,
+      "step": 4530
+    },
+    {
+      "epoch": 0.09064,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.01011962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.7623,
+      "loss/crossentropy": 2.5552597045898438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24949757009744644,
+      "step": 4532
+    },
+    {
+      "epoch": 0.09068,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.0117095947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.6999,
+      "loss/crossentropy": 2.1248819231987,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2795845717191696,
+      "step": 4534
+    },
+    {
+      "epoch": 0.09072,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.014655558268229167,
+      "learning_rate": 0.0001,
+      "loss": 4.7739,
+      "loss/crossentropy": 1.985984206199646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22764238715171814,
+      "step": 4536
+    },
+    {
+      "epoch": 0.09076,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.01519775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.64,
+      "loss/crossentropy": 2.220720648765564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2693287283182144,
+      "step": 4538
+    },
+    {
+      "epoch": 0.0908,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.01324462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.486,
+      "loss/crossentropy": 1.9541595578193665,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2419680804014206,
+      "step": 4540
+    },
+    {
+      "epoch": 0.09084,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.013916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.963,
+      "loss/crossentropy": 2.275113582611084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2562567666172981,
+      "step": 4542
+    },
+    {
+      "epoch": 0.09088,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.017122395833333335,
+      "learning_rate": 0.0001,
+      "loss": 4.7535,
+      "loss/crossentropy": 2.4411803483963013,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25798996537923813,
+      "step": 4544
+    },
+    {
+      "epoch": 0.09092,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.019823201497395835,
+      "learning_rate": 0.0001,
+      "loss": 4.7339,
+      "loss/crossentropy": 2.035769820213318,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25313758105039597,
+      "step": 4546
+    },
+    {
+      "epoch": 0.09096,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.018701171875,
+      "learning_rate": 0.0001,
+      "loss": 4.1331,
+      "loss/crossentropy": 1.9237529039382935,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24456002563238144,
+      "step": 4548
+    },
+    {
+      "epoch": 0.091,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.02281494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.6762,
+      "loss/crossentropy": 2.179704189300537,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2681535929441452,
+      "step": 4550
+    },
+    {
+      "epoch": 0.09104,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.024137369791666665,
+      "learning_rate": 0.0001,
+      "loss": 4.6899,
+      "loss/crossentropy": 2.013023316860199,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22643990069627762,
+      "step": 4552
+    },
+    {
+      "epoch": 0.09108,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.024201456705729166,
+      "learning_rate": 0.0001,
+      "loss": 4.6527,
+      "loss/crossentropy": 2.173883557319641,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26495447754859924,
+      "step": 4554
+    },
+    {
+      "epoch": 0.09112,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.025121053059895832,
+      "learning_rate": 0.0001,
+      "loss": 4.5323,
+      "loss/crossentropy": 1.9398870468139648,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24307211488485336,
+      "step": 4556
+    },
+    {
+      "epoch": 0.09116,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.025423177083333335,
+      "learning_rate": 0.0001,
+      "loss": 4.2028,
+      "loss/crossentropy": 1.8551223874092102,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21659143269062042,
+      "step": 4558
+    },
+    {
+      "epoch": 0.0912,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.0254058837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.7813,
+      "loss/crossentropy": 2.104207456111908,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26515253633260727,
+      "step": 4560
+    },
+    {
+      "epoch": 0.09124,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.023875935872395834,
+      "learning_rate": 0.0001,
+      "loss": 4.6241,
+      "loss/crossentropy": 2.31631863117218,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25454505532979965,
+      "step": 4562
+    },
+    {
+      "epoch": 0.09128,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.021800740559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.5732,
+      "loss/crossentropy": 2.331356406211853,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2667195200920105,
+      "step": 4564
+    },
+    {
+      "epoch": 0.09132,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.0176910400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.6714,
+      "loss/crossentropy": 2.2126184701919556,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24132181704044342,
+      "step": 4566
+    },
+    {
+      "epoch": 0.09136,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.013167317708333333,
+      "learning_rate": 0.0001,
+      "loss": 4.5753,
+      "loss/crossentropy": 2.330659508705139,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2838585078716278,
+      "step": 4568
+    },
+    {
+      "epoch": 0.0914,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.0132720947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4935,
+      "loss/crossentropy": 2.167214035987854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24021611362695694,
+      "step": 4570
+    },
+    {
+      "epoch": 0.09144,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.014046223958333333,
+      "learning_rate": 0.0001,
+      "loss": 4.5847,
+      "loss/crossentropy": 1.770102322101593,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2108836993575096,
+      "step": 4572
+    },
+    {
+      "epoch": 0.09148,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.015425618489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.2906,
+      "loss/crossentropy": 1.9293717741966248,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2186967208981514,
+      "step": 4574
+    },
+    {
+      "epoch": 0.09152,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.014188639322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.8128,
+      "loss/crossentropy": 2.4099135398864746,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25587645173072815,
+      "step": 4576
+    },
+    {
+      "epoch": 0.09156,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.06220296223958333,
+      "learning_rate": 0.0001,
+      "loss": 4.5727,
+      "loss/crossentropy": 1.7967870831489563,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2521464377641678,
+      "step": 4578
+    },
+    {
+      "epoch": 0.0916,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.0616851806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.7604,
+      "loss/crossentropy": 1.989583432674408,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22892683744430542,
+      "step": 4580
+    },
+    {
+      "epoch": 0.09164,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.0626953125,
+      "learning_rate": 0.0001,
+      "loss": 4.4986,
+      "loss/crossentropy": 2.167198657989502,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24552703648805618,
+      "step": 4582
+    },
+    {
+      "epoch": 0.09168,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.06129150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.8237,
+      "loss/crossentropy": 2.213137984275818,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2614079788327217,
+      "step": 4584
+    },
+    {
+      "epoch": 0.09172,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.06194661458333333,
+      "learning_rate": 0.0001,
+      "loss": 4.977,
+      "loss/crossentropy": 2.3586392998695374,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2527218610048294,
+      "step": 4586
+    },
+    {
+      "epoch": 0.09176,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.06054280598958333,
+      "learning_rate": 0.0001,
+      "loss": 4.5118,
+      "loss/crossentropy": 2.326598286628723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2684827446937561,
+      "step": 4588
+    },
+    {
+      "epoch": 0.0918,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.055562337239583336,
+      "learning_rate": 0.0001,
+      "loss": 4.4296,
+      "loss/crossentropy": 2.1365907192230225,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2705482095479965,
+      "step": 4590
+    },
+    {
+      "epoch": 0.09184,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.056005859375,
+      "learning_rate": 0.0001,
+      "loss": 4.6895,
+      "loss/crossentropy": 1.816661775112152,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2188858687877655,
+      "step": 4592
+    },
+    {
+      "epoch": 0.09188,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.020731608072916668,
+      "learning_rate": 0.0001,
+      "loss": 4.5953,
+      "loss/crossentropy": 2.1533923149108887,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2372177392244339,
+      "step": 4594
+    },
+    {
+      "epoch": 0.09192,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.04504801432291667,
+      "learning_rate": 0.0001,
+      "loss": 4.9472,
+      "loss/crossentropy": 2.2175614833831787,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27555912733078003,
+      "step": 4596
+    },
+    {
+      "epoch": 0.09196,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.04644775390625,
+      "learning_rate": 0.0001,
+      "loss": 5.11,
+      "loss/crossentropy": 2.583792209625244,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2970864772796631,
+      "step": 4598
+    },
+    {
+      "epoch": 0.092,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.046751912434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.5761,
+      "loss/crossentropy": 2.227868676185608,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.259210504591465,
+      "step": 4600
+    },
+    {
+      "epoch": 0.09204,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.047118123372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4125,
+      "loss/crossentropy": 2.146941900253296,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26443855464458466,
+      "step": 4602
+    },
+    {
+      "epoch": 0.09208,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.04439697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.8655,
+      "loss/crossentropy": 2.129795551300049,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2451881766319275,
+      "step": 4604
+    },
+    {
+      "epoch": 0.09212,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.0506256103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.895,
+      "loss/crossentropy": 2.2696213722229004,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26037923991680145,
+      "step": 4606
+    },
+    {
+      "epoch": 0.09216,
+      "grad_norm": 3.03125,
+      "grad_norm_var": 0.07105712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.7424,
+      "loss/crossentropy": 2.1916056871414185,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2648691013455391,
+      "step": 4608
+    },
+    {
+      "epoch": 0.0922,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.07119038899739584,
+      "learning_rate": 0.0001,
+      "loss": 4.5021,
+      "loss/crossentropy": 1.975761890411377,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22248996049165726,
+      "step": 4610
+    },
+    {
+      "epoch": 0.09224,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.053511555989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.6596,
+      "loss/crossentropy": 2.403375506401062,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24553030729293823,
+      "step": 4612
+    },
+    {
+      "epoch": 0.09228,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.049637858072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.4542,
+      "loss/crossentropy": 1.9468475580215454,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24023501574993134,
+      "step": 4614
+    },
+    {
+      "epoch": 0.09232,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.0534088134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.8095,
+      "loss/crossentropy": 2.013557195663452,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23829826712608337,
+      "step": 4616
+    },
+    {
+      "epoch": 0.09236,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0519439697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.5736,
+      "loss/crossentropy": 2.0552549958229065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2391991689801216,
+      "step": 4618
+    },
+    {
+      "epoch": 0.0924,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.053694661458333334,
+      "learning_rate": 0.0001,
+      "loss": 4.4091,
+      "loss/crossentropy": 1.8311110734939575,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23821169883012772,
+      "step": 4620
+    },
+    {
+      "epoch": 0.09244,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.04431966145833333,
+      "learning_rate": 0.0001,
+      "loss": 4.5034,
+      "loss/crossentropy": 2.0346454977989197,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23688799142837524,
+      "step": 4622
+    },
+    {
+      "epoch": 0.09248,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.016304524739583333,
+      "learning_rate": 0.0001,
+      "loss": 4.5329,
+      "loss/crossentropy": 1.8475716710090637,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23444947600364685,
+      "step": 4624
+    },
+    {
+      "epoch": 0.09252,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.0136383056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4315,
+      "loss/crossentropy": 2.075626790523529,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2663211151957512,
+      "step": 4626
+    },
+    {
+      "epoch": 0.09256,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.0130035400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.9214,
+      "loss/crossentropy": 2.413028359413147,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29516373574733734,
+      "step": 4628
+    },
+    {
+      "epoch": 0.0926,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.014143880208333333,
+      "learning_rate": 0.0001,
+      "loss": 4.6878,
+      "loss/crossentropy": 1.8549358248710632,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23275860399007797,
+      "step": 4630
+    },
+    {
+      "epoch": 0.09264,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.0076405843098958336,
+      "learning_rate": 0.0001,
+      "loss": 4.8594,
+      "loss/crossentropy": 2.273250460624695,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2511187419295311,
+      "step": 4632
+    },
+    {
+      "epoch": 0.09268,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.013093058268229167,
+      "learning_rate": 0.0001,
+      "loss": 4.336,
+      "loss/crossentropy": 1.929758369922638,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25462278723716736,
+      "step": 4634
+    },
+    {
+      "epoch": 0.09272,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.013158162434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.5888,
+      "loss/crossentropy": 2.0929598212242126,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23592843115329742,
+      "step": 4636
+    },
+    {
+      "epoch": 0.09276,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.015282185872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.4595,
+      "loss/crossentropy": 1.973824143409729,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24049720913171768,
+      "step": 4638
+    },
+    {
+      "epoch": 0.0928,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.014207967122395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2606,
+      "loss/crossentropy": 1.8606626987457275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21698658913373947,
+      "step": 4640
+    },
+    {
+      "epoch": 0.09284,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.01875,
+      "learning_rate": 0.0001,
+      "loss": 4.9827,
+      "loss/crossentropy": 2.241386890411377,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2831972986459732,
+      "step": 4642
+    },
+    {
+      "epoch": 0.09288,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.0166656494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.7046,
+      "loss/crossentropy": 2.2538920640945435,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2482161968946457,
+      "step": 4644
+    },
+    {
+      "epoch": 0.09292,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.01812744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.8557,
+      "loss/crossentropy": 2.067206382751465,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26665763556957245,
+      "step": 4646
+    },
+    {
+      "epoch": 0.09296,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.021751912434895833,
+      "learning_rate": 0.0001,
+      "loss": 4.9474,
+      "loss/crossentropy": 2.33401358127594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2752760946750641,
+      "step": 4648
+    },
+    {
+      "epoch": 0.093,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.04914449055989583,
+      "learning_rate": 0.0001,
+      "loss": 4.8658,
+      "loss/crossentropy": 2.1510735750198364,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24588150531053543,
+      "step": 4650
+    },
+    {
+      "epoch": 0.09304,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.049397786458333336,
+      "learning_rate": 0.0001,
+      "loss": 4.579,
+      "loss/crossentropy": 2.086448848247528,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.261296346783638,
+      "step": 4652
+    },
+    {
+      "epoch": 0.09308,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.0446685791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.7495,
+      "loss/crossentropy": 1.5951193571090698,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2312234491109848,
+      "step": 4654
+    },
+    {
+      "epoch": 0.09312,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.04052734375,
+      "learning_rate": 0.0001,
+      "loss": 4.2273,
+      "loss/crossentropy": 2.0098360776901245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22354383766651154,
+      "step": 4656
+    },
+    {
+      "epoch": 0.09316,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.04079488118489583,
+      "learning_rate": 0.0001,
+      "loss": 4.6047,
+      "loss/crossentropy": 2.263219714164734,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26900506019592285,
+      "step": 4658
+    },
+    {
+      "epoch": 0.0932,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.0485504150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.5575,
+      "loss/crossentropy": 1.9481555819511414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24665354192256927,
+      "step": 4660
+    },
+    {
+      "epoch": 0.09324,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.04988505045572917,
+      "learning_rate": 0.0001,
+      "loss": 4.5895,
+      "loss/crossentropy": 1.8751367926597595,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24356309324502945,
+      "step": 4662
+    },
+    {
+      "epoch": 0.09328,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.06110026041666667,
+      "learning_rate": 0.0001,
+      "loss": 4.7739,
+      "loss/crossentropy": 2.343896746635437,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2558425962924957,
+      "step": 4664
+    },
+    {
+      "epoch": 0.09332,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.032811482747395836,
+      "learning_rate": 0.0001,
+      "loss": 4.779,
+      "loss/crossentropy": 2.067797303199768,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25030098110437393,
+      "step": 4666
+    },
+    {
+      "epoch": 0.09336,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.029816691080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.7427,
+      "loss/crossentropy": 2.4221161603927612,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26934675872325897,
+      "step": 4668
+    },
+    {
+      "epoch": 0.0934,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.038182576497395836,
+      "learning_rate": 0.0001,
+      "loss": 4.845,
+      "loss/crossentropy": 2.2733672857284546,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2536270022392273,
+      "step": 4670
+    },
+    {
+      "epoch": 0.09344,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.0369537353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.7721,
+      "loss/crossentropy": 2.179157257080078,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24323320388793945,
+      "step": 4672
+    },
+    {
+      "epoch": 0.09348,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.03875325520833333,
+      "learning_rate": 0.0001,
+      "loss": 4.8574,
+      "loss/crossentropy": 2.3613970279693604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2567686140537262,
+      "step": 4674
+    },
+    {
+      "epoch": 0.09352,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.029683430989583332,
+      "learning_rate": 0.0001,
+      "loss": 4.787,
+      "loss/crossentropy": 2.0755810141563416,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2349473536014557,
+      "step": 4676
+    },
+    {
+      "epoch": 0.09356,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.03134358723958333,
+      "learning_rate": 0.0001,
+      "loss": 4.6235,
+      "loss/crossentropy": 1.9971619248390198,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22000454366207123,
+      "step": 4678
+    },
+    {
+      "epoch": 0.0936,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.020068359375,
+      "learning_rate": 0.0001,
+      "loss": 4.5307,
+      "loss/crossentropy": 1.9419977068901062,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21052303910255432,
+      "step": 4680
+    },
+    {
+      "epoch": 0.09364,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.021122233072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.6314,
+      "loss/crossentropy": 2.1175334453582764,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23567666858434677,
+      "step": 4682
+    },
+    {
+      "epoch": 0.09368,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.021122233072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.5955,
+      "loss/crossentropy": 2.0300605297088623,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29177258908748627,
+      "step": 4684
+    },
+    {
+      "epoch": 0.09372,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.009228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5583,
+      "loss/crossentropy": 2.052473723888397,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22480525821447372,
+      "step": 4686
+    },
+    {
+      "epoch": 0.09376,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.008275349934895834,
+      "learning_rate": 0.0001,
+      "loss": 4.4925,
+      "loss/crossentropy": 2.2506834268569946,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26876674592494965,
+      "step": 4688
+    },
+    {
+      "epoch": 0.0938,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.007515462239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.291,
+      "loss/crossentropy": 1.930393099784851,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22246946394443512,
+      "step": 4690
+    },
+    {
+      "epoch": 0.09384,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.007616170247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.9317,
+      "loss/crossentropy": 2.5477651357650757,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27808643877506256,
+      "step": 4692
+    },
+    {
+      "epoch": 0.09388,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.009033203125,
+      "learning_rate": 0.0001,
+      "loss": 4.5313,
+      "loss/crossentropy": 2.1059221625328064,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.253792941570282,
+      "step": 4694
+    },
+    {
+      "epoch": 0.09392,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0091217041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.607,
+      "loss/crossentropy": 2.0058358907699585,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23942459374666214,
+      "step": 4696
+    },
+    {
+      "epoch": 0.09396,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2024,
+      "loss/crossentropy": 1.9118947982788086,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2300974577665329,
+      "step": 4698
+    },
+    {
+      "epoch": 0.094,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.013337198893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.6541,
+      "loss/crossentropy": 2.1172733902931213,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24785596132278442,
+      "step": 4700
+    },
+    {
+      "epoch": 0.09404,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.016044108072916667,
+      "learning_rate": 0.0001,
+      "loss": 4.9005,
+      "loss/crossentropy": 2.0064170956611633,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25045372545719147,
+      "step": 4702
+    },
+    {
+      "epoch": 0.09408,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.015738932291666667,
+      "learning_rate": 0.0001,
+      "loss": 4.8139,
+      "loss/crossentropy": 2.5816656351089478,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27938568592071533,
+      "step": 4704
+    },
+    {
+      "epoch": 0.09412,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.016597493489583334,
+      "learning_rate": 0.0001,
+      "loss": 4.866,
+      "loss/crossentropy": 2.5768171548843384,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27975398302078247,
+      "step": 4706
+    },
+    {
+      "epoch": 0.09416,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0153717041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.6075,
+      "loss/crossentropy": 2.323893189430237,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2608431279659271,
+      "step": 4708
+    },
+    {
+      "epoch": 0.0942,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.01357421875,
+      "learning_rate": 0.0001,
+      "loss": 4.3564,
+      "loss/crossentropy": 1.7910810708999634,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21380099654197693,
+      "step": 4710
+    },
+    {
+      "epoch": 0.09424,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.015848795572916668,
+      "learning_rate": 0.0001,
+      "loss": 4.6374,
+      "loss/crossentropy": 2.214052677154541,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24833911657333374,
+      "step": 4712
+    },
+    {
+      "epoch": 0.09428,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.0158355712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.9987,
+      "loss/crossentropy": 2.0850380063056946,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2419341504573822,
+      "step": 4714
+    },
+    {
+      "epoch": 0.09432,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.014058430989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.2961,
+      "loss/crossentropy": 2.0707927346229553,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2499033510684967,
+      "step": 4716
+    },
+    {
+      "epoch": 0.09436,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.0125396728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4057,
+      "loss/crossentropy": 2.11221444606781,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25916673243045807,
+      "step": 4718
+    },
+    {
+      "epoch": 0.0944,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.012718709309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.6057,
+      "loss/crossentropy": 1.8400230407714844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22771906107664108,
+      "step": 4720
+    },
+    {
+      "epoch": 0.09444,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.008234659830729166,
+      "learning_rate": 0.0001,
+      "loss": 4.4929,
+      "loss/crossentropy": 2.2271865606307983,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2574647441506386,
+      "step": 4722
+    },
+    {
+      "epoch": 0.09448,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.008649698893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.4629,
+      "loss/crossentropy": 2.4532920122146606,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27140843868255615,
+      "step": 4724
+    },
+    {
+      "epoch": 0.09452,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.009504191080729167,
+      "learning_rate": 0.0001,
+      "loss": 4.731,
+      "loss/crossentropy": 1.951303780078888,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2333502620458603,
+      "step": 4726
+    },
+    {
+      "epoch": 0.09456,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.009137980143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.6334,
+      "loss/crossentropy": 2.209821343421936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24776015430688858,
+      "step": 4728
+    },
+    {
+      "epoch": 0.0946,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.010872395833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.861,
+      "loss/crossentropy": 2.434941530227661,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2747645229101181,
+      "step": 4730
+    },
+    {
+      "epoch": 0.09464,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.010602823893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.6145,
+      "loss/crossentropy": 2.051652252674103,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24515582621097565,
+      "step": 4732
+    },
+    {
+      "epoch": 0.09468,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.009797159830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.6804,
+      "loss/crossentropy": 2.0039377212524414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23073314130306244,
+      "step": 4734
+    },
+    {
+      "epoch": 0.09472,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.19903971354166666,
+      "learning_rate": 0.0001,
+      "loss": 4.8333,
+      "loss/crossentropy": 2.166410982608795,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25871724635362625,
+      "step": 4736
+    },
+    {
+      "epoch": 0.09476,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.2066802978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2739,
+      "loss/crossentropy": 1.9289153218269348,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23183659464120865,
+      "step": 4738
+    },
+    {
+      "epoch": 0.0948,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.20640360514322917,
+      "learning_rate": 0.0001,
+      "loss": 4.5903,
+      "loss/crossentropy": 2.270771861076355,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26335832476615906,
+      "step": 4740
+    },
+    {
+      "epoch": 0.09484,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.206494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.7142,
+      "loss/crossentropy": 2.395651936531067,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28527122735977173,
+      "step": 4742
+    },
+    {
+      "epoch": 0.09488,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.20437723795572918,
+      "learning_rate": 0.0001,
+      "loss": 4.5083,
+      "loss/crossentropy": 1.8597867488861084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.224149189889431,
+      "step": 4744
+    },
+    {
+      "epoch": 0.09492,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.20693257649739583,
+      "learning_rate": 0.0001,
+      "loss": 4.4415,
+      "loss/crossentropy": 1.7795116305351257,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23233074694871902,
+      "step": 4746
+    },
+    {
+      "epoch": 0.09496,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.20255533854166666,
+      "learning_rate": 0.0001,
+      "loss": 4.7749,
+      "loss/crossentropy": 1.9449282884597778,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2524118423461914,
+      "step": 4748
+    },
+    {
+      "epoch": 0.095,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.2005035400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.7047,
+      "loss/crossentropy": 2.195169448852539,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2349853590130806,
+      "step": 4750
+    },
+    {
+      "epoch": 0.09504,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.022468058268229167,
+      "learning_rate": 0.0001,
+      "loss": 4.5268,
+      "loss/crossentropy": 1.6628928184509277,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1975274682044983,
+      "step": 4752
+    },
+    {
+      "epoch": 0.09508,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.018488566080729168,
+      "learning_rate": 0.0001,
+      "loss": 4.5862,
+      "loss/crossentropy": 2.0991236567497253,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23341374844312668,
+      "step": 4754
+    },
+    {
+      "epoch": 0.09512,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.017024739583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.8338,
+      "loss/crossentropy": 2.350286066532135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2648291736841202,
+      "step": 4756
+    },
+    {
+      "epoch": 0.09516,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.017430623372395832,
+      "learning_rate": 0.0001,
+      "loss": 4.5788,
+      "loss/crossentropy": 2.02384877204895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24653150886297226,
+      "step": 4758
+    },
+    {
+      "epoch": 0.0952,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.01734619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.7052,
+      "loss/crossentropy": 2.264349341392517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25881223380565643,
+      "step": 4760
+    },
+    {
+      "epoch": 0.09524,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 0.03779296875,
+      "learning_rate": 0.0001,
+      "loss": 4.8216,
+      "loss/crossentropy": 2.197356939315796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2565019279718399,
+      "step": 4762
+    },
+    {
+      "epoch": 0.09528,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.038655598958333336,
+      "learning_rate": 0.0001,
+      "loss": 4.3849,
+      "loss/crossentropy": 1.8794787526130676,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21771979331970215,
+      "step": 4764
+    },
+    {
+      "epoch": 0.09532,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.03795572916666667,
+      "learning_rate": 0.0001,
+      "loss": 4.7391,
+      "loss/crossentropy": 2.107520580291748,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2350333333015442,
+      "step": 4766
+    },
+    {
+      "epoch": 0.09536,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.03297119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.5194,
+      "loss/crossentropy": 2.082249402999878,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24860269576311111,
+      "step": 4768
+    },
+    {
+      "epoch": 0.0954,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.03062744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3985,
+      "loss/crossentropy": 1.9632289409637451,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24099770188331604,
+      "step": 4770
+    },
+    {
+      "epoch": 0.09544,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.02994384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4526,
+      "loss/crossentropy": 2.3403327465057373,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2573448717594147,
+      "step": 4772
+    },
+    {
+      "epoch": 0.09548,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.03181050618489583,
+      "learning_rate": 0.0001,
+      "loss": 4.7711,
+      "loss/crossentropy": 2.302455425262451,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28050975501537323,
+      "step": 4774
+    },
+    {
+      "epoch": 0.09552,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.031103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.6092,
+      "loss/crossentropy": 2.2875213623046875,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23910009860992432,
+      "step": 4776
+    },
+    {
+      "epoch": 0.09556,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.022721354166666666,
+      "learning_rate": 0.0001,
+      "loss": 4.5237,
+      "loss/crossentropy": 2.0440263748168945,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24640783667564392,
+      "step": 4778
+    },
+    {
+      "epoch": 0.0956,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.02613525390625,
+      "learning_rate": 0.0001,
+      "loss": 4.6328,
+      "loss/crossentropy": 2.1597142219543457,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24566050618886948,
+      "step": 4780
+    },
+    {
+      "epoch": 0.09564,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.027765909830729168,
+      "learning_rate": 0.0001,
+      "loss": 4.7138,
+      "loss/crossentropy": 2.2846235036849976,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2598261833190918,
+      "step": 4782
+    },
+    {
+      "epoch": 0.09568,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.029683430989583332,
+      "learning_rate": 0.0001,
+      "loss": 4.4527,
+      "loss/crossentropy": 2.032243251800537,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2488306611776352,
+      "step": 4784
+    },
+    {
+      "epoch": 0.09572,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.037230428059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.7651,
+      "loss/crossentropy": 2.4999172687530518,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25949685275554657,
+      "step": 4786
+    },
+    {
+      "epoch": 0.09576,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.0375152587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.6068,
+      "loss/crossentropy": 2.1610575914382935,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2414923906326294,
+      "step": 4788
+    },
+    {
+      "epoch": 0.0958,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.03135477701822917,
+      "learning_rate": 0.0001,
+      "loss": 4.5309,
+      "loss/crossentropy": 1.8679919838905334,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21853071451187134,
+      "step": 4790
+    },
+    {
+      "epoch": 0.09584,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.031281534830729166,
+      "learning_rate": 0.0001,
+      "loss": 4.697,
+      "loss/crossentropy": 2.072615623474121,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24943216145038605,
+      "step": 4792
+    },
+    {
+      "epoch": 0.09588,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.032624308268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.5555,
+      "loss/crossentropy": 2.420115113258362,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2840816229581833,
+      "step": 4794
+    },
+    {
+      "epoch": 0.09592,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.0455078125,
+      "learning_rate": 0.0001,
+      "loss": 5.102,
+      "loss/crossentropy": 2.2809360027313232,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2739071249961853,
+      "step": 4796
+    },
+    {
+      "epoch": 0.09596,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.04319559733072917,
+      "learning_rate": 0.0001,
+      "loss": 4.5418,
+      "loss/crossentropy": 2.1809465289115906,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24680403620004654,
+      "step": 4798
+    },
+    {
+      "epoch": 0.096,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.044611612955729164,
+      "learning_rate": 0.0001,
+      "loss": 4.6994,
+      "loss/crossentropy": 1.9710316061973572,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2598233222961426,
+      "step": 4800
+    },
+    {
+      "epoch": 0.09604,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.035477701822916666,
+      "learning_rate": 0.0001,
+      "loss": 4.9811,
+      "loss/crossentropy": 2.4378572702407837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2665309011936188,
+      "step": 4802
+    },
+    {
+      "epoch": 0.09608,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.03759358723958333,
+      "learning_rate": 0.0001,
+      "loss": 4.604,
+      "loss/crossentropy": 1.8578800559043884,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24051420390605927,
+      "step": 4804
+    },
+    {
+      "epoch": 0.09612,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.034520467122395836,
+      "learning_rate": 0.0001,
+      "loss": 4.4567,
+      "loss/crossentropy": 2.191601037979126,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25017624348402023,
+      "step": 4806
+    },
+    {
+      "epoch": 0.09616,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.038374837239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.5777,
+      "loss/crossentropy": 2.2077550888061523,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31426818668842316,
+      "step": 4808
+    },
+    {
+      "epoch": 0.0962,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.030659993489583332,
+      "learning_rate": 0.0001,
+      "loss": 4.917,
+      "loss/crossentropy": 2.2983503341674805,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26998060941696167,
+      "step": 4810
+    },
+    {
+      "epoch": 0.09624,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.020824178059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.8691,
+      "loss/crossentropy": 2.0875505208969116,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2673826217651367,
+      "step": 4812
+    },
+    {
+      "epoch": 0.09628,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.031403605143229166,
+      "learning_rate": 0.0001,
+      "loss": 4.0736,
+      "loss/crossentropy": 1.9439318776130676,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2379719465970993,
+      "step": 4814
+    },
+    {
+      "epoch": 0.09632,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.03611551920572917,
+      "learning_rate": 0.0001,
+      "loss": 4.4698,
+      "loss/crossentropy": 1.9822518229484558,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24425340443849564,
+      "step": 4816
+    },
+    {
+      "epoch": 0.09636,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.03902994791666667,
+      "learning_rate": 0.0001,
+      "loss": 4.475,
+      "loss/crossentropy": 2.0238336324691772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2409110590815544,
+      "step": 4818
+    },
+    {
+      "epoch": 0.0964,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.03707275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.5401,
+      "loss/crossentropy": 2.454450249671936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26127950847148895,
+      "step": 4820
+    },
+    {
+      "epoch": 0.09644,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.038407389322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.5682,
+      "loss/crossentropy": 2.210579752922058,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23662539571523666,
+      "step": 4822
+    },
+    {
+      "epoch": 0.09648,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.034601847330729164,
+      "learning_rate": 0.0001,
+      "loss": 4.5527,
+      "loss/crossentropy": 1.8737664222717285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22884630411863327,
+      "step": 4824
+    },
+    {
+      "epoch": 0.09652,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.0335357666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.6221,
+      "loss/crossentropy": 2.1939562559127808,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25145241618156433,
+      "step": 4826
+    },
+    {
+      "epoch": 0.09656,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.0278472900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.6685,
+      "loss/crossentropy": 2.133625030517578,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25248992443084717,
+      "step": 4828
+    },
+    {
+      "epoch": 0.0966,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.0197906494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2854,
+      "loss/crossentropy": 2.259738326072693,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24727293848991394,
+      "step": 4830
+    },
+    {
+      "epoch": 0.09664,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.043635050455729164,
+      "learning_rate": 0.0001,
+      "loss": 4.7497,
+      "loss/crossentropy": 2.0186268091201782,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27940231561660767,
+      "step": 4832
+    },
+    {
+      "epoch": 0.09668,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.04185282389322917,
+      "learning_rate": 0.0001,
+      "loss": 4.5867,
+      "loss/crossentropy": 1.9686395525932312,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23021705448627472,
+      "step": 4834
+    },
+    {
+      "epoch": 0.09672,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.0424224853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.6157,
+      "loss/crossentropy": 2.1677820682525635,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2766892910003662,
+      "step": 4836
+    },
+    {
+      "epoch": 0.09676,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.04670308430989583,
+      "learning_rate": 0.0001,
+      "loss": 4.3751,
+      "loss/crossentropy": 1.8458788990974426,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23294126987457275,
+      "step": 4838
+    },
+    {
+      "epoch": 0.0968,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.0495269775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.8931,
+      "loss/crossentropy": 1.7898097038269043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23436476290225983,
+      "step": 4840
+    },
+    {
+      "epoch": 0.09684,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.04843343098958333,
+      "learning_rate": 0.0001,
+      "loss": 4.5169,
+      "loss/crossentropy": 2.4594497680664062,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2618473023176193,
+      "step": 4842
+    },
+    {
+      "epoch": 0.09688,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.04752197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4841,
+      "loss/crossentropy": 2.061558425426483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2523345798254013,
+      "step": 4844
+    },
+    {
+      "epoch": 0.09692,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.04462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5606,
+      "loss/crossentropy": 2.060658574104309,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2493506520986557,
+      "step": 4846
+    },
+    {
+      "epoch": 0.09696,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.016169230143229168,
+      "learning_rate": 0.0001,
+      "loss": 4.8028,
+      "loss/crossentropy": 2.201029062271118,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2629493921995163,
+      "step": 4848
+    },
+    {
+      "epoch": 0.097,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.01519775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.7404,
+      "loss/crossentropy": 2.199273705482483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27054519951343536,
+      "step": 4850
+    },
+    {
+      "epoch": 0.09704,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0189453125,
+      "learning_rate": 0.0001,
+      "loss": 4.3008,
+      "loss/crossentropy": 2.2466784715652466,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2678111642599106,
+      "step": 4852
+    },
+    {
+      "epoch": 0.09708,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.022054036458333332,
+      "learning_rate": 0.0001,
+      "loss": 4.6762,
+      "loss/crossentropy": 1.9152815341949463,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2441672906279564,
+      "step": 4854
+    },
+    {
+      "epoch": 0.09712,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.017577107747395834,
+      "learning_rate": 0.0001,
+      "loss": 4.3818,
+      "loss/crossentropy": 2.092438578605652,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24061349034309387,
+      "step": 4856
+    },
+    {
+      "epoch": 0.09716,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.01802978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3224,
+      "loss/crossentropy": 1.8620384335517883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23738765716552734,
+      "step": 4858
+    },
+    {
+      "epoch": 0.0972,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.04138997395833333,
+      "learning_rate": 0.0001,
+      "loss": 4.8798,
+      "loss/crossentropy": 2.199701428413391,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26521213352680206,
+      "step": 4860
+    },
+    {
+      "epoch": 0.09724,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.042170206705729164,
+      "learning_rate": 0.0001,
+      "loss": 4.7598,
+      "loss/crossentropy": 2.3637821674346924,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2707534506917,
+      "step": 4862
+    },
+    {
+      "epoch": 0.09728,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.045491536458333336,
+      "learning_rate": 0.0001,
+      "loss": 4.5184,
+      "loss/crossentropy": 2.207235813140869,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28080131858587265,
+      "step": 4864
+    },
+    {
+      "epoch": 0.09732,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0466461181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3798,
+      "loss/crossentropy": 1.901595950126648,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23459318280220032,
+      "step": 4866
+    },
+    {
+      "epoch": 0.09736,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.042769368489583334,
+      "learning_rate": 0.0001,
+      "loss": 4.4289,
+      "loss/crossentropy": 2.298312544822693,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2580094337463379,
+      "step": 4868
+    },
+    {
+      "epoch": 0.0974,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.03658854166666667,
+      "learning_rate": 0.0001,
+      "loss": 4.7576,
+      "loss/crossentropy": 2.29680597782135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24749226868152618,
+      "step": 4870
+    },
+    {
+      "epoch": 0.09744,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0361968994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.5839,
+      "loss/crossentropy": 2.1169378757476807,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2565944790840149,
+      "step": 4872
+    },
+    {
+      "epoch": 0.09748,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.04035542805989583,
+      "learning_rate": 0.0001,
+      "loss": 4.5364,
+      "loss/crossentropy": 2.0863184928894043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2527791038155556,
+      "step": 4874
+    },
+    {
+      "epoch": 0.09752,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.015625,
+      "learning_rate": 0.0001,
+      "loss": 4.6176,
+      "loss/crossentropy": 2.146193563938141,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23968011140823364,
+      "step": 4876
+    },
+    {
+      "epoch": 0.09756,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.01060791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.5598,
+      "loss/crossentropy": 2.1125508546829224,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2498009204864502,
+      "step": 4878
+    },
+    {
+      "epoch": 0.0976,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.004130045572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.772,
+      "loss/crossentropy": 2.13166081905365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2640175372362137,
+      "step": 4880
+    },
+    {
+      "epoch": 0.09764,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0051910400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.5276,
+      "loss/crossentropy": 2.047860622406006,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23980721086263657,
+      "step": 4882
+    },
+    {
+      "epoch": 0.09768,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.010477701822916666,
+      "learning_rate": 0.0001,
+      "loss": 4.8519,
+      "loss/crossentropy": 2.362569808959961,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27908293902873993,
+      "step": 4884
+    },
+    {
+      "epoch": 0.09772,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012398274739583333,
+      "learning_rate": 0.0001,
+      "loss": 4.2709,
+      "loss/crossentropy": 1.8781500458717346,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22341010719537735,
+      "step": 4886
+    },
+    {
+      "epoch": 0.09776,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.019807942708333335,
+      "learning_rate": 0.0001,
+      "loss": 4.9408,
+      "loss/crossentropy": 2.1138893365859985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2533891350030899,
+      "step": 4888
+    },
+    {
+      "epoch": 0.0978,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.0188140869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.5981,
+      "loss/crossentropy": 2.2212090492248535,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.279159352183342,
+      "step": 4890
+    },
+    {
+      "epoch": 0.09784,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.0177886962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.7245,
+      "loss/crossentropy": 1.9747707843780518,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24078013002872467,
+      "step": 4892
+    },
+    {
+      "epoch": 0.09788,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.020015462239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.4051,
+      "loss/crossentropy": 1.8607316613197327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25342857837677,
+      "step": 4894
+    },
+    {
+      "epoch": 0.09792,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.020963541666666665,
+      "learning_rate": 0.0001,
+      "loss": 4.6759,
+      "loss/crossentropy": 2.245271682739258,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24260863661766052,
+      "step": 4896
+    },
+    {
+      "epoch": 0.09796,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.017560831705729165,
+      "learning_rate": 0.0001,
+      "loss": 4.5006,
+      "loss/crossentropy": 2.0503702759742737,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25429168343544006,
+      "step": 4898
+    },
+    {
+      "epoch": 0.098,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.017236328125,
+      "learning_rate": 0.0001,
+      "loss": 4.421,
+      "loss/crossentropy": 1.7784077525138855,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2249627709388733,
+      "step": 4900
+    },
+    {
+      "epoch": 0.09804,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.0114898681640625,
+      "learning_rate": 0.0001,
+      "loss": 4.5398,
+      "loss/crossentropy": 1.9827336072921753,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22001181542873383,
+      "step": 4902
+    },
+    {
+      "epoch": 0.09808,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 1.638996378580729,
+      "learning_rate": 0.0001,
+      "loss": 4.8149,
+      "loss/crossentropy": 2.1003565788269043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2550586014986038,
+      "step": 4904
+    },
+    {
+      "epoch": 0.09812,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 1.6366933186848958,
+      "learning_rate": 0.0001,
+      "loss": 4.5099,
+      "loss/crossentropy": 1.9565055966377258,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25481177121400833,
+      "step": 4906
+    },
+    {
+      "epoch": 0.09816,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 1.6479777018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.6214,
+      "loss/crossentropy": 2.1693456172943115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23336851596832275,
+      "step": 4908
+    },
+    {
+      "epoch": 0.0982,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 1.6428995768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.7637,
+      "loss/crossentropy": 2.050541341304779,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25821831077337265,
+      "step": 4910
+    },
+    {
+      "epoch": 0.09824,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 1.62906494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.8515,
+      "loss/crossentropy": 2.168497681617737,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2674206495285034,
+      "step": 4912
+    },
+    {
+      "epoch": 0.09828,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 1.6097320556640624,
+      "learning_rate": 0.0001,
+      "loss": 5.1667,
+      "loss/crossentropy": 2.147577404975891,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26900771260261536,
+      "step": 4914
+    },
+    {
+      "epoch": 0.09832,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 1.6371734619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3958,
+      "loss/crossentropy": 2.4436198472976685,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2504816800355911,
+      "step": 4916
+    },
+    {
+      "epoch": 0.09836,
+      "grad_norm": 2.25,
+      "grad_norm_var": 1.6447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.7244,
+      "loss/crossentropy": 2.097359538078308,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24005448818206787,
+      "step": 4918
+    },
+    {
+      "epoch": 0.0984,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0267974853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2305,
+      "loss/crossentropy": 1.8467384576797485,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22932368516921997,
+      "step": 4920
+    },
+    {
+      "epoch": 0.09844,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.04263916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.4407,
+      "loss/crossentropy": 2.1454135179519653,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2634875178337097,
+      "step": 4922
+    },
+    {
+      "epoch": 0.09848,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.05164286295572917,
+      "learning_rate": 0.0001,
+      "loss": 4.5775,
+      "loss/crossentropy": 1.9837967157363892,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23980768024921417,
+      "step": 4924
+    },
+    {
+      "epoch": 0.09852,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.054032389322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.6888,
+      "loss/crossentropy": 2.099667489528656,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23482084274291992,
+      "step": 4926
+    },
+    {
+      "epoch": 0.09856,
+      "grad_norm": 3.125,
+      "grad_norm_var": 0.08787333170572917,
+      "learning_rate": 0.0001,
+      "loss": 4.5685,
+      "loss/crossentropy": 1.8467332124710083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.256004735827446,
+      "step": 4928
+    },
+    {
+      "epoch": 0.0986,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.08548075358072917,
+      "learning_rate": 0.0001,
+      "loss": 4.4474,
+      "loss/crossentropy": 2.036003887653351,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2377806007862091,
+      "step": 4930
+    },
+    {
+      "epoch": 0.09864,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.07834370930989583,
+      "learning_rate": 0.0001,
+      "loss": 4.8258,
+      "loss/crossentropy": 2.091457724571228,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24746537953615189,
+      "step": 4932
+    },
+    {
+      "epoch": 0.09868,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.07929280598958334,
+      "learning_rate": 0.0001,
+      "loss": 4.6583,
+      "loss/crossentropy": 2.245227336883545,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24141517281532288,
+      "step": 4934
+    },
+    {
+      "epoch": 0.09872,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.0661285400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.6514,
+      "loss/crossentropy": 2.295682668685913,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28035247325897217,
+      "step": 4936
+    },
+    {
+      "epoch": 0.09876,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.05524800618489583,
+      "learning_rate": 0.0001,
+      "loss": 4.5525,
+      "loss/crossentropy": 1.9251704812049866,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23609354346990585,
+      "step": 4938
+    },
+    {
+      "epoch": 0.0988,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.048314412434895836,
+      "learning_rate": 0.0001,
+      "loss": 5.1583,
+      "loss/crossentropy": 2.289652466773987,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30698196589946747,
+      "step": 4940
+    },
+    {
+      "epoch": 0.09884,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0534820556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.7759,
+      "loss/crossentropy": 2.3339043855667114,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2449272722005844,
+      "step": 4942
+    },
+    {
+      "epoch": 0.09888,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.01949462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.6678,
+      "loss/crossentropy": 2.0610195994377136,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2553517669439316,
+      "step": 4944
+    },
+    {
+      "epoch": 0.09892,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.018122355143229168,
+      "learning_rate": 0.0001,
+      "loss": 4.5849,
+      "loss/crossentropy": 2.1093825101852417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24767974764108658,
+      "step": 4946
+    },
+    {
+      "epoch": 0.09896,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.01416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3631,
+      "loss/crossentropy": 2.1742878556251526,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24640457332134247,
+      "step": 4948
+    },
+    {
+      "epoch": 0.099,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.0144195556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.8319,
+      "loss/crossentropy": 2.4237486124038696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2610916793346405,
+      "step": 4950
+    },
+    {
+      "epoch": 0.09904,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.015445963541666666,
+      "learning_rate": 0.0001,
+      "loss": 4.3397,
+      "loss/crossentropy": 2.197754144668579,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2275979220867157,
+      "step": 4952
+    },
+    {
+      "epoch": 0.09908,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.01529541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.4533,
+      "loss/crossentropy": 2.267225503921509,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23391032963991165,
+      "step": 4954
+    },
+    {
+      "epoch": 0.09912,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.009504191080729167,
+      "learning_rate": 0.0001,
+      "loss": 4.47,
+      "loss/crossentropy": 2.04893159866333,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23763196170330048,
+      "step": 4956
+    },
+    {
+      "epoch": 0.09916,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.010716756184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.97,
+      "loss/crossentropy": 2.4489223957061768,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27698180079460144,
+      "step": 4958
+    },
+    {
+      "epoch": 0.0992,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.011930338541666667,
+      "learning_rate": 0.0001,
+      "loss": 4.8604,
+      "loss/crossentropy": 2.3654375076293945,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2495984137058258,
+      "step": 4960
+    },
+    {
+      "epoch": 0.09924,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.011812337239583333,
+      "learning_rate": 0.0001,
+      "loss": 4.7671,
+      "loss/crossentropy": 1.8583308458328247,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21832667291164398,
+      "step": 4962
+    },
+    {
+      "epoch": 0.09928,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.01177978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4951,
+      "loss/crossentropy": 2.0762425661087036,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2280900850892067,
+      "step": 4964
+    },
+    {
+      "epoch": 0.09932,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.012532552083333334,
+      "learning_rate": 0.0001,
+      "loss": 4.7461,
+      "loss/crossentropy": 2.228640913963318,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26595622301101685,
+      "step": 4966
+    },
+    {
+      "epoch": 0.09936,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0111236572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.521,
+      "loss/crossentropy": 2.1447466611862183,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2542327791452408,
+      "step": 4968
+    },
+    {
+      "epoch": 0.0994,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.01226806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3959,
+      "loss/crossentropy": 2.0750887989997864,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2362435683608055,
+      "step": 4970
+    },
+    {
+      "epoch": 0.09944,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.009666951497395833,
+      "learning_rate": 0.0001,
+      "loss": 4.6705,
+      "loss/crossentropy": 1.9413353204727173,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24418477714061737,
+      "step": 4972
+    },
+    {
+      "epoch": 0.09948,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.00904541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.609,
+      "loss/crossentropy": 2.102766752243042,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24170882254838943,
+      "step": 4974
+    },
+    {
+      "epoch": 0.09952,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.0090240478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.7568,
+      "loss/crossentropy": 2.431061267852783,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2855495512485504,
+      "step": 4976
+    },
+    {
+      "epoch": 0.09956,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.010448201497395834,
+      "learning_rate": 0.0001,
+      "loss": 4.7449,
+      "loss/crossentropy": 1.9656312465667725,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26410341262817383,
+      "step": 4978
+    },
+    {
+      "epoch": 0.0996,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.01627197265625,
+      "learning_rate": 0.0001,
+      "loss": 5.0286,
+      "loss/crossentropy": 2.2365923523902893,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27946531772613525,
+      "step": 4980
+    },
+    {
+      "epoch": 0.09964,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.015184529622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.5791,
+      "loss/crossentropy": 2.203595757484436,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25247688591480255,
+      "step": 4982
+    },
+    {
+      "epoch": 0.09968,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.0163482666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.6824,
+      "loss/crossentropy": 2.1462446451187134,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2539386674761772,
+      "step": 4984
+    },
+    {
+      "epoch": 0.09972,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.014127604166666667,
+      "learning_rate": 0.0001,
+      "loss": 4.7121,
+      "loss/crossentropy": 2.4518587589263916,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24366125464439392,
+      "step": 4986
+    },
+    {
+      "epoch": 0.09976,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.01402587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5131,
+      "loss/crossentropy": 2.003869950771332,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22874485701322556,
+      "step": 4988
+    },
+    {
+      "epoch": 0.0998,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.017476399739583332,
+      "learning_rate": 0.0001,
+      "loss": 4.3221,
+      "loss/crossentropy": 2.0251912474632263,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2240670546889305,
+      "step": 4990
+    },
+    {
+      "epoch": 0.09984,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.020166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.5297,
+      "loss/crossentropy": 2.199779748916626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22916750609874725,
+      "step": 4992
+    },
+    {
+      "epoch": 0.09988,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.0207672119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.8066,
+      "loss/crossentropy": 2.3852288722991943,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26489999890327454,
+      "step": 4994
+    },
+    {
+      "epoch": 0.09992,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.03874409993489583,
+      "learning_rate": 0.0001,
+      "loss": 4.7245,
+      "loss/crossentropy": 1.9446094632148743,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24112944304943085,
+      "step": 4996
+    },
+    {
+      "epoch": 0.09996,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.04269205729166667,
+      "learning_rate": 0.0001,
+      "loss": 4.7063,
+      "loss/crossentropy": 2.585115671157837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26535044610500336,
+      "step": 4998
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 1.14400634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.6848,
+      "loss/crossentropy": 1.9871427416801453,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24747569859027863,
+      "step": 5000
+    },
+    {
+      "epoch": 0.10004,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 1.1424967447916667,
+      "learning_rate": 0.0001,
+      "loss": 4.6058,
+      "loss/crossentropy": 1.8981972336769104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24131165444850922,
+      "step": 5002
+    },
+    {
+      "epoch": 0.10008,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 1.1522786458333334,
+      "learning_rate": 0.0001,
+      "loss": 4.5731,
+      "loss/crossentropy": 2.323825240135193,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2584778293967247,
+      "step": 5004
+    },
+    {
+      "epoch": 0.10012,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 1.1363118489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.7065,
+      "loss/crossentropy": 1.728028118610382,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2997850477695465,
+      "step": 5006
+    },
+    {
+      "epoch": 0.10016,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 1.12437744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.8001,
+      "loss/crossentropy": 2.1486289501190186,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25335805118083954,
+      "step": 5008
+    },
+    {
+      "epoch": 0.1002,
+      "grad_norm": 2.5,
+      "grad_norm_var": 1.1099761962890624,
+      "learning_rate": 0.0001,
+      "loss": 4.936,
+      "loss/crossentropy": 2.3132145404815674,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26033517718315125,
+      "step": 5010
+    },
+    {
+      "epoch": 0.10024,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 1.1134928385416667,
+      "learning_rate": 0.0001,
+      "loss": 4.493,
+      "loss/crossentropy": 1.9233656525611877,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2471245899796486,
+      "step": 5012
+    },
+    {
+      "epoch": 0.10028,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 1.1301747639973958,
+      "learning_rate": 0.0001,
+      "loss": 4.5221,
+      "loss/crossentropy": 1.9435511827468872,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24730819463729858,
+      "step": 5014
+    },
+    {
+      "epoch": 0.10032,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.018928019205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3895,
+      "loss/crossentropy": 2.2031294107437134,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23213820159435272,
+      "step": 5016
+    },
+    {
+      "epoch": 0.10036,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.0276763916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.7247,
+      "loss/crossentropy": 2.285850405693054,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25809091329574585,
+      "step": 5018
+    },
+    {
+      "epoch": 0.1004,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.025211588541666666,
+      "learning_rate": 0.0001,
+      "loss": 4.7697,
+      "loss/crossentropy": 1.8660435676574707,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2436714917421341,
+      "step": 5020
+    },
+    {
+      "epoch": 0.10044,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.025877888997395834,
+      "learning_rate": 0.0001,
+      "loss": 4.3716,
+      "loss/crossentropy": 2.0659420490264893,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24248096346855164,
+      "step": 5022
+    },
+    {
+      "epoch": 0.10048,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.030516560872395834,
+      "learning_rate": 0.0001,
+      "loss": 4.7093,
+      "loss/crossentropy": 2.213133215904236,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2423061951994896,
+      "step": 5024
+    },
+    {
+      "epoch": 0.10052,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0313385009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.5288,
+      "loss/crossentropy": 2.3052343130111694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24757324904203415,
+      "step": 5026
+    },
+    {
+      "epoch": 0.10056,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.039094034830729166,
+      "learning_rate": 0.0001,
+      "loss": 4.8709,
+      "loss/crossentropy": 2.226990580558777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25892695784568787,
+      "step": 5028
+    },
+    {
+      "epoch": 0.1006,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.04163004557291667,
+      "learning_rate": 0.0001,
+      "loss": 4.9444,
+      "loss/crossentropy": 2.3460742235183716,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27760128676891327,
+      "step": 5030
+    },
+    {
+      "epoch": 0.10064,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.04345703125,
+      "learning_rate": 0.0001,
+      "loss": 4.3597,
+      "loss/crossentropy": 1.9782095551490784,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2349972277879715,
+      "step": 5032
+    },
+    {
+      "epoch": 0.10068,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.06148681640625,
+      "learning_rate": 0.0001,
+      "loss": 4.5602,
+      "loss/crossentropy": 1.847929298877716,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22401423752307892,
+      "step": 5034
+    },
+    {
+      "epoch": 0.10072,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.06073811848958333,
+      "learning_rate": 0.0001,
+      "loss": 4.4989,
+      "loss/crossentropy": 2.172071158885956,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2656550332903862,
+      "step": 5036
+    },
+    {
+      "epoch": 0.10076,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.07026265462239584,
+      "learning_rate": 0.0001,
+      "loss": 4.3892,
+      "loss/crossentropy": 2.3497499227523804,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26085225492715836,
+      "step": 5038
+    },
+    {
+      "epoch": 0.1008,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.07284749348958333,
+      "learning_rate": 0.0001,
+      "loss": 4.2583,
+      "loss/crossentropy": 2.0916348695755005,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21129868924617767,
+      "step": 5040
+    },
+    {
+      "epoch": 0.10084,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.07073160807291666,
+      "learning_rate": 0.0001,
+      "loss": 4.8931,
+      "loss/crossentropy": 2.243077278137207,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2657035142183304,
+      "step": 5042
+    },
+    {
+      "epoch": 0.10088,
+      "grad_norm": 10.0,
+      "grad_norm_var": 3.600194295247396,
+      "learning_rate": 0.0001,
+      "loss": 4.8887,
+      "loss/crossentropy": 1.9361066222190857,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.41255413740873337,
+      "step": 5044
+    },
+    {
+      "epoch": 0.10092,
+      "grad_norm": 3.703125,
+      "grad_norm_var": 3.6162760416666666,
+      "learning_rate": 0.0001,
+      "loss": 4.7632,
+      "loss/crossentropy": 2.0139313340187073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2591235190629959,
+      "step": 5046
+    },
+    {
+      "epoch": 0.10096,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 3.589704386393229,
+      "learning_rate": 0.0001,
+      "loss": 4.6642,
+      "loss/crossentropy": 2.335710287094116,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26457205414772034,
+      "step": 5048
+    },
+    {
+      "epoch": 0.101,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 3.604325358072917,
+      "learning_rate": 0.0001,
+      "loss": 4.3329,
+      "loss/crossentropy": 1.9269848465919495,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23867928236722946,
+      "step": 5050
+    },
+    {
+      "epoch": 0.10104,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 3.627415974934896,
+      "learning_rate": 0.0001,
+      "loss": 4.4982,
+      "loss/crossentropy": 1.9732608795166016,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24961821734905243,
+      "step": 5052
+    },
+    {
+      "epoch": 0.10108,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 3.6287506103515623,
+      "learning_rate": 0.0001,
+      "loss": 4.7218,
+      "loss/crossentropy": 2.2659696340560913,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24580512940883636,
+      "step": 5054
+    },
+    {
+      "epoch": 0.10112,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 3.57880859375,
+      "learning_rate": 0.0001,
+      "loss": 4.4318,
+      "loss/crossentropy": 1.6600720882415771,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20283473283052444,
+      "step": 5056
+    },
+    {
+      "epoch": 0.10116,
+      "grad_norm": 2.375,
+      "grad_norm_var": 3.602855428059896,
+      "learning_rate": 0.0001,
+      "loss": 4.8249,
+      "loss/crossentropy": 2.0175185799598694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23329314589500427,
+      "step": 5058
+    },
+    {
+      "epoch": 0.1012,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.146484375,
+      "learning_rate": 0.0001,
+      "loss": 4.5132,
+      "loss/crossentropy": 2.259668231010437,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2692077234387398,
+      "step": 5060
+    },
+    {
+      "epoch": 0.10124,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.02144775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.6392,
+      "loss/crossentropy": 2.260953903198242,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28225430846214294,
+      "step": 5062
+    },
+    {
+      "epoch": 0.10128,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.021686808268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.3624,
+      "loss/crossentropy": 1.9721892476081848,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2336483597755432,
+      "step": 5064
+    },
+    {
+      "epoch": 0.10132,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.021512858072916665,
+      "learning_rate": 0.0001,
+      "loss": 4.6257,
+      "loss/crossentropy": 2.188947319984436,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23959602415561676,
+      "step": 5066
+    },
+    {
+      "epoch": 0.10136,
+      "grad_norm": 3.078125,
+      "grad_norm_var": 0.04843343098958333,
+      "learning_rate": 0.0001,
+      "loss": 4.5601,
+      "loss/crossentropy": 1.7914190292358398,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2234276980161667,
+      "step": 5068
+    },
+    {
+      "epoch": 0.1014,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.049925740559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.5806,
+      "loss/crossentropy": 2.1124663949012756,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22523616254329681,
+      "step": 5070
+    },
+    {
+      "epoch": 0.10144,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.03764546712239583,
+      "learning_rate": 0.0001,
+      "loss": 4.7098,
+      "loss/crossentropy": 2.1146361231803894,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22483345866203308,
+      "step": 5072
+    },
+    {
+      "epoch": 0.10148,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.04121805826822917,
+      "learning_rate": 0.0001,
+      "loss": 4.5076,
+      "loss/crossentropy": 2.335755705833435,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.256914846599102,
+      "step": 5074
+    },
+    {
+      "epoch": 0.10152,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.04346415201822917,
+      "learning_rate": 0.0001,
+      "loss": 4.8665,
+      "loss/crossentropy": 2.2911819219589233,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26038119196891785,
+      "step": 5076
+    },
+    {
+      "epoch": 0.10156,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.047587076822916664,
+      "learning_rate": 0.0001,
+      "loss": 4.592,
+      "loss/crossentropy": 2.059769034385681,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24004006385803223,
+      "step": 5078
+    },
+    {
+      "epoch": 0.1016,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.04888916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.5028,
+      "loss/crossentropy": 2.141201138496399,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22454539686441422,
+      "step": 5080
+    },
+    {
+      "epoch": 0.10164,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.050146484375,
+      "learning_rate": 0.0001,
+      "loss": 4.7527,
+      "loss/crossentropy": 1.9538633823394775,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24477297067642212,
+      "step": 5082
+    },
+    {
+      "epoch": 0.10168,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.00992431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4799,
+      "loss/crossentropy": 2.1555078625679016,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2425938919186592,
+      "step": 5084
+    },
+    {
+      "epoch": 0.10172,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.009235636393229166,
+      "learning_rate": 0.0001,
+      "loss": 4.5583,
+      "loss/crossentropy": 2.2306214570999146,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2445632517337799,
+      "step": 5086
+    },
+    {
+      "epoch": 0.10176,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.010179646809895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1075,
+      "loss/crossentropy": 1.9713392853736877,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22239256650209427,
+      "step": 5088
+    },
+    {
+      "epoch": 0.1018,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.011799112955729166,
+      "learning_rate": 0.0001,
+      "loss": 4.5181,
+      "loss/crossentropy": 1.951128602027893,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24310563504695892,
+      "step": 5090
+    },
+    {
+      "epoch": 0.10184,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.011051432291666666,
+      "learning_rate": 0.0001,
+      "loss": 4.3617,
+      "loss/crossentropy": 2.0100057125091553,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23054375499486923,
+      "step": 5092
+    },
+    {
+      "epoch": 0.10188,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.011042277018229166,
+      "learning_rate": 0.0001,
+      "loss": 4.4573,
+      "loss/crossentropy": 2.1898789405822754,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24956130981445312,
+      "step": 5094
+    },
+    {
+      "epoch": 0.10192,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.021512858072916665,
+      "learning_rate": 0.0001,
+      "loss": 4.6576,
+      "loss/crossentropy": 1.5666239857673645,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20010025054216385,
+      "step": 5096
+    },
+    {
+      "epoch": 0.10196,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.022468058268229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2762,
+      "loss/crossentropy": 1.884951651096344,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2484818547964096,
+      "step": 5098
+    },
+    {
+      "epoch": 0.102,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.022554524739583335,
+      "learning_rate": 0.0001,
+      "loss": 4.5207,
+      "loss/crossentropy": 1.976080298423767,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23380043357610703,
+      "step": 5100
+    },
+    {
+      "epoch": 0.10204,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.0229156494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.592,
+      "loss/crossentropy": 2.1262658834457397,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2441401332616806,
+      "step": 5102
+    },
+    {
+      "epoch": 0.10208,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.08055013020833333,
+      "learning_rate": 0.0001,
+      "loss": 4.2735,
+      "loss/crossentropy": 1.7588757276535034,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2101762592792511,
+      "step": 5104
+    },
+    {
+      "epoch": 0.10212,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.08059488932291667,
+      "learning_rate": 0.0001,
+      "loss": 4.6721,
+      "loss/crossentropy": 2.292783260345459,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26928654313087463,
+      "step": 5106
+    },
+    {
+      "epoch": 0.10216,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.4834218343098958,
+      "learning_rate": 0.0001,
+      "loss": 4.7475,
+      "loss/crossentropy": 2.0670888423919678,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2644767463207245,
+      "step": 5108
+    },
+    {
+      "epoch": 0.1022,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.4729563395182292,
+      "learning_rate": 0.0001,
+      "loss": 4.4113,
+      "loss/crossentropy": 2.0522598028182983,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2410094290971756,
+      "step": 5110
+    },
+    {
+      "epoch": 0.10224,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.4729075113932292,
+      "learning_rate": 0.0001,
+      "loss": 4.4914,
+      "loss/crossentropy": 2.0756974816322327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27077721059322357,
+      "step": 5112
+    },
+    {
+      "epoch": 0.10228,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.47226155598958336,
+      "learning_rate": 0.0001,
+      "loss": 4.6524,
+      "loss/crossentropy": 2.1569767594337463,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23582034558057785,
+      "step": 5114
+    },
+    {
+      "epoch": 0.10232,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.4847819010416667,
+      "learning_rate": 0.0001,
+      "loss": 4.2821,
+      "loss/crossentropy": 1.9736077785491943,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23984025418758392,
+      "step": 5116
+    },
+    {
+      "epoch": 0.10236,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.4942698160807292,
+      "learning_rate": 0.0001,
+      "loss": 4.3047,
+      "loss/crossentropy": 2.1400066614151,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2347380667924881,
+      "step": 5118
+    },
+    {
+      "epoch": 0.1024,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.46923726399739585,
+      "learning_rate": 0.0001,
+      "loss": 4.3531,
+      "loss/crossentropy": 1.989999234676361,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22508147358894348,
+      "step": 5120
+    },
+    {
+      "epoch": 0.10244,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.4704498291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.8817,
+      "loss/crossentropy": 2.132390856742859,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3074522316455841,
+      "step": 5122
+    },
+    {
+      "epoch": 0.10248,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.011002604166666667,
+      "learning_rate": 0.0001,
+      "loss": 4.3606,
+      "loss/crossentropy": 1.7906856536865234,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2126675397157669,
+      "step": 5124
+    },
+    {
+      "epoch": 0.10252,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0112457275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.7045,
+      "loss/crossentropy": 2.0576369762420654,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27211636304855347,
+      "step": 5126
+    },
+    {
+      "epoch": 0.10256,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.01099853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5402,
+      "loss/crossentropy": 2.1174184679985046,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2344469577074051,
+      "step": 5128
+    },
+    {
+      "epoch": 0.1026,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.0090972900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.7227,
+      "loss/crossentropy": 1.9139717817306519,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2362786829471588,
+      "step": 5130
+    },
+    {
+      "epoch": 0.10264,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.008495076497395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4801,
+      "loss/crossentropy": 2.022357940673828,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2307513877749443,
+      "step": 5132
+    },
+    {
+      "epoch": 0.10268,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.007721964518229167,
+      "learning_rate": 0.0001,
+      "loss": 4.3963,
+      "loss/crossentropy": 2.038477897644043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22871223092079163,
+      "step": 5134
+    },
+    {
+      "epoch": 0.10272,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.007079060872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.7283,
+      "loss/crossentropy": 2.0895442962646484,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23719681799411774,
+      "step": 5136
+    },
+    {
+      "epoch": 0.10276,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.008820597330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.7059,
+      "loss/crossentropy": 2.17978835105896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2519204765558243,
+      "step": 5138
+    },
+    {
+      "epoch": 0.1028,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.00953369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.7318,
+      "loss/crossentropy": 2.19089937210083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24376338720321655,
+      "step": 5140
+    },
+    {
+      "epoch": 0.10284,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.011002604166666667,
+      "learning_rate": 0.0001,
+      "loss": 4.3961,
+      "loss/crossentropy": 2.018259823322296,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23707401752471924,
+      "step": 5142
+    },
+    {
+      "epoch": 0.10288,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.012287394205729166,
+      "learning_rate": 0.0001,
+      "loss": 4.7556,
+      "loss/crossentropy": 2.110401153564453,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23409561812877655,
+      "step": 5144
+    },
+    {
+      "epoch": 0.10292,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.012239583333333333,
+      "learning_rate": 0.0001,
+      "loss": 4.615,
+      "loss/crossentropy": 2.2096832990646362,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2580249160528183,
+      "step": 5146
+    },
+    {
+      "epoch": 0.10296,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.015803019205729168,
+      "learning_rate": 0.0001,
+      "loss": 4.5218,
+      "loss/crossentropy": 2.1825822591781616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23592937737703323,
+      "step": 5148
+    },
+    {
+      "epoch": 0.103,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.01451416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.6945,
+      "loss/crossentropy": 2.2440234422683716,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23741164803504944,
+      "step": 5150
+    },
+    {
+      "epoch": 0.10304,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.015071614583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.962,
+      "loss/crossentropy": 2.3818799257278442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2642097622156143,
+      "step": 5152
+    },
+    {
+      "epoch": 0.10308,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.014378865559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.5706,
+      "loss/crossentropy": 2.1905806064605713,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26945509016513824,
+      "step": 5154
+    },
+    {
+      "epoch": 0.10312,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.017671712239583335,
+      "learning_rate": 0.0001,
+      "loss": 4.5678,
+      "loss/crossentropy": 2.096913695335388,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2540033459663391,
+      "step": 5156
+    },
+    {
+      "epoch": 0.10316,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.015721638997395832,
+      "learning_rate": 0.0001,
+      "loss": 4.5286,
+      "loss/crossentropy": 1.7916489243507385,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22504088282585144,
+      "step": 5158
+    },
+    {
+      "epoch": 0.1032,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.014240519205729166,
+      "learning_rate": 0.0001,
+      "loss": 4.6256,
+      "loss/crossentropy": 1.9366079568862915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25474467873573303,
+      "step": 5160
+    },
+    {
+      "epoch": 0.10324,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.014533487955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.5054,
+      "loss/crossentropy": 2.0233771800994873,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23394355177879333,
+      "step": 5162
+    },
+    {
+      "epoch": 0.10328,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.010693359375,
+      "learning_rate": 0.0001,
+      "loss": 4.7803,
+      "loss/crossentropy": 2.442312717437744,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28390438854694366,
+      "step": 5164
+    },
+    {
+      "epoch": 0.10332,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.014435831705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.7849,
+      "loss/crossentropy": 1.9547526836395264,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2298036813735962,
+      "step": 5166
+    },
+    {
+      "epoch": 0.10336,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.018766276041666665,
+      "learning_rate": 0.0001,
+      "loss": 4.3225,
+      "loss/crossentropy": 1.7974739074707031,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21959447860717773,
+      "step": 5168
+    },
+    {
+      "epoch": 0.1034,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.020075480143229168,
+      "learning_rate": 0.0001,
+      "loss": 4.5438,
+      "loss/crossentropy": 1.9860564470291138,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24661653488874435,
+      "step": 5170
+    },
+    {
+      "epoch": 0.10344,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.019050089518229167,
+      "learning_rate": 0.0001,
+      "loss": 4.5084,
+      "loss/crossentropy": 1.6198940873146057,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20592768490314484,
+      "step": 5172
+    },
+    {
+      "epoch": 0.10348,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.019554646809895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4831,
+      "loss/crossentropy": 2.193474531173706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24437790364027023,
+      "step": 5174
+    },
+    {
+      "epoch": 0.10352,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.022557576497395832,
+      "learning_rate": 0.0001,
+      "loss": 4.9209,
+      "loss/crossentropy": 2.221992254257202,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2931511402130127,
+      "step": 5176
+    },
+    {
+      "epoch": 0.10356,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.022272745768229168,
+      "learning_rate": 0.0001,
+      "loss": 4.4611,
+      "loss/crossentropy": 2.006419837474823,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2335921749472618,
+      "step": 5178
+    },
+    {
+      "epoch": 0.1036,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.022175089518229166,
+      "learning_rate": 0.0001,
+      "loss": 4.4477,
+      "loss/crossentropy": 2.2861804962158203,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2549041658639908,
+      "step": 5180
+    },
+    {
+      "epoch": 0.10364,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.014378865559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.4079,
+      "loss/crossentropy": 2.217998743057251,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24741299450397491,
+      "step": 5182
+    },
+    {
+      "epoch": 0.10368,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.0123931884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.7344,
+      "loss/crossentropy": 2.1875526905059814,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23636415600776672,
+      "step": 5184
+    },
+    {
+      "epoch": 0.10372,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.014850870768229166,
+      "learning_rate": 0.0001,
+      "loss": 4.4629,
+      "loss/crossentropy": 1.8408135175704956,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23278112709522247,
+      "step": 5186
+    },
+    {
+      "epoch": 0.10376,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.014188639322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.6848,
+      "loss/crossentropy": 1.7936646342277527,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23859456181526184,
+      "step": 5188
+    },
+    {
+      "epoch": 0.1038,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.014058430989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.3825,
+      "loss/crossentropy": 2.0800318717956543,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23155340552330017,
+      "step": 5190
+    },
+    {
+      "epoch": 0.10384,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0066802978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.9645,
+      "loss/crossentropy": 2.277778387069702,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25029345601797104,
+      "step": 5192
+    },
+    {
+      "epoch": 0.10388,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0081695556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2486,
+      "loss/crossentropy": 1.9658478498458862,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22655482590198517,
+      "step": 5194
+    },
+    {
+      "epoch": 0.10392,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.01060791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.7503,
+      "loss/crossentropy": 2.214509129524231,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.254493810236454,
+      "step": 5196
+    },
+    {
+      "epoch": 0.10396,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0102203369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3648,
+      "loss/crossentropy": 1.9465742707252502,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23288530111312866,
+      "step": 5198
+    },
+    {
+      "epoch": 0.104,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.010277303059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.3216,
+      "loss/crossentropy": 2.062779188156128,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21423730999231339,
+      "step": 5200
+    },
+    {
+      "epoch": 0.10404,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.010350545247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1645,
+      "loss/crossentropy": 1.777470588684082,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22998665273189545,
+      "step": 5202
+    },
+    {
+      "epoch": 0.10408,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.05396728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.8979,
+      "loss/crossentropy": 2.2505980730056763,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30352361500263214,
+      "step": 5204
+    },
+    {
+      "epoch": 0.10412,
+      "grad_norm": 3.359375,
+      "grad_norm_var": 0.12148335774739584,
+      "learning_rate": 0.0001,
+      "loss": 4.4093,
+      "loss/crossentropy": 1.8989517092704773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2210940569639206,
+      "step": 5206
+    },
+    {
+      "epoch": 0.10416,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.12923177083333334,
+      "learning_rate": 0.0001,
+      "loss": 4.5535,
+      "loss/crossentropy": 2.378798723220825,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25575730204582214,
+      "step": 5208
+    },
+    {
+      "epoch": 0.1042,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.11901041666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.6763,
+      "loss/crossentropy": 1.7642006278038025,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22105304896831512,
+      "step": 5210
+    },
+    {
+      "epoch": 0.10424,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.11607666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.8945,
+      "loss/crossentropy": 2.188746988773346,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24192160367965698,
+      "step": 5212
+    },
+    {
+      "epoch": 0.10428,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.11298421223958334,
+      "learning_rate": 0.0001,
+      "loss": 4.6001,
+      "loss/crossentropy": 2.116630494594574,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2879187613725662,
+      "step": 5214
+    },
+    {
+      "epoch": 0.10432,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.11013895670572917,
+      "learning_rate": 0.0001,
+      "loss": 4.4932,
+      "loss/crossentropy": 1.8329599499702454,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21425092220306396,
+      "step": 5216
+    },
+    {
+      "epoch": 0.10436,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.1064849853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3966,
+      "loss/crossentropy": 2.1063259840011597,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23704984784126282,
+      "step": 5218
+    },
+    {
+      "epoch": 0.1044,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.08238525390625,
+      "learning_rate": 0.0001,
+      "loss": 4.4821,
+      "loss/crossentropy": 1.7994996309280396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21408653259277344,
+      "step": 5220
+    },
+    {
+      "epoch": 0.10444,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.0281158447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2437,
+      "loss/crossentropy": 1.880006492137909,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24362730979919434,
+      "step": 5222
+    },
+    {
+      "epoch": 0.10448,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.017853800455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2726,
+      "loss/crossentropy": 2.010268449783325,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22580985724925995,
+      "step": 5224
+    },
+    {
+      "epoch": 0.10452,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.0177886962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.6839,
+      "loss/crossentropy": 2.057162046432495,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23878254741430283,
+      "step": 5226
+    },
+    {
+      "epoch": 0.10456,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.019603474934895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3774,
+      "loss/crossentropy": 2.0615930557250977,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22890077531337738,
+      "step": 5228
+    },
+    {
+      "epoch": 0.1046,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.020817057291666666,
+      "learning_rate": 0.0001,
+      "loss": 4.7693,
+      "loss/crossentropy": 1.9013472199440002,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22539222240447998,
+      "step": 5230
+    },
+    {
+      "epoch": 0.10464,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.019482421875,
+      "learning_rate": 0.0001,
+      "loss": 4.5531,
+      "loss/crossentropy": 2.239185929298401,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24706681817770004,
+      "step": 5232
+    },
+    {
+      "epoch": 0.10468,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.020540364583333335,
+      "learning_rate": 0.0001,
+      "loss": 4.3902,
+      "loss/crossentropy": 1.9881523251533508,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24434638023376465,
+      "step": 5234
+    },
+    {
+      "epoch": 0.10472,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.020524088541666666,
+      "learning_rate": 0.0001,
+      "loss": 4.7908,
+      "loss/crossentropy": 1.9529212713241577,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2321469634771347,
+      "step": 5236
+    },
+    {
+      "epoch": 0.10476,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.01041259765625,
+      "learning_rate": 0.0001,
+      "loss": 4.8045,
+      "loss/crossentropy": 2.196424722671509,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24054434895515442,
+      "step": 5238
+    },
+    {
+      "epoch": 0.1048,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.009781901041666667,
+      "learning_rate": 0.0001,
+      "loss": 4.5895,
+      "loss/crossentropy": 2.082987070083618,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2760557308793068,
+      "step": 5240
+    },
+    {
+      "epoch": 0.10484,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.011652628580729166,
+      "learning_rate": 0.0001,
+      "loss": 4.5961,
+      "loss/crossentropy": 2.2369720935821533,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25305214524269104,
+      "step": 5242
+    },
+    {
+      "epoch": 0.10488,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0119537353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.7268,
+      "loss/crossentropy": 2.3372031450271606,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26765232533216476,
+      "step": 5244
+    },
+    {
+      "epoch": 0.10492,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.0113677978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.6148,
+      "loss/crossentropy": 2.23944628238678,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26508912444114685,
+      "step": 5246
+    },
+    {
+      "epoch": 0.10496,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.012105305989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.3453,
+      "loss/crossentropy": 2.2701858282089233,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2551077604293823,
+      "step": 5248
+    },
+    {
+      "epoch": 0.105,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.010887654622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.6455,
+      "loss/crossentropy": 2.293464779853821,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25016437470912933,
+      "step": 5250
+    },
+    {
+      "epoch": 0.10504,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.0109527587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5798,
+      "loss/crossentropy": 2.3072171211242676,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27824972569942474,
+      "step": 5252
+    },
+    {
+      "epoch": 0.10508,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.010498046875,
+      "learning_rate": 0.0001,
+      "loss": 4.5948,
+      "loss/crossentropy": 1.9855756759643555,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2312464788556099,
+      "step": 5254
+    },
+    {
+      "epoch": 0.10512,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.009129842122395834,
+      "learning_rate": 0.0001,
+      "loss": 4.8104,
+      "loss/crossentropy": 1.9584077596664429,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2386016771197319,
+      "step": 5256
+    },
+    {
+      "epoch": 0.10516,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.010741170247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.6103,
+      "loss/crossentropy": 2.2184669375419617,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25914129614830017,
+      "step": 5258
+    },
+    {
+      "epoch": 0.1052,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0122222900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1221,
+      "loss/crossentropy": 1.6798554062843323,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2037271112203598,
+      "step": 5260
+    },
+    {
+      "epoch": 0.10524,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.011995442708333333,
+      "learning_rate": 0.0001,
+      "loss": 4.2718,
+      "loss/crossentropy": 1.8675006031990051,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21148693561553955,
+      "step": 5262
+    },
+    {
+      "epoch": 0.10528,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.012336222330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.5136,
+      "loss/crossentropy": 1.9439855813980103,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24611759185791016,
+      "step": 5264
+    },
+    {
+      "epoch": 0.10532,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.013923136393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.5298,
+      "loss/crossentropy": 2.1550748348236084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2428945228457451,
+      "step": 5266
+    },
+    {
+      "epoch": 0.10536,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.0117584228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.6777,
+      "loss/crossentropy": 1.9924054741859436,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21866007149219513,
+      "step": 5268
+    },
+    {
+      "epoch": 0.1054,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.011571248372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.5359,
+      "loss/crossentropy": 2.0413920879364014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2579498365521431,
+      "step": 5270
+    },
+    {
+      "epoch": 0.10544,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.016380818684895833,
+      "learning_rate": 0.0001,
+      "loss": 4.6406,
+      "loss/crossentropy": 2.062632381916046,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24655035883188248,
+      "step": 5272
+    },
+    {
+      "epoch": 0.10548,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.021833292643229165,
+      "learning_rate": 0.0001,
+      "loss": 4.8133,
+      "loss/crossentropy": 2.0620386600494385,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3041655272245407,
+      "step": 5274
+    },
+    {
+      "epoch": 0.10552,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.016657511393229168,
+      "learning_rate": 0.0001,
+      "loss": 4.4775,
+      "loss/crossentropy": 1.966421365737915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22333864122629166,
+      "step": 5276
+    },
+    {
+      "epoch": 0.10556,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.017650349934895834,
+      "learning_rate": 0.0001,
+      "loss": 4.4323,
+      "loss/crossentropy": 1.9120238423347473,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22857370972633362,
+      "step": 5278
+    },
+    {
+      "epoch": 0.1056,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.019701131184895835,
+      "learning_rate": 0.0001,
+      "loss": 4.5179,
+      "loss/crossentropy": 2.084389805793762,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2556762397289276,
+      "step": 5280
+    },
+    {
+      "epoch": 0.10564,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.019758097330729165,
+      "learning_rate": 0.0001,
+      "loss": 4.4552,
+      "loss/crossentropy": 1.8707188367843628,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22955116629600525,
+      "step": 5282
+    },
+    {
+      "epoch": 0.10568,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.019775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3538,
+      "loss/crossentropy": 1.8243692517280579,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21088172495365143,
+      "step": 5284
+    },
+    {
+      "epoch": 0.10572,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.0192535400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.5046,
+      "loss/crossentropy": 2.111305356025696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2595943957567215,
+      "step": 5286
+    },
+    {
+      "epoch": 0.10576,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.015697224934895834,
+      "learning_rate": 0.0001,
+      "loss": 4.4598,
+      "loss/crossentropy": 2.3729283809661865,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2632211595773697,
+      "step": 5288
+    },
+    {
+      "epoch": 0.1058,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.007249959309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.6256,
+      "loss/crossentropy": 2.3542696237564087,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25645585358142853,
+      "step": 5290
+    },
+    {
+      "epoch": 0.10584,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.01002197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4863,
+      "loss/crossentropy": 2.0140068531036377,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23033145815134048,
+      "step": 5292
+    },
+    {
+      "epoch": 0.10588,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.0099761962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.8126,
+      "loss/crossentropy": 1.9499077796936035,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21458172798156738,
+      "step": 5294
+    },
+    {
+      "epoch": 0.10592,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0084625244140625,
+      "learning_rate": 0.0001,
+      "loss": 4.479,
+      "loss/crossentropy": 2.1434344053268433,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2552379444241524,
+      "step": 5296
+    },
+    {
+      "epoch": 0.10596,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.007819620768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.7149,
+      "loss/crossentropy": 1.9951340556144714,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21683495491743088,
+      "step": 5298
+    },
+    {
+      "epoch": 0.106,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.02330322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.7855,
+      "loss/crossentropy": 2.380235195159912,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2756577730178833,
+      "step": 5300
+    },
+    {
+      "epoch": 0.10604,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.02330322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.8596,
+      "loss/crossentropy": 2.298241972923279,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2598598450422287,
+      "step": 5302
+    },
+    {
+      "epoch": 0.10608,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.032136027018229166,
+      "learning_rate": 0.0001,
+      "loss": 4.8615,
+      "loss/crossentropy": 2.093233823776245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23809552192687988,
+      "step": 5304
+    },
+    {
+      "epoch": 0.10612,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.031493123372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.5271,
+      "loss/crossentropy": 2.0901564955711365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24874016642570496,
+      "step": 5306
+    },
+    {
+      "epoch": 0.10616,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.0304107666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3193,
+      "loss/crossentropy": 1.8029736280441284,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20973137766122818,
+      "step": 5308
+    },
+    {
+      "epoch": 0.1062,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.03316650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.4255,
+      "loss/crossentropy": 2.4068437814712524,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25310443341732025,
+      "step": 5310
+    },
+    {
+      "epoch": 0.10624,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.031029256184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4499,
+      "loss/crossentropy": 2.1125503182411194,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2455870360136032,
+      "step": 5312
+    },
+    {
+      "epoch": 0.10628,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0323150634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4752,
+      "loss/crossentropy": 2.0995737314224243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2418016716837883,
+      "step": 5314
+    },
+    {
+      "epoch": 0.10632,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5873,
+      "loss/crossentropy": 1.8753212690353394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24125799536705017,
+      "step": 5316
+    },
+    {
+      "epoch": 0.10636,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.021451822916666665,
+      "learning_rate": 0.0001,
+      "loss": 4.7269,
+      "loss/crossentropy": 2.0175408720970154,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23516137897968292,
+      "step": 5318
+    },
+    {
+      "epoch": 0.1064,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.00914306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4953,
+      "loss/crossentropy": 2.2671592235565186,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2356845736503601,
+      "step": 5320
+    },
+    {
+      "epoch": 0.10644,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.009468587239583333,
+      "learning_rate": 0.0001,
+      "loss": 4.5328,
+      "loss/crossentropy": 2.142452359199524,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26422591507434845,
+      "step": 5322
+    },
+    {
+      "epoch": 0.10648,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.007225545247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4328,
+      "loss/crossentropy": 1.9664896726608276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25856246054172516,
+      "step": 5324
+    },
+    {
+      "epoch": 0.10652,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.007222493489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.5531,
+      "loss/crossentropy": 2.168110191822052,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23274105042219162,
+      "step": 5326
+    },
+    {
+      "epoch": 0.10656,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.00552978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5242,
+      "loss/crossentropy": 2.006514251232147,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2532104551792145,
+      "step": 5328
+    },
+    {
+      "epoch": 0.1066,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.0070220947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.5593,
+      "loss/crossentropy": 2.462701439857483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2771739661693573,
+      "step": 5330
+    },
+    {
+      "epoch": 0.10664,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0076324462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4076,
+      "loss/crossentropy": 2.0889209508895874,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2479858472943306,
+      "step": 5332
+    },
+    {
+      "epoch": 0.10668,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.008356730143229166,
+      "learning_rate": 0.0001,
+      "loss": 4.329,
+      "loss/crossentropy": 1.8056100606918335,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22194840013980865,
+      "step": 5334
+    },
+    {
+      "epoch": 0.10672,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0075185139973958336,
+      "learning_rate": 0.0001,
+      "loss": 4.703,
+      "loss/crossentropy": 2.32460880279541,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28185322880744934,
+      "step": 5336
+    },
+    {
+      "epoch": 0.10676,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0075185139973958336,
+      "learning_rate": 0.0001,
+      "loss": 4.6388,
+      "loss/crossentropy": 2.238978862762451,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24314726889133453,
+      "step": 5338
+    },
+    {
+      "epoch": 0.1068,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.00914306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4161,
+      "loss/crossentropy": 1.8914734721183777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20767460763454437,
+      "step": 5340
+    },
+    {
+      "epoch": 0.10684,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.010445149739583333,
+      "learning_rate": 0.0001,
+      "loss": 4.5628,
+      "loss/crossentropy": 1.9704068899154663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23617815226316452,
+      "step": 5342
+    },
+    {
+      "epoch": 0.10688,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.013374837239583333,
+      "learning_rate": 0.0001,
+      "loss": 4.214,
+      "loss/crossentropy": 1.8539690971374512,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2218247577548027,
+      "step": 5344
+    },
+    {
+      "epoch": 0.10692,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.012596638997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.6077,
+      "loss/crossentropy": 1.982038140296936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23828908801078796,
+      "step": 5346
+    },
+    {
+      "epoch": 0.10696,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.013313802083333333,
+      "learning_rate": 0.0001,
+      "loss": 4.2879,
+      "loss/crossentropy": 1.8247870802879333,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22487633675336838,
+      "step": 5348
+    },
+    {
+      "epoch": 0.107,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.011253865559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.382,
+      "loss/crossentropy": 2.0704214572906494,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24332743138074875,
+      "step": 5350
+    },
+    {
+      "epoch": 0.10704,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.013509114583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.709,
+      "loss/crossentropy": 2.037345290184021,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2515557184815407,
+      "step": 5352
+    },
+    {
+      "epoch": 0.10708,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.020702107747395834,
+      "learning_rate": 0.0001,
+      "loss": 4.5427,
+      "loss/crossentropy": 2.0561426877975464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24030926823616028,
+      "step": 5354
+    },
+    {
+      "epoch": 0.10712,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.017943318684895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3108,
+      "loss/crossentropy": 1.6871100068092346,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2526541203260422,
+      "step": 5356
+    },
+    {
+      "epoch": 0.10716,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.01978759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.5238,
+      "loss/crossentropy": 2.0133201479911804,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23663413524627686,
+      "step": 5358
+    },
+    {
+      "epoch": 0.1072,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.0170562744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.557,
+      "loss/crossentropy": 2.0627574920654297,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24395380914211273,
+      "step": 5360
+    },
+    {
+      "epoch": 0.10724,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.020929972330729168,
+      "learning_rate": 0.0001,
+      "loss": 4.5495,
+      "loss/crossentropy": 2.280818462371826,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2806383967399597,
+      "step": 5362
+    },
+    {
+      "epoch": 0.10728,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.022684733072916668,
+      "learning_rate": 0.0001,
+      "loss": 4.4743,
+      "loss/crossentropy": 2.002636671066284,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23986798524856567,
+      "step": 5364
+    },
+    {
+      "epoch": 0.10732,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.022298177083333332,
+      "learning_rate": 0.0001,
+      "loss": 4.6816,
+      "loss/crossentropy": 2.042721927165985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23199205100536346,
+      "step": 5366
+    },
+    {
+      "epoch": 0.10736,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.021540323893229168,
+      "learning_rate": 0.0001,
+      "loss": 4.3225,
+      "loss/crossentropy": 2.1047908663749695,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23987916857004166,
+      "step": 5368
+    },
+    {
+      "epoch": 0.1074,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.014774576822916666,
+      "learning_rate": 0.0001,
+      "loss": 4.4827,
+      "loss/crossentropy": 2.0513075590133667,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.242512047290802,
+      "step": 5370
+    },
+    {
+      "epoch": 0.10744,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.015913899739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.4452,
+      "loss/crossentropy": 1.9151215553283691,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23007020354270935,
+      "step": 5372
+    },
+    {
+      "epoch": 0.10748,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.019303385416666666,
+      "learning_rate": 0.0001,
+      "loss": 4.2344,
+      "loss/crossentropy": 1.8759313821792603,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2068658247590065,
+      "step": 5374
+    },
+    {
+      "epoch": 0.10752,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.020807902018229168,
+      "learning_rate": 0.0001,
+      "loss": 4.4667,
+      "loss/crossentropy": 2.1500572562217712,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24392293393611908,
+      "step": 5376
+    },
+    {
+      "epoch": 0.10756,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.018355305989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.0917,
+      "loss/crossentropy": 1.6089633703231812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20190145075321198,
+      "step": 5378
+    },
+    {
+      "epoch": 0.1076,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.014452107747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.6521,
+      "loss/crossentropy": 2.1967561841011047,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2749984338879585,
+      "step": 5380
+    },
+    {
+      "epoch": 0.10764,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.011311848958333334,
+      "learning_rate": 0.0001,
+      "loss": 4.0843,
+      "loss/crossentropy": 1.8293656706809998,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20650158822536469,
+      "step": 5382
+    },
+    {
+      "epoch": 0.10768,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.011847941080729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3441,
+      "loss/crossentropy": 2.3964673280715942,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2814205437898636,
+      "step": 5384
+    },
+    {
+      "epoch": 0.10772,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.021842447916666667,
+      "learning_rate": 0.0001,
+      "loss": 4.8872,
+      "loss/crossentropy": 2.4995274543762207,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28704003244638443,
+      "step": 5386
+    },
+    {
+      "epoch": 0.10776,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.019462076822916667,
+      "learning_rate": 0.0001,
+      "loss": 4.6066,
+      "loss/crossentropy": 2.0650060176849365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23606212437152863,
+      "step": 5388
+    },
+    {
+      "epoch": 0.1078,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.01640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3723,
+      "loss/crossentropy": 2.3049341440200806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23550046980381012,
+      "step": 5390
+    },
+    {
+      "epoch": 0.10784,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.013505045572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.813,
+      "loss/crossentropy": 2.2687143087387085,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2854095697402954,
+      "step": 5392
+    },
+    {
+      "epoch": 0.10788,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.014557902018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.6267,
+      "loss/crossentropy": 2.029325544834137,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23444775491952896,
+      "step": 5394
+    },
+    {
+      "epoch": 0.10792,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.013895670572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.5214,
+      "loss/crossentropy": 2.2012031078338623,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27986764907836914,
+      "step": 5396
+    },
+    {
+      "epoch": 0.10796,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.012626139322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.65,
+      "loss/crossentropy": 2.2396020889282227,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24839117377996445,
+      "step": 5398
+    },
+    {
+      "epoch": 0.108,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.009566243489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.4634,
+      "loss/crossentropy": 2.1481886506080627,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2545444592833519,
+      "step": 5400
+    },
+    {
+      "epoch": 0.10804,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.006005859375,
+      "learning_rate": 0.0001,
+      "loss": 4.6109,
+      "loss/crossentropy": 1.9799351692199707,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23296385258436203,
+      "step": 5402
+    },
+    {
+      "epoch": 0.10808,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.006180826822916667,
+      "learning_rate": 0.0001,
+      "loss": 4.5612,
+      "loss/crossentropy": 1.845237910747528,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2316332384943962,
+      "step": 5404
+    },
+    {
+      "epoch": 0.10812,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.005692545572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.4825,
+      "loss/crossentropy": 2.078865647315979,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.251836359500885,
+      "step": 5406
+    },
+    {
+      "epoch": 0.10816,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.006245930989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.4409,
+      "loss/crossentropy": 2.031971752643585,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23658733069896698,
+      "step": 5408
+    },
+    {
+      "epoch": 0.1082,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.00504150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3034,
+      "loss/crossentropy": 1.8173908591270447,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2169174626469612,
+      "step": 5410
+    },
+    {
+      "epoch": 0.10824,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0086090087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.838,
+      "loss/crossentropy": 2.2501285672187805,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.240458145737648,
+      "step": 5412
+    },
+    {
+      "epoch": 0.10828,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.015168253580729167,
+      "learning_rate": 0.0001,
+      "loss": 4.5449,
+      "loss/crossentropy": 2.256573438644409,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26129382848739624,
+      "step": 5414
+    },
+    {
+      "epoch": 0.10832,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.0153717041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.7704,
+      "loss/crossentropy": 2.2014705538749695,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2334136888384819,
+      "step": 5416
+    },
+    {
+      "epoch": 0.10836,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.013016764322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.4046,
+      "loss/crossentropy": 1.8590609431266785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2271011471748352,
+      "step": 5418
+    },
+    {
+      "epoch": 0.1084,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.017378743489583334,
+      "learning_rate": 0.0001,
+      "loss": 4.9419,
+      "loss/crossentropy": 2.2923961877822876,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25833888351917267,
+      "step": 5420
+    },
+    {
+      "epoch": 0.10844,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.017723592122395833,
+      "learning_rate": 0.0001,
+      "loss": 4.4535,
+      "loss/crossentropy": 2.1932299733161926,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26209479570388794,
+      "step": 5422
+    },
+    {
+      "epoch": 0.10848,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.016185506184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.7057,
+      "loss/crossentropy": 2.3909924030303955,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2525275945663452,
+      "step": 5424
+    },
+    {
+      "epoch": 0.10852,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.011324055989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.7509,
+      "loss/crossentropy": 2.423817992210388,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2635423541069031,
+      "step": 5426
+    },
+    {
+      "epoch": 0.10856,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.009227498372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.7082,
+      "loss/crossentropy": 1.9641632437705994,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2460889369249344,
+      "step": 5428
+    },
+    {
+      "epoch": 0.1086,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.0077707926432291664,
+      "learning_rate": 0.0001,
+      "loss": 4.6347,
+      "loss/crossentropy": 2.027769148349762,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2357894703745842,
+      "step": 5430
+    },
+    {
+      "epoch": 0.10864,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.0128570556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4833,
+      "loss/crossentropy": 2.122319996356964,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2445123866200447,
+      "step": 5432
+    },
+    {
+      "epoch": 0.10868,
+      "grad_norm": 3.53125,
+      "grad_norm_var": 0.0993072509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.6332,
+      "loss/crossentropy": 1.8631052374839783,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23456327617168427,
+      "step": 5434
+    },
+    {
+      "epoch": 0.10872,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.1000885009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.6022,
+      "loss/crossentropy": 2.184281885623932,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24073782563209534,
+      "step": 5436
+    },
+    {
+      "epoch": 0.10876,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.09521077473958334,
+      "learning_rate": 0.0001,
+      "loss": 4.7912,
+      "loss/crossentropy": 1.9587833881378174,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22247321158647537,
+      "step": 5438
+    },
+    {
+      "epoch": 0.1088,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.09562886555989583,
+      "learning_rate": 0.0001,
+      "loss": 4.5185,
+      "loss/crossentropy": 2.334655284881592,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24817728251218796,
+      "step": 5440
+    },
+    {
+      "epoch": 0.10884,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.10161031087239583,
+      "learning_rate": 0.0001,
+      "loss": 4.3817,
+      "loss/crossentropy": 2.1424371004104614,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2622353136539459,
+      "step": 5442
+    },
+    {
+      "epoch": 0.10888,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.10598958333333333,
+      "learning_rate": 0.0001,
+      "loss": 4.5876,
+      "loss/crossentropy": 2.0363662242889404,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2450125440955162,
+      "step": 5444
+    },
+    {
+      "epoch": 0.10892,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.110791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.4015,
+      "loss/crossentropy": 2.0536006689071655,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23283181339502335,
+      "step": 5446
+    },
+    {
+      "epoch": 0.10896,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.10741780598958334,
+      "learning_rate": 0.0001,
+      "loss": 4.5194,
+      "loss/crossentropy": 2.2678059339523315,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24513862282037735,
+      "step": 5448
+    },
+    {
+      "epoch": 0.109,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.02115478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.7404,
+      "loss/crossentropy": 2.406686782836914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2901146113872528,
+      "step": 5450
+    },
+    {
+      "epoch": 0.10904,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.017801920572916668,
+      "learning_rate": 0.0001,
+      "loss": 4.4724,
+      "loss/crossentropy": 2.352605938911438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25215400755405426,
+      "step": 5452
+    },
+    {
+      "epoch": 0.10908,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.014546712239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.5593,
+      "loss/crossentropy": 1.9139850735664368,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22682765871286392,
+      "step": 5454
+    },
+    {
+      "epoch": 0.10912,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.01441650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.6163,
+      "loss/crossentropy": 2.3240445852279663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24574412405490875,
+      "step": 5456
+    },
+    {
+      "epoch": 0.10916,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0123687744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3356,
+      "loss/crossentropy": 1.9347040057182312,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2159302830696106,
+      "step": 5458
+    },
+    {
+      "epoch": 0.1092,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.012360636393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.6539,
+      "loss/crossentropy": 1.8933109641075134,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.284725621342659,
+      "step": 5460
+    },
+    {
+      "epoch": 0.10924,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0129058837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2488,
+      "loss/crossentropy": 2.3611297607421875,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2651517689228058,
+      "step": 5462
+    },
+    {
+      "epoch": 0.10928,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0235504150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.4564,
+      "loss/crossentropy": 1.828608751296997,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2337760180234909,
+      "step": 5464
+    },
+    {
+      "epoch": 0.10932,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.015412394205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.321,
+      "loss/crossentropy": 2.1374374628067017,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2525453567504883,
+      "step": 5466
+    },
+    {
+      "epoch": 0.10936,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0150543212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5307,
+      "loss/crossentropy": 1.8054441213607788,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2252344712615013,
+      "step": 5468
+    },
+    {
+      "epoch": 0.1094,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.022972615559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.616,
+      "loss/crossentropy": 2.1468498706817627,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2604861631989479,
+      "step": 5470
+    },
+    {
+      "epoch": 0.10944,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.022684733072916668,
+      "learning_rate": 0.0001,
+      "loss": 4.7298,
+      "loss/crossentropy": 2.2180548906326294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2603686898946762,
+      "step": 5472
+    },
+    {
+      "epoch": 0.10948,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.022261555989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.5263,
+      "loss/crossentropy": 1.9773708581924438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23676057159900665,
+      "step": 5474
+    },
+    {
+      "epoch": 0.10952,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.023485310872395835,
+      "learning_rate": 0.0001,
+      "loss": 4.6156,
+      "loss/crossentropy": 1.9277283549308777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22822558879852295,
+      "step": 5476
+    },
+    {
+      "epoch": 0.10956,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.022484334309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.5529,
+      "loss/crossentropy": 2.0625431537628174,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2565220817923546,
+      "step": 5478
+    },
+    {
+      "epoch": 0.1096,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.01441650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.7956,
+      "loss/crossentropy": 2.383894443511963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2563806623220444,
+      "step": 5480
+    },
+    {
+      "epoch": 0.10964,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0120758056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.5226,
+      "loss/crossentropy": 2.409442663192749,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2637571170926094,
+      "step": 5482
+    },
+    {
+      "epoch": 0.10968,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0151763916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.743,
+      "loss/crossentropy": 2.1789854764938354,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23846548050642014,
+      "step": 5484
+    },
+    {
+      "epoch": 0.10972,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.011839803059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4108,
+      "loss/crossentropy": 2.127842903137207,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2622772455215454,
+      "step": 5486
+    },
+    {
+      "epoch": 0.10976,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.01412353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.6032,
+      "loss/crossentropy": 2.107556462287903,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24714312702417374,
+      "step": 5488
+    },
+    {
+      "epoch": 0.1098,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.022459920247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.6525,
+      "loss/crossentropy": 2.1959601640701294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23568396270275116,
+      "step": 5490
+    },
+    {
+      "epoch": 0.10984,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.021826171875,
+      "learning_rate": 0.0001,
+      "loss": 4.5584,
+      "loss/crossentropy": 2.1246761083602905,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24083472788333893,
+      "step": 5492
+    },
+    {
+      "epoch": 0.10988,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.025031534830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.6436,
+      "loss/crossentropy": 2.091724157333374,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2527346611022949,
+      "step": 5494
+    },
+    {
+      "epoch": 0.10992,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.026056925455729168,
+      "learning_rate": 0.0001,
+      "loss": 4.3207,
+      "loss/crossentropy": 1.8898470997810364,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20916878432035446,
+      "step": 5496
+    },
+    {
+      "epoch": 0.10996,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.025951131184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.399,
+      "loss/crossentropy": 2.1901716589927673,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2409309297800064,
+      "step": 5498
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.017626953125,
+      "learning_rate": 0.0001,
+      "loss": 4.6897,
+      "loss/crossentropy": 2.1018574237823486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24559657275676727,
+      "step": 5500
+    },
+    {
+      "epoch": 0.11004,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.019449869791666668,
+      "learning_rate": 0.0001,
+      "loss": 3.8159,
+      "loss/crossentropy": 2.0575350522994995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23721858859062195,
+      "step": 5502
+    },
+    {
+      "epoch": 0.11008,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.019710286458333334,
+      "learning_rate": 0.0001,
+      "loss": 4.6347,
+      "loss/crossentropy": 2.1846336126327515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2378462702035904,
+      "step": 5504
+    },
+    {
+      "epoch": 0.11012,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.014574178059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4028,
+      "loss/crossentropy": 2.1359363198280334,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26211391389369965,
+      "step": 5506
+    },
+    {
+      "epoch": 0.11016,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.01416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.5621,
+      "loss/crossentropy": 2.236825942993164,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2557792589068413,
+      "step": 5508
+    },
+    {
+      "epoch": 0.1102,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.009521484375,
+      "learning_rate": 0.0001,
+      "loss": 4.3234,
+      "loss/crossentropy": 2.3140580654144287,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2554958164691925,
+      "step": 5510
+    },
+    {
+      "epoch": 0.11024,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.00904541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.4382,
+      "loss/crossentropy": 1.7190355062484741,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19964009523391724,
+      "step": 5512
+    },
+    {
+      "epoch": 0.11028,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.014631144205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3831,
+      "loss/crossentropy": 1.8326427340507507,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2199154868721962,
+      "step": 5514
+    },
+    {
+      "epoch": 0.11032,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.01549072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4494,
+      "loss/crossentropy": 1.9013121724128723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20293578505516052,
+      "step": 5516
+    },
+    {
+      "epoch": 0.11036,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0138671875,
+      "learning_rate": 0.0001,
+      "loss": 4.7056,
+      "loss/crossentropy": 2.0221983790397644,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23435519635677338,
+      "step": 5518
+    },
+    {
+      "epoch": 0.1104,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.0140533447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.6157,
+      "loss/crossentropy": 2.153970956802368,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22127598524093628,
+      "step": 5520
+    },
+    {
+      "epoch": 0.11044,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010835774739583333,
+      "learning_rate": 0.0001,
+      "loss": 4.4516,
+      "loss/crossentropy": 1.8674496412277222,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.232261061668396,
+      "step": 5522
+    },
+    {
+      "epoch": 0.11048,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.012495930989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.6528,
+      "loss/crossentropy": 2.1575759649276733,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22218701988458633,
+      "step": 5524
+    },
+    {
+      "epoch": 0.11052,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.012434895833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.5553,
+      "loss/crossentropy": 2.054452419281006,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25137215852737427,
+      "step": 5526
+    },
+    {
+      "epoch": 0.11056,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.022639973958333334,
+      "learning_rate": 0.0001,
+      "loss": 4.4789,
+      "loss/crossentropy": 1.966478705406189,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23289234936237335,
+      "step": 5528
+    },
+    {
+      "epoch": 0.1106,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.017626953125,
+      "learning_rate": 0.0001,
+      "loss": 4.687,
+      "loss/crossentropy": 2.171034336090088,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23373593389987946,
+      "step": 5530
+    },
+    {
+      "epoch": 0.11064,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.022362263997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.142,
+      "loss/crossentropy": 2.2416292428970337,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24835532158613205,
+      "step": 5532
+    },
+    {
+      "epoch": 0.11068,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.02720947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.6252,
+      "loss/crossentropy": 2.2599780559539795,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24948878586292267,
+      "step": 5534
+    },
+    {
+      "epoch": 0.11072,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.028902180989583335,
+      "learning_rate": 0.0001,
+      "loss": 4.6984,
+      "loss/crossentropy": 2.2292014360427856,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26616473495960236,
+      "step": 5536
+    },
+    {
+      "epoch": 0.11076,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.027424112955729166,
+      "learning_rate": 0.0001,
+      "loss": 4.336,
+      "loss/crossentropy": 1.9285388588905334,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22964124381542206,
+      "step": 5538
+    },
+    {
+      "epoch": 0.1108,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0251953125,
+      "learning_rate": 0.0001,
+      "loss": 4.2861,
+      "loss/crossentropy": 1.864789366722107,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2264304906129837,
+      "step": 5540
+    },
+    {
+      "epoch": 0.11084,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0250885009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.5163,
+      "loss/crossentropy": 1.8676912188529968,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21919699758291245,
+      "step": 5542
+    },
+    {
+      "epoch": 0.11088,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0190338134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3612,
+      "loss/crossentropy": 2.34523469209671,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25638002157211304,
+      "step": 5544
+    },
+    {
+      "epoch": 0.11092,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.020992024739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.5165,
+      "loss/crossentropy": 2.2903120517730713,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25111711025238037,
+      "step": 5546
+    },
+    {
+      "epoch": 0.11096,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.017671712239583335,
+      "learning_rate": 0.0001,
+      "loss": 4.5336,
+      "loss/crossentropy": 2.2106658220291138,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22340595722198486,
+      "step": 5548
+    },
+    {
+      "epoch": 0.111,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.013700358072916667,
+      "learning_rate": 0.0001,
+      "loss": 4.6305,
+      "loss/crossentropy": 2.0777581334114075,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2446284517645836,
+      "step": 5550
+    },
+    {
+      "epoch": 0.11104,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.010758463541666667,
+      "learning_rate": 0.0001,
+      "loss": 4.5507,
+      "loss/crossentropy": 2.131237506866455,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24113191664218903,
+      "step": 5552
+    },
+    {
+      "epoch": 0.11108,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.018778483072916668,
+      "learning_rate": 0.0001,
+      "loss": 4.6337,
+      "loss/crossentropy": 2.190987467765808,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2528213635087013,
+      "step": 5554
+    },
+    {
+      "epoch": 0.11112,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.019527180989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.4889,
+      "loss/crossentropy": 2.26843523979187,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27781064808368683,
+      "step": 5556
+    },
+    {
+      "epoch": 0.11116,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.019527180989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.2242,
+      "loss/crossentropy": 1.9507999420166016,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23172564804553986,
+      "step": 5558
+    },
+    {
+      "epoch": 0.1112,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.016966756184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.4821,
+      "loss/crossentropy": 2.0738234519958496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24497026205062866,
+      "step": 5560
+    },
+    {
+      "epoch": 0.11124,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.019261678059895832,
+      "learning_rate": 0.0001,
+      "loss": 4.9501,
+      "loss/crossentropy": 2.273179054260254,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2917838394641876,
+      "step": 5562
+    },
+    {
+      "epoch": 0.11128,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.017899576822916666,
+      "learning_rate": 0.0001,
+      "loss": 4.774,
+      "loss/crossentropy": 2.085157036781311,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2488701120018959,
+      "step": 5564
+    },
+    {
+      "epoch": 0.11132,
+      "grad_norm": 3.109375,
+      "grad_norm_var": 0.05950113932291667,
+      "learning_rate": 0.0001,
+      "loss": 4.2869,
+      "loss/crossentropy": 2.0528116822242737,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24717991054058075,
+      "step": 5566
+    },
+    {
+      "epoch": 0.11136,
+      "grad_norm": 7.0,
+      "grad_norm_var": 1.3981597900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.4443,
+      "loss/crossentropy": 2.0651500821113586,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26806148886680603,
+      "step": 5568
+    },
+    {
+      "epoch": 0.1114,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 1.4721588134765624,
+      "learning_rate": 0.0001,
+      "loss": 4.6162,
+      "loss/crossentropy": 2.1860616207122803,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23045460134744644,
+      "step": 5570
+    },
+    {
+      "epoch": 0.11144,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 1.460399373372396,
+      "learning_rate": 0.0001,
+      "loss": 4.3629,
+      "loss/crossentropy": 1.6931262016296387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20816385000944138,
+      "step": 5572
+    },
+    {
+      "epoch": 0.11148,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 1.4581858317057292,
+      "learning_rate": 0.0001,
+      "loss": 4.3376,
+      "loss/crossentropy": 2.199341118335724,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2490597665309906,
+      "step": 5574
+    },
+    {
+      "epoch": 0.11152,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 1.4561513264973958,
+      "learning_rate": 0.0001,
+      "loss": 4.6627,
+      "loss/crossentropy": 2.2010069489479065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24095547199249268,
+      "step": 5576
+    },
+    {
+      "epoch": 0.11156,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 1.4810129801432292,
+      "learning_rate": 0.0001,
+      "loss": 4.5243,
+      "loss/crossentropy": 1.9907150864601135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2122703418135643,
+      "step": 5578
+    },
+    {
+      "epoch": 0.1116,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 1.5023508707682292,
+      "learning_rate": 0.0001,
+      "loss": 4.2618,
+      "loss/crossentropy": 2.196335554122925,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25111766904592514,
+      "step": 5580
+    },
+    {
+      "epoch": 0.11164,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 1.5003000895182292,
+      "learning_rate": 0.0001,
+      "loss": 4.5104,
+      "loss/crossentropy": 2.0762988924980164,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23773670941591263,
+      "step": 5582
+    },
+    {
+      "epoch": 0.11168,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.15530192057291667,
+      "learning_rate": 0.0001,
+      "loss": 4.3166,
+      "loss/crossentropy": 2.0803143978118896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22162869572639465,
+      "step": 5584
+    },
+    {
+      "epoch": 0.11172,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.0084136962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5596,
+      "loss/crossentropy": 2.1821994185447693,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26280316710472107,
+      "step": 5586
+    },
+    {
+      "epoch": 0.11176,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.011847941080729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4328,
+      "loss/crossentropy": 2.1899439096450806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24677179753780365,
+      "step": 5588
+    },
+    {
+      "epoch": 0.1118,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0116851806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4361,
+      "loss/crossentropy": 2.334734559059143,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2481135129928589,
+      "step": 5590
+    },
+    {
+      "epoch": 0.11184,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.007291666666666667,
+      "learning_rate": 0.0001,
+      "loss": 4.7838,
+      "loss/crossentropy": 2.2976341247558594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2454354390501976,
+      "step": 5592
+    },
+    {
+      "epoch": 0.11188,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.007477823893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.7148,
+      "loss/crossentropy": 2.3243749141693115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2735731601715088,
+      "step": 5594
+    },
+    {
+      "epoch": 0.11192,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.007372029622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.7124,
+      "loss/crossentropy": 2.0328271985054016,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21919244527816772,
+      "step": 5596
+    },
+    {
+      "epoch": 0.11196,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0077789306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4584,
+      "loss/crossentropy": 2.249367594718933,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2577860951423645,
+      "step": 5598
+    },
+    {
+      "epoch": 0.112,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.009505208333333333,
+      "learning_rate": 0.0001,
+      "loss": 4.2594,
+      "loss/crossentropy": 1.9844761490821838,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2142435386776924,
+      "step": 5600
+    },
+    {
+      "epoch": 0.11204,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.0111328125,
+      "learning_rate": 0.0001,
+      "loss": 4.6043,
+      "loss/crossentropy": 2.1334372758865356,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23296397179365158,
+      "step": 5602
+    },
+    {
+      "epoch": 0.11208,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.0078084309895833336,
+      "learning_rate": 0.0001,
+      "loss": 4.5308,
+      "loss/crossentropy": 2.119946002960205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23398507386446,
+      "step": 5604
+    },
+    {
+      "epoch": 0.11212,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.008040364583333333,
+      "learning_rate": 0.0001,
+      "loss": 4.5011,
+      "loss/crossentropy": 2.0414544343948364,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23420867323875427,
+      "step": 5606
+    },
+    {
+      "epoch": 0.11216,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.009130859375,
+      "learning_rate": 0.0001,
+      "loss": 4.2092,
+      "loss/crossentropy": 1.9592725038528442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23922864347696304,
+      "step": 5608
+    },
+    {
+      "epoch": 0.1122,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.011188761393229166,
+      "learning_rate": 0.0001,
+      "loss": 4.4795,
+      "loss/crossentropy": 2.150269627571106,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2528265416622162,
+      "step": 5610
+    },
+    {
+      "epoch": 0.11224,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.3099772135416667,
+      "learning_rate": 0.0001,
+      "loss": 4.6234,
+      "loss/crossentropy": 2.0591378211975098,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2430240362882614,
+      "step": 5612
+    },
+    {
+      "epoch": 0.11228,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.30794169108072916,
+      "learning_rate": 0.0001,
+      "loss": 4.4251,
+      "loss/crossentropy": 2.2132861614227295,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22676552087068558,
+      "step": 5614
+    },
+    {
+      "epoch": 0.11232,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.3002237955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4332,
+      "loss/crossentropy": 1.9607917070388794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23068836331367493,
+      "step": 5616
+    },
+    {
+      "epoch": 0.11236,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.3021321614583333,
+      "learning_rate": 0.0001,
+      "loss": 4.549,
+      "loss/crossentropy": 2.2245940566062927,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25029121339321136,
+      "step": 5618
+    },
+    {
+      "epoch": 0.1124,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.30278218587239586,
+      "learning_rate": 0.0001,
+      "loss": 4.4539,
+      "loss/crossentropy": 2.2511253356933594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25143079459667206,
+      "step": 5620
+    },
+    {
+      "epoch": 0.11244,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.3026194254557292,
+      "learning_rate": 0.0001,
+      "loss": 4.4632,
+      "loss/crossentropy": 2.2945470809936523,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2196483463048935,
+      "step": 5622
+    },
+    {
+      "epoch": 0.11248,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.29273681640625,
+      "learning_rate": 0.0001,
+      "loss": 4.8769,
+      "loss/crossentropy": 2.2266393899917603,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24020668864250183,
+      "step": 5624
+    },
+    {
+      "epoch": 0.11252,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.29136454264322914,
+      "learning_rate": 0.0001,
+      "loss": 4.742,
+      "loss/crossentropy": 2.2835845947265625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27199871838092804,
+      "step": 5626
+    },
+    {
+      "epoch": 0.11256,
+      "grad_norm": 2.859375,
+      "grad_norm_var": 0.026383463541666666,
+      "learning_rate": 0.0001,
+      "loss": 4.4213,
+      "loss/crossentropy": 1.9576718211174011,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21716968715190887,
+      "step": 5628
+    },
+    {
+      "epoch": 0.1126,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.044611612955729164,
+      "learning_rate": 0.0001,
+      "loss": 4.7695,
+      "loss/crossentropy": 2.0955676436424255,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24259109795093536,
+      "step": 5630
+    },
+    {
+      "epoch": 0.11264,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.044840494791666664,
+      "learning_rate": 0.0001,
+      "loss": 4.4127,
+      "loss/crossentropy": 2.119523346424103,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2478165179491043,
+      "step": 5632
+    },
+    {
+      "epoch": 0.11268,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0444488525390625,
+      "learning_rate": 0.0001,
+      "loss": 4.5591,
+      "loss/crossentropy": 2.189425826072693,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24532486498355865,
+      "step": 5634
+    },
+    {
+      "epoch": 0.11272,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.0467193603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4969,
+      "loss/crossentropy": 2.215874433517456,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24533041566610336,
+      "step": 5636
+    },
+    {
+      "epoch": 0.11276,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.04868876139322917,
+      "learning_rate": 0.0001,
+      "loss": 4.5657,
+      "loss/crossentropy": 2.226451873779297,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2545652836561203,
+      "step": 5638
+    },
+    {
+      "epoch": 0.1128,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.05181884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4779,
+      "loss/crossentropy": 2.0343592762947083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2503928989171982,
+      "step": 5640
+    },
+    {
+      "epoch": 0.11284,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0535552978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.6253,
+      "loss/crossentropy": 2.142001748085022,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24624411761760712,
+      "step": 5642
+    },
+    {
+      "epoch": 0.11288,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.039383951822916666,
+      "learning_rate": 0.0001,
+      "loss": 4.4212,
+      "loss/crossentropy": 1.822394609451294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21315739303827286,
+      "step": 5644
+    },
+    {
+      "epoch": 0.11292,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.018863932291666666,
+      "learning_rate": 0.0001,
+      "loss": 4.1611,
+      "loss/crossentropy": 2.3221731185913086,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24407917261123657,
+      "step": 5646
+    },
+    {
+      "epoch": 0.11296,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.012531534830729166,
+      "learning_rate": 0.0001,
+      "loss": 4.5591,
+      "loss/crossentropy": 1.9784467816352844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22578515857458115,
+      "step": 5648
+    },
+    {
+      "epoch": 0.113,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.0142730712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.636,
+      "loss/crossentropy": 2.2148635387420654,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25031210482120514,
+      "step": 5650
+    },
+    {
+      "epoch": 0.11304,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.012369791666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.4644,
+      "loss/crossentropy": 1.9204095602035522,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22073335200548172,
+      "step": 5652
+    },
+    {
+      "epoch": 0.11308,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.009764607747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.6601,
+      "loss/crossentropy": 2.092605173587799,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24161123484373093,
+      "step": 5654
+    },
+    {
+      "epoch": 0.11312,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.0163726806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.5977,
+      "loss/crossentropy": 1.9546263217926025,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22741927206516266,
+      "step": 5656
+    },
+    {
+      "epoch": 0.11316,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.016974894205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.6585,
+      "loss/crossentropy": 2.256605863571167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2560018301010132,
+      "step": 5658
+    },
+    {
+      "epoch": 0.1132,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0157135009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3605,
+      "loss/crossentropy": 2.24527370929718,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.270521879196167,
+      "step": 5660
+    },
+    {
+      "epoch": 0.11324,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.014029947916666667,
+      "learning_rate": 0.0001,
+      "loss": 4.5429,
+      "loss/crossentropy": 1.8154722452163696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22786997258663177,
+      "step": 5662
+    },
+    {
+      "epoch": 0.11328,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.0143951416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.4079,
+      "loss/crossentropy": 2.135699689388275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.253468282520771,
+      "step": 5664
+    },
+    {
+      "epoch": 0.11332,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.015087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5381,
+      "loss/crossentropy": 2.15896338224411,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25651170313358307,
+      "step": 5666
+    },
+    {
+      "epoch": 0.11336,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.012035115559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.7877,
+      "loss/crossentropy": 2.115864336490631,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28546491265296936,
+      "step": 5668
+    },
+    {
+      "epoch": 0.1134,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.012214152018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.4283,
+      "loss/crossentropy": 2.2036256790161133,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2397892326116562,
+      "step": 5670
+    },
+    {
+      "epoch": 0.11344,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.008430989583333333,
+      "learning_rate": 0.0001,
+      "loss": 4.598,
+      "loss/crossentropy": 2.4966647624969482,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2984588146209717,
+      "step": 5672
+    },
+    {
+      "epoch": 0.11348,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.008202107747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.5289,
+      "loss/crossentropy": 2.051860749721527,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2748461365699768,
+      "step": 5674
+    },
+    {
+      "epoch": 0.11352,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.007938639322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.597,
+      "loss/crossentropy": 2.046416461467743,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2320190668106079,
+      "step": 5676
+    },
+    {
+      "epoch": 0.11356,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0054514567057291664,
+      "learning_rate": 0.0001,
+      "loss": 4.7389,
+      "loss/crossentropy": 2.2385342121124268,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21859309077262878,
+      "step": 5678
+    },
+    {
+      "epoch": 0.1136,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.006843058268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2487,
+      "loss/crossentropy": 1.8511550426483154,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22065181285142899,
+      "step": 5680
+    },
+    {
+      "epoch": 0.11364,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.031477864583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.642,
+      "loss/crossentropy": 2.304056167602539,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29507844150066376,
+      "step": 5682
+    },
+    {
+      "epoch": 0.11368,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.03329671223958333,
+      "learning_rate": 0.0001,
+      "loss": 4.5519,
+      "loss/crossentropy": 1.993275225162506,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33826952427625656,
+      "step": 5684
+    },
+    {
+      "epoch": 0.11372,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.03498433430989583,
+      "learning_rate": 0.0001,
+      "loss": 4.4729,
+      "loss/crossentropy": 2.1836347579956055,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24539872258901596,
+      "step": 5686
+    },
+    {
+      "epoch": 0.11376,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.03504231770833333,
+      "learning_rate": 0.0001,
+      "loss": 4.5908,
+      "loss/crossentropy": 1.9467885494232178,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2180013582110405,
+      "step": 5688
+    },
+    {
+      "epoch": 0.1138,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.04907938639322917,
+      "learning_rate": 0.0001,
+      "loss": 4.2888,
+      "loss/crossentropy": 1.9907563924789429,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2551623433828354,
+      "step": 5690
+    },
+    {
+      "epoch": 0.11384,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.049153645833333336,
+      "learning_rate": 0.0001,
+      "loss": 4.2186,
+      "loss/crossentropy": 1.9452654719352722,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24263620376586914,
+      "step": 5692
+    },
+    {
+      "epoch": 0.11388,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.050093587239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.5816,
+      "loss/crossentropy": 2.2448811531066895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2508034110069275,
+      "step": 5694
+    },
+    {
+      "epoch": 0.11392,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.055712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1868,
+      "loss/crossentropy": 1.991935908794403,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22709138691425323,
+      "step": 5696
+    },
+    {
+      "epoch": 0.11396,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.47700907389322916,
+      "learning_rate": 0.0001,
+      "loss": 4.8208,
+      "loss/crossentropy": 2.0479623675346375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23980189859867096,
+      "step": 5698
+    },
+    {
+      "epoch": 0.114,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.4779581705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4742,
+      "loss/crossentropy": 1.9319151639938354,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2295292615890503,
+      "step": 5700
+    },
+    {
+      "epoch": 0.11404,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.47330322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.32,
+      "loss/crossentropy": 1.985447645187378,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.246543288230896,
+      "step": 5702
+    },
+    {
+      "epoch": 0.11408,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.478271484375,
+      "learning_rate": 0.0001,
+      "loss": 4.6602,
+      "loss/crossentropy": 2.0016521215438843,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25966253876686096,
+      "step": 5704
+    },
+    {
+      "epoch": 0.11412,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.47038472493489586,
+      "learning_rate": 0.0001,
+      "loss": 4.9207,
+      "loss/crossentropy": 2.112374246120453,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3251144737005234,
+      "step": 5706
+    },
+    {
+      "epoch": 0.11416,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.4698720296223958,
+      "learning_rate": 0.0001,
+      "loss": 4.5223,
+      "loss/crossentropy": 2.0931158661842346,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2275928407907486,
+      "step": 5708
+    },
+    {
+      "epoch": 0.1142,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.46104227701822914,
+      "learning_rate": 0.0001,
+      "loss": 4.4481,
+      "loss/crossentropy": 1.9867743849754333,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2419440597295761,
+      "step": 5710
+    },
+    {
+      "epoch": 0.11424,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.4639312744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3874,
+      "loss/crossentropy": 2.1822216510772705,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2437271624803543,
+      "step": 5712
+    },
+    {
+      "epoch": 0.11428,
+      "grad_norm": 6.46875,
+      "grad_norm_var": 1.1088775634765624,
+      "learning_rate": 0.0001,
+      "loss": 4.5307,
+      "loss/crossentropy": 2.412580370903015,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3222763389348984,
+      "step": 5714
+    },
+    {
+      "epoch": 0.11432,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 1.1019490559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2491,
+      "loss/crossentropy": 1.676392376422882,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24499019235372543,
+      "step": 5716
+    },
+    {
+      "epoch": 0.11436,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 1.1039876302083333,
+      "learning_rate": 0.0001,
+      "loss": 4.4741,
+      "loss/crossentropy": 1.7818856835365295,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22935030609369278,
+      "step": 5718
+    },
+    {
+      "epoch": 0.1144,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 1.0944732666015624,
+      "learning_rate": 0.0001,
+      "loss": 4.5445,
+      "loss/crossentropy": 2.012014925479889,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.231942281126976,
+      "step": 5720
+    },
+    {
+      "epoch": 0.11444,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 1.1023834228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5061,
+      "loss/crossentropy": 2.3663965463638306,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.254768967628479,
+      "step": 5722
+    },
+    {
+      "epoch": 0.11448,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 1.107982381184896,
+      "learning_rate": 0.0001,
+      "loss": 4.5878,
+      "loss/crossentropy": 2.343206286430359,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26609115302562714,
+      "step": 5724
+    },
+    {
+      "epoch": 0.11452,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 1.1131337483723958,
+      "learning_rate": 0.0001,
+      "loss": 4.5047,
+      "loss/crossentropy": 1.8696978092193604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24282050132751465,
+      "step": 5726
+    },
+    {
+      "epoch": 0.11456,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 1.1170644124348958,
+      "learning_rate": 0.0001,
+      "loss": 4.2198,
+      "loss/crossentropy": 2.1430450677871704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23211582750082016,
+      "step": 5728
+    },
+    {
+      "epoch": 0.1146,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.013841756184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.6868,
+      "loss/crossentropy": 2.2231308221817017,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23736387491226196,
+      "step": 5730
+    },
+    {
+      "epoch": 0.11464,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.014069620768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.4911,
+      "loss/crossentropy": 1.8789280652999878,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24081497639417648,
+      "step": 5732
+    },
+    {
+      "epoch": 0.11468,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.014875284830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4478,
+      "loss/crossentropy": 2.0491732358932495,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20530463755130768,
+      "step": 5734
+    },
+    {
+      "epoch": 0.11472,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.012516276041666666,
+      "learning_rate": 0.0001,
+      "loss": 4.4352,
+      "loss/crossentropy": 2.067046642303467,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24701344966888428,
+      "step": 5736
+    },
+    {
+      "epoch": 0.11476,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.012791951497395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4515,
+      "loss/crossentropy": 2.0207647681236267,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2243770807981491,
+      "step": 5738
+    },
+    {
+      "epoch": 0.1148,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.013084920247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.9205,
+      "loss/crossentropy": 2.230514347553253,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24377264082431793,
+      "step": 5740
+    },
+    {
+      "epoch": 0.11484,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.013997395833333334,
+      "learning_rate": 0.0001,
+      "loss": 4.3831,
+      "loss/crossentropy": 2.38068687915802,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24760407954454422,
+      "step": 5742
+    },
+    {
+      "epoch": 0.11488,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.012483723958333333,
+      "learning_rate": 0.0001,
+      "loss": 4.5598,
+      "loss/crossentropy": 2.238909125328064,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2503668889403343,
+      "step": 5744
+    },
+    {
+      "epoch": 0.11492,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.015265909830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4848,
+      "loss/crossentropy": 1.7423101663589478,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20424582809209824,
+      "step": 5746
+    },
+    {
+      "epoch": 0.11496,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0181304931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.6754,
+      "loss/crossentropy": 2.5906827449798584,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24731651693582535,
+      "step": 5748
+    },
+    {
+      "epoch": 0.115,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.025830078125,
+      "learning_rate": 0.0001,
+      "loss": 4.7427,
+      "loss/crossentropy": 2.418861746788025,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2505089193582535,
+      "step": 5750
+    },
+    {
+      "epoch": 0.11504,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.03658447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.869,
+      "loss/crossentropy": 2.158658504486084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2381347045302391,
+      "step": 5752
+    },
+    {
+      "epoch": 0.11508,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.03860270182291667,
+      "learning_rate": 0.0001,
+      "loss": 4.5073,
+      "loss/crossentropy": 2.0938435196876526,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23639065027236938,
+      "step": 5754
+    },
+    {
+      "epoch": 0.11512,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0441314697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2227,
+      "loss/crossentropy": 2.023799479007721,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24510329961776733,
+      "step": 5756
+    },
+    {
+      "epoch": 0.11516,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.03854166666666667,
+      "learning_rate": 0.0001,
+      "loss": 4.5155,
+      "loss/crossentropy": 2.3589184284210205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2778936177492142,
+      "step": 5758
+    },
+    {
+      "epoch": 0.1152,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.041731770833333334,
+      "learning_rate": 0.0001,
+      "loss": 4.4209,
+      "loss/crossentropy": 2.4897998571395874,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23546544462442398,
+      "step": 5760
+    },
+    {
+      "epoch": 0.11524,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.041747029622395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4619,
+      "loss/crossentropy": 2.0426196455955505,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22708147019147873,
+      "step": 5762
+    },
+    {
+      "epoch": 0.11528,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.0397369384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.6661,
+      "loss/crossentropy": 2.2582051753997803,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25646351277828217,
+      "step": 5764
+    },
+    {
+      "epoch": 0.11532,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.03430582682291667,
+      "learning_rate": 0.0001,
+      "loss": 4.5929,
+      "loss/crossentropy": 2.181105613708496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2380223199725151,
+      "step": 5766
+    },
+    {
+      "epoch": 0.11536,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.026200358072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.5078,
+      "loss/crossentropy": 2.236580967903137,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26678355038166046,
+      "step": 5768
+    },
+    {
+      "epoch": 0.1154,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.020361328125,
+      "learning_rate": 0.0001,
+      "loss": 4.6232,
+      "loss/crossentropy": 1.9383749961853027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24920085072517395,
+      "step": 5770
+    },
+    {
+      "epoch": 0.11544,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.018529256184895832,
+      "learning_rate": 0.0001,
+      "loss": 4.6044,
+      "loss/crossentropy": 2.2856688499450684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.258881650865078,
+      "step": 5772
+    },
+    {
+      "epoch": 0.11548,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.0239410400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.8391,
+      "loss/crossentropy": 2.2897390127182007,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24453644454479218,
+      "step": 5774
+    },
+    {
+      "epoch": 0.11552,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.014549763997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.5698,
+      "loss/crossentropy": 2.0502785444259644,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24106161296367645,
+      "step": 5776
+    },
+    {
+      "epoch": 0.11556,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.01441650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.5665,
+      "loss/crossentropy": 2.344050645828247,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24690424650907516,
+      "step": 5778
+    },
+    {
+      "epoch": 0.1156,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.021239217122395834,
+      "learning_rate": 0.0001,
+      "loss": 4.5273,
+      "loss/crossentropy": 2.2274389266967773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25618284940719604,
+      "step": 5780
+    },
+    {
+      "epoch": 0.11564,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.022184244791666665,
+      "learning_rate": 0.0001,
+      "loss": 4.4888,
+      "loss/crossentropy": 2.063184678554535,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22606099396944046,
+      "step": 5782
+    },
+    {
+      "epoch": 0.11568,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0206451416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2333,
+      "loss/crossentropy": 2.093947410583496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23588553071022034,
+      "step": 5784
+    },
+    {
+      "epoch": 0.11572,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0246490478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.426,
+      "loss/crossentropy": 2.1599318981170654,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20322780311107635,
+      "step": 5786
+    },
+    {
+      "epoch": 0.11576,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.02506103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5398,
+      "loss/crossentropy": 2.274693012237549,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2704206556081772,
+      "step": 5788
+    },
+    {
+      "epoch": 0.1158,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.0222076416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.6755,
+      "loss/crossentropy": 1.9712103009223938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28300437331199646,
+      "step": 5790
+    },
+    {
+      "epoch": 0.11584,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.024344889322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.2035,
+      "loss/crossentropy": 2.027747690677643,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24419110268354416,
+      "step": 5792
+    },
+    {
+      "epoch": 0.11588,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.024251302083333332,
+      "learning_rate": 0.0001,
+      "loss": 4.5614,
+      "loss/crossentropy": 2.162364959716797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2522001415491104,
+      "step": 5794
+    },
+    {
+      "epoch": 0.11592,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.0166656494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.4711,
+      "loss/crossentropy": 2.558881998062134,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26168718934059143,
+      "step": 5796
+    },
+    {
+      "epoch": 0.11596,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0162994384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4108,
+      "loss/crossentropy": 2.1027071475982666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24049362540245056,
+      "step": 5798
+    },
+    {
+      "epoch": 0.116,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.015034993489583334,
+      "learning_rate": 0.0001,
+      "loss": 4.6059,
+      "loss/crossentropy": 2.488932490348816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2697457820177078,
+      "step": 5800
+    },
+    {
+      "epoch": 0.11604,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.0121734619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.6706,
+      "loss/crossentropy": 2.0441418886184692,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23980429768562317,
+      "step": 5802
+    },
+    {
+      "epoch": 0.11608,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.0107421875,
+      "learning_rate": 0.0001,
+      "loss": 4.8471,
+      "loss/crossentropy": 2.2873395681381226,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3105090409517288,
+      "step": 5804
+    },
+    {
+      "epoch": 0.11612,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.00826416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.5388,
+      "loss/crossentropy": 1.8773444890975952,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20640414953231812,
+      "step": 5806
+    },
+    {
+      "epoch": 0.11616,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.007502237955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.7338,
+      "loss/crossentropy": 2.3736027479171753,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2679155319929123,
+      "step": 5808
+    },
+    {
+      "epoch": 0.1162,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.010676066080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0496,
+      "loss/crossentropy": 2.071690082550049,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21868111193180084,
+      "step": 5810
+    },
+    {
+      "epoch": 0.11624,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.015425618489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.4029,
+      "loss/crossentropy": 2.0531184673309326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24457989633083344,
+      "step": 5812
+    },
+    {
+      "epoch": 0.11628,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.015718587239583335,
+      "learning_rate": 0.0001,
+      "loss": 4.6511,
+      "loss/crossentropy": 2.1161271929740906,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22809523344039917,
+      "step": 5814
+    },
+    {
+      "epoch": 0.11632,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0196929931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2736,
+      "loss/crossentropy": 1.6557151675224304,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20782940834760666,
+      "step": 5816
+    },
+    {
+      "epoch": 0.11636,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.03704020182291667,
+      "learning_rate": 0.0001,
+      "loss": 4.5977,
+      "loss/crossentropy": 2.2499040365219116,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2713513821363449,
+      "step": 5818
+    },
+    {
+      "epoch": 0.1164,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.03723958333333333,
+      "learning_rate": 0.0001,
+      "loss": 4.6417,
+      "loss/crossentropy": 2.3616446256637573,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2613145411014557,
+      "step": 5820
+    },
+    {
+      "epoch": 0.11644,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3744,
+      "loss/crossentropy": 2.0932790637016296,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23843754082918167,
+      "step": 5822
+    },
+    {
+      "epoch": 0.11648,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.03540751139322917,
+      "learning_rate": 0.0001,
+      "loss": 4.4089,
+      "loss/crossentropy": 2.128177046775818,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26579540967941284,
+      "step": 5824
+    },
+    {
+      "epoch": 0.11652,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.031281534830729166,
+      "learning_rate": 0.0001,
+      "loss": 4.6942,
+      "loss/crossentropy": 2.332372784614563,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27953268587589264,
+      "step": 5826
+    },
+    {
+      "epoch": 0.11656,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0263092041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.4615,
+      "loss/crossentropy": 2.360959053039551,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26249848306179047,
+      "step": 5828
+    },
+    {
+      "epoch": 0.1166,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.030745442708333334,
+      "learning_rate": 0.0001,
+      "loss": 3.9757,
+      "loss/crossentropy": 1.9800339341163635,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22766301035881042,
+      "step": 5830
+    },
+    {
+      "epoch": 0.11664,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.029442342122395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3287,
+      "loss/crossentropy": 2.082980155944824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2368428260087967,
+      "step": 5832
+    },
+    {
+      "epoch": 0.11668,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.011986287434895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4441,
+      "loss/crossentropy": 2.093027710914612,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21121951937675476,
+      "step": 5834
+    },
+    {
+      "epoch": 0.11672,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.011031087239583333,
+      "learning_rate": 0.0001,
+      "loss": 4.5008,
+      "loss/crossentropy": 2.1329175233840942,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25303877890110016,
+      "step": 5836
+    },
+    {
+      "epoch": 0.11676,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.022932942708333334,
+      "learning_rate": 0.0001,
+      "loss": 4.5174,
+      "loss/crossentropy": 1.79305762052536,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20448636263608932,
+      "step": 5838
+    },
+    {
+      "epoch": 0.1168,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.024933878580729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2722,
+      "loss/crossentropy": 1.957836627960205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22745755314826965,
+      "step": 5840
+    },
+    {
+      "epoch": 0.11684,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.025007120768229165,
+      "learning_rate": 0.0001,
+      "loss": 4.3773,
+      "loss/crossentropy": 2.1167174577713013,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2437898963689804,
+      "step": 5842
+    },
+    {
+      "epoch": 0.11688,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.024800618489583332,
+      "learning_rate": 0.0001,
+      "loss": 4.4856,
+      "loss/crossentropy": 2.3288447856903076,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2355937883257866,
+      "step": 5844
+    },
+    {
+      "epoch": 0.11692,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.022337849934895834,
+      "learning_rate": 0.0001,
+      "loss": 4.4774,
+      "loss/crossentropy": 2.2526416778564453,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24917340278625488,
+      "step": 5846
+    },
+    {
+      "epoch": 0.11696,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0198883056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.391,
+      "loss/crossentropy": 2.194224774837494,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2625589966773987,
+      "step": 5848
+    },
+    {
+      "epoch": 0.117,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.020116170247395832,
+      "learning_rate": 0.0001,
+      "loss": 4.4955,
+      "loss/crossentropy": 2.0156877040863037,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22422078251838684,
+      "step": 5850
+    },
+    {
+      "epoch": 0.11704,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.01962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.6231,
+      "loss/crossentropy": 2.2480785846710205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24705884605646133,
+      "step": 5852
+    },
+    {
+      "epoch": 0.11708,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0078521728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4817,
+      "loss/crossentropy": 2.0915993452072144,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24669666588306427,
+      "step": 5854
+    },
+    {
+      "epoch": 0.11712,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0062652587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4444,
+      "loss/crossentropy": 2.1283876299858093,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24228712916374207,
+      "step": 5856
+    },
+    {
+      "epoch": 0.11716,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.015523274739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.6582,
+      "loss/crossentropy": 2.028861939907074,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2281404584646225,
+      "step": 5858
+    },
+    {
+      "epoch": 0.1172,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.015852864583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.2293,
+      "loss/crossentropy": 2.154610753059387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24233703315258026,
+      "step": 5860
+    },
+    {
+      "epoch": 0.11724,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.015787760416666668,
+      "learning_rate": 0.0001,
+      "loss": 4.5059,
+      "loss/crossentropy": 1.9396602511405945,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22303076088428497,
+      "step": 5862
+    },
+    {
+      "epoch": 0.11728,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.021061197916666666,
+      "learning_rate": 0.0001,
+      "loss": 4.7447,
+      "loss/crossentropy": 2.053893029689789,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26897912472486496,
+      "step": 5864
+    },
+    {
+      "epoch": 0.11732,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0263336181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3529,
+      "loss/crossentropy": 1.990949273109436,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22180908918380737,
+      "step": 5866
+    },
+    {
+      "epoch": 0.11736,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.030085245768229168,
+      "learning_rate": 0.0001,
+      "loss": 4.6434,
+      "loss/crossentropy": 2.0929455161094666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24055248498916626,
+      "step": 5868
+    },
+    {
+      "epoch": 0.1174,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.03168843587239583,
+      "learning_rate": 0.0001,
+      "loss": 4.5465,
+      "loss/crossentropy": 2.1476733684539795,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2416895627975464,
+      "step": 5870
+    },
+    {
+      "epoch": 0.11744,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.027962239583333333,
+      "learning_rate": 0.0001,
+      "loss": 4.5633,
+      "loss/crossentropy": 1.74330335855484,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22001302987337112,
+      "step": 5872
+    },
+    {
+      "epoch": 0.11748,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.021507771809895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2725,
+      "loss/crossentropy": 1.8903921246528625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21950766444206238,
+      "step": 5874
+    },
+    {
+      "epoch": 0.11752,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.022069295247395832,
+      "learning_rate": 0.0001,
+      "loss": 4.3673,
+      "loss/crossentropy": 1.798406720161438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20271167159080505,
+      "step": 5876
+    },
+    {
+      "epoch": 0.11756,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.021089680989583335,
+      "learning_rate": 0.0001,
+      "loss": 4.7128,
+      "loss/crossentropy": 1.9651959538459778,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21256517618894577,
+      "step": 5878
+    },
+    {
+      "epoch": 0.1176,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.016844685872395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4158,
+      "loss/crossentropy": 2.039245307445526,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23225219547748566,
+      "step": 5880
+    },
+    {
+      "epoch": 0.11764,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.013329060872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.8741,
+      "loss/crossentropy": 2.500381350517273,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31757834553718567,
+      "step": 5882
+    },
+    {
+      "epoch": 0.11768,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.018195597330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.6463,
+      "loss/crossentropy": 2.0540305972099304,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2474198415875435,
+      "step": 5884
+    },
+    {
+      "epoch": 0.11772,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.016462198893229165,
+      "learning_rate": 0.0001,
+      "loss": 4.5047,
+      "loss/crossentropy": 2.072624385356903,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2482014298439026,
+      "step": 5886
+    },
+    {
+      "epoch": 0.11776,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.017853800455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.4025,
+      "loss/crossentropy": 2.153423309326172,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.234448604285717,
+      "step": 5888
+    },
+    {
+      "epoch": 0.1178,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.025325520833333334,
+      "learning_rate": 0.0001,
+      "loss": 4.2698,
+      "loss/crossentropy": 1.5941627621650696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20982372760772705,
+      "step": 5890
+    },
+    {
+      "epoch": 0.11784,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.025886027018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.5168,
+      "loss/crossentropy": 2.2858930826187134,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24936140328645706,
+      "step": 5892
+    },
+    {
+      "epoch": 0.11788,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.028076171875,
+      "learning_rate": 0.0001,
+      "loss": 4.536,
+      "loss/crossentropy": 2.333559274673462,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26152122020721436,
+      "step": 5894
+    },
+    {
+      "epoch": 0.11792,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.027106730143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.412,
+      "loss/crossentropy": 2.134613037109375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24929091334342957,
+      "step": 5896
+    },
+    {
+      "epoch": 0.11796,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.026423136393229168,
+      "learning_rate": 0.0001,
+      "loss": 4.4052,
+      "loss/crossentropy": 2.179764688014984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24523546546697617,
+      "step": 5898
+    },
+    {
+      "epoch": 0.118,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0204742431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3431,
+      "loss/crossentropy": 2.1184223294258118,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21715932339429855,
+      "step": 5900
+    },
+    {
+      "epoch": 0.11804,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.021882120768229166,
+      "learning_rate": 0.0001,
+      "loss": 4.5606,
+      "loss/crossentropy": 2.024593770503998,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23016374558210373,
+      "step": 5902
+    },
+    {
+      "epoch": 0.11808,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0204986572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3646,
+      "loss/crossentropy": 2.077186107635498,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2538782134652138,
+      "step": 5904
+    },
+    {
+      "epoch": 0.11812,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.013190714518229167,
+      "learning_rate": 0.0001,
+      "loss": 4.9135,
+      "loss/crossentropy": 2.2535945177078247,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25025132298469543,
+      "step": 5906
+    },
+    {
+      "epoch": 0.11816,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.01064453125,
+      "learning_rate": 0.0001,
+      "loss": 4.5512,
+      "loss/crossentropy": 2.5321284532546997,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2729681059718132,
+      "step": 5908
+    },
+    {
+      "epoch": 0.1182,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.006403605143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.332,
+      "loss/crossentropy": 2.043885111808777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23721691966056824,
+      "step": 5910
+    },
+    {
+      "epoch": 0.11824,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.005204264322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.3901,
+      "loss/crossentropy": 2.1343676447868347,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23325734585523605,
+      "step": 5912
+    },
+    {
+      "epoch": 0.11828,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0074045817057291664,
+      "learning_rate": 0.0001,
+      "loss": 4.3551,
+      "loss/crossentropy": 2.1164477467536926,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24526448547840118,
+      "step": 5914
+    },
+    {
+      "epoch": 0.11832,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0064737955729166664,
+      "learning_rate": 0.0001,
+      "loss": 4.5162,
+      "loss/crossentropy": 2.268216848373413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24837128818035126,
+      "step": 5916
+    },
+    {
+      "epoch": 0.11836,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0057037353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4636,
+      "loss/crossentropy": 2.074695885181427,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24956009536981583,
+      "step": 5918
+    },
+    {
+      "epoch": 0.1184,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.007372029622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3165,
+      "loss/crossentropy": 1.9369722604751587,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2316487580537796,
+      "step": 5920
+    },
+    {
+      "epoch": 0.11844,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.00611572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2789,
+      "loss/crossentropy": 2.2189531326293945,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23884039372205734,
+      "step": 5922
+    },
+    {
+      "epoch": 0.11848,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.007323201497395833,
+      "learning_rate": 0.0001,
+      "loss": 4.5965,
+      "loss/crossentropy": 2.3833028078079224,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2504591718316078,
+      "step": 5924
+    },
+    {
+      "epoch": 0.11852,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.007005818684895833,
+      "learning_rate": 0.0001,
+      "loss": 4.6346,
+      "loss/crossentropy": 2.0443845987319946,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23422807455062866,
+      "step": 5926
+    },
+    {
+      "epoch": 0.11856,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0073964436848958336,
+      "learning_rate": 0.0001,
+      "loss": 4.9061,
+      "loss/crossentropy": 2.223625063896179,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24985776841640472,
+      "step": 5928
+    },
+    {
+      "epoch": 0.1186,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.005126953125,
+      "learning_rate": 0.0001,
+      "loss": 4.493,
+      "loss/crossentropy": 2.1353545784950256,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22846727818250656,
+      "step": 5930
+    },
+    {
+      "epoch": 0.11864,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.005159505208333333,
+      "learning_rate": 0.0001,
+      "loss": 4.5786,
+      "loss/crossentropy": 2.0497827529907227,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21461189538240433,
+      "step": 5932
+    },
+    {
+      "epoch": 0.11868,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.004715983072916667,
+      "learning_rate": 0.0001,
+      "loss": 4.5071,
+      "loss/crossentropy": 1.9257362484931946,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2216043472290039,
+      "step": 5934
+    },
+    {
+      "epoch": 0.11872,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0034006754557291668,
+      "learning_rate": 0.0001,
+      "loss": 4.4212,
+      "loss/crossentropy": 2.0458216071128845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23162957280874252,
+      "step": 5936
+    },
+    {
+      "epoch": 0.11876,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0033274332682291666,
+      "learning_rate": 0.0001,
+      "loss": 4.6109,
+      "loss/crossentropy": 2.0786396861076355,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22303655743598938,
+      "step": 5938
+    },
+    {
+      "epoch": 0.1188,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.04327799479166667,
+      "learning_rate": 0.0001,
+      "loss": 4.4861,
+      "loss/crossentropy": 1.9800568222999573,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22236331552267075,
+      "step": 5940
+    },
+    {
+      "epoch": 0.11884,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0502349853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.056,
+      "loss/crossentropy": 1.882250189781189,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21664132922887802,
+      "step": 5942
+    },
+    {
+      "epoch": 0.11888,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.05025634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.5965,
+      "loss/crossentropy": 2.3682695627212524,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2698971778154373,
+      "step": 5944
+    },
+    {
+      "epoch": 0.11892,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0505523681640625,
+      "learning_rate": 0.0001,
+      "loss": 4.6364,
+      "loss/crossentropy": 2.225574493408203,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2642917186021805,
+      "step": 5946
+    },
+    {
+      "epoch": 0.11896,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.05032145182291667,
+      "learning_rate": 0.0001,
+      "loss": 4.3157,
+      "loss/crossentropy": 1.8634169697761536,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2170402556657791,
+      "step": 5948
+    },
+    {
+      "epoch": 0.119,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.05123291015625,
+      "learning_rate": 0.0001,
+      "loss": 5.0392,
+      "loss/crossentropy": 2.4500025510787964,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27845144271850586,
+      "step": 5950
+    },
+    {
+      "epoch": 0.11904,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.054108683268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.1179,
+      "loss/crossentropy": 2.1532052755355835,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24952176213264465,
+      "step": 5952
+    },
+    {
+      "epoch": 0.11908,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0561676025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2592,
+      "loss/crossentropy": 2.066560387611389,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2356283888220787,
+      "step": 5954
+    },
+    {
+      "epoch": 0.11912,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.01021728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.6187,
+      "loss/crossentropy": 1.9679089784622192,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23729706555604935,
+      "step": 5956
+    },
+    {
+      "epoch": 0.11916,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0071441650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3463,
+      "loss/crossentropy": 2.3490394353866577,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26400282233953476,
+      "step": 5958
+    },
+    {
+      "epoch": 0.1192,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.008576456705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3629,
+      "loss/crossentropy": 2.145757555961609,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23329483717679977,
+      "step": 5960
+    },
+    {
+      "epoch": 0.11924,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.005399576822916667,
+      "learning_rate": 0.0001,
+      "loss": 4.2376,
+      "loss/crossentropy": 2.0764617919921875,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22371648252010345,
+      "step": 5962
+    },
+    {
+      "epoch": 0.11928,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.010380045572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.8106,
+      "loss/crossentropy": 2.199389696121216,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23210449516773224,
+      "step": 5964
+    },
+    {
+      "epoch": 0.11932,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.01064453125,
+      "learning_rate": 0.0001,
+      "loss": 4.3515,
+      "loss/crossentropy": 1.9008439183235168,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21474837511777878,
+      "step": 5966
+    },
+    {
+      "epoch": 0.11936,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0148590087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4603,
+      "loss/crossentropy": 2.1779539585113525,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23485051095485687,
+      "step": 5968
+    },
+    {
+      "epoch": 0.1194,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.014176432291666667,
+      "learning_rate": 0.0001,
+      "loss": 4.428,
+      "loss/crossentropy": 2.267147421836853,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24924689531326294,
+      "step": 5970
+    },
+    {
+      "epoch": 0.11944,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.016292317708333334,
+      "learning_rate": 0.0001,
+      "loss": 4.0856,
+      "loss/crossentropy": 2.0618110299110413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24352984875440598,
+      "step": 5972
+    },
+    {
+      "epoch": 0.11948,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.023758951822916666,
+      "learning_rate": 0.0001,
+      "loss": 4.5802,
+      "loss/crossentropy": 2.1419676542282104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23680058121681213,
+      "step": 5974
+    },
+    {
+      "epoch": 0.11952,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.023368326822916667,
+      "learning_rate": 0.0001,
+      "loss": 4.6554,
+      "loss/crossentropy": 2.0376622080802917,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22515031695365906,
+      "step": 5976
+    },
+    {
+      "epoch": 0.11956,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.025739542643229165,
+      "learning_rate": 0.0001,
+      "loss": 4.4213,
+      "loss/crossentropy": 1.7675965428352356,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.219834603369236,
+      "step": 5978
+    },
+    {
+      "epoch": 0.1196,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.024030558268229165,
+      "learning_rate": 0.0001,
+      "loss": 4.2886,
+      "loss/crossentropy": 1.8919905424118042,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2256327122449875,
+      "step": 5980
+    },
+    {
+      "epoch": 0.11964,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0228424072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.6148,
+      "loss/crossentropy": 2.287980794906616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25571687519550323,
+      "step": 5982
+    },
+    {
+      "epoch": 0.11968,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0169830322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3527,
+      "loss/crossentropy": 2.1030094027519226,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23401429504156113,
+      "step": 5984
+    },
+    {
+      "epoch": 0.11972,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.0166412353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.8857,
+      "loss/crossentropy": 2.469533920288086,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2548582851886749,
+      "step": 5986
+    },
+    {
+      "epoch": 0.11976,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.018047841389973958,
+      "learning_rate": 0.0001,
+      "loss": 4.167,
+      "loss/crossentropy": 2.140324354171753,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22603372484445572,
+      "step": 5988
+    },
+    {
+      "epoch": 0.1198,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.015773264567057292,
+      "learning_rate": 0.0001,
+      "loss": 4.3848,
+      "loss/crossentropy": 2.1848061084747314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22712922096252441,
+      "step": 5990
+    },
+    {
+      "epoch": 0.11984,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.015380605061848959,
+      "learning_rate": 0.0001,
+      "loss": 4.6457,
+      "loss/crossentropy": 2.2872358560562134,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2352529615163803,
+      "step": 5992
+    },
+    {
+      "epoch": 0.11988,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.012318674723307292,
+      "learning_rate": 0.0001,
+      "loss": 4.6136,
+      "loss/crossentropy": 2.082811713218689,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24107889831066132,
+      "step": 5994
+    },
+    {
+      "epoch": 0.11992,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.013602447509765626,
+      "learning_rate": 0.0001,
+      "loss": 4.2089,
+      "loss/crossentropy": 2.23664391040802,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23429522663354874,
+      "step": 5996
+    },
+    {
+      "epoch": 0.11996,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.013242340087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5237,
+      "loss/crossentropy": 2.451270341873169,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2541816979646683,
+      "step": 5998
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.013561757405598958,
+      "learning_rate": 0.0001,
+      "loss": 4.5559,
+      "loss/crossentropy": 2.1744157671928406,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22349119931459427,
+      "step": 6000
+    },
+    {
+      "epoch": 0.12004,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.013171132405598958,
+      "learning_rate": 0.0001,
+      "loss": 4.5532,
+      "loss/crossentropy": 2.0316836833953857,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.259635865688324,
+      "step": 6002
+    },
+    {
+      "epoch": 0.12008,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.022337849934895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3674,
+      "loss/crossentropy": 2.0989437103271484,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25056491047143936,
+      "step": 6004
+    },
+    {
+      "epoch": 0.12012,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.02633056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0883,
+      "loss/crossentropy": 1.9609100818634033,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22969383746385574,
+      "step": 6006
+    },
+    {
+      "epoch": 0.12016,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.027665201822916666,
+      "learning_rate": 0.0001,
+      "loss": 4.2094,
+      "loss/crossentropy": 2.077883243560791,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22850078344345093,
+      "step": 6008
+    },
+    {
+      "epoch": 0.1202,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 1.6465779622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.5659,
+      "loss/crossentropy": 1.7967591285705566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2126556932926178,
+      "step": 6010
+    },
+    {
+      "epoch": 0.12024,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 1.62222900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.4018,
+      "loss/crossentropy": 1.6516226530075073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2188895046710968,
+      "step": 6012
+    },
+    {
+      "epoch": 0.12028,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 1.62984619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3303,
+      "loss/crossentropy": 2.161388635635376,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23447516560554504,
+      "step": 6014
+    },
+    {
+      "epoch": 0.12032,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 1.6235636393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.5112,
+      "loss/crossentropy": 1.9205461740493774,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23797179013490677,
+      "step": 6016
+    },
+    {
+      "epoch": 0.12036,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 1.616844685872396,
+      "learning_rate": 0.0001,
+      "loss": 4.8019,
+      "loss/crossentropy": 2.025223135948181,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25120896100997925,
+      "step": 6018
+    },
+    {
+      "epoch": 0.1204,
+      "grad_norm": 2.25,
+      "grad_norm_var": 1.6347005208333334,
+      "learning_rate": 0.0001,
+      "loss": 4.4819,
+      "loss/crossentropy": 1.957942008972168,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22228525578975677,
+      "step": 6020
+    },
+    {
+      "epoch": 0.12044,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 1.674201456705729,
+      "learning_rate": 0.0001,
+      "loss": 4.6193,
+      "loss/crossentropy": 2.3325445652008057,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26250994950532913,
+      "step": 6022
+    },
+    {
+      "epoch": 0.12048,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 1.6957997639973958,
+      "learning_rate": 0.0001,
+      "loss": 4.264,
+      "loss/crossentropy": 2.0049667954444885,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22990728169679642,
+      "step": 6024
+    },
+    {
+      "epoch": 0.12052,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.08680013020833334,
+      "learning_rate": 0.0001,
+      "loss": 4.5996,
+      "loss/crossentropy": 2.0473387241363525,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23628919571638107,
+      "step": 6026
+    },
+    {
+      "epoch": 0.12056,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.09107640584309896,
+      "learning_rate": 0.0001,
+      "loss": 4.109,
+      "loss/crossentropy": 2.013141930103302,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21728236973285675,
+      "step": 6028
+    },
+    {
+      "epoch": 0.1206,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.09145278930664062,
+      "learning_rate": 0.0001,
+      "loss": 4.3869,
+      "loss/crossentropy": 2.1269132494926453,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25139085948467255,
+      "step": 6030
+    },
+    {
+      "epoch": 0.12064,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.09058405558268229,
+      "learning_rate": 0.0001,
+      "loss": 4.5568,
+      "loss/crossentropy": 2.5267512798309326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27998340129852295,
+      "step": 6032
+    },
+    {
+      "epoch": 0.12068,
+      "grad_norm": 2.78125,
+      "grad_norm_var": 0.10746027628580729,
+      "learning_rate": 0.0001,
+      "loss": 4.2615,
+      "loss/crossentropy": 1.8502249717712402,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22863731533288956,
+      "step": 6034
+    },
+    {
+      "epoch": 0.12072,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.10850601196289063,
+      "learning_rate": 0.0001,
+      "loss": 4.2197,
+      "loss/crossentropy": 1.7754456400871277,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23809141665697098,
+      "step": 6036
+    },
+    {
+      "epoch": 0.12076,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.033760325113932295,
+      "learning_rate": 0.0001,
+      "loss": 4.4714,
+      "loss/crossentropy": 1.9596920609474182,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2111937776207924,
+      "step": 6038
+    },
+    {
+      "epoch": 0.1208,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.02995580037434896,
+      "learning_rate": 0.0001,
+      "loss": 4.3079,
+      "loss/crossentropy": 1.888563334941864,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21698038280010223,
+      "step": 6040
+    },
+    {
+      "epoch": 0.12084,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.031404368082682294,
+      "learning_rate": 0.0001,
+      "loss": 4.2412,
+      "loss/crossentropy": 2.1646993160247803,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22735021263360977,
+      "step": 6042
+    },
+    {
+      "epoch": 0.12088,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0258941650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.525,
+      "loss/crossentropy": 2.0790343284606934,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2193024456501007,
+      "step": 6044
+    },
+    {
+      "epoch": 0.12092,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.025211588541666666,
+      "learning_rate": 0.0001,
+      "loss": 4.3538,
+      "loss/crossentropy": 2.2733768224716187,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.249516561627388,
+      "step": 6046
+    },
+    {
+      "epoch": 0.12096,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.02431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.7286,
+      "loss/crossentropy": 2.5003533363342285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22928690910339355,
+      "step": 6048
+    },
+    {
+      "epoch": 0.121,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.03629150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.6269,
+      "loss/crossentropy": 1.9606900215148926,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2152085080742836,
+      "step": 6050
+    },
+    {
+      "epoch": 0.12104,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.0384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.5724,
+      "loss/crossentropy": 2.266395926475525,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26055608689785004,
+      "step": 6052
+    },
+    {
+      "epoch": 0.12108,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.037821451822916664,
+      "learning_rate": 0.0001,
+      "loss": 4.5841,
+      "loss/crossentropy": 2.1753041744232178,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2327822595834732,
+      "step": 6054
+    },
+    {
+      "epoch": 0.12112,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.041304524739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.2198,
+      "loss/crossentropy": 1.775630235671997,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2073052078485489,
+      "step": 6056
+    },
+    {
+      "epoch": 0.12116,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.03975321451822917,
+      "learning_rate": 0.0001,
+      "loss": 4.486,
+      "loss/crossentropy": 2.415855050086975,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.270721398293972,
+      "step": 6058
+    },
+    {
+      "epoch": 0.1212,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.039453125,
+      "learning_rate": 0.0001,
+      "loss": 4.6647,
+      "loss/crossentropy": 2.2122162580490112,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24723708629608154,
+      "step": 6060
+    },
+    {
+      "epoch": 0.12124,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.041792805989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.3803,
+      "loss/crossentropy": 1.9540830850601196,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21385761350393295,
+      "step": 6062
+    },
+    {
+      "epoch": 0.12128,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.0413726806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.979,
+      "loss/crossentropy": 1.923313319683075,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22986505925655365,
+      "step": 6064
+    },
+    {
+      "epoch": 0.12132,
+      "grad_norm": 3.03125,
+      "grad_norm_var": 0.05182291666666667,
+      "learning_rate": 0.0001,
+      "loss": 4.6895,
+      "loss/crossentropy": 2.4042444229125977,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30044034123420715,
+      "step": 6066
+    },
+    {
+      "epoch": 0.12136,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.049225870768229166,
+      "learning_rate": 0.0001,
+      "loss": 4.5857,
+      "loss/crossentropy": 2.2610549926757812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2628016769886017,
+      "step": 6068
+    },
+    {
+      "epoch": 0.1214,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0513671875,
+      "learning_rate": 0.0001,
+      "loss": 4.5024,
+      "loss/crossentropy": 1.9100797176361084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22663169354200363,
+      "step": 6070
+    },
+    {
+      "epoch": 0.12144,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.04830729166666667,
+      "learning_rate": 0.0001,
+      "loss": 4.5916,
+      "loss/crossentropy": 2.3971651792526245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26831263303756714,
+      "step": 6072
+    },
+    {
+      "epoch": 0.12148,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.0492095947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4553,
+      "loss/crossentropy": 2.106821596622467,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23786011338233948,
+      "step": 6074
+    },
+    {
+      "epoch": 0.12152,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.052469889322916664,
+      "learning_rate": 0.0001,
+      "loss": 4.4225,
+      "loss/crossentropy": 2.1920535564422607,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23812127113342285,
+      "step": 6076
+    },
+    {
+      "epoch": 0.12156,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.049088541666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.2314,
+      "loss/crossentropy": 2.3014419078826904,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2380141019821167,
+      "step": 6078
+    },
+    {
+      "epoch": 0.1216,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.04903055826822917,
+      "learning_rate": 0.0001,
+      "loss": 4.4558,
+      "loss/crossentropy": 2.1721781492233276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24158813059329987,
+      "step": 6080
+    },
+    {
+      "epoch": 0.12164,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.013232421875,
+      "learning_rate": 0.0001,
+      "loss": 4.5192,
+      "loss/crossentropy": 2.1230934858322144,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23764102160930634,
+      "step": 6082
+    },
+    {
+      "epoch": 0.12168,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.0166412353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2147,
+      "loss/crossentropy": 1.9570311307907104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24687693268060684,
+      "step": 6084
+    },
+    {
+      "epoch": 0.12172,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.014997355143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2063,
+      "loss/crossentropy": 2.272383213043213,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2567671462893486,
+      "step": 6086
+    },
+    {
+      "epoch": 0.12176,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.009598795572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.645,
+      "loss/crossentropy": 2.216492176055908,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24377377331256866,
+      "step": 6088
+    },
+    {
+      "epoch": 0.1218,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.007201131184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.5956,
+      "loss/crossentropy": 2.25177001953125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23176074773073196,
+      "step": 6090
+    },
+    {
+      "epoch": 0.12184,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.009098307291666666,
+      "learning_rate": 0.0001,
+      "loss": 4.1966,
+      "loss/crossentropy": 2.2452452182769775,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24528006464242935,
+      "step": 6092
+    },
+    {
+      "epoch": 0.12188,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0091949462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3548,
+      "loss/crossentropy": 2.078445553779602,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2276337668299675,
+      "step": 6094
+    },
+    {
+      "epoch": 0.12192,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.00982666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.4435,
+      "loss/crossentropy": 2.2938032150268555,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2784377336502075,
+      "step": 6096
+    },
+    {
+      "epoch": 0.12196,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0136627197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1588,
+      "loss/crossentropy": 2.028180480003357,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23716723918914795,
+      "step": 6098
+    },
+    {
+      "epoch": 0.122,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.020580037434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.6345,
+      "loss/crossentropy": 2.4959323406219482,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2510451450943947,
+      "step": 6100
+    },
+    {
+      "epoch": 0.12204,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.023856608072916667,
+      "learning_rate": 0.0001,
+      "loss": 4.6639,
+      "loss/crossentropy": 2.186043620109558,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23511512577533722,
+      "step": 6102
+    },
+    {
+      "epoch": 0.12208,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.024442545572916665,
+      "learning_rate": 0.0001,
+      "loss": 4.6727,
+      "loss/crossentropy": 2.6631078720092773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25689610838890076,
+      "step": 6104
+    },
+    {
+      "epoch": 0.12212,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.025609334309895832,
+      "learning_rate": 0.0001,
+      "loss": 4.7943,
+      "loss/crossentropy": 2.310486674308777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26937828958034515,
+      "step": 6106
+    },
+    {
+      "epoch": 0.12216,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.0247711181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.9892,
+      "loss/crossentropy": 2.2036240100860596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24968907237052917,
+      "step": 6108
+    },
+    {
+      "epoch": 0.1222,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.058251953125,
+      "learning_rate": 0.0001,
+      "loss": 4.3527,
+      "loss/crossentropy": 1.9434874057769775,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23247701674699783,
+      "step": 6110
+    },
+    {
+      "epoch": 0.12224,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.05579020182291667,
+      "learning_rate": 0.0001,
+      "loss": 4.4149,
+      "loss/crossentropy": 2.0525330305099487,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24810528755187988,
+      "step": 6112
+    },
+    {
+      "epoch": 0.12228,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.04482320149739583,
+      "learning_rate": 0.0001,
+      "loss": 4.5205,
+      "loss/crossentropy": 2.0085532665252686,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2514277398586273,
+      "step": 6114
+    },
+    {
+      "epoch": 0.12232,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.04120686848958333,
+      "learning_rate": 0.0001,
+      "loss": 4.5745,
+      "loss/crossentropy": 1.7393967509269714,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2239851951599121,
+      "step": 6116
+    },
+    {
+      "epoch": 0.12236,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.04327799479166667,
+      "learning_rate": 0.0001,
+      "loss": 4.7332,
+      "loss/crossentropy": 1.8714343905448914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22823140025138855,
+      "step": 6118
+    },
+    {
+      "epoch": 0.1224,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.04192708333333333,
+      "learning_rate": 0.0001,
+      "loss": 4.6177,
+      "loss/crossentropy": 1.9353562593460083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22261886298656464,
+      "step": 6120
+    },
+    {
+      "epoch": 0.12244,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0469635009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3154,
+      "loss/crossentropy": 2.0409420132637024,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2319282591342926,
+      "step": 6122
+    },
+    {
+      "epoch": 0.12248,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.04882405598958333,
+      "learning_rate": 0.0001,
+      "loss": 4.4764,
+      "loss/crossentropy": 2.2461307048797607,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25047267973423004,
+      "step": 6124
+    },
+    {
+      "epoch": 0.12252,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.014012654622395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4863,
+      "loss/crossentropy": 2.1060246229171753,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24702349305152893,
+      "step": 6126
+    },
+    {
+      "epoch": 0.12256,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.01718724568684896,
+      "learning_rate": 0.0001,
+      "loss": 3.9093,
+      "loss/crossentropy": 1.7481068968772888,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19611438363790512,
+      "step": 6128
+    },
+    {
+      "epoch": 0.1226,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.017329661051432292,
+      "learning_rate": 0.0001,
+      "loss": 4.5506,
+      "loss/crossentropy": 1.8341861963272095,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20660528540611267,
+      "step": 6130
+    },
+    {
+      "epoch": 0.12264,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.018507639567057293,
+      "learning_rate": 0.0001,
+      "loss": 4.3388,
+      "loss/crossentropy": 1.7332024574279785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21615543216466904,
+      "step": 6132
+    },
+    {
+      "epoch": 0.12268,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.013203684488932292,
+      "learning_rate": 0.0001,
+      "loss": 4.3027,
+      "loss/crossentropy": 1.8959643244743347,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.219951793551445,
+      "step": 6134
+    },
+    {
+      "epoch": 0.12272,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.012601470947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.6945,
+      "loss/crossentropy": 2.234723210334778,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2342153787612915,
+      "step": 6136
+    },
+    {
+      "epoch": 0.12276,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.011637115478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.7394,
+      "loss/crossentropy": 2.2297592759132385,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24097590148448944,
+      "step": 6138
+    },
+    {
+      "epoch": 0.1228,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.013586171468098958,
+      "learning_rate": 0.0001,
+      "loss": 4.592,
+      "loss/crossentropy": 2.013442814350128,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2532622739672661,
+      "step": 6140
+    },
+    {
+      "epoch": 0.12284,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.014833323160807292,
+      "learning_rate": 0.0001,
+      "loss": 4.417,
+      "loss/crossentropy": 2.3066656589508057,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24480555206537247,
+      "step": 6142
+    },
+    {
+      "epoch": 0.12288,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.014241536458333334,
+      "learning_rate": 0.0001,
+      "loss": 4.4156,
+      "loss/crossentropy": 1.9281310439109802,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22363708168268204,
+      "step": 6144
+    },
+    {
+      "epoch": 0.12292,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.016792805989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.4314,
+      "loss/crossentropy": 1.965875267982483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22153983265161514,
+      "step": 6146
+    },
+    {
+      "epoch": 0.12296,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.03536783854166667,
+      "learning_rate": 0.0001,
+      "loss": 4.6139,
+      "loss/crossentropy": 1.811126947402954,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21779045462608337,
+      "step": 6148
+    },
+    {
+      "epoch": 0.123,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.0311676025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.4855,
+      "loss/crossentropy": 2.5114762783050537,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2542252689599991,
+      "step": 6150
+    },
+    {
+      "epoch": 0.12304,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.028271484375,
+      "learning_rate": 0.0001,
+      "loss": 4.1269,
+      "loss/crossentropy": 1.8784565925598145,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24637068808078766,
+      "step": 6152
+    },
+    {
+      "epoch": 0.12308,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.028685506184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3377,
+      "loss/crossentropy": 1.6900760531425476,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2170872688293457,
+      "step": 6154
+    },
+    {
+      "epoch": 0.12312,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.02867431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.5705,
+      "loss/crossentropy": 2.413783550262451,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2624947279691696,
+      "step": 6156
+    },
+    {
+      "epoch": 0.12316,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.030126953125,
+      "learning_rate": 0.0001,
+      "loss": 4.3463,
+      "loss/crossentropy": 2.0598058104515076,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2071695551276207,
+      "step": 6158
+    },
+    {
+      "epoch": 0.1232,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.029781087239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.6702,
+      "loss/crossentropy": 2.0407246947288513,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24353720247745514,
+      "step": 6160
+    },
+    {
+      "epoch": 0.12324,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.028450520833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.3116,
+      "loss/crossentropy": 1.9608840346336365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20965111255645752,
+      "step": 6162
+    },
+    {
+      "epoch": 0.12328,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.01640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2146,
+      "loss/crossentropy": 2.0231454372406006,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.219897098839283,
+      "step": 6164
+    },
+    {
+      "epoch": 0.12332,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.015999348958333333,
+      "learning_rate": 0.0001,
+      "loss": 4.614,
+      "loss/crossentropy": 2.217663288116455,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2466331645846367,
+      "step": 6166
+    },
+    {
+      "epoch": 0.12336,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.01597900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.4135,
+      "loss/crossentropy": 1.6825732588768005,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1999632865190506,
+      "step": 6168
+    },
+    {
+      "epoch": 0.1234,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.020897420247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.4552,
+      "loss/crossentropy": 2.288950800895691,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26286834478378296,
+      "step": 6170
+    },
+    {
+      "epoch": 0.12344,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.023225911458333335,
+      "learning_rate": 0.0001,
+      "loss": 4.3376,
+      "loss/crossentropy": 1.8202561140060425,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22252517193555832,
+      "step": 6172
+    },
+    {
+      "epoch": 0.12348,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.019710286458333334,
+      "learning_rate": 0.0001,
+      "loss": 4.5023,
+      "loss/crossentropy": 2.1007159948349,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24983422458171844,
+      "step": 6174
+    },
+    {
+      "epoch": 0.12352,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0199371337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4513,
+      "loss/crossentropy": 2.0953084230422974,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25545743107795715,
+      "step": 6176
+    },
+    {
+      "epoch": 0.12356,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.019001261393229166,
+      "learning_rate": 0.0001,
+      "loss": 4.4315,
+      "loss/crossentropy": 2.107246518135071,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22741339355707169,
+      "step": 6178
+    },
+    {
+      "epoch": 0.1236,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.013118489583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.4982,
+      "loss/crossentropy": 2.1219626665115356,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22499094158411026,
+      "step": 6180
+    },
+    {
+      "epoch": 0.12364,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.013093058268229167,
+      "learning_rate": 0.0001,
+      "loss": 4.4263,
+      "loss/crossentropy": 1.8892266154289246,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2255062311887741,
+      "step": 6182
+    },
+    {
+      "epoch": 0.12368,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.015315755208333334,
+      "learning_rate": 0.0001,
+      "loss": 4.5884,
+      "loss/crossentropy": 2.0765860080718994,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24712087213993073,
+      "step": 6184
+    },
+    {
+      "epoch": 0.12372,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.010074869791666666,
+      "learning_rate": 0.0001,
+      "loss": 4.5431,
+      "loss/crossentropy": 2.301337718963623,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24419523775577545,
+      "step": 6186
+    },
+    {
+      "epoch": 0.12376,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.008373006184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.5043,
+      "loss/crossentropy": 2.0489712953567505,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25768278539180756,
+      "step": 6188
+    },
+    {
+      "epoch": 0.1238,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0128082275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2654,
+      "loss/crossentropy": 1.9328826069831848,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22916094958782196,
+      "step": 6190
+    },
+    {
+      "epoch": 0.12384,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.01171875,
+      "learning_rate": 0.0001,
+      "loss": 4.1574,
+      "loss/crossentropy": 1.6330446004867554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20600398629903793,
+      "step": 6192
+    },
+    {
+      "epoch": 0.12388,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0111724853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4769,
+      "loss/crossentropy": 1.988203227519989,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22711507230997086,
+      "step": 6194
+    },
+    {
+      "epoch": 0.12392,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.01590143839518229,
+      "learning_rate": 0.0001,
+      "loss": 4.125,
+      "loss/crossentropy": 2.036049246788025,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2110549360513687,
+      "step": 6196
+    },
+    {
+      "epoch": 0.12396,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.01624120076497396,
+      "learning_rate": 0.0001,
+      "loss": 4.5546,
+      "loss/crossentropy": 2.217681884765625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23705793917179108,
+      "step": 6198
+    },
+    {
+      "epoch": 0.124,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.015276845296223958,
+      "learning_rate": 0.0001,
+      "loss": 4.4803,
+      "loss/crossentropy": 2.3878824710845947,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24000737071037292,
+      "step": 6200
+    },
+    {
+      "epoch": 0.12404,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.015852610270182293,
+      "learning_rate": 0.0001,
+      "loss": 4.4754,
+      "loss/crossentropy": 2.2790249586105347,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26150786131620407,
+      "step": 6202
+    },
+    {
+      "epoch": 0.12408,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.015036773681640626,
+      "learning_rate": 0.0001,
+      "loss": 4.4703,
+      "loss/crossentropy": 2.251123785972595,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25231797993183136,
+      "step": 6204
+    },
+    {
+      "epoch": 0.12412,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.010802968343098959,
+      "learning_rate": 0.0001,
+      "loss": 4.5294,
+      "loss/crossentropy": 1.8977670073509216,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21783004701137543,
+      "step": 6206
+    },
+    {
+      "epoch": 0.12416,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.012894439697265624,
+      "learning_rate": 0.0001,
+      "loss": 4.6458,
+      "loss/crossentropy": 2.385319232940674,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29223111271858215,
+      "step": 6208
+    },
+    {
+      "epoch": 0.1242,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.015964508056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.6555,
+      "loss/crossentropy": 1.9274529218673706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22827968001365662,
+      "step": 6210
+    },
+    {
+      "epoch": 0.12424,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.01226806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.8232,
+      "loss/crossentropy": 2.1861478090286255,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2424854040145874,
+      "step": 6212
+    },
+    {
+      "epoch": 0.12428,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.01207275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3002,
+      "loss/crossentropy": 2.2234357595443726,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2471691370010376,
+      "step": 6214
+    },
+    {
+      "epoch": 0.12432,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.022541300455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.5216,
+      "loss/crossentropy": 2.365513563156128,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2640291824936867,
+      "step": 6216
+    },
+    {
+      "epoch": 0.12436,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.021955362955729165,
+      "learning_rate": 0.0001,
+      "loss": 4.282,
+      "loss/crossentropy": 1.964316964149475,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22341662645339966,
+      "step": 6218
+    },
+    {
+      "epoch": 0.1244,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.021805826822916666,
+      "learning_rate": 0.0001,
+      "loss": 4.7078,
+      "loss/crossentropy": 2.3704408407211304,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24599966406822205,
+      "step": 6220
+    },
+    {
+      "epoch": 0.12444,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0193756103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.7558,
+      "loss/crossentropy": 2.1461408138275146,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25205330550670624,
+      "step": 6222
+    },
+    {
+      "epoch": 0.12448,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.018745930989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.7032,
+      "loss/crossentropy": 2.3997104167938232,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23371660709381104,
+      "step": 6224
+    },
+    {
+      "epoch": 0.12452,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1842,
+      "loss/crossentropy": 1.7976875305175781,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21662414073944092,
+      "step": 6226
+    },
+    {
+      "epoch": 0.12456,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.02008056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2214,
+      "loss/crossentropy": 1.9929583668708801,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21378497034311295,
+      "step": 6228
+    },
+    {
+      "epoch": 0.1246,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.02017822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2982,
+      "loss/crossentropy": 1.8853623867034912,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21267645806074142,
+      "step": 6230
+    },
+    {
+      "epoch": 0.12464,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.008454386393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2638,
+      "loss/crossentropy": 2.2534812688827515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23003943264484406,
+      "step": 6232
+    },
+    {
+      "epoch": 0.12468,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.010863240559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2504,
+      "loss/crossentropy": 2.026564121246338,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2172931283712387,
+      "step": 6234
+    },
+    {
+      "epoch": 0.12472,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.01021728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3555,
+      "loss/crossentropy": 1.9323118925094604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22170638293027878,
+      "step": 6236
+    },
+    {
+      "epoch": 0.12476,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.007938639322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.479,
+      "loss/crossentropy": 2.056011915206909,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2656880244612694,
+      "step": 6238
+    },
+    {
+      "epoch": 0.1248,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.00830078125,
+      "learning_rate": 0.0001,
+      "loss": 4.6288,
+      "loss/crossentropy": 2.095108926296234,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23341026157140732,
+      "step": 6240
+    },
+    {
+      "epoch": 0.12484,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.011555989583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.737,
+      "loss/crossentropy": 2.0198334455490112,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24095112830400467,
+      "step": 6242
+    },
+    {
+      "epoch": 0.12488,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.012105305989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.3745,
+      "loss/crossentropy": 2.0397544503211975,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22567399591207504,
+      "step": 6244
+    },
+    {
+      "epoch": 0.12492,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0120513916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3633,
+      "loss/crossentropy": 1.9094319343566895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22690637409687042,
+      "step": 6246
+    },
+    {
+      "epoch": 0.12496,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.011295572916666666,
+      "learning_rate": 0.0001,
+      "loss": 4.6474,
+      "loss/crossentropy": 2.27765429019928,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25075703859329224,
+      "step": 6248
+    },
+    {
+      "epoch": 0.125,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.017023722330729168,
+      "learning_rate": 0.0001,
+      "loss": 4.5769,
+      "loss/crossentropy": 2.0027456283569336,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24267998337745667,
+      "step": 6250
+    },
+    {
+      "epoch": 0.12504,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.017723592122395833,
+      "learning_rate": 0.0001,
+      "loss": 4.5398,
+      "loss/crossentropy": 2.276857614517212,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24048637598752975,
+      "step": 6252
+    },
+    {
+      "epoch": 0.12508,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.018065388997395834,
+      "learning_rate": 0.0001,
+      "loss": 4.7603,
+      "loss/crossentropy": 2.1234214305877686,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23481453210115433,
+      "step": 6254
+    },
+    {
+      "epoch": 0.12512,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.019624837239583335,
+      "learning_rate": 0.0001,
+      "loss": 4.7537,
+      "loss/crossentropy": 2.201158881187439,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23473594337701797,
+      "step": 6256
+    },
+    {
+      "epoch": 0.12516,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.017220052083333333,
+      "learning_rate": 0.0001,
+      "loss": 4.7202,
+      "loss/crossentropy": 2.3437804579734802,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24881581962108612,
+      "step": 6258
+    },
+    {
+      "epoch": 0.1252,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.015461222330729166,
+      "learning_rate": 0.0001,
+      "loss": 4.5938,
+      "loss/crossentropy": 2.0984586477279663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24332843720912933,
+      "step": 6260
+    },
+    {
+      "epoch": 0.12524,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.016942342122395832,
+      "learning_rate": 0.0001,
+      "loss": 4.2317,
+      "loss/crossentropy": 1.7946080565452576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21510492265224457,
+      "step": 6262
+    },
+    {
+      "epoch": 0.12528,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.016161092122395835,
+      "learning_rate": 0.0001,
+      "loss": 4.3889,
+      "loss/crossentropy": 2.022711932659149,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22890903800725937,
+      "step": 6264
+    },
+    {
+      "epoch": 0.12532,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.009577433268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.288,
+      "loss/crossentropy": 1.9752087593078613,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21972095221281052,
+      "step": 6266
+    },
+    {
+      "epoch": 0.12536,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.008625284830729166,
+      "learning_rate": 0.0001,
+      "loss": 4.4952,
+      "loss/crossentropy": 1.7267251014709473,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21305006742477417,
+      "step": 6268
+    },
+    {
+      "epoch": 0.1254,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.008869425455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2164,
+      "loss/crossentropy": 2.249786615371704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24875187873840332,
+      "step": 6270
+    },
+    {
+      "epoch": 0.12544,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.012190755208333333,
+      "learning_rate": 0.0001,
+      "loss": 4.693,
+      "loss/crossentropy": 2.402994990348816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25536587834358215,
+      "step": 6272
+    },
+    {
+      "epoch": 0.12548,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.016243489583333333,
+      "learning_rate": 0.0001,
+      "loss": 4.606,
+      "loss/crossentropy": 2.240913987159729,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3013365715742111,
+      "step": 6274
+    },
+    {
+      "epoch": 0.12552,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.019071451822916665,
+      "learning_rate": 0.0001,
+      "loss": 4.4691,
+      "loss/crossentropy": 2.0767332911491394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22113215178251266,
+      "step": 6276
+    },
+    {
+      "epoch": 0.12556,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.02041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.608,
+      "loss/crossentropy": 1.8625048995018005,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20321352779865265,
+      "step": 6278
+    },
+    {
+      "epoch": 0.1256,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.024072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0386,
+      "loss/crossentropy": 1.9143638610839844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22085773944854736,
+      "step": 6280
+    },
+    {
+      "epoch": 0.12564,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.023192342122395834,
+      "learning_rate": 0.0001,
+      "loss": 4.372,
+      "loss/crossentropy": 2.3756210803985596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26027245819568634,
+      "step": 6282
+    },
+    {
+      "epoch": 0.12568,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.02427978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5526,
+      "loss/crossentropy": 1.9310896396636963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2086925357580185,
+      "step": 6284
+    },
+    {
+      "epoch": 0.12572,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.021675618489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.4751,
+      "loss/crossentropy": 2.1757423877716064,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.241075336933136,
+      "step": 6286
+    },
+    {
+      "epoch": 0.12576,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.021284993489583334,
+      "learning_rate": 0.0001,
+      "loss": 4.0946,
+      "loss/crossentropy": 2.3057546615600586,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24882248044013977,
+      "step": 6288
+    },
+    {
+      "epoch": 0.1258,
+      "grad_norm": 3.140625,
+      "grad_norm_var": 0.06883036295572917,
+      "learning_rate": 0.0001,
+      "loss": 4.5211,
+      "loss/crossentropy": 2.1551633477211,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2796258181333542,
+      "step": 6290
+    },
+    {
+      "epoch": 0.12584,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.06608072916666667,
+      "learning_rate": 0.0001,
+      "loss": 4.4915,
+      "loss/crossentropy": 2.0627459287643433,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24781616777181625,
+      "step": 6292
+    },
+    {
+      "epoch": 0.12588,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.06463216145833334,
+      "learning_rate": 0.0001,
+      "loss": 4.5896,
+      "loss/crossentropy": 1.795321524143219,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2113223671913147,
+      "step": 6294
+    },
+    {
+      "epoch": 0.12592,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.06004130045572917,
+      "learning_rate": 0.0001,
+      "loss": 4.4988,
+      "loss/crossentropy": 2.323319673538208,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24748852849006653,
+      "step": 6296
+    },
+    {
+      "epoch": 0.12596,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.05998942057291667,
+      "learning_rate": 0.0001,
+      "loss": 4.4801,
+      "loss/crossentropy": 2.053200662136078,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2278411090373993,
+      "step": 6298
+    },
+    {
+      "epoch": 0.126,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.060155232747395836,
+      "learning_rate": 0.0001,
+      "loss": 4.4605,
+      "loss/crossentropy": 2.102539896965027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26071713864803314,
+      "step": 6300
+    },
+    {
+      "epoch": 0.12604,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.059554036458333334,
+      "learning_rate": 0.0001,
+      "loss": 4.4613,
+      "loss/crossentropy": 1.9316620826721191,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23007915914058685,
+      "step": 6302
+    },
+    {
+      "epoch": 0.12608,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.06288655598958333,
+      "learning_rate": 0.0001,
+      "loss": 4.4194,
+      "loss/crossentropy": 1.9966526627540588,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2424854189157486,
+      "step": 6304
+    },
+    {
+      "epoch": 0.12612,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.022541300455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2382,
+      "loss/crossentropy": 1.895868957042694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2255951091647148,
+      "step": 6306
+    },
+    {
+      "epoch": 0.12616,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.023688761393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.7758,
+      "loss/crossentropy": 2.3897345066070557,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24461720883846283,
+      "step": 6308
+    },
+    {
+      "epoch": 0.1262,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.022858683268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.8275,
+      "loss/crossentropy": 2.059292197227478,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29264035820961,
+      "step": 6310
+    },
+    {
+      "epoch": 0.12624,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.0221343994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.5098,
+      "loss/crossentropy": 1.8142406344413757,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22127485275268555,
+      "step": 6312
+    },
+    {
+      "epoch": 0.12628,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0245025634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2349,
+      "loss/crossentropy": 2.0186127424240112,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.217330664396286,
+      "step": 6314
+    },
+    {
+      "epoch": 0.12632,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.024820963541666668,
+      "learning_rate": 0.0001,
+      "loss": 4.0102,
+      "loss/crossentropy": 1.9439889192581177,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2108919695019722,
+      "step": 6316
+    },
+    {
+      "epoch": 0.12636,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.024690755208333335,
+      "learning_rate": 0.0001,
+      "loss": 4.8672,
+      "loss/crossentropy": 2.617791175842285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2756097912788391,
+      "step": 6318
+    },
+    {
+      "epoch": 0.1264,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.017609659830729166,
+      "learning_rate": 0.0001,
+      "loss": 4.3918,
+      "loss/crossentropy": 1.9074286818504333,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22491587698459625,
+      "step": 6320
+    },
+    {
+      "epoch": 0.12644,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.016331990559895832,
+      "learning_rate": 0.0001,
+      "loss": 4.4304,
+      "loss/crossentropy": 2.1414809226989746,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22703612595796585,
+      "step": 6322
+    },
+    {
+      "epoch": 0.12648,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.0154205322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4982,
+      "loss/crossentropy": 2.291784942150116,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24490328133106232,
+      "step": 6324
+    },
+    {
+      "epoch": 0.12652,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0143218994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.228,
+      "loss/crossentropy": 1.938249409198761,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22611317038536072,
+      "step": 6326
+    },
+    {
+      "epoch": 0.12656,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0187408447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4932,
+      "loss/crossentropy": 1.7510024905204773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30439992994070053,
+      "step": 6328
+    },
+    {
+      "epoch": 0.1266,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.022907511393229166,
+      "learning_rate": 0.0001,
+      "loss": 4.4351,
+      "loss/crossentropy": 2.3168221712112427,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25660980492830276,
+      "step": 6330
+    },
+    {
+      "epoch": 0.12664,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.024409993489583334,
+      "learning_rate": 0.0001,
+      "loss": 4.5002,
+      "loss/crossentropy": 1.9957427978515625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21420849114656448,
+      "step": 6332
+    },
+    {
+      "epoch": 0.12668,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.025104777018229166,
+      "learning_rate": 0.0001,
+      "loss": 4.3583,
+      "loss/crossentropy": 2.4371464252471924,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2576342821121216,
+      "step": 6334
+    },
+    {
+      "epoch": 0.12672,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.021458943684895832,
+      "learning_rate": 0.0001,
+      "loss": 4.2497,
+      "loss/crossentropy": 1.910677433013916,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21360614150762558,
+      "step": 6336
+    },
+    {
+      "epoch": 0.12676,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.020750935872395834,
+      "learning_rate": 0.0001,
+      "loss": 4.2679,
+      "loss/crossentropy": 2.057362914085388,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23079442232847214,
+      "step": 6338
+    },
+    {
+      "epoch": 0.1268,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0191070556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1818,
+      "loss/crossentropy": 2.0249438881874084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24038050323724747,
+      "step": 6340
+    },
+    {
+      "epoch": 0.12684,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.019074503580729166,
+      "learning_rate": 0.0001,
+      "loss": 4.5374,
+      "loss/crossentropy": 2.0371533632278442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2355760782957077,
+      "step": 6342
+    },
+    {
+      "epoch": 0.12688,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0130523681640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4849,
+      "loss/crossentropy": 2.3101218938827515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24161705374717712,
+      "step": 6344
+    },
+    {
+      "epoch": 0.12692,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.010261027018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.6399,
+      "loss/crossentropy": 2.1990097761154175,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2537754699587822,
+      "step": 6346
+    },
+    {
+      "epoch": 0.12696,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0136383056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2697,
+      "loss/crossentropy": 2.1783597469329834,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2106732353568077,
+      "step": 6348
+    },
+    {
+      "epoch": 0.127,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.012987263997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.5127,
+      "loss/crossentropy": 2.2316598892211914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23841773718595505,
+      "step": 6350
+    },
+    {
+      "epoch": 0.12704,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.00943603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5651,
+      "loss/crossentropy": 2.0329924821853638,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22755083441734314,
+      "step": 6352
+    },
+    {
+      "epoch": 0.12708,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.014525349934895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1888,
+      "loss/crossentropy": 1.9174052476882935,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20000722259283066,
+      "step": 6354
+    },
+    {
+      "epoch": 0.12712,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0154693603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3321,
+      "loss/crossentropy": 1.968774676322937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2194407731294632,
+      "step": 6356
+    },
+    {
+      "epoch": 0.12716,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.0183013916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.5768,
+      "loss/crossentropy": 2.1767213344573975,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24483858048915863,
+      "step": 6358
+    },
+    {
+      "epoch": 0.1272,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.018391927083333332,
+      "learning_rate": 0.0001,
+      "loss": 4.5063,
+      "loss/crossentropy": 1.8390987515449524,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21677500754594803,
+      "step": 6360
+    },
+    {
+      "epoch": 0.12724,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.016039021809895835,
+      "learning_rate": 0.0001,
+      "loss": 4.2635,
+      "loss/crossentropy": 2.1923086643218994,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21362057328224182,
+      "step": 6362
+    },
+    {
+      "epoch": 0.12728,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.010347493489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.3104,
+      "loss/crossentropy": 2.030683994293213,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22884567826986313,
+      "step": 6364
+    },
+    {
+      "epoch": 0.12732,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.00816650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.452,
+      "loss/crossentropy": 2.330837845802307,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24233026057481766,
+      "step": 6366
+    },
+    {
+      "epoch": 0.12736,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.012040201822916667,
+      "learning_rate": 0.0001,
+      "loss": 4.6404,
+      "loss/crossentropy": 2.001612663269043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2332654967904091,
+      "step": 6368
+    },
+    {
+      "epoch": 0.1274,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.011750284830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.7101,
+      "loss/crossentropy": 1.9234941601753235,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23741928488016129,
+      "step": 6370
+    },
+    {
+      "epoch": 0.12744,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.012262980143229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2164,
+      "loss/crossentropy": 2.059934139251709,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24736596643924713,
+      "step": 6372
+    },
+    {
+      "epoch": 0.12748,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010798136393229166,
+      "learning_rate": 0.0001,
+      "loss": 4.3916,
+      "loss/crossentropy": 2.25182843208313,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24349220097064972,
+      "step": 6374
+    },
+    {
+      "epoch": 0.12752,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.010888671875,
+      "learning_rate": 0.0001,
+      "loss": 4.3742,
+      "loss/crossentropy": 1.845405638217926,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2138313353061676,
+      "step": 6376
+    },
+    {
+      "epoch": 0.12756,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.0164215087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.568,
+      "loss/crossentropy": 1.8833998441696167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25304871797561646,
+      "step": 6378
+    },
+    {
+      "epoch": 0.1276,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.017072550455729165,
+      "learning_rate": 0.0001,
+      "loss": 4.2282,
+      "loss/crossentropy": 2.1066314578056335,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2231372445821762,
+      "step": 6380
+    },
+    {
+      "epoch": 0.12764,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.018561808268229167,
+      "learning_rate": 0.0001,
+      "loss": 4.3381,
+      "loss/crossentropy": 2.024670898914337,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22260528802871704,
+      "step": 6382
+    },
+    {
+      "epoch": 0.12768,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.01763916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.8132,
+      "loss/crossentropy": 2.425115466117859,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2429298236966133,
+      "step": 6384
+    },
+    {
+      "epoch": 0.12772,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.016597493489583334,
+      "learning_rate": 0.0001,
+      "loss": 4.2387,
+      "loss/crossentropy": 2.1847925186157227,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2495381161570549,
+      "step": 6386
+    },
+    {
+      "epoch": 0.12776,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.015746053059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.6919,
+      "loss/crossentropy": 2.1463611125946045,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2303348332643509,
+      "step": 6388
+    },
+    {
+      "epoch": 0.1278,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.014839680989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.2714,
+      "loss/crossentropy": 1.9755331873893738,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2232731431722641,
+      "step": 6390
+    },
+    {
+      "epoch": 0.12784,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.015973917643229165,
+      "learning_rate": 0.0001,
+      "loss": 4.5255,
+      "loss/crossentropy": 2.0741465091705322,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22912710905075073,
+      "step": 6392
+    },
+    {
+      "epoch": 0.12788,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.009471638997395834,
+      "learning_rate": 0.0001,
+      "loss": 4.3834,
+      "loss/crossentropy": 2.1314439177513123,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22910036891698837,
+      "step": 6394
+    },
+    {
+      "epoch": 0.12792,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0107086181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1615,
+      "loss/crossentropy": 1.976862907409668,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.233707495033741,
+      "step": 6396
+    },
+    {
+      "epoch": 0.12796,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.010807291666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.3939,
+      "loss/crossentropy": 2.0534666180610657,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21424879133701324,
+      "step": 6398
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.008088175455729167,
+      "learning_rate": 0.0001,
+      "loss": 4.6647,
+      "loss/crossentropy": 2.2200992107391357,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2836414724588394,
+      "step": 6400
+    },
+    {
+      "epoch": 0.12804,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.007542928059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.3802,
+      "loss/crossentropy": 1.8642286658287048,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21438255906105042,
+      "step": 6402
+    },
+    {
+      "epoch": 0.12808,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.005464680989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.3868,
+      "loss/crossentropy": 1.9571613073349,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22038634121418,
+      "step": 6404
+    },
+    {
+      "epoch": 0.12812,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.008980305989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.6888,
+      "loss/crossentropy": 2.0420787930488586,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2532814294099808,
+      "step": 6406
+    },
+    {
+      "epoch": 0.12816,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.008349609375,
+      "learning_rate": 0.0001,
+      "loss": 4.3357,
+      "loss/crossentropy": 1.7511736750602722,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2067284658551216,
+      "step": 6408
+    },
+    {
+      "epoch": 0.1282,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0095123291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.5568,
+      "loss/crossentropy": 2.0310762524604797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2763645648956299,
+      "step": 6410
+    },
+    {
+      "epoch": 0.12824,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.008617146809895834,
+      "learning_rate": 0.0001,
+      "loss": 4.575,
+      "loss/crossentropy": 2.112699866294861,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2409205138683319,
+      "step": 6412
+    },
+    {
+      "epoch": 0.12828,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.14166259765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3023,
+      "loss/crossentropy": 1.8243904113769531,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2079915553331375,
+      "step": 6414
+    },
+    {
+      "epoch": 0.12832,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.14143473307291668,
+      "learning_rate": 0.0001,
+      "loss": 4.5395,
+      "loss/crossentropy": 1.8310211896896362,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21304991096258163,
+      "step": 6416
+    },
+    {
+      "epoch": 0.12836,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.14011128743489584,
+      "learning_rate": 0.0001,
+      "loss": 4.3806,
+      "loss/crossentropy": 1.9653990268707275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22207710891962051,
+      "step": 6418
+    },
+    {
+      "epoch": 0.1284,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.13982645670572916,
+      "learning_rate": 0.0001,
+      "loss": 4.5869,
+      "loss/crossentropy": 2.0583431124687195,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23768695443868637,
+      "step": 6420
+    },
+    {
+      "epoch": 0.12844,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.1391998291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.6512,
+      "loss/crossentropy": 2.1162279844284058,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23963302373886108,
+      "step": 6422
+    },
+    {
+      "epoch": 0.12848,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.13347066243489583,
+      "learning_rate": 0.0001,
+      "loss": 4.6268,
+      "loss/crossentropy": 2.068794012069702,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21967273205518723,
+      "step": 6424
+    },
+    {
+      "epoch": 0.12852,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.13585611979166667,
+      "learning_rate": 0.0001,
+      "loss": 4.4004,
+      "loss/crossentropy": 2.165997266769409,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22991405427455902,
+      "step": 6426
+    },
+    {
+      "epoch": 0.12856,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.14031473795572916,
+      "learning_rate": 0.0001,
+      "loss": 4.1816,
+      "loss/crossentropy": 1.9116491675376892,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22784583270549774,
+      "step": 6428
+    },
+    {
+      "epoch": 0.1286,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.0087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5268,
+      "loss/crossentropy": 2.296347498893738,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2618508040904999,
+      "step": 6430
+    },
+    {
+      "epoch": 0.12864,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0100250244140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0363,
+      "loss/crossentropy": 2.3961373567581177,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27667778730392456,
+      "step": 6432
+    },
+    {
+      "epoch": 0.12868,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.009598795572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.3993,
+      "loss/crossentropy": 1.9141033291816711,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22818633913993835,
+      "step": 6434
+    },
+    {
+      "epoch": 0.12872,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0070149739583333336,
+      "learning_rate": 0.0001,
+      "loss": 4.481,
+      "loss/crossentropy": 2.4431718587875366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24852856248617172,
+      "step": 6436
+    },
+    {
+      "epoch": 0.12876,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.006494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.5153,
+      "loss/crossentropy": 1.9343949556350708,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22181283682584763,
+      "step": 6438
+    },
+    {
+      "epoch": 0.1288,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.005549112955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4547,
+      "loss/crossentropy": 2.203734040260315,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23536919057369232,
+      "step": 6440
+    },
+    {
+      "epoch": 0.12884,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0055572509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.354,
+      "loss/crossentropy": 2.0419046878814697,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2121758982539177,
+      "step": 6442
+    },
+    {
+      "epoch": 0.12888,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.0054595947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.5738,
+      "loss/crossentropy": 2.3551554679870605,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23853591084480286,
+      "step": 6444
+    },
+    {
+      "epoch": 0.12892,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.004813639322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.4334,
+      "loss/crossentropy": 1.9079387784004211,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22546496242284775,
+      "step": 6446
+    },
+    {
+      "epoch": 0.12896,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.00390625,
+      "learning_rate": 0.0001,
+      "loss": 4.4188,
+      "loss/crossentropy": 1.9344156980514526,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21923892199993134,
+      "step": 6448
+    },
+    {
+      "epoch": 0.129,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.0047271728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3219,
+      "loss/crossentropy": 1.7627189755439758,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20726975798606873,
+      "step": 6450
+    },
+    {
+      "epoch": 0.12904,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.0048736572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4045,
+      "loss/crossentropy": 1.9988782405853271,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23156572133302689,
+      "step": 6452
+    },
+    {
+      "epoch": 0.12908,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.004621378580729167,
+      "learning_rate": 0.0001,
+      "loss": 4.6551,
+      "loss/crossentropy": 2.3970296382904053,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24723558872938156,
+      "step": 6454
+    },
+    {
+      "epoch": 0.12912,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.00455322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3785,
+      "loss/crossentropy": 2.009088099002838,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2188432812690735,
+      "step": 6456
+    },
+    {
+      "epoch": 0.12916,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.005882771809895834,
+      "learning_rate": 0.0001,
+      "loss": 4.9189,
+      "loss/crossentropy": 2.166573464870453,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22506655752658844,
+      "step": 6458
+    },
+    {
+      "epoch": 0.1292,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.00670166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3281,
+      "loss/crossentropy": 2.1567386388778687,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23143760859966278,
+      "step": 6460
+    },
+    {
+      "epoch": 0.12924,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.007307942708333333,
+      "learning_rate": 0.0001,
+      "loss": 3.9726,
+      "loss/crossentropy": 1.8458901643753052,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22627578675746918,
+      "step": 6462
+    },
+    {
+      "epoch": 0.12928,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.00738525390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1416,
+      "loss/crossentropy": 1.7933887243270874,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1944974958896637,
+      "step": 6464
+    },
+    {
+      "epoch": 0.12932,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.0077707926432291664,
+      "learning_rate": 0.0001,
+      "loss": 4.3712,
+      "loss/crossentropy": 2.14457631111145,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2461908757686615,
+      "step": 6466
+    },
+    {
+      "epoch": 0.12936,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.011563873291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2253,
+      "loss/crossentropy": 2.2191081047058105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24845656007528305,
+      "step": 6468
+    },
+    {
+      "epoch": 0.1294,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.012284088134765624,
+      "learning_rate": 0.0001,
+      "loss": 4.3992,
+      "loss/crossentropy": 2.2655181884765625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24961909651756287,
+      "step": 6470
+    },
+    {
+      "epoch": 0.12944,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.012617746988932291,
+      "learning_rate": 0.0001,
+      "loss": 4.2431,
+      "loss/crossentropy": 1.9992872476577759,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21180924773216248,
+      "step": 6472
+    },
+    {
+      "epoch": 0.12948,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009348297119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2865,
+      "loss/crossentropy": 2.1950103044509888,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23401658236980438,
+      "step": 6474
+    },
+    {
+      "epoch": 0.12952,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.010027821858723958,
+      "learning_rate": 0.0001,
+      "loss": 4.2614,
+      "loss/crossentropy": 2.148743689060211,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2146785706281662,
+      "step": 6476
+    },
+    {
+      "epoch": 0.12956,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.010253651936848959,
+      "learning_rate": 0.0001,
+      "loss": 4.7196,
+      "loss/crossentropy": 2.4312193393707275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2596089243888855,
+      "step": 6478
+    },
+    {
+      "epoch": 0.1296,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.011844635009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.7232,
+      "loss/crossentropy": 2.061104893684387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24564718455076218,
+      "step": 6480
+    },
+    {
+      "epoch": 0.12964,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.011224110921223959,
+      "learning_rate": 0.0001,
+      "loss": 4.6833,
+      "loss/crossentropy": 2.1994687914848328,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2352369725704193,
+      "step": 6482
+    },
+    {
+      "epoch": 0.12968,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.010692342122395834,
+      "learning_rate": 0.0001,
+      "loss": 4.2378,
+      "loss/crossentropy": 2.085163116455078,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2309635877609253,
+      "step": 6484
+    },
+    {
+      "epoch": 0.12972,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0098297119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.5964,
+      "loss/crossentropy": 2.3360198736190796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24846713244915009,
+      "step": 6486
+    },
+    {
+      "epoch": 0.12976,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.010602823893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.4454,
+      "loss/crossentropy": 2.089003086090088,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24980145692825317,
+      "step": 6488
+    },
+    {
+      "epoch": 0.1298,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0108795166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.4446,
+      "loss/crossentropy": 2.0884299874305725,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23875930905342102,
+      "step": 6490
+    },
+    {
+      "epoch": 0.12984,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.01002197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1661,
+      "loss/crossentropy": 1.9070702195167542,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21942836046218872,
+      "step": 6492
+    },
+    {
+      "epoch": 0.12988,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.02047119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.6558,
+      "loss/crossentropy": 2.1617711782455444,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22820374369621277,
+      "step": 6494
+    },
+    {
+      "epoch": 0.12992,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0203765869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0938,
+      "loss/crossentropy": 1.7782898545265198,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22162485867738724,
+      "step": 6496
+    },
+    {
+      "epoch": 0.12996,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.021028645833333335,
+      "learning_rate": 0.0001,
+      "loss": 4.4198,
+      "loss/crossentropy": 2.1364612579345703,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22490044683218002,
+      "step": 6498
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.020213826497395834,
+      "learning_rate": 0.0001,
+      "loss": 4.569,
+      "loss/crossentropy": 2.282773971557617,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24553906172513962,
+      "step": 6500
+    },
+    {
+      "epoch": 0.13004,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.020905558268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.4965,
+      "loss/crossentropy": 2.008001983165741,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22152648121118546,
+      "step": 6502
+    },
+    {
+      "epoch": 0.13008,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.021833292643229165,
+      "learning_rate": 0.0001,
+      "loss": 4.4535,
+      "loss/crossentropy": 2.1681981086730957,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26258186995983124,
+      "step": 6504
+    },
+    {
+      "epoch": 0.13012,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.021761067708333335,
+      "learning_rate": 0.0001,
+      "loss": 4.6254,
+      "loss/crossentropy": 2.5754435062408447,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26315446197986603,
+      "step": 6506
+    },
+    {
+      "epoch": 0.13016,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.021187337239583333,
+      "learning_rate": 0.0001,
+      "loss": 4.1505,
+      "loss/crossentropy": 1.7897658348083496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22884277999401093,
+      "step": 6508
+    },
+    {
+      "epoch": 0.1302,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.016844685872395834,
+      "learning_rate": 0.0001,
+      "loss": 4.6315,
+      "loss/crossentropy": 1.8081435561180115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21860769391059875,
+      "step": 6510
+    },
+    {
+      "epoch": 0.13024,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.0140289306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.7371,
+      "loss/crossentropy": 2.243759036064148,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2545628473162651,
+      "step": 6512
+    },
+    {
+      "epoch": 0.13028,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.03447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.6147,
+      "loss/crossentropy": 2.069986939430237,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20862630754709244,
+      "step": 6514
+    },
+    {
+      "epoch": 0.13032,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0349761962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4758,
+      "loss/crossentropy": 2.052341878414154,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2460884153842926,
+      "step": 6516
+    },
+    {
+      "epoch": 0.13036,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.0317047119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.6369,
+      "loss/crossentropy": 2.3376708030700684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25382688641548157,
+      "step": 6518
+    },
+    {
+      "epoch": 0.1304,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0329498291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3048,
+      "loss/crossentropy": 1.8329171538352966,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2135135680437088,
+      "step": 6520
+    },
+    {
+      "epoch": 0.13044,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.035791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.6582,
+      "loss/crossentropy": 2.077217698097229,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2265045866370201,
+      "step": 6522
+    },
+    {
+      "epoch": 0.13048,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.036783854166666664,
+      "learning_rate": 0.0001,
+      "loss": 4.2189,
+      "loss/crossentropy": 2.0677568912506104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2468869537115097,
+      "step": 6524
+    },
+    {
+      "epoch": 0.13052,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.03591206868489583,
+      "learning_rate": 0.0001,
+      "loss": 4.5246,
+      "loss/crossentropy": 2.0692074298858643,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2602302059531212,
+      "step": 6526
+    },
+    {
+      "epoch": 0.13056,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.03583984375,
+      "learning_rate": 0.0001,
+      "loss": 4.6709,
+      "loss/crossentropy": 1.9767250418663025,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2232941836118698,
+      "step": 6528
+    },
+    {
+      "epoch": 0.1306,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.01480712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4851,
+      "loss/crossentropy": 1.7538996934890747,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20786649733781815,
+      "step": 6530
+    },
+    {
+      "epoch": 0.13064,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.01578369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.5598,
+      "loss/crossentropy": 2.061249256134033,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23238955438137054,
+      "step": 6532
+    },
+    {
+      "epoch": 0.13068,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0167877197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.237,
+      "loss/crossentropy": 1.9083253145217896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22536182403564453,
+      "step": 6534
+    },
+    {
+      "epoch": 0.13072,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.017952473958333333,
+      "learning_rate": 0.0001,
+      "loss": 4.3004,
+      "loss/crossentropy": 2.052153766155243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2343958392739296,
+      "step": 6536
+    },
+    {
+      "epoch": 0.13076,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010091145833333334,
+      "learning_rate": 0.0001,
+      "loss": 4.3638,
+      "loss/crossentropy": 1.9979270100593567,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20102836191654205,
+      "step": 6538
+    },
+    {
+      "epoch": 0.1308,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.012031809488932291,
+      "learning_rate": 0.0001,
+      "loss": 4.019,
+      "loss/crossentropy": 1.618333637714386,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21146921813488007,
+      "step": 6540
+    },
+    {
+      "epoch": 0.13084,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.011940256754557291,
+      "learning_rate": 0.0001,
+      "loss": 4.2514,
+      "loss/crossentropy": 1.771790623664856,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21661869436502457,
+      "step": 6542
+    },
+    {
+      "epoch": 0.13088,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.007165273030598958,
+      "learning_rate": 0.0001,
+      "loss": 4.3528,
+      "loss/crossentropy": 2.2347733974456787,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2507361173629761,
+      "step": 6544
+    },
+    {
+      "epoch": 0.13092,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.012389882405598959,
+      "learning_rate": 0.0001,
+      "loss": 4.4333,
+      "loss/crossentropy": 2.226397395133972,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.257301464676857,
+      "step": 6546
+    },
+    {
+      "epoch": 0.13096,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.012776438395182292,
+      "learning_rate": 0.0001,
+      "loss": 4.1629,
+      "loss/crossentropy": 1.9884281158447266,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2213538959622383,
+      "step": 6548
+    },
+    {
+      "epoch": 0.131,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.01755549112955729,
+      "learning_rate": 0.0001,
+      "loss": 4.549,
+      "loss/crossentropy": 2.0790398120880127,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2259911745786667,
+      "step": 6550
+    },
+    {
+      "epoch": 0.13104,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.016294097900390624,
+      "learning_rate": 0.0001,
+      "loss": 4.5118,
+      "loss/crossentropy": 2.1309107542037964,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23984672129154205,
+      "step": 6552
+    },
+    {
+      "epoch": 0.13108,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.01744562784830729,
+      "learning_rate": 0.0001,
+      "loss": 4.4528,
+      "loss/crossentropy": 1.9651963114738464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21699358522891998,
+      "step": 6554
+    },
+    {
+      "epoch": 0.13112,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.014598592122395834,
+      "learning_rate": 0.0001,
+      "loss": 4.436,
+      "loss/crossentropy": 2.405247449874878,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24624405801296234,
+      "step": 6556
+    },
+    {
+      "epoch": 0.13116,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0154296875,
+      "learning_rate": 0.0001,
+      "loss": 4.3984,
+      "loss/crossentropy": 2.215611457824707,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24835016578435898,
+      "step": 6558
+    },
+    {
+      "epoch": 0.1312,
+      "grad_norm": 4.125,
+      "grad_norm_var": 0.2436920166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3876,
+      "loss/crossentropy": 2.1731653809547424,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24633550643920898,
+      "step": 6560
+    },
+    {
+      "epoch": 0.13124,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.24501953125,
+      "learning_rate": 0.0001,
+      "loss": 4.5101,
+      "loss/crossentropy": 2.0507587790489197,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25702129304409027,
+      "step": 6562
+    },
+    {
+      "epoch": 0.13128,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.2433502197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3464,
+      "loss/crossentropy": 2.2088446617126465,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22610870003700256,
+      "step": 6564
+    },
+    {
+      "epoch": 0.13132,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.23931884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3612,
+      "loss/crossentropy": 1.6911352276802063,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2188590243458748,
+      "step": 6566
+    },
+    {
+      "epoch": 0.13136,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.24976806640625,
+      "learning_rate": 0.0001,
+      "loss": 3.9795,
+      "loss/crossentropy": 1.807699978351593,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21844930946826935,
+      "step": 6568
+    },
+    {
+      "epoch": 0.1314,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.24514872233072918,
+      "learning_rate": 0.0001,
+      "loss": 4.4293,
+      "loss/crossentropy": 2.292602300643921,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.242776520550251,
+      "step": 6570
+    },
+    {
+      "epoch": 0.13144,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.24806315104166668,
+      "learning_rate": 0.0001,
+      "loss": 4.07,
+      "loss/crossentropy": 1.5262329578399658,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1674257293343544,
+      "step": 6572
+    },
+    {
+      "epoch": 0.13148,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.24504801432291667,
+      "learning_rate": 0.0001,
+      "loss": 4.4084,
+      "loss/crossentropy": 2.180319309234619,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2260192409157753,
+      "step": 6574
+    },
+    {
+      "epoch": 0.13152,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0216705322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.544,
+      "loss/crossentropy": 2.188440203666687,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23389383405447006,
+      "step": 6576
+    },
+    {
+      "epoch": 0.13156,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.03427327473958333,
+      "learning_rate": 0.0001,
+      "loss": 4.7595,
+      "loss/crossentropy": 2.20253586769104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3217846751213074,
+      "step": 6578
+    },
+    {
+      "epoch": 0.1316,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.034764607747395836,
+      "learning_rate": 0.0001,
+      "loss": 4.2771,
+      "loss/crossentropy": 2.1901479959487915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24521923065185547,
+      "step": 6580
+    },
+    {
+      "epoch": 0.13164,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.04170633951822917,
+      "learning_rate": 0.0001,
+      "loss": 4.7105,
+      "loss/crossentropy": 2.0719348192214966,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2640562355518341,
+      "step": 6582
+    },
+    {
+      "epoch": 0.13168,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.040339152018229164,
+      "learning_rate": 0.0001,
+      "loss": 4.1487,
+      "loss/crossentropy": 2.331762194633484,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24890532344579697,
+      "step": 6584
+    },
+    {
+      "epoch": 0.13172,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.04429931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.6013,
+      "loss/crossentropy": 1.5143779516220093,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1957392692565918,
+      "step": 6586
+    },
+    {
+      "epoch": 0.13176,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.04088134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4311,
+      "loss/crossentropy": 2.301910698413849,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27353671938180923,
+      "step": 6588
+    },
+    {
+      "epoch": 0.1318,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.11796773274739583,
+      "learning_rate": 0.0001,
+      "loss": 4.5035,
+      "loss/crossentropy": 2.018579602241516,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24261966347694397,
+      "step": 6590
+    },
+    {
+      "epoch": 0.13184,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.11685282389322917,
+      "learning_rate": 0.0001,
+      "loss": 4.6931,
+      "loss/crossentropy": 1.9749983549118042,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20727626234292984,
+      "step": 6592
+    },
+    {
+      "epoch": 0.13188,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.11741536458333333,
+      "learning_rate": 0.0001,
+      "loss": 4.1929,
+      "loss/crossentropy": 2.040414035320282,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22879169881343842,
+      "step": 6594
+    },
+    {
+      "epoch": 0.13192,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.11551106770833333,
+      "learning_rate": 0.0001,
+      "loss": 4.3331,
+      "loss/crossentropy": 1.941766619682312,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22591909021139145,
+      "step": 6596
+    },
+    {
+      "epoch": 0.13196,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.10946858723958333,
+      "learning_rate": 0.0001,
+      "loss": 4.3659,
+      "loss/crossentropy": 1.8487200140953064,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23654203116893768,
+      "step": 6598
+    },
+    {
+      "epoch": 0.132,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.10340067545572916,
+      "learning_rate": 0.0001,
+      "loss": 4.225,
+      "loss/crossentropy": 1.9297338724136353,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2016594037413597,
+      "step": 6600
+    },
+    {
+      "epoch": 0.13204,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.11005757649739584,
+      "learning_rate": 0.0001,
+      "loss": 4.3643,
+      "loss/crossentropy": 2.048487663269043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20044995844364166,
+      "step": 6602
+    },
+    {
+      "epoch": 0.13208,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.11295572916666667,
+      "learning_rate": 0.0001,
+      "loss": 4.425,
+      "loss/crossentropy": 2.3620439767837524,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24750277400016785,
+      "step": 6604
+    },
+    {
+      "epoch": 0.13212,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.0158203125,
+      "learning_rate": 0.0001,
+      "loss": 4.2578,
+      "loss/crossentropy": 2.1324113607406616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23109012842178345,
+      "step": 6606
+    },
+    {
+      "epoch": 0.13216,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.015620930989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.4775,
+      "loss/crossentropy": 2.021477997303009,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2152409330010414,
+      "step": 6608
+    },
+    {
+      "epoch": 0.1322,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.012669881184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.6221,
+      "loss/crossentropy": 2.1693036556243896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.234656922519207,
+      "step": 6610
+    },
+    {
+      "epoch": 0.13224,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.008854166666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.1469,
+      "loss/crossentropy": 2.1138017177581787,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2239084094762802,
+      "step": 6612
+    },
+    {
+      "epoch": 0.13228,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.008772786458333333,
+      "learning_rate": 0.0001,
+      "loss": 4.4862,
+      "loss/crossentropy": 2.2540252208709717,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2614182382822037,
+      "step": 6614
+    },
+    {
+      "epoch": 0.13232,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.008837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2131,
+      "loss/crossentropy": 2.033502757549286,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2177339717745781,
+      "step": 6616
+    },
+    {
+      "epoch": 0.13236,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010054524739583333,
+      "learning_rate": 0.0001,
+      "loss": 4.2648,
+      "loss/crossentropy": 2.2490646839141846,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22369390726089478,
+      "step": 6618
+    },
+    {
+      "epoch": 0.1324,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0150543212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3571,
+      "loss/crossentropy": 2.14319908618927,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24394190311431885,
+      "step": 6620
+    },
+    {
+      "epoch": 0.13244,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.01627197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.5136,
+      "loss/crossentropy": 2.045474410057068,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22738997638225555,
+      "step": 6622
+    },
+    {
+      "epoch": 0.13248,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.015816243489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.5773,
+      "loss/crossentropy": 2.1853290796279907,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2565506473183632,
+      "step": 6624
+    },
+    {
+      "epoch": 0.13252,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.017015584309895835,
+      "learning_rate": 0.0001,
+      "loss": 4.275,
+      "loss/crossentropy": 2.1161083579063416,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22949891537427902,
+      "step": 6626
+    },
+    {
+      "epoch": 0.13256,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0167633056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3154,
+      "loss/crossentropy": 2.167048454284668,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23087909072637558,
+      "step": 6628
+    },
+    {
+      "epoch": 0.1326,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0156158447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2777,
+      "loss/crossentropy": 2.1544610261917114,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25153525173664093,
+      "step": 6630
+    },
+    {
+      "epoch": 0.13264,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.018244425455729168,
+      "learning_rate": 0.0001,
+      "loss": 4.4778,
+      "loss/crossentropy": 2.0319228768348694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22067170590162277,
+      "step": 6632
+    },
+    {
+      "epoch": 0.13268,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0162261962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4398,
+      "loss/crossentropy": 1.9723476767539978,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21380966901779175,
+      "step": 6634
+    },
+    {
+      "epoch": 0.13272,
+      "grad_norm": 2.796875,
+      "grad_norm_var": 0.031590779622395836,
+      "learning_rate": 0.0001,
+      "loss": 4.3474,
+      "loss/crossentropy": 2.2833333015441895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23245477676391602,
+      "step": 6636
+    },
+    {
+      "epoch": 0.13276,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.03132222493489583,
+      "learning_rate": 0.0001,
+      "loss": 4.4522,
+      "loss/crossentropy": 2.13793683052063,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22665268182754517,
+      "step": 6638
+    },
+    {
+      "epoch": 0.1328,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.04592692057291667,
+      "learning_rate": 0.0001,
+      "loss": 4.6386,
+      "loss/crossentropy": 2.188693881034851,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2471916377544403,
+      "step": 6640
+    },
+    {
+      "epoch": 0.13284,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0438629150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.6394,
+      "loss/crossentropy": 2.169856071472168,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23427975177764893,
+      "step": 6642
+    },
+    {
+      "epoch": 0.13288,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.04365132649739583,
+      "learning_rate": 0.0001,
+      "loss": 4.6498,
+      "loss/crossentropy": 2.1600061655044556,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23366540670394897,
+      "step": 6644
+    },
+    {
+      "epoch": 0.13292,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.044188435872395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4521,
+      "loss/crossentropy": 1.822945475578308,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2203340008854866,
+      "step": 6646
+    },
+    {
+      "epoch": 0.13296,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.045735677083333336,
+      "learning_rate": 0.0001,
+      "loss": 4.3263,
+      "loss/crossentropy": 1.8908653259277344,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2320382297039032,
+      "step": 6648
+    },
+    {
+      "epoch": 0.133,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.04363606770833333,
+      "learning_rate": 0.0001,
+      "loss": 4.5065,
+      "loss/crossentropy": 2.126000165939331,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24388836324214935,
+      "step": 6650
+    },
+    {
+      "epoch": 0.13304,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.0242828369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.5787,
+      "loss/crossentropy": 2.434928297996521,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24083568900823593,
+      "step": 6652
+    },
+    {
+      "epoch": 0.13308,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.023824055989583332,
+      "learning_rate": 0.0001,
+      "loss": 4.5538,
+      "loss/crossentropy": 2.2186567783355713,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2220132276415825,
+      "step": 6654
+    },
+    {
+      "epoch": 0.13312,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.006257120768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.4934,
+      "loss/crossentropy": 1.849799931049347,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21551364660263062,
+      "step": 6656
+    },
+    {
+      "epoch": 0.13316,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0072662353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2237,
+      "loss/crossentropy": 2.082044243812561,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21274320781230927,
+      "step": 6658
+    },
+    {
+      "epoch": 0.1332,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010838826497395834,
+      "learning_rate": 0.0001,
+      "loss": 4.5884,
+      "loss/crossentropy": 2.1957098245620728,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23078418523073196,
+      "step": 6660
+    },
+    {
+      "epoch": 0.13324,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.16033528645833334,
+      "learning_rate": 0.0001,
+      "loss": 4.519,
+      "loss/crossentropy": 2.228309690952301,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27178408950567245,
+      "step": 6662
+    },
+    {
+      "epoch": 0.13328,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.156298828125,
+      "learning_rate": 0.0001,
+      "loss": 4.5987,
+      "loss/crossentropy": 1.8185940384864807,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2145048901438713,
+      "step": 6664
+    },
+    {
+      "epoch": 0.13332,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.1566558837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4722,
+      "loss/crossentropy": 2.198649048805237,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2262621968984604,
+      "step": 6666
+    },
+    {
+      "epoch": 0.13336,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.159130859375,
+      "learning_rate": 0.0001,
+      "loss": 4.5729,
+      "loss/crossentropy": 2.2075835466384888,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23531068861484528,
+      "step": 6668
+    },
+    {
+      "epoch": 0.1334,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.16243387858072916,
+      "learning_rate": 0.0001,
+      "loss": 4.2913,
+      "loss/crossentropy": 1.9719768166542053,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2206977903842926,
+      "step": 6670
+    },
+    {
+      "epoch": 0.13344,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.17021484375,
+      "learning_rate": 0.0001,
+      "loss": 4.2144,
+      "loss/crossentropy": 2.304553985595703,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23418369889259338,
+      "step": 6672
+    },
+    {
+      "epoch": 0.13348,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.1682281494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.4485,
+      "loss/crossentropy": 2.212409734725952,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23888318240642548,
+      "step": 6674
+    },
+    {
+      "epoch": 0.13352,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.19371337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.176,
+      "loss/crossentropy": 2.001897156238556,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2147163525223732,
+      "step": 6676
+    },
+    {
+      "epoch": 0.13356,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.043680826822916664,
+      "learning_rate": 0.0001,
+      "loss": 4.4245,
+      "loss/crossentropy": 2.216760039329529,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24913202226161957,
+      "step": 6678
+    },
+    {
+      "epoch": 0.1336,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.04597142537434896,
+      "learning_rate": 0.0001,
+      "loss": 4.1862,
+      "loss/crossentropy": 1.8190750479698181,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.200186125934124,
+      "step": 6680
+    },
+    {
+      "epoch": 0.13364,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.047548166910807294,
+      "learning_rate": 0.0001,
+      "loss": 4.7271,
+      "loss/crossentropy": 2.311274528503418,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22400956600904465,
+      "step": 6682
+    },
+    {
+      "epoch": 0.13368,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.5293841044108073,
+      "learning_rate": 0.0001,
+      "loss": 4.3852,
+      "loss/crossentropy": 2.0381893515586853,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23101551085710526,
+      "step": 6684
+    },
+    {
+      "epoch": 0.13372,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.5331776936848959,
+      "learning_rate": 0.0001,
+      "loss": 4.1185,
+      "loss/crossentropy": 1.7441503405570984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20208899676799774,
+      "step": 6686
+    },
+    {
+      "epoch": 0.13376,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.5252593994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3835,
+      "loss/crossentropy": 1.8874938488006592,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2255900353193283,
+      "step": 6688
+    },
+    {
+      "epoch": 0.1338,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.5284006754557292,
+      "learning_rate": 0.0001,
+      "loss": 4.2563,
+      "loss/crossentropy": 2.2768125534057617,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21636968851089478,
+      "step": 6690
+    },
+    {
+      "epoch": 0.13384,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.5063954671223958,
+      "learning_rate": 0.0001,
+      "loss": 4.278,
+      "loss/crossentropy": 2.0253931283950806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28937554359436035,
+      "step": 6692
+    },
+    {
+      "epoch": 0.13388,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.5079661051432292,
+      "learning_rate": 0.0001,
+      "loss": 4.4205,
+      "loss/crossentropy": 2.1940718293190002,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23384064435958862,
+      "step": 6694
+    },
+    {
+      "epoch": 0.13392,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.4987993876139323,
+      "learning_rate": 0.0001,
+      "loss": 4.5066,
+      "loss/crossentropy": 2.071534514427185,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2146512120962143,
+      "step": 6696
+    },
+    {
+      "epoch": 0.13396,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.5017534891764323,
+      "learning_rate": 0.0001,
+      "loss": 4.348,
+      "loss/crossentropy": 1.8530714511871338,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21837462484836578,
+      "step": 6698
+    },
+    {
+      "epoch": 0.134,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.013396962483723959,
+      "learning_rate": 0.0001,
+      "loss": 4.521,
+      "loss/crossentropy": 2.210664451122284,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2199932038784027,
+      "step": 6700
+    },
+    {
+      "epoch": 0.13404,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.011506144205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4226,
+      "loss/crossentropy": 1.8530223965644836,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.209881991147995,
+      "step": 6702
+    },
+    {
+      "epoch": 0.13408,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.011839803059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4302,
+      "loss/crossentropy": 1.8609183430671692,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22096765041351318,
+      "step": 6704
+    },
+    {
+      "epoch": 0.13412,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.021214803059895832,
+      "learning_rate": 0.0001,
+      "loss": 4.8429,
+      "loss/crossentropy": 2.33315110206604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26414938271045685,
+      "step": 6706
+    },
+    {
+      "epoch": 0.13416,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.04798075358072917,
+      "learning_rate": 0.0001,
+      "loss": 4.6054,
+      "loss/crossentropy": 2.2656116485595703,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2769838646054268,
+      "step": 6708
+    },
+    {
+      "epoch": 0.1342,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0466705322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4875,
+      "loss/crossentropy": 2.2131590843200684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23640615493059158,
+      "step": 6710
+    },
+    {
+      "epoch": 0.13424,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.044831339518229166,
+      "learning_rate": 0.0001,
+      "loss": 4.1554,
+      "loss/crossentropy": 1.8667671084403992,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.212738998234272,
+      "step": 6712
+    },
+    {
+      "epoch": 0.13428,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.04480692545572917,
+      "learning_rate": 0.0001,
+      "loss": 4.0958,
+      "loss/crossentropy": 1.9699830412864685,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23507894575595856,
+      "step": 6714
+    },
+    {
+      "epoch": 0.13432,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.04632161458333333,
+      "learning_rate": 0.0001,
+      "loss": 4.1571,
+      "loss/crossentropy": 1.8108918070793152,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2192004919052124,
+      "step": 6716
+    },
+    {
+      "epoch": 0.13436,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.042740885416666666,
+      "learning_rate": 0.0001,
+      "loss": 4.4983,
+      "loss/crossentropy": 2.0528674125671387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22221814841032028,
+      "step": 6718
+    },
+    {
+      "epoch": 0.1344,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.046305338541666664,
+      "learning_rate": 0.0001,
+      "loss": 4.2544,
+      "loss/crossentropy": 1.7881956696510315,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21330178529024124,
+      "step": 6720
+    },
+    {
+      "epoch": 0.13444,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.039159138997395836,
+      "learning_rate": 0.0001,
+      "loss": 4.5424,
+      "loss/crossentropy": 2.2016018629074097,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22042546421289444,
+      "step": 6722
+    },
+    {
+      "epoch": 0.13448,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.006322224934895833,
+      "learning_rate": 0.0001,
+      "loss": 4.3159,
+      "loss/crossentropy": 1.9661846160888672,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21542846411466599,
+      "step": 6724
+    },
+    {
+      "epoch": 0.13452,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.031037394205729166,
+      "learning_rate": 0.0001,
+      "loss": 4.4473,
+      "loss/crossentropy": 2.1073482036590576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23370730876922607,
+      "step": 6726
+    },
+    {
+      "epoch": 0.13456,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.030826822916666666,
+      "learning_rate": 0.0001,
+      "loss": 4.6069,
+      "loss/crossentropy": 2.1937917470932007,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22414422780275345,
+      "step": 6728
+    },
+    {
+      "epoch": 0.1346,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.030436197916666668,
+      "learning_rate": 0.0001,
+      "loss": 4.2937,
+      "loss/crossentropy": 2.078732967376709,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21227504312992096,
+      "step": 6730
+    },
+    {
+      "epoch": 0.13464,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.030301920572916665,
+      "learning_rate": 0.0001,
+      "loss": 4.3043,
+      "loss/crossentropy": 1.9002525806427002,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21004119515419006,
+      "step": 6732
+    },
+    {
+      "epoch": 0.13468,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.030794270833333335,
+      "learning_rate": 0.0001,
+      "loss": 4.5156,
+      "loss/crossentropy": 2.055518925189972,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21610142290592194,
+      "step": 6734
+    },
+    {
+      "epoch": 0.13472,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.028473917643229166,
+      "learning_rate": 0.0001,
+      "loss": 4.4889,
+      "loss/crossentropy": 2.0521084666252136,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23528365790843964,
+      "step": 6736
+    },
+    {
+      "epoch": 0.13476,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.02730712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5051,
+      "loss/crossentropy": 2.2536301612854004,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21544227004051208,
+      "step": 6738
+    },
+    {
+      "epoch": 0.1348,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.027179972330729166,
+      "learning_rate": 0.0001,
+      "loss": 4.5164,
+      "loss/crossentropy": 2.2610143423080444,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2532905787229538,
+      "step": 6740
+    },
+    {
+      "epoch": 0.13484,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.00836181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1301,
+      "loss/crossentropy": 1.8840081095695496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21514790505170822,
+      "step": 6742
+    },
+    {
+      "epoch": 0.13488,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0084136962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2532,
+      "loss/crossentropy": 2.0841002464294434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2328692153096199,
+      "step": 6744
+    },
+    {
+      "epoch": 0.13492,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.010798136393229166,
+      "learning_rate": 0.0001,
+      "loss": 4.6335,
+      "loss/crossentropy": 2.507196068763733,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2736222892999649,
+      "step": 6746
+    },
+    {
+      "epoch": 0.13496,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.016527303059895835,
+      "learning_rate": 0.0001,
+      "loss": 4.189,
+      "loss/crossentropy": 2.0180357098579407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22328195720911026,
+      "step": 6748
+    },
+    {
+      "epoch": 0.135,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.016097005208333334,
+      "learning_rate": 0.0001,
+      "loss": 4.2854,
+      "loss/crossentropy": 2.2457560300827026,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26572495698928833,
+      "step": 6750
+    },
+    {
+      "epoch": 0.13504,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.031493123372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.885,
+      "loss/crossentropy": 2.1280853748321533,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22716474533081055,
+      "step": 6752
+    },
+    {
+      "epoch": 0.13508,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.03216145833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.3901,
+      "loss/crossentropy": 2.1843650341033936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24630828201770782,
+      "step": 6754
+    },
+    {
+      "epoch": 0.13512,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.03390299479166667,
+      "learning_rate": 0.0001,
+      "loss": 4.3584,
+      "loss/crossentropy": 1.7955012917518616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20614011585712433,
+      "step": 6756
+    },
+    {
+      "epoch": 0.13516,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.028563435872395834,
+      "learning_rate": 0.0001,
+      "loss": 4.3546,
+      "loss/crossentropy": 1.9315852522850037,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22918011993169785,
+      "step": 6758
+    },
+    {
+      "epoch": 0.1352,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.028055826822916668,
+      "learning_rate": 0.0001,
+      "loss": 4.2659,
+      "loss/crossentropy": 1.982887327671051,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21635843813419342,
+      "step": 6760
+    },
+    {
+      "epoch": 0.13524,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.029215494791666668,
+      "learning_rate": 0.0001,
+      "loss": 4.2828,
+      "loss/crossentropy": 2.25021892786026,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22931701689958572,
+      "step": 6762
+    },
+    {
+      "epoch": 0.13528,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.02252197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.5991,
+      "loss/crossentropy": 2.5220746994018555,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2406606376171112,
+      "step": 6764
+    },
+    {
+      "epoch": 0.13532,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.022386678059895835,
+      "learning_rate": 0.0001,
+      "loss": 4.5143,
+      "loss/crossentropy": 1.8115127086639404,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.220379076898098,
+      "step": 6766
+    },
+    {
+      "epoch": 0.13536,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010856119791666667,
+      "learning_rate": 0.0001,
+      "loss": 4.5113,
+      "loss/crossentropy": 1.8998088240623474,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22993376106023788,
+      "step": 6768
+    },
+    {
+      "epoch": 0.1354,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.01051025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.28,
+      "loss/crossentropy": 2.0183660984039307,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21991200000047684,
+      "step": 6770
+    },
+    {
+      "epoch": 0.13544,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.009847005208333334,
+      "learning_rate": 0.0001,
+      "loss": 4.6224,
+      "loss/crossentropy": 2.1927448511123657,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2536699026823044,
+      "step": 6772
+    },
+    {
+      "epoch": 0.13548,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.014404296875,
+      "learning_rate": 0.0001,
+      "loss": 4.196,
+      "loss/crossentropy": 1.92184317111969,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20716708898544312,
+      "step": 6774
+    },
+    {
+      "epoch": 0.13552,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.016813151041666665,
+      "learning_rate": 0.0001,
+      "loss": 4.7779,
+      "loss/crossentropy": 2.2437468767166138,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26381388306617737,
+      "step": 6776
+    },
+    {
+      "epoch": 0.13556,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.016722615559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.2907,
+      "loss/crossentropy": 2.087414026260376,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21566492319107056,
+      "step": 6778
+    },
+    {
+      "epoch": 0.1356,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.015653483072916665,
+      "learning_rate": 0.0001,
+      "loss": 4.4273,
+      "loss/crossentropy": 2.1936367750167847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22492723166942596,
+      "step": 6780
+    },
+    {
+      "epoch": 0.13564,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.014046223958333333,
+      "learning_rate": 0.0001,
+      "loss": 4.2992,
+      "loss/crossentropy": 1.7642306685447693,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20992937684059143,
+      "step": 6782
+    },
+    {
+      "epoch": 0.13568,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.014371744791666667,
+      "learning_rate": 0.0001,
+      "loss": 4.3593,
+      "loss/crossentropy": 2.01781964302063,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22552715986967087,
+      "step": 6784
+    },
+    {
+      "epoch": 0.13572,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.016402180989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.4708,
+      "loss/crossentropy": 2.0788158774375916,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24181769788265228,
+      "step": 6786
+    },
+    {
+      "epoch": 0.13576,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.01539306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2163,
+      "loss/crossentropy": 2.0424017310142517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21705424785614014,
+      "step": 6788
+    },
+    {
+      "epoch": 0.1358,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.019025675455729165,
+      "learning_rate": 0.0001,
+      "loss": 4.1182,
+      "loss/crossentropy": 1.6175345182418823,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1899409219622612,
+      "step": 6790
+    },
+    {
+      "epoch": 0.13584,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.016852823893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2914,
+      "loss/crossentropy": 2.004386007785797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22395263612270355,
+      "step": 6792
+    },
+    {
+      "epoch": 0.13588,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.019245402018229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2182,
+      "loss/crossentropy": 1.9224132895469666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2326791137456894,
+      "step": 6794
+    },
+    {
+      "epoch": 0.13592,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.019755045572916668,
+      "learning_rate": 0.0001,
+      "loss": 4.4768,
+      "loss/crossentropy": 1.8331453204154968,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21586360037326813,
+      "step": 6796
+    },
+    {
+      "epoch": 0.13596,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0197906494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3059,
+      "loss/crossentropy": 2.535244107246399,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24744001775979996,
+      "step": 6798
+    },
+    {
+      "epoch": 0.136,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.017704264322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.4444,
+      "loss/crossentropy": 2.0433249473571777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2502296343445778,
+      "step": 6800
+    },
+    {
+      "epoch": 0.13604,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0221588134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4619,
+      "loss/crossentropy": 2.35608172416687,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2536633685231209,
+      "step": 6802
+    },
+    {
+      "epoch": 0.13608,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.022362263997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.4493,
+      "loss/crossentropy": 2.1230265498161316,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2276441976428032,
+      "step": 6804
+    },
+    {
+      "epoch": 0.13612,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.015925089518229168,
+      "learning_rate": 0.0001,
+      "loss": 4.2125,
+      "loss/crossentropy": 2.0186346769332886,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22158686816692352,
+      "step": 6806
+    },
+    {
+      "epoch": 0.13616,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.0151763916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.6065,
+      "loss/crossentropy": 2.4136343002319336,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22668009996414185,
+      "step": 6808
+    },
+    {
+      "epoch": 0.1362,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012495930989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.3372,
+      "loss/crossentropy": 2.1241788268089294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23394256830215454,
+      "step": 6810
+    },
+    {
+      "epoch": 0.13624,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.017145792643229168,
+      "learning_rate": 0.0001,
+      "loss": 4.6063,
+      "loss/crossentropy": 1.9051874279975891,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2560805454850197,
+      "step": 6812
+    },
+    {
+      "epoch": 0.13628,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.015458170572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.2419,
+      "loss/crossentropy": 1.9248363375663757,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22457116842269897,
+      "step": 6814
+    },
+    {
+      "epoch": 0.13632,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.022264607747395835,
+      "learning_rate": 0.0001,
+      "loss": 4.5953,
+      "loss/crossentropy": 2.3065048456192017,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.245137557387352,
+      "step": 6816
+    },
+    {
+      "epoch": 0.13636,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0152496337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.6096,
+      "loss/crossentropy": 2.152611255645752,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22794383764266968,
+      "step": 6818
+    },
+    {
+      "epoch": 0.1364,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.016242472330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.129,
+      "loss/crossentropy": 1.9548735618591309,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20764236897230148,
+      "step": 6820
+    },
+    {
+      "epoch": 0.13644,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.016136678059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.3501,
+      "loss/crossentropy": 2.11979341506958,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22793393582105637,
+      "step": 6822
+    },
+    {
+      "epoch": 0.13648,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.016748046875,
+      "learning_rate": 0.0001,
+      "loss": 4.3338,
+      "loss/crossentropy": 2.351949691772461,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2366446554660797,
+      "step": 6824
+    },
+    {
+      "epoch": 0.13652,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.018843587239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.3764,
+      "loss/crossentropy": 2.1197460889816284,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24759536981582642,
+      "step": 6826
+    },
+    {
+      "epoch": 0.13656,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.013264973958333334,
+      "learning_rate": 0.0001,
+      "loss": 4.1204,
+      "loss/crossentropy": 1.8491687178611755,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20777105540037155,
+      "step": 6828
+    },
+    {
+      "epoch": 0.1366,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.013459269205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4867,
+      "loss/crossentropy": 1.964136004447937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2368694394826889,
+      "step": 6830
+    },
+    {
+      "epoch": 0.13664,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.006859334309895834,
+      "learning_rate": 0.0001,
+      "loss": 4.469,
+      "loss/crossentropy": 1.8988104462623596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19548720866441727,
+      "step": 6832
+    },
+    {
+      "epoch": 0.13668,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0059855143229166664,
+      "learning_rate": 0.0001,
+      "loss": 4.3104,
+      "loss/crossentropy": 1.757002353668213,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21120281517505646,
+      "step": 6834
+    },
+    {
+      "epoch": 0.13672,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.005952962239583333,
+      "learning_rate": 0.0001,
+      "loss": 4.4239,
+      "loss/crossentropy": 1.9414420127868652,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22160177677869797,
+      "step": 6836
+    },
+    {
+      "epoch": 0.13676,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0072174072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4294,
+      "loss/crossentropy": 2.280028223991394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24696747958660126,
+      "step": 6838
+    },
+    {
+      "epoch": 0.1368,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.006787109375,
+      "learning_rate": 0.0001,
+      "loss": 4.523,
+      "loss/crossentropy": 2.106986403465271,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24022039771080017,
+      "step": 6840
+    },
+    {
+      "epoch": 0.13684,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0349517822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2609,
+      "loss/crossentropy": 1.9540700912475586,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1968660056591034,
+      "step": 6842
+    },
+    {
+      "epoch": 0.13688,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.03394775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2181,
+      "loss/crossentropy": 1.6771780252456665,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20230162143707275,
+      "step": 6844
+    },
+    {
+      "epoch": 0.13692,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.04419657389322917,
+      "learning_rate": 0.0001,
+      "loss": 4.7567,
+      "loss/crossentropy": 2.059873402118683,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.276339128613472,
+      "step": 6846
+    },
+    {
+      "epoch": 0.13696,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.04684244791666667,
+      "learning_rate": 0.0001,
+      "loss": 4.2512,
+      "loss/crossentropy": 1.7943353056907654,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1893974468111992,
+      "step": 6848
+    },
+    {
+      "epoch": 0.137,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.046223958333333336,
+      "learning_rate": 0.0001,
+      "loss": 4.687,
+      "loss/crossentropy": 2.314136028289795,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24898302555084229,
+      "step": 6850
+    },
+    {
+      "epoch": 0.13704,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.04383036295572917,
+      "learning_rate": 0.0001,
+      "loss": 4.4257,
+      "loss/crossentropy": 2.0062466263771057,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23169535398483276,
+      "step": 6852
+    },
+    {
+      "epoch": 0.13708,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.044873046875,
+      "learning_rate": 0.0001,
+      "loss": 4.5787,
+      "loss/crossentropy": 2.3600821495056152,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25484780967235565,
+      "step": 6854
+    },
+    {
+      "epoch": 0.13712,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.043675740559895836,
+      "learning_rate": 0.0001,
+      "loss": 4.2113,
+      "loss/crossentropy": 1.885023295879364,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.255520723760128,
+      "step": 6856
+    },
+    {
+      "epoch": 0.13716,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.019806925455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.6493,
+      "loss/crossentropy": 2.2864162921905518,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22469403594732285,
+      "step": 6858
+    },
+    {
+      "epoch": 0.1372,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.018961588541666668,
+      "learning_rate": 0.0001,
+      "loss": 4.4017,
+      "loss/crossentropy": 1.908643126487732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22016740590333939,
+      "step": 6860
+    },
+    {
+      "epoch": 0.13724,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.010282389322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.7255,
+      "loss/crossentropy": 2.1028786301612854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23632919788360596,
+      "step": 6862
+    },
+    {
+      "epoch": 0.13728,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.01539306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4043,
+      "loss/crossentropy": 2.0363592505455017,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22131157666444778,
+      "step": 6864
+    },
+    {
+      "epoch": 0.13732,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0158843994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.4357,
+      "loss/crossentropy": 2.030495524406433,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.233994759619236,
+      "step": 6866
+    },
+    {
+      "epoch": 0.13736,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.018033854166666665,
+      "learning_rate": 0.0001,
+      "loss": 4.3036,
+      "loss/crossentropy": 1.6365603804588318,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1909056007862091,
+      "step": 6868
+    },
+    {
+      "epoch": 0.1374,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.015217081705729166,
+      "learning_rate": 0.0001,
+      "loss": 4.655,
+      "loss/crossentropy": 2.205111026763916,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2267644703388214,
+      "step": 6870
+    },
+    {
+      "epoch": 0.13744,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.016437784830729166,
+      "learning_rate": 0.0001,
+      "loss": 4.4207,
+      "loss/crossentropy": 2.0179646015167236,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2307143434882164,
+      "step": 6872
+    },
+    {
+      "epoch": 0.13748,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.018290201822916668,
+      "learning_rate": 0.0001,
+      "loss": 4.2154,
+      "loss/crossentropy": 1.9697463512420654,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2207925096154213,
+      "step": 6874
+    },
+    {
+      "epoch": 0.13752,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.019189453125,
+      "learning_rate": 0.0001,
+      "loss": 4.5691,
+      "loss/crossentropy": 2.186875820159912,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25762687623500824,
+      "step": 6876
+    },
+    {
+      "epoch": 0.13756,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.017867024739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.2713,
+      "loss/crossentropy": 2.3203837871551514,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23470903187990189,
+      "step": 6878
+    },
+    {
+      "epoch": 0.1376,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0066640218098958336,
+      "learning_rate": 0.0001,
+      "loss": 4.4154,
+      "loss/crossentropy": 2.2642472982406616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2454553246498108,
+      "step": 6880
+    },
+    {
+      "epoch": 0.13764,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.008687337239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.4076,
+      "loss/crossentropy": 1.9313859343528748,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21797804534435272,
+      "step": 6882
+    },
+    {
+      "epoch": 0.13768,
+      "grad_norm": 3.4375,
+      "grad_norm_var": 0.11204020182291667,
+      "learning_rate": 0.0001,
+      "loss": 4.7846,
+      "loss/crossentropy": 2.5469977855682373,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2651800066232681,
+      "step": 6884
+    },
+    {
+      "epoch": 0.13772,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.11030171712239584,
+      "learning_rate": 0.0001,
+      "loss": 4.4893,
+      "loss/crossentropy": 2.549328088760376,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2566085457801819,
+      "step": 6886
+    },
+    {
+      "epoch": 0.13776,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.1126617431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4927,
+      "loss/crossentropy": 2.3094369769096375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24709751456975937,
+      "step": 6888
+    },
+    {
+      "epoch": 0.1378,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.1097808837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2412,
+      "loss/crossentropy": 1.5071046948432922,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19776180386543274,
+      "step": 6890
+    },
+    {
+      "epoch": 0.13784,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.1101226806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3705,
+      "loss/crossentropy": 2.0064221620559692,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20836234837770462,
+      "step": 6892
+    },
+    {
+      "epoch": 0.13788,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.10816650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3608,
+      "loss/crossentropy": 2.1216301321983337,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22151901572942734,
+      "step": 6894
+    },
+    {
+      "epoch": 0.13792,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.1073150634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4168,
+      "loss/crossentropy": 1.8417679071426392,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21999332308769226,
+      "step": 6896
+    },
+    {
+      "epoch": 0.13796,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.10695699055989584,
+      "learning_rate": 0.0001,
+      "loss": 4.7005,
+      "loss/crossentropy": 2.4651769399642944,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27029043436050415,
+      "step": 6898
+    },
+    {
+      "epoch": 0.138,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.014188639322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.2985,
+      "loss/crossentropy": 1.7225988507270813,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20930497348308563,
+      "step": 6900
+    },
+    {
+      "epoch": 0.13804,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.015360514322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.1156,
+      "loss/crossentropy": 2.1218297481536865,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21276423335075378,
+      "step": 6902
+    },
+    {
+      "epoch": 0.13808,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.010856119791666667,
+      "learning_rate": 0.0001,
+      "loss": 4.2706,
+      "loss/crossentropy": 2.040019452571869,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2438269630074501,
+      "step": 6904
+    },
+    {
+      "epoch": 0.13812,
+      "grad_norm": 15.8125,
+      "grad_norm_var": 11.600536092122395,
+      "learning_rate": 0.0001,
+      "loss": 4.5041,
+      "loss/crossentropy": 1.8229625225067139,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22197778522968292,
+      "step": 6906
+    },
+    {
+      "epoch": 0.13816,
+      "grad_norm": 2.25,
+      "grad_norm_var": 11.543973795572917,
+      "learning_rate": 0.0001,
+      "loss": 4.7087,
+      "loss/crossentropy": 2.453408360481262,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24046239256858826,
+      "step": 6908
+    },
+    {
+      "epoch": 0.1382,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 11.55152587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4806,
+      "loss/crossentropy": 2.2724320888519287,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23739346861839294,
+      "step": 6910
+    },
+    {
+      "epoch": 0.13824,
+      "grad_norm": 2.125,
+      "grad_norm_var": 11.562272135416666,
+      "learning_rate": 0.0001,
+      "loss": 4.2747,
+      "loss/crossentropy": 2.2382686138153076,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22692064195871353,
+      "step": 6912
+    },
+    {
+      "epoch": 0.13828,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 11.54869384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.6867,
+      "loss/crossentropy": 2.021562337875366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24746088683605194,
+      "step": 6914
+    },
+    {
+      "epoch": 0.13832,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 11.546556599934895,
+      "learning_rate": 0.0001,
+      "loss": 4.3247,
+      "loss/crossentropy": 2.1071943044662476,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2144002616405487,
+      "step": 6916
+    },
+    {
+      "epoch": 0.13836,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 11.550846354166667,
+      "learning_rate": 0.0001,
+      "loss": 4.2686,
+      "loss/crossentropy": 1.9641517400741577,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22584721446037292,
+      "step": 6918
+    },
+    {
+      "epoch": 0.1384,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 11.559130859375,
+      "learning_rate": 0.0001,
+      "loss": 4.3194,
+      "loss/crossentropy": 2.4430564641952515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24351171404123306,
+      "step": 6920
+    },
+    {
+      "epoch": 0.13844,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.022484334309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4202,
+      "loss/crossentropy": 2.2237725257873535,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2248261496424675,
+      "step": 6922
+    },
+    {
+      "epoch": 0.13848,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.015412394205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.2028,
+      "loss/crossentropy": 1.7291913628578186,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19805100560188293,
+      "step": 6924
+    },
+    {
+      "epoch": 0.13852,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.013923136393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.3972,
+      "loss/crossentropy": 1.807108223438263,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20386626571416855,
+      "step": 6926
+    },
+    {
+      "epoch": 0.13856,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.01422119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.5188,
+      "loss/crossentropy": 2.510676622390747,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24666880816221237,
+      "step": 6928
+    },
+    {
+      "epoch": 0.1386,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007515462239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.2006,
+      "loss/crossentropy": 1.9420115947723389,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2214776575565338,
+      "step": 6930
+    },
+    {
+      "epoch": 0.13864,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.010837554931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4445,
+      "loss/crossentropy": 2.2288190722465515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23719585686922073,
+      "step": 6932
+    },
+    {
+      "epoch": 0.13868,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.011043294270833334,
+      "learning_rate": 0.0001,
+      "loss": 4.071,
+      "loss/crossentropy": 2.04274183511734,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21653369069099426,
+      "step": 6934
+    },
+    {
+      "epoch": 0.13872,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.0121002197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4041,
+      "loss/crossentropy": 1.9149779081344604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19941364973783493,
+      "step": 6936
+    },
+    {
+      "epoch": 0.13876,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.011554972330729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2577,
+      "loss/crossentropy": 1.7983179092407227,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18979590386152267,
+      "step": 6938
+    },
+    {
+      "epoch": 0.1388,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.023164876302083335,
+      "learning_rate": 0.0001,
+      "loss": 4.3314,
+      "loss/crossentropy": 2.1919915080070496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22652295976877213,
+      "step": 6940
+    },
+    {
+      "epoch": 0.13884,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.023152669270833332,
+      "learning_rate": 0.0001,
+      "loss": 4.494,
+      "loss/crossentropy": 2.0362821221351624,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24296525120735168,
+      "step": 6942
+    },
+    {
+      "epoch": 0.13888,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.023653157552083335,
+      "learning_rate": 0.0001,
+      "loss": 4.4135,
+      "loss/crossentropy": 2.0371538400650024,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2217850610613823,
+      "step": 6944
+    },
+    {
+      "epoch": 0.13892,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.0268310546875,
+      "learning_rate": 0.0001,
+      "loss": 4.3371,
+      "loss/crossentropy": 1.9137988686561584,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2218162938952446,
+      "step": 6946
+    },
+    {
+      "epoch": 0.13896,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.03438898722330729,
+      "learning_rate": 0.0001,
+      "loss": 4.6521,
+      "loss/crossentropy": 2.3215843439102173,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2576482892036438,
+      "step": 6948
+    },
+    {
+      "epoch": 0.139,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.026590983072916668,
+      "learning_rate": 0.0001,
+      "loss": 4.6004,
+      "loss/crossentropy": 2.169154405593872,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24629274010658264,
+      "step": 6950
+    },
+    {
+      "epoch": 0.13904,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.03173421223958333,
+      "learning_rate": 0.0001,
+      "loss": 4.3549,
+      "loss/crossentropy": 2.1355135440826416,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2316850870847702,
+      "step": 6952
+    },
+    {
+      "epoch": 0.13908,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.03183186848958333,
+      "learning_rate": 0.0001,
+      "loss": 4.3462,
+      "loss/crossentropy": 2.264985144138336,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23467915505170822,
+      "step": 6954
+    },
+    {
+      "epoch": 0.13912,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.025926717122395835,
+      "learning_rate": 0.0001,
+      "loss": 4.6559,
+      "loss/crossentropy": 2.1007654666900635,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2376948595046997,
+      "step": 6956
+    },
+    {
+      "epoch": 0.13916,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.02603759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4871,
+      "loss/crossentropy": 2.284608840942383,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24811238050460815,
+      "step": 6958
+    },
+    {
+      "epoch": 0.1392,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.025911458333333335,
+      "learning_rate": 0.0001,
+      "loss": 4.2338,
+      "loss/crossentropy": 1.657732367515564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19541934877634048,
+      "step": 6960
+    },
+    {
+      "epoch": 0.13924,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.022391764322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.3832,
+      "loss/crossentropy": 1.9607325792312622,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2269211858510971,
+      "step": 6962
+    },
+    {
+      "epoch": 0.13928,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.006571451822916667,
+      "learning_rate": 0.0001,
+      "loss": 4.3759,
+      "loss/crossentropy": 1.7454752326011658,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21780573576688766,
+      "step": 6964
+    },
+    {
+      "epoch": 0.13932,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.010350545247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.7664,
+      "loss/crossentropy": 2.001866638660431,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24929454922676086,
+      "step": 6966
+    },
+    {
+      "epoch": 0.13936,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.008006795247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4181,
+      "loss/crossentropy": 1.9167855978012085,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22114858031272888,
+      "step": 6968
+    },
+    {
+      "epoch": 0.1394,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.008003743489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.2284,
+      "loss/crossentropy": 2.0324739813804626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23271320760250092,
+      "step": 6970
+    },
+    {
+      "epoch": 0.13944,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.016307576497395834,
+      "learning_rate": 0.0001,
+      "loss": 4.5375,
+      "loss/crossentropy": 2.162013590335846,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22832238674163818,
+      "step": 6972
+    },
+    {
+      "epoch": 0.13948,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.018944295247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1406,
+      "loss/crossentropy": 2.074672818183899,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2402098923921585,
+      "step": 6974
+    },
+    {
+      "epoch": 0.13952,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.019758097330729165,
+      "learning_rate": 0.0001,
+      "loss": 4.4221,
+      "loss/crossentropy": 1.9982299208641052,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21126113086938858,
+      "step": 6976
+    },
+    {
+      "epoch": 0.13956,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.021751912434895833,
+      "learning_rate": 0.0001,
+      "loss": 4.3391,
+      "loss/crossentropy": 1.944049894809723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2296794354915619,
+      "step": 6978
+    },
+    {
+      "epoch": 0.1396,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.02047119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.4344,
+      "loss/crossentropy": 2.308506488800049,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2589666247367859,
+      "step": 6980
+    },
+    {
+      "epoch": 0.13964,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.017724609375,
+      "learning_rate": 0.0001,
+      "loss": 4.2867,
+      "loss/crossentropy": 2.129163682460785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21905581653118134,
+      "step": 6982
+    },
+    {
+      "epoch": 0.13968,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.017902628580729166,
+      "learning_rate": 0.0001,
+      "loss": 4.3023,
+      "loss/crossentropy": 1.8560669422149658,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2119361013174057,
+      "step": 6984
+    },
+    {
+      "epoch": 0.13972,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.018602498372395835,
+      "learning_rate": 0.0001,
+      "loss": 4.1212,
+      "loss/crossentropy": 1.8194095492362976,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20872193574905396,
+      "step": 6986
+    },
+    {
+      "epoch": 0.13976,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.005501302083333334,
+      "learning_rate": 0.0001,
+      "loss": 4.2612,
+      "loss/crossentropy": 2.0200153589248657,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20160746574401855,
+      "step": 6988
+    },
+    {
+      "epoch": 0.1398,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.007298787434895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2757,
+      "loss/crossentropy": 1.982479751110077,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23396103084087372,
+      "step": 6990
+    },
+    {
+      "epoch": 0.13984,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.006494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.5075,
+      "loss/crossentropy": 2.17054283618927,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23827539384365082,
+      "step": 6992
+    },
+    {
+      "epoch": 0.13988,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1794,
+      "loss/crossentropy": 1.619499921798706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1917721927165985,
+      "step": 6994
+    },
+    {
+      "epoch": 0.13992,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.005353800455729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3833,
+      "loss/crossentropy": 2.1082500219345093,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23768241703510284,
+      "step": 6996
+    },
+    {
+      "epoch": 0.13996,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.004076131184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.6731,
+      "loss/crossentropy": 1.8480825424194336,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21046122163534164,
+      "step": 6998
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.008463541666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.5285,
+      "loss/crossentropy": 2.0547631978988647,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22980494797229767,
+      "step": 7000
+    },
+    {
+      "epoch": 0.14004,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.007835896809895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4625,
+      "loss/crossentropy": 2.0695141553878784,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2100546732544899,
+      "step": 7002
+    },
+    {
+      "epoch": 0.14008,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.012277984619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3716,
+      "loss/crossentropy": 2.105263113975525,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24552470445632935,
+      "step": 7004
+    },
+    {
+      "epoch": 0.14012,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.014427693684895833,
+      "learning_rate": 0.0001,
+      "loss": 4.3566,
+      "loss/crossentropy": 2.03000670671463,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24780434370040894,
+      "step": 7006
+    },
+    {
+      "epoch": 0.14016,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.0173980712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4758,
+      "loss/crossentropy": 2.288944959640503,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2760937511920929,
+      "step": 7008
+    },
+    {
+      "epoch": 0.1402,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.06544570922851563,
+      "learning_rate": 0.0001,
+      "loss": 3.9326,
+      "loss/crossentropy": 1.790147304534912,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1905786320567131,
+      "step": 7010
+    },
+    {
+      "epoch": 0.14024,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.07765884399414062,
+      "learning_rate": 0.0001,
+      "loss": 4.6364,
+      "loss/crossentropy": 2.008346378803253,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23138166218996048,
+      "step": 7012
+    },
+    {
+      "epoch": 0.14028,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.07974014282226563,
+      "learning_rate": 0.0001,
+      "loss": 4.2304,
+      "loss/crossentropy": 1.9694496393203735,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21347863972187042,
+      "step": 7014
+    },
+    {
+      "epoch": 0.14032,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.07948989868164062,
+      "learning_rate": 0.0001,
+      "loss": 4.3685,
+      "loss/crossentropy": 2.0907286405563354,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23349857330322266,
+      "step": 7016
+    },
+    {
+      "epoch": 0.14036,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.07850316365559896,
+      "learning_rate": 0.0001,
+      "loss": 4.6454,
+      "loss/crossentropy": 2.161414623260498,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23841089010238647,
+      "step": 7018
+    },
+    {
+      "epoch": 0.1404,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.07315266927083333,
+      "learning_rate": 0.0001,
+      "loss": 4.4868,
+      "loss/crossentropy": 2.1402887105941772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25493840873241425,
+      "step": 7020
+    },
+    {
+      "epoch": 0.14044,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.06809666951497396,
+      "learning_rate": 0.0001,
+      "loss": 4.1248,
+      "loss/crossentropy": 2.0703811049461365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24300381541252136,
+      "step": 7022
+    },
+    {
+      "epoch": 0.14048,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.0716875712076823,
+      "learning_rate": 0.0001,
+      "loss": 4.6165,
+      "loss/crossentropy": 2.152569532394409,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23347805440425873,
+      "step": 7024
+    },
+    {
+      "epoch": 0.14052,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.03453369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1642,
+      "loss/crossentropy": 2.0831095576286316,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2198955938220024,
+      "step": 7026
+    },
+    {
+      "epoch": 0.14056,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.019498697916666665,
+      "learning_rate": 0.0001,
+      "loss": 4.4929,
+      "loss/crossentropy": 2.1631508469581604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23971489816904068,
+      "step": 7028
+    },
+    {
+      "epoch": 0.1406,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.019481404622395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4687,
+      "loss/crossentropy": 2.1683043241500854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2628757208585739,
+      "step": 7030
+    },
+    {
+      "epoch": 0.14064,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.07377827962239583,
+      "learning_rate": 0.0001,
+      "loss": 4.522,
+      "loss/crossentropy": 2.021001398563385,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22922180593013763,
+      "step": 7032
+    },
+    {
+      "epoch": 0.14068,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0725494384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.7729,
+      "loss/crossentropy": 2.267430543899536,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27525072544813156,
+      "step": 7034
+    },
+    {
+      "epoch": 0.14072,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0788726806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3118,
+      "loss/crossentropy": 2.066729426383972,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21959475427865982,
+      "step": 7036
+    },
+    {
+      "epoch": 0.14076,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.08162333170572916,
+      "learning_rate": 0.0001,
+      "loss": 4.3203,
+      "loss/crossentropy": 1.7972697019577026,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21773608028888702,
+      "step": 7038
+    },
+    {
+      "epoch": 0.1408,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.08068745930989583,
+      "learning_rate": 0.0001,
+      "loss": 4.0745,
+      "loss/crossentropy": 1.751904845237732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19929176568984985,
+      "step": 7040
+    },
+    {
+      "epoch": 0.14084,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.07649332682291667,
+      "learning_rate": 0.0001,
+      "loss": 4.4188,
+      "loss/crossentropy": 1.8432873487472534,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21235650032758713,
+      "step": 7042
+    },
+    {
+      "epoch": 0.14088,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.07618815104166667,
+      "learning_rate": 0.0001,
+      "loss": 4.2343,
+      "loss/crossentropy": 1.9589285850524902,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21506928652524948,
+      "step": 7044
+    },
+    {
+      "epoch": 0.14092,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0788726806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.229,
+      "loss/crossentropy": 2.3658028841018677,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23405643552541733,
+      "step": 7046
+    },
+    {
+      "epoch": 0.14096,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.008837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3922,
+      "loss/crossentropy": 2.088135540485382,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22823208570480347,
+      "step": 7048
+    },
+    {
+      "epoch": 0.141,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.006810506184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2367,
+      "loss/crossentropy": 1.9309821724891663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21409911662340164,
+      "step": 7050
+    },
+    {
+      "epoch": 0.14104,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.006636555989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.5237,
+      "loss/crossentropy": 2.5411492586135864,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25017087161540985,
+      "step": 7052
+    },
+    {
+      "epoch": 0.14108,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.007352701822916667,
+      "learning_rate": 0.0001,
+      "loss": 4.4998,
+      "loss/crossentropy": 2.3210322856903076,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23967693746089935,
+      "step": 7054
+    },
+    {
+      "epoch": 0.14112,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.009370930989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.4607,
+      "loss/crossentropy": 2.054674744606018,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23270255327224731,
+      "step": 7056
+    },
+    {
+      "epoch": 0.14116,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.008968098958333334,
+      "learning_rate": 0.0001,
+      "loss": 4.5423,
+      "loss/crossentropy": 2.545789122581482,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25560182332992554,
+      "step": 7058
+    },
+    {
+      "epoch": 0.1412,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.008861287434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.2682,
+      "loss/crossentropy": 2.262348175048828,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24763159453868866,
+      "step": 7060
+    },
+    {
+      "epoch": 0.14124,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.008447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.618,
+      "loss/crossentropy": 2.1045475602149963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2472379505634308,
+      "step": 7062
+    },
+    {
+      "epoch": 0.14128,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0093170166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2312,
+      "loss/crossentropy": 1.5632115006446838,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19010942429304123,
+      "step": 7064
+    },
+    {
+      "epoch": 0.14132,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0116851806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2638,
+      "loss/crossentropy": 2.0847875475883484,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21371345967054367,
+      "step": 7066
+    },
+    {
+      "epoch": 0.14136,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.018016560872395834,
+      "learning_rate": 0.0001,
+      "loss": 4.5171,
+      "loss/crossentropy": 2.2243804931640625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23065787553787231,
+      "step": 7068
+    },
+    {
+      "epoch": 0.1414,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.016927083333333332,
+      "learning_rate": 0.0001,
+      "loss": 4.2812,
+      "loss/crossentropy": 1.9477753639221191,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21616356819868088,
+      "step": 7070
+    },
+    {
+      "epoch": 0.14144,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.024169921875,
+      "learning_rate": 0.0001,
+      "loss": 4.7005,
+      "loss/crossentropy": 2.2598072290420532,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25795431435108185,
+      "step": 7072
+    },
+    {
+      "epoch": 0.14148,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.022652180989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.4345,
+      "loss/crossentropy": 1.8817242980003357,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21974974125623703,
+      "step": 7074
+    },
+    {
+      "epoch": 0.14152,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0256744384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.5688,
+      "loss/crossentropy": 2.5275847911834717,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24608048796653748,
+      "step": 7076
+    },
+    {
+      "epoch": 0.14156,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0265045166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3937,
+      "loss/crossentropy": 2.400865852832794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22960034757852554,
+      "step": 7078
+    },
+    {
+      "epoch": 0.1416,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0237457275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.4005,
+      "loss/crossentropy": 1.908901333808899,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22271078824996948,
+      "step": 7080
+    },
+    {
+      "epoch": 0.14164,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.02213134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1384,
+      "loss/crossentropy": 2.3330780267715454,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23944097012281418,
+      "step": 7082
+    },
+    {
+      "epoch": 0.14168,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.018941243489583332,
+      "learning_rate": 0.0001,
+      "loss": 4.3516,
+      "loss/crossentropy": 2.332213521003723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23576530069112778,
+      "step": 7084
+    },
+    {
+      "epoch": 0.14172,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.022435506184895832,
+      "learning_rate": 0.0001,
+      "loss": 4.4654,
+      "loss/crossentropy": 2.2269067764282227,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25341375917196274,
+      "step": 7086
+    },
+    {
+      "epoch": 0.14176,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.01470947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2491,
+      "loss/crossentropy": 2.461983561515808,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24491792172193527,
+      "step": 7088
+    },
+    {
+      "epoch": 0.1418,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.013952382405598958,
+      "learning_rate": 0.0001,
+      "loss": 4.2348,
+      "loss/crossentropy": 2.428719997406006,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2394469603896141,
+      "step": 7090
+    },
+    {
+      "epoch": 0.14184,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011433664957682292,
+      "learning_rate": 0.0001,
+      "loss": 4.2242,
+      "loss/crossentropy": 2.32351291179657,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23441501706838608,
+      "step": 7092
+    },
+    {
+      "epoch": 0.14188,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.011482493082682291,
+      "learning_rate": 0.0001,
+      "loss": 4.4155,
+      "loss/crossentropy": 2.1165764331817627,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2350979596376419,
+      "step": 7094
+    },
+    {
+      "epoch": 0.14192,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.014288075764973958,
+      "learning_rate": 0.0001,
+      "loss": 4.4952,
+      "loss/crossentropy": 2.150681734085083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2475891336798668,
+      "step": 7096
+    },
+    {
+      "epoch": 0.14196,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.013079579671223958,
+      "learning_rate": 0.0001,
+      "loss": 4.2753,
+      "loss/crossentropy": 2.038177013397217,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2293965071439743,
+      "step": 7098
+    },
+    {
+      "epoch": 0.142,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.014062245686848959,
+      "learning_rate": 0.0001,
+      "loss": 4.298,
+      "loss/crossentropy": 1.899521827697754,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20856370776891708,
+      "step": 7100
+    },
+    {
+      "epoch": 0.14204,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.010593414306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3449,
+      "loss/crossentropy": 1.9807924032211304,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22756796330213547,
+      "step": 7102
+    },
+    {
+      "epoch": 0.14208,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009421539306640626,
+      "learning_rate": 0.0001,
+      "loss": 4.2893,
+      "loss/crossentropy": 2.158667206764221,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22594892233610153,
+      "step": 7104
+    },
+    {
+      "epoch": 0.14212,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.007811482747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3751,
+      "loss/crossentropy": 2.3133270144462585,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22925584018230438,
+      "step": 7106
+    },
+    {
+      "epoch": 0.14216,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010107421875,
+      "learning_rate": 0.0001,
+      "loss": 4.2762,
+      "loss/crossentropy": 1.9796301126480103,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21650104224681854,
+      "step": 7108
+    },
+    {
+      "epoch": 0.1422,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.010054524739583333,
+      "learning_rate": 0.0001,
+      "loss": 4.2778,
+      "loss/crossentropy": 2.092659056186676,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23407263308763504,
+      "step": 7110
+    },
+    {
+      "epoch": 0.14224,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.009056599934895833,
+      "learning_rate": 0.0001,
+      "loss": 4.6078,
+      "loss/crossentropy": 1.970819890499115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2262839823961258,
+      "step": 7112
+    },
+    {
+      "epoch": 0.14228,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.011923980712890626,
+      "learning_rate": 0.0001,
+      "loss": 4.0642,
+      "loss/crossentropy": 1.5877107381820679,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.182430237531662,
+      "step": 7114
+    },
+    {
+      "epoch": 0.14232,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.020336659749348958,
+      "learning_rate": 0.0001,
+      "loss": 4.6997,
+      "loss/crossentropy": 2.3208755254745483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25018931180238724,
+      "step": 7116
+    },
+    {
+      "epoch": 0.14236,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.01904271443684896,
+      "learning_rate": 0.0001,
+      "loss": 4.4483,
+      "loss/crossentropy": 2.1348973512649536,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23516173660755157,
+      "step": 7118
+    },
+    {
+      "epoch": 0.1424,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.02533543904622396,
+      "learning_rate": 0.0001,
+      "loss": 4.3443,
+      "loss/crossentropy": 2.1442995071411133,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23094037175178528,
+      "step": 7120
+    },
+    {
+      "epoch": 0.14244,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.025608062744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3297,
+      "loss/crossentropy": 2.24001145362854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2479296177625656,
+      "step": 7122
+    },
+    {
+      "epoch": 0.14248,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.038917795817057295,
+      "learning_rate": 0.0001,
+      "loss": 4.4122,
+      "loss/crossentropy": 1.8284733891487122,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23670874536037445,
+      "step": 7124
+    },
+    {
+      "epoch": 0.14252,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.04146499633789062,
+      "learning_rate": 0.0001,
+      "loss": 4.6223,
+      "loss/crossentropy": 2.1003533601760864,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24235840141773224,
+      "step": 7126
+    },
+    {
+      "epoch": 0.14256,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.04201024373372396,
+      "learning_rate": 0.0001,
+      "loss": 4.2591,
+      "loss/crossentropy": 2.3661316633224487,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.258208692073822,
+      "step": 7128
+    },
+    {
+      "epoch": 0.1426,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.030989583333333334,
+      "learning_rate": 0.0001,
+      "loss": 4.3288,
+      "loss/crossentropy": 2.2374593019485474,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2543400973081589,
+      "step": 7130
+    },
+    {
+      "epoch": 0.14264,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.02945556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.5939,
+      "loss/crossentropy": 1.9141342639923096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23257911205291748,
+      "step": 7132
+    },
+    {
+      "epoch": 0.14268,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.031224568684895832,
+      "learning_rate": 0.0001,
+      "loss": 4.1893,
+      "loss/crossentropy": 1.992666780948639,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2298717424273491,
+      "step": 7134
+    },
+    {
+      "epoch": 0.14272,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.033503214518229164,
+      "learning_rate": 0.0001,
+      "loss": 4.2665,
+      "loss/crossentropy": 1.9794283509254456,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2112378552556038,
+      "step": 7136
+    },
+    {
+      "epoch": 0.14276,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.0382232666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3801,
+      "loss/crossentropy": 2.1011139154434204,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27278490364551544,
+      "step": 7138
+    },
+    {
+      "epoch": 0.1428,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.02340087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.6989,
+      "loss/crossentropy": 2.3489880561828613,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24445360153913498,
+      "step": 7140
+    },
+    {
+      "epoch": 0.14284,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.0173736572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3418,
+      "loss/crossentropy": 2.011172831058502,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.210151307284832,
+      "step": 7142
+    },
+    {
+      "epoch": 0.14288,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.017366536458333335,
+      "learning_rate": 0.0001,
+      "loss": 4.3488,
+      "loss/crossentropy": 1.963642418384552,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22939135879278183,
+      "step": 7144
+    },
+    {
+      "epoch": 0.14292,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.016194661458333332,
+      "learning_rate": 0.0001,
+      "loss": 4.5166,
+      "loss/crossentropy": 2.2739341259002686,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23235367238521576,
+      "step": 7146
+    },
+    {
+      "epoch": 0.14296,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.018973795572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.5977,
+      "loss/crossentropy": 2.282576322555542,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24144183099269867,
+      "step": 7148
+    },
+    {
+      "epoch": 0.143,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.02329279581705729,
+      "learning_rate": 0.0001,
+      "loss": 4.141,
+      "loss/crossentropy": 1.7847901582717896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19184929877519608,
+      "step": 7150
+    },
+    {
+      "epoch": 0.14304,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.021740468343098958,
+      "learning_rate": 0.0001,
+      "loss": 4.3379,
+      "loss/crossentropy": 2.165170907974243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.223430335521698,
+      "step": 7152
+    },
+    {
+      "epoch": 0.14308,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.018873850504557293,
+      "learning_rate": 0.0001,
+      "loss": 4.3407,
+      "loss/crossentropy": 2.0395787954330444,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21662698686122894,
+      "step": 7154
+    },
+    {
+      "epoch": 0.14312,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.01907323201497396,
+      "learning_rate": 0.0001,
+      "loss": 4.4936,
+      "loss/crossentropy": 2.014316141605377,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23184175789356232,
+      "step": 7156
+    },
+    {
+      "epoch": 0.14316,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.019419097900390626,
+      "learning_rate": 0.0001,
+      "loss": 4.5612,
+      "loss/crossentropy": 2.2581117153167725,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24055174738168716,
+      "step": 7158
+    },
+    {
+      "epoch": 0.1432,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.021022288004557292,
+      "learning_rate": 0.0001,
+      "loss": 4.3853,
+      "loss/crossentropy": 2.0905630588531494,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2719188630580902,
+      "step": 7160
+    },
+    {
+      "epoch": 0.14324,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.025233713785807292,
+      "learning_rate": 0.0001,
+      "loss": 4.6485,
+      "loss/crossentropy": 2.414529800415039,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24188701063394547,
+      "step": 7162
+    },
+    {
+      "epoch": 0.14328,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.02240778605143229,
+      "learning_rate": 0.0001,
+      "loss": 4.3605,
+      "loss/crossentropy": 2.028432607650757,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2300800457596779,
+      "step": 7164
+    },
+    {
+      "epoch": 0.14332,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.03299153645833333,
+      "learning_rate": 0.0001,
+      "loss": 4.2947,
+      "loss/crossentropy": 2.096144199371338,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23311930894851685,
+      "step": 7166
+    },
+    {
+      "epoch": 0.14336,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.029622395833333332,
+      "learning_rate": 0.0001,
+      "loss": 4.4375,
+      "loss/crossentropy": 2.259281277656555,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25558799505233765,
+      "step": 7168
+    },
+    {
+      "epoch": 0.1434,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0219390869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.4364,
+      "loss/crossentropy": 2.0766254663467407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22725434601306915,
+      "step": 7170
+    },
+    {
+      "epoch": 0.14344,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.024312337239583332,
+      "learning_rate": 0.0001,
+      "loss": 4.4695,
+      "loss/crossentropy": 2.26702618598938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22584324330091476,
+      "step": 7172
+    },
+    {
+      "epoch": 0.14348,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.027534993489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.3417,
+      "loss/crossentropy": 2.1933096647262573,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23658160120248795,
+      "step": 7174
+    },
+    {
+      "epoch": 0.14352,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.027372233072916665,
+      "learning_rate": 0.0001,
+      "loss": 4.351,
+      "loss/crossentropy": 2.2003660202026367,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22349942475557327,
+      "step": 7176
+    },
+    {
+      "epoch": 0.14356,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.025145467122395834,
+      "learning_rate": 0.0001,
+      "loss": 4.684,
+      "loss/crossentropy": 2.4630067348480225,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.253003865480423,
+      "step": 7178
+    },
+    {
+      "epoch": 0.1436,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.027082316080729165,
+      "learning_rate": 0.0001,
+      "loss": 4.6875,
+      "loss/crossentropy": 2.264480948448181,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27604997158050537,
+      "step": 7180
+    },
+    {
+      "epoch": 0.14364,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.00712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3064,
+      "loss/crossentropy": 2.1641955375671387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22587595880031586,
+      "step": 7182
+    },
+    {
+      "epoch": 0.14368,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.00943603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.6137,
+      "loss/crossentropy": 2.1432350873947144,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24324018508195877,
+      "step": 7184
+    },
+    {
+      "epoch": 0.14372,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.014611562093098959,
+      "learning_rate": 0.0001,
+      "loss": 4.3394,
+      "loss/crossentropy": 1.7448238134384155,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1751260682940483,
+      "step": 7186
+    },
+    {
+      "epoch": 0.14376,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.015295155843098958,
+      "learning_rate": 0.0001,
+      "loss": 4.3602,
+      "loss/crossentropy": 2.3202184438705444,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2314094603061676,
+      "step": 7188
+    },
+    {
+      "epoch": 0.1438,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.019681549072265624,
+      "learning_rate": 0.0001,
+      "loss": 4.187,
+      "loss/crossentropy": 1.970094919204712,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24998464435338974,
+      "step": 7190
+    },
+    {
+      "epoch": 0.14384,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.02075780232747396,
+      "learning_rate": 0.0001,
+      "loss": 4.215,
+      "loss/crossentropy": 2.1331114768981934,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2008385732769966,
+      "step": 7192
+    },
+    {
+      "epoch": 0.14388,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.021345774332682293,
+      "learning_rate": 0.0001,
+      "loss": 4.3976,
+      "loss/crossentropy": 2.1659106016159058,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22314336150884628,
+      "step": 7194
+    },
+    {
+      "epoch": 0.14392,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.018507639567057293,
+      "learning_rate": 0.0001,
+      "loss": 4.6324,
+      "loss/crossentropy": 2.3382883071899414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26393643021583557,
+      "step": 7196
+    },
+    {
+      "epoch": 0.14396,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.020499420166015626,
+      "learning_rate": 0.0001,
+      "loss": 4.7988,
+      "loss/crossentropy": 2.1325554847717285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2457222416996956,
+      "step": 7198
+    },
+    {
+      "epoch": 0.144,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.03144709269205729,
+      "learning_rate": 0.0001,
+      "loss": 4.3175,
+      "loss/crossentropy": 1.7927106022834778,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20284093916416168,
+      "step": 7200
+    },
+    {
+      "epoch": 0.14404,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.03050715128580729,
+      "learning_rate": 0.0001,
+      "loss": 4.1676,
+      "loss/crossentropy": 1.9799031615257263,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23475942015647888,
+      "step": 7202
+    },
+    {
+      "epoch": 0.14408,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.02939020792643229,
+      "learning_rate": 0.0001,
+      "loss": 4.2627,
+      "loss/crossentropy": 2.0590370893478394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2160736471414566,
+      "step": 7204
+    },
+    {
+      "epoch": 0.14412,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.02800267537434896,
+      "learning_rate": 0.0001,
+      "loss": 4.4397,
+      "loss/crossentropy": 1.9866149425506592,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23413674533367157,
+      "step": 7206
+    },
+    {
+      "epoch": 0.14416,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.03551610310872396,
+      "learning_rate": 0.0001,
+      "loss": 4.526,
+      "loss/crossentropy": 2.1320748925209045,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24469739198684692,
+      "step": 7208
+    },
+    {
+      "epoch": 0.1442,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.034708404541015626,
+      "learning_rate": 0.0001,
+      "loss": 4.5834,
+      "loss/crossentropy": 2.225857973098755,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23838083446025848,
+      "step": 7210
+    },
+    {
+      "epoch": 0.14424,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.03794733683268229,
+      "learning_rate": 0.0001,
+      "loss": 3.9952,
+      "loss/crossentropy": 1.9118528962135315,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2064301297068596,
+      "step": 7212
+    },
+    {
+      "epoch": 0.14428,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.03806940714518229,
+      "learning_rate": 0.0001,
+      "loss": 4.182,
+      "loss/crossentropy": 1.8142234086990356,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2172461450099945,
+      "step": 7214
+    },
+    {
+      "epoch": 0.14432,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.02540868123372396,
+      "learning_rate": 0.0001,
+      "loss": 4.4446,
+      "loss/crossentropy": 1.9308255910873413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2236044555902481,
+      "step": 7216
+    },
+    {
+      "epoch": 0.14436,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.022391764322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.4597,
+      "loss/crossentropy": 1.9821211695671082,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21110422909259796,
+      "step": 7218
+    },
+    {
+      "epoch": 0.1444,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.023176066080729165,
+      "learning_rate": 0.0001,
+      "loss": 4.6732,
+      "loss/crossentropy": 2.216045379638672,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23433538526296616,
+      "step": 7220
+    },
+    {
+      "epoch": 0.14444,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.024442545572916665,
+      "learning_rate": 0.0001,
+      "loss": 4.6089,
+      "loss/crossentropy": 2.2303662300109863,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2390453889966011,
+      "step": 7222
+    },
+    {
+      "epoch": 0.14448,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.017020670572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.0805,
+      "loss/crossentropy": 2.2152082920074463,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22196897864341736,
+      "step": 7224
+    },
+    {
+      "epoch": 0.14452,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.014860026041666667,
+      "learning_rate": 0.0001,
+      "loss": 4.2843,
+      "loss/crossentropy": 2.134513795375824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21964208781719208,
+      "step": 7226
+    },
+    {
+      "epoch": 0.14456,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.015262858072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.0289,
+      "loss/crossentropy": 1.6803861260414124,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18217483162879944,
+      "step": 7228
+    },
+    {
+      "epoch": 0.1446,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0152252197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1353,
+      "loss/crossentropy": 1.6597792506217957,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20317095518112183,
+      "step": 7230
+    },
+    {
+      "epoch": 0.14464,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.017704264322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.2978,
+      "loss/crossentropy": 1.776586651802063,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21835225820541382,
+      "step": 7232
+    },
+    {
+      "epoch": 0.14468,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.017154947916666666,
+      "learning_rate": 0.0001,
+      "loss": 4.1092,
+      "loss/crossentropy": 1.7347259521484375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21964067220687866,
+      "step": 7234
+    },
+    {
+      "epoch": 0.14472,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.015380859375,
+      "learning_rate": 0.0001,
+      "loss": 4.2069,
+      "loss/crossentropy": 1.79097181558609,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21776803582906723,
+      "step": 7236
+    },
+    {
+      "epoch": 0.14476,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0069244384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0335,
+      "loss/crossentropy": 2.051329553127289,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21669812500476837,
+      "step": 7238
+    },
+    {
+      "epoch": 0.1448,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0070220947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3206,
+      "loss/crossentropy": 1.965324580669403,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2283879667520523,
+      "step": 7240
+    },
+    {
+      "epoch": 0.14484,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.010692342122395834,
+      "learning_rate": 0.0001,
+      "loss": 4.5952,
+      "loss/crossentropy": 2.248784363269806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24906039983034134,
+      "step": 7242
+    },
+    {
+      "epoch": 0.14488,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010529581705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3321,
+      "loss/crossentropy": 1.9946890473365784,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22583268582820892,
+      "step": 7244
+    },
+    {
+      "epoch": 0.14492,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010660807291666666,
+      "learning_rate": 0.0001,
+      "loss": 4.1436,
+      "loss/crossentropy": 2.2306413650512695,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22068945318460464,
+      "step": 7246
+    },
+    {
+      "epoch": 0.14496,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.00758056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2159,
+      "loss/crossentropy": 2.110253095626831,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22045104205608368,
+      "step": 7248
+    },
+    {
+      "epoch": 0.145,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.008210245768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0693,
+      "loss/crossentropy": 1.928157925605774,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2043258175253868,
+      "step": 7250
+    },
+    {
+      "epoch": 0.14504,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.009110514322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.2458,
+      "loss/crossentropy": 2.2674691677093506,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23258862644433975,
+      "step": 7252
+    },
+    {
+      "epoch": 0.14508,
+      "grad_norm": 2.765625,
+      "grad_norm_var": 0.03052978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5107,
+      "loss/crossentropy": 2.2825024127960205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.263367660343647,
+      "step": 7254
+    },
+    {
+      "epoch": 0.14512,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.040185546875,
+      "learning_rate": 0.0001,
+      "loss": 4.2394,
+      "loss/crossentropy": 2.1546168327331543,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23073332011699677,
+      "step": 7256
+    },
+    {
+      "epoch": 0.14516,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.040816243489583334,
+      "learning_rate": 0.0001,
+      "loss": 4.5504,
+      "loss/crossentropy": 2.0490044951438904,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24161820113658905,
+      "step": 7258
+    },
+    {
+      "epoch": 0.1452,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.04038798014322917,
+      "learning_rate": 0.0001,
+      "loss": 4.6468,
+      "loss/crossentropy": 2.115446150302887,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2537624090909958,
+      "step": 7260
+    },
+    {
+      "epoch": 0.14524,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.22388407389322917,
+      "learning_rate": 0.0001,
+      "loss": 4.1457,
+      "loss/crossentropy": 2.0302165746688843,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21026766300201416,
+      "step": 7262
+    },
+    {
+      "epoch": 0.14528,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.2228179931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4077,
+      "loss/crossentropy": 2.102527379989624,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2354147508740425,
+      "step": 7264
+    },
+    {
+      "epoch": 0.14532,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.21614176432291668,
+      "learning_rate": 0.0001,
+      "loss": 4.1108,
+      "loss/crossentropy": 2.0095282793045044,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22679834067821503,
+      "step": 7266
+    },
+    {
+      "epoch": 0.14536,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.20706278483072918,
+      "learning_rate": 0.0001,
+      "loss": 4.3849,
+      "loss/crossentropy": 1.8988603353500366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21238050609827042,
+      "step": 7268
+    },
+    {
+      "epoch": 0.1454,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.19975484212239583,
+      "learning_rate": 0.0001,
+      "loss": 4.6614,
+      "loss/crossentropy": 2.186660885810852,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24453039467334747,
+      "step": 7270
+    },
+    {
+      "epoch": 0.14544,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.20244852701822916,
+      "learning_rate": 0.0001,
+      "loss": 4.1409,
+      "loss/crossentropy": 1.8927155137062073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2147793024778366,
+      "step": 7272
+    },
+    {
+      "epoch": 0.14548,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.20608317057291667,
+      "learning_rate": 0.0001,
+      "loss": 4.1375,
+      "loss/crossentropy": 1.8969642519950867,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21910040825605392,
+      "step": 7274
+    },
+    {
+      "epoch": 0.14552,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.20545247395833333,
+      "learning_rate": 0.0001,
+      "loss": 4.3325,
+      "loss/crossentropy": 2.090053617954254,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22576116025447845,
+      "step": 7276
+    },
+    {
+      "epoch": 0.14556,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.010480753580729167,
+      "learning_rate": 0.0001,
+      "loss": 4.7263,
+      "loss/crossentropy": 2.1606650352478027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22937766462564468,
+      "step": 7278
+    },
+    {
+      "epoch": 0.1456,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.013451131184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0924,
+      "loss/crossentropy": 1.9946333765983582,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20145532488822937,
+      "step": 7280
+    },
+    {
+      "epoch": 0.14564,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.013834635416666666,
+      "learning_rate": 0.0001,
+      "loss": 4.6519,
+      "loss/crossentropy": 2.0958545207977295,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22669509798288345,
+      "step": 7282
+    },
+    {
+      "epoch": 0.14568,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.016109212239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.4224,
+      "loss/crossentropy": 2.0515894889831543,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2520884945988655,
+      "step": 7284
+    },
+    {
+      "epoch": 0.14572,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.016063435872395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1572,
+      "loss/crossentropy": 2.034587264060974,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21523287147283554,
+      "step": 7286
+    },
+    {
+      "epoch": 0.14576,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.016080729166666665,
+      "learning_rate": 0.0001,
+      "loss": 3.8649,
+      "loss/crossentropy": 1.6578314900398254,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19603776931762695,
+      "step": 7288
+    },
+    {
+      "epoch": 0.1458,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.014167277018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2443,
+      "loss/crossentropy": 2.0019100308418274,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22880420833826065,
+      "step": 7290
+    },
+    {
+      "epoch": 0.14584,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.018723297119140624,
+      "learning_rate": 0.0001,
+      "loss": 3.981,
+      "loss/crossentropy": 2.068517565727234,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19675085693597794,
+      "step": 7292
+    },
+    {
+      "epoch": 0.14588,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.013734690348307292,
+      "learning_rate": 0.0001,
+      "loss": 4.4037,
+      "loss/crossentropy": 2.000797212123871,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20464950054883957,
+      "step": 7294
+    },
+    {
+      "epoch": 0.14592,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.013734690348307292,
+      "learning_rate": 0.0001,
+      "loss": 4.4132,
+      "loss/crossentropy": 2.1914668679237366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2406519278883934,
+      "step": 7296
+    },
+    {
+      "epoch": 0.14596,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.01685358683268229,
+      "learning_rate": 0.0001,
+      "loss": 4.4737,
+      "loss/crossentropy": 2.123211979866028,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.250032439827919,
+      "step": 7298
+    },
+    {
+      "epoch": 0.146,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.028507232666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.718,
+      "loss/crossentropy": 2.0686148405075073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22581970691680908,
+      "step": 7300
+    },
+    {
+      "epoch": 0.14604,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.029288482666015626,
+      "learning_rate": 0.0001,
+      "loss": 4.2295,
+      "loss/crossentropy": 2.141907751560211,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21503636240959167,
+      "step": 7302
+    },
+    {
+      "epoch": 0.14608,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.028436024983723957,
+      "learning_rate": 0.0001,
+      "loss": 4.1493,
+      "loss/crossentropy": 1.6741206645965576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20490698516368866,
+      "step": 7304
+    },
+    {
+      "epoch": 0.14612,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.027854156494140626,
+      "learning_rate": 0.0001,
+      "loss": 4.1259,
+      "loss/crossentropy": 1.8094561696052551,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21328043192625046,
+      "step": 7306
+    },
+    {
+      "epoch": 0.14616,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.020970662434895832,
+      "learning_rate": 0.0001,
+      "loss": 4.3715,
+      "loss/crossentropy": 2.204083800315857,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24204879999160767,
+      "step": 7308
+    },
+    {
+      "epoch": 0.1462,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.021126302083333333,
+      "learning_rate": 0.0001,
+      "loss": 4.2614,
+      "loss/crossentropy": 2.0166819095611572,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2219880372285843,
+      "step": 7310
+    },
+    {
+      "epoch": 0.14624,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.022684733072916668,
+      "learning_rate": 0.0001,
+      "loss": 4.3768,
+      "loss/crossentropy": 2.4667757749557495,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2880199924111366,
+      "step": 7312
+    },
+    {
+      "epoch": 0.14628,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.019287109375,
+      "learning_rate": 0.0001,
+      "loss": 4.2005,
+      "loss/crossentropy": 2.1497310400009155,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21792854368686676,
+      "step": 7314
+    },
+    {
+      "epoch": 0.14632,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0062896728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3911,
+      "loss/crossentropy": 2.179584264755249,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23200812935829163,
+      "step": 7316
+    },
+    {
+      "epoch": 0.14636,
+      "grad_norm": 6.0625,
+      "grad_norm_var": 0.9582590738932292,
+      "learning_rate": 0.0001,
+      "loss": 4.1545,
+      "loss/crossentropy": 1.4067250490188599,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.204126738011837,
+      "step": 7318
+    },
+    {
+      "epoch": 0.1464,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.9473592122395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2439,
+      "loss/crossentropy": 1.9879435896873474,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.247541606426239,
+      "step": 7320
+    },
+    {
+      "epoch": 0.14644,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.9377919514973958,
+      "learning_rate": 0.0001,
+      "loss": 4.1795,
+      "loss/crossentropy": 1.62649005651474,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2054424211382866,
+      "step": 7322
+    },
+    {
+      "epoch": 0.14648,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.9344228108723959,
+      "learning_rate": 0.0001,
+      "loss": 4.1479,
+      "loss/crossentropy": 1.9012435674667358,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2240126132965088,
+      "step": 7324
+    },
+    {
+      "epoch": 0.14652,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.948193359375,
+      "learning_rate": 0.0001,
+      "loss": 3.6841,
+      "loss/crossentropy": 1.8239200115203857,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19691675901412964,
+      "step": 7326
+    },
+    {
+      "epoch": 0.14656,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.94068603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2223,
+      "loss/crossentropy": 1.9689037799835205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21301231533288956,
+      "step": 7328
+    },
+    {
+      "epoch": 0.1466,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.9363433837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3733,
+      "loss/crossentropy": 2.1064560413360596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23990632593631744,
+      "step": 7330
+    },
+    {
+      "epoch": 0.14664,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.9397125244140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2765,
+      "loss/crossentropy": 2.0871987342834473,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23108436167240143,
+      "step": 7332
+    },
+    {
+      "epoch": 0.14668,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0209869384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4517,
+      "loss/crossentropy": 2.3061007857322693,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24406791478395462,
+      "step": 7334
+    },
+    {
+      "epoch": 0.14672,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.016943359375,
+      "learning_rate": 0.0001,
+      "loss": 4.4278,
+      "loss/crossentropy": 1.9471244812011719,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20859277993440628,
+      "step": 7336
+    },
+    {
+      "epoch": 0.14676,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.0131744384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.188,
+      "loss/crossentropy": 2.0628740191459656,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25501881539821625,
+      "step": 7338
+    },
+    {
+      "epoch": 0.1468,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.014481353759765624,
+      "learning_rate": 0.0001,
+      "loss": 4.4329,
+      "loss/crossentropy": 1.8065250515937805,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20222238451242447,
+      "step": 7340
+    },
+    {
+      "epoch": 0.14684,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012910715738932292,
+      "learning_rate": 0.0001,
+      "loss": 4.5431,
+      "loss/crossentropy": 2.134244918823242,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22319403290748596,
+      "step": 7342
+    },
+    {
+      "epoch": 0.14688,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012359364827473959,
+      "learning_rate": 0.0001,
+      "loss": 4.3098,
+      "loss/crossentropy": 2.5807924270629883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26865454018116,
+      "step": 7344
+    },
+    {
+      "epoch": 0.14692,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.015421295166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.684,
+      "loss/crossentropy": 2.4128278493881226,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25771988928318024,
+      "step": 7346
+    },
+    {
+      "epoch": 0.14696,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.017561594645182293,
+      "learning_rate": 0.0001,
+      "loss": 4.0915,
+      "loss/crossentropy": 1.7323983907699585,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2146565169095993,
+      "step": 7348
+    },
+    {
+      "epoch": 0.147,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.015553538004557292,
+      "learning_rate": 0.0001,
+      "loss": 4.5511,
+      "loss/crossentropy": 2.036192536354065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21692586690187454,
+      "step": 7350
+    },
+    {
+      "epoch": 0.14704,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.012320709228515626,
+      "learning_rate": 0.0001,
+      "loss": 4.4173,
+      "loss/crossentropy": 1.9586528539657593,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21543999761343002,
+      "step": 7352
+    },
+    {
+      "epoch": 0.14708,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.011805979410807292,
+      "learning_rate": 0.0001,
+      "loss": 4.4506,
+      "loss/crossentropy": 2.3444113731384277,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24080512672662735,
+      "step": 7354
+    },
+    {
+      "epoch": 0.14712,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0097076416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.4865,
+      "loss/crossentropy": 2.3060439825057983,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24567800760269165,
+      "step": 7356
+    },
+    {
+      "epoch": 0.14716,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.01279296875,
+      "learning_rate": 0.0001,
+      "loss": 4.5975,
+      "loss/crossentropy": 2.2267106771469116,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2345375493168831,
+      "step": 7358
+    },
+    {
+      "epoch": 0.1472,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.014481608072916667,
+      "learning_rate": 0.0001,
+      "loss": 4.7052,
+      "loss/crossentropy": 2.2470518350601196,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2417411357164383,
+      "step": 7360
+    },
+    {
+      "epoch": 0.14724,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.012953440348307291,
+      "learning_rate": 0.0001,
+      "loss": 4.3299,
+      "loss/crossentropy": 2.200543165206909,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24434109032154083,
+      "step": 7362
+    },
+    {
+      "epoch": 0.14728,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.010227203369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3282,
+      "loss/crossentropy": 1.995844304561615,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22591491788625717,
+      "step": 7364
+    },
+    {
+      "epoch": 0.14732,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.011580149332682291,
+      "learning_rate": 0.0001,
+      "loss": 4.3354,
+      "loss/crossentropy": 1.9180658459663391,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22776676714420319,
+      "step": 7366
+    },
+    {
+      "epoch": 0.14736,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.011840565999348959,
+      "learning_rate": 0.0001,
+      "loss": 4.3662,
+      "loss/crossentropy": 2.473931312561035,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26284685730934143,
+      "step": 7368
+    },
+    {
+      "epoch": 0.1474,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011744944254557292,
+      "learning_rate": 0.0001,
+      "loss": 4.3167,
+      "loss/crossentropy": 2.0392738580703735,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22323766350746155,
+      "step": 7370
+    },
+    {
+      "epoch": 0.14744,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.012094879150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3032,
+      "loss/crossentropy": 1.9847410917282104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22324485331773758,
+      "step": 7372
+    },
+    {
+      "epoch": 0.14748,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.008459218343098958,
+      "learning_rate": 0.0001,
+      "loss": 4.2794,
+      "loss/crossentropy": 1.930326521396637,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24155349284410477,
+      "step": 7374
+    },
+    {
+      "epoch": 0.14752,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0061724344889322914,
+      "learning_rate": 0.0001,
+      "loss": 4.0648,
+      "loss/crossentropy": 1.825449824333191,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2087552770972252,
+      "step": 7376
+    },
+    {
+      "epoch": 0.14756,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.004325358072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.2405,
+      "loss/crossentropy": 2.156645655632019,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2519562169909477,
+      "step": 7378
+    },
+    {
+      "epoch": 0.1476,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.008333333333333333,
+      "learning_rate": 0.0001,
+      "loss": 4.5316,
+      "loss/crossentropy": 2.255813479423523,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23737946152687073,
+      "step": 7380
+    },
+    {
+      "epoch": 0.14764,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.00797119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.5795,
+      "loss/crossentropy": 2.47933566570282,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23890959471464157,
+      "step": 7382
+    },
+    {
+      "epoch": 0.14768,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.007710774739583333,
+      "learning_rate": 0.0001,
+      "loss": 4.5328,
+      "loss/crossentropy": 2.139566659927368,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22334300726652145,
+      "step": 7384
+    },
+    {
+      "epoch": 0.14772,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.007673136393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1581,
+      "loss/crossentropy": 1.6182149052619934,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19345169514417648,
+      "step": 7386
+    },
+    {
+      "epoch": 0.14776,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.010758463541666667,
+      "learning_rate": 0.0001,
+      "loss": 4.4571,
+      "loss/crossentropy": 1.9984004497528076,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24823245406150818,
+      "step": 7388
+    },
+    {
+      "epoch": 0.1478,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.010445149739583333,
+      "learning_rate": 0.0001,
+      "loss": 4.3218,
+      "loss/crossentropy": 2.5892586708068848,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2489030361175537,
+      "step": 7390
+    },
+    {
+      "epoch": 0.14784,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.009598795572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.6077,
+      "loss/crossentropy": 2.3723479509353638,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23511765897274017,
+      "step": 7392
+    },
+    {
+      "epoch": 0.14788,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.0145660400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3902,
+      "loss/crossentropy": 2.001940071582794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22256013005971909,
+      "step": 7394
+    },
+    {
+      "epoch": 0.14792,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.013850911458333334,
+      "learning_rate": 0.0001,
+      "loss": 4.4422,
+      "loss/crossentropy": 2.2255555391311646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23905682563781738,
+      "step": 7396
+    },
+    {
+      "epoch": 0.14796,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.013277180989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.6217,
+      "loss/crossentropy": 2.5670583248138428,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25223178416490555,
+      "step": 7398
+    },
+    {
+      "epoch": 0.148,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 2.3002278645833334,
+      "learning_rate": 0.0001,
+      "loss": 4.6365,
+      "loss/crossentropy": 2.16109037399292,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2601509317755699,
+      "step": 7400
+    },
+    {
+      "epoch": 0.14804,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 2.3012847900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.4169,
+      "loss/crossentropy": 2.1484315395355225,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21474219858646393,
+      "step": 7402
+    },
+    {
+      "epoch": 0.14808,
+      "grad_norm": 4.3125,
+      "grad_norm_var": 2.48385009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.7452,
+      "loss/crossentropy": 2.183099091053009,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25882233679294586,
+      "step": 7404
+    },
+    {
+      "epoch": 0.14812,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 2.491097005208333,
+      "learning_rate": 0.0001,
+      "loss": 3.9739,
+      "loss/crossentropy": 1.795831561088562,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20870305597782135,
+      "step": 7406
+    },
+    {
+      "epoch": 0.14816,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 2.5058553059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3433,
+      "loss/crossentropy": 2.202280640602112,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2150556966662407,
+      "step": 7408
+    },
+    {
+      "epoch": 0.1482,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 2.5274251302083335,
+      "learning_rate": 0.0001,
+      "loss": 4.3063,
+      "loss/crossentropy": 2.2616937160491943,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24934212118387222,
+      "step": 7410
+    },
+    {
+      "epoch": 0.14824,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 2.51627197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3303,
+      "loss/crossentropy": 1.909091055393219,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23191271722316742,
+      "step": 7412
+    },
+    {
+      "epoch": 0.14828,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 2.5269765218098956,
+      "learning_rate": 0.0001,
+      "loss": 4.4224,
+      "loss/crossentropy": 2.1383588314056396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22893162816762924,
+      "step": 7414
+    },
+    {
+      "epoch": 0.14832,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.2983062744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3664,
+      "loss/crossentropy": 2.0628907680511475,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23075110465288162,
+      "step": 7416
+    },
+    {
+      "epoch": 0.14836,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.29704488118489586,
+      "learning_rate": 0.0001,
+      "loss": 4.2109,
+      "loss/crossentropy": 1.8996745347976685,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21171879768371582,
+      "step": 7418
+    },
+    {
+      "epoch": 0.1484,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0032867431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4241,
+      "loss/crossentropy": 2.2367645502090454,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2563689202070236,
+      "step": 7420
+    },
+    {
+      "epoch": 0.14844,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.003934733072916667,
+      "learning_rate": 0.0001,
+      "loss": 4.2326,
+      "loss/crossentropy": 2.274489164352417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2462785318493843,
+      "step": 7422
+    },
+    {
+      "epoch": 0.14848,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.00562744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.5288,
+      "loss/crossentropy": 2.227620482444763,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24729500710964203,
+      "step": 7424
+    },
+    {
+      "epoch": 0.14852,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.006864420572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.0913,
+      "loss/crossentropy": 1.9911785125732422,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.216292105615139,
+      "step": 7426
+    },
+    {
+      "epoch": 0.14856,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.008199055989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.4071,
+      "loss/crossentropy": 1.8715736865997314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2289058193564415,
+      "step": 7428
+    },
+    {
+      "epoch": 0.1486,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.007991536458333334,
+      "learning_rate": 0.0001,
+      "loss": 4.4671,
+      "loss/crossentropy": 2.0192378759384155,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2327374964952469,
+      "step": 7430
+    },
+    {
+      "epoch": 0.14864,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.008226521809895833,
+      "learning_rate": 0.0001,
+      "loss": 4.213,
+      "loss/crossentropy": 1.9407023191452026,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22586089372634888,
+      "step": 7432
+    },
+    {
+      "epoch": 0.14868,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.01011962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.7816,
+      "loss/crossentropy": 2.6349592208862305,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2605717331171036,
+      "step": 7434
+    },
+    {
+      "epoch": 0.14872,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.009989420572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.5511,
+      "loss/crossentropy": 2.3015077114105225,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2292959839105606,
+      "step": 7436
+    },
+    {
+      "epoch": 0.14876,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.013231404622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1663,
+      "loss/crossentropy": 2.081148624420166,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21435046195983887,
+      "step": 7438
+    },
+    {
+      "epoch": 0.1488,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011735026041666667,
+      "learning_rate": 0.0001,
+      "loss": 4.4342,
+      "loss/crossentropy": 2.235422372817993,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23794984817504883,
+      "step": 7440
+    },
+    {
+      "epoch": 0.14884,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.011295572916666666,
+      "learning_rate": 0.0001,
+      "loss": 4.4595,
+      "loss/crossentropy": 2.1423263549804688,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2200494110584259,
+      "step": 7442
+    },
+    {
+      "epoch": 0.14888,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.018342081705729166,
+      "learning_rate": 0.0001,
+      "loss": 4.514,
+      "loss/crossentropy": 2.3156672716140747,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25282321125268936,
+      "step": 7444
+    },
+    {
+      "epoch": 0.14892,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.020051066080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.3107,
+      "loss/crossentropy": 1.7777396440505981,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2104710191488266,
+      "step": 7446
+    },
+    {
+      "epoch": 0.14896,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.023341623942057292,
+      "learning_rate": 0.0001,
+      "loss": 3.874,
+      "loss/crossentropy": 1.8015541434288025,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1976182758808136,
+      "step": 7448
+    },
+    {
+      "epoch": 0.149,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.04948298136393229,
+      "learning_rate": 0.0001,
+      "loss": 4.5647,
+      "loss/crossentropy": 2.36995792388916,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23767317831516266,
+      "step": 7450
+    },
+    {
+      "epoch": 0.14904,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.050142161051432294,
+      "learning_rate": 0.0001,
+      "loss": 4.7568,
+      "loss/crossentropy": 2.2867971062660217,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2666979879140854,
+      "step": 7452
+    },
+    {
+      "epoch": 0.14908,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.043794504801432294,
+      "learning_rate": 0.0001,
+      "loss": 4.773,
+      "loss/crossentropy": 2.247913956642151,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2122594192624092,
+      "step": 7454
+    },
+    {
+      "epoch": 0.14912,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.04197362263997396,
+      "learning_rate": 0.0001,
+      "loss": 4.273,
+      "loss/crossentropy": 2.0020886063575745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22448039799928665,
+      "step": 7456
+    },
+    {
+      "epoch": 0.14916,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.043702952067057294,
+      "learning_rate": 0.0001,
+      "loss": 4.3139,
+      "loss/crossentropy": 2.2290679216384888,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21489766240119934,
+      "step": 7458
+    },
+    {
+      "epoch": 0.1492,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.049344635009765624,
+      "learning_rate": 0.0001,
+      "loss": 4.5919,
+      "loss/crossentropy": 2.0447877049446106,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22640856355428696,
+      "step": 7460
+    },
+    {
+      "epoch": 0.14924,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.045904286702473956,
+      "learning_rate": 0.0001,
+      "loss": 4.4614,
+      "loss/crossentropy": 2.395468831062317,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2395479902625084,
+      "step": 7462
+    },
+    {
+      "epoch": 0.14928,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.04019775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1552,
+      "loss/crossentropy": 2.117182433605194,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22302603721618652,
+      "step": 7464
+    },
+    {
+      "epoch": 0.14932,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.019498697916666665,
+      "learning_rate": 0.0001,
+      "loss": 4.4046,
+      "loss/crossentropy": 2.206045985221863,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2548774778842926,
+      "step": 7466
+    },
+    {
+      "epoch": 0.14936,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0194244384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1531,
+      "loss/crossentropy": 1.9933450818061829,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22455794364213943,
+      "step": 7468
+    },
+    {
+      "epoch": 0.1494,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.017731730143229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2744,
+      "loss/crossentropy": 2.216577649116516,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24432705342769623,
+      "step": 7470
+    },
+    {
+      "epoch": 0.14944,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.018163045247395832,
+      "learning_rate": 0.0001,
+      "loss": 4.6072,
+      "loss/crossentropy": 2.4282405376434326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2599884122610092,
+      "step": 7472
+    },
+    {
+      "epoch": 0.14948,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.018089803059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2833,
+      "loss/crossentropy": 2.2677053213119507,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24509359151124954,
+      "step": 7474
+    },
+    {
+      "epoch": 0.14952,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.007877604166666666,
+      "learning_rate": 0.0001,
+      "loss": 4.1271,
+      "loss/crossentropy": 1.9610475897789001,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19867657870054245,
+      "step": 7476
+    },
+    {
+      "epoch": 0.14956,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.012565104166666667,
+      "learning_rate": 0.0001,
+      "loss": 4.6286,
+      "loss/crossentropy": 2.1379209756851196,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24773475527763367,
+      "step": 7478
+    },
+    {
+      "epoch": 0.1496,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0125152587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3593,
+      "loss/crossentropy": 2.295411467552185,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2417762354016304,
+      "step": 7480
+    },
+    {
+      "epoch": 0.14964,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.013374837239583333,
+      "learning_rate": 0.0001,
+      "loss": 4.4839,
+      "loss/crossentropy": 2.0983279943466187,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22220823168754578,
+      "step": 7482
+    },
+    {
+      "epoch": 0.14968,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.015086873372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.3977,
+      "loss/crossentropy": 2.130508065223694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23387905955314636,
+      "step": 7484
+    },
+    {
+      "epoch": 0.14972,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.015869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.545,
+      "loss/crossentropy": 2.0727924704551697,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23198049515485764,
+      "step": 7486
+    },
+    {
+      "epoch": 0.14976,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.015523274739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.3722,
+      "loss/crossentropy": 2.1407171487808228,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23349495232105255,
+      "step": 7488
+    },
+    {
+      "epoch": 0.1498,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.016063435872395834,
+      "learning_rate": 0.0001,
+      "loss": 4.3483,
+      "loss/crossentropy": 2.123879909515381,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2230711579322815,
+      "step": 7490
+    },
+    {
+      "epoch": 0.14984,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.02194188435872396,
+      "learning_rate": 0.0001,
+      "loss": 3.8687,
+      "loss/crossentropy": 2.1362847685813904,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22217128425836563,
+      "step": 7492
+    },
+    {
+      "epoch": 0.14988,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.01625544230143229,
+      "learning_rate": 0.0001,
+      "loss": 4.4753,
+      "loss/crossentropy": 2.2171897292137146,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.237472265958786,
+      "step": 7494
+    },
+    {
+      "epoch": 0.14992,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.016841379801432292,
+      "learning_rate": 0.0001,
+      "loss": 4.2439,
+      "loss/crossentropy": 1.9622138142585754,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19700734317302704,
+      "step": 7496
+    },
+    {
+      "epoch": 0.14996,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.01666234334309896,
+      "learning_rate": 0.0001,
+      "loss": 4.1296,
+      "loss/crossentropy": 2.300232410430908,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26260019838809967,
+      "step": 7498
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.013765207926432292,
+      "learning_rate": 0.0001,
+      "loss": 4.4466,
+      "loss/crossentropy": 1.9829946756362915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21770135313272476,
+      "step": 7500
+    },
+    {
+      "epoch": 0.15004,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.010935211181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.017,
+      "loss/crossentropy": 1.8421878218650818,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20965785533189774,
+      "step": 7502
+    },
+    {
+      "epoch": 0.15008,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.010267893473307291,
+      "learning_rate": 0.0001,
+      "loss": 4.3111,
+      "loss/crossentropy": 2.0559566020965576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21752064675092697,
+      "step": 7504
+    },
+    {
+      "epoch": 0.15012,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.009834543863932291,
+      "learning_rate": 0.0001,
+      "loss": 4.5526,
+      "loss/crossentropy": 1.862777590751648,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2026660442352295,
+      "step": 7506
+    },
+    {
+      "epoch": 0.15016,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0064198811848958336,
+      "learning_rate": 0.0001,
+      "loss": 4.0238,
+      "loss/crossentropy": 2.340041399002075,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23489895462989807,
+      "step": 7508
+    },
+    {
+      "epoch": 0.1502,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.006224568684895833,
+      "learning_rate": 0.0001,
+      "loss": 4.5434,
+      "loss/crossentropy": 2.3596150875091553,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23775358498096466,
+      "step": 7510
+    },
+    {
+      "epoch": 0.15024,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.005980428059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4637,
+      "loss/crossentropy": 1.883503019809723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22191710770130157,
+      "step": 7512
+    },
+    {
+      "epoch": 0.15028,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.006723785400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1485,
+      "loss/crossentropy": 2.12862491607666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2177818939089775,
+      "step": 7514
+    },
+    {
+      "epoch": 0.15032,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.005936431884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.167,
+      "loss/crossentropy": 1.9326343536376953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20991001278162003,
+      "step": 7516
+    },
+    {
+      "epoch": 0.15036,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.008070627848307291,
+      "learning_rate": 0.0001,
+      "loss": 4.4552,
+      "loss/crossentropy": 2.1532761454582214,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2308126464486122,
+      "step": 7518
+    },
+    {
+      "epoch": 0.1504,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.007867177327473959,
+      "learning_rate": 0.0001,
+      "loss": 4.42,
+      "loss/crossentropy": 1.876515507698059,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1940685734152794,
+      "step": 7520
+    },
+    {
+      "epoch": 0.15044,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.008937327067057292,
+      "learning_rate": 0.0001,
+      "loss": 4.3515,
+      "loss/crossentropy": 2.3677611351013184,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24787116795778275,
+      "step": 7522
+    },
+    {
+      "epoch": 0.15048,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.010892486572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0489,
+      "loss/crossentropy": 1.8500076532363892,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19271107017993927,
+      "step": 7524
+    },
+    {
+      "epoch": 0.15052,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.010432688395182292,
+      "learning_rate": 0.0001,
+      "loss": 4.3298,
+      "loss/crossentropy": 1.8013980984687805,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2181655913591385,
+      "step": 7526
+    },
+    {
+      "epoch": 0.15056,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.010361480712890624,
+      "learning_rate": 0.0001,
+      "loss": 4.4958,
+      "loss/crossentropy": 2.6469568014144897,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2441466525197029,
+      "step": 7528
+    },
+    {
+      "epoch": 0.1506,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.009227498372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1317,
+      "loss/crossentropy": 1.7992960214614868,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2209397256374359,
+      "step": 7530
+    },
+    {
+      "epoch": 0.15064,
+      "grad_norm": 2.625,
+      "grad_norm_var": 0.023258463541666666,
+      "learning_rate": 0.0001,
+      "loss": 4.9501,
+      "loss/crossentropy": 2.1914783120155334,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23764144629240036,
+      "step": 7532
+    },
+    {
+      "epoch": 0.15068,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.02906494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.4833,
+      "loss/crossentropy": 2.0035120844841003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21483591943979263,
+      "step": 7534
+    },
+    {
+      "epoch": 0.15072,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.03235677083333333,
+      "learning_rate": 0.0001,
+      "loss": 4.2674,
+      "loss/crossentropy": 2.0218639969825745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20919294655323029,
+      "step": 7536
+    },
+    {
+      "epoch": 0.15076,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.03219401041666667,
+      "learning_rate": 0.0001,
+      "loss": 4.1091,
+      "loss/crossentropy": 2.006688416004181,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21247616410255432,
+      "step": 7538
+    },
+    {
+      "epoch": 0.1508,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.03542378743489583,
+      "learning_rate": 0.0001,
+      "loss": 4.5424,
+      "loss/crossentropy": 2.1565613746643066,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30109211802482605,
+      "step": 7540
+    },
+    {
+      "epoch": 0.15084,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.03603413899739583,
+      "learning_rate": 0.0001,
+      "loss": 4.3621,
+      "loss/crossentropy": 1.9676685333251953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23605409264564514,
+      "step": 7542
+    },
+    {
+      "epoch": 0.15088,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.051493326822916664,
+      "learning_rate": 0.0001,
+      "loss": 4.4178,
+      "loss/crossentropy": 2.1150137186050415,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22903620451688766,
+      "step": 7544
+    },
+    {
+      "epoch": 0.15092,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.051253255208333334,
+      "learning_rate": 0.0001,
+      "loss": 4.3186,
+      "loss/crossentropy": 2.1894554495811462,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24362927675247192,
+      "step": 7546
+    },
+    {
+      "epoch": 0.15096,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0410308837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4529,
+      "loss/crossentropy": 1.9624019861221313,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22628726810216904,
+      "step": 7548
+    },
+    {
+      "epoch": 0.151,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.04338277180989583,
+      "learning_rate": 0.0001,
+      "loss": 4.2795,
+      "loss/crossentropy": 2.1547625064849854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2295984849333763,
+      "step": 7550
+    },
+    {
+      "epoch": 0.15104,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.03857014973958333,
+      "learning_rate": 0.0001,
+      "loss": 4.4638,
+      "loss/crossentropy": 2.229305863380432,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2427791878581047,
+      "step": 7552
+    },
+    {
+      "epoch": 0.15108,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.037093098958333334,
+      "learning_rate": 0.0001,
+      "loss": 4.7319,
+      "loss/crossentropy": 2.4998362064361572,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23718395829200745,
+      "step": 7554
+    },
+    {
+      "epoch": 0.15112,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.033722941080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2118,
+      "loss/crossentropy": 2.319428563117981,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26024487614631653,
+      "step": 7556
+    },
+    {
+      "epoch": 0.15116,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0357818603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2451,
+      "loss/crossentropy": 1.861966609954834,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2121218591928482,
+      "step": 7558
+    },
+    {
+      "epoch": 0.1512,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.015751139322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.5963,
+      "loss/crossentropy": 2.1688510179519653,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22341617196798325,
+      "step": 7560
+    },
+    {
+      "epoch": 0.15124,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.013932291666666667,
+      "learning_rate": 0.0001,
+      "loss": 4.4808,
+      "loss/crossentropy": 2.2833406925201416,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22648481279611588,
+      "step": 7562
+    },
+    {
+      "epoch": 0.15128,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0138671875,
+      "learning_rate": 0.0001,
+      "loss": 4.4126,
+      "loss/crossentropy": 2.224393129348755,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23414459079504013,
+      "step": 7564
+    },
+    {
+      "epoch": 0.15132,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0098297119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.253,
+      "loss/crossentropy": 2.003828763961792,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2166791632771492,
+      "step": 7566
+    },
+    {
+      "epoch": 0.15136,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.01168212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0904,
+      "loss/crossentropy": 2.1813005208969116,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28873007744550705,
+      "step": 7568
+    },
+    {
+      "epoch": 0.1514,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.028696441650390626,
+      "learning_rate": 0.0001,
+      "loss": 4.3824,
+      "loss/crossentropy": 2.3818061351776123,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24916332960128784,
+      "step": 7570
+    },
+    {
+      "epoch": 0.15144,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.030460357666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1825,
+      "loss/crossentropy": 2.041518449783325,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22352956235408783,
+      "step": 7572
+    },
+    {
+      "epoch": 0.15148,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.0326904296875,
+      "learning_rate": 0.0001,
+      "loss": 3.8531,
+      "loss/crossentropy": 1.856759488582611,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20227209478616714,
+      "step": 7574
+    },
+    {
+      "epoch": 0.15152,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.03157450358072917,
+      "learning_rate": 0.0001,
+      "loss": 4.237,
+      "loss/crossentropy": 1.9612281918525696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22188737243413925,
+      "step": 7576
+    },
+    {
+      "epoch": 0.15156,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0337554931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0405,
+      "loss/crossentropy": 1.9610649943351746,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2282957062125206,
+      "step": 7578
+    },
+    {
+      "epoch": 0.1516,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.03400472005208333,
+      "learning_rate": 0.0001,
+      "loss": 4.1253,
+      "loss/crossentropy": 1.9239189624786377,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22926658391952515,
+      "step": 7580
+    },
+    {
+      "epoch": 0.15164,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.033299763997395836,
+      "learning_rate": 0.0001,
+      "loss": 4.548,
+      "loss/crossentropy": 2.425737738609314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24260805547237396,
+      "step": 7582
+    },
+    {
+      "epoch": 0.15168,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.03242085774739583,
+      "learning_rate": 0.0001,
+      "loss": 4.2894,
+      "loss/crossentropy": 1.9602521061897278,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23231150209903717,
+      "step": 7584
+    },
+    {
+      "epoch": 0.15172,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.007954661051432292,
+      "learning_rate": 0.0001,
+      "loss": 4.3131,
+      "loss/crossentropy": 2.3671375513076782,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24326395988464355,
+      "step": 7586
+    },
+    {
+      "epoch": 0.15176,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007675933837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1297,
+      "loss/crossentropy": 2.0128119587898254,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20082567632198334,
+      "step": 7588
+    },
+    {
+      "epoch": 0.1518,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0085845947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.5813,
+      "loss/crossentropy": 2.3719125986099243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.243422269821167,
+      "step": 7590
+    },
+    {
+      "epoch": 0.15184,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.010575358072916667,
+      "learning_rate": 0.0001,
+      "loss": 4.6177,
+      "loss/crossentropy": 2.088695764541626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23578013479709625,
+      "step": 7592
+    },
+    {
+      "epoch": 0.15188,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.01099853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3719,
+      "loss/crossentropy": 2.151872456073761,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2496410757303238,
+      "step": 7594
+    },
+    {
+      "epoch": 0.15192,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.017281087239583333,
+      "learning_rate": 0.0001,
+      "loss": 4.6041,
+      "loss/crossentropy": 2.020963430404663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25003430247306824,
+      "step": 7596
+    },
+    {
+      "epoch": 0.15196,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.0199615478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4439,
+      "loss/crossentropy": 1.9290395379066467,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20771078765392303,
+      "step": 7598
+    },
+    {
+      "epoch": 0.152,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.020536295572916665,
+      "learning_rate": 0.0001,
+      "loss": 4.2461,
+      "loss/crossentropy": 2.0420188307762146,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2375432327389717,
+      "step": 7600
+    },
+    {
+      "epoch": 0.15204,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.019169108072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.4291,
+      "loss/crossentropy": 2.474969744682312,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2535504847764969,
+      "step": 7602
+    },
+    {
+      "epoch": 0.15208,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.015550740559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.3487,
+      "loss/crossentropy": 2.177125334739685,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24919381737709045,
+      "step": 7604
+    },
+    {
+      "epoch": 0.15212,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.015510050455729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4719,
+      "loss/crossentropy": 2.348747491836548,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22336618602275848,
+      "step": 7606
+    },
+    {
+      "epoch": 0.15216,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.014676920572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.4502,
+      "loss/crossentropy": 1.718321442604065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19830843806266785,
+      "step": 7608
+    },
+    {
+      "epoch": 0.1522,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.015901692708333335,
+      "learning_rate": 0.0001,
+      "loss": 4.4505,
+      "loss/crossentropy": 2.2954800128936768,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23203244805335999,
+      "step": 7610
+    },
+    {
+      "epoch": 0.15224,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.013570149739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.1141,
+      "loss/crossentropy": 1.6918454766273499,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18312305957078934,
+      "step": 7612
+    },
+    {
+      "epoch": 0.15228,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.012727864583333333,
+      "learning_rate": 0.0001,
+      "loss": 4.1312,
+      "loss/crossentropy": 1.887774109840393,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2151220142841339,
+      "step": 7614
+    },
+    {
+      "epoch": 0.15232,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.01373291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1274,
+      "loss/crossentropy": 2.0903998613357544,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22486191242933273,
+      "step": 7616
+    },
+    {
+      "epoch": 0.15236,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.016310373942057293,
+      "learning_rate": 0.0001,
+      "loss": 4.0659,
+      "loss/crossentropy": 1.929358720779419,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20021560043096542,
+      "step": 7618
+    },
+    {
+      "epoch": 0.1524,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.03484064737955729,
+      "learning_rate": 0.0001,
+      "loss": 4.5726,
+      "loss/crossentropy": 1.9812661409378052,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21395482122898102,
+      "step": 7620
+    },
+    {
+      "epoch": 0.15244,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.03358942667643229,
+      "learning_rate": 0.0001,
+      "loss": 4.15,
+      "loss/crossentropy": 2.148552179336548,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22805052995681763,
+      "step": 7622
+    },
+    {
+      "epoch": 0.15248,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.03416926066080729,
+      "learning_rate": 0.0001,
+      "loss": 4.3471,
+      "loss/crossentropy": 2.012804687023163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2248745858669281,
+      "step": 7624
+    },
+    {
+      "epoch": 0.15252,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.03463312784830729,
+      "learning_rate": 0.0001,
+      "loss": 4.3948,
+      "loss/crossentropy": 2.3378156423568726,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2457970678806305,
+      "step": 7626
+    },
+    {
+      "epoch": 0.15256,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.036382802327473956,
+      "learning_rate": 0.0001,
+      "loss": 4.5414,
+      "loss/crossentropy": 2.0815274119377136,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22083784639835358,
+      "step": 7628
+    },
+    {
+      "epoch": 0.1526,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.039589182535807295,
+      "learning_rate": 0.0001,
+      "loss": 4.2609,
+      "loss/crossentropy": 2.172307014465332,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25988608598709106,
+      "step": 7630
+    },
+    {
+      "epoch": 0.15264,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.03921305338541667,
+      "learning_rate": 0.0001,
+      "loss": 4.3711,
+      "loss/crossentropy": 1.973683476448059,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20379862189292908,
+      "step": 7632
+    },
+    {
+      "epoch": 0.15268,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.039406077067057295,
+      "learning_rate": 0.0001,
+      "loss": 4.3394,
+      "loss/crossentropy": 2.175020456314087,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22480874508619308,
+      "step": 7634
+    },
+    {
+      "epoch": 0.15272,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.022304026285807292,
+      "learning_rate": 0.0001,
+      "loss": 4.2176,
+      "loss/crossentropy": 2.063227415084839,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2324352264404297,
+      "step": 7636
+    },
+    {
+      "epoch": 0.15276,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.019769032796223957,
+      "learning_rate": 0.0001,
+      "loss": 4.4654,
+      "loss/crossentropy": 1.9297555088996887,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2171614021062851,
+      "step": 7638
+    },
+    {
+      "epoch": 0.1528,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.019421132405598958,
+      "learning_rate": 0.0001,
+      "loss": 4.4698,
+      "loss/crossentropy": 1.9808542132377625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21765583008527756,
+      "step": 7640
+    },
+    {
+      "epoch": 0.15284,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.018790435791015626,
+      "learning_rate": 0.0001,
+      "loss": 4.4145,
+      "loss/crossentropy": 2.009860336780548,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20963389426469803,
+      "step": 7642
+    },
+    {
+      "epoch": 0.15288,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.017319488525390624,
+      "learning_rate": 0.0001,
+      "loss": 4.2649,
+      "loss/crossentropy": 1.912703514099121,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23796956986188889,
+      "step": 7644
+    },
+    {
+      "epoch": 0.15292,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.012341054280598958,
+      "learning_rate": 0.0001,
+      "loss": 4.4965,
+      "loss/crossentropy": 2.2945642471313477,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2621624022722244,
+      "step": 7646
+    },
+    {
+      "epoch": 0.15296,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011067708333333334,
+      "learning_rate": 0.0001,
+      "loss": 4.2573,
+      "loss/crossentropy": 2.185902237892151,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23890018463134766,
+      "step": 7648
+    },
+    {
+      "epoch": 0.153,
+      "grad_norm": 2.546875,
+      "grad_norm_var": 0.018648274739583335,
+      "learning_rate": 0.0001,
+      "loss": 4.5815,
+      "loss/crossentropy": 1.8625503778457642,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2206135168671608,
+      "step": 7650
+    },
+    {
+      "epoch": 0.15304,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.020684560139973957,
+      "learning_rate": 0.0001,
+      "loss": 3.9142,
+      "loss/crossentropy": 1.8815893530845642,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21166741847991943,
+      "step": 7652
+    },
+    {
+      "epoch": 0.15308,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.020979563395182293,
+      "learning_rate": 0.0001,
+      "loss": 4.2277,
+      "loss/crossentropy": 2.156697630882263,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22994951903820038,
+      "step": 7654
+    },
+    {
+      "epoch": 0.15312,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.02005182902018229,
+      "learning_rate": 0.0001,
+      "loss": 4.366,
+      "loss/crossentropy": 1.6800576448440552,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21905134618282318,
+      "step": 7656
+    },
+    {
+      "epoch": 0.15316,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.02061945597330729,
+      "learning_rate": 0.0001,
+      "loss": 4.1446,
+      "loss/crossentropy": 1.9215145707130432,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2034405767917633,
+      "step": 7658
+    },
+    {
+      "epoch": 0.1532,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.02380956013997396,
+      "learning_rate": 0.0001,
+      "loss": 4.6475,
+      "loss/crossentropy": 2.153718650341034,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25605182349681854,
+      "step": 7660
+    },
+    {
+      "epoch": 0.15324,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.02466608683268229,
+      "learning_rate": 0.0001,
+      "loss": 4.2676,
+      "loss/crossentropy": 1.8782889246940613,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20574549585580826,
+      "step": 7662
+    },
+    {
+      "epoch": 0.15328,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.031040191650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2171,
+      "loss/crossentropy": 2.003354489803314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24152260273694992,
+      "step": 7664
+    },
+    {
+      "epoch": 0.15332,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.022299957275390626,
+      "learning_rate": 0.0001,
+      "loss": 4.5828,
+      "loss/crossentropy": 2.217758059501648,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24700726568698883,
+      "step": 7666
+    },
+    {
+      "epoch": 0.15336,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.017757161458333334,
+      "learning_rate": 0.0001,
+      "loss": 4.3418,
+      "loss/crossentropy": 1.934537410736084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2163502648472786,
+      "step": 7668
+    },
+    {
+      "epoch": 0.1534,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.020340983072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.3,
+      "loss/crossentropy": 2.007661819458008,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21769292652606964,
+      "step": 7670
+    },
+    {
+      "epoch": 0.15344,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.020653279622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3082,
+      "loss/crossentropy": 2.1586949825286865,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2453690618276596,
+      "step": 7672
+    },
+    {
+      "epoch": 0.15348,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0197174072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4433,
+      "loss/crossentropy": 2.2201706171035767,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22639702260494232,
+      "step": 7674
+    },
+    {
+      "epoch": 0.15352,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.015104166666666667,
+      "learning_rate": 0.0001,
+      "loss": 4.2337,
+      "loss/crossentropy": 2.059146285057068,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22411910444498062,
+      "step": 7676
+    },
+    {
+      "epoch": 0.15356,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.015397135416666667,
+      "learning_rate": 0.0001,
+      "loss": 4.2318,
+      "loss/crossentropy": 1.9768954515457153,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21798508614301682,
+      "step": 7678
+    },
+    {
+      "epoch": 0.1536,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.003413899739583333,
+      "learning_rate": 0.0001,
+      "loss": 4.4823,
+      "loss/crossentropy": 1.8555094003677368,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19453337788581848,
+      "step": 7680
+    },
+    {
+      "epoch": 0.15364,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0036783854166666666,
+      "learning_rate": 0.0001,
+      "loss": 4.0379,
+      "loss/crossentropy": 1.5948917865753174,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17746463418006897,
+      "step": 7682
+    },
+    {
+      "epoch": 0.15368,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0034464518229166668,
+      "learning_rate": 0.0001,
+      "loss": 4.2901,
+      "loss/crossentropy": 1.8917757868766785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2103574424982071,
+      "step": 7684
+    },
+    {
+      "epoch": 0.15372,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0030670166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3326,
+      "loss/crossentropy": 2.090232729911804,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2199823334813118,
+      "step": 7686
+    },
+    {
+      "epoch": 0.15376,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.004979451497395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3695,
+      "loss/crossentropy": 1.8155178427696228,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21044857800006866,
+      "step": 7688
+    },
+    {
+      "epoch": 0.1538,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0063629150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.5003,
+      "loss/crossentropy": 2.31532621383667,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2437051385641098,
+      "step": 7690
+    },
+    {
+      "epoch": 0.15384,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0065419514973958336,
+      "learning_rate": 0.0001,
+      "loss": 4.3815,
+      "loss/crossentropy": 1.9688079357147217,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22686263918876648,
+      "step": 7692
+    },
+    {
+      "epoch": 0.15388,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.010001373291015626,
+      "learning_rate": 0.0001,
+      "loss": 3.9937,
+      "loss/crossentropy": 1.9029017686843872,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20635761320590973,
+      "step": 7694
+    },
+    {
+      "epoch": 0.15392,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.010587565104166667,
+      "learning_rate": 0.0001,
+      "loss": 3.91,
+      "loss/crossentropy": 1.9817028641700745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2080235257744789,
+      "step": 7696
+    },
+    {
+      "epoch": 0.15396,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.017463175455729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3301,
+      "loss/crossentropy": 2.3392014503479004,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26226382702589035,
+      "step": 7698
+    },
+    {
+      "epoch": 0.154,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.021320597330729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0381,
+      "loss/crossentropy": 1.7265403866767883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21790936589241028,
+      "step": 7700
+    },
+    {
+      "epoch": 0.15404,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.02276585896809896,
+      "learning_rate": 0.0001,
+      "loss": 4.1725,
+      "loss/crossentropy": 2.024384081363678,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21987473219633102,
+      "step": 7702
+    },
+    {
+      "epoch": 0.15408,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.03117650349934896,
+      "learning_rate": 0.0001,
+      "loss": 4.6702,
+      "loss/crossentropy": 2.1840893030166626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24231631308794022,
+      "step": 7704
+    },
+    {
+      "epoch": 0.15412,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.03001683553059896,
+      "learning_rate": 0.0001,
+      "loss": 4.2197,
+      "loss/crossentropy": 2.1950928568840027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21808429062366486,
+      "step": 7706
+    },
+    {
+      "epoch": 0.15416,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.03029352823893229,
+      "learning_rate": 0.0001,
+      "loss": 4.3215,
+      "loss/crossentropy": 1.9541537165641785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22109197825193405,
+      "step": 7708
+    },
+    {
+      "epoch": 0.1542,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0257476806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1966,
+      "loss/crossentropy": 2.0232877135276794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.214762382209301,
+      "step": 7710
+    },
+    {
+      "epoch": 0.15424,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.023859659830729168,
+      "learning_rate": 0.0001,
+      "loss": 4.0012,
+      "loss/crossentropy": 2.003768503665924,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21820105612277985,
+      "step": 7712
+    },
+    {
+      "epoch": 0.15428,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.020026652018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2471,
+      "loss/crossentropy": 2.007221221923828,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2203196883201599,
+      "step": 7714
+    },
+    {
+      "epoch": 0.15432,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.015900675455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.3201,
+      "loss/crossentropy": 2.0134615898132324,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21916545927524567,
+      "step": 7716
+    },
+    {
+      "epoch": 0.15436,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.016013336181640626,
+      "learning_rate": 0.0001,
+      "loss": 4.3821,
+      "loss/crossentropy": 1.9012999534606934,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25758640468120575,
+      "step": 7718
+    },
+    {
+      "epoch": 0.1544,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.008727773030598959,
+      "learning_rate": 0.0001,
+      "loss": 4.1555,
+      "loss/crossentropy": 2.074169874191284,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22667942196130753,
+      "step": 7720
+    },
+    {
+      "epoch": 0.15444,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.009059397379557292,
+      "learning_rate": 0.0001,
+      "loss": 4.4355,
+      "loss/crossentropy": 2.070925295352936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20353248715400696,
+      "step": 7722
+    },
+    {
+      "epoch": 0.15448,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.011163075764973959,
+      "learning_rate": 0.0001,
+      "loss": 4.5676,
+      "loss/crossentropy": 2.289568066596985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23175117373466492,
+      "step": 7724
+    },
+    {
+      "epoch": 0.15452,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.012827301025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.6152,
+      "loss/crossentropy": 2.21374249458313,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23585008084774017,
+      "step": 7726
+    },
+    {
+      "epoch": 0.15456,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.010791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.5575,
+      "loss/crossentropy": 2.15897136926651,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24143048375844955,
+      "step": 7728
+    },
+    {
+      "epoch": 0.1546,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.01142578125,
+      "learning_rate": 0.0001,
+      "loss": 4.4722,
+      "loss/crossentropy": 2.134206771850586,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24344487488269806,
+      "step": 7730
+    },
+    {
+      "epoch": 0.15464,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.011180623372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.5827,
+      "loss/crossentropy": 2.3832513093948364,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24878299236297607,
+      "step": 7732
+    },
+    {
+      "epoch": 0.15468,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0102203369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0722,
+      "loss/crossentropy": 1.917544960975647,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20134451985359192,
+      "step": 7734
+    },
+    {
+      "epoch": 0.15472,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.011767578125,
+      "learning_rate": 0.0001,
+      "loss": 4.4499,
+      "loss/crossentropy": 2.1081286668777466,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23577219247817993,
+      "step": 7736
+    },
+    {
+      "epoch": 0.15476,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.0137603759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.7942,
+      "loss/crossentropy": 2.214662790298462,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23744845390319824,
+      "step": 7738
+    },
+    {
+      "epoch": 0.1548,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.013895670572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.4325,
+      "loss/crossentropy": 1.995256781578064,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21037640422582626,
+      "step": 7740
+    },
+    {
+      "epoch": 0.15484,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.024006144205729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0701,
+      "loss/crossentropy": 2.2877765893936157,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23090071976184845,
+      "step": 7742
+    },
+    {
+      "epoch": 0.15488,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.022652180989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.6167,
+      "loss/crossentropy": 2.23935329914093,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24617131054401398,
+      "step": 7744
+    },
+    {
+      "epoch": 0.15492,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.023005167643229168,
+      "learning_rate": 0.0001,
+      "loss": 4.4886,
+      "loss/crossentropy": 2.15006422996521,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2235095053911209,
+      "step": 7746
+    },
+    {
+      "epoch": 0.15496,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.024144490559895832,
+      "learning_rate": 0.0001,
+      "loss": 4.1873,
+      "loss/crossentropy": 1.9917905926704407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21606986224651337,
+      "step": 7748
+    },
+    {
+      "epoch": 0.155,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.025804646809895835,
+      "learning_rate": 0.0001,
+      "loss": 4.6347,
+      "loss/crossentropy": 2.303179979324341,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2541813999414444,
+      "step": 7750
+    },
+    {
+      "epoch": 0.15504,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.022786458333333332,
+      "learning_rate": 0.0001,
+      "loss": 4.3647,
+      "loss/crossentropy": 2.231510281562805,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2479737550020218,
+      "step": 7752
+    },
+    {
+      "epoch": 0.15508,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.01778132120768229,
+      "learning_rate": 0.0001,
+      "loss": 4.1714,
+      "loss/crossentropy": 2.0530437231063843,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20220057666301727,
+      "step": 7754
+    },
+    {
+      "epoch": 0.15512,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.015909830729166668,
+      "learning_rate": 0.0001,
+      "loss": 4.1978,
+      "loss/crossentropy": 2.0889222025871277,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22389977425336838,
+      "step": 7756
+    },
+    {
+      "epoch": 0.15516,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.0131256103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4748,
+      "loss/crossentropy": 2.3807711601257324,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2397037297487259,
+      "step": 7758
+    },
+    {
+      "epoch": 0.1552,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.01297607421875,
+      "learning_rate": 0.0001,
+      "loss": 4.3382,
+      "loss/crossentropy": 1.8144067525863647,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19052604585886002,
+      "step": 7760
+    },
+    {
+      "epoch": 0.15524,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.013240559895833334,
+      "learning_rate": 0.0001,
+      "loss": 4.5358,
+      "loss/crossentropy": 2.295349955558777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2168404459953308,
+      "step": 7762
+    },
+    {
+      "epoch": 0.15528,
+      "grad_norm": 11.8125,
+      "grad_norm_var": 5.868936920166016,
+      "learning_rate": 0.0001,
+      "loss": 4.1706,
+      "loss/crossentropy": 1.7281805276870728,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.230300635099411,
+      "step": 7764
+    },
+    {
+      "epoch": 0.15532,
+      "grad_norm": 2.375,
+      "grad_norm_var": 5.861083730061849,
+      "learning_rate": 0.0001,
+      "loss": 4.3867,
+      "loss/crossentropy": 2.2153135538101196,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24419061839580536,
+      "step": 7766
+    },
+    {
+      "epoch": 0.15536,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 5.890169270833334,
+      "learning_rate": 0.0001,
+      "loss": 4.2047,
+      "loss/crossentropy": 2.0259060859680176,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2119704708456993,
+      "step": 7768
+    },
+    {
+      "epoch": 0.1554,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 5.871726226806641,
+      "learning_rate": 0.0001,
+      "loss": 4.3405,
+      "loss/crossentropy": 2.2399297952651978,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23252833634614944,
+      "step": 7770
+    },
+    {
+      "epoch": 0.15544,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 5.86380615234375,
+      "learning_rate": 0.0001,
+      "loss": 4.3287,
+      "loss/crossentropy": 2.0974661111831665,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2085232511162758,
+      "step": 7772
+    },
+    {
+      "epoch": 0.15548,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 5.861717732747396,
+      "learning_rate": 0.0001,
+      "loss": 4.4141,
+      "loss/crossentropy": 2.0121108293533325,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24354346096515656,
+      "step": 7774
+    },
+    {
+      "epoch": 0.15552,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 5.843431599934896,
+      "learning_rate": 0.0001,
+      "loss": 4.3167,
+      "loss/crossentropy": 2.1463273763656616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22797952592372894,
+      "step": 7776
+    },
+    {
+      "epoch": 0.15556,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 5.875705718994141,
+      "learning_rate": 0.0001,
+      "loss": 3.9814,
+      "loss/crossentropy": 1.6362827122211456,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18291430547833443,
+      "step": 7778
+    },
+    {
+      "epoch": 0.1556,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.016283162434895835,
+      "learning_rate": 0.0001,
+      "loss": 4.3534,
+      "loss/crossentropy": 2.4894620180130005,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24631594866514206,
+      "step": 7780
+    },
+    {
+      "epoch": 0.15564,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0119781494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.5799,
+      "loss/crossentropy": 2.346967577934265,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20964853465557098,
+      "step": 7782
+    },
+    {
+      "epoch": 0.15568,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.010282135009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.5036,
+      "loss/crossentropy": 2.0165189504623413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22410035878419876,
+      "step": 7784
+    },
+    {
+      "epoch": 0.15572,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.009417470296223958,
+      "learning_rate": 0.0001,
+      "loss": 4.0735,
+      "loss/crossentropy": 1.7486848831176758,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20792805403470993,
+      "step": 7786
+    },
+    {
+      "epoch": 0.15576,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009905751546223958,
+      "learning_rate": 0.0001,
+      "loss": 4.1321,
+      "loss/crossentropy": 1.9615037441253662,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22578728944063187,
+      "step": 7788
+    },
+    {
+      "epoch": 0.1558,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.009069569905598958,
+      "learning_rate": 0.0001,
+      "loss": 4.326,
+      "loss/crossentropy": 1.8386783003807068,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21898606419563293,
+      "step": 7790
+    },
+    {
+      "epoch": 0.15584,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.011533355712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.6429,
+      "loss/crossentropy": 2.1383039951324463,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2322411835193634,
+      "step": 7792
+    },
+    {
+      "epoch": 0.15588,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.008918253580729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3568,
+      "loss/crossentropy": 2.510488271713257,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24672146886587143,
+      "step": 7794
+    },
+    {
+      "epoch": 0.15592,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.009309895833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.4328,
+      "loss/crossentropy": 2.03993421792984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23064473271369934,
+      "step": 7796
+    },
+    {
+      "epoch": 0.15596,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.007306925455729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4833,
+      "loss/crossentropy": 2.305809736251831,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21916814893484116,
+      "step": 7798
+    },
+    {
+      "epoch": 0.156,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.007796223958333333,
+      "learning_rate": 0.0001,
+      "loss": 4.1218,
+      "loss/crossentropy": 1.8330454230308533,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20832987129688263,
+      "step": 7800
+    },
+    {
+      "epoch": 0.15604,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.007420857747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2579,
+      "loss/crossentropy": 1.9194663166999817,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22163349390029907,
+      "step": 7802
+    },
+    {
+      "epoch": 0.15608,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.006917317708333333,
+      "learning_rate": 0.0001,
+      "loss": 4.2219,
+      "loss/crossentropy": 1.798878252506256,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19532181322574615,
+      "step": 7804
+    },
+    {
+      "epoch": 0.15612,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0067942301432291664,
+      "learning_rate": 0.0001,
+      "loss": 4.2949,
+      "loss/crossentropy": 1.730432152748108,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20092248916625977,
+      "step": 7806
+    },
+    {
+      "epoch": 0.15616,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.01304931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3109,
+      "loss/crossentropy": 2.1426968574523926,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22185371816158295,
+      "step": 7808
+    },
+    {
+      "epoch": 0.1562,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.012848917643229167,
+      "learning_rate": 0.0001,
+      "loss": 4.4308,
+      "loss/crossentropy": 1.983969271183014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22203146666288376,
+      "step": 7810
+    },
+    {
+      "epoch": 0.15624,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.016532389322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.2114,
+      "loss/crossentropy": 2.2948192954063416,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23453453928232193,
+      "step": 7812
+    },
+    {
+      "epoch": 0.15628,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0175933837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.287,
+      "loss/crossentropy": 1.9190048575401306,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22457829862833023,
+      "step": 7814
+    },
+    {
+      "epoch": 0.15632,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.01529541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1721,
+      "loss/crossentropy": 2.0268847346305847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21628276258707047,
+      "step": 7816
+    },
+    {
+      "epoch": 0.15636,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.017411295572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.3553,
+      "loss/crossentropy": 2.0050706267356873,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20108875632286072,
+      "step": 7818
+    },
+    {
+      "epoch": 0.1564,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.024470774332682292,
+      "learning_rate": 0.0001,
+      "loss": 4.2059,
+      "loss/crossentropy": 2.096635937690735,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22623063623905182,
+      "step": 7820
+    },
+    {
+      "epoch": 0.15644,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.024580637613932293,
+      "learning_rate": 0.0001,
+      "loss": 4.3901,
+      "loss/crossentropy": 1.9188768863677979,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21088901162147522,
+      "step": 7822
+    },
+    {
+      "epoch": 0.15648,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.018304189046223957,
+      "learning_rate": 0.0001,
+      "loss": 4.4062,
+      "loss/crossentropy": 2.3639817237854004,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2608294039964676,
+      "step": 7824
+    },
+    {
+      "epoch": 0.15652,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.017765045166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2123,
+      "loss/crossentropy": 1.9716956615447998,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23327408730983734,
+      "step": 7826
+    },
+    {
+      "epoch": 0.15656,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.08131688435872396,
+      "learning_rate": 0.0001,
+      "loss": 4.1377,
+      "loss/crossentropy": 2.004276990890503,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22930586338043213,
+      "step": 7828
+    },
+    {
+      "epoch": 0.1566,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.08247858683268229,
+      "learning_rate": 0.0001,
+      "loss": 4.4703,
+      "loss/crossentropy": 1.7996181845664978,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21030305325984955,
+      "step": 7830
+    },
+    {
+      "epoch": 0.15664,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.09555435180664062,
+      "learning_rate": 0.0001,
+      "loss": 4.7938,
+      "loss/crossentropy": 2.192178189754486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2232302725315094,
+      "step": 7832
+    },
+    {
+      "epoch": 0.15668,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.09538345336914063,
+      "learning_rate": 0.0001,
+      "loss": 4.2381,
+      "loss/crossentropy": 1.7093925476074219,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19697313755750656,
+      "step": 7834
+    },
+    {
+      "epoch": 0.15672,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.09130859375,
+      "learning_rate": 0.0001,
+      "loss": 4.2534,
+      "loss/crossentropy": 1.915247917175293,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21003933250904083,
+      "step": 7836
+    },
+    {
+      "epoch": 0.15676,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.091162109375,
+      "learning_rate": 0.0001,
+      "loss": 4.3825,
+      "loss/crossentropy": 2.188641667366028,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23260314762592316,
+      "step": 7838
+    },
+    {
+      "epoch": 0.1568,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.09501953125,
+      "learning_rate": 0.0001,
+      "loss": 4.7329,
+      "loss/crossentropy": 2.3316495418548584,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24601806700229645,
+      "step": 7840
+    },
+    {
+      "epoch": 0.15684,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.0918121337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2934,
+      "loss/crossentropy": 2.140946924686432,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23532958328723907,
+      "step": 7842
+    },
+    {
+      "epoch": 0.15688,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.031525675455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2738,
+      "loss/crossentropy": 2.372095465660095,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2615740895271301,
+      "step": 7844
+    },
+    {
+      "epoch": 0.15692,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.028727213541666668,
+      "learning_rate": 0.0001,
+      "loss": 4.1954,
+      "loss/crossentropy": 1.8433185815811157,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2160269021987915,
+      "step": 7846
+    },
+    {
+      "epoch": 0.15696,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.016071573893229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2489,
+      "loss/crossentropy": 2.012324333190918,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1964300200343132,
+      "step": 7848
+    },
+    {
+      "epoch": 0.157,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.014090983072916667,
+      "learning_rate": 0.0001,
+      "loss": 4.486,
+      "loss/crossentropy": 2.0325884222984314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22080926597118378,
+      "step": 7850
+    },
+    {
+      "epoch": 0.15704,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.013036092122395834,
+      "learning_rate": 0.0001,
+      "loss": 4.3784,
+      "loss/crossentropy": 2.3786104917526245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24448946118354797,
+      "step": 7852
+    },
+    {
+      "epoch": 0.15708,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.011693318684895834,
+      "learning_rate": 0.0001,
+      "loss": 4.5173,
+      "loss/crossentropy": 2.304913640022278,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24708375334739685,
+      "step": 7854
+    },
+    {
+      "epoch": 0.15712,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.006371053059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.5711,
+      "loss/crossentropy": 1.8640215396881104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19824489206075668,
+      "step": 7856
+    },
+    {
+      "epoch": 0.15716,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.00625,
+      "learning_rate": 0.0001,
+      "loss": 4.6174,
+      "loss/crossentropy": 2.444548487663269,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2542492523789406,
+      "step": 7858
+    },
+    {
+      "epoch": 0.1572,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.00592041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0522,
+      "loss/crossentropy": 1.991346299648285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2159758359193802,
+      "step": 7860
+    },
+    {
+      "epoch": 0.15724,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.005101521809895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2627,
+      "loss/crossentropy": 2.3172048926353455,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2506742626428604,
+      "step": 7862
+    },
+    {
+      "epoch": 0.15728,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0050201416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.5415,
+      "loss/crossentropy": 1.9302632212638855,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2104542776942253,
+      "step": 7864
+    },
+    {
+      "epoch": 0.15732,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.020726521809895832,
+      "learning_rate": 0.0001,
+      "loss": 4.7831,
+      "loss/crossentropy": 2.457883358001709,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23538483679294586,
+      "step": 7866
+    },
+    {
+      "epoch": 0.15736,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.019677734375,
+      "learning_rate": 0.0001,
+      "loss": 4.1912,
+      "loss/crossentropy": 1.9857566952705383,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2131117358803749,
+      "step": 7868
+    },
+    {
+      "epoch": 0.1574,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.019140625,
+      "learning_rate": 0.0001,
+      "loss": 4.4264,
+      "loss/crossentropy": 2.0062127113342285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22806233912706375,
+      "step": 7870
+    },
+    {
+      "epoch": 0.15744,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.019489542643229166,
+      "learning_rate": 0.0001,
+      "loss": 4.6181,
+      "loss/crossentropy": 2.283127784729004,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23883548378944397,
+      "step": 7872
+    },
+    {
+      "epoch": 0.15748,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.019136555989583335,
+      "learning_rate": 0.0001,
+      "loss": 4.4442,
+      "loss/crossentropy": 2.110979437828064,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.209333136677742,
+      "step": 7874
+    },
+    {
+      "epoch": 0.15752,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.019856770833333332,
+      "learning_rate": 0.0001,
+      "loss": 4.2107,
+      "loss/crossentropy": 1.8705166578292847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20791998505592346,
+      "step": 7876
+    },
+    {
+      "epoch": 0.15756,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.019017537434895832,
+      "learning_rate": 0.0001,
+      "loss": 4.2956,
+      "loss/crossentropy": 2.133803129196167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23219672590494156,
+      "step": 7878
+    },
+    {
+      "epoch": 0.1576,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.019627888997395832,
+      "learning_rate": 0.0001,
+      "loss": 4.5028,
+      "loss/crossentropy": 2.1062549352645874,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2266012579202652,
+      "step": 7880
+    },
+    {
+      "epoch": 0.15764,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.0058553059895833336,
+      "learning_rate": 0.0001,
+      "loss": 4.5078,
+      "loss/crossentropy": 2.0088155269622803,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2079932913184166,
+      "step": 7882
+    },
+    {
+      "epoch": 0.15768,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0065582275390625,
+      "learning_rate": 0.0001,
+      "loss": 3.9204,
+      "loss/crossentropy": 1.6621176600456238,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19370710104703903,
+      "step": 7884
+    },
+    {
+      "epoch": 0.15772,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.007136027018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.3462,
+      "loss/crossentropy": 1.7729167938232422,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18562481552362442,
+      "step": 7886
+    },
+    {
+      "epoch": 0.15776,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.006843058268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.3897,
+      "loss/crossentropy": 2.132485508918762,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23432063311338425,
+      "step": 7888
+    },
+    {
+      "epoch": 0.1578,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.0091949462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3731,
+      "loss/crossentropy": 2.122144937515259,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23042195290327072,
+      "step": 7890
+    },
+    {
+      "epoch": 0.15784,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.034956868489583334,
+      "learning_rate": 0.0001,
+      "loss": 4.4025,
+      "loss/crossentropy": 1.855578601360321,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20124691724777222,
+      "step": 7892
+    },
+    {
+      "epoch": 0.15788,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.035054524739583336,
+      "learning_rate": 0.0001,
+      "loss": 4.172,
+      "loss/crossentropy": 2.02128005027771,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20766886323690414,
+      "step": 7894
+    },
+    {
+      "epoch": 0.15792,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.03681233723958333,
+      "learning_rate": 0.0001,
+      "loss": 4.3096,
+      "loss/crossentropy": 2.112824857234955,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22250613570213318,
+      "step": 7896
+    },
+    {
+      "epoch": 0.15796,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.03662821451822917,
+      "learning_rate": 0.0001,
+      "loss": 4.5595,
+      "loss/crossentropy": 2.2290207147598267,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2531846910715103,
+      "step": 7898
+    },
+    {
+      "epoch": 0.158,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.03882548014322917,
+      "learning_rate": 0.0001,
+      "loss": 4.3586,
+      "loss/crossentropy": 2.135373592376709,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2266940325498581,
+      "step": 7900
+    },
+    {
+      "epoch": 0.15804,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.04052734375,
+      "learning_rate": 0.0001,
+      "loss": 4.1719,
+      "loss/crossentropy": 1.7298616170883179,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1986929401755333,
+      "step": 7902
+    },
+    {
+      "epoch": 0.15808,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.04173075358072917,
+      "learning_rate": 0.0001,
+      "loss": 4.403,
+      "loss/crossentropy": 2.18759286403656,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21622422337532043,
+      "step": 7904
+    },
+    {
+      "epoch": 0.15812,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.03889058430989583,
+      "learning_rate": 0.0001,
+      "loss": 4.4765,
+      "loss/crossentropy": 2.0889216661453247,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21440055221319199,
+      "step": 7906
+    },
+    {
+      "epoch": 0.15816,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.010204060872395834,
+      "learning_rate": 0.0001,
+      "loss": 4.2231,
+      "loss/crossentropy": 1.7791658639907837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1967781037092209,
+      "step": 7908
+    },
+    {
+      "epoch": 0.1582,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.012386067708333334,
+      "learning_rate": 0.0001,
+      "loss": 4.2314,
+      "loss/crossentropy": 2.2144845724105835,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22152486443519592,
+      "step": 7910
+    },
+    {
+      "epoch": 0.15824,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.012723795572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.6236,
+      "loss/crossentropy": 2.316117286682129,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24549901485443115,
+      "step": 7912
+    },
+    {
+      "epoch": 0.15828,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0115875244140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0867,
+      "loss/crossentropy": 1.8642511367797852,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22441796958446503,
+      "step": 7914
+    },
+    {
+      "epoch": 0.15832,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0098541259765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1365,
+      "loss/crossentropy": 1.807969868183136,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22138798981904984,
+      "step": 7916
+    },
+    {
+      "epoch": 0.15836,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.013866170247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.5625,
+      "loss/crossentropy": 2.266697645187378,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22544697672128677,
+      "step": 7918
+    },
+    {
+      "epoch": 0.1584,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.012262980143229166,
+      "learning_rate": 0.0001,
+      "loss": 4.5632,
+      "loss/crossentropy": 2.0871587991714478,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21997228264808655,
+      "step": 7920
+    },
+    {
+      "epoch": 0.15844,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0126861572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.5567,
+      "loss/crossentropy": 2.1993759870529175,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2416691780090332,
+      "step": 7922
+    },
+    {
+      "epoch": 0.15848,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.020340983072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.2997,
+      "loss/crossentropy": 1.8039653897285461,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21721573173999786,
+      "step": 7924
+    },
+    {
+      "epoch": 0.15852,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.0225982666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.4795,
+      "loss/crossentropy": 2.188117265701294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.231883242726326,
+      "step": 7926
+    },
+    {
+      "epoch": 0.15856,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.021923828125,
+      "learning_rate": 0.0001,
+      "loss": 4.6694,
+      "loss/crossentropy": 2.3920425176620483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24655026197433472,
+      "step": 7928
+    },
+    {
+      "epoch": 0.1586,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.017561848958333334,
+      "learning_rate": 0.0001,
+      "loss": 4.1001,
+      "loss/crossentropy": 2.286831498146057,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23398208618164062,
+      "step": 7930
+    },
+    {
+      "epoch": 0.15864,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.016825358072916668,
+      "learning_rate": 0.0001,
+      "loss": 4.0007,
+      "loss/crossentropy": 2.075824797153473,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22423933446407318,
+      "step": 7932
+    },
+    {
+      "epoch": 0.15868,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0155426025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.5697,
+      "loss/crossentropy": 2.197165012359619,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22215355187654495,
+      "step": 7934
+    },
+    {
+      "epoch": 0.15872,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.017854817708333335,
+      "learning_rate": 0.0001,
+      "loss": 4.2899,
+      "loss/crossentropy": 1.8253535032272339,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20457974076271057,
+      "step": 7936
+    },
+    {
+      "epoch": 0.15876,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.01783447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4774,
+      "loss/crossentropy": 1.842383086681366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20742817968130112,
+      "step": 7938
+    },
+    {
+      "epoch": 0.1588,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.014208984375,
+      "learning_rate": 0.0001,
+      "loss": 4.3801,
+      "loss/crossentropy": 2.2086315155029297,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2086438685655594,
+      "step": 7940
+    },
+    {
+      "epoch": 0.15884,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0067047119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3718,
+      "loss/crossentropy": 2.3381282091140747,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2546956539154053,
+      "step": 7942
+    },
+    {
+      "epoch": 0.15888,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.005248006184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.3439,
+      "loss/crossentropy": 2.1159931421279907,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23917824029922485,
+      "step": 7944
+    },
+    {
+      "epoch": 0.15892,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0069081624348958336,
+      "learning_rate": 0.0001,
+      "loss": 4.2056,
+      "loss/crossentropy": 1.7507159113883972,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1870383694767952,
+      "step": 7946
+    },
+    {
+      "epoch": 0.15896,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0073964436848958336,
+      "learning_rate": 0.0001,
+      "loss": 4.2118,
+      "loss/crossentropy": 2.213807225227356,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22211921215057373,
+      "step": 7948
+    },
+    {
+      "epoch": 0.159,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009598795572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.4316,
+      "loss/crossentropy": 2.0570366978645325,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2662791311740875,
+      "step": 7950
+    },
+    {
+      "epoch": 0.15904,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.009748331705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.2041,
+      "loss/crossentropy": 1.8775206208229065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22045740485191345,
+      "step": 7952
+    },
+    {
+      "epoch": 0.15908,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.009601847330729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2575,
+      "loss/crossentropy": 2.3483108282089233,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22626785188913345,
+      "step": 7954
+    },
+    {
+      "epoch": 0.15912,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.009496053059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.193,
+      "loss/crossentropy": 1.9531084895133972,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22194421291351318,
+      "step": 7956
+    },
+    {
+      "epoch": 0.15916,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.011213175455729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4819,
+      "loss/crossentropy": 2.057813823223114,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21116723865270615,
+      "step": 7958
+    },
+    {
+      "epoch": 0.1592,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.01148681640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3478,
+      "loss/crossentropy": 1.9398415088653564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2257898524403572,
+      "step": 7960
+    },
+    {
+      "epoch": 0.15924,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0111480712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1942,
+      "loss/crossentropy": 1.9200270175933838,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2119666188955307,
+      "step": 7962
+    },
+    {
+      "epoch": 0.15928,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.010640462239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.4513,
+      "loss/crossentropy": 2.157149076461792,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23271384835243225,
+      "step": 7964
+    },
+    {
+      "epoch": 0.15932,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0083404541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2064,
+      "loss/crossentropy": 2.0308582186698914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23934553563594818,
+      "step": 7966
+    },
+    {
+      "epoch": 0.15936,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.015543619791666666,
+      "learning_rate": 0.0001,
+      "loss": 4.3977,
+      "loss/crossentropy": 2.1855711936950684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.236568883061409,
+      "step": 7968
+    },
+    {
+      "epoch": 0.1594,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.07629292805989583,
+      "learning_rate": 0.0001,
+      "loss": 4.719,
+      "loss/crossentropy": 2.4479551315307617,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24693088978528976,
+      "step": 7970
+    },
+    {
+      "epoch": 0.15944,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.07333882649739583,
+      "learning_rate": 0.0001,
+      "loss": 4.5221,
+      "loss/crossentropy": 2.439974784851074,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2502682954072952,
+      "step": 7972
+    },
+    {
+      "epoch": 0.15948,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.07625325520833333,
+      "learning_rate": 0.0001,
+      "loss": 4.4024,
+      "loss/crossentropy": 1.9899010062217712,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23031124472618103,
+      "step": 7974
+    },
+    {
+      "epoch": 0.15952,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.07517903645833333,
+      "learning_rate": 0.0001,
+      "loss": 4.2522,
+      "loss/crossentropy": 1.830255150794983,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1984459012746811,
+      "step": 7976
+    },
+    {
+      "epoch": 0.15956,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.07088114420572916,
+      "learning_rate": 0.0001,
+      "loss": 4.3832,
+      "loss/crossentropy": 1.9675705432891846,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21336429566144943,
+      "step": 7978
+    },
+    {
+      "epoch": 0.1596,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.06965738932291667,
+      "learning_rate": 0.0001,
+      "loss": 4.4784,
+      "loss/crossentropy": 2.030815005302429,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2102503478527069,
+      "step": 7980
+    },
+    {
+      "epoch": 0.15964,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.07285054524739583,
+      "learning_rate": 0.0001,
+      "loss": 4.492,
+      "loss/crossentropy": 2.4932440519332886,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.246867336332798,
+      "step": 7982
+    },
+    {
+      "epoch": 0.15968,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.07330322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2085,
+      "loss/crossentropy": 1.5839802622795105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17231802642345428,
+      "step": 7984
+    },
+    {
+      "epoch": 0.15972,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.0091217041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.4045,
+      "loss/crossentropy": 1.821477472782135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2239791452884674,
+      "step": 7986
+    },
+    {
+      "epoch": 0.15976,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0093414306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0655,
+      "loss/crossentropy": 2.013838052749634,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2335103377699852,
+      "step": 7988
+    },
+    {
+      "epoch": 0.1598,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.006376139322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.4989,
+      "loss/crossentropy": 1.9412779211997986,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20336110144853592,
+      "step": 7990
+    },
+    {
+      "epoch": 0.15984,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.008234659830729166,
+      "learning_rate": 0.0001,
+      "loss": 3.9015,
+      "loss/crossentropy": 1.6653677225112915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2025909200310707,
+      "step": 7992
+    },
+    {
+      "epoch": 0.15988,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.00865478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4299,
+      "loss/crossentropy": 2.0069726705551147,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21300900727510452,
+      "step": 7994
+    },
+    {
+      "epoch": 0.15992,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.009919230143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.4395,
+      "loss/crossentropy": 2.1118472814559937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25853876769542694,
+      "step": 7996
+    },
+    {
+      "epoch": 0.15996,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.013622029622395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1922,
+      "loss/crossentropy": 1.608262836933136,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1958206593990326,
+      "step": 7998
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0138580322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2896,
+      "loss/crossentropy": 1.6572073101997375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19102878868579865,
+      "step": 8000
+    },
+    {
+      "epoch": 0.16004,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0113677978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0811,
+      "loss/crossentropy": 1.9421688318252563,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19628942012786865,
+      "step": 8002
+    },
+    {
+      "epoch": 0.16008,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0134918212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5344,
+      "loss/crossentropy": 2.2197489738464355,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.248014435172081,
+      "step": 8004
+    },
+    {
+      "epoch": 0.16012,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0126617431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4625,
+      "loss/crossentropy": 2.200868308544159,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2476908192038536,
+      "step": 8006
+    },
+    {
+      "epoch": 0.16016,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.01162109375,
+      "learning_rate": 0.0001,
+      "loss": 4.8151,
+      "loss/crossentropy": 2.3793649673461914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2356039509177208,
+      "step": 8008
+    },
+    {
+      "epoch": 0.1602,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0112945556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1849,
+      "loss/crossentropy": 2.130257308483124,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.224358968436718,
+      "step": 8010
+    },
+    {
+      "epoch": 0.16024,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.009566243489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.5948,
+      "loss/crossentropy": 2.370365023612976,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23811528086662292,
+      "step": 8012
+    },
+    {
+      "epoch": 0.16028,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.008812459309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4326,
+      "loss/crossentropy": 1.985486626625061,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19639353454113007,
+      "step": 8014
+    },
+    {
+      "epoch": 0.16032,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.01129150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.4335,
+      "loss/crossentropy": 2.2128632068634033,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2340361252427101,
+      "step": 8016
+    },
+    {
+      "epoch": 0.16036,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009956868489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.3688,
+      "loss/crossentropy": 1.830498456954956,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19760622829198837,
+      "step": 8018
+    },
+    {
+      "epoch": 0.1604,
+      "grad_norm": 4.375,
+      "grad_norm_var": 0.3061757405598958,
+      "learning_rate": 0.0001,
+      "loss": 4.6443,
+      "loss/crossentropy": 1.9595977067947388,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2138657420873642,
+      "step": 8020
+    },
+    {
+      "epoch": 0.16044,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.30684305826822916,
+      "learning_rate": 0.0001,
+      "loss": 4.3101,
+      "loss/crossentropy": 2.400893449783325,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2615668326616287,
+      "step": 8022
+    },
+    {
+      "epoch": 0.16048,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.3080963134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.6359,
+      "loss/crossentropy": 2.5079843997955322,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.251235693693161,
+      "step": 8024
+    },
+    {
+      "epoch": 0.16052,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.30686848958333335,
+      "learning_rate": 0.0001,
+      "loss": 4.4973,
+      "loss/crossentropy": 2.573891043663025,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2646481841802597,
+      "step": 8026
+    },
+    {
+      "epoch": 0.16056,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.3064605712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4177,
+      "loss/crossentropy": 2.0088363885879517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22888045758008957,
+      "step": 8028
+    },
+    {
+      "epoch": 0.1606,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.31115697224934896,
+      "learning_rate": 0.0001,
+      "loss": 4.214,
+      "loss/crossentropy": 2.4596647024154663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2507014721632004,
+      "step": 8030
+    },
+    {
+      "epoch": 0.16064,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.3093462626139323,
+      "learning_rate": 0.0001,
+      "loss": 4.6826,
+      "loss/crossentropy": 2.204255223274231,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22913406044244766,
+      "step": 8032
+    },
+    {
+      "epoch": 0.16068,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.30677261352539065,
+      "learning_rate": 0.0001,
+      "loss": 4.6603,
+      "loss/crossentropy": 2.285408139228821,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23185817897319794,
+      "step": 8034
+    },
+    {
+      "epoch": 0.16072,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.012520345052083333,
+      "learning_rate": 0.0001,
+      "loss": 3.9513,
+      "loss/crossentropy": 2.0788660645484924,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20811481028795242,
+      "step": 8036
+    },
+    {
+      "epoch": 0.16076,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.01307373046875,
+      "learning_rate": 0.0001,
+      "loss": 4.5368,
+      "loss/crossentropy": 2.398258686065674,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2411409318447113,
+      "step": 8038
+    },
+    {
+      "epoch": 0.1608,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.015579986572265624,
+      "learning_rate": 0.0001,
+      "loss": 4.0415,
+      "loss/crossentropy": 2.3101454973220825,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22493668645620346,
+      "step": 8040
+    },
+    {
+      "epoch": 0.16084,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.015134429931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0612,
+      "loss/crossentropy": 1.9015105962753296,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22360052913427353,
+      "step": 8042
+    },
+    {
+      "epoch": 0.16088,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.017704010009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.216,
+      "loss/crossentropy": 2.0112481117248535,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24062782526016235,
+      "step": 8044
+    },
+    {
+      "epoch": 0.16092,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0161773681640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3638,
+      "loss/crossentropy": 2.2024285793304443,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21959003806114197,
+      "step": 8046
+    },
+    {
+      "epoch": 0.16096,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.014412434895833333,
+      "learning_rate": 0.0001,
+      "loss": 4.4825,
+      "loss/crossentropy": 2.1069058775901794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21980682760477066,
+      "step": 8048
+    },
+    {
+      "epoch": 0.161,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.012540690104166667,
+      "learning_rate": 0.0001,
+      "loss": 4.2106,
+      "loss/crossentropy": 1.8380340337753296,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2201671525835991,
+      "step": 8050
+    },
+    {
+      "epoch": 0.16104,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.010965728759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3778,
+      "loss/crossentropy": 2.276741087436676,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23787930607795715,
+      "step": 8052
+    },
+    {
+      "epoch": 0.16108,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.010680898030598959,
+      "learning_rate": 0.0001,
+      "loss": 4.0784,
+      "loss/crossentropy": 1.631809651851654,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19654212146997452,
+      "step": 8054
+    },
+    {
+      "epoch": 0.16112,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.007453409830729166,
+      "learning_rate": 0.0001,
+      "loss": 4.5927,
+      "loss/crossentropy": 2.067444145679474,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21420229971408844,
+      "step": 8056
+    },
+    {
+      "epoch": 0.16116,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.011767578125,
+      "learning_rate": 0.0001,
+      "loss": 4.562,
+      "loss/crossentropy": 2.384890556335449,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2312106341123581,
+      "step": 8058
+    },
+    {
+      "epoch": 0.1612,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.010986328125,
+      "learning_rate": 0.0001,
+      "loss": 4.363,
+      "loss/crossentropy": 2.3683160543441772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2527216002345085,
+      "step": 8060
+    },
+    {
+      "epoch": 0.16124,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.011253865559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4576,
+      "loss/crossentropy": 2.1545952558517456,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23928315192461014,
+      "step": 8062
+    },
+    {
+      "epoch": 0.16128,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.013695271809895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2285,
+      "loss/crossentropy": 2.0792208313941956,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2226722463965416,
+      "step": 8064
+    },
+    {
+      "epoch": 0.16132,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.013068644205729167,
+      "learning_rate": 0.0001,
+      "loss": 3.8788,
+      "loss/crossentropy": 2.181519627571106,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2193506434559822,
+      "step": 8066
+    },
+    {
+      "epoch": 0.16136,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.015067545572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.4445,
+      "loss/crossentropy": 1.7798657417297363,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23413337767124176,
+      "step": 8068
+    },
+    {
+      "epoch": 0.1614,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.017235310872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0731,
+      "loss/crossentropy": 2.1232666969299316,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23351240158081055,
+      "step": 8070
+    },
+    {
+      "epoch": 0.16144,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.017609659830729166,
+      "learning_rate": 0.0001,
+      "loss": 4.6975,
+      "loss/crossentropy": 2.34002685546875,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2533388137817383,
+      "step": 8072
+    },
+    {
+      "epoch": 0.16148,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.011637369791666666,
+      "learning_rate": 0.0001,
+      "loss": 4.3487,
+      "loss/crossentropy": 2.066399872303009,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21156150847673416,
+      "step": 8074
+    },
+    {
+      "epoch": 0.16152,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.010677083333333334,
+      "learning_rate": 0.0001,
+      "loss": 4.3646,
+      "loss/crossentropy": 2.2298463582992554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2307809516787529,
+      "step": 8076
+    },
+    {
+      "epoch": 0.16156,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0105865478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1946,
+      "loss/crossentropy": 1.9858508110046387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21499747782945633,
+      "step": 8078
+    },
+    {
+      "epoch": 0.1616,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.011149088541666666,
+      "learning_rate": 0.0001,
+      "loss": 3.9984,
+      "loss/crossentropy": 1.5669215321540833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1889330968260765,
+      "step": 8080
+    },
+    {
+      "epoch": 0.16164,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.011881510416666666,
+      "learning_rate": 0.0001,
+      "loss": 4.4743,
+      "loss/crossentropy": 2.296878218650818,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2405889928340912,
+      "step": 8082
+    },
+    {
+      "epoch": 0.16168,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.007307942708333333,
+      "learning_rate": 0.0001,
+      "loss": 4.2205,
+      "loss/crossentropy": 1.954626441001892,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19809379428625107,
+      "step": 8084
+    },
+    {
+      "epoch": 0.16172,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.008687337239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.5923,
+      "loss/crossentropy": 1.8640353083610535,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.226090669631958,
+      "step": 8086
+    },
+    {
+      "epoch": 0.16176,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.009663899739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.3363,
+      "loss/crossentropy": 2.1914591789245605,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24432511627674103,
+      "step": 8088
+    },
+    {
+      "epoch": 0.1618,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.009798177083333333,
+      "learning_rate": 0.0001,
+      "loss": 4.7291,
+      "loss/crossentropy": 2.192594051361084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23682628571987152,
+      "step": 8090
+    },
+    {
+      "epoch": 0.16184,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.010993448893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.4592,
+      "loss/crossentropy": 2.210235595703125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.230379119515419,
+      "step": 8092
+    },
+    {
+      "epoch": 0.16188,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.011115519205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4111,
+      "loss/crossentropy": 2.214667320251465,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26643867790699005,
+      "step": 8094
+    },
+    {
+      "epoch": 0.16192,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.009586588541666666,
+      "learning_rate": 0.0001,
+      "loss": 4.3131,
+      "loss/crossentropy": 1.9808599948883057,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20679324120283127,
+      "step": 8096
+    },
+    {
+      "epoch": 0.16196,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.008333333333333333,
+      "learning_rate": 0.0001,
+      "loss": 4.4683,
+      "loss/crossentropy": 2.076589345932007,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23430196940898895,
+      "step": 8098
+    },
+    {
+      "epoch": 0.162,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.006883748372395833,
+      "learning_rate": 0.0001,
+      "loss": 4.5645,
+      "loss/crossentropy": 2.364492177963257,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23023054748773575,
+      "step": 8100
+    },
+    {
+      "epoch": 0.16204,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.005890909830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.6595,
+      "loss/crossentropy": 2.2908111214637756,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21239468455314636,
+      "step": 8102
+    },
+    {
+      "epoch": 0.16208,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.008958943684895833,
+      "learning_rate": 0.0001,
+      "loss": 4.5503,
+      "loss/crossentropy": 1.8306183218955994,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21511316299438477,
+      "step": 8104
+    },
+    {
+      "epoch": 0.16212,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010724894205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.1582,
+      "loss/crossentropy": 2.1121758222579956,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23471853882074356,
+      "step": 8106
+    },
+    {
+      "epoch": 0.16216,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.012355295817057292,
+      "learning_rate": 0.0001,
+      "loss": 4.1464,
+      "loss/crossentropy": 1.7613067030906677,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2025228664278984,
+      "step": 8108
+    },
+    {
+      "epoch": 0.1622,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.013844553629557292,
+      "learning_rate": 0.0001,
+      "loss": 4.0797,
+      "loss/crossentropy": 2.1413058042526245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21817607432603836,
+      "step": 8110
+    },
+    {
+      "epoch": 0.16224,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.013641103108723959,
+      "learning_rate": 0.0001,
+      "loss": 4.5739,
+      "loss/crossentropy": 2.375948429107666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30982857942581177,
+      "step": 8112
+    },
+    {
+      "epoch": 0.16228,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.013396962483723959,
+      "learning_rate": 0.0001,
+      "loss": 4.076,
+      "loss/crossentropy": 1.9669193029403687,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21873307973146439,
+      "step": 8114
+    },
+    {
+      "epoch": 0.16232,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.015592193603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4024,
+      "loss/crossentropy": 2.2028547525405884,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.229970782995224,
+      "step": 8116
+    },
+    {
+      "epoch": 0.16236,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.01622289021809896,
+      "learning_rate": 0.0001,
+      "loss": 4.3359,
+      "loss/crossentropy": 2.082156002521515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2262335941195488,
+      "step": 8118
+    },
+    {
+      "epoch": 0.1624,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.012379709879557292,
+      "learning_rate": 0.0001,
+      "loss": 4.2737,
+      "loss/crossentropy": 2.0538666248321533,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22941745817661285,
+      "step": 8120
+    },
+    {
+      "epoch": 0.16244,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.015750885009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3615,
+      "loss/crossentropy": 2.338989734649658,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2434876710176468,
+      "step": 8122
+    },
+    {
+      "epoch": 0.16248,
+      "grad_norm": 2.375,
+      "grad_norm_var": 2.602311197916667,
+      "learning_rate": 0.0001,
+      "loss": 4.5472,
+      "loss/crossentropy": 2.277916193008423,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29285988211631775,
+      "step": 8124
+    },
+    {
+      "epoch": 0.16252,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 2.567577107747396,
+      "learning_rate": 0.0001,
+      "loss": 4.3437,
+      "loss/crossentropy": 2.1196334958076477,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20728030800819397,
+      "step": 8126
+    },
+    {
+      "epoch": 0.16256,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 2.5655558268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.4216,
+      "loss/crossentropy": 2.0216450095176697,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.229092076420784,
+      "step": 8128
+    },
+    {
+      "epoch": 0.1626,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 2.5546834309895834,
+      "learning_rate": 0.0001,
+      "loss": 4.5163,
+      "loss/crossentropy": 2.098921537399292,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22539281845092773,
+      "step": 8130
+    },
+    {
+      "epoch": 0.16264,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 2.580052693684896,
+      "learning_rate": 0.0001,
+      "loss": 4.1066,
+      "loss/crossentropy": 2.0186068415641785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20286529511213303,
+      "step": 8132
+    },
+    {
+      "epoch": 0.16268,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 2.5584706624348956,
+      "learning_rate": 0.0001,
+      "loss": 4.7847,
+      "loss/crossentropy": 2.157357335090637,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23342902958393097,
+      "step": 8134
+    },
+    {
+      "epoch": 0.16272,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 2.5516998291015627,
+      "learning_rate": 0.0001,
+      "loss": 4.3489,
+      "loss/crossentropy": 2.1885476112365723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22314801812171936,
+      "step": 8136
+    },
+    {
+      "epoch": 0.16276,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 2.5796160380045574,
+      "learning_rate": 0.0001,
+      "loss": 4.3403,
+      "loss/crossentropy": 2.2208757400512695,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2295266091823578,
+      "step": 8138
+    },
+    {
+      "epoch": 0.1628,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.027522532145182292,
+      "learning_rate": 0.0001,
+      "loss": 4.0473,
+      "loss/crossentropy": 1.7885233163833618,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2008970081806183,
+      "step": 8140
+    },
+    {
+      "epoch": 0.16284,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.022989654541015626,
+      "learning_rate": 0.0001,
+      "loss": 4.0835,
+      "loss/crossentropy": 2.152435064315796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2311881259083748,
+      "step": 8142
+    },
+    {
+      "epoch": 0.16288,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.016001129150390626,
+      "learning_rate": 0.0001,
+      "loss": 4.3384,
+      "loss/crossentropy": 2.0818406343460083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24117180705070496,
+      "step": 8144
+    },
+    {
+      "epoch": 0.16292,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.02513402303059896,
+      "learning_rate": 0.0001,
+      "loss": 4.4246,
+      "loss/crossentropy": 2.1775856614112854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23246955126523972,
+      "step": 8146
+    },
+    {
+      "epoch": 0.16296,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.033607737223307295,
+      "learning_rate": 0.0001,
+      "loss": 3.9993,
+      "loss/crossentropy": 1.7029682397842407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19564303010702133,
+      "step": 8148
+    },
+    {
+      "epoch": 0.163,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.03468195597330729,
+      "learning_rate": 0.0001,
+      "loss": 4.5631,
+      "loss/crossentropy": 2.070194900035858,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2382609099149704,
+      "step": 8150
+    },
+    {
+      "epoch": 0.16304,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.036043039957682294,
+      "learning_rate": 0.0001,
+      "loss": 4.19,
+      "loss/crossentropy": 1.9214876890182495,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20894134789705276,
+      "step": 8152
+    },
+    {
+      "epoch": 0.16308,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.03243815104166667,
+      "learning_rate": 0.0001,
+      "loss": 4.3622,
+      "loss/crossentropy": 1.7311474084854126,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1962232068181038,
+      "step": 8154
+    },
+    {
+      "epoch": 0.16312,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0323150634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4623,
+      "loss/crossentropy": 2.2161877155303955,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22318138182163239,
+      "step": 8156
+    },
+    {
+      "epoch": 0.16316,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.03186747233072917,
+      "learning_rate": 0.0001,
+      "loss": 4.7343,
+      "loss/crossentropy": 2.21865177154541,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.234280064702034,
+      "step": 8158
+    },
+    {
+      "epoch": 0.1632,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.032835896809895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3277,
+      "loss/crossentropy": 1.9517142176628113,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1964063122868538,
+      "step": 8160
+    },
+    {
+      "epoch": 0.16324,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.023713175455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.3878,
+      "loss/crossentropy": 2.261234760284424,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24549759924411774,
+      "step": 8162
+    },
+    {
+      "epoch": 0.16328,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012751261393229166,
+      "learning_rate": 0.0001,
+      "loss": 4.159,
+      "loss/crossentropy": 1.9791623950004578,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2018037736415863,
+      "step": 8164
+    },
+    {
+      "epoch": 0.16332,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.005125935872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.6231,
+      "loss/crossentropy": 2.3916029930114746,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24446460604667664,
+      "step": 8166
+    },
+    {
+      "epoch": 0.16336,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.007991536458333334,
+      "learning_rate": 0.0001,
+      "loss": 4.3251,
+      "loss/crossentropy": 2.204437553882599,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22334939241409302,
+      "step": 8168
+    },
+    {
+      "epoch": 0.1634,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.007111612955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3236,
+      "loss/crossentropy": 2.2013272047042847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2068365067243576,
+      "step": 8170
+    },
+    {
+      "epoch": 0.16344,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0065826416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2597,
+      "loss/crossentropy": 1.8648701310157776,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19690127670764923,
+      "step": 8172
+    },
+    {
+      "epoch": 0.16348,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.008857981363932291,
+      "learning_rate": 0.0001,
+      "loss": 4.1324,
+      "loss/crossentropy": 2.221195936203003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1900760903954506,
+      "step": 8174
+    },
+    {
+      "epoch": 0.16352,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.009445953369140624,
+      "learning_rate": 0.0001,
+      "loss": 4.4874,
+      "loss/crossentropy": 1.8648499846458435,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1982845515012741,
+      "step": 8176
+    },
+    {
+      "epoch": 0.16356,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010762278238932292,
+      "learning_rate": 0.0001,
+      "loss": 4.588,
+      "loss/crossentropy": 2.3032894134521484,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25082532316446304,
+      "step": 8178
+    },
+    {
+      "epoch": 0.1636,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.010931142171223958,
+      "learning_rate": 0.0001,
+      "loss": 4.3375,
+      "loss/crossentropy": 2.0634626150131226,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23546822369098663,
+      "step": 8180
+    },
+    {
+      "epoch": 0.16364,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.012737782796223958,
+      "learning_rate": 0.0001,
+      "loss": 4.2402,
+      "loss/crossentropy": 1.7406468391418457,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20627902448177338,
+      "step": 8182
+    },
+    {
+      "epoch": 0.16368,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.012668609619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.4103,
+      "loss/crossentropy": 2.3812272548675537,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25258868932724,
+      "step": 8184
+    },
+    {
+      "epoch": 0.16372,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.012499745686848958,
+      "learning_rate": 0.0001,
+      "loss": 4.3584,
+      "loss/crossentropy": 2.20754611492157,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21015550196170807,
+      "step": 8186
+    },
+    {
+      "epoch": 0.16376,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0140777587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1902,
+      "loss/crossentropy": 2.081672966480255,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19573034346103668,
+      "step": 8188
+    },
+    {
+      "epoch": 0.1638,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.009905751546223958,
+      "learning_rate": 0.0001,
+      "loss": 4.5267,
+      "loss/crossentropy": 2.184974491596222,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2477174997329712,
+      "step": 8190
+    },
+    {
+      "epoch": 0.16384,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010322825113932291,
+      "learning_rate": 0.0001,
+      "loss": 4.3804,
+      "loss/crossentropy": 2.1048192977905273,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22900478541851044,
+      "step": 8192
+    },
+    {
+      "epoch": 0.16388,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.010135650634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1068,
+      "loss/crossentropy": 1.960956335067749,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23504969477653503,
+      "step": 8194
+    },
+    {
+      "epoch": 0.16392,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.017114003499348957,
+      "learning_rate": 0.0001,
+      "loss": 4.7698,
+      "loss/crossentropy": 2.158856213092804,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2200375646352768,
+      "step": 8196
+    },
+    {
+      "epoch": 0.16396,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.017286936442057293,
+      "learning_rate": 0.0001,
+      "loss": 4.0533,
+      "loss/crossentropy": 2.0954058170318604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22729048877954483,
+      "step": 8198
+    },
+    {
+      "epoch": 0.164,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.01599299112955729,
+      "learning_rate": 0.0001,
+      "loss": 4.3492,
+      "loss/crossentropy": 2.1452964544296265,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21412881463766098,
+      "step": 8200
+    },
+    {
+      "epoch": 0.16404,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.02072321573893229,
+      "learning_rate": 0.0001,
+      "loss": 4.0204,
+      "loss/crossentropy": 1.9737866520881653,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19449186325073242,
+      "step": 8202
+    },
+    {
+      "epoch": 0.16408,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.020344034830729166,
+      "learning_rate": 0.0001,
+      "loss": 4.5119,
+      "loss/crossentropy": 2.213072657585144,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2333887815475464,
+      "step": 8204
+    },
+    {
+      "epoch": 0.16412,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0203277587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2139,
+      "loss/crossentropy": 1.8702161312103271,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21368274092674255,
+      "step": 8206
+    },
+    {
+      "epoch": 0.16416,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.02102635701497396,
+      "learning_rate": 0.0001,
+      "loss": 4.242,
+      "loss/crossentropy": 2.177275776863098,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21248809248209,
+      "step": 8208
+    },
+    {
+      "epoch": 0.1642,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.021945953369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3151,
+      "loss/crossentropy": 2.3422038555145264,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2288341298699379,
+      "step": 8210
+    },
+    {
+      "epoch": 0.16424,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.014212799072265626,
+      "learning_rate": 0.0001,
+      "loss": 4.2195,
+      "loss/crossentropy": 2.094432234764099,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2224871665239334,
+      "step": 8212
+    },
+    {
+      "epoch": 0.16428,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.020072174072265626,
+      "learning_rate": 0.0001,
+      "loss": 4.5045,
+      "loss/crossentropy": 2.3371682167053223,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2710718363523483,
+      "step": 8214
+    },
+    {
+      "epoch": 0.16432,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.034242502848307294,
+      "learning_rate": 0.0001,
+      "loss": 4.6926,
+      "loss/crossentropy": 1.8700988292694092,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20795634388923645,
+      "step": 8216
+    },
+    {
+      "epoch": 0.16436,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.02575658162434896,
+      "learning_rate": 0.0001,
+      "loss": 4.3266,
+      "loss/crossentropy": 2.080985188484192,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23497651517391205,
+      "step": 8218
+    },
+    {
+      "epoch": 0.1644,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.026364898681640624,
+      "learning_rate": 0.0001,
+      "loss": 4.3873,
+      "loss/crossentropy": 2.3486984968185425,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25393396615982056,
+      "step": 8220
+    },
+    {
+      "epoch": 0.16444,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.028148396809895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3446,
+      "loss/crossentropy": 2.0084245800971985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21738936007022858,
+      "step": 8222
+    },
+    {
+      "epoch": 0.16448,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.02846247355143229,
+      "learning_rate": 0.0001,
+      "loss": 3.8584,
+      "loss/crossentropy": 1.536482572555542,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17787320166826248,
+      "step": 8224
+    },
+    {
+      "epoch": 0.16452,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.024857330322265624,
+      "learning_rate": 0.0001,
+      "loss": 4.494,
+      "loss/crossentropy": 2.1727080941200256,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22967493534088135,
+      "step": 8226
+    },
+    {
+      "epoch": 0.16456,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.02490208943684896,
+      "learning_rate": 0.0001,
+      "loss": 4.1895,
+      "loss/crossentropy": 1.6915069222450256,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19113170355558395,
+      "step": 8228
+    },
+    {
+      "epoch": 0.1646,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.022739410400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.4378,
+      "loss/crossentropy": 2.1645957231521606,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21431444585323334,
+      "step": 8230
+    },
+    {
+      "epoch": 0.16464,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.00858154296875,
+      "learning_rate": 0.0001,
+      "loss": 3.8088,
+      "loss/crossentropy": 1.8797736763954163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19344569742679596,
+      "step": 8232
+    },
+    {
+      "epoch": 0.16468,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0148345947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3119,
+      "loss/crossentropy": 2.092893421649933,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21476523578166962,
+      "step": 8234
+    },
+    {
+      "epoch": 0.16472,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0143310546875,
+      "learning_rate": 0.0001,
+      "loss": 4.4145,
+      "loss/crossentropy": 2.336071252822876,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25474119186401367,
+      "step": 8236
+    },
+    {
+      "epoch": 0.16476,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.014410146077473958,
+      "learning_rate": 0.0001,
+      "loss": 4.3054,
+      "loss/crossentropy": 2.008872926235199,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22672003507614136,
+      "step": 8238
+    },
+    {
+      "epoch": 0.1648,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.011822255452473958,
+      "learning_rate": 0.0001,
+      "loss": 4.4997,
+      "loss/crossentropy": 2.2229605317115784,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2392737865447998,
+      "step": 8240
+    },
+    {
+      "epoch": 0.16484,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.011525217692057292,
+      "learning_rate": 0.0001,
+      "loss": 4.5894,
+      "loss/crossentropy": 2.1929808855056763,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22832375764846802,
+      "step": 8242
+    },
+    {
+      "epoch": 0.16488,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.010625966389973958,
+      "learning_rate": 0.0001,
+      "loss": 4.642,
+      "loss/crossentropy": 2.1557860374450684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24640469253063202,
+      "step": 8244
+    },
+    {
+      "epoch": 0.16492,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009582265218098959,
+      "learning_rate": 0.0001,
+      "loss": 4.2508,
+      "loss/crossentropy": 2.2462236881256104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23639583587646484,
+      "step": 8246
+    },
+    {
+      "epoch": 0.16496,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009501139322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.1476,
+      "loss/crossentropy": 2.1238350868225098,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23066531121730804,
+      "step": 8248
+    },
+    {
+      "epoch": 0.165,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0049479166666666664,
+      "learning_rate": 0.0001,
+      "loss": 4.1666,
+      "loss/crossentropy": 1.6866248846054077,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20677632093429565,
+      "step": 8250
+    },
+    {
+      "epoch": 0.16504,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.005338541666666667,
+      "learning_rate": 0.0001,
+      "loss": 4.1494,
+      "loss/crossentropy": 2.0640709400177,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2148914858698845,
+      "step": 8252
+    },
+    {
+      "epoch": 0.16508,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.00513916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.4138,
+      "loss/crossentropy": 2.003119468688965,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21817681193351746,
+      "step": 8254
+    },
+    {
+      "epoch": 0.16512,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.007763671875,
+      "learning_rate": 0.0001,
+      "loss": 4.276,
+      "loss/crossentropy": 2.053581953048706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22257455438375473,
+      "step": 8256
+    },
+    {
+      "epoch": 0.16516,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.008381144205729166,
+      "learning_rate": 0.0001,
+      "loss": 4.1314,
+      "loss/crossentropy": 1.788454830646515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21449437737464905,
+      "step": 8258
+    },
+    {
+      "epoch": 0.1652,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.009032185872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3986,
+      "loss/crossentropy": 1.8791787028312683,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1877290904521942,
+      "step": 8260
+    },
+    {
+      "epoch": 0.16524,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.009761555989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.4326,
+      "loss/crossentropy": 2.2346811294555664,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.221164733171463,
+      "step": 8262
+    },
+    {
+      "epoch": 0.16528,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009227498372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.3296,
+      "loss/crossentropy": 2.1032413244247437,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21829386800527573,
+      "step": 8264
+    },
+    {
+      "epoch": 0.16532,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.008690388997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3188,
+      "loss/crossentropy": 2.1452749967575073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24733971804380417,
+      "step": 8266
+    },
+    {
+      "epoch": 0.16536,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.008373006184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.2093,
+      "loss/crossentropy": 1.8318313956260681,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22139593213796616,
+      "step": 8268
+    },
+    {
+      "epoch": 0.1654,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0074045817057291664,
+      "learning_rate": 0.0001,
+      "loss": 4.3494,
+      "loss/crossentropy": 1.9865980744361877,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22354473173618317,
+      "step": 8270
+    },
+    {
+      "epoch": 0.16544,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.005464680989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.318,
+      "loss/crossentropy": 2.140509843826294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22832269966602325,
+      "step": 8272
+    },
+    {
+      "epoch": 0.16548,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.005052693684895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1632,
+      "loss/crossentropy": 2.3763319849967957,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22866757214069366,
+      "step": 8274
+    },
+    {
+      "epoch": 0.16552,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.004198201497395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2366,
+      "loss/crossentropy": 1.785762071609497,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19961480796337128,
+      "step": 8276
+    },
+    {
+      "epoch": 0.16556,
+      "grad_norm": 7.34375,
+      "grad_norm_var": 1.7181477864583334,
+      "learning_rate": 0.0001,
+      "loss": 4.5317,
+      "loss/crossentropy": 2.0500977635383606,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2972872704267502,
+      "step": 8278
+    },
+    {
+      "epoch": 0.1656,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 1.6987589518229167,
+      "learning_rate": 0.0001,
+      "loss": 4.4247,
+      "loss/crossentropy": 2.3665153980255127,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24115745723247528,
+      "step": 8280
+    },
+    {
+      "epoch": 0.16564,
+      "grad_norm": 2.25,
+      "grad_norm_var": 1.6973592122395833,
+      "learning_rate": 0.0001,
+      "loss": 4.326,
+      "loss/crossentropy": 1.9636226892471313,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2048879787325859,
+      "step": 8282
+    },
+    {
+      "epoch": 0.16568,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 1.6948720296223958,
+      "learning_rate": 0.0001,
+      "loss": 4.4303,
+      "loss/crossentropy": 2.2624993324279785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22186043858528137,
+      "step": 8284
+    },
+    {
+      "epoch": 0.16572,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 1.7048886617024739,
+      "learning_rate": 0.0001,
+      "loss": 4.0816,
+      "loss/crossentropy": 2.098397970199585,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.219336099922657,
+      "step": 8286
+    },
+    {
+      "epoch": 0.16576,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 1.6997393290201823,
+      "learning_rate": 0.0001,
+      "loss": 4.4705,
+      "loss/crossentropy": 1.9795190691947937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2145320549607277,
+      "step": 8288
+    },
+    {
+      "epoch": 0.1658,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 1.6936927795410157,
+      "learning_rate": 0.0001,
+      "loss": 4.4625,
+      "loss/crossentropy": 2.1028788089752197,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22656689584255219,
+      "step": 8290
+    },
+    {
+      "epoch": 0.16584,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 1.7015398661295573,
+      "learning_rate": 0.0001,
+      "loss": 4.1901,
+      "loss/crossentropy": 2.2936136722564697,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23688847571611404,
+      "step": 8292
+    },
+    {
+      "epoch": 0.16588,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.01862360636393229,
+      "learning_rate": 0.0001,
+      "loss": 4.1665,
+      "loss/crossentropy": 1.870754897594452,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20914533734321594,
+      "step": 8294
+    },
+    {
+      "epoch": 0.16592,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.007999674479166666,
+      "learning_rate": 0.0001,
+      "loss": 3.9884,
+      "loss/crossentropy": 2.034530520439148,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22234197705984116,
+      "step": 8296
+    },
+    {
+      "epoch": 0.16596,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0073811848958333336,
+      "learning_rate": 0.0001,
+      "loss": 4.3674,
+      "loss/crossentropy": 1.974400520324707,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20479100942611694,
+      "step": 8298
+    },
+    {
+      "epoch": 0.166,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0078765869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3157,
+      "loss/crossentropy": 2.180828809738159,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2229994386434555,
+      "step": 8300
+    },
+    {
+      "epoch": 0.16604,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.007624308268229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0704,
+      "loss/crossentropy": 2.098487079143524,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19709115475416183,
+      "step": 8302
+    },
+    {
+      "epoch": 0.16608,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.007673136393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2292,
+      "loss/crossentropy": 1.999358892440796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20891200006008148,
+      "step": 8304
+    },
+    {
+      "epoch": 0.16612,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.009022776285807292,
+      "learning_rate": 0.0001,
+      "loss": 4.4263,
+      "loss/crossentropy": 2.40866219997406,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22355159372091293,
+      "step": 8306
+    },
+    {
+      "epoch": 0.16616,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.008156077067057291,
+      "learning_rate": 0.0001,
+      "loss": 4.0532,
+      "loss/crossentropy": 1.774325966835022,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2058887928724289,
+      "step": 8308
+    },
+    {
+      "epoch": 0.1662,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.007352447509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4951,
+      "loss/crossentropy": 1.9870773553848267,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21368569880723953,
+      "step": 8310
+    },
+    {
+      "epoch": 0.16624,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.01121826171875,
+      "learning_rate": 0.0001,
+      "loss": 4.3496,
+      "loss/crossentropy": 1.9224175810813904,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2279355376958847,
+      "step": 8312
+    },
+    {
+      "epoch": 0.16628,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.012230428059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4074,
+      "loss/crossentropy": 2.011419177055359,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2363004833459854,
+      "step": 8314
+    },
+    {
+      "epoch": 0.16632,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.01444091796875,
+      "learning_rate": 0.0001,
+      "loss": 4.6627,
+      "loss/crossentropy": 1.9777795672416687,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21830055862665176,
+      "step": 8316
+    },
+    {
+      "epoch": 0.16636,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.013508860270182292,
+      "learning_rate": 0.0001,
+      "loss": 4.3798,
+      "loss/crossentropy": 2.3334981203079224,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21089734882116318,
+      "step": 8318
+    },
+    {
+      "epoch": 0.1664,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.015148671468098958,
+      "learning_rate": 0.0001,
+      "loss": 4.3792,
+      "loss/crossentropy": 1.9137234687805176,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21954041719436646,
+      "step": 8320
+    },
+    {
+      "epoch": 0.16644,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.013492838541666666,
+      "learning_rate": 0.0001,
+      "loss": 4.3826,
+      "loss/crossentropy": 2.1358155608177185,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21321023255586624,
+      "step": 8322
+    },
+    {
+      "epoch": 0.16648,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.013688151041666667,
+      "learning_rate": 0.0001,
+      "loss": 4.3454,
+      "loss/crossentropy": 2.1747822165489197,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23682481050491333,
+      "step": 8324
+    },
+    {
+      "epoch": 0.16652,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.013505045572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.2685,
+      "loss/crossentropy": 2.2818257808685303,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24226247519254684,
+      "step": 8326
+    },
+    {
+      "epoch": 0.16656,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.010155232747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.4806,
+      "loss/crossentropy": 1.9066791534423828,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20610930025577545,
+      "step": 8328
+    },
+    {
+      "epoch": 0.1666,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.008284505208333333,
+      "learning_rate": 0.0001,
+      "loss": 4.4799,
+      "loss/crossentropy": 2.2431830763816833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22407780587673187,
+      "step": 8330
+    },
+    {
+      "epoch": 0.16664,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0067708333333333336,
+      "learning_rate": 0.0001,
+      "loss": 4.5894,
+      "loss/crossentropy": 2.569726347923279,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2647472620010376,
+      "step": 8332
+    },
+    {
+      "epoch": 0.16668,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.006371053059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.2307,
+      "loss/crossentropy": 1.8364137411117554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2113686427474022,
+      "step": 8334
+    },
+    {
+      "epoch": 0.16672,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.007968902587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0369,
+      "loss/crossentropy": 2.0510441064834595,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2268069088459015,
+      "step": 8336
+    },
+    {
+      "epoch": 0.16676,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.007968902587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2528,
+      "loss/crossentropy": 1.7306728959083557,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20341812074184418,
+      "step": 8338
+    },
+    {
+      "epoch": 0.1668,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.023361968994140624,
+      "learning_rate": 0.0001,
+      "loss": 4.8146,
+      "loss/crossentropy": 2.398088574409485,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2426001876592636,
+      "step": 8340
+    },
+    {
+      "epoch": 0.16684,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.023128000895182292,
+      "learning_rate": 0.0001,
+      "loss": 4.0659,
+      "loss/crossentropy": 2.1259487867355347,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22666773200035095,
+      "step": 8342
+    },
+    {
+      "epoch": 0.16688,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.022141265869140624,
+      "learning_rate": 0.0001,
+      "loss": 4.2857,
+      "loss/crossentropy": 1.9744665026664734,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2887374758720398,
+      "step": 8344
+    },
+    {
+      "epoch": 0.16692,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.02316869099934896,
+      "learning_rate": 0.0001,
+      "loss": 4.1989,
+      "loss/crossentropy": 1.9841225743293762,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2008478343486786,
+      "step": 8346
+    },
+    {
+      "epoch": 0.16696,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.1913469950358073,
+      "learning_rate": 0.0001,
+      "loss": 4.5208,
+      "loss/crossentropy": 2.060486137866974,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23954987525939941,
+      "step": 8348
+    },
+    {
+      "epoch": 0.167,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.19230931599934895,
+      "learning_rate": 0.0001,
+      "loss": 4.2185,
+      "loss/crossentropy": 1.9356245398521423,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1987425833940506,
+      "step": 8350
+    },
+    {
+      "epoch": 0.16704,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.1835845947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1875,
+      "loss/crossentropy": 1.9968677163124084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21396416425704956,
+      "step": 8352
+    },
+    {
+      "epoch": 0.16708,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.1820465087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1696,
+      "loss/crossentropy": 2.1678614616394043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22860293090343475,
+      "step": 8354
+    },
+    {
+      "epoch": 0.16712,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.17534891764322916,
+      "learning_rate": 0.0001,
+      "loss": 4.1761,
+      "loss/crossentropy": 1.7752264142036438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20555391907691956,
+      "step": 8356
+    },
+    {
+      "epoch": 0.16716,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.1767578125,
+      "learning_rate": 0.0001,
+      "loss": 4.2391,
+      "loss/crossentropy": 1.7073925137519836,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19325412809848785,
+      "step": 8358
+    },
+    {
+      "epoch": 0.1672,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.17827860514322916,
+      "learning_rate": 0.0001,
+      "loss": 4.7185,
+      "loss/crossentropy": 2.2200660705566406,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21834757924079895,
+      "step": 8360
+    },
+    {
+      "epoch": 0.16724,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.17594401041666666,
+      "learning_rate": 0.0001,
+      "loss": 4.3551,
+      "loss/crossentropy": 2.3598183393478394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25592371821403503,
+      "step": 8362
+    },
+    {
+      "epoch": 0.16728,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.014644368489583334,
+      "learning_rate": 0.0001,
+      "loss": 4.3429,
+      "loss/crossentropy": 2.342926025390625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2480776235461235,
+      "step": 8364
+    },
+    {
+      "epoch": 0.16732,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.014411417643229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2802,
+      "loss/crossentropy": 1.8505961894989014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21376194059848785,
+      "step": 8366
+    },
+    {
+      "epoch": 0.16736,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0146636962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4199,
+      "loss/crossentropy": 2.2064108848571777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21334534883499146,
+      "step": 8368
+    },
+    {
+      "epoch": 0.1674,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.0338043212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5344,
+      "loss/crossentropy": 2.1280709505081177,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24501197040081024,
+      "step": 8370
+    },
+    {
+      "epoch": 0.16744,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.03312886555989583,
+      "learning_rate": 0.0001,
+      "loss": 4.3382,
+      "loss/crossentropy": 2.2134695053100586,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2368277609348297,
+      "step": 8372
+    },
+    {
+      "epoch": 0.16748,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.03319905598958333,
+      "learning_rate": 0.0001,
+      "loss": 4.1197,
+      "loss/crossentropy": 1.6942040920257568,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21226602047681808,
+      "step": 8374
+    },
+    {
+      "epoch": 0.16752,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.025755818684895834,
+      "learning_rate": 0.0001,
+      "loss": 4.5004,
+      "loss/crossentropy": 2.1554355025291443,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2655039578676224,
+      "step": 8376
+    },
+    {
+      "epoch": 0.16756,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.024800618489583332,
+      "learning_rate": 0.0001,
+      "loss": 4.2446,
+      "loss/crossentropy": 1.8107115030288696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1756090670824051,
+      "step": 8378
+    },
+    {
+      "epoch": 0.1676,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.026496378580729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0745,
+      "loss/crossentropy": 1.655519425868988,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1760415881872177,
+      "step": 8380
+    },
+    {
+      "epoch": 0.16764,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.026460774739583335,
+      "learning_rate": 0.0001,
+      "loss": 4.2369,
+      "loss/crossentropy": 2.1618025302886963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21886380016803741,
+      "step": 8382
+    },
+    {
+      "epoch": 0.16768,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.028270467122395834,
+      "learning_rate": 0.0001,
+      "loss": 4.0798,
+      "loss/crossentropy": 2.043319880962372,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21368451416492462,
+      "step": 8384
+    },
+    {
+      "epoch": 0.16772,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.010724894205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4035,
+      "loss/crossentropy": 2.3043102025985718,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25452760607004166,
+      "step": 8386
+    },
+    {
+      "epoch": 0.16776,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.011546834309895834,
+      "learning_rate": 0.0001,
+      "loss": 4.5633,
+      "loss/crossentropy": 2.3627192974090576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2438303530216217,
+      "step": 8388
+    },
+    {
+      "epoch": 0.1678,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011335245768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.4339,
+      "loss/crossentropy": 2.0960012674331665,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20813053101301193,
+      "step": 8390
+    },
+    {
+      "epoch": 0.16784,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.013923136393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0401,
+      "loss/crossentropy": 1.9132550358772278,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19716795533895493,
+      "step": 8392
+    },
+    {
+      "epoch": 0.16788,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.014742024739583333,
+      "learning_rate": 0.0001,
+      "loss": 4.2371,
+      "loss/crossentropy": 2.1912059783935547,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.233934685587883,
+      "step": 8394
+    },
+    {
+      "epoch": 0.16792,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.012691243489583334,
+      "learning_rate": 0.0001,
+      "loss": 4.4501,
+      "loss/crossentropy": 2.165616512298584,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23475389927625656,
+      "step": 8396
+    },
+    {
+      "epoch": 0.16796,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.01265869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.4841,
+      "loss/crossentropy": 2.209625542163849,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24280209839344025,
+      "step": 8398
+    },
+    {
+      "epoch": 0.168,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.013606516520182292,
+      "learning_rate": 0.0001,
+      "loss": 3.8959,
+      "loss/crossentropy": 1.972103476524353,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2072778418660164,
+      "step": 8400
+    },
+    {
+      "epoch": 0.16804,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.009663645426432292,
+      "learning_rate": 0.0001,
+      "loss": 4.4646,
+      "loss/crossentropy": 2.402593731880188,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23028723895549774,
+      "step": 8402
+    },
+    {
+      "epoch": 0.16808,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.008754221598307292,
+      "learning_rate": 0.0001,
+      "loss": 4.392,
+      "loss/crossentropy": 2.218156576156616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24104444682598114,
+      "step": 8404
+    },
+    {
+      "epoch": 0.16812,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.008722941080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2713,
+      "loss/crossentropy": 1.9802079796791077,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2134247124195099,
+      "step": 8406
+    },
+    {
+      "epoch": 0.16816,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.008577219645182292,
+      "learning_rate": 0.0001,
+      "loss": 3.9765,
+      "loss/crossentropy": 2.0322983264923096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21680974960327148,
+      "step": 8408
+    },
+    {
+      "epoch": 0.1682,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008194732666015624,
+      "learning_rate": 0.0001,
+      "loss": 4.2457,
+      "loss/crossentropy": 2.11838436126709,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2319432571530342,
+      "step": 8410
+    },
+    {
+      "epoch": 0.16824,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.013398996988932292,
+      "learning_rate": 0.0001,
+      "loss": 4.2204,
+      "loss/crossentropy": 2.3801279067993164,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23981253802776337,
+      "step": 8412
+    },
+    {
+      "epoch": 0.16828,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.013042958577473958,
+      "learning_rate": 0.0001,
+      "loss": 4.2766,
+      "loss/crossentropy": 2.0953307151794434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22835668921470642,
+      "step": 8414
+    },
+    {
+      "epoch": 0.16832,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.011774698893229166,
+      "learning_rate": 0.0001,
+      "loss": 4.528,
+      "loss/crossentropy": 2.0654338598251343,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2100541964173317,
+      "step": 8416
+    },
+    {
+      "epoch": 0.16836,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.018184407552083334,
+      "learning_rate": 0.0001,
+      "loss": 4.5734,
+      "loss/crossentropy": 2.383318305015564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.246080219745636,
+      "step": 8418
+    },
+    {
+      "epoch": 0.1684,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.018318684895833333,
+      "learning_rate": 0.0001,
+      "loss": 4.2764,
+      "loss/crossentropy": 2.103544294834137,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20659767091274261,
+      "step": 8420
+    },
+    {
+      "epoch": 0.16844,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.017437489827473958,
+      "learning_rate": 0.0001,
+      "loss": 4.2737,
+      "loss/crossentropy": 2.064394950866699,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23458171635866165,
+      "step": 8422
+    },
+    {
+      "epoch": 0.16848,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.014872233072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.4133,
+      "loss/crossentropy": 2.087044835090637,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23527196049690247,
+      "step": 8424
+    },
+    {
+      "epoch": 0.16852,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0133941650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.5867,
+      "loss/crossentropy": 2.41584312915802,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2425323873758316,
+      "step": 8426
+    },
+    {
+      "epoch": 0.16856,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.010400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3292,
+      "loss/crossentropy": 2.2542352080345154,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2330578714609146,
+      "step": 8428
+    },
+    {
+      "epoch": 0.1686,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.010724894205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.6628,
+      "loss/crossentropy": 2.453263282775879,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.280864879488945,
+      "step": 8430
+    },
+    {
+      "epoch": 0.16864,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.01197509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0443,
+      "loss/crossentropy": 1.9185429811477661,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19230744242668152,
+      "step": 8432
+    },
+    {
+      "epoch": 0.16868,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.005956013997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1683,
+      "loss/crossentropy": 2.020436644554138,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21756108105182648,
+      "step": 8434
+    },
+    {
+      "epoch": 0.16872,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.007616170247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.7727,
+      "loss/crossentropy": 2.3050636053085327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24950820207595825,
+      "step": 8436
+    },
+    {
+      "epoch": 0.16876,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.01510009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4091,
+      "loss/crossentropy": 2.2787232398986816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24267014116048813,
+      "step": 8438
+    },
+    {
+      "epoch": 0.1688,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.015327962239583333,
+      "learning_rate": 0.0001,
+      "loss": 4.2927,
+      "loss/crossentropy": 2.185176372528076,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2157401591539383,
+      "step": 8440
+    },
+    {
+      "epoch": 0.16884,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.01558837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2087,
+      "loss/crossentropy": 2.0673694610595703,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23145683109760284,
+      "step": 8442
+    },
+    {
+      "epoch": 0.16888,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.01754150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0229,
+      "loss/crossentropy": 1.9011740684509277,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21629629284143448,
+      "step": 8444
+    },
+    {
+      "epoch": 0.16892,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.018651326497395832,
+      "learning_rate": 0.0001,
+      "loss": 4.4821,
+      "loss/crossentropy": 2.055977463722229,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2275683432817459,
+      "step": 8446
+    },
+    {
+      "epoch": 0.16896,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0171295166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2152,
+      "loss/crossentropy": 1.752756416797638,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19437911361455917,
+      "step": 8448
+    },
+    {
+      "epoch": 0.169,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.015706380208333332,
+      "learning_rate": 0.0001,
+      "loss": 4.3631,
+      "loss/crossentropy": 2.479012131690979,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23593680560588837,
+      "step": 8450
+    },
+    {
+      "epoch": 0.16904,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.015523274739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.3245,
+      "loss/crossentropy": 2.065472185611725,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2148493528366089,
+      "step": 8452
+    },
+    {
+      "epoch": 0.16908,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.015901692708333335,
+      "learning_rate": 0.0001,
+      "loss": 4.3955,
+      "loss/crossentropy": 2.214062213897705,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25444991141557693,
+      "step": 8454
+    },
+    {
+      "epoch": 0.16912,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.0155426025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2059,
+      "loss/crossentropy": 1.7410383224487305,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21134207397699356,
+      "step": 8456
+    },
+    {
+      "epoch": 0.16916,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.020542144775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.196,
+      "loss/crossentropy": 1.9303107857704163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18110749125480652,
+      "step": 8458
+    },
+    {
+      "epoch": 0.1692,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.01810480753580729,
+      "learning_rate": 0.0001,
+      "loss": 4.3916,
+      "loss/crossentropy": 1.9091919660568237,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21320972591638565,
+      "step": 8460
+    },
+    {
+      "epoch": 0.16924,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.025099436442057293,
+      "learning_rate": 0.0001,
+      "loss": 4.4075,
+      "loss/crossentropy": 1.8021087050437927,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21294282376766205,
+      "step": 8462
+    },
+    {
+      "epoch": 0.16928,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.027186838785807292,
+      "learning_rate": 0.0001,
+      "loss": 3.97,
+      "loss/crossentropy": 2.018254518508911,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21781788766384125,
+      "step": 8464
+    },
+    {
+      "epoch": 0.16932,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.02906061808268229,
+      "learning_rate": 0.0001,
+      "loss": 4.5665,
+      "loss/crossentropy": 2.0593737959861755,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24223209917545319,
+      "step": 8466
+    },
+    {
+      "epoch": 0.16936,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.029504140218098957,
+      "learning_rate": 0.0001,
+      "loss": 4.3116,
+      "loss/crossentropy": 2.218974232673645,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22679369151592255,
+      "step": 8468
+    },
+    {
+      "epoch": 0.1694,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.022332509358723957,
+      "learning_rate": 0.0001,
+      "loss": 4.3313,
+      "loss/crossentropy": 2.182355046272278,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23690057545900345,
+      "step": 8470
+    },
+    {
+      "epoch": 0.16944,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.024930572509765624,
+      "learning_rate": 0.0001,
+      "loss": 4.6241,
+      "loss/crossentropy": 2.1669063568115234,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2458687722682953,
+      "step": 8472
+    },
+    {
+      "epoch": 0.16948,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.021100870768229165,
+      "learning_rate": 0.0001,
+      "loss": 4.3181,
+      "loss/crossentropy": 2.0656558871269226,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20421817898750305,
+      "step": 8474
+    },
+    {
+      "epoch": 0.16952,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.024625651041666665,
+      "learning_rate": 0.0001,
+      "loss": 4.4916,
+      "loss/crossentropy": 2.1470741033554077,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20798998326063156,
+      "step": 8476
+    },
+    {
+      "epoch": 0.16956,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.014972941080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2616,
+      "loss/crossentropy": 2.206741452217102,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2327529340982437,
+      "step": 8478
+    },
+    {
+      "epoch": 0.1696,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.012596638997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3338,
+      "loss/crossentropy": 2.4421777725219727,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2385794073343277,
+      "step": 8480
+    },
+    {
+      "epoch": 0.16964,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0115234375,
+      "learning_rate": 0.0001,
+      "loss": 3.963,
+      "loss/crossentropy": 1.8545736074447632,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18871797621250153,
+      "step": 8482
+    },
+    {
+      "epoch": 0.16968,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0118072509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1352,
+      "loss/crossentropy": 1.807646930217743,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19625309854745865,
+      "step": 8484
+    },
+    {
+      "epoch": 0.16972,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.012202962239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.5029,
+      "loss/crossentropy": 1.923595905303955,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2233208492398262,
+      "step": 8486
+    },
+    {
+      "epoch": 0.16976,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0080718994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3251,
+      "loss/crossentropy": 2.1018277406692505,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2556355446577072,
+      "step": 8488
+    },
+    {
+      "epoch": 0.1698,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.01099853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1127,
+      "loss/crossentropy": 1.9638542532920837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2101784572005272,
+      "step": 8490
+    },
+    {
+      "epoch": 0.16984,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.008703358968098958,
+      "learning_rate": 0.0001,
+      "loss": 3.9463,
+      "loss/crossentropy": 1.7770507335662842,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2004244327545166,
+      "step": 8492
+    },
+    {
+      "epoch": 0.16988,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.008573150634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3901,
+      "loss/crossentropy": 2.4064877033233643,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23509501665830612,
+      "step": 8494
+    },
+    {
+      "epoch": 0.16992,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.008713531494140624,
+      "learning_rate": 0.0001,
+      "loss": 4.3197,
+      "loss/crossentropy": 2.183193802833557,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22985967248678207,
+      "step": 8496
+    },
+    {
+      "epoch": 0.16996,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.009126536051432292,
+      "learning_rate": 0.0001,
+      "loss": 4.4983,
+      "loss/crossentropy": 2.241714060306549,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22633583843708038,
+      "step": 8498
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0097808837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.32,
+      "loss/crossentropy": 2.212075114250183,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20900271832942963,
+      "step": 8500
+    },
+    {
+      "epoch": 0.17004,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0071685791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1885,
+      "loss/crossentropy": 2.1283940076828003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21419794112443924,
+      "step": 8502
+    },
+    {
+      "epoch": 0.17008,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.00731201171875,
+      "learning_rate": 0.0001,
+      "loss": 4.3498,
+      "loss/crossentropy": 2.1958925127983093,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23375380039215088,
+      "step": 8504
+    },
+    {
+      "epoch": 0.17012,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0065673828125,
+      "learning_rate": 0.0001,
+      "loss": 4.3396,
+      "loss/crossentropy": 1.7883376479148865,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2097182646393776,
+      "step": 8506
+    },
+    {
+      "epoch": 0.17016,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.006811269124348958,
+      "learning_rate": 0.0001,
+      "loss": 4.5606,
+      "loss/crossentropy": 2.114001750946045,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2059268057346344,
+      "step": 8508
+    },
+    {
+      "epoch": 0.1702,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.006929270426432292,
+      "learning_rate": 0.0001,
+      "loss": 4.3178,
+      "loss/crossentropy": 2.2413129806518555,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2517316862940788,
+      "step": 8510
+    },
+    {
+      "epoch": 0.17024,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.015083567301432291,
+      "learning_rate": 0.0001,
+      "loss": 4.4202,
+      "loss/crossentropy": 2.0968031883239746,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2454070746898651,
+      "step": 8512
+    },
+    {
+      "epoch": 0.17028,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.01945978800455729,
+      "learning_rate": 0.0001,
+      "loss": 4.407,
+      "loss/crossentropy": 1.9813454151153564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21238256990909576,
+      "step": 8514
+    },
+    {
+      "epoch": 0.17032,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.019755045572916668,
+      "learning_rate": 0.0001,
+      "loss": 4.4724,
+      "loss/crossentropy": 2.261451005935669,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2393329069018364,
+      "step": 8516
+    },
+    {
+      "epoch": 0.17036,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.020654296875,
+      "learning_rate": 0.0001,
+      "loss": 4.5377,
+      "loss/crossentropy": 1.9661999344825745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2088441476225853,
+      "step": 8518
+    },
+    {
+      "epoch": 0.1704,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.022163899739583333,
+      "learning_rate": 0.0001,
+      "loss": 3.9288,
+      "loss/crossentropy": 1.8275291323661804,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20091407746076584,
+      "step": 8520
+    },
+    {
+      "epoch": 0.17044,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.023346964518229166,
+      "learning_rate": 0.0001,
+      "loss": 4.3154,
+      "loss/crossentropy": 2.266944646835327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2514675632119179,
+      "step": 8522
+    },
+    {
+      "epoch": 0.17048,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.02271728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3571,
+      "loss/crossentropy": 2.244120240211487,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25534868240356445,
+      "step": 8524
+    },
+    {
+      "epoch": 0.17052,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.025048828125,
+      "learning_rate": 0.0001,
+      "loss": 4.3037,
+      "loss/crossentropy": 1.8628552556037903,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21675898134708405,
+      "step": 8526
+    },
+    {
+      "epoch": 0.17056,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.01597900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3883,
+      "loss/crossentropy": 1.971808135509491,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2143295481801033,
+      "step": 8528
+    },
+    {
+      "epoch": 0.1706,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0114410400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3481,
+      "loss/crossentropy": 1.959191381931305,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24014096707105637,
+      "step": 8530
+    },
+    {
+      "epoch": 0.17064,
+      "grad_norm": 3.0,
+      "grad_norm_var": 0.05729878743489583,
+      "learning_rate": 0.0001,
+      "loss": 4.2633,
+      "loss/crossentropy": 1.836454451084137,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20957449078559875,
+      "step": 8532
+    },
+    {
+      "epoch": 0.17068,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.06005452473958333,
+      "learning_rate": 0.0001,
+      "loss": 4.1254,
+      "loss/crossentropy": 2.003947675228119,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23538918048143387,
+      "step": 8534
+    },
+    {
+      "epoch": 0.17072,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.05788472493489583,
+      "learning_rate": 0.0001,
+      "loss": 4.4412,
+      "loss/crossentropy": 2.0154194831848145,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2278473973274231,
+      "step": 8536
+    },
+    {
+      "epoch": 0.17076,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.05548502604166667,
+      "learning_rate": 0.0001,
+      "loss": 4.2426,
+      "loss/crossentropy": 2.1560275554656982,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21601636707782745,
+      "step": 8538
+    },
+    {
+      "epoch": 0.1708,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.05689697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2258,
+      "loss/crossentropy": 2.1494773626327515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22883395850658417,
+      "step": 8540
+    },
+    {
+      "epoch": 0.17084,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.05607808430989583,
+      "learning_rate": 0.0001,
+      "loss": 4.5201,
+      "loss/crossentropy": 2.20908784866333,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26363062113523483,
+      "step": 8542
+    },
+    {
+      "epoch": 0.17088,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0566802978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2477,
+      "loss/crossentropy": 1.7932087182998657,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18538028001785278,
+      "step": 8544
+    },
+    {
+      "epoch": 0.17092,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.05625,
+      "learning_rate": 0.0001,
+      "loss": 4.471,
+      "loss/crossentropy": 2.069899260997772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22498925775289536,
+      "step": 8546
+    },
+    {
+      "epoch": 0.17096,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.019169108072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.474,
+      "loss/crossentropy": 2.008604884147644,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21895557641983032,
+      "step": 8548
+    },
+    {
+      "epoch": 0.171,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.018798828125,
+      "learning_rate": 0.0001,
+      "loss": 4.1973,
+      "loss/crossentropy": 2.0652626156806946,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23341115564107895,
+      "step": 8550
+    },
+    {
+      "epoch": 0.17104,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.015607706705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.383,
+      "loss/crossentropy": 1.9780349135398865,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22891747951507568,
+      "step": 8552
+    },
+    {
+      "epoch": 0.17108,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.022980753580729166,
+      "learning_rate": 0.0001,
+      "loss": 4.4424,
+      "loss/crossentropy": 2.163089871406555,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2447328343987465,
+      "step": 8554
+    },
+    {
+      "epoch": 0.17112,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.022945149739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.1503,
+      "loss/crossentropy": 2.2946064472198486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24353116750717163,
+      "step": 8556
+    },
+    {
+      "epoch": 0.17116,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.022459920247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.3049,
+      "loss/crossentropy": 2.0238161087036133,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2153262495994568,
+      "step": 8558
+    },
+    {
+      "epoch": 0.1712,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.02604955037434896,
+      "learning_rate": 0.0001,
+      "loss": 3.8636,
+      "loss/crossentropy": 1.5683120489120483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17940818518400192,
+      "step": 8560
+    },
+    {
+      "epoch": 0.17124,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.027337392171223957,
+      "learning_rate": 0.0001,
+      "loss": 4.3867,
+      "loss/crossentropy": 1.9956589937210083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22829821705818176,
+      "step": 8562
+    },
+    {
+      "epoch": 0.17128,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.036834462483723955,
+      "learning_rate": 0.0001,
+      "loss": 4.7055,
+      "loss/crossentropy": 2.2242285013198853,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23567892611026764,
+      "step": 8564
+    },
+    {
+      "epoch": 0.17132,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.033719635009765624,
+      "learning_rate": 0.0001,
+      "loss": 4.274,
+      "loss/crossentropy": 2.2143776416778564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2457498088479042,
+      "step": 8566
+    },
+    {
+      "epoch": 0.17136,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.033782704671223955,
+      "learning_rate": 0.0001,
+      "loss": 4.354,
+      "loss/crossentropy": 1.852292537689209,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24004538357257843,
+      "step": 8568
+    },
+    {
+      "epoch": 0.1714,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.02676976521809896,
+      "learning_rate": 0.0001,
+      "loss": 4.4429,
+      "loss/crossentropy": 2.311089515686035,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22059186547994614,
+      "step": 8570
+    },
+    {
+      "epoch": 0.17144,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.027675120035807292,
+      "learning_rate": 0.0001,
+      "loss": 4.1697,
+      "loss/crossentropy": 1.9324169754981995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21051711589097977,
+      "step": 8572
+    },
+    {
+      "epoch": 0.17148,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.02904052734375,
+      "learning_rate": 0.0001,
+      "loss": 4.2457,
+      "loss/crossentropy": 1.982999861240387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2084646075963974,
+      "step": 8574
+    },
+    {
+      "epoch": 0.17152,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.02505671183268229,
+      "learning_rate": 0.0001,
+      "loss": 4.2503,
+      "loss/crossentropy": 2.1837204694747925,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22953644394874573,
+      "step": 8576
+    },
+    {
+      "epoch": 0.17156,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.024102528889973957,
+      "learning_rate": 0.0001,
+      "loss": 4.4448,
+      "loss/crossentropy": 2.2588841319084167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22851599752902985,
+      "step": 8578
+    },
+    {
+      "epoch": 0.1716,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.011502838134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2033,
+      "loss/crossentropy": 1.982733964920044,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21284686028957367,
+      "step": 8580
+    },
+    {
+      "epoch": 0.17164,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.012143707275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1238,
+      "loss/crossentropy": 2.443873167037964,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23966002464294434,
+      "step": 8582
+    },
+    {
+      "epoch": 0.17168,
+      "grad_norm": 5.09375,
+      "grad_norm_var": 0.5580645243326823,
+      "learning_rate": 0.0001,
+      "loss": 4.2654,
+      "loss/crossentropy": 2.462417483329773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2341725453734398,
+      "step": 8584
+    },
+    {
+      "epoch": 0.17172,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.5521705627441407,
+      "learning_rate": 0.0001,
+      "loss": 4.183,
+      "loss/crossentropy": 1.8569464683532715,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.224809430539608,
+      "step": 8586
+    },
+    {
+      "epoch": 0.17176,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.55804443359375,
+      "learning_rate": 0.0001,
+      "loss": 4.2762,
+      "loss/crossentropy": 2.017254650592804,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2259984165430069,
+      "step": 8588
+    },
+    {
+      "epoch": 0.1718,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.550066884358724,
+      "learning_rate": 0.0001,
+      "loss": 4.463,
+      "loss/crossentropy": 1.9804525971412659,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20844170451164246,
+      "step": 8590
+    },
+    {
+      "epoch": 0.17184,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.544781239827474,
+      "learning_rate": 0.0001,
+      "loss": 4.2676,
+      "loss/crossentropy": 2.2506592869758606,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22404606640338898,
+      "step": 8592
+    },
+    {
+      "epoch": 0.17188,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.5579335530598958,
+      "learning_rate": 0.0001,
+      "loss": 4.197,
+      "loss/crossentropy": 1.9515153765678406,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.216490276157856,
+      "step": 8594
+    },
+    {
+      "epoch": 0.17192,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.5598052978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4037,
+      "loss/crossentropy": 2.090883791446686,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22496677190065384,
+      "step": 8596
+    },
+    {
+      "epoch": 0.17196,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.5523844401041667,
+      "learning_rate": 0.0001,
+      "loss": 4.8429,
+      "loss/crossentropy": 2.4344359636306763,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2748369127511978,
+      "step": 8598
+    },
+    {
+      "epoch": 0.172,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.05840250651041667,
+      "learning_rate": 0.0001,
+      "loss": 4.4359,
+      "loss/crossentropy": 1.9280555844306946,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20140594244003296,
+      "step": 8600
+    },
+    {
+      "epoch": 0.17204,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.05537007649739583,
+      "learning_rate": 0.0001,
+      "loss": 4.2802,
+      "loss/crossentropy": 2.045006573200226,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21741003543138504,
+      "step": 8602
+    },
+    {
+      "epoch": 0.17208,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.04461034138997396,
+      "learning_rate": 0.0001,
+      "loss": 4.1622,
+      "loss/crossentropy": 2.092265546321869,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20971956849098206,
+      "step": 8604
+    },
+    {
+      "epoch": 0.17212,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.048130035400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.16,
+      "loss/crossentropy": 1.794768512248993,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20680297911167145,
+      "step": 8606
+    },
+    {
+      "epoch": 0.17216,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.05690078735351563,
+      "learning_rate": 0.0001,
+      "loss": 4.2213,
+      "loss/crossentropy": 1.9316805601119995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2163936346769333,
+      "step": 8608
+    },
+    {
+      "epoch": 0.1722,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.05872294108072917,
+      "learning_rate": 0.0001,
+      "loss": 4.5981,
+      "loss/crossentropy": 2.2786675691604614,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2411205694079399,
+      "step": 8610
+    },
+    {
+      "epoch": 0.17224,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.05852457682291667,
+      "learning_rate": 0.0001,
+      "loss": 4.5688,
+      "loss/crossentropy": 1.9211469888687134,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23195043951272964,
+      "step": 8612
+    },
+    {
+      "epoch": 0.17228,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.037393951416015626,
+      "learning_rate": 0.0001,
+      "loss": 4.1776,
+      "loss/crossentropy": 1.900360643863678,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2339140772819519,
+      "step": 8614
+    },
+    {
+      "epoch": 0.17232,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.03743464152018229,
+      "learning_rate": 0.0001,
+      "loss": 4.4408,
+      "loss/crossentropy": 1.976994514465332,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.218951515853405,
+      "step": 8616
+    },
+    {
+      "epoch": 0.17236,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.035982004801432294,
+      "learning_rate": 0.0001,
+      "loss": 4.1988,
+      "loss/crossentropy": 2.045244038105011,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20886047929525375,
+      "step": 8618
+    },
+    {
+      "epoch": 0.1724,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.03240534464518229,
+      "learning_rate": 0.0001,
+      "loss": 4.5082,
+      "loss/crossentropy": 2.21256685256958,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21809116005897522,
+      "step": 8620
+    },
+    {
+      "epoch": 0.17244,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.02641779581705729,
+      "learning_rate": 0.0001,
+      "loss": 4.6127,
+      "loss/crossentropy": 2.300337314605713,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22060109674930573,
+      "step": 8622
+    },
+    {
+      "epoch": 0.17248,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.016932932535807292,
+      "learning_rate": 0.0001,
+      "loss": 4.2411,
+      "loss/crossentropy": 1.8734883666038513,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20403072237968445,
+      "step": 8624
+    },
+    {
+      "epoch": 0.17252,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.01962865193684896,
+      "learning_rate": 0.0001,
+      "loss": 4.6907,
+      "loss/crossentropy": 2.1382813453674316,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27397096157073975,
+      "step": 8626
+    },
+    {
+      "epoch": 0.17256,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.019760894775390624,
+      "learning_rate": 0.0001,
+      "loss": 4.0917,
+      "loss/crossentropy": 1.9718505144119263,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20658842474222183,
+      "step": 8628
+    },
+    {
+      "epoch": 0.1726,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.016706339518229165,
+      "learning_rate": 0.0001,
+      "loss": 4.5233,
+      "loss/crossentropy": 2.0957319736480713,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2187560573220253,
+      "step": 8630
+    },
+    {
+      "epoch": 0.17264,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.019954427083333334,
+      "learning_rate": 0.0001,
+      "loss": 4.0986,
+      "loss/crossentropy": 2.0504234433174133,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22376062721014023,
+      "step": 8632
+    },
+    {
+      "epoch": 0.17268,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.05233968098958333,
+      "learning_rate": 0.0001,
+      "loss": 4.4513,
+      "loss/crossentropy": 2.057171046733856,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23209689557552338,
+      "step": 8634
+    },
+    {
+      "epoch": 0.17272,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.052611287434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3473,
+      "loss/crossentropy": 1.9635317921638489,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21854296326637268,
+      "step": 8636
+    },
+    {
+      "epoch": 0.17276,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.0546539306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3124,
+      "loss/crossentropy": 1.8973188400268555,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24171262234449387,
+      "step": 8638
+    },
+    {
+      "epoch": 0.1728,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.05347900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2068,
+      "loss/crossentropy": 1.6730469465255737,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17975886166095734,
+      "step": 8640
+    },
+    {
+      "epoch": 0.17284,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.04702123006184896,
+      "learning_rate": 0.0001,
+      "loss": 4.4222,
+      "loss/crossentropy": 2.2531689405441284,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2170827016234398,
+      "step": 8642
+    },
+    {
+      "epoch": 0.17288,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.045873769124348956,
+      "learning_rate": 0.0001,
+      "loss": 4.0249,
+      "loss/crossentropy": 2.0913639068603516,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22586838155984879,
+      "step": 8644
+    },
+    {
+      "epoch": 0.17292,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.04533869425455729,
+      "learning_rate": 0.0001,
+      "loss": 4.45,
+      "loss/crossentropy": 2.163489580154419,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2290281057357788,
+      "step": 8646
+    },
+    {
+      "epoch": 0.17296,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.04267552693684896,
+      "learning_rate": 0.0001,
+      "loss": 4.3198,
+      "loss/crossentropy": 2.0669034719467163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22411519289016724,
+      "step": 8648
+    },
+    {
+      "epoch": 0.173,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.008213043212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0474,
+      "loss/crossentropy": 1.9942336678504944,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21262076497077942,
+      "step": 8650
+    },
+    {
+      "epoch": 0.17304,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.009474436442057291,
+      "learning_rate": 0.0001,
+      "loss": 4.2701,
+      "loss/crossentropy": 2.046514868736267,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2376151606440544,
+      "step": 8652
+    },
+    {
+      "epoch": 0.17308,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.005995432535807292,
+      "learning_rate": 0.0001,
+      "loss": 4.3308,
+      "loss/crossentropy": 1.8385429382324219,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19642101973295212,
+      "step": 8654
+    },
+    {
+      "epoch": 0.17312,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.007252756754557292,
+      "learning_rate": 0.0001,
+      "loss": 4.446,
+      "loss/crossentropy": 2.259633481502533,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24319174885749817,
+      "step": 8656
+    },
+    {
+      "epoch": 0.17316,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.005301920572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.2067,
+      "loss/crossentropy": 1.9811018109321594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21801364421844482,
+      "step": 8658
+    },
+    {
+      "epoch": 0.1732,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0059722900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2158,
+      "loss/crossentropy": 2.1726362705230713,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23193368315696716,
+      "step": 8660
+    },
+    {
+      "epoch": 0.17324,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.006086222330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.2587,
+      "loss/crossentropy": 1.9915854930877686,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2111910656094551,
+      "step": 8662
+    },
+    {
+      "epoch": 0.17328,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.026851399739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.5001,
+      "loss/crossentropy": 1.9651137590408325,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.242890365421772,
+      "step": 8664
+    },
+    {
+      "epoch": 0.17332,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.026008097330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3848,
+      "loss/crossentropy": 1.865262508392334,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21065659821033478,
+      "step": 8666
+    },
+    {
+      "epoch": 0.17336,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.02535400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3402,
+      "loss/crossentropy": 1.9073076248168945,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1959603875875473,
+      "step": 8668
+    },
+    {
+      "epoch": 0.1734,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.024332682291666668,
+      "learning_rate": 0.0001,
+      "loss": 4.2965,
+      "loss/crossentropy": 2.167983889579773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22176912426948547,
+      "step": 8670
+    },
+    {
+      "epoch": 0.17344,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.5460896809895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4551,
+      "loss/crossentropy": 1.7029761672019958,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.222617506980896,
+      "step": 8672
+    },
+    {
+      "epoch": 0.17348,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.5439849853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3414,
+      "loss/crossentropy": 2.053748309612274,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23398682475090027,
+      "step": 8674
+    },
+    {
+      "epoch": 0.17352,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.5408274332682291,
+      "learning_rate": 0.0001,
+      "loss": 4.7882,
+      "loss/crossentropy": 2.309812903404236,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2961876690387726,
+      "step": 8676
+    },
+    {
+      "epoch": 0.17356,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.5421953837076823,
+      "learning_rate": 0.0001,
+      "loss": 4.4416,
+      "loss/crossentropy": 2.1045809984207153,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22072184830904007,
+      "step": 8678
+    },
+    {
+      "epoch": 0.1736,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.5490435282389323,
+      "learning_rate": 0.0001,
+      "loss": 4.4437,
+      "loss/crossentropy": 2.2114070653915405,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2208957076072693,
+      "step": 8680
+    },
+    {
+      "epoch": 0.17364,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.5490435282389323,
+      "learning_rate": 0.0001,
+      "loss": 4.5931,
+      "loss/crossentropy": 2.1773669719696045,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22639526426792145,
+      "step": 8682
+    },
+    {
+      "epoch": 0.17368,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.5398272196451823,
+      "learning_rate": 0.0001,
+      "loss": 3.9397,
+      "loss/crossentropy": 1.4213417768478394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17653951048851013,
+      "step": 8684
+    },
+    {
+      "epoch": 0.17372,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.5425374348958333,
+      "learning_rate": 0.0001,
+      "loss": 4.2711,
+      "loss/crossentropy": 1.968630075454712,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2007492408156395,
+      "step": 8686
+    },
+    {
+      "epoch": 0.17376,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.030304972330729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2297,
+      "loss/crossentropy": 2.0826632976531982,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21895240992307663,
+      "step": 8688
+    },
+    {
+      "epoch": 0.1738,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.029002888997395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4188,
+      "loss/crossentropy": 2.2756701707839966,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22170037031173706,
+      "step": 8690
+    },
+    {
+      "epoch": 0.17384,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010773722330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.51,
+      "loss/crossentropy": 2.329536557197571,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24717354029417038,
+      "step": 8692
+    },
+    {
+      "epoch": 0.17388,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.010033162434895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0776,
+      "loss/crossentropy": 2.077241063117981,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.215146966278553,
+      "step": 8694
+    },
+    {
+      "epoch": 0.17392,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0091949462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3482,
+      "loss/crossentropy": 2.2363221645355225,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23428452014923096,
+      "step": 8696
+    },
+    {
+      "epoch": 0.17396,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.01024169921875,
+      "learning_rate": 0.0001,
+      "loss": 4.1321,
+      "loss/crossentropy": 2.055815279483795,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22674524784088135,
+      "step": 8698
+    },
+    {
+      "epoch": 0.174,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0098541259765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2789,
+      "loss/crossentropy": 2.205570936203003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23787499964237213,
+      "step": 8700
+    },
+    {
+      "epoch": 0.17404,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009895579020182291,
+      "learning_rate": 0.0001,
+      "loss": 4.4636,
+      "loss/crossentropy": 2.262540578842163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23921719938516617,
+      "step": 8702
+    },
+    {
+      "epoch": 0.17408,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.010114542643229167,
+      "learning_rate": 0.0001,
+      "loss": 4.3372,
+      "loss/crossentropy": 2.5464816093444824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23302219063043594,
+      "step": 8704
+    },
+    {
+      "epoch": 0.17412,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.00897216796875,
+      "learning_rate": 0.0001,
+      "loss": 4.2355,
+      "loss/crossentropy": 2.050383508205414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23393510282039642,
+      "step": 8706
+    },
+    {
+      "epoch": 0.17416,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0076812744140625,
+      "learning_rate": 0.0001,
+      "loss": 3.9943,
+      "loss/crossentropy": 1.9034642577171326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20801686495542526,
+      "step": 8708
+    },
+    {
+      "epoch": 0.1742,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.007045237223307291,
+      "learning_rate": 0.0001,
+      "loss": 4.2801,
+      "loss/crossentropy": 2.313044309616089,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24881915748119354,
+      "step": 8710
+    },
+    {
+      "epoch": 0.17424,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.009388987223307292,
+      "learning_rate": 0.0001,
+      "loss": 4.3754,
+      "loss/crossentropy": 1.973829746246338,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21400053054094315,
+      "step": 8712
+    },
+    {
+      "epoch": 0.17428,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.010109202067057291,
+      "learning_rate": 0.0001,
+      "loss": 4.2936,
+      "loss/crossentropy": 1.831783950328827,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20855721831321716,
+      "step": 8714
+    },
+    {
+      "epoch": 0.17432,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.009683990478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2681,
+      "loss/crossentropy": 2.0173734426498413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22365443408489227,
+      "step": 8716
+    },
+    {
+      "epoch": 0.17436,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.007155100504557292,
+      "learning_rate": 0.0001,
+      "loss": 4.4177,
+      "loss/crossentropy": 1.6534234285354614,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20528900623321533,
+      "step": 8718
+    },
+    {
+      "epoch": 0.1744,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.006396484375,
+      "learning_rate": 0.0001,
+      "loss": 4.3658,
+      "loss/crossentropy": 1.8113531470298767,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21433213353157043,
+      "step": 8720
+    },
+    {
+      "epoch": 0.17444,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0064280192057291664,
+      "learning_rate": 0.0001,
+      "loss": 4.5135,
+      "loss/crossentropy": 2.0750836730003357,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22202756255865097,
+      "step": 8722
+    },
+    {
+      "epoch": 0.17448,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0056955973307291664,
+      "learning_rate": 0.0001,
+      "loss": 4.2512,
+      "loss/crossentropy": 2.1388206481933594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24137140065431595,
+      "step": 8724
+    },
+    {
+      "epoch": 0.17452,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.006745402018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0401,
+      "loss/crossentropy": 2.068696141242981,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2215234711766243,
+      "step": 8726
+    },
+    {
+      "epoch": 0.17456,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.005280558268229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0277,
+      "loss/crossentropy": 1.6970900893211365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21108710020780563,
+      "step": 8728
+    },
+    {
+      "epoch": 0.1746,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.005582682291666667,
+      "learning_rate": 0.0001,
+      "loss": 3.8555,
+      "loss/crossentropy": 1.8847576975822449,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2151619866490364,
+      "step": 8730
+    },
+    {
+      "epoch": 0.17464,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.009403483072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.4088,
+      "loss/crossentropy": 2.4103721380233765,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2589537426829338,
+      "step": 8732
+    },
+    {
+      "epoch": 0.17468,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.01011962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1415,
+      "loss/crossentropy": 1.8340824842453003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22282177209854126,
+      "step": 8734
+    },
+    {
+      "epoch": 0.17472,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010302734375,
+      "learning_rate": 0.0001,
+      "loss": 4.2472,
+      "loss/crossentropy": 1.88236665725708,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2193661779165268,
+      "step": 8736
+    },
+    {
+      "epoch": 0.17476,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.05056050618489583,
+      "learning_rate": 0.0001,
+      "loss": 4.441,
+      "loss/crossentropy": 1.8121293783187866,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21855003386735916,
+      "step": 8738
+    },
+    {
+      "epoch": 0.1748,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.05090738932291667,
+      "learning_rate": 0.0001,
+      "loss": 4.1422,
+      "loss/crossentropy": 2.215694308280945,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2405528798699379,
+      "step": 8740
+    },
+    {
+      "epoch": 0.17484,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.05090738932291667,
+      "learning_rate": 0.0001,
+      "loss": 4.2431,
+      "loss/crossentropy": 2.124837279319763,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22565175592899323,
+      "step": 8742
+    },
+    {
+      "epoch": 0.17488,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0476959228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.6783,
+      "loss/crossentropy": 2.2531429529190063,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.259593665599823,
+      "step": 8744
+    },
+    {
+      "epoch": 0.17492,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.04537760416666667,
+      "learning_rate": 0.0001,
+      "loss": 4.3546,
+      "loss/crossentropy": 2.403178572654724,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2475409209728241,
+      "step": 8746
+    },
+    {
+      "epoch": 0.17496,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.04527587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5698,
+      "loss/crossentropy": 1.7886858582496643,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21677076816558838,
+      "step": 8748
+    },
+    {
+      "epoch": 0.175,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.04397379557291667,
+      "learning_rate": 0.0001,
+      "loss": 4.1374,
+      "loss/crossentropy": 1.9257569313049316,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2098483294248581,
+      "step": 8750
+    },
+    {
+      "epoch": 0.17504,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.04396870930989583,
+      "learning_rate": 0.0001,
+      "loss": 4.407,
+      "loss/crossentropy": 2.1609140634536743,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22846446931362152,
+      "step": 8752
+    },
+    {
+      "epoch": 0.17508,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0081939697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2683,
+      "loss/crossentropy": 2.529700756072998,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2515157088637352,
+      "step": 8754
+    },
+    {
+      "epoch": 0.17512,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.008548736572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4965,
+      "loss/crossentropy": 2.1920565366744995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24070476740598679,
+      "step": 8756
+    },
+    {
+      "epoch": 0.17516,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0097900390625,
+      "learning_rate": 0.0001,
+      "loss": 3.8416,
+      "loss/crossentropy": 1.7714558839797974,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1787928193807602,
+      "step": 8758
+    },
+    {
+      "epoch": 0.1752,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.009666951497395833,
+      "learning_rate": 0.0001,
+      "loss": 4.4457,
+      "loss/crossentropy": 1.986818790435791,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20736730098724365,
+      "step": 8760
+    },
+    {
+      "epoch": 0.17524,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010453287760416667,
+      "learning_rate": 0.0001,
+      "loss": 4.2731,
+      "loss/crossentropy": 1.8152282238006592,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22459527850151062,
+      "step": 8762
+    },
+    {
+      "epoch": 0.17528,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0069163004557291664,
+      "learning_rate": 0.0001,
+      "loss": 4.4058,
+      "loss/crossentropy": 2.2312777042388916,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23017627000808716,
+      "step": 8764
+    },
+    {
+      "epoch": 0.17532,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.013252766927083333,
+      "learning_rate": 0.0001,
+      "loss": 4.4714,
+      "loss/crossentropy": 2.107849955558777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21224269270896912,
+      "step": 8766
+    },
+    {
+      "epoch": 0.17536,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.013206990559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1641,
+      "loss/crossentropy": 2.1588711738586426,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2200954109430313,
+      "step": 8768
+    },
+    {
+      "epoch": 0.1754,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0130126953125,
+      "learning_rate": 0.0001,
+      "loss": 4.4056,
+      "loss/crossentropy": 2.1355313062667847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24048195779323578,
+      "step": 8770
+    },
+    {
+      "epoch": 0.17544,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.011557769775390626,
+      "learning_rate": 0.0001,
+      "loss": 4.3505,
+      "loss/crossentropy": 2.477591037750244,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24459562450647354,
+      "step": 8772
+    },
+    {
+      "epoch": 0.17548,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.00845947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4184,
+      "loss/crossentropy": 2.2577285766601562,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23038798570632935,
+      "step": 8774
+    },
+    {
+      "epoch": 0.17552,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.009813435872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2681,
+      "loss/crossentropy": 2.239536762237549,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22290733456611633,
+      "step": 8776
+    },
+    {
+      "epoch": 0.17556,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.007373046875,
+      "learning_rate": 0.0001,
+      "loss": 4.4351,
+      "loss/crossentropy": 1.9139958024024963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2175864800810814,
+      "step": 8778
+    },
+    {
+      "epoch": 0.1756,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.00888671875,
+      "learning_rate": 0.0001,
+      "loss": 4.0756,
+      "loss/crossentropy": 2.0622661113739014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2223067432641983,
+      "step": 8780
+    },
+    {
+      "epoch": 0.17564,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.004524739583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.4552,
+      "loss/crossentropy": 2.222475051879883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21181221306324005,
+      "step": 8782
+    },
+    {
+      "epoch": 0.17568,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.007002512613932292,
+      "learning_rate": 0.0001,
+      "loss": 4.1696,
+      "loss/crossentropy": 2.2612074613571167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24079158157110214,
+      "step": 8784
+    },
+    {
+      "epoch": 0.17572,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.005863189697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0889,
+      "loss/crossentropy": 2.1629387736320496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21437199413776398,
+      "step": 8786
+    },
+    {
+      "epoch": 0.17576,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.009492746988932292,
+      "learning_rate": 0.0001,
+      "loss": 4.6583,
+      "loss/crossentropy": 2.145151972770691,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23797442018985748,
+      "step": 8788
+    },
+    {
+      "epoch": 0.1758,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.015457916259765624,
+      "learning_rate": 0.0001,
+      "loss": 4.5188,
+      "loss/crossentropy": 2.0366984605789185,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2191104218363762,
+      "step": 8790
+    },
+    {
+      "epoch": 0.17584,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.014422353108723958,
+      "learning_rate": 0.0001,
+      "loss": 4.1604,
+      "loss/crossentropy": 2.1049715280532837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.230033777654171,
+      "step": 8792
+    },
+    {
+      "epoch": 0.17588,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.014338938395182292,
+      "learning_rate": 0.0001,
+      "loss": 4.4327,
+      "loss/crossentropy": 2.2549991607666016,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23084092140197754,
+      "step": 8794
+    },
+    {
+      "epoch": 0.17592,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.015547688802083333,
+      "learning_rate": 0.0001,
+      "loss": 4.4653,
+      "loss/crossentropy": 1.9873813390731812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19879335910081863,
+      "step": 8796
+    },
+    {
+      "epoch": 0.17596,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.015677897135416667,
+      "learning_rate": 0.0001,
+      "loss": 4.6371,
+      "loss/crossentropy": 2.0723283886909485,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20538055896759033,
+      "step": 8798
+    },
+    {
+      "epoch": 0.176,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.013099924723307291,
+      "learning_rate": 0.0001,
+      "loss": 4.0313,
+      "loss/crossentropy": 2.090642750263214,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2329270914196968,
+      "step": 8800
+    },
+    {
+      "epoch": 0.17604,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.012672678629557291,
+      "learning_rate": 0.0001,
+      "loss": 4.4626,
+      "loss/crossentropy": 2.3432271480560303,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2456662431359291,
+      "step": 8802
+    },
+    {
+      "epoch": 0.17608,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.010465240478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5133,
+      "loss/crossentropy": 2.1210837364196777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22580894827842712,
+      "step": 8804
+    },
+    {
+      "epoch": 0.17612,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.005147043863932292,
+      "learning_rate": 0.0001,
+      "loss": 4.024,
+      "loss/crossentropy": 2.142494797706604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22426588833332062,
+      "step": 8806
+    },
+    {
+      "epoch": 0.17616,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0049435933430989586,
+      "learning_rate": 0.0001,
+      "loss": 4.2877,
+      "loss/crossentropy": 1.9163227677345276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21529845893383026,
+      "step": 8808
+    },
+    {
+      "epoch": 0.1762,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.004937489827473958,
+      "learning_rate": 0.0001,
+      "loss": 4.4776,
+      "loss/crossentropy": 2.1478612422943115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2354428842663765,
+      "step": 8810
+    },
+    {
+      "epoch": 0.17624,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0033854166666666668,
+      "learning_rate": 0.0001,
+      "loss": 4.435,
+      "loss/crossentropy": 2.1546601057052612,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22553270310163498,
+      "step": 8812
+    },
+    {
+      "epoch": 0.17628,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.004233551025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1661,
+      "loss/crossentropy": 2.0559862852096558,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19714127480983734,
+      "step": 8814
+    },
+    {
+      "epoch": 0.17632,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.004078928629557292,
+      "learning_rate": 0.0001,
+      "loss": 4.3543,
+      "loss/crossentropy": 2.1340363025665283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22422882914543152,
+      "step": 8816
+    },
+    {
+      "epoch": 0.17636,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0038937886555989584,
+      "learning_rate": 0.0001,
+      "loss": 4.4464,
+      "loss/crossentropy": 2.265942335128784,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23249086737632751,
+      "step": 8818
+    },
+    {
+      "epoch": 0.1764,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0031939188639322916,
+      "learning_rate": 0.0001,
+      "loss": 4.3187,
+      "loss/crossentropy": 2.245513081550598,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22869569063186646,
+      "step": 8820
+    },
+    {
+      "epoch": 0.17644,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.006461334228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.179,
+      "loss/crossentropy": 1.851025104522705,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21989689767360687,
+      "step": 8822
+    },
+    {
+      "epoch": 0.17648,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.006266021728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.276,
+      "loss/crossentropy": 2.2972241640090942,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22793132066726685,
+      "step": 8824
+    },
+    {
+      "epoch": 0.17652,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.006276194254557292,
+      "learning_rate": 0.0001,
+      "loss": 4.3955,
+      "loss/crossentropy": 2.248735189437866,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2133597657084465,
+      "step": 8826
+    },
+    {
+      "epoch": 0.17656,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.008314768473307291,
+      "learning_rate": 0.0001,
+      "loss": 4.4423,
+      "loss/crossentropy": 2.4173099994659424,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23985996842384338,
+      "step": 8828
+    },
+    {
+      "epoch": 0.1766,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.008698527018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.5425,
+      "loss/crossentropy": 2.5017653703689575,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26119648665189743,
+      "step": 8830
+    },
+    {
+      "epoch": 0.17664,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0090484619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2474,
+      "loss/crossentropy": 1.9006813764572144,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19938994944095612,
+      "step": 8832
+    },
+    {
+      "epoch": 0.17668,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.009496053059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1643,
+      "loss/crossentropy": 2.101746916770935,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20544035732746124,
+      "step": 8834
+    },
+    {
+      "epoch": 0.17672,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.010542805989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.3211,
+      "loss/crossentropy": 2.1605160236358643,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23061934113502502,
+      "step": 8836
+    },
+    {
+      "epoch": 0.17676,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.008967081705729166,
+      "learning_rate": 0.0001,
+      "loss": 4.4357,
+      "loss/crossentropy": 1.963772177696228,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20736530423164368,
+      "step": 8838
+    },
+    {
+      "epoch": 0.1768,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.009733072916666667,
+      "learning_rate": 0.0001,
+      "loss": 4.3572,
+      "loss/crossentropy": 2.154300093650818,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22221273183822632,
+      "step": 8840
+    },
+    {
+      "epoch": 0.17684,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009501139322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.4876,
+      "loss/crossentropy": 2.1576497554779053,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21834726631641388,
+      "step": 8842
+    },
+    {
+      "epoch": 0.17688,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0115875244140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1192,
+      "loss/crossentropy": 2.1316112279891968,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2166184037923813,
+      "step": 8844
+    },
+    {
+      "epoch": 0.17692,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0098052978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2421,
+      "loss/crossentropy": 2.068525493144989,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23183748871088028,
+      "step": 8846
+    },
+    {
+      "epoch": 0.17696,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.01051025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.5128,
+      "loss/crossentropy": 2.185767650604248,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25881427526474,
+      "step": 8848
+    },
+    {
+      "epoch": 0.177,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010619099934895833,
+      "learning_rate": 0.0001,
+      "loss": 4.3073,
+      "loss/crossentropy": 1.979454517364502,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20669714361429214,
+      "step": 8850
+    },
+    {
+      "epoch": 0.17704,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009373982747395834,
+      "learning_rate": 0.0001,
+      "loss": 4.3356,
+      "loss/crossentropy": 2.3473092317581177,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23232270777225494,
+      "step": 8852
+    },
+    {
+      "epoch": 0.17708,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.007616170247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3732,
+      "loss/crossentropy": 2.461324691772461,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.242417573928833,
+      "step": 8854
+    },
+    {
+      "epoch": 0.17712,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.008385976155598959,
+      "learning_rate": 0.0001,
+      "loss": 4.1107,
+      "loss/crossentropy": 1.5953214168548584,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17768454551696777,
+      "step": 8856
+    },
+    {
+      "epoch": 0.17716,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.008377838134765624,
+      "learning_rate": 0.0001,
+      "loss": 4.4289,
+      "loss/crossentropy": 2.1969146728515625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21466109156608582,
+      "step": 8858
+    },
+    {
+      "epoch": 0.1772,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.008459218343098958,
+      "learning_rate": 0.0001,
+      "loss": 4.2154,
+      "loss/crossentropy": 2.243234634399414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22729554027318954,
+      "step": 8860
+    },
+    {
+      "epoch": 0.17724,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.010628000895182291,
+      "learning_rate": 0.0001,
+      "loss": 4.2062,
+      "loss/crossentropy": 2.1855397820472717,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24905066192150116,
+      "step": 8862
+    },
+    {
+      "epoch": 0.17728,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.010628000895182291,
+      "learning_rate": 0.0001,
+      "loss": 4.4075,
+      "loss/crossentropy": 2.320886254310608,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22135943174362183,
+      "step": 8864
+    },
+    {
+      "epoch": 0.17732,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.011736806233723958,
+      "learning_rate": 0.0001,
+      "loss": 4.1444,
+      "loss/crossentropy": 2.140891909599304,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2109990492463112,
+      "step": 8866
+    },
+    {
+      "epoch": 0.17736,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.013038889567057291,
+      "learning_rate": 0.0001,
+      "loss": 4.1009,
+      "loss/crossentropy": 2.147824764251709,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2131204530596733,
+      "step": 8868
+    },
+    {
+      "epoch": 0.1774,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.039249420166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.7136,
+      "loss/crossentropy": 2.187807321548462,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2335718423128128,
+      "step": 8870
+    },
+    {
+      "epoch": 0.17744,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.037287394205729164,
+      "learning_rate": 0.0001,
+      "loss": 4.5265,
+      "loss/crossentropy": 2.3127458095550537,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2385788857936859,
+      "step": 8872
+    },
+    {
+      "epoch": 0.17748,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0369537353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1364,
+      "loss/crossentropy": 1.859586775302887,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.204575777053833,
+      "step": 8874
+    },
+    {
+      "epoch": 0.17752,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.03585611979166667,
+      "learning_rate": 0.0001,
+      "loss": 4.1993,
+      "loss/crossentropy": 1.9626107215881348,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21016598492860794,
+      "step": 8876
+    },
+    {
+      "epoch": 0.17756,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.034077962239583336,
+      "learning_rate": 0.0001,
+      "loss": 4.462,
+      "loss/crossentropy": 2.1130539774894714,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21995095163583755,
+      "step": 8878
+    },
+    {
+      "epoch": 0.1776,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.035374959309895836,
+      "learning_rate": 0.0001,
+      "loss": 4.4677,
+      "loss/crossentropy": 1.8914743065834045,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20437531173229218,
+      "step": 8880
+    },
+    {
+      "epoch": 0.17764,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.032763671875,
+      "learning_rate": 0.0001,
+      "loss": 4.4975,
+      "loss/crossentropy": 2.2135708332061768,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2478664517402649,
+      "step": 8882
+    },
+    {
+      "epoch": 0.17768,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.028922526041666667,
+      "learning_rate": 0.0001,
+      "loss": 4.3763,
+      "loss/crossentropy": 2.194110333919525,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2366471290588379,
+      "step": 8884
+    },
+    {
+      "epoch": 0.17772,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.009822336832682292,
+      "learning_rate": 0.0001,
+      "loss": 4.2844,
+      "loss/crossentropy": 2.4365748167037964,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25842973589897156,
+      "step": 8886
+    },
+    {
+      "epoch": 0.17776,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010001373291015626,
+      "learning_rate": 0.0001,
+      "loss": 4.0574,
+      "loss/crossentropy": 1.9800177216529846,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21982619166374207,
+      "step": 8888
+    },
+    {
+      "epoch": 0.1778,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.009956614176432291,
+      "learning_rate": 0.0001,
+      "loss": 4.3465,
+      "loss/crossentropy": 2.1437748670578003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21494000405073166,
+      "step": 8890
+    },
+    {
+      "epoch": 0.17784,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.008957672119140624,
+      "learning_rate": 0.0001,
+      "loss": 4.4051,
+      "loss/crossentropy": 2.0610267519950867,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22473593801259995,
+      "step": 8892
+    },
+    {
+      "epoch": 0.17788,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.013952382405598958,
+      "learning_rate": 0.0001,
+      "loss": 4.4637,
+      "loss/crossentropy": 2.261958599090576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2250354364514351,
+      "step": 8894
+    },
+    {
+      "epoch": 0.17792,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012237294514973959,
+      "learning_rate": 0.0001,
+      "loss": 4.3508,
+      "loss/crossentropy": 2.3689773082733154,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22185539454221725,
+      "step": 8896
+    },
+    {
+      "epoch": 0.17796,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.013034820556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.7432,
+      "loss/crossentropy": 2.612341523170471,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2561237961053848,
+      "step": 8898
+    },
+    {
+      "epoch": 0.178,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010990142822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.5876,
+      "loss/crossentropy": 2.1230576038360596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23115848749876022,
+      "step": 8900
+    },
+    {
+      "epoch": 0.17804,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009626261393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2154,
+      "loss/crossentropy": 2.200004458427429,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22002413868904114,
+      "step": 8902
+    },
+    {
+      "epoch": 0.17808,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009989420572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.3438,
+      "loss/crossentropy": 2.323713779449463,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22923698276281357,
+      "step": 8904
+    },
+    {
+      "epoch": 0.17812,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.00982666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3079,
+      "loss/crossentropy": 2.038426458835602,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21465667337179184,
+      "step": 8906
+    },
+    {
+      "epoch": 0.17816,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.01217041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.4498,
+      "loss/crossentropy": 2.3639097213745117,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2378860041499138,
+      "step": 8908
+    },
+    {
+      "epoch": 0.1782,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.00699462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.261,
+      "loss/crossentropy": 1.8291080594062805,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19894887506961823,
+      "step": 8910
+    },
+    {
+      "epoch": 0.17824,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.011263020833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.5606,
+      "loss/crossentropy": 2.3113714456558228,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24526391178369522,
+      "step": 8912
+    },
+    {
+      "epoch": 0.17828,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011449178059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3016,
+      "loss/crossentropy": 2.114617943763733,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21305133402347565,
+      "step": 8914
+    },
+    {
+      "epoch": 0.17832,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.014975738525390626,
+      "learning_rate": 0.0001,
+      "loss": 3.9333,
+      "loss/crossentropy": 1.6893808841705322,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19195494800806046,
+      "step": 8916
+    },
+    {
+      "epoch": 0.17836,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.016721343994140624,
+      "learning_rate": 0.0001,
+      "loss": 4.2501,
+      "loss/crossentropy": 1.829396367073059,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21181780844926834,
+      "step": 8918
+    },
+    {
+      "epoch": 0.1784,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.01869481404622396,
+      "learning_rate": 0.0001,
+      "loss": 4.1542,
+      "loss/crossentropy": 1.8910154104232788,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20975399017333984,
+      "step": 8920
+    },
+    {
+      "epoch": 0.17844,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.01953709920247396,
+      "learning_rate": 0.0001,
+      "loss": 4.3327,
+      "loss/crossentropy": 2.0501255989074707,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22336408495903015,
+      "step": 8922
+    },
+    {
+      "epoch": 0.17848,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.01740086873372396,
+      "learning_rate": 0.0001,
+      "loss": 4.3035,
+      "loss/crossentropy": 2.3023892641067505,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23251917958259583,
+      "step": 8924
+    },
+    {
+      "epoch": 0.17852,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.01822077433268229,
+      "learning_rate": 0.0001,
+      "loss": 4.0592,
+      "loss/crossentropy": 2.0030421018600464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19997069239616394,
+      "step": 8926
+    },
+    {
+      "epoch": 0.17856,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.013741048177083333,
+      "learning_rate": 0.0001,
+      "loss": 4.2568,
+      "loss/crossentropy": 2.2309017181396484,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22145257890224457,
+      "step": 8928
+    },
+    {
+      "epoch": 0.1786,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.013728841145833334,
+      "learning_rate": 0.0001,
+      "loss": 4.4366,
+      "loss/crossentropy": 2.1135157346725464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21580957621335983,
+      "step": 8930
+    },
+    {
+      "epoch": 0.17864,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.011146799723307291,
+      "learning_rate": 0.0001,
+      "loss": 4.3473,
+      "loss/crossentropy": 2.098900556564331,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22469021379947662,
+      "step": 8932
+    },
+    {
+      "epoch": 0.17868,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0103179931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2702,
+      "loss/crossentropy": 2.1558337211608887,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24765773862600327,
+      "step": 8934
+    },
+    {
+      "epoch": 0.17872,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.006004842122395834,
+      "learning_rate": 0.0001,
+      "loss": 4.3318,
+      "loss/crossentropy": 2.141040623188019,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22490675747394562,
+      "step": 8936
+    },
+    {
+      "epoch": 0.17876,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0096099853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4126,
+      "loss/crossentropy": 2.1036806106567383,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23260055482387543,
+      "step": 8938
+    },
+    {
+      "epoch": 0.1788,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.013036092122395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4151,
+      "loss/crossentropy": 1.9403663277626038,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21941428631544113,
+      "step": 8940
+    },
+    {
+      "epoch": 0.17884,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.014717610677083333,
+      "learning_rate": 0.0001,
+      "loss": 4.2,
+      "loss/crossentropy": 1.8589079976081848,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21533852070569992,
+      "step": 8942
+    },
+    {
+      "epoch": 0.17888,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.013816070556640626,
+      "learning_rate": 0.0001,
+      "loss": 4.2476,
+      "loss/crossentropy": 2.3136903643608093,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23139026761054993,
+      "step": 8944
+    },
+    {
+      "epoch": 0.17892,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012835439046223958,
+      "learning_rate": 0.0001,
+      "loss": 4.2669,
+      "loss/crossentropy": 2.305663585662842,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24263737350702286,
+      "step": 8946
+    },
+    {
+      "epoch": 0.17896,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.02195002237955729,
+      "learning_rate": 0.0001,
+      "loss": 4.3059,
+      "loss/crossentropy": 2.0450612902641296,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2124529778957367,
+      "step": 8948
+    },
+    {
+      "epoch": 0.179,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.020393880208333333,
+      "learning_rate": 0.0001,
+      "loss": 4.4229,
+      "loss/crossentropy": 2.435065984725952,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24603386223316193,
+      "step": 8950
+    },
+    {
+      "epoch": 0.17904,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.02295099894205729,
+      "learning_rate": 0.0001,
+      "loss": 4.2604,
+      "loss/crossentropy": 2.092818021774292,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20905248820781708,
+      "step": 8952
+    },
+    {
+      "epoch": 0.17908,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.024621327718098957,
+      "learning_rate": 0.0001,
+      "loss": 4.2539,
+      "loss/crossentropy": 2.08588969707489,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2142588049173355,
+      "step": 8954
+    },
+    {
+      "epoch": 0.17912,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.021144358317057292,
+      "learning_rate": 0.0001,
+      "loss": 4.3958,
+      "loss/crossentropy": 1.9161878824234009,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20626582205295563,
+      "step": 8956
+    },
+    {
+      "epoch": 0.17916,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.020182037353515626,
+      "learning_rate": 0.0001,
+      "loss": 4.3726,
+      "loss/crossentropy": 2.3072937726974487,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22028075903654099,
+      "step": 8958
+    },
+    {
+      "epoch": 0.1792,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.021109771728515626,
+      "learning_rate": 0.0001,
+      "loss": 4.4876,
+      "loss/crossentropy": 2.3053938150405884,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22212930023670197,
+      "step": 8960
+    },
+    {
+      "epoch": 0.17924,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.02269261678059896,
+      "learning_rate": 0.0001,
+      "loss": 4.5137,
+      "loss/crossentropy": 1.9130414128303528,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20036083459854126,
+      "step": 8962
+    },
+    {
+      "epoch": 0.17928,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.013242340087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1299,
+      "loss/crossentropy": 2.3808066844940186,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2116570845246315,
+      "step": 8964
+    },
+    {
+      "epoch": 0.17932,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.009549713134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.296,
+      "loss/crossentropy": 2.180716395378113,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2275484874844551,
+      "step": 8966
+    },
+    {
+      "epoch": 0.17936,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.008837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2597,
+      "loss/crossentropy": 2.027850866317749,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2011229619383812,
+      "step": 8968
+    },
+    {
+      "epoch": 0.1794,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.017447916666666667,
+      "learning_rate": 0.0001,
+      "loss": 4.3167,
+      "loss/crossentropy": 2.077622890472412,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21361806988716125,
+      "step": 8970
+    },
+    {
+      "epoch": 0.17944,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.017513020833333334,
+      "learning_rate": 0.0001,
+      "loss": 4.1972,
+      "loss/crossentropy": 2.004905104637146,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21110886335372925,
+      "step": 8972
+    },
+    {
+      "epoch": 0.17948,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.019466145833333334,
+      "learning_rate": 0.0001,
+      "loss": 4.5026,
+      "loss/crossentropy": 2.1859925389289856,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23011507838964462,
+      "step": 8974
+    },
+    {
+      "epoch": 0.17952,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.019840494791666666,
+      "learning_rate": 0.0001,
+      "loss": 4.2908,
+      "loss/crossentropy": 1.8372295498847961,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2171105071902275,
+      "step": 8976
+    },
+    {
+      "epoch": 0.17956,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.020197550455729168,
+      "learning_rate": 0.0001,
+      "loss": 4.3887,
+      "loss/crossentropy": 2.127313494682312,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22017831355333328,
+      "step": 8978
+    },
+    {
+      "epoch": 0.1796,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.022098795572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.3389,
+      "loss/crossentropy": 2.121580421924591,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24898843467235565,
+      "step": 8980
+    },
+    {
+      "epoch": 0.17964,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0226470947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1127,
+      "loss/crossentropy": 2.0973563194274902,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21355029940605164,
+      "step": 8982
+    },
+    {
+      "epoch": 0.17968,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.018896484375,
+      "learning_rate": 0.0001,
+      "loss": 4.3898,
+      "loss/crossentropy": 2.1109927892684937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2096879929304123,
+      "step": 8984
+    },
+    {
+      "epoch": 0.17972,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0126617431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1264,
+      "loss/crossentropy": 1.954129159450531,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20745816081762314,
+      "step": 8986
+    },
+    {
+      "epoch": 0.17976,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.013483683268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.5357,
+      "loss/crossentropy": 1.875806748867035,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19150983542203903,
+      "step": 8988
+    },
+    {
+      "epoch": 0.1798,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.01451416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3042,
+      "loss/crossentropy": 2.3101617097854614,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2393202781677246,
+      "step": 8990
+    },
+    {
+      "epoch": 0.17984,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.013374837239583333,
+      "learning_rate": 0.0001,
+      "loss": 4.4954,
+      "loss/crossentropy": 2.3156551122665405,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23346291482448578,
+      "step": 8992
+    },
+    {
+      "epoch": 0.17988,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.013966623942057292,
+      "learning_rate": 0.0001,
+      "loss": 4.3675,
+      "loss/crossentropy": 2.4437999725341797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2172817811369896,
+      "step": 8994
+    },
+    {
+      "epoch": 0.17992,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.009437815348307291,
+      "learning_rate": 0.0001,
+      "loss": 4.3122,
+      "loss/crossentropy": 2.0451250076293945,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22556670010089874,
+      "step": 8996
+    },
+    {
+      "epoch": 0.17996,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.009852854410807292,
+      "learning_rate": 0.0001,
+      "loss": 4.5654,
+      "loss/crossentropy": 2.3135393857955933,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2313593551516533,
+      "step": 8998
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010530344645182292,
+      "learning_rate": 0.0001,
+      "loss": 4.3001,
+      "loss/crossentropy": 1.9934805035591125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20815817266702652,
+      "step": 9000
+    },
+    {
+      "epoch": 0.18004,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.011982981363932292,
+      "learning_rate": 0.0001,
+      "loss": 4.6193,
+      "loss/crossentropy": 2.1770662665367126,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23489046096801758,
+      "step": 9002
+    },
+    {
+      "epoch": 0.18008,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.009779612223307291,
+      "learning_rate": 0.0001,
+      "loss": 4.3708,
+      "loss/crossentropy": 1.9791364073753357,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2139936238527298,
+      "step": 9004
+    },
+    {
+      "epoch": 0.18012,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.011445871988932292,
+      "learning_rate": 0.0001,
+      "loss": 4.298,
+      "loss/crossentropy": 2.2092931270599365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22620443254709244,
+      "step": 9006
+    },
+    {
+      "epoch": 0.18016,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.011034901936848958,
+      "learning_rate": 0.0001,
+      "loss": 4.3114,
+      "loss/crossentropy": 2.123443365097046,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21943332999944687,
+      "step": 9008
+    },
+    {
+      "epoch": 0.1802,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.009212239583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.4666,
+      "loss/crossentropy": 2.243329405784607,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24108020961284637,
+      "step": 9010
+    },
+    {
+      "epoch": 0.18024,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.008426920572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.3687,
+      "loss/crossentropy": 2.366227388381958,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21806316077709198,
+      "step": 9012
+    },
+    {
+      "epoch": 0.18028,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009663899739583334,
+      "learning_rate": 0.0001,
+      "loss": 3.8255,
+      "loss/crossentropy": 1.768812358379364,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1861925944685936,
+      "step": 9014
+    },
+    {
+      "epoch": 0.18032,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.012225087483723958,
+      "learning_rate": 0.0001,
+      "loss": 4.1236,
+      "loss/crossentropy": 1.9376537799835205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18691913783550262,
+      "step": 9016
+    },
+    {
+      "epoch": 0.18036,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.010709381103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2748,
+      "loss/crossentropy": 2.3026299476623535,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23757526278495789,
+      "step": 9018
+    },
+    {
+      "epoch": 0.1804,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.010057576497395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0026,
+      "loss/crossentropy": 1.9697216153144836,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21453910320997238,
+      "step": 9020
+    },
+    {
+      "epoch": 0.18044,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.011058553059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.5457,
+      "loss/crossentropy": 2.257638931274414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24236120283603668,
+      "step": 9022
+    },
+    {
+      "epoch": 0.18048,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0115142822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3323,
+      "loss/crossentropy": 2.244320869445801,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25429578125476837,
+      "step": 9024
+    },
+    {
+      "epoch": 0.18052,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.012214152018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2854,
+      "loss/crossentropy": 2.1105872988700867,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23081901669502258,
+      "step": 9026
+    },
+    {
+      "epoch": 0.18056,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.012839508056640626,
+      "learning_rate": 0.0001,
+      "loss": 4.2561,
+      "loss/crossentropy": 1.9657647609710693,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22111424803733826,
+      "step": 9028
+    },
+    {
+      "epoch": 0.1806,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.012308502197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4019,
+      "loss/crossentropy": 2.0759438276290894,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22763221710920334,
+      "step": 9030
+    },
+    {
+      "epoch": 0.18064,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010888417561848959,
+      "learning_rate": 0.0001,
+      "loss": 3.9999,
+      "loss/crossentropy": 2.0250572562217712,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2279352843761444,
+      "step": 9032
+    },
+    {
+      "epoch": 0.18068,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.010534413655598958,
+      "learning_rate": 0.0001,
+      "loss": 4.3772,
+      "loss/crossentropy": 2.270031213760376,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23242096602916718,
+      "step": 9034
+    },
+    {
+      "epoch": 0.18072,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.011922200520833334,
+      "learning_rate": 0.0001,
+      "loss": 4.4304,
+      "loss/crossentropy": 2.10041344165802,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23192601650953293,
+      "step": 9036
+    },
+    {
+      "epoch": 0.18076,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.009227498372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1041,
+      "loss/crossentropy": 2.0255953073501587,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23707614094018936,
+      "step": 9038
+    },
+    {
+      "epoch": 0.1808,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.010406239827473959,
+      "learning_rate": 0.0001,
+      "loss": 4.1399,
+      "loss/crossentropy": 1.8162729740142822,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19466694444417953,
+      "step": 9040
+    },
+    {
+      "epoch": 0.18084,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.010170237223307291,
+      "learning_rate": 0.0001,
+      "loss": 4.5543,
+      "loss/crossentropy": 2.1271599531173706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22750889509916306,
+      "step": 9042
+    },
+    {
+      "epoch": 0.18088,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.009447224934895833,
+      "learning_rate": 0.0001,
+      "loss": 4.5377,
+      "loss/crossentropy": 2.3638603687286377,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2456393539905548,
+      "step": 9044
+    },
+    {
+      "epoch": 0.18092,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.019437662760416665,
+      "learning_rate": 0.0001,
+      "loss": 4.3413,
+      "loss/crossentropy": 1.5851669907569885,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20994101464748383,
+      "step": 9046
+    },
+    {
+      "epoch": 0.18096,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.017789459228515624,
+      "learning_rate": 0.0001,
+      "loss": 4.2284,
+      "loss/crossentropy": 1.7990906834602356,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18919725716114044,
+      "step": 9048
+    },
+    {
+      "epoch": 0.181,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.017490386962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.5415,
+      "loss/crossentropy": 2.0975595712661743,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23837832361459732,
+      "step": 9050
+    },
+    {
+      "epoch": 0.18104,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.018070475260416666,
+      "learning_rate": 0.0001,
+      "loss": 4.1304,
+      "loss/crossentropy": 2.1970856189727783,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23219536244869232,
+      "step": 9052
+    },
+    {
+      "epoch": 0.18108,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.017292277018229166,
+      "learning_rate": 0.0001,
+      "loss": 4.211,
+      "loss/crossentropy": 2.0846009850502014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20775136351585388,
+      "step": 9054
+    },
+    {
+      "epoch": 0.18112,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.015075429280598959,
+      "learning_rate": 0.0001,
+      "loss": 4.0415,
+      "loss/crossentropy": 1.662496030330658,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18223516643047333,
+      "step": 9056
+    },
+    {
+      "epoch": 0.18116,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.016078440348307292,
+      "learning_rate": 0.0001,
+      "loss": 4.2561,
+      "loss/crossentropy": 2.126902401447296,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2193778082728386,
+      "step": 9058
+    },
+    {
+      "epoch": 0.1812,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.062459309895833336,
+      "learning_rate": 0.0001,
+      "loss": 4.0343,
+      "loss/crossentropy": 1.9529814720153809,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19026879966259003,
+      "step": 9060
+    },
+    {
+      "epoch": 0.18124,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.05981852213541667,
+      "learning_rate": 0.0001,
+      "loss": 4.1911,
+      "loss/crossentropy": 2.140450179576874,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23864319920539856,
+      "step": 9062
+    },
+    {
+      "epoch": 0.18128,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.06083882649739583,
+      "learning_rate": 0.0001,
+      "loss": 4.3442,
+      "loss/crossentropy": 2.4139195680618286,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2360195592045784,
+      "step": 9064
+    },
+    {
+      "epoch": 0.18132,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.06083882649739583,
+      "learning_rate": 0.0001,
+      "loss": 4.5122,
+      "loss/crossentropy": 2.190356135368347,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22671552002429962,
+      "step": 9066
+    },
+    {
+      "epoch": 0.18136,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.05916315714518229,
+      "learning_rate": 0.0001,
+      "loss": 4.281,
+      "loss/crossentropy": 1.9428812861442566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22318705916404724,
+      "step": 9068
+    },
+    {
+      "epoch": 0.1814,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.059242502848307295,
+      "learning_rate": 0.0001,
+      "loss": 4.4628,
+      "loss/crossentropy": 2.296473503112793,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22005227208137512,
+      "step": 9070
+    },
+    {
+      "epoch": 0.18144,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.05919774373372396,
+      "learning_rate": 0.0001,
+      "loss": 4.5848,
+      "loss/crossentropy": 2.2118901014328003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22189343720674515,
+      "step": 9072
+    },
+    {
+      "epoch": 0.18148,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.05810114542643229,
+      "learning_rate": 0.0001,
+      "loss": 4.285,
+      "loss/crossentropy": 1.8919037580490112,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1986190229654312,
+      "step": 9074
+    },
+    {
+      "epoch": 0.18152,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.011161295572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.3046,
+      "loss/crossentropy": 1.975312054157257,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20403321832418442,
+      "step": 9076
+    },
+    {
+      "epoch": 0.18156,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0075032552083333336,
+      "learning_rate": 0.0001,
+      "loss": 4.2175,
+      "loss/crossentropy": 1.8076966404914856,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19708115607500076,
+      "step": 9078
+    },
+    {
+      "epoch": 0.1816,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.007112630208333333,
+      "learning_rate": 0.0001,
+      "loss": 4.2166,
+      "loss/crossentropy": 2.101171374320984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22123181074857712,
+      "step": 9080
+    },
+    {
+      "epoch": 0.18164,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.007225545247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.2292,
+      "loss/crossentropy": 2.09942090511322,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21761803328990936,
+      "step": 9082
+    },
+    {
+      "epoch": 0.18168,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.008072916666666667,
+      "learning_rate": 0.0001,
+      "loss": 4.4802,
+      "loss/crossentropy": 2.4418424367904663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24908769130706787,
+      "step": 9084
+    },
+    {
+      "epoch": 0.18172,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3408,
+      "loss/crossentropy": 1.944950520992279,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20515284687280655,
+      "step": 9086
+    },
+    {
+      "epoch": 0.18176,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.007991536458333334,
+      "learning_rate": 0.0001,
+      "loss": 4.3117,
+      "loss/crossentropy": 1.870418667793274,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20427027344703674,
+      "step": 9088
+    },
+    {
+      "epoch": 0.1818,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.007682291666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.1323,
+      "loss/crossentropy": 1.9338520169258118,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19928501546382904,
+      "step": 9090
+    },
+    {
+      "epoch": 0.18184,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.008876291910807292,
+      "learning_rate": 0.0001,
+      "loss": 3.9734,
+      "loss/crossentropy": 1.7826221585273743,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19778436422348022,
+      "step": 9092
+    },
+    {
+      "epoch": 0.18188,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.007458241780598959,
+      "learning_rate": 0.0001,
+      "loss": 4.3537,
+      "loss/crossentropy": 2.2922680377960205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23613491654396057,
+      "step": 9094
+    },
+    {
+      "epoch": 0.18192,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.15102513631184897,
+      "learning_rate": 0.0001,
+      "loss": 4.6789,
+      "loss/crossentropy": 2.1802788972854614,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1969536542892456,
+      "step": 9096
+    },
+    {
+      "epoch": 0.18196,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.1498308817545573,
+      "learning_rate": 0.0001,
+      "loss": 4.1715,
+      "loss/crossentropy": 1.9129992723464966,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1942882016301155,
+      "step": 9098
+    },
+    {
+      "epoch": 0.182,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.14738337198893228,
+      "learning_rate": 0.0001,
+      "loss": 4.3374,
+      "loss/crossentropy": 1.8288249969482422,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19879616051912308,
+      "step": 9100
+    },
+    {
+      "epoch": 0.18204,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.14580663045247397,
+      "learning_rate": 0.0001,
+      "loss": 4.2824,
+      "loss/crossentropy": 2.039812684059143,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21374420076608658,
+      "step": 9102
+    },
+    {
+      "epoch": 0.18208,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.15449193318684895,
+      "learning_rate": 0.0001,
+      "loss": 4.1862,
+      "loss/crossentropy": 2.1177414059638977,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2193453460931778,
+      "step": 9104
+    },
+    {
+      "epoch": 0.18212,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.16704076131184895,
+      "learning_rate": 0.0001,
+      "loss": 4.4869,
+      "loss/crossentropy": 2.171097159385681,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21211445331573486,
+      "step": 9106
+    },
+    {
+      "epoch": 0.18216,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.16413548787434895,
+      "learning_rate": 0.0001,
+      "loss": 4.1422,
+      "loss/crossentropy": 1.9781638383865356,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21307373046875,
+      "step": 9108
+    },
+    {
+      "epoch": 0.1822,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.17157363891601562,
+      "learning_rate": 0.0001,
+      "loss": 4.2454,
+      "loss/crossentropy": 2.0868560075759888,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20633937418460846,
+      "step": 9110
+    },
+    {
+      "epoch": 0.18224,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.04592463175455729,
+      "learning_rate": 0.0001,
+      "loss": 4.1153,
+      "loss/crossentropy": 2.079172134399414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20709815621376038,
+      "step": 9112
+    },
+    {
+      "epoch": 0.18228,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.04835383097330729,
+      "learning_rate": 0.0001,
+      "loss": 4.1836,
+      "loss/crossentropy": 2.100913643836975,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21831409633159637,
+      "step": 9114
+    },
+    {
+      "epoch": 0.18232,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.049478912353515626,
+      "learning_rate": 0.0001,
+      "loss": 4.4087,
+      "loss/crossentropy": 2.018395781517029,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22545243054628372,
+      "step": 9116
+    },
+    {
+      "epoch": 0.18236,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.04278132120768229,
+      "learning_rate": 0.0001,
+      "loss": 4.3421,
+      "loss/crossentropy": 1.9177632331848145,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23346271365880966,
+      "step": 9118
+    },
+    {
+      "epoch": 0.1824,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.038358306884765624,
+      "learning_rate": 0.0001,
+      "loss": 3.9924,
+      "loss/crossentropy": 2.1122357845306396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22593770176172256,
+      "step": 9120
+    },
+    {
+      "epoch": 0.18244,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.014021555582682291,
+      "learning_rate": 0.0001,
+      "loss": 4.1399,
+      "loss/crossentropy": 1.7920495867729187,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2127150148153305,
+      "step": 9122
+    },
+    {
+      "epoch": 0.18248,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0096099853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2061,
+      "loss/crossentropy": 2.2180920839309692,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25088224560022354,
+      "step": 9124
+    },
+    {
+      "epoch": 0.18252,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.008185831705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3287,
+      "loss/crossentropy": 1.8999969959259033,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21616832166910172,
+      "step": 9126
+    },
+    {
+      "epoch": 0.18256,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0081451416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.5351,
+      "loss/crossentropy": 1.9424527287483215,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22016742825508118,
+      "step": 9128
+    },
+    {
+      "epoch": 0.1826,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0049468994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.4432,
+      "loss/crossentropy": 2.0539366006851196,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26818516850471497,
+      "step": 9130
+    },
+    {
+      "epoch": 0.18264,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.00458984375,
+      "learning_rate": 0.0001,
+      "loss": 4.1634,
+      "loss/crossentropy": 1.9180519580841064,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2067112922668457,
+      "step": 9132
+    },
+    {
+      "epoch": 0.18268,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.005052693684895833,
+      "learning_rate": 0.0001,
+      "loss": 3.9913,
+      "loss/crossentropy": 1.7417545318603516,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1970825269818306,
+      "step": 9134
+    },
+    {
+      "epoch": 0.18272,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.004833984375,
+      "learning_rate": 0.0001,
+      "loss": 4.016,
+      "loss/crossentropy": 1.9498217701911926,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21281737089157104,
+      "step": 9136
+    },
+    {
+      "epoch": 0.18276,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.005615234375,
+      "learning_rate": 0.0001,
+      "loss": 4.3842,
+      "loss/crossentropy": 2.140692949295044,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24291902035474777,
+      "step": 9138
+    },
+    {
+      "epoch": 0.1828,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0069976806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1785,
+      "loss/crossentropy": 2.3510342836380005,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23326712846755981,
+      "step": 9140
+    },
+    {
+      "epoch": 0.18284,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0072743733723958336,
+      "learning_rate": 0.0001,
+      "loss": 4.0707,
+      "loss/crossentropy": 2.049591898918152,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21118033677339554,
+      "step": 9142
+    },
+    {
+      "epoch": 0.18288,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.005631510416666667,
+      "learning_rate": 0.0001,
+      "loss": 4.263,
+      "loss/crossentropy": 1.949703335762024,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22161198407411575,
+      "step": 9144
+    },
+    {
+      "epoch": 0.18292,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.005399576822916667,
+      "learning_rate": 0.0001,
+      "loss": 4.5602,
+      "loss/crossentropy": 2.1442413330078125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21281076222658157,
+      "step": 9146
+    },
+    {
+      "epoch": 0.18296,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.007111612955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0848,
+      "loss/crossentropy": 2.103494882583618,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20892268419265747,
+      "step": 9148
+    },
+    {
+      "epoch": 0.183,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.01060791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.133,
+      "loss/crossentropy": 2.1544495224952698,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2008143737912178,
+      "step": 9150
+    },
+    {
+      "epoch": 0.18304,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.01109619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2799,
+      "loss/crossentropy": 2.021821677684784,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20975126326084137,
+      "step": 9152
+    },
+    {
+      "epoch": 0.18308,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.0102203369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2858,
+      "loss/crossentropy": 2.109215199947357,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19855067878961563,
+      "step": 9154
+    },
+    {
+      "epoch": 0.18312,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.011165364583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.2181,
+      "loss/crossentropy": 1.7631941437721252,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20584283769130707,
+      "step": 9156
+    },
+    {
+      "epoch": 0.18316,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0116851806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4005,
+      "loss/crossentropy": 2.131524443626404,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2151477411389351,
+      "step": 9158
+    },
+    {
+      "epoch": 0.1832,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.01334228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2729,
+      "loss/crossentropy": 2.018375277519226,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21982873231172562,
+      "step": 9160
+    },
+    {
+      "epoch": 0.18324,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0154205322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3302,
+      "loss/crossentropy": 2.217617154121399,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2257251739501953,
+      "step": 9162
+    },
+    {
+      "epoch": 0.18328,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.014606730143229166,
+      "learning_rate": 0.0001,
+      "loss": 4.227,
+      "loss/crossentropy": 1.8632460832595825,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19728046655654907,
+      "step": 9164
+    },
+    {
+      "epoch": 0.18332,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.012137858072916667,
+      "learning_rate": 0.0001,
+      "loss": 4.5335,
+      "loss/crossentropy": 2.2818111181259155,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22397568821907043,
+      "step": 9166
+    },
+    {
+      "epoch": 0.18336,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.01177978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4104,
+      "loss/crossentropy": 2.1209938526153564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22490206360816956,
+      "step": 9168
+    },
+    {
+      "epoch": 0.1834,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.00924072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.244,
+      "loss/crossentropy": 2.141623795032501,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21710850298404694,
+      "step": 9170
+    },
+    {
+      "epoch": 0.18344,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.007323201497395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2063,
+      "loss/crossentropy": 2.165239691734314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.235738106071949,
+      "step": 9172
+    },
+    {
+      "epoch": 0.18348,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008103179931640624,
+      "learning_rate": 0.0001,
+      "loss": 4.0186,
+      "loss/crossentropy": 1.8649475574493408,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2088451236486435,
+      "step": 9174
+    },
+    {
+      "epoch": 0.18352,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.007012685139973958,
+      "learning_rate": 0.0001,
+      "loss": 4.2853,
+      "loss/crossentropy": 2.312218189239502,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.236283540725708,
+      "step": 9176
+    },
+    {
+      "epoch": 0.18356,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.008699544270833333,
+      "learning_rate": 0.0001,
+      "loss": 3.9626,
+      "loss/crossentropy": 2.0149282217025757,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21434535831212997,
+      "step": 9178
+    },
+    {
+      "epoch": 0.1836,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0086822509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.5081,
+      "loss/crossentropy": 2.499966621398926,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26318275928497314,
+      "step": 9180
+    },
+    {
+      "epoch": 0.18364,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0084136962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4498,
+      "loss/crossentropy": 2.0454984307289124,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2202616035938263,
+      "step": 9182
+    },
+    {
+      "epoch": 0.18368,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.010835520426432292,
+      "learning_rate": 0.0001,
+      "loss": 4.0754,
+      "loss/crossentropy": 2.1708725094795227,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20958629250526428,
+      "step": 9184
+    },
+    {
+      "epoch": 0.18372,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.010792795817057292,
+      "learning_rate": 0.0001,
+      "loss": 4.2104,
+      "loss/crossentropy": 1.8261350989341736,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1952563151717186,
+      "step": 9186
+    },
+    {
+      "epoch": 0.18376,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.010009511311848959,
+      "learning_rate": 0.0001,
+      "loss": 4.1247,
+      "loss/crossentropy": 2.036627769470215,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22490298002958298,
+      "step": 9188
+    },
+    {
+      "epoch": 0.1838,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.010501861572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0088,
+      "loss/crossentropy": 1.7977086305618286,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19282807409763336,
+      "step": 9190
+    },
+    {
+      "epoch": 0.18384,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.011211903889973958,
+      "learning_rate": 0.0001,
+      "loss": 4.1774,
+      "loss/crossentropy": 2.170135021209717,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23149186372756958,
+      "step": 9192
+    },
+    {
+      "epoch": 0.18388,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.012035115559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.4155,
+      "loss/crossentropy": 2.1453020572662354,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25120319426059723,
+      "step": 9194
+    },
+    {
+      "epoch": 0.18392,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.01224365234375,
+      "learning_rate": 0.0001,
+      "loss": 4.3628,
+      "loss/crossentropy": 1.8794063925743103,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20883548259735107,
+      "step": 9196
+    },
+    {
+      "epoch": 0.18396,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.0141998291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3384,
+      "loss/crossentropy": 2.021254241466522,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21939975768327713,
+      "step": 9198
+    },
+    {
+      "epoch": 0.184,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.009557851155598958,
+      "learning_rate": 0.0001,
+      "loss": 4.158,
+      "loss/crossentropy": 2.195580303668976,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2108924463391304,
+      "step": 9200
+    },
+    {
+      "epoch": 0.18404,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.014085896809895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0781,
+      "loss/crossentropy": 1.6260902881622314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20428159832954407,
+      "step": 9202
+    },
+    {
+      "epoch": 0.18408,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.015363566080729167,
+      "learning_rate": 0.0001,
+      "loss": 4.2525,
+      "loss/crossentropy": 2.138678550720215,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21641074120998383,
+      "step": 9204
+    },
+    {
+      "epoch": 0.18412,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.01825129191080729,
+      "learning_rate": 0.0001,
+      "loss": 4.4706,
+      "loss/crossentropy": 2.047194480895996,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20891964435577393,
+      "step": 9206
+    },
+    {
+      "epoch": 0.18416,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.019606272379557293,
+      "learning_rate": 0.0001,
+      "loss": 4.2312,
+      "loss/crossentropy": 2.189277768135071,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2150397077202797,
+      "step": 9208
+    },
+    {
+      "epoch": 0.1842,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.017618560791015626,
+      "learning_rate": 0.0001,
+      "loss": 4.2792,
+      "loss/crossentropy": 2.184122920036316,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23278063535690308,
+      "step": 9210
+    },
+    {
+      "epoch": 0.18424,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.019139607747395832,
+      "learning_rate": 0.0001,
+      "loss": 4.3471,
+      "loss/crossentropy": 2.413718104362488,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24333150684833527,
+      "step": 9212
+    },
+    {
+      "epoch": 0.18428,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.019017537434895832,
+      "learning_rate": 0.0001,
+      "loss": 3.8486,
+      "loss/crossentropy": 1.8086814880371094,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19976364076137543,
+      "step": 9214
+    },
+    {
+      "epoch": 0.18432,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.019978841145833332,
+      "learning_rate": 0.0001,
+      "loss": 4.5222,
+      "loss/crossentropy": 2.2418206930160522,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.229690782725811,
+      "step": 9216
+    },
+    {
+      "epoch": 0.18436,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.016355133056640624,
+      "learning_rate": 0.0001,
+      "loss": 4.0817,
+      "loss/crossentropy": 1.8083258867263794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20999443531036377,
+      "step": 9218
+    },
+    {
+      "epoch": 0.1844,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.01693115234375,
+      "learning_rate": 0.0001,
+      "loss": 3.772,
+      "loss/crossentropy": 1.8117709755897522,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20436270534992218,
+      "step": 9220
+    },
+    {
+      "epoch": 0.18444,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.010550944010416667,
+      "learning_rate": 0.0001,
+      "loss": 4.2149,
+      "loss/crossentropy": 2.024270534515381,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20097267627716064,
+      "step": 9222
+    },
+    {
+      "epoch": 0.18448,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.011449178059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.4756,
+      "loss/crossentropy": 2.2385981678962708,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21131044626235962,
+      "step": 9224
+    },
+    {
+      "epoch": 0.18452,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.011271158854166666,
+      "learning_rate": 0.0001,
+      "loss": 4.3745,
+      "loss/crossentropy": 2.127749502658844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22855369001626968,
+      "step": 9226
+    },
+    {
+      "epoch": 0.18456,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.011237589518229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2645,
+      "loss/crossentropy": 2.1877033710479736,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2339663878083229,
+      "step": 9228
+    },
+    {
+      "epoch": 0.1846,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.011139933268229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2642,
+      "loss/crossentropy": 1.931507408618927,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20912021398544312,
+      "step": 9230
+    },
+    {
+      "epoch": 0.18464,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.011579386393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.3309,
+      "loss/crossentropy": 1.8101251125335693,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23724676668643951,
+      "step": 9232
+    },
+    {
+      "epoch": 0.18468,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0111480712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.201,
+      "loss/crossentropy": 2.015208065509796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23482007533311844,
+      "step": 9234
+    },
+    {
+      "epoch": 0.18472,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.010282135009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.5357,
+      "loss/crossentropy": 2.2742738723754883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23632052540779114,
+      "step": 9236
+    },
+    {
+      "epoch": 0.18476,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.016806793212890626,
+      "learning_rate": 0.0001,
+      "loss": 4.3574,
+      "loss/crossentropy": 1.7254774570465088,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2104022428393364,
+      "step": 9238
+    },
+    {
+      "epoch": 0.1848,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.016585032145182293,
+      "learning_rate": 0.0001,
+      "loss": 4.3477,
+      "loss/crossentropy": 2.181770443916321,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22411521524190903,
+      "step": 9240
+    },
+    {
+      "epoch": 0.18484,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.020157877604166666,
+      "learning_rate": 0.0001,
+      "loss": 3.9517,
+      "loss/crossentropy": 1.8449691534042358,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2091139778494835,
+      "step": 9242
+    },
+    {
+      "epoch": 0.18488,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.017195383707682293,
+      "learning_rate": 0.0001,
+      "loss": 4.3389,
+      "loss/crossentropy": 2.0917609333992004,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2644127458333969,
+      "step": 9244
+    },
+    {
+      "epoch": 0.18492,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.01793390909830729,
+      "learning_rate": 0.0001,
+      "loss": 4.083,
+      "loss/crossentropy": 2.060012102127075,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21294714510440826,
+      "step": 9246
+    },
+    {
+      "epoch": 0.18496,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.018387603759765624,
+      "learning_rate": 0.0001,
+      "loss": 4.6598,
+      "loss/crossentropy": 2.059940278530121,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22648683190345764,
+      "step": 9248
+    },
+    {
+      "epoch": 0.185,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.01862360636393229,
+      "learning_rate": 0.0001,
+      "loss": 4.4696,
+      "loss/crossentropy": 1.8423291444778442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22509171068668365,
+      "step": 9250
+    },
+    {
+      "epoch": 0.18504,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.016410064697265626,
+      "learning_rate": 0.0001,
+      "loss": 4.4629,
+      "loss/crossentropy": 2.2559698820114136,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23896963894367218,
+      "step": 9252
+    },
+    {
+      "epoch": 0.18508,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.01579767862955729,
+      "learning_rate": 0.0001,
+      "loss": 4.6027,
+      "loss/crossentropy": 2.0583658814430237,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22763275355100632,
+      "step": 9254
+    },
+    {
+      "epoch": 0.18512,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.016658274332682292,
+      "learning_rate": 0.0001,
+      "loss": 4.2654,
+      "loss/crossentropy": 1.816649854183197,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18608735501766205,
+      "step": 9256
+    },
+    {
+      "epoch": 0.18516,
+      "grad_norm": 4.65625,
+      "grad_norm_var": 0.4073150634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2837,
+      "loss/crossentropy": 2.0920958518981934,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22859029471874237,
+      "step": 9258
+    },
+    {
+      "epoch": 0.1852,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.4108306884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2517,
+      "loss/crossentropy": 2.1475982666015625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21678777784109116,
+      "step": 9260
+    },
+    {
+      "epoch": 0.18524,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.4100901285807292,
+      "learning_rate": 0.0001,
+      "loss": 4.4258,
+      "loss/crossentropy": 2.22346031665802,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2333592176437378,
+      "step": 9262
+    },
+    {
+      "epoch": 0.18528,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.41646703084309894,
+      "learning_rate": 0.0001,
+      "loss": 4.2134,
+      "loss/crossentropy": 1.7652028799057007,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18461769074201584,
+      "step": 9264
+    },
+    {
+      "epoch": 0.18532,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.4225006103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2362,
+      "loss/crossentropy": 2.1549625396728516,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22428707033395767,
+      "step": 9266
+    },
+    {
+      "epoch": 0.18536,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.4281972249348958,
+      "learning_rate": 0.0001,
+      "loss": 4.3685,
+      "loss/crossentropy": 2.1677842140197754,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22887174785137177,
+      "step": 9268
+    },
+    {
+      "epoch": 0.1854,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.42428792317708336,
+      "learning_rate": 0.0001,
+      "loss": 4.1517,
+      "loss/crossentropy": 1.8352131247520447,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19628287106752396,
+      "step": 9270
+    },
+    {
+      "epoch": 0.18544,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.43038304646809894,
+      "learning_rate": 0.0001,
+      "loss": 4.101,
+      "loss/crossentropy": 2.104279100894928,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20481227338314056,
+      "step": 9272
+    },
+    {
+      "epoch": 0.18548,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.009795888264973959,
+      "learning_rate": 0.0001,
+      "loss": 4.2011,
+      "loss/crossentropy": 1.8447301387786865,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2227916270494461,
+      "step": 9274
+    },
+    {
+      "epoch": 0.18552,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.013651275634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3544,
+      "loss/crossentropy": 2.303207755088806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24895529448986053,
+      "step": 9276
+    },
+    {
+      "epoch": 0.18556,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.014115142822265624,
+      "learning_rate": 0.0001,
+      "loss": 4.699,
+      "loss/crossentropy": 2.248077630996704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23077847063541412,
+      "step": 9278
+    },
+    {
+      "epoch": 0.1856,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.017438761393229165,
+      "learning_rate": 0.0001,
+      "loss": 4.4943,
+      "loss/crossentropy": 2.261389970779419,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23669905960559845,
+      "step": 9280
+    },
+    {
+      "epoch": 0.18564,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.016056060791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3805,
+      "loss/crossentropy": 2.402338147163391,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2453952580690384,
+      "step": 9282
+    },
+    {
+      "epoch": 0.18568,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.01587702433268229,
+      "learning_rate": 0.0001,
+      "loss": 4.2253,
+      "loss/crossentropy": 2.011174201965332,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22673364728689194,
+      "step": 9284
+    },
+    {
+      "epoch": 0.18572,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.016810862223307292,
+      "learning_rate": 0.0001,
+      "loss": 4.3897,
+      "loss/crossentropy": 1.9700093269348145,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22711393237113953,
+      "step": 9286
+    },
+    {
+      "epoch": 0.18576,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.014922841389973959,
+      "learning_rate": 0.0001,
+      "loss": 3.9609,
+      "loss/crossentropy": 2.0253939032554626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21902555227279663,
+      "step": 9288
+    },
+    {
+      "epoch": 0.1858,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.016961415608723957,
+      "learning_rate": 0.0001,
+      "loss": 4.1799,
+      "loss/crossentropy": 1.741984784603119,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1879509538412094,
+      "step": 9290
+    },
+    {
+      "epoch": 0.18584,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.022564442952473958,
+      "learning_rate": 0.0001,
+      "loss": 4.7983,
+      "loss/crossentropy": 2.3156943321228027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2687100023031235,
+      "step": 9292
+    },
+    {
+      "epoch": 0.18588,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.025833892822265624,
+      "learning_rate": 0.0001,
+      "loss": 4.2894,
+      "loss/crossentropy": 2.0826371908187866,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21234872937202454,
+      "step": 9294
+    },
+    {
+      "epoch": 0.18592,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.025921376546223958,
+      "learning_rate": 0.0001,
+      "loss": 4.6473,
+      "loss/crossentropy": 2.4080610275268555,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2670409381389618,
+      "step": 9296
+    },
+    {
+      "epoch": 0.18596,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.02958958943684896,
+      "learning_rate": 0.0001,
+      "loss": 4.1997,
+      "loss/crossentropy": 1.722270905971527,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20094333589076996,
+      "step": 9298
+    },
+    {
+      "epoch": 0.186,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.028527577718098957,
+      "learning_rate": 0.0001,
+      "loss": 4.3247,
+      "loss/crossentropy": 2.0514711141586304,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.209539495408535,
+      "step": 9300
+    },
+    {
+      "epoch": 0.18604,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 2.5507850646972656,
+      "learning_rate": 0.0001,
+      "loss": 4.9819,
+      "loss/crossentropy": 2.4808409214019775,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2367517650127411,
+      "step": 9302
+    },
+    {
+      "epoch": 0.18608,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 2.536018880208333,
+      "learning_rate": 0.0001,
+      "loss": 4.1105,
+      "loss/crossentropy": 2.2539944648742676,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22834083437919617,
+      "step": 9304
+    },
+    {
+      "epoch": 0.18612,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 2.5449544270833333,
+      "learning_rate": 0.0001,
+      "loss": 4.2383,
+      "loss/crossentropy": 1.9335210919380188,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20264607667922974,
+      "step": 9306
+    },
+    {
+      "epoch": 0.18616,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 2.5707194010416665,
+      "learning_rate": 0.0001,
+      "loss": 4.2285,
+      "loss/crossentropy": 1.90863037109375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20882528275251389,
+      "step": 9308
+    },
+    {
+      "epoch": 0.1862,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 2.5589996337890626,
+      "learning_rate": 0.0001,
+      "loss": 4.2195,
+      "loss/crossentropy": 2.1449084281921387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2329169511795044,
+      "step": 9310
+    },
+    {
+      "epoch": 0.18624,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 2.567341105143229,
+      "learning_rate": 0.0001,
+      "loss": 4.1806,
+      "loss/crossentropy": 1.9571366906166077,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2148708775639534,
+      "step": 9312
+    },
+    {
+      "epoch": 0.18628,
+      "grad_norm": 2.25,
+      "grad_norm_var": 2.542252604166667,
+      "learning_rate": 0.0001,
+      "loss": 4.4691,
+      "loss/crossentropy": 2.174731135368347,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2280728593468666,
+      "step": 9314
+    },
+    {
+      "epoch": 0.18632,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 2.5416575113932294,
+      "learning_rate": 0.0001,
+      "loss": 4.4344,
+      "loss/crossentropy": 1.9569833874702454,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20855706185102463,
+      "step": 9316
+    },
+    {
+      "epoch": 0.18636,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0090972900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3162,
+      "loss/crossentropy": 2.153563976287842,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22807130962610245,
+      "step": 9318
+    },
+    {
+      "epoch": 0.1864,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.010741933186848959,
+      "learning_rate": 0.0001,
+      "loss": 4.3028,
+      "loss/crossentropy": 1.9219058752059937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2295902967453003,
+      "step": 9320
+    },
+    {
+      "epoch": 0.18644,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.011926015218098959,
+      "learning_rate": 0.0001,
+      "loss": 4.29,
+      "loss/crossentropy": 2.1993675231933594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.223759263753891,
+      "step": 9322
+    },
+    {
+      "epoch": 0.18648,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.011730702718098958,
+      "learning_rate": 0.0001,
+      "loss": 4.2254,
+      "loss/crossentropy": 2.173740863800049,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23405525833368301,
+      "step": 9324
+    },
+    {
+      "epoch": 0.18652,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.012564849853515626,
+      "learning_rate": 0.0001,
+      "loss": 4.3717,
+      "loss/crossentropy": 2.026577115058899,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2265823632478714,
+      "step": 9326
+    },
+    {
+      "epoch": 0.18656,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.030326080322265626,
+      "learning_rate": 0.0001,
+      "loss": 4.4664,
+      "loss/crossentropy": 2.3629637956619263,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23659023642539978,
+      "step": 9328
+    },
+    {
+      "epoch": 0.1866,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.03050715128580729,
+      "learning_rate": 0.0001,
+      "loss": 4.3245,
+      "loss/crossentropy": 2.100727915763855,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21468425542116165,
+      "step": 9330
+    },
+    {
+      "epoch": 0.18664,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.025923411051432293,
+      "learning_rate": 0.0001,
+      "loss": 4.5291,
+      "loss/crossentropy": 2.163568615913391,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23402437567710876,
+      "step": 9332
+    },
+    {
+      "epoch": 0.18668,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.02934748331705729,
+      "learning_rate": 0.0001,
+      "loss": 3.9687,
+      "loss/crossentropy": 1.9579994082450867,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19713342934846878,
+      "step": 9334
+    },
+    {
+      "epoch": 0.18672,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.029412587483723957,
+      "learning_rate": 0.0001,
+      "loss": 4.2586,
+      "loss/crossentropy": 1.805375874042511,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22647518664598465,
+      "step": 9336
+    },
+    {
+      "epoch": 0.18676,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.02797215779622396,
+      "learning_rate": 0.0001,
+      "loss": 4.2025,
+      "loss/crossentropy": 2.0764458775520325,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2270444855093956,
+      "step": 9338
+    },
+    {
+      "epoch": 0.1868,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.030987294514973958,
+      "learning_rate": 0.0001,
+      "loss": 3.8731,
+      "loss/crossentropy": 1.5654467940330505,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15944529324769974,
+      "step": 9340
+    },
+    {
+      "epoch": 0.18684,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.02976048787434896,
+      "learning_rate": 0.0001,
+      "loss": 4.3831,
+      "loss/crossentropy": 2.007612407207489,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2061111181974411,
+      "step": 9342
+    },
+    {
+      "epoch": 0.18688,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.010910797119140624,
+      "learning_rate": 0.0001,
+      "loss": 4.0835,
+      "loss/crossentropy": 1.981432855129242,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19339826703071594,
+      "step": 9344
+    },
+    {
+      "epoch": 0.18692,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.00965576171875,
+      "learning_rate": 0.0001,
+      "loss": 4.1291,
+      "loss/crossentropy": 1.804275631904602,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1968545839190483,
+      "step": 9346
+    },
+    {
+      "epoch": 0.18696,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.008763631184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.2109,
+      "loss/crossentropy": 1.9718617796897888,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21482623368501663,
+      "step": 9348
+    },
+    {
+      "epoch": 0.187,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.008397420247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1672,
+      "loss/crossentropy": 1.8358338475227356,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20042669028043747,
+      "step": 9350
+    },
+    {
+      "epoch": 0.18704,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.009089914957682292,
+      "learning_rate": 0.0001,
+      "loss": 4.2045,
+      "loss/crossentropy": 1.9447709321975708,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21207460761070251,
+      "step": 9352
+    },
+    {
+      "epoch": 0.18708,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011356353759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3134,
+      "loss/crossentropy": 1.8921163082122803,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19945629686117172,
+      "step": 9354
+    },
+    {
+      "epoch": 0.18712,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.009580230712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2018,
+      "loss/crossentropy": 2.1323947310447693,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2258753925561905,
+      "step": 9356
+    },
+    {
+      "epoch": 0.18716,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009277089436848959,
+      "learning_rate": 0.0001,
+      "loss": 4.4278,
+      "loss/crossentropy": 2.064914345741272,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23450962454080582,
+      "step": 9358
+    },
+    {
+      "epoch": 0.1872,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.005783843994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0173,
+      "loss/crossentropy": 2.1264703273773193,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2169734537601471,
+      "step": 9360
+    },
+    {
+      "epoch": 0.18724,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.004784138997395834,
+      "learning_rate": 0.0001,
+      "loss": 4.2926,
+      "loss/crossentropy": 1.7044150233268738,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19169463962316513,
+      "step": 9362
+    },
+    {
+      "epoch": 0.18728,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.004541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.5367,
+      "loss/crossentropy": 1.9475398659706116,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22891414165496826,
+      "step": 9364
+    },
+    {
+      "epoch": 0.18732,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.008017730712890626,
+      "learning_rate": 0.0001,
+      "loss": 4.1416,
+      "loss/crossentropy": 2.134114623069763,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2241017445921898,
+      "step": 9366
+    },
+    {
+      "epoch": 0.18736,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.007342274983723958,
+      "learning_rate": 0.0001,
+      "loss": 4.2185,
+      "loss/crossentropy": 1.8317620158195496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20189791917800903,
+      "step": 9368
+    },
+    {
+      "epoch": 0.1874,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.005041249593098958,
+      "learning_rate": 0.0001,
+      "loss": 4.3518,
+      "loss/crossentropy": 2.257538855075836,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2343401461839676,
+      "step": 9370
+    },
+    {
+      "epoch": 0.18744,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.005228424072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0535,
+      "loss/crossentropy": 2.3722634315490723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23045828938484192,
+      "step": 9372
+    },
+    {
+      "epoch": 0.18748,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.005222320556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2293,
+      "loss/crossentropy": 2.1880545020103455,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2345728725194931,
+      "step": 9374
+    },
+    {
+      "epoch": 0.18752,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.008255767822265624,
+      "learning_rate": 0.0001,
+      "loss": 4.5037,
+      "loss/crossentropy": 1.8883287906646729,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25232937932014465,
+      "step": 9376
+    },
+    {
+      "epoch": 0.18756,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.009124501546223959,
+      "learning_rate": 0.0001,
+      "loss": 4.2963,
+      "loss/crossentropy": 1.9253730773925781,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20324261486530304,
+      "step": 9378
+    },
+    {
+      "epoch": 0.1876,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009211985270182292,
+      "learning_rate": 0.0001,
+      "loss": 4.2745,
+      "loss/crossentropy": 1.961540937423706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22623597085475922,
+      "step": 9380
+    },
+    {
+      "epoch": 0.18764,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.007835896809895833,
+      "learning_rate": 0.0001,
+      "loss": 4.347,
+      "loss/crossentropy": 2.190120279788971,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24154195934534073,
+      "step": 9382
+    },
+    {
+      "epoch": 0.18768,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.007445271809895833,
+      "learning_rate": 0.0001,
+      "loss": 4.6159,
+      "loss/crossentropy": 2.0888350009918213,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22750811278820038,
+      "step": 9384
+    },
+    {
+      "epoch": 0.18772,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.008128865559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3171,
+      "loss/crossentropy": 1.724816918373108,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21245518326759338,
+      "step": 9386
+    },
+    {
+      "epoch": 0.18776,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.009307607014973959,
+      "learning_rate": 0.0001,
+      "loss": 4.0868,
+      "loss/crossentropy": 1.6542762517929077,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1961463838815689,
+      "step": 9388
+    },
+    {
+      "epoch": 0.1878,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009714508056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4748,
+      "loss/crossentropy": 2.3528761863708496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23583710938692093,
+      "step": 9390
+    },
+    {
+      "epoch": 0.18784,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009012603759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.394,
+      "loss/crossentropy": 2.181188702583313,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24201467633247375,
+      "step": 9392
+    },
+    {
+      "epoch": 0.18788,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.008414459228515626,
+      "learning_rate": 0.0001,
+      "loss": 4.3476,
+      "loss/crossentropy": 2.280683398246765,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22065220028162003,
+      "step": 9394
+    },
+    {
+      "epoch": 0.18792,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.009275054931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1603,
+      "loss/crossentropy": 1.9839438199996948,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22708184272050858,
+      "step": 9396
+    },
+    {
+      "epoch": 0.18796,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0070879618326822914,
+      "learning_rate": 0.0001,
+      "loss": 4.4322,
+      "loss/crossentropy": 2.311874270439148,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22249652445316315,
+      "step": 9398
+    },
+    {
+      "epoch": 0.188,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.008503214518229166,
+      "learning_rate": 0.0001,
+      "loss": 4.3808,
+      "loss/crossentropy": 2.430112838745117,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23544297367334366,
+      "step": 9400
+    },
+    {
+      "epoch": 0.18804,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.00892333984375,
+      "learning_rate": 0.0001,
+      "loss": 4.2393,
+      "loss/crossentropy": 2.146397888660431,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23211465775966644,
+      "step": 9402
+    },
+    {
+      "epoch": 0.18808,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.008318837483723958,
+      "learning_rate": 0.0001,
+      "loss": 4.5071,
+      "loss/crossentropy": 2.4402170181274414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25594406574964523,
+      "step": 9404
+    },
+    {
+      "epoch": 0.18812,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.008294423421223959,
+      "learning_rate": 0.0001,
+      "loss": 4.3848,
+      "loss/crossentropy": 2.141621232032776,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2156461626291275,
+      "step": 9406
+    },
+    {
+      "epoch": 0.18816,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.006290435791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2511,
+      "loss/crossentropy": 2.0266456604003906,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20194754749536514,
+      "step": 9408
+    },
+    {
+      "epoch": 0.1882,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.007062784830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.2831,
+      "loss/crossentropy": 2.304496645927429,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22345459461212158,
+      "step": 9410
+    },
+    {
+      "epoch": 0.18824,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.007523600260416667,
+      "learning_rate": 0.0001,
+      "loss": 4.4085,
+      "loss/crossentropy": 1.8486470580101013,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20493299514055252,
+      "step": 9412
+    },
+    {
+      "epoch": 0.18828,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0075927734375,
+      "learning_rate": 0.0001,
+      "loss": 4.4562,
+      "loss/crossentropy": 2.0593990683555603,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22942744940519333,
+      "step": 9414
+    },
+    {
+      "epoch": 0.18832,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.006528472900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1551,
+      "loss/crossentropy": 1.7936111688613892,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1776513010263443,
+      "step": 9416
+    },
+    {
+      "epoch": 0.18836,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.005997467041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.4484,
+      "loss/crossentropy": 2.0676616430282593,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21748381853103638,
+      "step": 9418
+    },
+    {
+      "epoch": 0.1884,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.005236562093098958,
+      "learning_rate": 0.0001,
+      "loss": 4.3415,
+      "loss/crossentropy": 2.0317665934562683,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2215682566165924,
+      "step": 9420
+    },
+    {
+      "epoch": 0.18844,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.006359608968098959,
+      "learning_rate": 0.0001,
+      "loss": 4.5623,
+      "loss/crossentropy": 2.3345483541488647,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24850556254386902,
+      "step": 9422
+    },
+    {
+      "epoch": 0.18848,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.007043202718098958,
+      "learning_rate": 0.0001,
+      "loss": 4.3526,
+      "loss/crossentropy": 2.0603779554367065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20997442305088043,
+      "step": 9424
+    },
+    {
+      "epoch": 0.18852,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.006591796875,
+      "learning_rate": 0.0001,
+      "loss": 4.4944,
+      "loss/crossentropy": 1.9450209140777588,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21465806663036346,
+      "step": 9426
+    },
+    {
+      "epoch": 0.18856,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0065266927083333336,
+      "learning_rate": 0.0001,
+      "loss": 4.2452,
+      "loss/crossentropy": 2.3568087816238403,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24619507789611816,
+      "step": 9428
+    },
+    {
+      "epoch": 0.1886,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.007828776041666667,
+      "learning_rate": 0.0001,
+      "loss": 4.1976,
+      "loss/crossentropy": 2.2047020196914673,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2519628629088402,
+      "step": 9430
+    },
+    {
+      "epoch": 0.18864,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.010188547770182292,
+      "learning_rate": 0.0001,
+      "loss": 4.3168,
+      "loss/crossentropy": 2.1594117879867554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21274243295192719,
+      "step": 9432
+    },
+    {
+      "epoch": 0.18868,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.009936269124348958,
+      "learning_rate": 0.0001,
+      "loss": 4.3847,
+      "loss/crossentropy": 2.120336890220642,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2092270776629448,
+      "step": 9434
+    },
+    {
+      "epoch": 0.18872,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.011120351155598958,
+      "learning_rate": 0.0001,
+      "loss": 4.2725,
+      "loss/crossentropy": 2.13198459148407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.238224595785141,
+      "step": 9436
+    },
+    {
+      "epoch": 0.18876,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.010009511311848959,
+      "learning_rate": 0.0001,
+      "loss": 4.3706,
+      "loss/crossentropy": 1.9252395629882812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22254322469234467,
+      "step": 9438
+    },
+    {
+      "epoch": 0.1888,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0123291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0574,
+      "loss/crossentropy": 2.123266577720642,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22276867926120758,
+      "step": 9440
+    },
+    {
+      "epoch": 0.18884,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.010741170247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2195,
+      "loss/crossentropy": 1.9219747185707092,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18884174525737762,
+      "step": 9442
+    },
+    {
+      "epoch": 0.18888,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010587565104166667,
+      "learning_rate": 0.0001,
+      "loss": 4.2603,
+      "loss/crossentropy": 2.0122207403182983,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20913395285606384,
+      "step": 9444
+    },
+    {
+      "epoch": 0.18892,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.010632069905598958,
+      "learning_rate": 0.0001,
+      "loss": 4.0871,
+      "loss/crossentropy": 2.0255361199378967,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1982431635260582,
+      "step": 9446
+    },
+    {
+      "epoch": 0.18896,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.008770497639973958,
+      "learning_rate": 0.0001,
+      "loss": 4.4085,
+      "loss/crossentropy": 1.8060500025749207,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19787351042032242,
+      "step": 9448
+    },
+    {
+      "epoch": 0.189,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.007389068603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3799,
+      "loss/crossentropy": 2.340656042098999,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22321298718452454,
+      "step": 9450
+    },
+    {
+      "epoch": 0.18904,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.00966796875,
+      "learning_rate": 0.0001,
+      "loss": 4.3973,
+      "loss/crossentropy": 2.35786235332489,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2356308028101921,
+      "step": 9452
+    },
+    {
+      "epoch": 0.18908,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.01002197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2622,
+      "loss/crossentropy": 2.344806671142578,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21194154769182205,
+      "step": 9454
+    },
+    {
+      "epoch": 0.18912,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.008786773681640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4861,
+      "loss/crossentropy": 2.2449493408203125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21038557589054108,
+      "step": 9456
+    },
+    {
+      "epoch": 0.18916,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.014518229166666667,
+      "learning_rate": 0.0001,
+      "loss": 4.0343,
+      "loss/crossentropy": 1.977162778377533,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20586547255516052,
+      "step": 9458
+    },
+    {
+      "epoch": 0.1892,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.015860748291015626,
+      "learning_rate": 0.0001,
+      "loss": 4.3453,
+      "loss/crossentropy": 2.0941065549850464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2197253629565239,
+      "step": 9460
+    },
+    {
+      "epoch": 0.18924,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0143463134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3479,
+      "loss/crossentropy": 2.5145565271377563,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25138507783412933,
+      "step": 9462
+    },
+    {
+      "epoch": 0.18928,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.014289347330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4076,
+      "loss/crossentropy": 2.2870718240737915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2298167496919632,
+      "step": 9464
+    },
+    {
+      "epoch": 0.18932,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.013678995768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.3588,
+      "loss/crossentropy": 2.2095978260040283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.217079259455204,
+      "step": 9466
+    },
+    {
+      "epoch": 0.18936,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.011726633707682291,
+      "learning_rate": 0.0001,
+      "loss": 4.3735,
+      "loss/crossentropy": 1.9591819047927856,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21825183182954788,
+      "step": 9468
+    },
+    {
+      "epoch": 0.1894,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.011352284749348959,
+      "learning_rate": 0.0001,
+      "loss": 4.5072,
+      "loss/crossentropy": 2.266845226287842,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24172072112560272,
+      "step": 9470
+    },
+    {
+      "epoch": 0.18944,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.011437733968098959,
+      "learning_rate": 0.0001,
+      "loss": 4.3448,
+      "loss/crossentropy": 2.0732688903808594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23074156790971756,
+      "step": 9472
+    },
+    {
+      "epoch": 0.18948,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.0090728759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.6697,
+      "loss/crossentropy": 1.935340702533722,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2194545865058899,
+      "step": 9474
+    },
+    {
+      "epoch": 0.18952,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.007594553629557291,
+      "learning_rate": 0.0001,
+      "loss": 4.1518,
+      "loss/crossentropy": 1.795669674873352,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18794939666986465,
+      "step": 9476
+    },
+    {
+      "epoch": 0.18956,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.008841705322265626,
+      "learning_rate": 0.0001,
+      "loss": 3.7745,
+      "loss/crossentropy": 1.9292446970939636,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20147182047367096,
+      "step": 9478
+    },
+    {
+      "epoch": 0.1896,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0067779541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1955,
+      "loss/crossentropy": 2.0110061168670654,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2096811756491661,
+      "step": 9480
+    },
+    {
+      "epoch": 0.18964,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.008034006754557291,
+      "learning_rate": 0.0001,
+      "loss": 4.2937,
+      "loss/crossentropy": 2.1543468236923218,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2142205834388733,
+      "step": 9482
+    },
+    {
+      "epoch": 0.18968,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.009723917643229166,
+      "learning_rate": 0.0001,
+      "loss": 4.0126,
+      "loss/crossentropy": 2.0860520601272583,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21716032177209854,
+      "step": 9484
+    },
+    {
+      "epoch": 0.18972,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.010241444905598958,
+      "learning_rate": 0.0001,
+      "loss": 3.9412,
+      "loss/crossentropy": 1.7190409302711487,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19397014379501343,
+      "step": 9486
+    },
+    {
+      "epoch": 0.18976,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.017116038004557292,
+      "learning_rate": 0.0001,
+      "loss": 4.4711,
+      "loss/crossentropy": 2.178081512451172,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2280048429965973,
+      "step": 9488
+    },
+    {
+      "epoch": 0.1898,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.013063303629557292,
+      "learning_rate": 0.0001,
+      "loss": 4.267,
+      "loss/crossentropy": 2.0749863982200623,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22755203396081924,
+      "step": 9490
+    },
+    {
+      "epoch": 0.18984,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.014098866780598959,
+      "learning_rate": 0.0001,
+      "loss": 4.3147,
+      "loss/crossentropy": 2.214204430580139,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23071825504302979,
+      "step": 9492
+    },
+    {
+      "epoch": 0.18988,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.015794881184895835,
+      "learning_rate": 0.0001,
+      "loss": 4.0041,
+      "loss/crossentropy": 1.6625414490699768,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1857389286160469,
+      "step": 9494
+    },
+    {
+      "epoch": 0.18992,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.020539347330729166,
+      "learning_rate": 0.0001,
+      "loss": 4.4474,
+      "loss/crossentropy": 1.8537201285362244,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21718977391719818,
+      "step": 9496
+    },
+    {
+      "epoch": 0.18996,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.024074045817057292,
+      "learning_rate": 0.0001,
+      "loss": 4.0488,
+      "loss/crossentropy": 1.716725468635559,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18544895946979523,
+      "step": 9498
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.024192047119140626,
+      "learning_rate": 0.0001,
+      "loss": 4.1035,
+      "loss/crossentropy": 1.94467431306839,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.210828959941864,
+      "step": 9500
+    },
+    {
+      "epoch": 0.19004,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.023298136393229165,
+      "learning_rate": 0.0001,
+      "loss": 4.2383,
+      "loss/crossentropy": 1.9377062320709229,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22141354531049728,
+      "step": 9502
+    },
+    {
+      "epoch": 0.19008,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.016950480143229165,
+      "learning_rate": 0.0001,
+      "loss": 4.3717,
+      "loss/crossentropy": 2.2015358209609985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22363336384296417,
+      "step": 9504
+    },
+    {
+      "epoch": 0.19012,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.017235310872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2332,
+      "loss/crossentropy": 2.373024582862854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2402234748005867,
+      "step": 9506
+    },
+    {
+      "epoch": 0.19016,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.016649373372395835,
+      "learning_rate": 0.0001,
+      "loss": 4.0519,
+      "loss/crossentropy": 2.1573110222816467,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24207139760255814,
+      "step": 9508
+    },
+    {
+      "epoch": 0.1902,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.014788564046223958,
+      "learning_rate": 0.0001,
+      "loss": 4.4041,
+      "loss/crossentropy": 2.324455976486206,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24618541449308395,
+      "step": 9510
+    },
+    {
+      "epoch": 0.19024,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010603586832682291,
+      "learning_rate": 0.0001,
+      "loss": 4.0065,
+      "loss/crossentropy": 1.7480111718177795,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2049119919538498,
+      "step": 9512
+    },
+    {
+      "epoch": 0.19028,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.006300608317057292,
+      "learning_rate": 0.0001,
+      "loss": 4.3644,
+      "loss/crossentropy": 1.9925439953804016,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2314591035246849,
+      "step": 9514
+    },
+    {
+      "epoch": 0.19032,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.007094065348307292,
+      "learning_rate": 0.0001,
+      "loss": 4.0702,
+      "loss/crossentropy": 2.133601188659668,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2329796403646469,
+      "step": 9516
+    },
+    {
+      "epoch": 0.19036,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.010198720296223958,
+      "learning_rate": 0.0001,
+      "loss": 4.246,
+      "loss/crossentropy": 2.093464970588684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22636859863996506,
+      "step": 9518
+    },
+    {
+      "epoch": 0.1904,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009757232666015626,
+      "learning_rate": 0.0001,
+      "loss": 4.4766,
+      "loss/crossentropy": 2.6137614250183105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27037859708070755,
+      "step": 9520
+    },
+    {
+      "epoch": 0.19044,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.012300364176432292,
+      "learning_rate": 0.0001,
+      "loss": 4.5499,
+      "loss/crossentropy": 2.008640229701996,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21647297590970993,
+      "step": 9522
+    },
+    {
+      "epoch": 0.19048,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.012286122639973958,
+      "learning_rate": 0.0001,
+      "loss": 4.2467,
+      "loss/crossentropy": 2.12644362449646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23370585590600967,
+      "step": 9524
+    },
+    {
+      "epoch": 0.19052,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011425526936848958,
+      "learning_rate": 0.0001,
+      "loss": 4.5409,
+      "loss/crossentropy": 2.2338638305664062,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23637598007917404,
+      "step": 9526
+    },
+    {
+      "epoch": 0.19056,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.011785634358723958,
+      "learning_rate": 0.0001,
+      "loss": 4.1405,
+      "loss/crossentropy": 2.0632832646369934,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20682457089424133,
+      "step": 9528
+    },
+    {
+      "epoch": 0.1906,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.012726847330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4058,
+      "loss/crossentropy": 2.219120740890503,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22551076859235764,
+      "step": 9530
+    },
+    {
+      "epoch": 0.19064,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.011860911051432292,
+      "learning_rate": 0.0001,
+      "loss": 4.2985,
+      "loss/crossentropy": 2.4633371829986572,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24491076171398163,
+      "step": 9532
+    },
+    {
+      "epoch": 0.19068,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008211008707682292,
+      "learning_rate": 0.0001,
+      "loss": 4.2152,
+      "loss/crossentropy": 2.2408339977264404,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22265981882810593,
+      "step": 9534
+    },
+    {
+      "epoch": 0.19072,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.009544881184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.326,
+      "loss/crossentropy": 1.9779353141784668,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2159332111477852,
+      "step": 9536
+    },
+    {
+      "epoch": 0.19076,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.0082275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3522,
+      "loss/crossentropy": 2.0315812826156616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20710154622793198,
+      "step": 9538
+    },
+    {
+      "epoch": 0.1908,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.011787923177083333,
+      "learning_rate": 0.0001,
+      "loss": 4.18,
+      "loss/crossentropy": 2.085337817668915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22174393385648727,
+      "step": 9540
+    },
+    {
+      "epoch": 0.19084,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.016507975260416665,
+      "learning_rate": 0.0001,
+      "loss": 4.4386,
+      "loss/crossentropy": 2.3449169397354126,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3135230466723442,
+      "step": 9542
+    },
+    {
+      "epoch": 0.19088,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.016169230143229168,
+      "learning_rate": 0.0001,
+      "loss": 4.3943,
+      "loss/crossentropy": 2.1083431243896484,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21395261585712433,
+      "step": 9544
+    },
+    {
+      "epoch": 0.19092,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.015730539957682293,
+      "learning_rate": 0.0001,
+      "loss": 4.2091,
+      "loss/crossentropy": 2.0386710166931152,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22487390786409378,
+      "step": 9546
+    },
+    {
+      "epoch": 0.19096,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.015240224202473958,
+      "learning_rate": 0.0001,
+      "loss": 4.2377,
+      "loss/crossentropy": 1.9533087611198425,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2223353162407875,
+      "step": 9548
+    },
+    {
+      "epoch": 0.191,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.015317535400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.4183,
+      "loss/crossentropy": 2.35861599445343,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26162558794021606,
+      "step": 9550
+    },
+    {
+      "epoch": 0.19104,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.014388020833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.2507,
+      "loss/crossentropy": 1.9529705047607422,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2125585675239563,
+      "step": 9552
+    },
+    {
+      "epoch": 0.19108,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011844889322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.2041,
+      "loss/crossentropy": 2.0364453196525574,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21163037419319153,
+      "step": 9554
+    },
+    {
+      "epoch": 0.19112,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009749348958333333,
+      "learning_rate": 0.0001,
+      "loss": 4.2758,
+      "loss/crossentropy": 2.1321409940719604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24250106513500214,
+      "step": 9556
+    },
+    {
+      "epoch": 0.19116,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0073931376139322914,
+      "learning_rate": 0.0001,
+      "loss": 4.1848,
+      "loss/crossentropy": 2.024011969566345,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20132286846637726,
+      "step": 9558
+    },
+    {
+      "epoch": 0.1912,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.006980133056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3726,
+      "loss/crossentropy": 2.106776535511017,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23114337772130966,
+      "step": 9560
+    },
+    {
+      "epoch": 0.19124,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.007645416259765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4642,
+      "loss/crossentropy": 1.9228236079216003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22426098585128784,
+      "step": 9562
+    },
+    {
+      "epoch": 0.19128,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.008090972900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2384,
+      "loss/crossentropy": 2.3033370971679688,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20113499462604523,
+      "step": 9564
+    },
+    {
+      "epoch": 0.19132,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.008973948160807292,
+      "learning_rate": 0.0001,
+      "loss": 4.0736,
+      "loss/crossentropy": 1.6983963251113892,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19003060460090637,
+      "step": 9566
+    },
+    {
+      "epoch": 0.19136,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.011295318603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5283,
+      "loss/crossentropy": 2.502004861831665,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24949797987937927,
+      "step": 9568
+    },
+    {
+      "epoch": 0.1914,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011793772379557291,
+      "learning_rate": 0.0001,
+      "loss": 4.2004,
+      "loss/crossentropy": 1.9981504678726196,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2152494639158249,
+      "step": 9570
+    },
+    {
+      "epoch": 0.19144,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.012931060791015626,
+      "learning_rate": 0.0001,
+      "loss": 4.1211,
+      "loss/crossentropy": 2.1489784717559814,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22105325013399124,
+      "step": 9572
+    },
+    {
+      "epoch": 0.19148,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.013016510009765624,
+      "learning_rate": 0.0001,
+      "loss": 4.2234,
+      "loss/crossentropy": 1.974421203136444,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23960395902395248,
+      "step": 9574
+    },
+    {
+      "epoch": 0.19152,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.012670644124348958,
+      "learning_rate": 0.0001,
+      "loss": 4.4119,
+      "loss/crossentropy": 2.379599928855896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26731471717357635,
+      "step": 9576
+    },
+    {
+      "epoch": 0.19156,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.012444814046223959,
+      "learning_rate": 0.0001,
+      "loss": 4.3718,
+      "loss/crossentropy": 2.26211154460907,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22957566380500793,
+      "step": 9578
+    },
+    {
+      "epoch": 0.1916,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.011871083577473959,
+      "learning_rate": 0.0001,
+      "loss": 4.4594,
+      "loss/crossentropy": 2.282869577407837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2404445931315422,
+      "step": 9580
+    },
+    {
+      "epoch": 0.19164,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.010457102457682292,
+      "learning_rate": 0.0001,
+      "loss": 4.1592,
+      "loss/crossentropy": 1.9183810949325562,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18672315031290054,
+      "step": 9582
+    },
+    {
+      "epoch": 0.19168,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0076812744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1801,
+      "loss/crossentropy": 2.2722173929214478,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23262400180101395,
+      "step": 9584
+    },
+    {
+      "epoch": 0.19172,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0075266520182291664,
+      "learning_rate": 0.0001,
+      "loss": 4.4032,
+      "loss/crossentropy": 2.426178455352783,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23461100459098816,
+      "step": 9586
+    },
+    {
+      "epoch": 0.19176,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.00687255859375,
+      "learning_rate": 0.0001,
+      "loss": 4.344,
+      "loss/crossentropy": 2.2266165018081665,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22530251741409302,
+      "step": 9588
+    },
+    {
+      "epoch": 0.1918,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.005541737874348958,
+      "learning_rate": 0.0001,
+      "loss": 4.1633,
+      "loss/crossentropy": 1.786275327205658,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19257958233356476,
+      "step": 9590
+    },
+    {
+      "epoch": 0.19184,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.008868153889973958,
+      "learning_rate": 0.0001,
+      "loss": 4.2367,
+      "loss/crossentropy": 1.8497431874275208,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21965742111206055,
+      "step": 9592
+    },
+    {
+      "epoch": 0.19188,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.008143870035807292,
+      "learning_rate": 0.0001,
+      "loss": 4.5612,
+      "loss/crossentropy": 2.492846131324768,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21851783990859985,
+      "step": 9594
+    },
+    {
+      "epoch": 0.19192,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.010027821858723958,
+      "learning_rate": 0.0001,
+      "loss": 4.6838,
+      "loss/crossentropy": 2.3447986841201782,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26567359268665314,
+      "step": 9596
+    },
+    {
+      "epoch": 0.19196,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.010654449462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1513,
+      "loss/crossentropy": 1.656063199043274,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1903422325849533,
+      "step": 9598
+    },
+    {
+      "epoch": 0.192,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.008329264322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.3252,
+      "loss/crossentropy": 2.1816134452819824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23314762860536575,
+      "step": 9600
+    },
+    {
+      "epoch": 0.19204,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010007476806640625,
+      "learning_rate": 0.0001,
+      "loss": 3.8537,
+      "loss/crossentropy": 1.787261426448822,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20938758552074432,
+      "step": 9602
+    },
+    {
+      "epoch": 0.19208,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010526275634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2374,
+      "loss/crossentropy": 1.9722678065299988,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23085469752550125,
+      "step": 9604
+    },
+    {
+      "epoch": 0.19212,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.012113444010416667,
+      "learning_rate": 0.0001,
+      "loss": 4.0954,
+      "loss/crossentropy": 2.04559987783432,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21647220849990845,
+      "step": 9606
+    },
+    {
+      "epoch": 0.19216,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.008934529622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.5397,
+      "loss/crossentropy": 2.5426105260849,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23680832237005234,
+      "step": 9608
+    },
+    {
+      "epoch": 0.1922,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.008890787760416666,
+      "learning_rate": 0.0001,
+      "loss": 4.245,
+      "loss/crossentropy": 2.1339075565338135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22501112520694733,
+      "step": 9610
+    },
+    {
+      "epoch": 0.19224,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.006550852457682292,
+      "learning_rate": 0.0001,
+      "loss": 4.0924,
+      "loss/crossentropy": 1.9554831981658936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20187357813119888,
+      "step": 9612
+    },
+    {
+      "epoch": 0.19228,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.009065500895182292,
+      "learning_rate": 0.0001,
+      "loss": 4.5578,
+      "loss/crossentropy": 2.2996249198913574,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24954287707805634,
+      "step": 9614
+    },
+    {
+      "epoch": 0.19232,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.009065500895182292,
+      "learning_rate": 0.0001,
+      "loss": 4.3753,
+      "loss/crossentropy": 2.2439414262771606,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22833245247602463,
+      "step": 9616
+    },
+    {
+      "epoch": 0.19236,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.007933553059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2297,
+      "loss/crossentropy": 1.8900890946388245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21077623218297958,
+      "step": 9618
+    },
+    {
+      "epoch": 0.1924,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.007804361979166666,
+      "learning_rate": 0.0001,
+      "loss": 4.287,
+      "loss/crossentropy": 2.110726058483124,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21897974610328674,
+      "step": 9620
+    },
+    {
+      "epoch": 0.19244,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0080474853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0135,
+      "loss/crossentropy": 1.7363090515136719,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19939633458852768,
+      "step": 9622
+    },
+    {
+      "epoch": 0.19248,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.01014404296875,
+      "learning_rate": 0.0001,
+      "loss": 4.3643,
+      "loss/crossentropy": 1.8148014545440674,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21637701243162155,
+      "step": 9624
+    },
+    {
+      "epoch": 0.19252,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0092041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2451,
+      "loss/crossentropy": 2.2895134687423706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2297188639640808,
+      "step": 9626
+    },
+    {
+      "epoch": 0.19256,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.006296539306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4778,
+      "loss/crossentropy": 2.117924213409424,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24057473242282867,
+      "step": 9628
+    },
+    {
+      "epoch": 0.1926,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.005421702067057292,
+      "learning_rate": 0.0001,
+      "loss": 4.2791,
+      "loss/crossentropy": 1.8709319829940796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2106107696890831,
+      "step": 9630
+    },
+    {
+      "epoch": 0.19264,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0079742431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0174,
+      "loss/crossentropy": 1.5156871676445007,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17487400770187378,
+      "step": 9632
+    },
+    {
+      "epoch": 0.19268,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0183837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.6281,
+      "loss/crossentropy": 2.153126537799835,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22009101510047913,
+      "step": 9634
+    },
+    {
+      "epoch": 0.19272,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.017853800455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.443,
+      "loss/crossentropy": 2.0890414714813232,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2233077436685562,
+      "step": 9636
+    },
+    {
+      "epoch": 0.19276,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.01858495076497396,
+      "learning_rate": 0.0001,
+      "loss": 4.2572,
+      "loss/crossentropy": 2.1218496561050415,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22576671838760376,
+      "step": 9638
+    },
+    {
+      "epoch": 0.1928,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.020918528238932293,
+      "learning_rate": 0.0001,
+      "loss": 4.2522,
+      "loss/crossentropy": 2.1131649017333984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21846124529838562,
+      "step": 9640
+    },
+    {
+      "epoch": 0.19284,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.02067845662434896,
+      "learning_rate": 0.0001,
+      "loss": 4.6643,
+      "loss/crossentropy": 2.3941714763641357,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.246211439371109,
+      "step": 9642
+    },
+    {
+      "epoch": 0.19288,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.02687352498372396,
+      "learning_rate": 0.0001,
+      "loss": 4.0641,
+      "loss/crossentropy": 1.9579638838768005,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.198220357298851,
+      "step": 9644
+    },
+    {
+      "epoch": 0.19292,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.02754491170247396,
+      "learning_rate": 0.0001,
+      "loss": 4.564,
+      "loss/crossentropy": 2.388027787208557,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23349716514348984,
+      "step": 9646
+    },
+    {
+      "epoch": 0.19296,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.02520726521809896,
+      "learning_rate": 0.0001,
+      "loss": 4.3678,
+      "loss/crossentropy": 1.8203087449073792,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18923642486333847,
+      "step": 9648
+    },
+    {
+      "epoch": 0.193,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.015476226806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2119,
+      "loss/crossentropy": 1.996269702911377,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2096453383564949,
+      "step": 9650
+    },
+    {
+      "epoch": 0.19304,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.01572240193684896,
+      "learning_rate": 0.0001,
+      "loss": 4.2921,
+      "loss/crossentropy": 1.806606113910675,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19824761897325516,
+      "step": 9652
+    },
+    {
+      "epoch": 0.19308,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.01651178995768229,
+      "learning_rate": 0.0001,
+      "loss": 4.1681,
+      "loss/crossentropy": 2.190830111503601,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22319861501455307,
+      "step": 9654
+    },
+    {
+      "epoch": 0.19312,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.014048004150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.4579,
+      "loss/crossentropy": 1.9721493124961853,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2011292800307274,
+      "step": 9656
+    },
+    {
+      "epoch": 0.19316,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.014277903238932292,
+      "learning_rate": 0.0001,
+      "loss": 4.5673,
+      "loss/crossentropy": 2.1256929636001587,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2728194147348404,
+      "step": 9658
+    },
+    {
+      "epoch": 0.1932,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.008345286051432291,
+      "learning_rate": 0.0001,
+      "loss": 4.3206,
+      "loss/crossentropy": 2.090156316757202,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2240355908870697,
+      "step": 9660
+    },
+    {
+      "epoch": 0.19324,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0108795166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0231,
+      "loss/crossentropy": 2.2930272817611694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23188824206590652,
+      "step": 9662
+    },
+    {
+      "epoch": 0.19328,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0122222900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0075,
+      "loss/crossentropy": 1.9754068851470947,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18698863685131073,
+      "step": 9664
+    },
+    {
+      "epoch": 0.19332,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0118560791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1668,
+      "loss/crossentropy": 1.8987788558006287,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22531752288341522,
+      "step": 9666
+    },
+    {
+      "epoch": 0.19336,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.011678059895833334,
+      "learning_rate": 0.0001,
+      "loss": 4.3185,
+      "loss/crossentropy": 2.2699583768844604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2572309076786041,
+      "step": 9668
+    },
+    {
+      "epoch": 0.1934,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0072509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3267,
+      "loss/crossentropy": 1.6649349927902222,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23388104140758514,
+      "step": 9670
+    },
+    {
+      "epoch": 0.19344,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.007258097330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3846,
+      "loss/crossentropy": 2.173617362976074,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20848755538463593,
+      "step": 9672
+    },
+    {
+      "epoch": 0.19348,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.005890909830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3252,
+      "loss/crossentropy": 2.2690787315368652,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23493453860282898,
+      "step": 9674
+    },
+    {
+      "epoch": 0.19352,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.007079060872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3621,
+      "loss/crossentropy": 2.00560861825943,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22538188099861145,
+      "step": 9676
+    },
+    {
+      "epoch": 0.19356,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.004369862874348958,
+      "learning_rate": 0.0001,
+      "loss": 4.1264,
+      "loss/crossentropy": 1.960120975971222,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21056914329528809,
+      "step": 9678
+    },
+    {
+      "epoch": 0.1936,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.004689280192057292,
+      "learning_rate": 0.0001,
+      "loss": 3.9868,
+      "loss/crossentropy": 1.8921862840652466,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22250327467918396,
+      "step": 9680
+    },
+    {
+      "epoch": 0.19364,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.004839833577473958,
+      "learning_rate": 0.0001,
+      "loss": 4.296,
+      "loss/crossentropy": 1.9474233984947205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19719959795475006,
+      "step": 9682
+    },
+    {
+      "epoch": 0.19368,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0072100321451822914,
+      "learning_rate": 0.0001,
+      "loss": 4.2985,
+      "loss/crossentropy": 2.3391844034194946,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22263485193252563,
+      "step": 9684
+    },
+    {
+      "epoch": 0.19372,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.008715565999348958,
+      "learning_rate": 0.0001,
+      "loss": 4.3641,
+      "loss/crossentropy": 2.190012037754059,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22531607002019882,
+      "step": 9686
+    },
+    {
+      "epoch": 0.19376,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.008283487955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.2601,
+      "loss/crossentropy": 1.9935640096664429,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21438511461019516,
+      "step": 9688
+    },
+    {
+      "epoch": 0.1938,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.009186808268229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1756,
+      "loss/crossentropy": 1.7482191324234009,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20972990244627,
+      "step": 9690
+    },
+    {
+      "epoch": 0.19384,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.008353678385416667,
+      "learning_rate": 0.0001,
+      "loss": 4.475,
+      "loss/crossentropy": 2.2413275241851807,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24103231728076935,
+      "step": 9692
+    },
+    {
+      "epoch": 0.19388,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.008519490559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.2886,
+      "loss/crossentropy": 2.2846572399139404,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2420315518975258,
+      "step": 9694
+    },
+    {
+      "epoch": 0.19392,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.007968902587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1665,
+      "loss/crossentropy": 1.7977504134178162,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20145538449287415,
+      "step": 9696
+    },
+    {
+      "epoch": 0.19396,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.007535552978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.205,
+      "loss/crossentropy": 2.0343902111053467,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21595563739538193,
+      "step": 9698
+    },
+    {
+      "epoch": 0.194,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0052487691243489586,
+      "learning_rate": 0.0001,
+      "loss": 3.9524,
+      "loss/crossentropy": 1.8828233480453491,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19624938070774078,
+      "step": 9700
+    },
+    {
+      "epoch": 0.19404,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.005602773030598958,
+      "learning_rate": 0.0001,
+      "loss": 4.1569,
+      "loss/crossentropy": 1.9177573323249817,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19467243552207947,
+      "step": 9702
+    },
+    {
+      "epoch": 0.19408,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.006613922119140625,
+      "learning_rate": 0.0001,
+      "loss": 3.8881,
+      "loss/crossentropy": 1.8025588393211365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18757501989603043,
+      "step": 9704
+    },
+    {
+      "epoch": 0.19412,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.006723785400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3195,
+      "loss/crossentropy": 2.025633454322815,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21623078733682632,
+      "step": 9706
+    },
+    {
+      "epoch": 0.19416,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009129842122395834,
+      "learning_rate": 0.0001,
+      "loss": 3.8244,
+      "loss/crossentropy": 1.9421055316925049,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20563311874866486,
+      "step": 9708
+    },
+    {
+      "epoch": 0.1942,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.009191640218098958,
+      "learning_rate": 0.0001,
+      "loss": 4.0977,
+      "loss/crossentropy": 1.9820671081542969,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22029083967208862,
+      "step": 9710
+    },
+    {
+      "epoch": 0.19424,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.010341135660807292,
+      "learning_rate": 0.0001,
+      "loss": 4.2037,
+      "loss/crossentropy": 1.9491158723831177,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19559209793806076,
+      "step": 9712
+    },
+    {
+      "epoch": 0.19428,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010416412353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1592,
+      "loss/crossentropy": 1.8653306365013123,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18629660457372665,
+      "step": 9714
+    },
+    {
+      "epoch": 0.19432,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.01950251261393229,
+      "learning_rate": 0.0001,
+      "loss": 4.2567,
+      "loss/crossentropy": 1.6897491812705994,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1948539912700653,
+      "step": 9716
+    },
+    {
+      "epoch": 0.19436,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.020776112874348957,
+      "learning_rate": 0.0001,
+      "loss": 4.4387,
+      "loss/crossentropy": 2.07690966129303,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2147115021944046,
+      "step": 9718
+    },
+    {
+      "epoch": 0.1944,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.020189412434895835,
+      "learning_rate": 0.0001,
+      "loss": 4.269,
+      "loss/crossentropy": 2.0356597304344177,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20626354217529297,
+      "step": 9720
+    },
+    {
+      "epoch": 0.19444,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0204254150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0548,
+      "loss/crossentropy": 1.7709991931915283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18573015183210373,
+      "step": 9722
+    },
+    {
+      "epoch": 0.19448,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.01830012003580729,
+      "learning_rate": 0.0001,
+      "loss": 4.117,
+      "loss/crossentropy": 2.0255925059318542,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22114800661802292,
+      "step": 9724
+    },
+    {
+      "epoch": 0.19452,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.01871337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4655,
+      "loss/crossentropy": 2.0046772956848145,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22649522870779037,
+      "step": 9726
+    },
+    {
+      "epoch": 0.19456,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.015827433268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.179,
+      "loss/crossentropy": 2.2746634483337402,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23017344623804092,
+      "step": 9728
+    },
+    {
+      "epoch": 0.1946,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.015184529622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.203,
+      "loss/crossentropy": 2.0191025137901306,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20670472085475922,
+      "step": 9730
+    },
+    {
+      "epoch": 0.19464,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008177693684895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1657,
+      "loss/crossentropy": 2.3198455572128296,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24985665082931519,
+      "step": 9732
+    },
+    {
+      "epoch": 0.19468,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0073150634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3886,
+      "loss/crossentropy": 2.5229711532592773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22920701652765274,
+      "step": 9734
+    },
+    {
+      "epoch": 0.19472,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0072662353515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9991,
+      "loss/crossentropy": 2.081319808959961,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21542686223983765,
+      "step": 9736
+    },
+    {
+      "epoch": 0.19476,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.008373006184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.0698,
+      "loss/crossentropy": 2.2170007824897766,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21852095425128937,
+      "step": 9738
+    },
+    {
+      "epoch": 0.1948,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.008265940348307292,
+      "learning_rate": 0.0001,
+      "loss": 4.1528,
+      "loss/crossentropy": 1.830683708190918,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1910252571105957,
+      "step": 9740
+    },
+    {
+      "epoch": 0.19484,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.008420562744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3543,
+      "loss/crossentropy": 2.1303864121437073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.228670135140419,
+      "step": 9742
+    },
+    {
+      "epoch": 0.19488,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007486724853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1504,
+      "loss/crossentropy": 2.2621915340423584,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22027206420898438,
+      "step": 9744
+    },
+    {
+      "epoch": 0.19492,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0110260009765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9564,
+      "loss/crossentropy": 2.044555902481079,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20469766855239868,
+      "step": 9746
+    },
+    {
+      "epoch": 0.19496,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0115142822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1677,
+      "loss/crossentropy": 1.911176860332489,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22036674618721008,
+      "step": 9748
+    },
+    {
+      "epoch": 0.195,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.010545857747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0904,
+      "loss/crossentropy": 1.8650219440460205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19878911972045898,
+      "step": 9750
+    },
+    {
+      "epoch": 0.19504,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.02088623046875,
+      "learning_rate": 0.0001,
+      "loss": 4.3648,
+      "loss/crossentropy": 2.196391463279724,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20845064520835876,
+      "step": 9752
+    },
+    {
+      "epoch": 0.19508,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.021201324462890626,
+      "learning_rate": 0.0001,
+      "loss": 4.0132,
+      "loss/crossentropy": 2.222484588623047,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2144618257880211,
+      "step": 9754
+    },
+    {
+      "epoch": 0.19512,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.020685831705729168,
+      "learning_rate": 0.0001,
+      "loss": 4.4078,
+      "loss/crossentropy": 2.416514754295349,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2370949387550354,
+      "step": 9756
+    },
+    {
+      "epoch": 0.19516,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.020566558837890624,
+      "learning_rate": 0.0001,
+      "loss": 4.0119,
+      "loss/crossentropy": 2.22190260887146,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23128122836351395,
+      "step": 9758
+    },
+    {
+      "epoch": 0.1952,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.024008941650390626,
+      "learning_rate": 0.0001,
+      "loss": 4.4253,
+      "loss/crossentropy": 2.446492910385132,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23962965607643127,
+      "step": 9760
+    },
+    {
+      "epoch": 0.19524,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.019245402018229166,
+      "learning_rate": 0.0001,
+      "loss": 4.1676,
+      "loss/crossentropy": 2.1458136439323425,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19657295942306519,
+      "step": 9762
+    },
+    {
+      "epoch": 0.19528,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0197418212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.6029,
+      "loss/crossentropy": 1.9773340225219727,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22887050360441208,
+      "step": 9764
+    },
+    {
+      "epoch": 0.19532,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.020334625244140626,
+      "learning_rate": 0.0001,
+      "loss": 4.1583,
+      "loss/crossentropy": 2.0938061475753784,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19823112338781357,
+      "step": 9766
+    },
+    {
+      "epoch": 0.19536,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.015636952718098958,
+      "learning_rate": 0.0001,
+      "loss": 4.2628,
+      "loss/crossentropy": 1.9068174958229065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19924984872341156,
+      "step": 9768
+    },
+    {
+      "epoch": 0.1954,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012422688802083333,
+      "learning_rate": 0.0001,
+      "loss": 4.3567,
+      "loss/crossentropy": 2.045413613319397,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19175175577402115,
+      "step": 9770
+    },
+    {
+      "epoch": 0.19544,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.013106282552083333,
+      "learning_rate": 0.0001,
+      "loss": 4.1972,
+      "loss/crossentropy": 2.0262961983680725,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22137057781219482,
+      "step": 9772
+    },
+    {
+      "epoch": 0.19548,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.012992350260416667,
+      "learning_rate": 0.0001,
+      "loss": 4.0624,
+      "loss/crossentropy": 2.146475672721863,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2284562587738037,
+      "step": 9774
+    },
+    {
+      "epoch": 0.19552,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.011885579427083333,
+      "learning_rate": 0.0001,
+      "loss": 4.154,
+      "loss/crossentropy": 2.0316100120544434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26275022327899933,
+      "step": 9776
+    },
+    {
+      "epoch": 0.19556,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.011359659830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.5424,
+      "loss/crossentropy": 2.1386696100234985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22695952653884888,
+      "step": 9778
+    },
+    {
+      "epoch": 0.1956,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.024405924479166667,
+      "learning_rate": 0.0001,
+      "loss": 4.2391,
+      "loss/crossentropy": 2.3000820875167847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24045731872320175,
+      "step": 9780
+    },
+    {
+      "epoch": 0.19564,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.02415949503580729,
+      "learning_rate": 0.0001,
+      "loss": 4.7778,
+      "loss/crossentropy": 2.1272310614585876,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2318389192223549,
+      "step": 9782
+    },
+    {
+      "epoch": 0.19568,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.025655110677083332,
+      "learning_rate": 0.0001,
+      "loss": 4.1232,
+      "loss/crossentropy": 1.8953965306282043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.213637076318264,
+      "step": 9784
+    },
+    {
+      "epoch": 0.19572,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0261871337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3459,
+      "loss/crossentropy": 2.0738734006881714,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2100028172135353,
+      "step": 9786
+    },
+    {
+      "epoch": 0.19576,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.027378082275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2599,
+      "loss/crossentropy": 1.9454593658447266,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18557409197092056,
+      "step": 9788
+    },
+    {
+      "epoch": 0.1958,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.026569620768229166,
+      "learning_rate": 0.0001,
+      "loss": 4.1655,
+      "loss/crossentropy": 2.101949095726013,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22479471564292908,
+      "step": 9790
+    },
+    {
+      "epoch": 0.19584,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.025935872395833334,
+      "learning_rate": 0.0001,
+      "loss": 4.3841,
+      "loss/crossentropy": 1.9524416327476501,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20574256777763367,
+      "step": 9792
+    },
+    {
+      "epoch": 0.19588,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.06084391276041667,
+      "learning_rate": 0.0001,
+      "loss": 4.1596,
+      "loss/crossentropy": 1.9490719437599182,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19411193579435349,
+      "step": 9794
+    },
+    {
+      "epoch": 0.19592,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.050675455729166666,
+      "learning_rate": 0.0001,
+      "loss": 4.3438,
+      "loss/crossentropy": 2.095793664455414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21078093349933624,
+      "step": 9796
+    },
+    {
+      "epoch": 0.19596,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0508056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0828,
+      "loss/crossentropy": 2.032066822052002,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21181444078683853,
+      "step": 9798
+    },
+    {
+      "epoch": 0.196,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.048378245035807295,
+      "learning_rate": 0.0001,
+      "loss": 4.0019,
+      "loss/crossentropy": 2.1378380060195923,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.224056214094162,
+      "step": 9800
+    },
+    {
+      "epoch": 0.19604,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.04784520467122396,
+      "learning_rate": 0.0001,
+      "loss": 4.2932,
+      "loss/crossentropy": 2.120614767074585,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2104548141360283,
+      "step": 9802
+    },
+    {
+      "epoch": 0.19608,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.048954010009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1975,
+      "loss/crossentropy": 1.8837141394615173,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19755827635526657,
+      "step": 9804
+    },
+    {
+      "epoch": 0.19612,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.04793675740559896,
+      "learning_rate": 0.0001,
+      "loss": 4.1645,
+      "loss/crossentropy": 1.9325945973396301,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2005266472697258,
+      "step": 9806
+    },
+    {
+      "epoch": 0.19616,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.04839045206705729,
+      "learning_rate": 0.0001,
+      "loss": 4.2211,
+      "loss/crossentropy": 1.789370834827423,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20060084015130997,
+      "step": 9808
+    },
+    {
+      "epoch": 0.1962,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.01709162394205729,
+      "learning_rate": 0.0001,
+      "loss": 4.2463,
+      "loss/crossentropy": 1.9807876348495483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21805762499570847,
+      "step": 9810
+    },
+    {
+      "epoch": 0.19624,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.018888092041015624,
+      "learning_rate": 0.0001,
+      "loss": 4.1339,
+      "loss/crossentropy": 2.210070848464966,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23082506656646729,
+      "step": 9812
+    },
+    {
+      "epoch": 0.19628,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.01693903605143229,
+      "learning_rate": 0.0001,
+      "loss": 4.1633,
+      "loss/crossentropy": 1.8644117712974548,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20300551503896713,
+      "step": 9814
+    },
+    {
+      "epoch": 0.19632,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.01634496053059896,
+      "learning_rate": 0.0001,
+      "loss": 4.2064,
+      "loss/crossentropy": 1.6804233193397522,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20438820868730545,
+      "step": 9816
+    },
+    {
+      "epoch": 0.19636,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.01757990519205729,
+      "learning_rate": 0.0001,
+      "loss": 4.4339,
+      "loss/crossentropy": 2.2513452768325806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24517202377319336,
+      "step": 9818
+    },
+    {
+      "epoch": 0.1964,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.01683527628580729,
+      "learning_rate": 0.0001,
+      "loss": 4.2238,
+      "loss/crossentropy": 2.3133161067962646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22146832942962646,
+      "step": 9820
+    },
+    {
+      "epoch": 0.19644,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.016949208577473958,
+      "learning_rate": 0.0001,
+      "loss": 4.5003,
+      "loss/crossentropy": 2.3226611614227295,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2327018678188324,
+      "step": 9822
+    },
+    {
+      "epoch": 0.19648,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.016228993733723957,
+      "learning_rate": 0.0001,
+      "loss": 4.1832,
+      "loss/crossentropy": 2.234626054763794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23657850921154022,
+      "step": 9824
+    },
+    {
+      "epoch": 0.19652,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.008941396077473959,
+      "learning_rate": 0.0001,
+      "loss": 4.4576,
+      "loss/crossentropy": 2.3478844165802,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24796272069215775,
+      "step": 9826
+    },
+    {
+      "epoch": 0.19656,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.006982167561848958,
+      "learning_rate": 0.0001,
+      "loss": 4.3678,
+      "loss/crossentropy": 2.094748795032501,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21118396520614624,
+      "step": 9828
+    },
+    {
+      "epoch": 0.1966,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.007289377848307291,
+      "learning_rate": 0.0001,
+      "loss": 4.2916,
+      "loss/crossentropy": 2.166300058364868,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21773284673690796,
+      "step": 9830
+    },
+    {
+      "epoch": 0.19664,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.007212066650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.5253,
+      "loss/crossentropy": 2.198317289352417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22636590898036957,
+      "step": 9832
+    },
+    {
+      "epoch": 0.19668,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.006278228759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3405,
+      "loss/crossentropy": 2.3081597089767456,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.231675922870636,
+      "step": 9834
+    },
+    {
+      "epoch": 0.19672,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.005206044514973958,
+      "learning_rate": 0.0001,
+      "loss": 4.1181,
+      "loss/crossentropy": 2.048017203807831,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20089948922395706,
+      "step": 9836
+    },
+    {
+      "epoch": 0.19676,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.01768773396809896,
+      "learning_rate": 0.0001,
+      "loss": 4.0515,
+      "loss/crossentropy": 1.886117160320282,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21702590584754944,
+      "step": 9838
+    },
+    {
+      "epoch": 0.1968,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.017895253499348958,
+      "learning_rate": 0.0001,
+      "loss": 3.9827,
+      "loss/crossentropy": 1.902605414390564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19976041465997696,
+      "step": 9840
+    },
+    {
+      "epoch": 0.19684,
+      "grad_norm": 4.28125,
+      "grad_norm_var": 0.309179433186849,
+      "learning_rate": 0.0001,
+      "loss": 4.0874,
+      "loss/crossentropy": 1.7959995865821838,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2023170217871666,
+      "step": 9842
+    },
+    {
+      "epoch": 0.19688,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.3066993713378906,
+      "learning_rate": 0.0001,
+      "loss": 4.5535,
+      "loss/crossentropy": 2.1481738090515137,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22133169323205948,
+      "step": 9844
+    },
+    {
+      "epoch": 0.19692,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.30752741495768227,
+      "learning_rate": 0.0001,
+      "loss": 4.0269,
+      "loss/crossentropy": 1.9328197240829468,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20346572250127792,
+      "step": 9846
+    },
+    {
+      "epoch": 0.19696,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.31202367146809895,
+      "learning_rate": 0.0001,
+      "loss": 4.0299,
+      "loss/crossentropy": 2.096457004547119,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21341010928153992,
+      "step": 9848
+    },
+    {
+      "epoch": 0.197,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.3146522521972656,
+      "learning_rate": 0.0001,
+      "loss": 4.3105,
+      "loss/crossentropy": 1.9698969721794128,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1855439990758896,
+      "step": 9850
+    },
+    {
+      "epoch": 0.19704,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.31038004557291665,
+      "learning_rate": 0.0001,
+      "loss": 4.2165,
+      "loss/crossentropy": 1.987346351146698,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23657751083374023,
+      "step": 9852
+    },
+    {
+      "epoch": 0.19708,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.3109169006347656,
+      "learning_rate": 0.0001,
+      "loss": 4.0346,
+      "loss/crossentropy": 1.9535572528839111,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21131790429353714,
+      "step": 9854
+    },
+    {
+      "epoch": 0.19712,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.3090349833170573,
+      "learning_rate": 0.0001,
+      "loss": 4.5828,
+      "loss/crossentropy": 2.08541601896286,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2290833741426468,
+      "step": 9856
+    },
+    {
+      "epoch": 0.19716,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.016155751546223958,
+      "learning_rate": 0.0001,
+      "loss": 4.2774,
+      "loss/crossentropy": 2.1930192708969116,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23072391748428345,
+      "step": 9858
+    },
+    {
+      "epoch": 0.1972,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.016658274332682292,
+      "learning_rate": 0.0001,
+      "loss": 4.3923,
+      "loss/crossentropy": 1.8369358777999878,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20687467604875565,
+      "step": 9860
+    },
+    {
+      "epoch": 0.19724,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.017380523681640624,
+      "learning_rate": 0.0001,
+      "loss": 4.0287,
+      "loss/crossentropy": 1.6707186102867126,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1808091625571251,
+      "step": 9862
+    },
+    {
+      "epoch": 0.19728,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.017651112874348958,
+      "learning_rate": 0.0001,
+      "loss": 4.1184,
+      "loss/crossentropy": 1.8352991342544556,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19169726222753525,
+      "step": 9864
+    },
+    {
+      "epoch": 0.19732,
+      "grad_norm": 2.734375,
+      "grad_norm_var": 0.042909495035807294,
+      "learning_rate": 0.0001,
+      "loss": 4.2857,
+      "loss/crossentropy": 1.9427489638328552,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20076656341552734,
+      "step": 9866
+    },
+    {
+      "epoch": 0.19736,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.04237035115559896,
+      "learning_rate": 0.0001,
+      "loss": 4.3725,
+      "loss/crossentropy": 2.2190250158309937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24385111033916473,
+      "step": 9868
+    },
+    {
+      "epoch": 0.1974,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.04182510375976563,
+      "learning_rate": 0.0001,
+      "loss": 4.2192,
+      "loss/crossentropy": 2.0958147644996643,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23264098167419434,
+      "step": 9870
+    },
+    {
+      "epoch": 0.19744,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.04228897094726562,
+      "learning_rate": 0.0001,
+      "loss": 4.3571,
+      "loss/crossentropy": 2.542472720146179,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23458892852067947,
+      "step": 9872
+    },
+    {
+      "epoch": 0.19748,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.03728408813476562,
+      "learning_rate": 0.0001,
+      "loss": 4.2767,
+      "loss/crossentropy": 2.1562893390655518,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2243279591202736,
+      "step": 9874
+    },
+    {
+      "epoch": 0.19752,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.034395090738932294,
+      "learning_rate": 0.0001,
+      "loss": 4.0114,
+      "loss/crossentropy": 1.9851951599121094,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2369420975446701,
+      "step": 9876
+    },
+    {
+      "epoch": 0.19756,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.032714589436848955,
+      "learning_rate": 0.0001,
+      "loss": 4.3017,
+      "loss/crossentropy": 1.8751549124717712,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20777788758277893,
+      "step": 9878
+    },
+    {
+      "epoch": 0.1976,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0318267822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1328,
+      "loss/crossentropy": 2.010055720806122,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21204733848571777,
+      "step": 9880
+    },
+    {
+      "epoch": 0.19764,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.008942667643229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1122,
+      "loss/crossentropy": 2.04589307308197,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21330490708351135,
+      "step": 9882
+    },
+    {
+      "epoch": 0.19768,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010138956705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.1663,
+      "loss/crossentropy": 1.9441133737564087,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22589778900146484,
+      "step": 9884
+    },
+    {
+      "epoch": 0.19772,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.008296457926432292,
+      "learning_rate": 0.0001,
+      "loss": 4.4239,
+      "loss/crossentropy": 2.305395483970642,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24267761409282684,
+      "step": 9886
+    },
+    {
+      "epoch": 0.19776,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.007269032796223958,
+      "learning_rate": 0.0001,
+      "loss": 4.386,
+      "loss/crossentropy": 2.096014082431793,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21358592063188553,
+      "step": 9888
+    },
+    {
+      "epoch": 0.1978,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.008149973551432292,
+      "learning_rate": 0.0001,
+      "loss": 4.2315,
+      "loss/crossentropy": 2.1115033626556396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22994756698608398,
+      "step": 9890
+    },
+    {
+      "epoch": 0.19784,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.008654530843098958,
+      "learning_rate": 0.0001,
+      "loss": 4.1419,
+      "loss/crossentropy": 1.6122692227363586,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1897209882736206,
+      "step": 9892
+    },
+    {
+      "epoch": 0.19788,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.012933095296223959,
+      "learning_rate": 0.0001,
+      "loss": 4.1679,
+      "loss/crossentropy": 1.4790136218070984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17421242594718933,
+      "step": 9894
+    },
+    {
+      "epoch": 0.19792,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.013700358072916667,
+      "learning_rate": 0.0001,
+      "loss": 4.2921,
+      "loss/crossentropy": 1.8394885063171387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20840360969305038,
+      "step": 9896
+    },
+    {
+      "epoch": 0.19796,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0135894775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3993,
+      "loss/crossentropy": 2.152444541454315,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23834071308374405,
+      "step": 9898
+    },
+    {
+      "epoch": 0.198,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.012189737955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.5268,
+      "loss/crossentropy": 2.367082357406616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23409543931484222,
+      "step": 9900
+    },
+    {
+      "epoch": 0.19804,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.011799112955729166,
+      "learning_rate": 0.0001,
+      "loss": 4.4599,
+      "loss/crossentropy": 2.136048436164856,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21797242760658264,
+      "step": 9902
+    },
+    {
+      "epoch": 0.19808,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011945597330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.2262,
+      "loss/crossentropy": 2.035883128643036,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.215388223528862,
+      "step": 9904
+    },
+    {
+      "epoch": 0.19812,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.011970774332682291,
+      "learning_rate": 0.0001,
+      "loss": 4.2024,
+      "loss/crossentropy": 2.0339369773864746,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2085876688361168,
+      "step": 9906
+    },
+    {
+      "epoch": 0.19816,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009655507405598958,
+      "learning_rate": 0.0001,
+      "loss": 4.4441,
+      "loss/crossentropy": 1.829429566860199,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20085029304027557,
+      "step": 9908
+    },
+    {
+      "epoch": 0.1982,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.008861287434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1108,
+      "loss/crossentropy": 2.1721774339675903,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2118668630719185,
+      "step": 9910
+    },
+    {
+      "epoch": 0.19824,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.008861287434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3511,
+      "loss/crossentropy": 1.958261251449585,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22304313629865646,
+      "step": 9912
+    },
+    {
+      "epoch": 0.19828,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.009870402018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.5819,
+      "loss/crossentropy": 2.4651763439178467,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2435118407011032,
+      "step": 9914
+    },
+    {
+      "epoch": 0.19832,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.012532297770182292,
+      "learning_rate": 0.0001,
+      "loss": 4.0973,
+      "loss/crossentropy": 2.173910617828369,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21142201125621796,
+      "step": 9916
+    },
+    {
+      "epoch": 0.19836,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.013239542643229166,
+      "learning_rate": 0.0001,
+      "loss": 3.9283,
+      "loss/crossentropy": 1.9559763073921204,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19051063805818558,
+      "step": 9918
+    },
+    {
+      "epoch": 0.1984,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.015895334879557292,
+      "learning_rate": 0.0001,
+      "loss": 4.2516,
+      "loss/crossentropy": 2.238003969192505,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2338731735944748,
+      "step": 9920
+    },
+    {
+      "epoch": 0.19844,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.015233357747395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4624,
+      "loss/crossentropy": 2.269726276397705,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19430368393659592,
+      "step": 9922
+    },
+    {
+      "epoch": 0.19848,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.015697224934895834,
+      "learning_rate": 0.0001,
+      "loss": 4.436,
+      "loss/crossentropy": 2.351833701133728,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24469508230686188,
+      "step": 9924
+    },
+    {
+      "epoch": 0.19852,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.013398996988932292,
+      "learning_rate": 0.0001,
+      "loss": 4.3833,
+      "loss/crossentropy": 2.0600146055221558,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21375566720962524,
+      "step": 9926
+    },
+    {
+      "epoch": 0.19856,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.012562815348307292,
+      "learning_rate": 0.0001,
+      "loss": 4.2957,
+      "loss/crossentropy": 2.1543694734573364,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20916736871004105,
+      "step": 9928
+    },
+    {
+      "epoch": 0.1986,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010782623291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3111,
+      "loss/crossentropy": 1.8952317833900452,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2216978445649147,
+      "step": 9930
+    },
+    {
+      "epoch": 0.19864,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008695475260416667,
+      "learning_rate": 0.0001,
+      "loss": 4.0542,
+      "loss/crossentropy": 2.102243661880493,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22314947098493576,
+      "step": 9932
+    },
+    {
+      "epoch": 0.19868,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0072021484375,
+      "learning_rate": 0.0001,
+      "loss": 4.1653,
+      "loss/crossentropy": 1.9036884307861328,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19729873538017273,
+      "step": 9934
+    },
+    {
+      "epoch": 0.19872,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.006459299723307292,
+      "learning_rate": 0.0001,
+      "loss": 4.17,
+      "loss/crossentropy": 1.823796033859253,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18089265376329422,
+      "step": 9936
+    },
+    {
+      "epoch": 0.19876,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.006302642822265625,
+      "learning_rate": 0.0001,
+      "loss": 3.8742,
+      "loss/crossentropy": 2.055707633495331,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20620816200971603,
+      "step": 9938
+    },
+    {
+      "epoch": 0.1988,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0055010477701822914,
+      "learning_rate": 0.0001,
+      "loss": 4.256,
+      "loss/crossentropy": 2.0490049719810486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21709006279706955,
+      "step": 9940
+    },
+    {
+      "epoch": 0.19884,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.005891672770182292,
+      "learning_rate": 0.0001,
+      "loss": 4.2733,
+      "loss/crossentropy": 2.164198637008667,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2335330694913864,
+      "step": 9942
+    },
+    {
+      "epoch": 0.19888,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.006941731770833333,
+      "learning_rate": 0.0001,
+      "loss": 4.1463,
+      "loss/crossentropy": 1.9218478202819824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1945241093635559,
+      "step": 9944
+    },
+    {
+      "epoch": 0.19892,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.007328033447265625,
+      "learning_rate": 0.0001,
+      "loss": 3.8591,
+      "loss/crossentropy": 1.9456552267074585,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20599794387817383,
+      "step": 9946
+    },
+    {
+      "epoch": 0.19896,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.007749176025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0385,
+      "loss/crossentropy": 2.0472273230552673,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20920252054929733,
+      "step": 9948
+    },
+    {
+      "epoch": 0.199,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.007860310872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.308,
+      "loss/crossentropy": 2.2252047061920166,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2360471710562706,
+      "step": 9950
+    },
+    {
+      "epoch": 0.19904,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.005537923177083333,
+      "learning_rate": 0.0001,
+      "loss": 4.0828,
+      "loss/crossentropy": 1.646431565284729,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1692601889371872,
+      "step": 9952
+    },
+    {
+      "epoch": 0.19908,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0054443359375,
+      "learning_rate": 0.0001,
+      "loss": 3.8818,
+      "loss/crossentropy": 2.0114784836769104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20070793479681015,
+      "step": 9954
+    },
+    {
+      "epoch": 0.19912,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.00677490234375,
+      "learning_rate": 0.0001,
+      "loss": 4.324,
+      "loss/crossentropy": 2.001778781414032,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21009314060211182,
+      "step": 9956
+    },
+    {
+      "epoch": 0.19916,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.006037394205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3493,
+      "loss/crossentropy": 2.1330565214157104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22442802786827087,
+      "step": 9958
+    },
+    {
+      "epoch": 0.1992,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.005635579427083333,
+      "learning_rate": 0.0001,
+      "loss": 4.193,
+      "loss/crossentropy": 1.9146793484687805,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21145610511302948,
+      "step": 9960
+    },
+    {
+      "epoch": 0.19924,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.004705556233723958,
+      "learning_rate": 0.0001,
+      "loss": 4.3262,
+      "loss/crossentropy": 2.5224483013153076,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23354032635688782,
+      "step": 9962
+    },
+    {
+      "epoch": 0.19928,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.004552968343098958,
+      "learning_rate": 0.0001,
+      "loss": 4.3663,
+      "loss/crossentropy": 2.245160937309265,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21848157793283463,
+      "step": 9964
+    },
+    {
+      "epoch": 0.19932,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.004622141520182292,
+      "learning_rate": 0.0001,
+      "loss": 4.3506,
+      "loss/crossentropy": 2.122299015522003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21853189170360565,
+      "step": 9966
+    },
+    {
+      "epoch": 0.19936,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.005716705322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4524,
+      "loss/crossentropy": 2.31084668636322,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23691194504499435,
+      "step": 9968
+    },
+    {
+      "epoch": 0.1994,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.007834625244140626,
+      "learning_rate": 0.0001,
+      "loss": 4.274,
+      "loss/crossentropy": 2.2242307662963867,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21957046538591385,
+      "step": 9970
+    },
+    {
+      "epoch": 0.19944,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.007063547770182292,
+      "learning_rate": 0.0001,
+      "loss": 4.2264,
+      "loss/crossentropy": 1.792852759361267,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19996580481529236,
+      "step": 9972
+    },
+    {
+      "epoch": 0.19948,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0063250223795572914,
+      "learning_rate": 0.0001,
+      "loss": 4.3029,
+      "loss/crossentropy": 1.9593411087989807,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21858739852905273,
+      "step": 9974
+    },
+    {
+      "epoch": 0.19952,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.0067827860514322914,
+      "learning_rate": 0.0001,
+      "loss": 4.3067,
+      "loss/crossentropy": 2.259618401527405,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22094043344259262,
+      "step": 9976
+    },
+    {
+      "epoch": 0.19956,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0063168843587239586,
+      "learning_rate": 0.0001,
+      "loss": 4.373,
+      "loss/crossentropy": 2.31876802444458,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22025877982378006,
+      "step": 9978
+    },
+    {
+      "epoch": 0.1996,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.006109364827473958,
+      "learning_rate": 0.0001,
+      "loss": 4.15,
+      "loss/crossentropy": 1.7625555396080017,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21313250809907913,
+      "step": 9980
+    },
+    {
+      "epoch": 0.19964,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.007342274983723958,
+      "learning_rate": 0.0001,
+      "loss": 4.4628,
+      "loss/crossentropy": 2.19295072555542,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23694515973329544,
+      "step": 9982
+    },
+    {
+      "epoch": 0.19968,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.008739217122395834,
+      "learning_rate": 0.0001,
+      "loss": 4.187,
+      "loss/crossentropy": 2.4073877334594727,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22179137915372849,
+      "step": 9984
+    },
+    {
+      "epoch": 0.19972,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008125813802083333,
+      "learning_rate": 0.0001,
+      "loss": 3.9804,
+      "loss/crossentropy": 1.8942558765411377,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20588286221027374,
+      "step": 9986
+    },
+    {
+      "epoch": 0.19976,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.012188466389973958,
+      "learning_rate": 0.0001,
+      "loss": 4.109,
+      "loss/crossentropy": 2.220746397972107,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2144280970096588,
+      "step": 9988
+    },
+    {
+      "epoch": 0.1998,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.01260986328125,
+      "learning_rate": 0.0001,
+      "loss": 4.2279,
+      "loss/crossentropy": 1.9511706233024597,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1998429372906685,
+      "step": 9990
+    },
+    {
+      "epoch": 0.19984,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.012247467041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.143,
+      "loss/crossentropy": 2.249726891517639,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24011528491973877,
+      "step": 9992
+    },
+    {
+      "epoch": 0.19988,
+      "grad_norm": 2.6875,
+      "grad_norm_var": 0.03792292277018229,
+      "learning_rate": 0.0001,
+      "loss": 4.3104,
+      "loss/crossentropy": 1.957375943660736,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24464774131774902,
+      "step": 9994
+    },
+    {
+      "epoch": 0.19992,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.039033762613932294,
+      "learning_rate": 0.0001,
+      "loss": 4.5492,
+      "loss/crossentropy": 2.265984058380127,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24291887879371643,
+      "step": 9996
+    },
+    {
+      "epoch": 0.19996,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.03920873006184896,
+      "learning_rate": 0.0001,
+      "loss": 4.1432,
+      "loss/crossentropy": 1.8064668774604797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21571090072393417,
+      "step": 9998
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.03875732421875,
+      "learning_rate": 0.0001,
+      "loss": 4.2295,
+      "loss/crossentropy": 1.9072380661964417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19980185478925705,
+      "step": 10000
+    },
+    {
+      "epoch": 0.20004,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.036641438802083336,
+      "learning_rate": 0.0001,
+      "loss": 4.3919,
+      "loss/crossentropy": 2.074933707714081,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22161328792572021,
+      "step": 10002
+    },
+    {
+      "epoch": 0.20008,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.027581532796223957,
+      "learning_rate": 0.0001,
+      "loss": 4.0447,
+      "loss/crossentropy": 1.9344687461853027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1825423538684845,
+      "step": 10004
+    },
+    {
+      "epoch": 0.20012,
+      "grad_norm": 3.640625,
+      "grad_norm_var": 0.162158203125,
+      "learning_rate": 0.0001,
+      "loss": 4.1116,
+      "loss/crossentropy": 1.866003930568695,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22012518346309662,
+      "step": 10006
+    },
+    {
+      "epoch": 0.20016,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.162158203125,
+      "learning_rate": 0.0001,
+      "loss": 4.3517,
+      "loss/crossentropy": 2.145058751106262,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22737383097410202,
+      "step": 10008
+    },
+    {
+      "epoch": 0.2002,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.15038960774739582,
+      "learning_rate": 0.0001,
+      "loss": 3.9755,
+      "loss/crossentropy": 1.7828176617622375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22082456946372986,
+      "step": 10010
+    },
+    {
+      "epoch": 0.20024,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.1537994384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3143,
+      "loss/crossentropy": 2.1222537755966187,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2176828756928444,
+      "step": 10012
+    },
+    {
+      "epoch": 0.20028,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.15230712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.53,
+      "loss/crossentropy": 2.119267463684082,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24002012610435486,
+      "step": 10014
+    },
+    {
+      "epoch": 0.20032,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.15458577473958332,
+      "learning_rate": 0.0001,
+      "loss": 4.3528,
+      "loss/crossentropy": 2.198129415512085,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23767977952957153,
+      "step": 10016
+    },
+    {
+      "epoch": 0.20036,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.15608317057291668,
+      "learning_rate": 0.0001,
+      "loss": 4.0228,
+      "loss/crossentropy": 1.7466872334480286,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19916100800037384,
+      "step": 10018
+    },
+    {
+      "epoch": 0.2004,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.15812352498372395,
+      "learning_rate": 0.0001,
+      "loss": 4.1989,
+      "loss/crossentropy": 1.9834936261177063,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1891954466700554,
+      "step": 10020
+    },
+    {
+      "epoch": 0.20044,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.005840810139973959,
+      "learning_rate": 0.0001,
+      "loss": 4.0987,
+      "loss/crossentropy": 1.8997412323951721,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19624030590057373,
+      "step": 10022
+    },
+    {
+      "epoch": 0.20048,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0065915425618489586,
+      "learning_rate": 0.0001,
+      "loss": 4.2531,
+      "loss/crossentropy": 2.0051563382148743,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2069488987326622,
+      "step": 10024
+    },
+    {
+      "epoch": 0.20052,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.008107248942057292,
+      "learning_rate": 0.0001,
+      "loss": 4.4215,
+      "loss/crossentropy": 1.8662462830543518,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2053019255399704,
+      "step": 10026
+    },
+    {
+      "epoch": 0.20056,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.02585627237955729,
+      "learning_rate": 0.0001,
+      "loss": 4.1444,
+      "loss/crossentropy": 2.1043936014175415,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22167576104402542,
+      "step": 10028
+    },
+    {
+      "epoch": 0.2006,
+      "grad_norm": 5.78125,
+      "grad_norm_var": 0.859185536702474,
+      "learning_rate": 0.0001,
+      "loss": 4.3685,
+      "loss/crossentropy": 2.1248152256011963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22400003671646118,
+      "step": 10030
+    },
+    {
+      "epoch": 0.20064,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.852441151936849,
+      "learning_rate": 0.0001,
+      "loss": 4.1971,
+      "loss/crossentropy": 2.326894521713257,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24597454071044922,
+      "step": 10032
+    },
+    {
+      "epoch": 0.20068,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.8413164774576823,
+      "learning_rate": 0.0001,
+      "loss": 4.1801,
+      "loss/crossentropy": 1.8590435981750488,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20270948112010956,
+      "step": 10034
+    },
+    {
+      "epoch": 0.20072,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.82972412109375,
+      "learning_rate": 0.0001,
+      "loss": 4.5098,
+      "loss/crossentropy": 2.0383604168891907,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21504472196102142,
+      "step": 10036
+    },
+    {
+      "epoch": 0.20076,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.81461181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.599,
+      "loss/crossentropy": 2.42622447013855,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2143569439649582,
+      "step": 10038
+    },
+    {
+      "epoch": 0.2008,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.825640614827474,
+      "learning_rate": 0.0001,
+      "loss": 4.3216,
+      "loss/crossentropy": 2.0825703144073486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21215181052684784,
+      "step": 10040
+    },
+    {
+      "epoch": 0.20084,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.8326515197753906,
+      "learning_rate": 0.0001,
+      "loss": 3.9615,
+      "loss/crossentropy": 2.1682112216949463,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23538567870855331,
+      "step": 10042
+    },
+    {
+      "epoch": 0.20088,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.8466957092285157,
+      "learning_rate": 0.0001,
+      "loss": 4.2244,
+      "loss/crossentropy": 2.0161439180374146,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21078374981880188,
+      "step": 10044
+    },
+    {
+      "epoch": 0.20092,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.018143463134765624,
+      "learning_rate": 0.0001,
+      "loss": 4.5491,
+      "loss/crossentropy": 2.2262184619903564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2855340391397476,
+      "step": 10046
+    },
+    {
+      "epoch": 0.20096,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.011321767171223959,
+      "learning_rate": 0.0001,
+      "loss": 4.5078,
+      "loss/crossentropy": 2.454360246658325,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2287754938006401,
+      "step": 10048
+    },
+    {
+      "epoch": 0.201,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007940419514973958,
+      "learning_rate": 0.0001,
+      "loss": 4.1146,
+      "loss/crossentropy": 1.970844566822052,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20197579264640808,
+      "step": 10050
+    },
+    {
+      "epoch": 0.20104,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.006211090087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3573,
+      "loss/crossentropy": 2.0942559242248535,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21927295625209808,
+      "step": 10052
+    },
+    {
+      "epoch": 0.20108,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.005771636962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.036,
+      "loss/crossentropy": 1.868508517742157,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1838754191994667,
+      "step": 10054
+    },
+    {
+      "epoch": 0.20112,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0058095296223958336,
+      "learning_rate": 0.0001,
+      "loss": 4.2101,
+      "loss/crossentropy": 2.028561532497406,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21768346428871155,
+      "step": 10056
+    },
+    {
+      "epoch": 0.20116,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.006669108072916667,
+      "learning_rate": 0.0001,
+      "loss": 3.9803,
+      "loss/crossentropy": 2.3005030155181885,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23332027345895767,
+      "step": 10058
+    },
+    {
+      "epoch": 0.2012,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0065826416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.089,
+      "loss/crossentropy": 1.7793474793434143,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18733646720647812,
+      "step": 10060
+    },
+    {
+      "epoch": 0.20124,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.005125935872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0197,
+      "loss/crossentropy": 2.0612844228744507,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21674886345863342,
+      "step": 10062
+    },
+    {
+      "epoch": 0.20128,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.004500071207682292,
+      "learning_rate": 0.0001,
+      "loss": 4.237,
+      "loss/crossentropy": 2.3219568729400635,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20989079773426056,
+      "step": 10064
+    },
+    {
+      "epoch": 0.20132,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.004659016927083333,
+      "learning_rate": 0.0001,
+      "loss": 4.0959,
+      "loss/crossentropy": 2.0941001176834106,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19992788136005402,
+      "step": 10066
+    },
+    {
+      "epoch": 0.20136,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0038330078125,
+      "learning_rate": 0.0001,
+      "loss": 4.053,
+      "loss/crossentropy": 1.7448341250419617,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2108384072780609,
+      "step": 10068
+    },
+    {
+      "epoch": 0.2014,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0059397379557291664,
+      "learning_rate": 0.0001,
+      "loss": 4.263,
+      "loss/crossentropy": 2.0475903749465942,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19875742495059967,
+      "step": 10070
+    },
+    {
+      "epoch": 0.20144,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0499755859375,
+      "learning_rate": 0.0001,
+      "loss": 4.1014,
+      "loss/crossentropy": 2.0624433755874634,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20577051490545273,
+      "step": 10072
+    },
+    {
+      "epoch": 0.20148,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.04807840983072917,
+      "learning_rate": 0.0001,
+      "loss": 4.3612,
+      "loss/crossentropy": 2.194978952407837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19964048266410828,
+      "step": 10074
+    },
+    {
+      "epoch": 0.20152,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.04691162109375,
+      "learning_rate": 0.0001,
+      "loss": 4.4025,
+      "loss/crossentropy": 2.011984169483185,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22259068489074707,
+      "step": 10076
+    },
+    {
+      "epoch": 0.20156,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0458740234375,
+      "learning_rate": 0.0001,
+      "loss": 4.3539,
+      "loss/crossentropy": 2.098285675048828,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2232050970196724,
+      "step": 10078
+    },
+    {
+      "epoch": 0.2016,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.04423726399739583,
+      "learning_rate": 0.0001,
+      "loss": 4.2772,
+      "loss/crossentropy": 2.313677191734314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2224883735179901,
+      "step": 10080
+    },
+    {
+      "epoch": 0.20164,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.04454523722330729,
+      "learning_rate": 0.0001,
+      "loss": 4.1523,
+      "loss/crossentropy": 2.3712470531463623,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23282987624406815,
+      "step": 10082
+    },
+    {
+      "epoch": 0.20168,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.043342844645182295,
+      "learning_rate": 0.0001,
+      "loss": 4.3173,
+      "loss/crossentropy": 2.204525947570801,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2384341061115265,
+      "step": 10084
+    },
+    {
+      "epoch": 0.20172,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.04528401692708333,
+      "learning_rate": 0.0001,
+      "loss": 3.9497,
+      "loss/crossentropy": 1.8102558851242065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17472535371780396,
+      "step": 10086
+    },
+    {
+      "epoch": 0.20176,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.005711873372395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2119,
+      "loss/crossentropy": 2.041890263557434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21201669424772263,
+      "step": 10088
+    },
+    {
+      "epoch": 0.2018,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.006026204427083333,
+      "learning_rate": 0.0001,
+      "loss": 4.3869,
+      "loss/crossentropy": 1.9904854893684387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23453570157289505,
+      "step": 10090
+    },
+    {
+      "epoch": 0.20184,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.00716552734375,
+      "learning_rate": 0.0001,
+      "loss": 4.4324,
+      "loss/crossentropy": 2.199320912361145,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22087062150239944,
+      "step": 10092
+    },
+    {
+      "epoch": 0.20188,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0073150634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.392,
+      "loss/crossentropy": 2.321953535079956,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2463374137878418,
+      "step": 10094
+    },
+    {
+      "epoch": 0.20192,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.006076812744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3276,
+      "loss/crossentropy": 2.1109840869903564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23713566362857819,
+      "step": 10096
+    },
+    {
+      "epoch": 0.20196,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.006322987874348958,
+      "learning_rate": 0.0001,
+      "loss": 3.9339,
+      "loss/crossentropy": 1.9126858711242676,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20392102003097534,
+      "step": 10098
+    },
+    {
+      "epoch": 0.202,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.006268056233723959,
+      "learning_rate": 0.0001,
+      "loss": 4.2902,
+      "loss/crossentropy": 2.073318660259247,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20465004444122314,
+      "step": 10100
+    },
+    {
+      "epoch": 0.20204,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.009837849934895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1883,
+      "loss/crossentropy": 1.7532709836959839,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22591036558151245,
+      "step": 10102
+    },
+    {
+      "epoch": 0.20208,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.010660807291666666,
+      "learning_rate": 0.0001,
+      "loss": 4.3088,
+      "loss/crossentropy": 2.1874141693115234,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.227812297642231,
+      "step": 10104
+    },
+    {
+      "epoch": 0.20212,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.014095052083333334,
+      "learning_rate": 0.0001,
+      "loss": 4.4738,
+      "loss/crossentropy": 2.522923469543457,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24104881286621094,
+      "step": 10106
+    },
+    {
+      "epoch": 0.20216,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.013963826497395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2823,
+      "loss/crossentropy": 1.9359918236732483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19563085585832596,
+      "step": 10108
+    },
+    {
+      "epoch": 0.2022,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.015315755208333334,
+      "learning_rate": 0.0001,
+      "loss": 4.3582,
+      "loss/crossentropy": 2.390757203102112,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24748887866735458,
+      "step": 10110
+    },
+    {
+      "epoch": 0.20224,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.015013631184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.5576,
+      "loss/crossentropy": 2.419153571128845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2228071466088295,
+      "step": 10112
+    },
+    {
+      "epoch": 0.20228,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.015404256184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0235,
+      "loss/crossentropy": 1.7460771799087524,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19431117177009583,
+      "step": 10114
+    },
+    {
+      "epoch": 0.20232,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.01754150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.601,
+      "loss/crossentropy": 2.308094024658203,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24104833602905273,
+      "step": 10116
+    },
+    {
+      "epoch": 0.20236,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.013374837239583333,
+      "learning_rate": 0.0001,
+      "loss": 4.1696,
+      "loss/crossentropy": 2.0273314118385315,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2325449138879776,
+      "step": 10118
+    },
+    {
+      "epoch": 0.2024,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.013792928059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.016,
+      "loss/crossentropy": 2.1261476278305054,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21481933444738388,
+      "step": 10120
+    },
+    {
+      "epoch": 0.20244,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0107818603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.478,
+      "loss/crossentropy": 2.5006214380264282,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25569023191928864,
+      "step": 10122
+    },
+    {
+      "epoch": 0.20248,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.011775461832682292,
+      "learning_rate": 0.0001,
+      "loss": 4.3208,
+      "loss/crossentropy": 2.2033116817474365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2601129561662674,
+      "step": 10124
+    },
+    {
+      "epoch": 0.20252,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.011572011311848958,
+      "learning_rate": 0.0001,
+      "loss": 4.1098,
+      "loss/crossentropy": 2.2208765745162964,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20958464592695236,
+      "step": 10126
+    },
+    {
+      "epoch": 0.20256,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.011848958333333333,
+      "learning_rate": 0.0001,
+      "loss": 3.943,
+      "loss/crossentropy": 1.7752392888069153,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22697371244430542,
+      "step": 10128
+    },
+    {
+      "epoch": 0.2026,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.011937459309895834,
+      "learning_rate": 0.0001,
+      "loss": 4.0529,
+      "loss/crossentropy": 1.6444379687309265,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18721824884414673,
+      "step": 10130
+    },
+    {
+      "epoch": 0.20264,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.007271321614583334,
+      "learning_rate": 0.0001,
+      "loss": 4.1796,
+      "loss/crossentropy": 1.7681297659873962,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24296899884939194,
+      "step": 10132
+    },
+    {
+      "epoch": 0.20268,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007614898681640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1593,
+      "loss/crossentropy": 1.8710024952888489,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1891346201300621,
+      "step": 10134
+    },
+    {
+      "epoch": 0.20272,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.007515462239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.0822,
+      "loss/crossentropy": 2.0258530974388123,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22128118574619293,
+      "step": 10136
+    },
+    {
+      "epoch": 0.20276,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 1.7479237874348958,
+      "learning_rate": 0.0001,
+      "loss": 4.3091,
+      "loss/crossentropy": 2.3366141319274902,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2612725794315338,
+      "step": 10138
+    },
+    {
+      "epoch": 0.2028,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 1.739208730061849,
+      "learning_rate": 0.0001,
+      "loss": 4.357,
+      "loss/crossentropy": 2.152353823184967,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2239176332950592,
+      "step": 10140
+    },
+    {
+      "epoch": 0.20284,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 1.7346433003743489,
+      "learning_rate": 0.0001,
+      "loss": 4.2552,
+      "loss/crossentropy": 1.8189843893051147,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21529949456453323,
+      "step": 10142
+    },
+    {
+      "epoch": 0.20288,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 1.7219970703125,
+      "learning_rate": 0.0001,
+      "loss": 4.3179,
+      "loss/crossentropy": 2.308253049850464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21748851984739304,
+      "step": 10144
+    },
+    {
+      "epoch": 0.20292,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 1.7289377848307292,
+      "learning_rate": 0.0001,
+      "loss": 4.1762,
+      "loss/crossentropy": 1.953293800354004,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1948333978652954,
+      "step": 10146
+    },
+    {
+      "epoch": 0.20296,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 1.726512654622396,
+      "learning_rate": 0.0001,
+      "loss": 3.9573,
+      "loss/crossentropy": 2.1122325658798218,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2085581198334694,
+      "step": 10148
+    },
+    {
+      "epoch": 0.203,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 1.716387685139974,
+      "learning_rate": 0.0001,
+      "loss": 4.4209,
+      "loss/crossentropy": 2.450512409210205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23670841753482819,
+      "step": 10150
+    },
+    {
+      "epoch": 0.20304,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 1.7187327067057292,
+      "learning_rate": 0.0001,
+      "loss": 4.1131,
+      "loss/crossentropy": 2.3858957290649414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23381955921649933,
+      "step": 10152
+    },
+    {
+      "epoch": 0.20308,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0060618082682291664,
+      "learning_rate": 0.0001,
+      "loss": 4.2154,
+      "loss/crossentropy": 2.1214572191238403,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22399096935987473,
+      "step": 10154
+    },
+    {
+      "epoch": 0.20312,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0082916259765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0487,
+      "loss/crossentropy": 1.9299064874649048,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19886164367198944,
+      "step": 10156
+    },
+    {
+      "epoch": 0.20316,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.009877268473307292,
+      "learning_rate": 0.0001,
+      "loss": 4.108,
+      "loss/crossentropy": 1.9582993388175964,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1859385445713997,
+      "step": 10158
+    },
+    {
+      "epoch": 0.2032,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.008790842692057292,
+      "learning_rate": 0.0001,
+      "loss": 4.408,
+      "loss/crossentropy": 2.0555814504623413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25231435894966125,
+      "step": 10160
+    },
+    {
+      "epoch": 0.20324,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009090169270833334,
+      "learning_rate": 0.0001,
+      "loss": 4.2159,
+      "loss/crossentropy": 1.9849395155906677,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20529592037200928,
+      "step": 10162
+    },
+    {
+      "epoch": 0.20328,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0071441650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.5619,
+      "loss/crossentropy": 1.9529971480369568,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1997845619916916,
+      "step": 10164
+    },
+    {
+      "epoch": 0.20332,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.009330240885416667,
+      "learning_rate": 0.0001,
+      "loss": 4.3117,
+      "loss/crossentropy": 1.9885541200637817,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19847019761800766,
+      "step": 10166
+    },
+    {
+      "epoch": 0.20336,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.010013834635416666,
+      "learning_rate": 0.0001,
+      "loss": 4.3431,
+      "loss/crossentropy": 1.8039852380752563,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19332807511091232,
+      "step": 10168
+    },
+    {
+      "epoch": 0.2034,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.01004638671875,
+      "learning_rate": 0.0001,
+      "loss": 4.4287,
+      "loss/crossentropy": 2.1633352041244507,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22904948145151138,
+      "step": 10170
+    },
+    {
+      "epoch": 0.20344,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0070953369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1148,
+      "loss/crossentropy": 2.2646039724349976,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22390951961278915,
+      "step": 10172
+    },
+    {
+      "epoch": 0.20348,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.005741119384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4114,
+      "loss/crossentropy": 2.604608416557312,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25521157681941986,
+      "step": 10174
+    },
+    {
+      "epoch": 0.20352,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.016257476806640626,
+      "learning_rate": 0.0001,
+      "loss": 4.3319,
+      "loss/crossentropy": 2.286113977432251,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23809745162725449,
+      "step": 10176
+    },
+    {
+      "epoch": 0.20356,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.01777928670247396,
+      "learning_rate": 0.0001,
+      "loss": 3.9951,
+      "loss/crossentropy": 2.0746694207191467,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21103999018669128,
+      "step": 10178
+    },
+    {
+      "epoch": 0.2036,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.018381500244140626,
+      "learning_rate": 0.0001,
+      "loss": 3.9892,
+      "loss/crossentropy": 1.9076440930366516,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2048335075378418,
+      "step": 10180
+    },
+    {
+      "epoch": 0.20364,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0178863525390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1768,
+      "loss/crossentropy": 1.8896904587745667,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20643991231918335,
+      "step": 10182
+    },
+    {
+      "epoch": 0.20368,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.016747029622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.215,
+      "loss/crossentropy": 2.156657338142395,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22853681445121765,
+      "step": 10184
+    },
+    {
+      "epoch": 0.20372,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.017772420247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.6307,
+      "loss/crossentropy": 2.3767203092575073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2422076091170311,
+      "step": 10186
+    },
+    {
+      "epoch": 0.20376,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.01834894816080729,
+      "learning_rate": 0.0001,
+      "loss": 4.3402,
+      "loss/crossentropy": 2.5728834867477417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22912005335092545,
+      "step": 10188
+    },
+    {
+      "epoch": 0.2038,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.02005182902018229,
+      "learning_rate": 0.0001,
+      "loss": 4.11,
+      "loss/crossentropy": 2.1542125940322876,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19575025141239166,
+      "step": 10190
+    },
+    {
+      "epoch": 0.20384,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.004610188802083333,
+      "learning_rate": 0.0001,
+      "loss": 3.9114,
+      "loss/crossentropy": 1.5845852494239807,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1845107451081276,
+      "step": 10192
+    },
+    {
+      "epoch": 0.20388,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.010957590738932292,
+      "learning_rate": 0.0001,
+      "loss": 4.2193,
+      "loss/crossentropy": 2.1166247129440308,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2352868989109993,
+      "step": 10194
+    },
+    {
+      "epoch": 0.20392,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.010941314697265624,
+      "learning_rate": 0.0001,
+      "loss": 4.3374,
+      "loss/crossentropy": 2.0129401683807373,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22231722623109818,
+      "step": 10196
+    },
+    {
+      "epoch": 0.20396,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010595703125,
+      "learning_rate": 0.0001,
+      "loss": 3.9709,
+      "loss/crossentropy": 1.7132073044776917,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18660317361354828,
+      "step": 10198
+    },
+    {
+      "epoch": 0.204,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.016743977864583332,
+      "learning_rate": 0.0001,
+      "loss": 4.3959,
+      "loss/crossentropy": 2.0064845085144043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20879874378442764,
+      "step": 10200
+    },
+    {
+      "epoch": 0.20404,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.01803766886393229,
+      "learning_rate": 0.0001,
+      "loss": 4.3558,
+      "loss/crossentropy": 2.4205944538116455,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24201631546020508,
+      "step": 10202
+    },
+    {
+      "epoch": 0.20408,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.019694010416666668,
+      "learning_rate": 0.0001,
+      "loss": 3.8977,
+      "loss/crossentropy": 2.0392255187034607,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21578273177146912,
+      "step": 10204
+    },
+    {
+      "epoch": 0.20412,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.019230143229166666,
+      "learning_rate": 0.0001,
+      "loss": 4.1414,
+      "loss/crossentropy": 1.8652849197387695,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1896674558520317,
+      "step": 10206
+    },
+    {
+      "epoch": 0.20416,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.02332331339518229,
+      "learning_rate": 0.0001,
+      "loss": 4.7698,
+      "loss/crossentropy": 2.016683042049408,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23144973814487457,
+      "step": 10208
+    },
+    {
+      "epoch": 0.2042,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.020401763916015624,
+      "learning_rate": 0.0001,
+      "loss": 4.3691,
+      "loss/crossentropy": 1.9395010471343994,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19717292487621307,
+      "step": 10210
+    },
+    {
+      "epoch": 0.20424,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0210845947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0907,
+      "loss/crossentropy": 1.7626919150352478,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.200415201485157,
+      "step": 10212
+    },
+    {
+      "epoch": 0.20428,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.021955362955729165,
+      "learning_rate": 0.0001,
+      "loss": 4.3411,
+      "loss/crossentropy": 2.3014339208602905,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22324562072753906,
+      "step": 10214
+    },
+    {
+      "epoch": 0.20432,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.019636027018229165,
+      "learning_rate": 0.0001,
+      "loss": 4.1237,
+      "loss/crossentropy": 1.906779408454895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2223847657442093,
+      "step": 10216
+    },
+    {
+      "epoch": 0.20436,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.01789118448893229,
+      "learning_rate": 0.0001,
+      "loss": 4.4555,
+      "loss/crossentropy": 2.085246205329895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21039249747991562,
+      "step": 10218
+    },
+    {
+      "epoch": 0.2044,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.014806874593098958,
+      "learning_rate": 0.0001,
+      "loss": 4.4477,
+      "loss/crossentropy": 2.213107645511627,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22880114614963531,
+      "step": 10220
+    },
+    {
+      "epoch": 0.20444,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.013392893473307292,
+      "learning_rate": 0.0001,
+      "loss": 3.9015,
+      "loss/crossentropy": 1.9510034322738647,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20935232937335968,
+      "step": 10222
+    },
+    {
+      "epoch": 0.20448,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.008906809488932292,
+      "learning_rate": 0.0001,
+      "loss": 4.1237,
+      "loss/crossentropy": 1.8595823645591736,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20829308032989502,
+      "step": 10224
+    },
+    {
+      "epoch": 0.20452,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008990224202473958,
+      "learning_rate": 0.0001,
+      "loss": 4.1287,
+      "loss/crossentropy": 1.8250519037246704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18982955813407898,
+      "step": 10226
+    },
+    {
+      "epoch": 0.20456,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0074859619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3656,
+      "loss/crossentropy": 2.410372495651245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22655458748340607,
+      "step": 10228
+    },
+    {
+      "epoch": 0.2046,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.007533518473307291,
+      "learning_rate": 0.0001,
+      "loss": 4.2581,
+      "loss/crossentropy": 2.321051836013794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2280098795890808,
+      "step": 10230
+    },
+    {
+      "epoch": 0.20464,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.007380167643229167,
+      "learning_rate": 0.0001,
+      "loss": 3.9927,
+      "loss/crossentropy": 2.266388177871704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2341095432639122,
+      "step": 10232
+    },
+    {
+      "epoch": 0.20468,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.007478841145833333,
+      "learning_rate": 0.0001,
+      "loss": 4.5263,
+      "loss/crossentropy": 2.390430450439453,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23684432357549667,
+      "step": 10234
+    },
+    {
+      "epoch": 0.20472,
+      "grad_norm": 3.390625,
+      "grad_norm_var": 0.11719563802083334,
+      "learning_rate": 0.0001,
+      "loss": 4.4269,
+      "loss/crossentropy": 2.07179594039917,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21902770549058914,
+      "step": 10236
+    },
+    {
+      "epoch": 0.20476,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.12704264322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.3473,
+      "loss/crossentropy": 1.686942458152771,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20007195323705673,
+      "step": 10238
+    },
+    {
+      "epoch": 0.2048,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.26913248697916664,
+      "learning_rate": 0.0001,
+      "loss": 4.2375,
+      "loss/crossentropy": 2.0421791076660156,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2187313288450241,
+      "step": 10240
+    },
+    {
+      "epoch": 0.20484,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.26201985677083334,
+      "learning_rate": 0.0001,
+      "loss": 4.1641,
+      "loss/crossentropy": 1.9503712058067322,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19879397749900818,
+      "step": 10242
+    },
+    {
+      "epoch": 0.20488,
+      "grad_norm": 3.21875,
+      "grad_norm_var": 0.82620849609375,
+      "learning_rate": 0.0001,
+      "loss": 4.413,
+      "loss/crossentropy": 2.1453936100006104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22767861932516098,
+      "step": 10244
+    },
+    {
+      "epoch": 0.20492,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.808221181233724,
+      "learning_rate": 0.0001,
+      "loss": 4.1924,
+      "loss/crossentropy": 1.9007731080055237,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20069654285907745,
+      "step": 10246
+    },
+    {
+      "epoch": 0.20496,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.8184832255045573,
+      "learning_rate": 0.0001,
+      "loss": 4.0685,
+      "loss/crossentropy": 2.0545393228530884,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1983654722571373,
+      "step": 10248
+    },
+    {
+      "epoch": 0.205,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.804272206624349,
+      "learning_rate": 0.0001,
+      "loss": 4.3391,
+      "loss/crossentropy": 2.158636450767517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24239712953567505,
+      "step": 10250
+    },
+    {
+      "epoch": 0.20504,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.7762794494628906,
+      "learning_rate": 0.0001,
+      "loss": 4.1608,
+      "loss/crossentropy": 2.1119120121002197,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2136205956339836,
+      "step": 10252
+    },
+    {
+      "epoch": 0.20508,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.7838417053222656,
+      "learning_rate": 0.0001,
+      "loss": 4.3184,
+      "loss/crossentropy": 2.0690027475357056,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20616184175014496,
+      "step": 10254
+    },
+    {
+      "epoch": 0.20512,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.6926798502604167,
+      "learning_rate": 0.0001,
+      "loss": 4.0635,
+      "loss/crossentropy": 2.178507924079895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21580064296722412,
+      "step": 10256
+    },
+    {
+      "epoch": 0.20516,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.6889719645182292,
+      "learning_rate": 0.0001,
+      "loss": 4.1433,
+      "loss/crossentropy": 2.292190670967102,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22163032740354538,
+      "step": 10258
+    },
+    {
+      "epoch": 0.2052,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.03568115234375,
+      "learning_rate": 0.0001,
+      "loss": 4.0138,
+      "loss/crossentropy": 2.066649317741394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2081037238240242,
+      "step": 10260
+    },
+    {
+      "epoch": 0.20524,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.05111490885416667,
+      "learning_rate": 0.0001,
+      "loss": 4.4315,
+      "loss/crossentropy": 1.9017595052719116,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19284649938344955,
+      "step": 10262
+    },
+    {
+      "epoch": 0.20528,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.04326960245768229,
+      "learning_rate": 0.0001,
+      "loss": 4.0974,
+      "loss/crossentropy": 2.1215697526931763,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2244422286748886,
+      "step": 10264
+    },
+    {
+      "epoch": 0.20532,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.02539647420247396,
+      "learning_rate": 0.0001,
+      "loss": 4.3231,
+      "loss/crossentropy": 2.170191764831543,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2308938354253769,
+      "step": 10266
+    },
+    {
+      "epoch": 0.20536,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.02535985310872396,
+      "learning_rate": 0.0001,
+      "loss": 4.3472,
+      "loss/crossentropy": 2.0430655479431152,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2435612976551056,
+      "step": 10268
+    },
+    {
+      "epoch": 0.2054,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.025394439697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.536,
+      "loss/crossentropy": 2.3141634464263916,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2234780564904213,
+      "step": 10270
+    },
+    {
+      "epoch": 0.20544,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.023371378580729168,
+      "learning_rate": 0.0001,
+      "loss": 4.1944,
+      "loss/crossentropy": 2.310709834098816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2505229711532593,
+      "step": 10272
+    },
+    {
+      "epoch": 0.20548,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.02434056599934896,
+      "learning_rate": 0.0001,
+      "loss": 4.0664,
+      "loss/crossentropy": 1.9158611297607422,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2062816470861435,
+      "step": 10274
+    },
+    {
+      "epoch": 0.20552,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.02697728474934896,
+      "learning_rate": 0.0001,
+      "loss": 4.0545,
+      "loss/crossentropy": 2.0835859179496765,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21051569283008575,
+      "step": 10276
+    },
+    {
+      "epoch": 0.20556,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.006951649983723958,
+      "learning_rate": 0.0001,
+      "loss": 4.4047,
+      "loss/crossentropy": 1.9533037543296814,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20007294416427612,
+      "step": 10278
+    },
+    {
+      "epoch": 0.2056,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.010701497395833334,
+      "learning_rate": 0.0001,
+      "loss": 4.4499,
+      "loss/crossentropy": 2.3090076446533203,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2295464426279068,
+      "step": 10280
+    },
+    {
+      "epoch": 0.20564,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.014989217122395834,
+      "learning_rate": 0.0001,
+      "loss": 4.2708,
+      "loss/crossentropy": 2.2951393127441406,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22297964990139008,
+      "step": 10282
+    },
+    {
+      "epoch": 0.20568,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.014574178059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1153,
+      "loss/crossentropy": 2.311514675617218,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2287898138165474,
+      "step": 10284
+    },
+    {
+      "epoch": 0.20572,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.014989217122395834,
+      "learning_rate": 0.0001,
+      "loss": 4.2287,
+      "loss/crossentropy": 2.277890205383301,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21169421076774597,
+      "step": 10286
+    },
+    {
+      "epoch": 0.20576,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.015143839518229167,
+      "learning_rate": 0.0001,
+      "loss": 4.4448,
+      "loss/crossentropy": 2.070693612098694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21682647615671158,
+      "step": 10288
+    },
+    {
+      "epoch": 0.2058,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.014788564046223958,
+      "learning_rate": 0.0001,
+      "loss": 4.205,
+      "loss/crossentropy": 2.223360061645508,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24135399609804153,
+      "step": 10290
+    },
+    {
+      "epoch": 0.20584,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.018155670166015624,
+      "learning_rate": 0.0001,
+      "loss": 4.3304,
+      "loss/crossentropy": 2.430101752281189,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21579021215438843,
+      "step": 10292
+    },
+    {
+      "epoch": 0.20588,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.018173980712890624,
+      "learning_rate": 0.0001,
+      "loss": 4.1615,
+      "loss/crossentropy": 1.960309624671936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22619223594665527,
+      "step": 10294
+    },
+    {
+      "epoch": 0.20592,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.024217732747395835,
+      "learning_rate": 0.0001,
+      "loss": 4.8862,
+      "loss/crossentropy": 2.0035970211029053,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20148956030607224,
+      "step": 10296
+    },
+    {
+      "epoch": 0.20596,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.021968587239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.175,
+      "loss/crossentropy": 1.976987361907959,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21035870164632797,
+      "step": 10298
+    },
+    {
+      "epoch": 0.206,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.021675618489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.1596,
+      "loss/crossentropy": 2.0631470680236816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21725613623857498,
+      "step": 10300
+    },
+    {
+      "epoch": 0.20604,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.02693456013997396,
+      "learning_rate": 0.0001,
+      "loss": 4.4734,
+      "loss/crossentropy": 2.2747987508773804,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2295055389404297,
+      "step": 10302
+    },
+    {
+      "epoch": 0.20608,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.029255167643229166,
+      "learning_rate": 0.0001,
+      "loss": 4.1255,
+      "loss/crossentropy": 2.0811264514923096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21890189498662949,
+      "step": 10304
+    },
+    {
+      "epoch": 0.20612,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.028319295247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4391,
+      "loss/crossentropy": 2.2474766969680786,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2430667206645012,
+      "step": 10306
+    },
+    {
+      "epoch": 0.20616,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.024933878580729166,
+      "learning_rate": 0.0001,
+      "loss": 4.3016,
+      "loss/crossentropy": 1.899698257446289,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21101202815771103,
+      "step": 10308
+    },
+    {
+      "epoch": 0.2062,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.023851521809895835,
+      "learning_rate": 0.0001,
+      "loss": 3.9769,
+      "loss/crossentropy": 1.6432967782020569,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19004638493061066,
+      "step": 10310
+    },
+    {
+      "epoch": 0.20624,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0149658203125,
+      "learning_rate": 0.0001,
+      "loss": 4.3901,
+      "loss/crossentropy": 1.9606398940086365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24546240270137787,
+      "step": 10312
+    },
+    {
+      "epoch": 0.20628,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.015021769205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0988,
+      "loss/crossentropy": 2.140414595603943,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2428836077451706,
+      "step": 10314
+    },
+    {
+      "epoch": 0.20632,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.018393707275390626,
+      "learning_rate": 0.0001,
+      "loss": 4.3768,
+      "loss/crossentropy": 2.0886260271072388,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2201090231537819,
+      "step": 10316
+    },
+    {
+      "epoch": 0.20636,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.1039947509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.5696,
+      "loss/crossentropy": 2.3409098386764526,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24670831114053726,
+      "step": 10318
+    },
+    {
+      "epoch": 0.2064,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.10114313761393229,
+      "learning_rate": 0.0001,
+      "loss": 4.4153,
+      "loss/crossentropy": 2.233125150203705,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21548831462860107,
+      "step": 10320
+    },
+    {
+      "epoch": 0.20644,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.1031206766764323,
+      "learning_rate": 0.0001,
+      "loss": 4.0617,
+      "loss/crossentropy": 2.131038188934326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20244313776493073,
+      "step": 10322
+    },
+    {
+      "epoch": 0.20648,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.10423965454101562,
+      "learning_rate": 0.0001,
+      "loss": 4.3428,
+      "loss/crossentropy": 2.1683152318000793,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21852879226207733,
+      "step": 10324
+    },
+    {
+      "epoch": 0.20652,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.10465672810872396,
+      "learning_rate": 0.0001,
+      "loss": 4.4117,
+      "loss/crossentropy": 2.2986634969711304,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24344030022621155,
+      "step": 10326
+    },
+    {
+      "epoch": 0.20656,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.10876057942708334,
+      "learning_rate": 0.0001,
+      "loss": 4.063,
+      "loss/crossentropy": 2.2356297969818115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21090354025363922,
+      "step": 10328
+    },
+    {
+      "epoch": 0.2066,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.10851949055989583,
+      "learning_rate": 0.0001,
+      "loss": 4.4235,
+      "loss/crossentropy": 2.6431000232696533,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26503315567970276,
+      "step": 10330
+    },
+    {
+      "epoch": 0.20664,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.10668919881184896,
+      "learning_rate": 0.0001,
+      "loss": 4.2277,
+      "loss/crossentropy": 2.0261669754981995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21528497338294983,
+      "step": 10332
+    },
+    {
+      "epoch": 0.20668,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.007755279541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1092,
+      "loss/crossentropy": 1.5593605041503906,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18069667369127274,
+      "step": 10334
+    },
+    {
+      "epoch": 0.20672,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.0087646484375,
+      "learning_rate": 0.0001,
+      "loss": 4.1836,
+      "loss/crossentropy": 1.953243374824524,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19982123374938965,
+      "step": 10336
+    },
+    {
+      "epoch": 0.20676,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0088531494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.4369,
+      "loss/crossentropy": 2.067806303501129,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2209654077887535,
+      "step": 10338
+    },
+    {
+      "epoch": 0.2068,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0088043212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2458,
+      "loss/crossentropy": 1.9948397874832153,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21432363241910934,
+      "step": 10340
+    },
+    {
+      "epoch": 0.20684,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.010231272379557291,
+      "learning_rate": 0.0001,
+      "loss": 3.7643,
+      "loss/crossentropy": 1.7932568788528442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19161275029182434,
+      "step": 10342
+    },
+    {
+      "epoch": 0.20688,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.0162017822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.5325,
+      "loss/crossentropy": 2.021036922931671,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21150880306959152,
+      "step": 10344
+    },
+    {
+      "epoch": 0.20692,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.014623769124348958,
+      "learning_rate": 0.0001,
+      "loss": 3.9351,
+      "loss/crossentropy": 2.0004186630249023,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18971875309944153,
+      "step": 10346
+    },
+    {
+      "epoch": 0.20696,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.014898427327473958,
+      "learning_rate": 0.0001,
+      "loss": 4.1234,
+      "loss/crossentropy": 2.2949434518814087,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2356676608324051,
+      "step": 10348
+    },
+    {
+      "epoch": 0.207,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.013108062744140624,
+      "learning_rate": 0.0001,
+      "loss": 4.1755,
+      "loss/crossentropy": 2.161319613456726,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24161628633737564,
+      "step": 10350
+    },
+    {
+      "epoch": 0.20704,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.013285319010416666,
+      "learning_rate": 0.0001,
+      "loss": 4.1432,
+      "loss/crossentropy": 2.027602195739746,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19196761399507523,
+      "step": 10352
+    },
+    {
+      "epoch": 0.20708,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.013703409830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3658,
+      "loss/crossentropy": 2.163583278656006,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21044812351465225,
+      "step": 10354
+    },
+    {
+      "epoch": 0.20712,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.014045969645182291,
+      "learning_rate": 0.0001,
+      "loss": 4.0419,
+      "loss/crossentropy": 2.055150866508484,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21587203443050385,
+      "step": 10356
+    },
+    {
+      "epoch": 0.20716,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.0192047119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2947,
+      "loss/crossentropy": 2.368631362915039,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2406565323472023,
+      "step": 10358
+    },
+    {
+      "epoch": 0.2072,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.016486612955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.6473,
+      "loss/crossentropy": 2.5399086475372314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2554238885641098,
+      "step": 10360
+    },
+    {
+      "epoch": 0.20724,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.015366363525390624,
+      "learning_rate": 0.0001,
+      "loss": 4.2346,
+      "loss/crossentropy": 2.0829185843467712,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2278829663991928,
+      "step": 10362
+    },
+    {
+      "epoch": 0.20728,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.014371490478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2583,
+      "loss/crossentropy": 1.9829052090644836,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19677383452653885,
+      "step": 10364
+    },
+    {
+      "epoch": 0.20732,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.014385732014973958,
+      "learning_rate": 0.0001,
+      "loss": 4.3801,
+      "loss/crossentropy": 2.2335458993911743,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23407185822725296,
+      "step": 10366
+    },
+    {
+      "epoch": 0.20736,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.012149810791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2074,
+      "loss/crossentropy": 1.824280858039856,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.203867107629776,
+      "step": 10368
+    },
+    {
+      "epoch": 0.2074,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.013898722330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0666,
+      "loss/crossentropy": 2.1007773876190186,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20312541723251343,
+      "step": 10370
+    },
+    {
+      "epoch": 0.20744,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.013350168863932291,
+      "learning_rate": 0.0001,
+      "loss": 4.3692,
+      "loss/crossentropy": 1.953888475894928,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20697196573019028,
+      "step": 10372
+    },
+    {
+      "epoch": 0.20748,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.03509089152018229,
+      "learning_rate": 0.0001,
+      "loss": 4.2786,
+      "loss/crossentropy": 2.186478853225708,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24410755187273026,
+      "step": 10374
+    },
+    {
+      "epoch": 0.20752,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.03706232706705729,
+      "learning_rate": 0.0001,
+      "loss": 3.9974,
+      "loss/crossentropy": 1.9011998772621155,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1983819529414177,
+      "step": 10376
+    },
+    {
+      "epoch": 0.20756,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.03695246378580729,
+      "learning_rate": 0.0001,
+      "loss": 4.5884,
+      "loss/crossentropy": 2.599787950515747,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25155550241470337,
+      "step": 10378
+    },
+    {
+      "epoch": 0.2076,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.037021636962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2702,
+      "loss/crossentropy": 2.0120421648025513,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21281170845031738,
+      "step": 10380
+    },
+    {
+      "epoch": 0.20764,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.036710357666015624,
+      "learning_rate": 0.0001,
+      "loss": 4.4915,
+      "loss/crossentropy": 2.0685967803001404,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23026303946971893,
+      "step": 10382
+    },
+    {
+      "epoch": 0.20768,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0445068359375,
+      "learning_rate": 0.0001,
+      "loss": 4.1171,
+      "loss/crossentropy": 1.8392394185066223,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19240978360176086,
+      "step": 10384
+    },
+    {
+      "epoch": 0.20772,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.04523518880208333,
+      "learning_rate": 0.0001,
+      "loss": 4.2428,
+      "loss/crossentropy": 1.786954402923584,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20547360181808472,
+      "step": 10386
+    },
+    {
+      "epoch": 0.20776,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0452301025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2084,
+      "loss/crossentropy": 1.6960806250572205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1748766005039215,
+      "step": 10388
+    },
+    {
+      "epoch": 0.2078,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.018693033854166666,
+      "learning_rate": 0.0001,
+      "loss": 4.2491,
+      "loss/crossentropy": 2.271879196166992,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25173740088939667,
+      "step": 10390
+    },
+    {
+      "epoch": 0.20784,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.015778605143229166,
+      "learning_rate": 0.0001,
+      "loss": 4.5199,
+      "loss/crossentropy": 2.2860567569732666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23291928321123123,
+      "step": 10392
+    },
+    {
+      "epoch": 0.20788,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0148193359375,
+      "learning_rate": 0.0001,
+      "loss": 4.1871,
+      "loss/crossentropy": 1.9925037026405334,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.212161086499691,
+      "step": 10394
+    },
+    {
+      "epoch": 0.20792,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.016306304931640626,
+      "learning_rate": 0.0001,
+      "loss": 4.3728,
+      "loss/crossentropy": 2.1189831495285034,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19486143440008163,
+      "step": 10396
+    },
+    {
+      "epoch": 0.20796,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0168121337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2325,
+      "loss/crossentropy": 2.170132279396057,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23083829134702682,
+      "step": 10398
+    },
+    {
+      "epoch": 0.208,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012129465738932291,
+      "learning_rate": 0.0001,
+      "loss": 4.0773,
+      "loss/crossentropy": 1.8486035466194153,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18999971449375153,
+      "step": 10400
+    },
+    {
+      "epoch": 0.20804,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.011240386962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.138,
+      "loss/crossentropy": 1.727788269519806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19234652817249298,
+      "step": 10402
+    },
+    {
+      "epoch": 0.20808,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.01046142578125,
+      "learning_rate": 0.0001,
+      "loss": 4.4102,
+      "loss/crossentropy": 2.196265935897827,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21071433275938034,
+      "step": 10404
+    },
+    {
+      "epoch": 0.20812,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.007111612955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3378,
+      "loss/crossentropy": 1.7230273485183716,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19570616632699966,
+      "step": 10406
+    },
+    {
+      "epoch": 0.20816,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.00567626953125,
+      "learning_rate": 0.0001,
+      "loss": 3.9693,
+      "loss/crossentropy": 1.8815646767616272,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2055530995130539,
+      "step": 10408
+    },
+    {
+      "epoch": 0.2082,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.006400299072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1974,
+      "loss/crossentropy": 2.2131329774856567,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22710958123207092,
+      "step": 10410
+    },
+    {
+      "epoch": 0.20824,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.006103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1904,
+      "loss/crossentropy": 1.5199981927871704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16881641000509262,
+      "step": 10412
+    },
+    {
+      "epoch": 0.20828,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.005356597900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2456,
+      "loss/crossentropy": 2.305867075920105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2523300349712372,
+      "step": 10414
+    },
+    {
+      "epoch": 0.20832,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.004571278889973958,
+      "learning_rate": 0.0001,
+      "loss": 4.2727,
+      "loss/crossentropy": 1.989980161190033,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19124356657266617,
+      "step": 10416
+    },
+    {
+      "epoch": 0.20836,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.004255167643229167,
+      "learning_rate": 0.0001,
+      "loss": 4.4745,
+      "loss/crossentropy": 2.165328025817871,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23673278093338013,
+      "step": 10418
+    },
+    {
+      "epoch": 0.2084,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.008074696858723958,
+      "learning_rate": 0.0001,
+      "loss": 4.5429,
+      "loss/crossentropy": 2.2451056241989136,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23130114376544952,
+      "step": 10420
+    },
+    {
+      "epoch": 0.20844,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.0106689453125,
+      "learning_rate": 0.0001,
+      "loss": 3.9259,
+      "loss/crossentropy": 2.1694064140319824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21424317359924316,
+      "step": 10422
+    },
+    {
+      "epoch": 0.20848,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.012059529622395834,
+      "learning_rate": 0.0001,
+      "loss": 4.3558,
+      "loss/crossentropy": 2.140601873397827,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22452331334352493,
+      "step": 10424
+    },
+    {
+      "epoch": 0.20852,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.011476389567057292,
+      "learning_rate": 0.0001,
+      "loss": 4.3058,
+      "loss/crossentropy": 2.2076770067214966,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21608063578605652,
+      "step": 10426
+    },
+    {
+      "epoch": 0.20856,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.011579386393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.151,
+      "loss/crossentropy": 2.1738568544387817,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2288571298122406,
+      "step": 10428
+    },
+    {
+      "epoch": 0.2086,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.012165323893229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2576,
+      "loss/crossentropy": 2.2586612701416016,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22107571363449097,
+      "step": 10430
+    },
+    {
+      "epoch": 0.20864,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.014511871337890624,
+      "learning_rate": 0.0001,
+      "loss": 4.5441,
+      "loss/crossentropy": 2.336306095123291,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23462744057178497,
+      "step": 10432
+    },
+    {
+      "epoch": 0.20868,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0148590087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1625,
+      "loss/crossentropy": 2.3164994716644287,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21541497856378555,
+      "step": 10434
+    },
+    {
+      "epoch": 0.20872,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0121734619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2497,
+      "loss/crossentropy": 1.848636507987976,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20993927866220474,
+      "step": 10436
+    },
+    {
+      "epoch": 0.20876,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.010509999593098958,
+      "learning_rate": 0.0001,
+      "loss": 4.4985,
+      "loss/crossentropy": 2.234964370727539,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21675845235586166,
+      "step": 10438
+    },
+    {
+      "epoch": 0.2088,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.009159088134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2389,
+      "loss/crossentropy": 1.9301238656044006,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21708600223064423,
+      "step": 10440
+    },
+    {
+      "epoch": 0.20884,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.010721842447916666,
+      "learning_rate": 0.0001,
+      "loss": 4.1706,
+      "loss/crossentropy": 2.231620192527771,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23156532645225525,
+      "step": 10442
+    },
+    {
+      "epoch": 0.20888,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010237375895182291,
+      "learning_rate": 0.0001,
+      "loss": 4.18,
+      "loss/crossentropy": 1.8612747192382812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1914522647857666,
+      "step": 10444
+    },
+    {
+      "epoch": 0.20892,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009907786051432292,
+      "learning_rate": 0.0001,
+      "loss": 4.1713,
+      "loss/crossentropy": 2.2229456305503845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23199696838855743,
+      "step": 10446
+    },
+    {
+      "epoch": 0.20896,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.009708658854166666,
+      "learning_rate": 0.0001,
+      "loss": 4.0556,
+      "loss/crossentropy": 1.9055940508842468,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2158679962158203,
+      "step": 10448
+    },
+    {
+      "epoch": 0.209,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.011774698893229166,
+      "learning_rate": 0.0001,
+      "loss": 3.8876,
+      "loss/crossentropy": 1.5537404417991638,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17176809161901474,
+      "step": 10450
+    },
+    {
+      "epoch": 0.20904,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.0123931884765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9222,
+      "loss/crossentropy": 2.319318413734436,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22001250088214874,
+      "step": 10452
+    },
+    {
+      "epoch": 0.20908,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.010383097330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.421,
+      "loss/crossentropy": 2.2334396839141846,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.222968190908432,
+      "step": 10454
+    },
+    {
+      "epoch": 0.20912,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0110260009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.5172,
+      "loss/crossentropy": 2.5762773752212524,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26605312526226044,
+      "step": 10456
+    },
+    {
+      "epoch": 0.20916,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.010503896077473958,
+      "learning_rate": 0.0001,
+      "loss": 4.1109,
+      "loss/crossentropy": 1.7634761333465576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18641388416290283,
+      "step": 10458
+    },
+    {
+      "epoch": 0.2092,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.013224029541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.126,
+      "loss/crossentropy": 2.28191876411438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22916647791862488,
+      "step": 10460
+    },
+    {
+      "epoch": 0.20924,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.013903554280598958,
+      "learning_rate": 0.0001,
+      "loss": 4.1984,
+      "loss/crossentropy": 1.9712265729904175,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2293689101934433,
+      "step": 10462
+    },
+    {
+      "epoch": 0.20928,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012292226155598959,
+      "learning_rate": 0.0001,
+      "loss": 4.3826,
+      "loss/crossentropy": 2.425857424736023,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.255269430577755,
+      "step": 10464
+    },
+    {
+      "epoch": 0.20932,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011031087239583333,
+      "learning_rate": 0.0001,
+      "loss": 4.1919,
+      "loss/crossentropy": 2.0697131752967834,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22233501076698303,
+      "step": 10466
+    },
+    {
+      "epoch": 0.20936,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.008854166666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.4017,
+      "loss/crossentropy": 1.8367178440093994,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2063748985528946,
+      "step": 10468
+    },
+    {
+      "epoch": 0.2094,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.018192545572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.1864,
+      "loss/crossentropy": 1.977232813835144,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20568673312664032,
+      "step": 10470
+    },
+    {
+      "epoch": 0.20944,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.018195597330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.1622,
+      "loss/crossentropy": 2.1934465169906616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2175101339817047,
+      "step": 10472
+    },
+    {
+      "epoch": 0.20948,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.016047159830729168,
+      "learning_rate": 0.0001,
+      "loss": 4.3005,
+      "loss/crossentropy": 1.7997339367866516,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20166774094104767,
+      "step": 10474
+    },
+    {
+      "epoch": 0.20952,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.014671834309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1167,
+      "loss/crossentropy": 1.9334582090377808,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19190441817045212,
+      "step": 10476
+    },
+    {
+      "epoch": 0.20956,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.015148671468098958,
+      "learning_rate": 0.0001,
+      "loss": 3.9434,
+      "loss/crossentropy": 1.7263885140419006,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17968116700649261,
+      "step": 10478
+    },
+    {
+      "epoch": 0.2096,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.01661961873372396,
+      "learning_rate": 0.0001,
+      "loss": 4.1153,
+      "loss/crossentropy": 2.1710296869277954,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19717055559158325,
+      "step": 10480
+    },
+    {
+      "epoch": 0.20964,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.016947174072265626,
+      "learning_rate": 0.0001,
+      "loss": 4.2089,
+      "loss/crossentropy": 2.043896973133087,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21403591334819794,
+      "step": 10482
+    },
+    {
+      "epoch": 0.20968,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.02459691365559896,
+      "learning_rate": 0.0001,
+      "loss": 3.9237,
+      "loss/crossentropy": 1.7813313603401184,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20273292809724808,
+      "step": 10484
+    },
+    {
+      "epoch": 0.20972,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.013155110677083333,
+      "learning_rate": 0.0001,
+      "loss": 4.0985,
+      "loss/crossentropy": 1.8628470301628113,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1945454627275467,
+      "step": 10486
+    },
+    {
+      "epoch": 0.20976,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.013044230143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2279,
+      "loss/crossentropy": 2.2047882080078125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23798923939466476,
+      "step": 10488
+    },
+    {
+      "epoch": 0.2098,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.012697092692057292,
+      "learning_rate": 0.0001,
+      "loss": 4.1327,
+      "loss/crossentropy": 1.8838441967964172,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2224583625793457,
+      "step": 10490
+    },
+    {
+      "epoch": 0.20984,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.011201731363932292,
+      "learning_rate": 0.0001,
+      "loss": 4.1985,
+      "loss/crossentropy": 1.9326539039611816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2116800993680954,
+      "step": 10492
+    },
+    {
+      "epoch": 0.20988,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.017829386393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2023,
+      "loss/crossentropy": 1.7767577171325684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20496132969856262,
+      "step": 10494
+    },
+    {
+      "epoch": 0.20992,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.016996256510416665,
+      "learning_rate": 0.0001,
+      "loss": 4.1665,
+      "loss/crossentropy": 1.996176838874817,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1979241669178009,
+      "step": 10496
+    },
+    {
+      "epoch": 0.20996,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.016136678059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1467,
+      "loss/crossentropy": 2.5282262563705444,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23003733158111572,
+      "step": 10498
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.008760579427083333,
+      "learning_rate": 0.0001,
+      "loss": 4.1981,
+      "loss/crossentropy": 2.3120675086975098,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22543538361787796,
+      "step": 10500
+    },
+    {
+      "epoch": 0.21004,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.008766428629557291,
+      "learning_rate": 0.0001,
+      "loss": 4.1216,
+      "loss/crossentropy": 2.2110280990600586,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23066550493240356,
+      "step": 10502
+    },
+    {
+      "epoch": 0.21008,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.009242502848307292,
+      "learning_rate": 0.0001,
+      "loss": 4.0804,
+      "loss/crossentropy": 1.6792908906936646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20689093321561813,
+      "step": 10504
+    },
+    {
+      "epoch": 0.21012,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.011116536458333333,
+      "learning_rate": 0.0001,
+      "loss": 4.137,
+      "loss/crossentropy": 1.9797767400741577,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2185518741607666,
+      "step": 10506
+    },
+    {
+      "epoch": 0.21016,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.011533355712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0243,
+      "loss/crossentropy": 2.0001984238624573,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22058366239070892,
+      "step": 10508
+    },
+    {
+      "epoch": 0.2102,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.006414540608723958,
+      "learning_rate": 0.0001,
+      "loss": 4.3707,
+      "loss/crossentropy": 2.1622806787490845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21733585745096207,
+      "step": 10510
+    },
+    {
+      "epoch": 0.21024,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.00740966796875,
+      "learning_rate": 0.0001,
+      "loss": 4.0392,
+      "loss/crossentropy": 2.0224735736846924,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20596059411764145,
+      "step": 10512
+    },
+    {
+      "epoch": 0.21028,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.011244455973307291,
+      "learning_rate": 0.0001,
+      "loss": 4.0712,
+      "loss/crossentropy": 2.099945902824402,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19777391105890274,
+      "step": 10514
+    },
+    {
+      "epoch": 0.21032,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.011823527018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1246,
+      "loss/crossentropy": 1.9806578159332275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19959519058465958,
+      "step": 10516
+    },
+    {
+      "epoch": 0.21036,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.012239329020182292,
+      "learning_rate": 0.0001,
+      "loss": 4.1182,
+      "loss/crossentropy": 2.0056468844413757,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21810457110404968,
+      "step": 10518
+    },
+    {
+      "epoch": 0.2104,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.01785456339518229,
+      "learning_rate": 0.0001,
+      "loss": 4.1085,
+      "loss/crossentropy": 1.9525137543678284,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19195258617401123,
+      "step": 10520
+    },
+    {
+      "epoch": 0.21044,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.01580988566080729,
+      "learning_rate": 0.0001,
+      "loss": 4.4648,
+      "loss/crossentropy": 2.2391252517700195,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23115740716457367,
+      "step": 10522
+    },
+    {
+      "epoch": 0.21048,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.016727701822916666,
+      "learning_rate": 0.0001,
+      "loss": 4.6467,
+      "loss/crossentropy": 2.550819158554077,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2846238762140274,
+      "step": 10524
+    },
+    {
+      "epoch": 0.21052,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0176177978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2787,
+      "loss/crossentropy": 2.181081712245941,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23952369391918182,
+      "step": 10526
+    },
+    {
+      "epoch": 0.21056,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.01784032185872396,
+      "learning_rate": 0.0001,
+      "loss": 4.1627,
+      "loss/crossentropy": 2.3070446848869324,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21255015581846237,
+      "step": 10528
+    },
+    {
+      "epoch": 0.2106,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.014235178629557291,
+      "learning_rate": 0.0001,
+      "loss": 4.3184,
+      "loss/crossentropy": 1.9560331106185913,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19652864336967468,
+      "step": 10530
+    },
+    {
+      "epoch": 0.21064,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0127593994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.5728,
+      "loss/crossentropy": 2.2470709085464478,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25076867640018463,
+      "step": 10532
+    },
+    {
+      "epoch": 0.21068,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011767323811848958,
+      "learning_rate": 0.0001,
+      "loss": 4.2652,
+      "loss/crossentropy": 2.0689820051193237,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21815705299377441,
+      "step": 10534
+    },
+    {
+      "epoch": 0.21072,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.013499959309895834,
+      "learning_rate": 0.0001,
+      "loss": 3.9524,
+      "loss/crossentropy": 2.0633797645568848,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20865381509065628,
+      "step": 10536
+    },
+    {
+      "epoch": 0.21076,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.014029947916666667,
+      "learning_rate": 0.0001,
+      "loss": 4.2644,
+      "loss/crossentropy": 2.3332748413085938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21530094742774963,
+      "step": 10538
+    },
+    {
+      "epoch": 0.2108,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.011942545572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.1452,
+      "loss/crossentropy": 2.093048572540283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23568043112754822,
+      "step": 10540
+    },
+    {
+      "epoch": 0.21084,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.009992472330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.5464,
+      "loss/crossentropy": 2.0059397220611572,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.253838449716568,
+      "step": 10542
+    },
+    {
+      "epoch": 0.21088,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009284464518229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1417,
+      "loss/crossentropy": 2.278248429298401,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23320979624986649,
+      "step": 10544
+    },
+    {
+      "epoch": 0.21092,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.008540598551432292,
+      "learning_rate": 0.0001,
+      "loss": 4.0951,
+      "loss/crossentropy": 2.063527822494507,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21290308982133865,
+      "step": 10546
+    },
+    {
+      "epoch": 0.21096,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.007511138916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1456,
+      "loss/crossentropy": 2.092045545578003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21721098572015762,
+      "step": 10548
+    },
+    {
+      "epoch": 0.211,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.06824111938476562,
+      "learning_rate": 0.0001,
+      "loss": 4.0586,
+      "loss/crossentropy": 1.9876770973205566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21775592118501663,
+      "step": 10550
+    },
+    {
+      "epoch": 0.21104,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.06347249348958334,
+      "learning_rate": 0.0001,
+      "loss": 4.2372,
+      "loss/crossentropy": 2.292428970336914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24829304218292236,
+      "step": 10552
+    },
+    {
+      "epoch": 0.21108,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.06552632649739583,
+      "learning_rate": 0.0001,
+      "loss": 4.3412,
+      "loss/crossentropy": 2.257239043712616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2195405438542366,
+      "step": 10554
+    },
+    {
+      "epoch": 0.21112,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.06616923014322916,
+      "learning_rate": 0.0001,
+      "loss": 4.404,
+      "loss/crossentropy": 2.1424754858016968,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21044345945119858,
+      "step": 10556
+    },
+    {
+      "epoch": 0.21116,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.06642964680989584,
+      "learning_rate": 0.0001,
+      "loss": 4.2812,
+      "loss/crossentropy": 1.777747094631195,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19684316962957382,
+      "step": 10558
+    },
+    {
+      "epoch": 0.2112,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.06655171712239584,
+      "learning_rate": 0.0001,
+      "loss": 4.1239,
+      "loss/crossentropy": 2.0860772728919983,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20560266077518463,
+      "step": 10560
+    },
+    {
+      "epoch": 0.21124,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.06787007649739583,
+      "learning_rate": 0.0001,
+      "loss": 4.2409,
+      "loss/crossentropy": 1.9594369530677795,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2015877440571785,
+      "step": 10562
+    },
+    {
+      "epoch": 0.21128,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.06467692057291667,
+      "learning_rate": 0.0001,
+      "loss": 4.2539,
+      "loss/crossentropy": 2.069046676158905,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21409741044044495,
+      "step": 10564
+    },
+    {
+      "epoch": 0.21132,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.009464263916015625,
+      "learning_rate": 0.0001,
+      "loss": 3.822,
+      "loss/crossentropy": 1.7300589084625244,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18024658411741257,
+      "step": 10566
+    },
+    {
+      "epoch": 0.21136,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0077288309733072914,
+      "learning_rate": 0.0001,
+      "loss": 4.2608,
+      "loss/crossentropy": 2.048615336418152,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.228649340569973,
+      "step": 10568
+    },
+    {
+      "epoch": 0.2114,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.007085927327473958,
+      "learning_rate": 0.0001,
+      "loss": 4.4622,
+      "loss/crossentropy": 1.9678268432617188,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20117972791194916,
+      "step": 10570
+    },
+    {
+      "epoch": 0.21144,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.006534576416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3919,
+      "loss/crossentropy": 2.595113754272461,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24565115571022034,
+      "step": 10572
+    },
+    {
+      "epoch": 0.21148,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.008658599853515626,
+      "learning_rate": 0.0001,
+      "loss": 4.0873,
+      "loss/crossentropy": 2.096100628376007,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20913005620241165,
+      "step": 10574
+    },
+    {
+      "epoch": 0.21152,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.012223052978515624,
+      "learning_rate": 0.0001,
+      "loss": 4.181,
+      "loss/crossentropy": 2.0096259713172913,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21125900745391846,
+      "step": 10576
+    },
+    {
+      "epoch": 0.21156,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.011572011311848958,
+      "learning_rate": 0.0001,
+      "loss": 3.9738,
+      "loss/crossentropy": 1.9990533590316772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20894166082143784,
+      "step": 10578
+    },
+    {
+      "epoch": 0.2116,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.013606516520182292,
+      "learning_rate": 0.0001,
+      "loss": 4.0404,
+      "loss/crossentropy": 2.1385116577148438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19753456860780716,
+      "step": 10580
+    },
+    {
+      "epoch": 0.21164,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.012482706705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.5056,
+      "loss/crossentropy": 2.2081239819526672,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23155531287193298,
+      "step": 10582
+    },
+    {
+      "epoch": 0.21168,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.014339192708333334,
+      "learning_rate": 0.0001,
+      "loss": 4.161,
+      "loss/crossentropy": 1.6915860772132874,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18060292303562164,
+      "step": 10584
+    },
+    {
+      "epoch": 0.21172,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.019245402018229166,
+      "learning_rate": 0.0001,
+      "loss": 4.083,
+      "loss/crossentropy": 1.5825872421264648,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19387374818325043,
+      "step": 10586
+    },
+    {
+      "epoch": 0.21176,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0253082275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.6102,
+      "loss/crossentropy": 2.386876940727234,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23594971746206284,
+      "step": 10588
+    },
+    {
+      "epoch": 0.2118,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.02759577433268229,
+      "learning_rate": 0.0001,
+      "loss": 4.106,
+      "loss/crossentropy": 2.202653169631958,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22666631639003754,
+      "step": 10590
+    },
+    {
+      "epoch": 0.21184,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.026041412353515626,
+      "learning_rate": 0.0001,
+      "loss": 4.2285,
+      "loss/crossentropy": 2.013135075569153,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22155165672302246,
+      "step": 10592
+    },
+    {
+      "epoch": 0.21188,
+      "grad_norm": 2.25,
+      "grad_norm_var": 2.762861887613932,
+      "learning_rate": 0.0001,
+      "loss": 4.1353,
+      "loss/crossentropy": 1.6701499223709106,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21808167546987534,
+      "step": 10594
+    },
+    {
+      "epoch": 0.21192,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 2.7677996317545572,
+      "learning_rate": 0.0001,
+      "loss": 4.1438,
+      "loss/crossentropy": 2.074462592601776,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22254129499197006,
+      "step": 10596
+    },
+    {
+      "epoch": 0.21196,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 2.787275950113932,
+      "learning_rate": 0.0001,
+      "loss": 4.2329,
+      "loss/crossentropy": 2.182308316230774,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21201200038194656,
+      "step": 10598
+    },
+    {
+      "epoch": 0.212,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 2.784148915608724,
+      "learning_rate": 0.0001,
+      "loss": 4.0688,
+      "loss/crossentropy": 1.9777602553367615,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21454624831676483,
+      "step": 10600
+    },
+    {
+      "epoch": 0.21204,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 2.778930409749349,
+      "learning_rate": 0.0001,
+      "loss": 4.2707,
+      "loss/crossentropy": 2.272592306137085,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22290532290935516,
+      "step": 10602
+    },
+    {
+      "epoch": 0.21208,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 2.784010569254557,
+      "learning_rate": 0.0001,
+      "loss": 4.3406,
+      "loss/crossentropy": 2.110643744468689,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22552277147769928,
+      "step": 10604
+    },
+    {
+      "epoch": 0.21212,
+      "grad_norm": 2.125,
+      "grad_norm_var": 2.7753326416015627,
+      "learning_rate": 0.0001,
+      "loss": 4.3485,
+      "loss/crossentropy": 2.279823422431946,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22356732934713364,
+      "step": 10606
+    },
+    {
+      "epoch": 0.21216,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 2.768701171875,
+      "learning_rate": 0.0001,
+      "loss": 4.3203,
+      "loss/crossentropy": 2.0985517501831055,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22894078493118286,
+      "step": 10608
+    },
+    {
+      "epoch": 0.2122,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.006884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2295,
+      "loss/crossentropy": 2.251029133796692,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24473578482866287,
+      "step": 10610
+    },
+    {
+      "epoch": 0.21224,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.005785115559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2451,
+      "loss/crossentropy": 2.1706892251968384,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21725767105817795,
+      "step": 10612
+    },
+    {
+      "epoch": 0.21228,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.004378255208333333,
+      "learning_rate": 0.0001,
+      "loss": 4.3319,
+      "loss/crossentropy": 2.0709590315818787,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22771050035953522,
+      "step": 10614
+    },
+    {
+      "epoch": 0.21232,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.005163319905598958,
+      "learning_rate": 0.0001,
+      "loss": 4.2507,
+      "loss/crossentropy": 2.0437510013580322,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2055979147553444,
+      "step": 10616
+    },
+    {
+      "epoch": 0.21236,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.005204010009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2492,
+      "loss/crossentropy": 2.023369252681732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23950359225273132,
+      "step": 10618
+    },
+    {
+      "epoch": 0.2124,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.004325103759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3522,
+      "loss/crossentropy": 2.051850199699402,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24145089089870453,
+      "step": 10620
+    },
+    {
+      "epoch": 0.21244,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0044247945149739586,
+      "learning_rate": 0.0001,
+      "loss": 4.1743,
+      "loss/crossentropy": 1.9617546796798706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2095404863357544,
+      "step": 10622
+    },
+    {
+      "epoch": 0.21248,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.004662831624348958,
+      "learning_rate": 0.0001,
+      "loss": 4.2115,
+      "loss/crossentropy": 1.9792284965515137,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21156969666481018,
+      "step": 10624
+    },
+    {
+      "epoch": 0.21252,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.004662831624348958,
+      "learning_rate": 0.0001,
+      "loss": 4.4402,
+      "loss/crossentropy": 1.936375379562378,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21360263973474503,
+      "step": 10626
+    },
+    {
+      "epoch": 0.21256,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.0071408589680989586,
+      "learning_rate": 0.0001,
+      "loss": 4.0695,
+      "loss/crossentropy": 2.11286723613739,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20090097934007645,
+      "step": 10628
+    },
+    {
+      "epoch": 0.2126,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007120513916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2555,
+      "loss/crossentropy": 2.1020379066467285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21139420568943024,
+      "step": 10630
+    },
+    {
+      "epoch": 0.21264,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0064453125,
+      "learning_rate": 0.0001,
+      "loss": 4.2027,
+      "loss/crossentropy": 2.220509111881256,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22286564111709595,
+      "step": 10632
+    },
+    {
+      "epoch": 0.21268,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.0080078125,
+      "learning_rate": 0.0001,
+      "loss": 4.0012,
+      "loss/crossentropy": 2.058075189590454,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22242067009210587,
+      "step": 10634
+    },
+    {
+      "epoch": 0.21272,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.005269368489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.1825,
+      "loss/crossentropy": 1.9641217589378357,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20702090859413147,
+      "step": 10636
+    },
+    {
+      "epoch": 0.21276,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.005231730143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1309,
+      "loss/crossentropy": 1.9234120845794678,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20697829127311707,
+      "step": 10638
+    },
+    {
+      "epoch": 0.2128,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0067047119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2551,
+      "loss/crossentropy": 2.1050453782081604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21321691572666168,
+      "step": 10640
+    },
+    {
+      "epoch": 0.21284,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.009323883056640624,
+      "learning_rate": 0.0001,
+      "loss": 4.0852,
+      "loss/crossentropy": 2.21867573261261,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21468330919742584,
+      "step": 10642
+    },
+    {
+      "epoch": 0.21288,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.007319895426432291,
+      "learning_rate": 0.0001,
+      "loss": 4.4209,
+      "loss/crossentropy": 2.213089942932129,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21172764152288437,
+      "step": 10644
+    },
+    {
+      "epoch": 0.21292,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012556711832682291,
+      "learning_rate": 0.0001,
+      "loss": 4.3261,
+      "loss/crossentropy": 2.0370752811431885,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23427169024944305,
+      "step": 10646
+    },
+    {
+      "epoch": 0.21296,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.016035715738932293,
+      "learning_rate": 0.0001,
+      "loss": 4.2622,
+      "loss/crossentropy": 2.080373227596283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2148713618516922,
+      "step": 10648
+    },
+    {
+      "epoch": 0.213,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.014134724934895834,
+      "learning_rate": 0.0001,
+      "loss": 4.5408,
+      "loss/crossentropy": 2.3023130893707275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22768481075763702,
+      "step": 10650
+    },
+    {
+      "epoch": 0.21304,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.016185506184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.372,
+      "loss/crossentropy": 2.14642870426178,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22463608533143997,
+      "step": 10652
+    },
+    {
+      "epoch": 0.21308,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.016377766927083332,
+      "learning_rate": 0.0001,
+      "loss": 4.2164,
+      "loss/crossentropy": 2.2469639778137207,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2505262568593025,
+      "step": 10654
+    },
+    {
+      "epoch": 0.21312,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.015915679931640624,
+      "learning_rate": 0.0001,
+      "loss": 4.243,
+      "loss/crossentropy": 2.0431448221206665,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21013544499874115,
+      "step": 10656
+    },
+    {
+      "epoch": 0.21316,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.013631184895833334,
+      "learning_rate": 0.0001,
+      "loss": 3.9872,
+      "loss/crossentropy": 1.6768526434898376,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18652133643627167,
+      "step": 10658
+    },
+    {
+      "epoch": 0.2132,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.017175038655598957,
+      "learning_rate": 0.0001,
+      "loss": 3.9442,
+      "loss/crossentropy": 1.3748261332511902,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15284860879182816,
+      "step": 10660
+    },
+    {
+      "epoch": 0.21324,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.012282053629557291,
+      "learning_rate": 0.0001,
+      "loss": 4.1451,
+      "loss/crossentropy": 1.9126732349395752,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21141140908002853,
+      "step": 10662
+    },
+    {
+      "epoch": 0.21328,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.010908762613932291,
+      "learning_rate": 0.0001,
+      "loss": 4.2392,
+      "loss/crossentropy": 1.97357976436615,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21326088905334473,
+      "step": 10664
+    },
+    {
+      "epoch": 0.21332,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011201985677083333,
+      "learning_rate": 0.0001,
+      "loss": 4.2825,
+      "loss/crossentropy": 2.1782814860343933,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21183249354362488,
+      "step": 10666
+    },
+    {
+      "epoch": 0.21336,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.009098307291666666,
+      "learning_rate": 0.0001,
+      "loss": 4.1251,
+      "loss/crossentropy": 2.1700649857521057,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2310282066464424,
+      "step": 10668
+    },
+    {
+      "epoch": 0.2134,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0087554931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.479,
+      "loss/crossentropy": 2.249666213989258,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21025578677654266,
+      "step": 10670
+    },
+    {
+      "epoch": 0.21344,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.009287261962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3445,
+      "loss/crossentropy": 1.993752121925354,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20829375833272934,
+      "step": 10672
+    },
+    {
+      "epoch": 0.21348,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.009244791666666667,
+      "learning_rate": 0.0001,
+      "loss": 4.0814,
+      "loss/crossentropy": 2.0472013354301453,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21456149220466614,
+      "step": 10674
+    },
+    {
+      "epoch": 0.21352,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.007222239176432292,
+      "learning_rate": 0.0001,
+      "loss": 3.9497,
+      "loss/crossentropy": 2.0347819328308105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2133965790271759,
+      "step": 10676
+    },
+    {
+      "epoch": 0.21356,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.008168284098307292,
+      "learning_rate": 0.0001,
+      "loss": 4.1528,
+      "loss/crossentropy": 1.86410254240036,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22139018028974533,
+      "step": 10678
+    },
+    {
+      "epoch": 0.2136,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.005641428629557291,
+      "learning_rate": 0.0001,
+      "loss": 4.3134,
+      "loss/crossentropy": 2.3244482278823853,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24854417145252228,
+      "step": 10680
+    },
+    {
+      "epoch": 0.21364,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.006304677327473958,
+      "learning_rate": 0.0001,
+      "loss": 4.7182,
+      "loss/crossentropy": 2.518718123435974,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.274397537112236,
+      "step": 10682
+    },
+    {
+      "epoch": 0.21368,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.005928548177083334,
+      "learning_rate": 0.0001,
+      "loss": 4.1289,
+      "loss/crossentropy": 1.8111079931259155,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2039540931582451,
+      "step": 10684
+    },
+    {
+      "epoch": 0.21372,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.008129628499348958,
+      "learning_rate": 0.0001,
+      "loss": 4.3446,
+      "loss/crossentropy": 2.186485230922699,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22497782111167908,
+      "step": 10686
+    },
+    {
+      "epoch": 0.21376,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.009877268473307292,
+      "learning_rate": 0.0001,
+      "loss": 4.4708,
+      "loss/crossentropy": 2.1850993633270264,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22541093826293945,
+      "step": 10688
+    },
+    {
+      "epoch": 0.2138,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0090576171875,
+      "learning_rate": 0.0001,
+      "loss": 4.3991,
+      "loss/crossentropy": 1.9756001830101013,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23400776088237762,
+      "step": 10690
+    },
+    {
+      "epoch": 0.21384,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010652669270833333,
+      "learning_rate": 0.0001,
+      "loss": 4.0283,
+      "loss/crossentropy": 1.7454423904418945,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18263398110866547,
+      "step": 10692
+    },
+    {
+      "epoch": 0.21388,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010066731770833334,
+      "learning_rate": 0.0001,
+      "loss": 3.9364,
+      "loss/crossentropy": 1.5824024081230164,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18881092965602875,
+      "step": 10694
+    },
+    {
+      "epoch": 0.21392,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.009383138020833333,
+      "learning_rate": 0.0001,
+      "loss": 4.3587,
+      "loss/crossentropy": 2.1171644926071167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2173115462064743,
+      "step": 10696
+    },
+    {
+      "epoch": 0.21396,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.008698527018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2102,
+      "loss/crossentropy": 1.9327389001846313,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22025491297245026,
+      "step": 10698
+    },
+    {
+      "epoch": 0.214,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.009952799479166666,
+      "learning_rate": 0.0001,
+      "loss": 3.9622,
+      "loss/crossentropy": 1.9806901216506958,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2089216560125351,
+      "step": 10700
+    },
+    {
+      "epoch": 0.21404,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.007503000895182291,
+      "learning_rate": 0.0001,
+      "loss": 4.2503,
+      "loss/crossentropy": 2.216805338859558,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21268506348133087,
+      "step": 10702
+    },
+    {
+      "epoch": 0.21408,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0055084228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5121,
+      "loss/crossentropy": 2.3998383283615112,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26092807948589325,
+      "step": 10704
+    },
+    {
+      "epoch": 0.21412,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.007869466145833334,
+      "learning_rate": 0.0001,
+      "loss": 4.4273,
+      "loss/crossentropy": 2.0581844449043274,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2316882163286209,
+      "step": 10706
+    },
+    {
+      "epoch": 0.21416,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.006258138020833333,
+      "learning_rate": 0.0001,
+      "loss": 4.2684,
+      "loss/crossentropy": 2.3091371059417725,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23538943380117416,
+      "step": 10708
+    },
+    {
+      "epoch": 0.2142,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0067291259765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9567,
+      "loss/crossentropy": 1.7134324312210083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1941806674003601,
+      "step": 10710
+    },
+    {
+      "epoch": 0.21424,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0060699462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3165,
+      "loss/crossentropy": 2.2040648460388184,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20669078081846237,
+      "step": 10712
+    },
+    {
+      "epoch": 0.21428,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.020612589518229165,
+      "learning_rate": 0.0001,
+      "loss": 4.3207,
+      "loss/crossentropy": 2.3040376901626587,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20685256272554398,
+      "step": 10714
+    },
+    {
+      "epoch": 0.21432,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.020318349202473957,
+      "learning_rate": 0.0001,
+      "loss": 4.2674,
+      "loss/crossentropy": 2.189309239387512,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21499691903591156,
+      "step": 10716
+    },
+    {
+      "epoch": 0.21436,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.020216623942057293,
+      "learning_rate": 0.0001,
+      "loss": 4.4144,
+      "loss/crossentropy": 2.1955957412719727,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20932748913764954,
+      "step": 10718
+    },
+    {
+      "epoch": 0.2144,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.020271809895833333,
+      "learning_rate": 0.0001,
+      "loss": 4.2135,
+      "loss/crossentropy": 1.7800896763801575,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1939619928598404,
+      "step": 10720
+    },
+    {
+      "epoch": 0.21444,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.020335896809895834,
+      "learning_rate": 0.0001,
+      "loss": 4.2757,
+      "loss/crossentropy": 1.8974847197532654,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19434937089681625,
+      "step": 10722
+    },
+    {
+      "epoch": 0.21448,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.021455891927083335,
+      "learning_rate": 0.0001,
+      "loss": 4.2688,
+      "loss/crossentropy": 2.1631242632865906,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21765826642513275,
+      "step": 10724
+    },
+    {
+      "epoch": 0.21452,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0203521728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.6421,
+      "loss/crossentropy": 2.241260290145874,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2334459200501442,
+      "step": 10726
+    },
+    {
+      "epoch": 0.21456,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0226470947265625,
+      "learning_rate": 0.0001,
+      "loss": 3.9478,
+      "loss/crossentropy": 1.9063156247138977,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22118454426527023,
+      "step": 10728
+    },
+    {
+      "epoch": 0.2146,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011310831705729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0446,
+      "loss/crossentropy": 1.8109349012374878,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18581955134868622,
+      "step": 10730
+    },
+    {
+      "epoch": 0.21464,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.009749348958333333,
+      "learning_rate": 0.0001,
+      "loss": 4.3262,
+      "loss/crossentropy": 2.1113163232803345,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21339301764965057,
+      "step": 10732
+    },
+    {
+      "epoch": 0.21468,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0108551025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.5308,
+      "loss/crossentropy": 2.1731717586517334,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22267260402441025,
+      "step": 10734
+    },
+    {
+      "epoch": 0.21472,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010087076822916667,
+      "learning_rate": 0.0001,
+      "loss": 4.2363,
+      "loss/crossentropy": 2.322808027267456,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21375543624162674,
+      "step": 10736
+    },
+    {
+      "epoch": 0.21476,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.010358683268229167,
+      "learning_rate": 0.0001,
+      "loss": 4.109,
+      "loss/crossentropy": 1.8200489282608032,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19796901941299438,
+      "step": 10738
+    },
+    {
+      "epoch": 0.2148,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009455362955729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0918,
+      "loss/crossentropy": 2.353461265563965,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22755976021289825,
+      "step": 10740
+    },
+    {
+      "epoch": 0.21484,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0071451822916666664,
+      "learning_rate": 0.0001,
+      "loss": 4.2925,
+      "loss/crossentropy": 2.3200663328170776,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23450587689876556,
+      "step": 10742
+    },
+    {
+      "epoch": 0.21488,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.006403605143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.187,
+      "loss/crossentropy": 1.7293912768363953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17938270419836044,
+      "step": 10744
+    },
+    {
+      "epoch": 0.21492,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.008656565348307292,
+      "learning_rate": 0.0001,
+      "loss": 4.3213,
+      "loss/crossentropy": 1.9759944081306458,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20631389319896698,
+      "step": 10746
+    },
+    {
+      "epoch": 0.21496,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.006870269775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3658,
+      "loss/crossentropy": 2.2232764959335327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21919700503349304,
+      "step": 10748
+    },
+    {
+      "epoch": 0.215,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.004898834228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2466,
+      "loss/crossentropy": 2.0827722549438477,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20569747686386108,
+      "step": 10750
+    },
+    {
+      "epoch": 0.21504,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.005509440104166667,
+      "learning_rate": 0.0001,
+      "loss": 4.2373,
+      "loss/crossentropy": 2.0712032318115234,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20504066348075867,
+      "step": 10752
+    },
+    {
+      "epoch": 0.21508,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.007963053385416667,
+      "learning_rate": 0.0001,
+      "loss": 4.3466,
+      "loss/crossentropy": 2.2717082500457764,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22957370430231094,
+      "step": 10754
+    },
+    {
+      "epoch": 0.21512,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.0133056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2821,
+      "loss/crossentropy": 2.263810157775879,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2291572391986847,
+      "step": 10756
+    },
+    {
+      "epoch": 0.21516,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.013724517822265626,
+      "learning_rate": 0.0001,
+      "loss": 4.2157,
+      "loss/crossentropy": 2.103231191635132,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21839886158704758,
+      "step": 10758
+    },
+    {
+      "epoch": 0.2152,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.013962554931640624,
+      "learning_rate": 0.0001,
+      "loss": 4.4772,
+      "loss/crossentropy": 2.1272148489952087,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2280830293893814,
+      "step": 10760
+    },
+    {
+      "epoch": 0.21524,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.012962849934895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1192,
+      "loss/crossentropy": 2.133938789367676,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2162996605038643,
+      "step": 10762
+    },
+    {
+      "epoch": 0.21528,
+      "grad_norm": 2.828125,
+      "grad_norm_var": 0.04528401692708333,
+      "learning_rate": 0.0001,
+      "loss": 4.5351,
+      "loss/crossentropy": 2.305683732032776,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23797836154699326,
+      "step": 10764
+    },
+    {
+      "epoch": 0.21532,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.04370829264322917,
+      "learning_rate": 0.0001,
+      "loss": 3.9886,
+      "loss/crossentropy": 1.7248413562774658,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17226764559745789,
+      "step": 10766
+    },
+    {
+      "epoch": 0.21536,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.041715240478515624,
+      "learning_rate": 0.0001,
+      "loss": 4.6422,
+      "loss/crossentropy": 2.275644540786743,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2433001920580864,
+      "step": 10768
+    },
+    {
+      "epoch": 0.2154,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.043338775634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2836,
+      "loss/crossentropy": 1.908457100391388,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19758973270654678,
+      "step": 10770
+    },
+    {
+      "epoch": 0.21544,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.07423909505208333,
+      "learning_rate": 0.0001,
+      "loss": 4.2317,
+      "loss/crossentropy": 1.9401238560676575,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22177566587924957,
+      "step": 10772
+    },
+    {
+      "epoch": 0.21548,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.07476170857747395,
+      "learning_rate": 0.0001,
+      "loss": 4.295,
+      "loss/crossentropy": 2.247257351875305,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21560464799404144,
+      "step": 10774
+    },
+    {
+      "epoch": 0.21552,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.07644424438476563,
+      "learning_rate": 0.0001,
+      "loss": 4.1482,
+      "loss/crossentropy": 1.9191248416900635,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.214834064245224,
+      "step": 10776
+    },
+    {
+      "epoch": 0.21556,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0785888671875,
+      "learning_rate": 0.0001,
+      "loss": 4.0644,
+      "loss/crossentropy": 1.8963102102279663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20552606880664825,
+      "step": 10778
+    },
+    {
+      "epoch": 0.2156,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.047078450520833336,
+      "learning_rate": 0.0001,
+      "loss": 4.265,
+      "loss/crossentropy": 2.0517951250076294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1963544860482216,
+      "step": 10780
+    },
+    {
+      "epoch": 0.21564,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.046727498372395836,
+      "learning_rate": 0.0001,
+      "loss": 4.3571,
+      "loss/crossentropy": 2.1484888792037964,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23963741213083267,
+      "step": 10782
+    },
+    {
+      "epoch": 0.21568,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.06716206868489584,
+      "learning_rate": 0.0001,
+      "loss": 4.6827,
+      "loss/crossentropy": 2.012593388557434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2115195021033287,
+      "step": 10784
+    },
+    {
+      "epoch": 0.21572,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.06651102701822917,
+      "learning_rate": 0.0001,
+      "loss": 4.018,
+      "loss/crossentropy": 2.139856696128845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23609659075737,
+      "step": 10786
+    },
+    {
+      "epoch": 0.21576,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.03216120402018229,
+      "learning_rate": 0.0001,
+      "loss": 4.5007,
+      "loss/crossentropy": 2.0139951705932617,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22799161076545715,
+      "step": 10788
+    },
+    {
+      "epoch": 0.2158,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.035676829020182294,
+      "learning_rate": 0.0001,
+      "loss": 4.227,
+      "loss/crossentropy": 1.8325074315071106,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18156912177801132,
+      "step": 10790
+    },
+    {
+      "epoch": 0.21584,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.04155654907226562,
+      "learning_rate": 0.0001,
+      "loss": 3.8042,
+      "loss/crossentropy": 1.7531892657279968,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18498936295509338,
+      "step": 10792
+    },
+    {
+      "epoch": 0.21588,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.03681233723958333,
+      "learning_rate": 0.0001,
+      "loss": 4.3955,
+      "loss/crossentropy": 2.09742671251297,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21543975174427032,
+      "step": 10794
+    },
+    {
+      "epoch": 0.21592,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.03532613118489583,
+      "learning_rate": 0.0001,
+      "loss": 4.4752,
+      "loss/crossentropy": 2.4303117990493774,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24365779757499695,
+      "step": 10796
+    },
+    {
+      "epoch": 0.21596,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0371002197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1973,
+      "loss/crossentropy": 2.007950007915497,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20470578223466873,
+      "step": 10798
+    },
+    {
+      "epoch": 0.216,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.026949055989583335,
+      "learning_rate": 0.0001,
+      "loss": 3.9975,
+      "loss/crossentropy": 1.895507276058197,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2005787119269371,
+      "step": 10800
+    },
+    {
+      "epoch": 0.21604,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.026691691080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2543,
+      "loss/crossentropy": 2.174374043941498,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2282470539212227,
+      "step": 10802
+    },
+    {
+      "epoch": 0.21608,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.02789484659830729,
+      "learning_rate": 0.0001,
+      "loss": 4.0642,
+      "loss/crossentropy": 1.8878389596939087,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19730794429779053,
+      "step": 10804
+    },
+    {
+      "epoch": 0.21612,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.023884073893229166,
+      "learning_rate": 0.0001,
+      "loss": 4.1733,
+      "loss/crossentropy": 1.9874022006988525,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21521702408790588,
+      "step": 10806
+    },
+    {
+      "epoch": 0.21616,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.01962458292643229,
+      "learning_rate": 0.0001,
+      "loss": 4.0783,
+      "loss/crossentropy": 2.0173474550247192,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21552105993032455,
+      "step": 10808
+    },
+    {
+      "epoch": 0.2162,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.018344879150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0583,
+      "loss/crossentropy": 2.175139367580414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20279064774513245,
+      "step": 10810
+    },
+    {
+      "epoch": 0.21624,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.022849273681640626,
+      "learning_rate": 0.0001,
+      "loss": 4.5125,
+      "loss/crossentropy": 2.6407864093780518,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2939087897539139,
+      "step": 10812
+    },
+    {
+      "epoch": 0.21628,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.024930826822916665,
+      "learning_rate": 0.0001,
+      "loss": 4.0869,
+      "loss/crossentropy": 1.947974681854248,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2102196291089058,
+      "step": 10814
+    },
+    {
+      "epoch": 0.21632,
+      "grad_norm": 9.875,
+      "grad_norm_var": 3.7874745686848956,
+      "learning_rate": 0.0001,
+      "loss": 4.3404,
+      "loss/crossentropy": 1.9040276408195496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20579400658607483,
+      "step": 10816
+    },
+    {
+      "epoch": 0.21636,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 3.7853190104166665,
+      "learning_rate": 0.0001,
+      "loss": 3.7119,
+      "loss/crossentropy": 1.7941421270370483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19792088121175766,
+      "step": 10818
+    },
+    {
+      "epoch": 0.2164,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 3.7839088439941406,
+      "learning_rate": 0.0001,
+      "loss": 4.2573,
+      "loss/crossentropy": 2.2164441347122192,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2903618812561035,
+      "step": 10820
+    },
+    {
+      "epoch": 0.21644,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 3.77445068359375,
+      "learning_rate": 0.0001,
+      "loss": 4.2537,
+      "loss/crossentropy": 1.9888933897018433,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20800678431987762,
+      "step": 10822
+    },
+    {
+      "epoch": 0.21648,
+      "grad_norm": 1.875,
+      "grad_norm_var": 3.79569091796875,
+      "learning_rate": 0.0001,
+      "loss": 3.9218,
+      "loss/crossentropy": 1.9026559591293335,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18179579824209213,
+      "step": 10824
+    },
+    {
+      "epoch": 0.21652,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 3.7974202473958334,
+      "learning_rate": 0.0001,
+      "loss": 4.1267,
+      "loss/crossentropy": 1.9020920991897583,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20755057036876678,
+      "step": 10826
+    },
+    {
+      "epoch": 0.21656,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 3.8177286783854165,
+      "learning_rate": 0.0001,
+      "loss": 4.4689,
+      "loss/crossentropy": 2.3464537858963013,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23209182918071747,
+      "step": 10828
+    },
+    {
+      "epoch": 0.2166,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 3.7860877990722654,
+      "learning_rate": 0.0001,
+      "loss": 4.4486,
+      "loss/crossentropy": 2.305312991142273,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2548810988664627,
+      "step": 10830
+    },
+    {
+      "epoch": 0.21664,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.026775868733723958,
+      "learning_rate": 0.0001,
+      "loss": 3.9818,
+      "loss/crossentropy": 1.4548576474189758,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1718049794435501,
+      "step": 10832
+    },
+    {
+      "epoch": 0.21668,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.02504247029622396,
+      "learning_rate": 0.0001,
+      "loss": 4.4909,
+      "loss/crossentropy": 2.2954181432724,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22977370768785477,
+      "step": 10834
+    },
+    {
+      "epoch": 0.21672,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.022930653889973958,
+      "learning_rate": 0.0001,
+      "loss": 4.0433,
+      "loss/crossentropy": 2.1974023580551147,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21474600583314896,
+      "step": 10836
+    },
+    {
+      "epoch": 0.21676,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.02304865519205729,
+      "learning_rate": 0.0001,
+      "loss": 4.5905,
+      "loss/crossentropy": 2.1211158633232117,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19471213221549988,
+      "step": 10838
+    },
+    {
+      "epoch": 0.2168,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0204010009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0868,
+      "loss/crossentropy": 1.9942908883094788,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1899253949522972,
+      "step": 10840
+    },
+    {
+      "epoch": 0.21684,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.023656209309895832,
+      "learning_rate": 0.0001,
+      "loss": 4.0787,
+      "loss/crossentropy": 1.9848283529281616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1777423620223999,
+      "step": 10842
+    },
+    {
+      "epoch": 0.21688,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.02545140584309896,
+      "learning_rate": 0.0001,
+      "loss": 4.1365,
+      "loss/crossentropy": 1.84994775056839,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19193057715892792,
+      "step": 10844
+    },
+    {
+      "epoch": 0.21692,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.02163670857747396,
+      "learning_rate": 0.0001,
+      "loss": 4.2869,
+      "loss/crossentropy": 1.8300130367279053,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19747256487607956,
+      "step": 10846
+    },
+    {
+      "epoch": 0.21696,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.010676829020182292,
+      "learning_rate": 0.0001,
+      "loss": 4.1624,
+      "loss/crossentropy": 1.9316250681877136,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22893256694078445,
+      "step": 10848
+    },
+    {
+      "epoch": 0.217,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009085845947265626,
+      "learning_rate": 0.0001,
+      "loss": 4.0294,
+      "loss/crossentropy": 1.9862067103385925,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22470611333847046,
+      "step": 10850
+    },
+    {
+      "epoch": 0.21704,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.009372711181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3979,
+      "loss/crossentropy": 2.196234107017517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22592387348413467,
+      "step": 10852
+    },
+    {
+      "epoch": 0.21708,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.2918291727701823,
+      "learning_rate": 0.0001,
+      "loss": 4.5516,
+      "loss/crossentropy": 2.3097496032714844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29694322496652603,
+      "step": 10854
+    },
+    {
+      "epoch": 0.21712,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.2865191141764323,
+      "learning_rate": 0.0001,
+      "loss": 3.8687,
+      "loss/crossentropy": 1.883777916431427,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21657781302928925,
+      "step": 10856
+    },
+    {
+      "epoch": 0.21716,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.2916338602701823,
+      "learning_rate": 0.0001,
+      "loss": 4.1915,
+      "loss/crossentropy": 2.0483964681625366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22432015091180801,
+      "step": 10858
+    },
+    {
+      "epoch": 0.2172,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.2870839436848958,
+      "learning_rate": 0.0001,
+      "loss": 4.1503,
+      "loss/crossentropy": 2.2700769901275635,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21760191768407822,
+      "step": 10860
+    },
+    {
+      "epoch": 0.21724,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.2878214518229167,
+      "learning_rate": 0.0001,
+      "loss": 4.574,
+      "loss/crossentropy": 2.3758704662323,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21792180836200714,
+      "step": 10862
+    },
+    {
+      "epoch": 0.21728,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.2934641520182292,
+      "learning_rate": 0.0001,
+      "loss": 4.2895,
+      "loss/crossentropy": 2.1972473859786987,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22460343688726425,
+      "step": 10864
+    },
+    {
+      "epoch": 0.21732,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.29321187337239585,
+      "learning_rate": 0.0001,
+      "loss": 4.3693,
+      "loss/crossentropy": 1.9811919331550598,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23008134216070175,
+      "step": 10866
+    },
+    {
+      "epoch": 0.21736,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.2992327372233073,
+      "learning_rate": 0.0001,
+      "loss": 4.2237,
+      "loss/crossentropy": 2.032214403152466,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1958027482032776,
+      "step": 10868
+    },
+    {
+      "epoch": 0.2174,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.035162099202473956,
+      "learning_rate": 0.0001,
+      "loss": 4.0956,
+      "loss/crossentropy": 2.08488667011261,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21948565542697906,
+      "step": 10870
+    },
+    {
+      "epoch": 0.21744,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.03401260375976563,
+      "learning_rate": 0.0001,
+      "loss": 4.2923,
+      "loss/crossentropy": 2.304922103881836,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1977907121181488,
+      "step": 10872
+    },
+    {
+      "epoch": 0.21748,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.005582427978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1762,
+      "loss/crossentropy": 1.991280436515808,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2196018323302269,
+      "step": 10874
+    },
+    {
+      "epoch": 0.21752,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.004965972900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.4116,
+      "loss/crossentropy": 1.955183207988739,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21059879660606384,
+      "step": 10876
+    },
+    {
+      "epoch": 0.21756,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0045562744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2105,
+      "loss/crossentropy": 2.1443604230880737,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23160522431135178,
+      "step": 10878
+    },
+    {
+      "epoch": 0.2176,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.004889933268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2374,
+      "loss/crossentropy": 2.0859211683273315,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2296902909874916,
+      "step": 10880
+    },
+    {
+      "epoch": 0.21764,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0059234619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3303,
+      "loss/crossentropy": 2.220987915992737,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22863281518220901,
+      "step": 10882
+    },
+    {
+      "epoch": 0.21768,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.005625152587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2195,
+      "loss/crossentropy": 2.1172574758529663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.197882778942585,
+      "step": 10884
+    },
+    {
+      "epoch": 0.21772,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.02047704060872396,
+      "learning_rate": 0.0001,
+      "loss": 4.3907,
+      "loss/crossentropy": 1.7861940264701843,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1984492540359497,
+      "step": 10886
+    },
+    {
+      "epoch": 0.21776,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0218414306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0445,
+      "loss/crossentropy": 2.1119033098220825,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20427606999874115,
+      "step": 10888
+    },
+    {
+      "epoch": 0.2178,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.02235692342122396,
+      "learning_rate": 0.0001,
+      "loss": 3.923,
+      "loss/crossentropy": 1.7849717140197754,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20817570388317108,
+      "step": 10890
+    },
+    {
+      "epoch": 0.21784,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.025131988525390624,
+      "learning_rate": 0.0001,
+      "loss": 4.2015,
+      "loss/crossentropy": 2.0543535351753235,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20579595863819122,
+      "step": 10892
+    },
+    {
+      "epoch": 0.21788,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.02643000284830729,
+      "learning_rate": 0.0001,
+      "loss": 4.1527,
+      "loss/crossentropy": 2.085465431213379,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23541826009750366,
+      "step": 10894
+    },
+    {
+      "epoch": 0.21792,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.03343073527018229,
+      "learning_rate": 0.0001,
+      "loss": 4.3288,
+      "loss/crossentropy": 1.9521069526672363,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21404746174812317,
+      "step": 10896
+    },
+    {
+      "epoch": 0.21796,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.03172378540039063,
+      "learning_rate": 0.0001,
+      "loss": 4.039,
+      "loss/crossentropy": 1.8710272908210754,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20500759035348892,
+      "step": 10898
+    },
+    {
+      "epoch": 0.218,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.032083892822265626,
+      "learning_rate": 0.0001,
+      "loss": 4.0695,
+      "loss/crossentropy": 2.275243639945984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24073244631290436,
+      "step": 10900
+    },
+    {
+      "epoch": 0.21804,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0181060791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3508,
+      "loss/crossentropy": 2.2093106508255005,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23254899680614471,
+      "step": 10902
+    },
+    {
+      "epoch": 0.21808,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.01702855428059896,
+      "learning_rate": 0.0001,
+      "loss": 4.2805,
+      "loss/crossentropy": 2.0315810441970825,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20614907145500183,
+      "step": 10904
+    },
+    {
+      "epoch": 0.21812,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.016478474934895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3461,
+      "loss/crossentropy": 2.0397735834121704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2093086987733841,
+      "step": 10906
+    },
+    {
+      "epoch": 0.21816,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.013508097330729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0286,
+      "loss/crossentropy": 1.9616519808769226,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19433944672346115,
+      "step": 10908
+    },
+    {
+      "epoch": 0.2182,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.01749445597330729,
+      "learning_rate": 0.0001,
+      "loss": 4.2157,
+      "loss/crossentropy": 2.1907248497009277,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23218485713005066,
+      "step": 10910
+    },
+    {
+      "epoch": 0.21824,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009981282552083333,
+      "learning_rate": 0.0001,
+      "loss": 4.163,
+      "loss/crossentropy": 2.2785152196884155,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21416624635457993,
+      "step": 10912
+    },
+    {
+      "epoch": 0.21828,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.00955810546875,
+      "learning_rate": 0.0001,
+      "loss": 4.2528,
+      "loss/crossentropy": 2.151498794555664,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23632780462503433,
+      "step": 10914
+    },
+    {
+      "epoch": 0.21832,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008829752604166666,
+      "learning_rate": 0.0001,
+      "loss": 4.3937,
+      "loss/crossentropy": 2.816411852836609,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2243700549006462,
+      "step": 10916
+    },
+    {
+      "epoch": 0.21836,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.014662424723307291,
+      "learning_rate": 0.0001,
+      "loss": 4.4091,
+      "loss/crossentropy": 2.2203346490859985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22302204370498657,
+      "step": 10918
+    },
+    {
+      "epoch": 0.2184,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.014662424723307291,
+      "learning_rate": 0.0001,
+      "loss": 4.3534,
+      "loss/crossentropy": 2.0416316390037537,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20798437297344208,
+      "step": 10920
+    },
+    {
+      "epoch": 0.21844,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.014426422119140626,
+      "learning_rate": 0.0001,
+      "loss": 4.1014,
+      "loss/crossentropy": 2.2251007556915283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2149331197142601,
+      "step": 10922
+    },
+    {
+      "epoch": 0.21848,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.014631907145182291,
+      "learning_rate": 0.0001,
+      "loss": 4.2346,
+      "loss/crossentropy": 1.976640522480011,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21704353392124176,
+      "step": 10924
+    },
+    {
+      "epoch": 0.21852,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.011006418863932292,
+      "learning_rate": 0.0001,
+      "loss": 4.3284,
+      "loss/crossentropy": 2.041864037513733,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19621634483337402,
+      "step": 10926
+    },
+    {
+      "epoch": 0.21856,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.009696451822916667,
+      "learning_rate": 0.0001,
+      "loss": 4.241,
+      "loss/crossentropy": 1.7188060879707336,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1992729976773262,
+      "step": 10928
+    },
+    {
+      "epoch": 0.2186,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.010512034098307291,
+      "learning_rate": 0.0001,
+      "loss": 4.0333,
+      "loss/crossentropy": 1.664458692073822,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20099520683288574,
+      "step": 10930
+    },
+    {
+      "epoch": 0.21864,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.011161041259765626,
+      "learning_rate": 0.0001,
+      "loss": 4.1952,
+      "loss/crossentropy": 2.199326276779175,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23000852018594742,
+      "step": 10932
+    },
+    {
+      "epoch": 0.21868,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.005694325764973958,
+      "learning_rate": 0.0001,
+      "loss": 4.1399,
+      "loss/crossentropy": 1.980049967765808,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2211422473192215,
+      "step": 10934
+    },
+    {
+      "epoch": 0.21872,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.005411529541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0876,
+      "loss/crossentropy": 1.7884072661399841,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.203270323574543,
+      "step": 10936
+    },
+    {
+      "epoch": 0.21876,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.005527496337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1923,
+      "loss/crossentropy": 2.1758522987365723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21960537880659103,
+      "step": 10938
+    },
+    {
+      "epoch": 0.2188,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.004416656494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1708,
+      "loss/crossentropy": 2.255508303642273,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2171120047569275,
+      "step": 10940
+    },
+    {
+      "epoch": 0.21884,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.004308827718098958,
+      "learning_rate": 0.0001,
+      "loss": 4.1294,
+      "loss/crossentropy": 1.9758012890815735,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19945884495973587,
+      "step": 10942
+    },
+    {
+      "epoch": 0.21888,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0055582682291666664,
+      "learning_rate": 0.0001,
+      "loss": 4.2122,
+      "loss/crossentropy": 2.2980172634124756,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20864219218492508,
+      "step": 10944
+    },
+    {
+      "epoch": 0.21892,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.006666819254557292,
+      "learning_rate": 0.0001,
+      "loss": 3.919,
+      "loss/crossentropy": 1.5307916402816772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1799665167927742,
+      "step": 10946
+    },
+    {
+      "epoch": 0.21896,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.006648508707682291,
+      "learning_rate": 0.0001,
+      "loss": 4.3796,
+      "loss/crossentropy": 2.3332602977752686,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22964681684970856,
+      "step": 10948
+    },
+    {
+      "epoch": 0.219,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.007380930582682291,
+      "learning_rate": 0.0001,
+      "loss": 4.1207,
+      "loss/crossentropy": 2.2509007453918457,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22141733020544052,
+      "step": 10950
+    },
+    {
+      "epoch": 0.21904,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.008949534098307291,
+      "learning_rate": 0.0001,
+      "loss": 4.4879,
+      "loss/crossentropy": 2.3677018880844116,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25676336884498596,
+      "step": 10952
+    },
+    {
+      "epoch": 0.21908,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.009934234619140624,
+      "learning_rate": 0.0001,
+      "loss": 4.4283,
+      "loss/crossentropy": 2.2807843685150146,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21486903727054596,
+      "step": 10954
+    },
+    {
+      "epoch": 0.21912,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009232330322265624,
+      "learning_rate": 0.0001,
+      "loss": 4.3126,
+      "loss/crossentropy": 2.3883347511291504,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21362561732530594,
+      "step": 10956
+    },
+    {
+      "epoch": 0.21916,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009405263264973958,
+      "learning_rate": 0.0001,
+      "loss": 4.0692,
+      "loss/crossentropy": 1.8598107695579529,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19806977361440659,
+      "step": 10958
+    },
+    {
+      "epoch": 0.2192,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.007738240559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.3259,
+      "loss/crossentropy": 2.1962249875068665,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20640095323324203,
+      "step": 10960
+    },
+    {
+      "epoch": 0.21924,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.007958984375,
+      "learning_rate": 0.0001,
+      "loss": 4.1774,
+      "loss/crossentropy": 2.061887502670288,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22463876008987427,
+      "step": 10962
+    },
+    {
+      "epoch": 0.21928,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010237375895182291,
+      "learning_rate": 0.0001,
+      "loss": 3.9749,
+      "loss/crossentropy": 1.9112628102302551,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19327937066555023,
+      "step": 10964
+    },
+    {
+      "epoch": 0.21932,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.010640207926432292,
+      "learning_rate": 0.0001,
+      "loss": 4.3078,
+      "loss/crossentropy": 2.1655001640319824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22729264944791794,
+      "step": 10966
+    },
+    {
+      "epoch": 0.21936,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.010400136311848959,
+      "learning_rate": 0.0001,
+      "loss": 4.1641,
+      "loss/crossentropy": 2.199634552001953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23565081506967545,
+      "step": 10968
+    },
+    {
+      "epoch": 0.2194,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.010155995686848959,
+      "learning_rate": 0.0001,
+      "loss": 4.4298,
+      "loss/crossentropy": 2.4142041206359863,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23020224273204803,
+      "step": 10970
+    },
+    {
+      "epoch": 0.21944,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011193593343098959,
+      "learning_rate": 0.0001,
+      "loss": 4.3856,
+      "loss/crossentropy": 2.2480571269989014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.224863201379776,
+      "step": 10972
+    },
+    {
+      "epoch": 0.21948,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.012238566080729167,
+      "learning_rate": 0.0001,
+      "loss": 4.099,
+      "loss/crossentropy": 2.1786953806877136,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21182173490524292,
+      "step": 10974
+    },
+    {
+      "epoch": 0.21952,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.013337961832682292,
+      "learning_rate": 0.0001,
+      "loss": 4.2826,
+      "loss/crossentropy": 2.3767744302749634,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22259585559368134,
+      "step": 10976
+    },
+    {
+      "epoch": 0.21956,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.010794830322265626,
+      "learning_rate": 0.0001,
+      "loss": 4.405,
+      "loss/crossentropy": 1.9838140606880188,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21830464899539948,
+      "step": 10978
+    },
+    {
+      "epoch": 0.2196,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0092193603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3811,
+      "loss/crossentropy": 2.0167009234428406,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21239794790744781,
+      "step": 10980
+    },
+    {
+      "epoch": 0.21964,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.019269816080729165,
+      "learning_rate": 0.0001,
+      "loss": 4.3117,
+      "loss/crossentropy": 2.001616358757019,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21801268309354782,
+      "step": 10982
+    },
+    {
+      "epoch": 0.21968,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.019432576497395833,
+      "learning_rate": 0.0001,
+      "loss": 4.6266,
+      "loss/crossentropy": 2.259532332420349,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20131191611289978,
+      "step": 10984
+    },
+    {
+      "epoch": 0.21972,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0215728759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1159,
+      "loss/crossentropy": 1.9223415851593018,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21203956007957458,
+      "step": 10986
+    },
+    {
+      "epoch": 0.21976,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.020295206705729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0177,
+      "loss/crossentropy": 2.3779542446136475,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23436614871025085,
+      "step": 10988
+    },
+    {
+      "epoch": 0.2198,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.02195002237955729,
+      "learning_rate": 0.0001,
+      "loss": 4.0412,
+      "loss/crossentropy": 1.8552000522613525,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20212603360414505,
+      "step": 10990
+    },
+    {
+      "epoch": 0.21984,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0206451416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2264,
+      "loss/crossentropy": 1.79928320646286,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19676420837640762,
+      "step": 10992
+    },
+    {
+      "epoch": 0.21988,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.018619791666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.2426,
+      "loss/crossentropy": 1.886910319328308,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21270643919706345,
+      "step": 10994
+    },
+    {
+      "epoch": 0.21992,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.019066365559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0775,
+      "loss/crossentropy": 1.5682110786437988,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19159042835235596,
+      "step": 10996
+    },
+    {
+      "epoch": 0.21996,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.008104451497395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2933,
+      "loss/crossentropy": 2.3335143327713013,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23225059360265732,
+      "step": 10998
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.007222493489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.3146,
+      "loss/crossentropy": 2.2234359979629517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21465980261564255,
+      "step": 11000
+    },
+    {
+      "epoch": 0.22004,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.00601806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.6023,
+      "loss/crossentropy": 2.3676271438598633,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2674448639154434,
+      "step": 11002
+    },
+    {
+      "epoch": 0.22008,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.006180826822916667,
+      "learning_rate": 0.0001,
+      "loss": 4.1064,
+      "loss/crossentropy": 1.9216612577438354,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2036839798092842,
+      "step": 11004
+    },
+    {
+      "epoch": 0.22012,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.004369099934895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0427,
+      "loss/crossentropy": 1.6732578873634338,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18962469696998596,
+      "step": 11006
+    },
+    {
+      "epoch": 0.22016,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.009761555989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.2704,
+      "loss/crossentropy": 2.0608668327331543,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23521222919225693,
+      "step": 11008
+    },
+    {
+      "epoch": 0.2202,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010872395833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.1808,
+      "loss/crossentropy": 1.8038227558135986,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1811496466398239,
+      "step": 11010
+    },
+    {
+      "epoch": 0.22024,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.010789998372395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3407,
+      "loss/crossentropy": 1.8687627911567688,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20196181535720825,
+      "step": 11012
+    },
+    {
+      "epoch": 0.22028,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.01138916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0608,
+      "loss/crossentropy": 1.8011687397956848,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18384280055761337,
+      "step": 11014
+    },
+    {
+      "epoch": 0.22032,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.012727610270182292,
+      "learning_rate": 0.0001,
+      "loss": 4.0229,
+      "loss/crossentropy": 1.8108493089675903,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19042345136404037,
+      "step": 11016
+    },
+    {
+      "epoch": 0.22036,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.012611643473307291,
+      "learning_rate": 0.0001,
+      "loss": 4.2755,
+      "loss/crossentropy": 2.4056872129440308,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22949732840061188,
+      "step": 11018
+    },
+    {
+      "epoch": 0.2204,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.011451975504557291,
+      "learning_rate": 0.0001,
+      "loss": 4.2768,
+      "loss/crossentropy": 2.0417627692222595,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21793337166309357,
+      "step": 11020
+    },
+    {
+      "epoch": 0.22044,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.011549631754557291,
+      "learning_rate": 0.0001,
+      "loss": 4.2767,
+      "loss/crossentropy": 1.8312503099441528,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19038879871368408,
+      "step": 11022
+    },
+    {
+      "epoch": 0.22048,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.01941095987955729,
+      "learning_rate": 0.0001,
+      "loss": 4.1846,
+      "loss/crossentropy": 2.0615866780281067,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.224861241877079,
+      "step": 11024
+    },
+    {
+      "epoch": 0.22052,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.02702611287434896,
+      "learning_rate": 0.0001,
+      "loss": 4.0216,
+      "loss/crossentropy": 1.8578996062278748,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20426958799362183,
+      "step": 11026
+    },
+    {
+      "epoch": 0.22056,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.028364817301432293,
+      "learning_rate": 0.0001,
+      "loss": 3.9303,
+      "loss/crossentropy": 2.01333224773407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2140725627541542,
+      "step": 11028
+    },
+    {
+      "epoch": 0.2206,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.0296142578125,
+      "learning_rate": 0.0001,
+      "loss": 3.8142,
+      "loss/crossentropy": 1.9607431292533875,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19605513662099838,
+      "step": 11030
+    },
+    {
+      "epoch": 0.22064,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.026712799072265626,
+      "learning_rate": 0.0001,
+      "loss": 4.1336,
+      "loss/crossentropy": 1.8091335892677307,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18755877763032913,
+      "step": 11032
+    },
+    {
+      "epoch": 0.22068,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0282958984375,
+      "learning_rate": 0.0001,
+      "loss": 4.0206,
+      "loss/crossentropy": 1.5547168254852295,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18891388177871704,
+      "step": 11034
+    },
+    {
+      "epoch": 0.22072,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.029605865478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1658,
+      "loss/crossentropy": 2.0881760120391846,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21140296012163162,
+      "step": 11036
+    },
+    {
+      "epoch": 0.22076,
+      "grad_norm": 2.953125,
+      "grad_norm_var": 0.07485936482747396,
+      "learning_rate": 0.0001,
+      "loss": 4.0865,
+      "loss/crossentropy": 2.0874351263046265,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2202945575118065,
+      "step": 11038
+    },
+    {
+      "epoch": 0.2208,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.06611302693684896,
+      "learning_rate": 0.0001,
+      "loss": 4.3998,
+      "loss/crossentropy": 2.2349069118499756,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2247784063220024,
+      "step": 11040
+    },
+    {
+      "epoch": 0.22084,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.060373687744140626,
+      "learning_rate": 0.0001,
+      "loss": 4.4679,
+      "loss/crossentropy": 2.0404593348503113,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23377884924411774,
+      "step": 11042
+    },
+    {
+      "epoch": 0.22088,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.05997314453125,
+      "learning_rate": 0.0001,
+      "loss": 4.115,
+      "loss/crossentropy": 2.418761968612671,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2288268655538559,
+      "step": 11044
+    },
+    {
+      "epoch": 0.22092,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.05608495076497396,
+      "learning_rate": 0.0001,
+      "loss": 4.514,
+      "loss/crossentropy": 2.3263691663742065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23381420969963074,
+      "step": 11046
+    },
+    {
+      "epoch": 0.22096,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.058166249593098955,
+      "learning_rate": 0.0001,
+      "loss": 4.3659,
+      "loss/crossentropy": 2.0020187497138977,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1993579939007759,
+      "step": 11048
+    },
+    {
+      "epoch": 0.221,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.05726318359375,
+      "learning_rate": 0.0001,
+      "loss": 4.1869,
+      "loss/crossentropy": 2.3417880535125732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21232012659311295,
+      "step": 11050
+    },
+    {
+      "epoch": 0.22104,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.06346817016601562,
+      "learning_rate": 0.0001,
+      "loss": 4.9882,
+      "loss/crossentropy": 2.2952964305877686,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22096505016088486,
+      "step": 11052
+    },
+    {
+      "epoch": 0.22108,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.028562164306640624,
+      "learning_rate": 0.0001,
+      "loss": 4.2431,
+      "loss/crossentropy": 1.7963152527809143,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18996400386095047,
+      "step": 11054
+    },
+    {
+      "epoch": 0.22112,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.028433990478515626,
+      "learning_rate": 0.0001,
+      "loss": 4.4276,
+      "loss/crossentropy": 2.1328593492507935,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21772069483995438,
+      "step": 11056
+    },
+    {
+      "epoch": 0.22116,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.0318267822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.317,
+      "loss/crossentropy": 1.8048993349075317,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17359213531017303,
+      "step": 11058
+    },
+    {
+      "epoch": 0.2212,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.0309234619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0173,
+      "loss/crossentropy": 1.9539333581924438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20828361809253693,
+      "step": 11060
+    },
+    {
+      "epoch": 0.22124,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.032291412353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0288,
+      "loss/crossentropy": 1.8231340050697327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20730753242969513,
+      "step": 11062
+    },
+    {
+      "epoch": 0.22128,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.033878326416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2756,
+      "loss/crossentropy": 1.9315263032913208,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20077265799045563,
+      "step": 11064
+    },
+    {
+      "epoch": 0.22132,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.034795888264973956,
+      "learning_rate": 0.0001,
+      "loss": 4.2137,
+      "loss/crossentropy": 1.9278368949890137,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2203744500875473,
+      "step": 11066
+    },
+    {
+      "epoch": 0.22136,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.016812896728515624,
+      "learning_rate": 0.0001,
+      "loss": 4.38,
+      "loss/crossentropy": 1.9314138889312744,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2024018093943596,
+      "step": 11068
+    },
+    {
+      "epoch": 0.2214,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.019618479410807292,
+      "learning_rate": 0.0001,
+      "loss": 3.9142,
+      "loss/crossentropy": 1.822297751903534,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20530518889427185,
+      "step": 11070
+    },
+    {
+      "epoch": 0.22144,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.02112401326497396,
+      "learning_rate": 0.0001,
+      "loss": 4.1412,
+      "loss/crossentropy": 2.0021358132362366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1963522955775261,
+      "step": 11072
+    },
+    {
+      "epoch": 0.22148,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.018293253580729165,
+      "learning_rate": 0.0001,
+      "loss": 4.5595,
+      "loss/crossentropy": 2.3780601024627686,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24999547004699707,
+      "step": 11074
+    },
+    {
+      "epoch": 0.22152,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.01727472941080729,
+      "learning_rate": 0.0001,
+      "loss": 4.2572,
+      "loss/crossentropy": 1.8128371238708496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1940363049507141,
+      "step": 11076
+    },
+    {
+      "epoch": 0.22156,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0147857666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1388,
+      "loss/crossentropy": 1.836561381816864,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19289480894804,
+      "step": 11078
+    },
+    {
+      "epoch": 0.2216,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0125640869140625,
+      "learning_rate": 0.0001,
+      "loss": 3.9891,
+      "loss/crossentropy": 2.203549385070801,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23240803182125092,
+      "step": 11080
+    },
+    {
+      "epoch": 0.22164,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.0147857666015625,
+      "learning_rate": 0.0001,
+      "loss": 3.9827,
+      "loss/crossentropy": 2.126620829105377,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21564993262290955,
+      "step": 11082
+    },
+    {
+      "epoch": 0.22168,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.015672810872395835,
+      "learning_rate": 0.0001,
+      "loss": 4.6468,
+      "loss/crossentropy": 2.5914783477783203,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23293393850326538,
+      "step": 11084
+    },
+    {
+      "epoch": 0.22172,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.016752115885416665,
+      "learning_rate": 0.0001,
+      "loss": 4.0881,
+      "loss/crossentropy": 2.097515106201172,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21570277214050293,
+      "step": 11086
+    },
+    {
+      "epoch": 0.22176,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.01565526326497396,
+      "learning_rate": 0.0001,
+      "loss": 4.2469,
+      "loss/crossentropy": 1.9408356547355652,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22177009284496307,
+      "step": 11088
+    },
+    {
+      "epoch": 0.2218,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.014713287353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.378,
+      "loss/crossentropy": 1.9851223826408386,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20538930594921112,
+      "step": 11090
+    },
+    {
+      "epoch": 0.22184,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.014694976806640624,
+      "learning_rate": 0.0001,
+      "loss": 4.2424,
+      "loss/crossentropy": 2.261754631996155,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21294979751110077,
+      "step": 11092
+    },
+    {
+      "epoch": 0.22188,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.014924875895182292,
+      "learning_rate": 0.0001,
+      "loss": 4.1847,
+      "loss/crossentropy": 1.978752851486206,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21184594929218292,
+      "step": 11094
+    },
+    {
+      "epoch": 0.22192,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.014918772379557292,
+      "learning_rate": 0.0001,
+      "loss": 4.515,
+      "loss/crossentropy": 2.1992926597595215,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2428576499223709,
+      "step": 11096
+    },
+    {
+      "epoch": 0.22196,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010131581624348959,
+      "learning_rate": 0.0001,
+      "loss": 4.1475,
+      "loss/crossentropy": 1.9407767057418823,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25351718813180923,
+      "step": 11098
+    },
+    {
+      "epoch": 0.222,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.009993235270182291,
+      "learning_rate": 0.0001,
+      "loss": 4.2485,
+      "loss/crossentropy": 2.2973347902297974,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2211253046989441,
+      "step": 11100
+    },
+    {
+      "epoch": 0.22204,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.007380930582682291,
+      "learning_rate": 0.0001,
+      "loss": 4.466,
+      "loss/crossentropy": 2.323284387588501,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23907601833343506,
+      "step": 11102
+    },
+    {
+      "epoch": 0.22208,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.00631103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3745,
+      "loss/crossentropy": 1.9557610750198364,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20206698775291443,
+      "step": 11104
+    },
+    {
+      "epoch": 0.22212,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.006083170572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.1432,
+      "loss/crossentropy": 2.130094528198242,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22658853232860565,
+      "step": 11106
+    },
+    {
+      "epoch": 0.22216,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.006078084309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0207,
+      "loss/crossentropy": 1.6461073160171509,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.183306485414505,
+      "step": 11108
+    },
+    {
+      "epoch": 0.2222,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.008310699462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2901,
+      "loss/crossentropy": 1.9869291186332703,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20724371075630188,
+      "step": 11110
+    },
+    {
+      "epoch": 0.22224,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.007696278889973958,
+      "learning_rate": 0.0001,
+      "loss": 4.2361,
+      "loss/crossentropy": 2.314267873764038,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22680803388357162,
+      "step": 11112
+    },
+    {
+      "epoch": 0.22228,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0104888916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.37,
+      "loss/crossentropy": 2.061935067176819,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1931627094745636,
+      "step": 11114
+    },
+    {
+      "epoch": 0.22232,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009601847330729166,
+      "learning_rate": 0.0001,
+      "loss": 4.1174,
+      "loss/crossentropy": 2.157355546951294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2328970953822136,
+      "step": 11116
+    },
+    {
+      "epoch": 0.22236,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.012214914957682291,
+      "learning_rate": 0.0001,
+      "loss": 4.2092,
+      "loss/crossentropy": 1.8968737125396729,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19509851932525635,
+      "step": 11118
+    },
+    {
+      "epoch": 0.2224,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.013516998291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3433,
+      "loss/crossentropy": 2.1147825717926025,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22639968246221542,
+      "step": 11120
+    },
+    {
+      "epoch": 0.22244,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.01519775390625,
+      "learning_rate": 0.0001,
+      "loss": 3.9818,
+      "loss/crossentropy": 1.9084222316741943,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16847409307956696,
+      "step": 11122
+    },
+    {
+      "epoch": 0.22248,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.015028635660807291,
+      "learning_rate": 0.0001,
+      "loss": 4.0188,
+      "loss/crossentropy": 1.8558747172355652,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2013881430029869,
+      "step": 11124
+    },
+    {
+      "epoch": 0.22252,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.015755208333333333,
+      "learning_rate": 0.0001,
+      "loss": 4.3236,
+      "loss/crossentropy": 2.2021052837371826,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2129388153553009,
+      "step": 11126
+    },
+    {
+      "epoch": 0.22256,
+      "grad_norm": 1.8125,
+      "grad_norm_var": 0.019212849934895835,
+      "learning_rate": 0.0001,
+      "loss": 4.2892,
+      "loss/crossentropy": 2.267111897468567,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22087457031011581,
+      "step": 11128
+    },
+    {
+      "epoch": 0.2226,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.02021052042643229,
+      "learning_rate": 0.0001,
+      "loss": 4.3781,
+      "loss/crossentropy": 1.860244870185852,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21286892145872116,
+      "step": 11130
+    },
+    {
+      "epoch": 0.22264,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.021142578125,
+      "learning_rate": 0.0001,
+      "loss": 3.7832,
+      "loss/crossentropy": 1.8312670588493347,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20548796653747559,
+      "step": 11132
+    },
+    {
+      "epoch": 0.22268,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.019087473551432293,
+      "learning_rate": 0.0001,
+      "loss": 4.1916,
+      "loss/crossentropy": 2.2309051752090454,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2244933694601059,
+      "step": 11134
+    },
+    {
+      "epoch": 0.22272,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.01761449178059896,
+      "learning_rate": 0.0001,
+      "loss": 4.1125,
+      "loss/crossentropy": 2.1003236770629883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22222542017698288,
+      "step": 11136
+    },
+    {
+      "epoch": 0.22276,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.015965779622395832,
+      "learning_rate": 0.0001,
+      "loss": 3.9391,
+      "loss/crossentropy": 1.95048588514328,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20614230632781982,
+      "step": 11138
+    },
+    {
+      "epoch": 0.2228,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.016364542643229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1608,
+      "loss/crossentropy": 2.2856240272521973,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23748096823692322,
+      "step": 11140
+    },
+    {
+      "epoch": 0.22284,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.012455240885416666,
+      "learning_rate": 0.0001,
+      "loss": 4.2187,
+      "loss/crossentropy": 2.3873090744018555,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23743586987257004,
+      "step": 11142
+    },
+    {
+      "epoch": 0.22288,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.008234659830729166,
+      "learning_rate": 0.0001,
+      "loss": 4.1283,
+      "loss/crossentropy": 2.248707115650177,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22035827487707138,
+      "step": 11144
+    },
+    {
+      "epoch": 0.22292,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0067860921223958336,
+      "learning_rate": 0.0001,
+      "loss": 4.3561,
+      "loss/crossentropy": 1.9792630672454834,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2102261707186699,
+      "step": 11146
+    },
+    {
+      "epoch": 0.22296,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.006156158447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0479,
+      "loss/crossentropy": 2.284587323665619,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20941253006458282,
+      "step": 11148
+    },
+    {
+      "epoch": 0.223,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.006361643473307292,
+      "learning_rate": 0.0001,
+      "loss": 4.1792,
+      "loss/crossentropy": 1.9800288677215576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20009202510118484,
+      "step": 11150
+    },
+    {
+      "epoch": 0.22304,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0064165751139322914,
+      "learning_rate": 0.0001,
+      "loss": 4.1859,
+      "loss/crossentropy": 1.925826370716095,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18616148829460144,
+      "step": 11152
+    },
+    {
+      "epoch": 0.22308,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0053708394368489586,
+      "learning_rate": 0.0001,
+      "loss": 3.9281,
+      "loss/crossentropy": 2.158196806907654,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21486472338438034,
+      "step": 11154
+    },
+    {
+      "epoch": 0.22312,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.004393513997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3342,
+      "loss/crossentropy": 2.0873841047286987,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22101643681526184,
+      "step": 11156
+    },
+    {
+      "epoch": 0.22316,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0044748942057291664,
+      "learning_rate": 0.0001,
+      "loss": 4.3772,
+      "loss/crossentropy": 1.9485042691230774,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19757962226867676,
+      "step": 11158
+    },
+    {
+      "epoch": 0.2232,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.004279581705729166,
+      "learning_rate": 0.0001,
+      "loss": 4.331,
+      "loss/crossentropy": 2.055552661418915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19271192699670792,
+      "step": 11160
+    },
+    {
+      "epoch": 0.22324,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0033599853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0958,
+      "loss/crossentropy": 1.9115247130393982,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1922249048948288,
+      "step": 11162
+    },
+    {
+      "epoch": 0.22328,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.0043609619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2582,
+      "loss/crossentropy": 2.1461042761802673,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2185581848025322,
+      "step": 11164
+    },
+    {
+      "epoch": 0.22332,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.004808553059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1212,
+      "loss/crossentropy": 2.1625128984451294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20505116879940033,
+      "step": 11166
+    },
+    {
+      "epoch": 0.22336,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.003692372639973958,
+      "learning_rate": 0.0001,
+      "loss": 4.0317,
+      "loss/crossentropy": 1.9978403449058533,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2015106976032257,
+      "step": 11168
+    },
+    {
+      "epoch": 0.2234,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.003794097900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0108,
+      "loss/crossentropy": 1.9462909698486328,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20814144611358643,
+      "step": 11170
+    },
+    {
+      "epoch": 0.22344,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.004423014322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.2147,
+      "loss/crossentropy": 2.262490153312683,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21692577749490738,
+      "step": 11172
+    },
+    {
+      "epoch": 0.22348,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0052734375,
+      "learning_rate": 0.0001,
+      "loss": 4.4079,
+      "loss/crossentropy": 2.092001974582672,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22943469882011414,
+      "step": 11174
+    },
+    {
+      "epoch": 0.22352,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0052734375,
+      "learning_rate": 0.0001,
+      "loss": 4.273,
+      "loss/crossentropy": 2.302329421043396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21188674122095108,
+      "step": 11176
+    },
+    {
+      "epoch": 0.22356,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.008701324462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0762,
+      "loss/crossentropy": 1.7181463837623596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19460663199424744,
+      "step": 11178
+    },
+    {
+      "epoch": 0.2236,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.007895660400390626,
+      "learning_rate": 0.0001,
+      "loss": 4.0061,
+      "loss/crossentropy": 1.8402328491210938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1928708627820015,
+      "step": 11180
+    },
+    {
+      "epoch": 0.22364,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.008103179931640624,
+      "learning_rate": 0.0001,
+      "loss": 4.2968,
+      "loss/crossentropy": 1.9490735530853271,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2492925077676773,
+      "step": 11182
+    },
+    {
+      "epoch": 0.22368,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.00867919921875,
+      "learning_rate": 0.0001,
+      "loss": 4.113,
+      "loss/crossentropy": 2.367197036743164,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25882700830698013,
+      "step": 11184
+    },
+    {
+      "epoch": 0.22372,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0090972900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0782,
+      "loss/crossentropy": 2.1192296743392944,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2208312824368477,
+      "step": 11186
+    },
+    {
+      "epoch": 0.22376,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.010772450764973959,
+      "learning_rate": 0.0001,
+      "loss": 4.7239,
+      "loss/crossentropy": 2.2150460481643677,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21440500020980835,
+      "step": 11188
+    },
+    {
+      "epoch": 0.2238,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0121490478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4245,
+      "loss/crossentropy": 2.0837016105651855,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2777148336172104,
+      "step": 11190
+    },
+    {
+      "epoch": 0.22384,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.014240519205729166,
+      "learning_rate": 0.0001,
+      "loss": 3.918,
+      "loss/crossentropy": 1.7995057106018066,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18295922130346298,
+      "step": 11192
+    },
+    {
+      "epoch": 0.22388,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.014357248942057291,
+      "learning_rate": 0.0001,
+      "loss": 4.3361,
+      "loss/crossentropy": 2.2150347232818604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24010418355464935,
+      "step": 11194
+    },
+    {
+      "epoch": 0.22392,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.014357248942057291,
+      "learning_rate": 0.0001,
+      "loss": 4.1711,
+      "loss/crossentropy": 2.1161770820617676,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2124250829219818,
+      "step": 11196
+    },
+    {
+      "epoch": 0.22396,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.013578033447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4875,
+      "loss/crossentropy": 2.224100112915039,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23966734111309052,
+      "step": 11198
+    },
+    {
+      "epoch": 0.224,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.011921946207682292,
+      "learning_rate": 0.0001,
+      "loss": 4.377,
+      "loss/crossentropy": 2.196989417076111,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21709956228733063,
+      "step": 11200
+    },
+    {
+      "epoch": 0.22404,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011574045817057291,
+      "learning_rate": 0.0001,
+      "loss": 4.0457,
+      "loss/crossentropy": 1.6592280864715576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1827787384390831,
+      "step": 11202
+    },
+    {
+      "epoch": 0.22408,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.008790842692057292,
+      "learning_rate": 0.0001,
+      "loss": 4.0564,
+      "loss/crossentropy": 1.9689467549324036,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20387520641088486,
+      "step": 11204
+    },
+    {
+      "epoch": 0.22412,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.03997802734375,
+      "learning_rate": 0.0001,
+      "loss": 4.3494,
+      "loss/crossentropy": 2.1846379041671753,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2068176046013832,
+      "step": 11206
+    },
+    {
+      "epoch": 0.22416,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.037353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3948,
+      "loss/crossentropy": 1.8361602425575256,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22452392429113388,
+      "step": 11208
+    },
+    {
+      "epoch": 0.2242,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3597,
+      "loss/crossentropy": 2.5658172369003296,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2602368891239166,
+      "step": 11210
+    },
+    {
+      "epoch": 0.22424,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.04119440714518229,
+      "learning_rate": 0.0001,
+      "loss": 4.0555,
+      "loss/crossentropy": 1.8167916536331177,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20330622047185898,
+      "step": 11212
+    },
+    {
+      "epoch": 0.22428,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.04070612589518229,
+      "learning_rate": 0.0001,
+      "loss": 4.2374,
+      "loss/crossentropy": 1.9151161313056946,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20695462822914124,
+      "step": 11214
+    },
+    {
+      "epoch": 0.22432,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.04528401692708333,
+      "learning_rate": 0.0001,
+      "loss": 3.9807,
+      "loss/crossentropy": 1.9590752720832825,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1945583075284958,
+      "step": 11216
+    },
+    {
+      "epoch": 0.22436,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.04522298177083333,
+      "learning_rate": 0.0001,
+      "loss": 4.2803,
+      "loss/crossentropy": 1.9777710437774658,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20104076713323593,
+      "step": 11218
+    },
+    {
+      "epoch": 0.2244,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.04744847615559896,
+      "learning_rate": 0.0001,
+      "loss": 4.0279,
+      "loss/crossentropy": 1.725940465927124,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19038556516170502,
+      "step": 11220
+    },
+    {
+      "epoch": 0.22444,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.012963612874348959,
+      "learning_rate": 0.0001,
+      "loss": 4.4067,
+      "loss/crossentropy": 2.505728602409363,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2434321641921997,
+      "step": 11222
+    },
+    {
+      "epoch": 0.22448,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.010518137613932292,
+      "learning_rate": 0.0001,
+      "loss": 4.2902,
+      "loss/crossentropy": 2.0733951330184937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21223794668912888,
+      "step": 11224
+    },
+    {
+      "epoch": 0.22452,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.010628255208333333,
+      "learning_rate": 0.0001,
+      "loss": 4.1548,
+      "loss/crossentropy": 2.105073928833008,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22444891929626465,
+      "step": 11226
+    },
+    {
+      "epoch": 0.22456,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009745025634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2361,
+      "loss/crossentropy": 1.9164994359016418,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20893585681915283,
+      "step": 11228
+    },
+    {
+      "epoch": 0.2246,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.014902496337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3753,
+      "loss/crossentropy": 2.0922536849975586,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22757402062416077,
+      "step": 11230
+    },
+    {
+      "epoch": 0.22464,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.01190185546875,
+      "learning_rate": 0.0001,
+      "loss": 3.9791,
+      "loss/crossentropy": 1.886056363582611,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2054021805524826,
+      "step": 11232
+    },
+    {
+      "epoch": 0.22468,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.014300282796223958,
+      "learning_rate": 0.0001,
+      "loss": 3.9194,
+      "loss/crossentropy": 2.028389871120453,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2109459862112999,
+      "step": 11234
+    },
+    {
+      "epoch": 0.22472,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.012813313802083334,
+      "learning_rate": 0.0001,
+      "loss": 4.2084,
+      "loss/crossentropy": 1.9590765833854675,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19774210453033447,
+      "step": 11236
+    },
+    {
+      "epoch": 0.22476,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.012800852457682291,
+      "learning_rate": 0.0001,
+      "loss": 4.0153,
+      "loss/crossentropy": 1.8284756541252136,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17485490441322327,
+      "step": 11238
+    },
+    {
+      "epoch": 0.2248,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.014625803629557291,
+      "learning_rate": 0.0001,
+      "loss": 4.3494,
+      "loss/crossentropy": 2.1119225025177,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22235971689224243,
+      "step": 11240
+    },
+    {
+      "epoch": 0.22484,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.015209706624348958,
+      "learning_rate": 0.0001,
+      "loss": 4.3303,
+      "loss/crossentropy": 2.3050538301467896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2249491587281227,
+      "step": 11242
+    },
+    {
+      "epoch": 0.22488,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.015360514322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.1235,
+      "loss/crossentropy": 1.9806398153305054,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20141054689884186,
+      "step": 11244
+    },
+    {
+      "epoch": 0.22492,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.007879384358723958,
+      "learning_rate": 0.0001,
+      "loss": 4.0793,
+      "loss/crossentropy": 2.0702012181282043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1980852484703064,
+      "step": 11246
+    },
+    {
+      "epoch": 0.22496,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.008113606770833334,
+      "learning_rate": 0.0001,
+      "loss": 3.9997,
+      "loss/crossentropy": 2.205365300178528,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22107956558465958,
+      "step": 11248
+    },
+    {
+      "epoch": 0.225,
+      "grad_norm": 2.890625,
+      "grad_norm_var": 0.05272598266601562,
+      "learning_rate": 0.0001,
+      "loss": 4.0748,
+      "loss/crossentropy": 2.196021556854248,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.199070006608963,
+      "step": 11250
+    },
+    {
+      "epoch": 0.22504,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.05241673787434896,
+      "learning_rate": 0.0001,
+      "loss": 4.2062,
+      "loss/crossentropy": 2.0661864280700684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22117173671722412,
+      "step": 11252
+    },
+    {
+      "epoch": 0.22508,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.050687408447265624,
+      "learning_rate": 0.0001,
+      "loss": 3.9176,
+      "loss/crossentropy": 1.825901210308075,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.174439437687397,
+      "step": 11254
+    },
+    {
+      "epoch": 0.22512,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.061470286051432295,
+      "learning_rate": 0.0001,
+      "loss": 4.7052,
+      "loss/crossentropy": 2.3710498809814453,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2357504665851593,
+      "step": 11256
+    },
+    {
+      "epoch": 0.22516,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.06520182291666667,
+      "learning_rate": 0.0001,
+      "loss": 3.9419,
+      "loss/crossentropy": 2.1523157358169556,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2179568186402321,
+      "step": 11258
+    },
+    {
+      "epoch": 0.2252,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.06559015909830729,
+      "learning_rate": 0.0001,
+      "loss": 4.1953,
+      "loss/crossentropy": 2.0974292755126953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1980036199092865,
+      "step": 11260
+    },
+    {
+      "epoch": 0.22524,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.06339925130208333,
+      "learning_rate": 0.0001,
+      "loss": 4.4409,
+      "loss/crossentropy": 2.2679883241653442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22085320204496384,
+      "step": 11262
+    },
+    {
+      "epoch": 0.22528,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.06108373006184896,
+      "learning_rate": 0.0001,
+      "loss": 4.2972,
+      "loss/crossentropy": 2.0697352290153503,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21976519376039505,
+      "step": 11264
+    },
+    {
+      "epoch": 0.22532,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.025465647379557293,
+      "learning_rate": 0.0001,
+      "loss": 4.4149,
+      "loss/crossentropy": 2.1807767748832703,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21304857730865479,
+      "step": 11266
+    },
+    {
+      "epoch": 0.22536,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.02541071573893229,
+      "learning_rate": 0.0001,
+      "loss": 4.4395,
+      "loss/crossentropy": 2.178891122341156,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22129195928573608,
+      "step": 11268
+    },
+    {
+      "epoch": 0.2254,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.023607381184895835,
+      "learning_rate": 0.0001,
+      "loss": 4.2908,
+      "loss/crossentropy": 2.1025387048721313,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20581847429275513,
+      "step": 11270
+    },
+    {
+      "epoch": 0.22544,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.013055165608723959,
+      "learning_rate": 0.0001,
+      "loss": 4.2192,
+      "loss/crossentropy": 1.8710424900054932,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1900918111205101,
+      "step": 11272
+    },
+    {
+      "epoch": 0.22548,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.012373606363932291,
+      "learning_rate": 0.0001,
+      "loss": 4.498,
+      "loss/crossentropy": 1.8837561011314392,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21019018441438675,
+      "step": 11274
+    },
+    {
+      "epoch": 0.22552,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.013132476806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3139,
+      "loss/crossentropy": 2.2795380353927612,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23858627676963806,
+      "step": 11276
+    },
+    {
+      "epoch": 0.22556,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0158111572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0532,
+      "loss/crossentropy": 2.0241262316703796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1990864798426628,
+      "step": 11278
+    },
+    {
+      "epoch": 0.2256,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.022541300455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2935,
+      "loss/crossentropy": 1.9307058453559875,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2100597321987152,
+      "step": 11280
+    },
+    {
+      "epoch": 0.22564,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.020414225260416665,
+      "learning_rate": 0.0001,
+      "loss": 4.2523,
+      "loss/crossentropy": 2.08352792263031,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21930356323719025,
+      "step": 11282
+    },
+    {
+      "epoch": 0.22568,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.021361287434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.4277,
+      "loss/crossentropy": 2.393290877342224,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2560981214046478,
+      "step": 11284
+    },
+    {
+      "epoch": 0.22572,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.023738606770833334,
+      "learning_rate": 0.0001,
+      "loss": 4.1612,
+      "loss/crossentropy": 2.20097017288208,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20318082720041275,
+      "step": 11286
+    },
+    {
+      "epoch": 0.22576,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0221099853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2307,
+      "loss/crossentropy": 2.031981647014618,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20313747972249985,
+      "step": 11288
+    },
+    {
+      "epoch": 0.2258,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0187255859375,
+      "learning_rate": 0.0001,
+      "loss": 4.3513,
+      "loss/crossentropy": 2.197006046772003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2301759570837021,
+      "step": 11290
+    },
+    {
+      "epoch": 0.22584,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0197021484375,
+      "learning_rate": 0.0001,
+      "loss": 4.2903,
+      "loss/crossentropy": 2.294014096260071,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21280069649219513,
+      "step": 11292
+    },
+    {
+      "epoch": 0.22588,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.018089803059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0985,
+      "loss/crossentropy": 2.0551719665527344,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20910422503948212,
+      "step": 11294
+    },
+    {
+      "epoch": 0.22592,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.010636393229166667,
+      "learning_rate": 0.0001,
+      "loss": 4.2109,
+      "loss/crossentropy": 2.1044358015060425,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20463456213474274,
+      "step": 11296
+    },
+    {
+      "epoch": 0.22596,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.012631988525390625,
+      "learning_rate": 0.0001,
+      "loss": 4.198,
+      "loss/crossentropy": 2.1013529300689697,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21202504634857178,
+      "step": 11298
+    },
+    {
+      "epoch": 0.226,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.009329986572265626,
+      "learning_rate": 0.0001,
+      "loss": 4.4227,
+      "loss/crossentropy": 2.1887649297714233,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2048492729663849,
+      "step": 11300
+    },
+    {
+      "epoch": 0.22604,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.009525299072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0952,
+      "loss/crossentropy": 1.8243364691734314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2013952136039734,
+      "step": 11302
+    },
+    {
+      "epoch": 0.22608,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0137451171875,
+      "learning_rate": 0.0001,
+      "loss": 4.4636,
+      "loss/crossentropy": 2.2550541162490845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2493698000907898,
+      "step": 11304
+    },
+    {
+      "epoch": 0.22612,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.0149169921875,
+      "learning_rate": 0.0001,
+      "loss": 3.9407,
+      "loss/crossentropy": 1.8744492530822754,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19957631081342697,
+      "step": 11306
+    },
+    {
+      "epoch": 0.22616,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.013681793212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0717,
+      "loss/crossentropy": 2.026526629924774,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20474696904420853,
+      "step": 11308
+    },
+    {
+      "epoch": 0.2262,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.015881093343098958,
+      "learning_rate": 0.0001,
+      "loss": 4.3898,
+      "loss/crossentropy": 2.3802725076675415,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.247590571641922,
+      "step": 11310
+    },
+    {
+      "epoch": 0.22624,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.016123199462890626,
+      "learning_rate": 0.0001,
+      "loss": 4.171,
+      "loss/crossentropy": 1.9359605312347412,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.207523413002491,
+      "step": 11312
+    },
+    {
+      "epoch": 0.22628,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0129547119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1634,
+      "loss/crossentropy": 2.3035311698913574,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21815945208072662,
+      "step": 11314
+    },
+    {
+      "epoch": 0.22632,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0225341796875,
+      "learning_rate": 0.0001,
+      "loss": 4.3189,
+      "loss/crossentropy": 1.8083779215812683,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20705801248550415,
+      "step": 11316
+    },
+    {
+      "epoch": 0.22636,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0205718994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2912,
+      "loss/crossentropy": 2.306682825088501,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22666794806718826,
+      "step": 11318
+    },
+    {
+      "epoch": 0.2264,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.01737060546875,
+      "learning_rate": 0.0001,
+      "loss": 4.0519,
+      "loss/crossentropy": 1.809365153312683,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20744331926107407,
+      "step": 11320
+    },
+    {
+      "epoch": 0.22644,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.017894490559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.2428,
+      "loss/crossentropy": 2.289853572845459,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23791835457086563,
+      "step": 11322
+    },
+    {
+      "epoch": 0.22648,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.027341461181640624,
+      "learning_rate": 0.0001,
+      "loss": 4.5668,
+      "loss/crossentropy": 2.0609280467033386,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2166810780763626,
+      "step": 11324
+    },
+    {
+      "epoch": 0.22652,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.02535400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0567,
+      "loss/crossentropy": 2.1174912452697754,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20515341311693192,
+      "step": 11326
+    },
+    {
+      "epoch": 0.22656,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.029319000244140626,
+      "learning_rate": 0.0001,
+      "loss": 4.0318,
+      "loss/crossentropy": 1.99192476272583,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20127686113119125,
+      "step": 11328
+    },
+    {
+      "epoch": 0.2266,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.031060536702473957,
+      "learning_rate": 0.0001,
+      "loss": 3.9552,
+      "loss/crossentropy": 1.7805609107017517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17500489950180054,
+      "step": 11330
+    },
+    {
+      "epoch": 0.22664,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.02398656209309896,
+      "learning_rate": 0.0001,
+      "loss": 4.1417,
+      "loss/crossentropy": 2.024085283279419,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20659280568361282,
+      "step": 11332
+    },
+    {
+      "epoch": 0.22668,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.023631795247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.2674,
+      "loss/crossentropy": 2.174618899822235,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23250284045934677,
+      "step": 11334
+    },
+    {
+      "epoch": 0.22672,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.02451171875,
+      "learning_rate": 0.0001,
+      "loss": 4.213,
+      "loss/crossentropy": 1.886943757534027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18855806440114975,
+      "step": 11336
+    },
+    {
+      "epoch": 0.22676,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.02182184855143229,
+      "learning_rate": 0.0001,
+      "loss": 4.1281,
+      "loss/crossentropy": 2.060371160507202,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1951771154999733,
+      "step": 11338
+    },
+    {
+      "epoch": 0.2268,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.005576324462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1692,
+      "loss/crossentropy": 2.15705668926239,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23738030344247818,
+      "step": 11340
+    },
+    {
+      "epoch": 0.22684,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.005576324462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2096,
+      "loss/crossentropy": 2.1922959089279175,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24581187963485718,
+      "step": 11342
+    },
+    {
+      "epoch": 0.22688,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.007303873697916667,
+      "learning_rate": 0.0001,
+      "loss": 3.9786,
+      "loss/crossentropy": 2.1590365171432495,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20936457812786102,
+      "step": 11344
+    },
+    {
+      "epoch": 0.22692,
+      "grad_norm": 2.53125,
+      "grad_norm_var": 0.020783487955729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2382,
+      "loss/crossentropy": 1.8087702989578247,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18874070048332214,
+      "step": 11346
+    },
+    {
+      "epoch": 0.22696,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.021732584635416666,
+      "learning_rate": 0.0001,
+      "loss": 4.4981,
+      "loss/crossentropy": 2.5205971002578735,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24057865887880325,
+      "step": 11348
+    },
+    {
+      "epoch": 0.227,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.02061945597330729,
+      "learning_rate": 0.0001,
+      "loss": 4.16,
+      "loss/crossentropy": 2.114508092403412,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20631257444620132,
+      "step": 11350
+    },
+    {
+      "epoch": 0.22704,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.020881144205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4344,
+      "loss/crossentropy": 2.385537028312683,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.233476921916008,
+      "step": 11352
+    },
+    {
+      "epoch": 0.22708,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0217926025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3335,
+      "loss/crossentropy": 2.4251039028167725,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24502182751893997,
+      "step": 11354
+    },
+    {
+      "epoch": 0.22712,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.025690714518229168,
+      "learning_rate": 0.0001,
+      "loss": 4.345,
+      "loss/crossentropy": 2.278030514717102,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22328510880470276,
+      "step": 11356
+    },
+    {
+      "epoch": 0.22716,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.025031534830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.2835,
+      "loss/crossentropy": 1.7373265027999878,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17635879665613174,
+      "step": 11358
+    },
+    {
+      "epoch": 0.2272,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.024461873372395835,
+      "learning_rate": 0.0001,
+      "loss": 4.3644,
+      "loss/crossentropy": 2.1546722650527954,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22285740077495575,
+      "step": 11360
+    },
+    {
+      "epoch": 0.22724,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0143218994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.5415,
+      "loss/crossentropy": 2.341711401939392,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23693612217903137,
+      "step": 11362
+    },
+    {
+      "epoch": 0.22728,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.0461090087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3131,
+      "loss/crossentropy": 1.8764930367469788,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19202134013175964,
+      "step": 11364
+    },
+    {
+      "epoch": 0.22732,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0450439453125,
+      "learning_rate": 0.0001,
+      "loss": 4.1172,
+      "loss/crossentropy": 2.035600185394287,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2188200280070305,
+      "step": 11366
+    },
+    {
+      "epoch": 0.22736,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.045481109619140626,
+      "learning_rate": 0.0001,
+      "loss": 4.0469,
+      "loss/crossentropy": 2.120850682258606,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22664117813110352,
+      "step": 11368
+    },
+    {
+      "epoch": 0.2274,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.050455729166666664,
+      "learning_rate": 0.0001,
+      "loss": 3.9552,
+      "loss/crossentropy": 1.9753262996673584,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20000861585140228,
+      "step": 11370
+    },
+    {
+      "epoch": 0.22744,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.06503499348958333,
+      "learning_rate": 0.0001,
+      "loss": 4.2474,
+      "loss/crossentropy": 2.0264564156532288,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2026129513978958,
+      "step": 11372
+    },
+    {
+      "epoch": 0.22748,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.07055562337239583,
+      "learning_rate": 0.0001,
+      "loss": 3.8546,
+      "loss/crossentropy": 1.9269734025001526,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22436296939849854,
+      "step": 11374
+    },
+    {
+      "epoch": 0.22752,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.07017822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4164,
+      "loss/crossentropy": 2.1881991624832153,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22483092546463013,
+      "step": 11376
+    },
+    {
+      "epoch": 0.22756,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.07446187337239583,
+      "learning_rate": 0.0001,
+      "loss": 4.0277,
+      "loss/crossentropy": 2.0142401456832886,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21913451701402664,
+      "step": 11378
+    },
+    {
+      "epoch": 0.2276,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.05676676432291667,
+      "learning_rate": 0.0001,
+      "loss": 4.3374,
+      "loss/crossentropy": 2.0229859352111816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.238722562789917,
+      "step": 11380
+    },
+    {
+      "epoch": 0.22764,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.05724054972330729,
+      "learning_rate": 0.0001,
+      "loss": 4.0223,
+      "loss/crossentropy": 2.043630540370941,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21258120238780975,
+      "step": 11382
+    },
+    {
+      "epoch": 0.22768,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.06020685831705729,
+      "learning_rate": 0.0001,
+      "loss": 4.0752,
+      "loss/crossentropy": 2.702946662902832,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24362115561962128,
+      "step": 11384
+    },
+    {
+      "epoch": 0.22772,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.05449600219726562,
+      "learning_rate": 0.0001,
+      "loss": 3.9806,
+      "loss/crossentropy": 2.243411898612976,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22256402671337128,
+      "step": 11386
+    },
+    {
+      "epoch": 0.22776,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.033699289957682295,
+      "learning_rate": 0.0001,
+      "loss": 3.7808,
+      "loss/crossentropy": 2.1451956033706665,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21471337974071503,
+      "step": 11388
+    },
+    {
+      "epoch": 0.2278,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.030775705973307293,
+      "learning_rate": 0.0001,
+      "loss": 4.2865,
+      "loss/crossentropy": 2.426279664039612,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22846391052007675,
+      "step": 11390
+    },
+    {
+      "epoch": 0.22784,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.033455149332682295,
+      "learning_rate": 0.0001,
+      "loss": 4.2878,
+      "loss/crossentropy": 2.0890655517578125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23008010536432266,
+      "step": 11392
+    },
+    {
+      "epoch": 0.22788,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.030452219645182292,
+      "learning_rate": 0.0001,
+      "loss": 4.2124,
+      "loss/crossentropy": 2.3496296405792236,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23718956112861633,
+      "step": 11394
+    },
+    {
+      "epoch": 0.22792,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.013152821858723959,
+      "learning_rate": 0.0001,
+      "loss": 4.0025,
+      "loss/crossentropy": 2.02141535282135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21102941036224365,
+      "step": 11396
+    },
+    {
+      "epoch": 0.22796,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.01275634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4817,
+      "loss/crossentropy": 2.213461995124817,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21838735044002533,
+      "step": 11398
+    },
+    {
+      "epoch": 0.228,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.11743062337239583,
+      "learning_rate": 0.0001,
+      "loss": 4.2313,
+      "loss/crossentropy": 2.161481499671936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24635545909404755,
+      "step": 11400
+    },
+    {
+      "epoch": 0.22804,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.1158843994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.5372,
+      "loss/crossentropy": 2.0998951196670532,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2205812931060791,
+      "step": 11402
+    },
+    {
+      "epoch": 0.22808,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.11588109334309896,
+      "learning_rate": 0.0001,
+      "loss": 4.1606,
+      "loss/crossentropy": 1.8785207867622375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20330313593149185,
+      "step": 11404
+    },
+    {
+      "epoch": 0.22812,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.11588109334309896,
+      "learning_rate": 0.0001,
+      "loss": 4.2408,
+      "loss/crossentropy": 2.254656672477722,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.220963753759861,
+      "step": 11406
+    },
+    {
+      "epoch": 0.22816,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.11553726196289063,
+      "learning_rate": 0.0001,
+      "loss": 4.1356,
+      "loss/crossentropy": 1.7936404347419739,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1878006011247635,
+      "step": 11408
+    },
+    {
+      "epoch": 0.2282,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.12290445963541667,
+      "learning_rate": 0.0001,
+      "loss": 4.2199,
+      "loss/crossentropy": 1.8537682890892029,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1893099844455719,
+      "step": 11410
+    },
+    {
+      "epoch": 0.22824,
+      "grad_norm": 3.234375,
+      "grad_norm_var": 0.1904205322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1628,
+      "loss/crossentropy": 2.0987170338630676,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2072230949997902,
+      "step": 11412
+    },
+    {
+      "epoch": 0.22828,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.1907958984375,
+      "learning_rate": 0.0001,
+      "loss": 4.2473,
+      "loss/crossentropy": 2.049329698085785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2142597660422325,
+      "step": 11414
+    },
+    {
+      "epoch": 0.22832,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.09533284505208334,
+      "learning_rate": 0.0001,
+      "loss": 4.283,
+      "loss/crossentropy": 2.066833019256592,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22086824476718903,
+      "step": 11416
+    },
+    {
+      "epoch": 0.22836,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.09519755045572917,
+      "learning_rate": 0.0001,
+      "loss": 4.1837,
+      "loss/crossentropy": 2.01213002204895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23191991448402405,
+      "step": 11418
+    },
+    {
+      "epoch": 0.2284,
+      "grad_norm": 5.6875,
+      "grad_norm_var": 0.8556495666503906,
+      "learning_rate": 0.0001,
+      "loss": 4.1977,
+      "loss/crossentropy": 2.1349334716796875,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22979671508073807,
+      "step": 11420
+    },
+    {
+      "epoch": 0.22844,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.8487709045410157,
+      "learning_rate": 0.0001,
+      "loss": 4.2101,
+      "loss/crossentropy": 1.9201850295066833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21903745830059052,
+      "step": 11422
+    },
+    {
+      "epoch": 0.22848,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.8546376546223958,
+      "learning_rate": 0.0001,
+      "loss": 4.1927,
+      "loss/crossentropy": 2.011150360107422,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2059009075164795,
+      "step": 11424
+    },
+    {
+      "epoch": 0.22852,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.8533322652180989,
+      "learning_rate": 0.0001,
+      "loss": 4.3762,
+      "loss/crossentropy": 2.1529648303985596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23028475046157837,
+      "step": 11426
+    },
+    {
+      "epoch": 0.22856,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.8176177978515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9657,
+      "loss/crossentropy": 2.1602566838264465,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2262946516275406,
+      "step": 11428
+    },
+    {
+      "epoch": 0.2286,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.829766591389974,
+      "learning_rate": 0.0001,
+      "loss": 3.9369,
+      "loss/crossentropy": 1.7230631113052368,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19715693593025208,
+      "step": 11430
+    },
+    {
+      "epoch": 0.22864,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.8287737528483073,
+      "learning_rate": 0.0001,
+      "loss": 4.3791,
+      "loss/crossentropy": 2.4396276473999023,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24283458292484283,
+      "step": 11432
+    },
+    {
+      "epoch": 0.22868,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.8311480204264323,
+      "learning_rate": 0.0001,
+      "loss": 4.2692,
+      "loss/crossentropy": 1.8615645170211792,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21288780122995377,
+      "step": 11434
+    },
+    {
+      "epoch": 0.22872,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.012941233317057292,
+      "learning_rate": 0.0001,
+      "loss": 4.394,
+      "loss/crossentropy": 1.8972707390785217,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18510619550943375,
+      "step": 11436
+    },
+    {
+      "epoch": 0.22876,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.011980946858723958,
+      "learning_rate": 0.0001,
+      "loss": 4.1459,
+      "loss/crossentropy": 2.217754364013672,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21217356622219086,
+      "step": 11438
+    },
+    {
+      "epoch": 0.2288,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.014818318684895833,
+      "learning_rate": 0.0001,
+      "loss": 3.7955,
+      "loss/crossentropy": 1.6481398940086365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17804963141679764,
+      "step": 11440
+    },
+    {
+      "epoch": 0.22884,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.01375732421875,
+      "learning_rate": 0.0001,
+      "loss": 4.3291,
+      "loss/crossentropy": 2.026508390903473,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20739784091711044,
+      "step": 11442
+    },
+    {
+      "epoch": 0.22888,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.012334950764973958,
+      "learning_rate": 0.0001,
+      "loss": 4.2086,
+      "loss/crossentropy": 2.05685293674469,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22639908641576767,
+      "step": 11444
+    },
+    {
+      "epoch": 0.22892,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.01197509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4689,
+      "loss/crossentropy": 2.073192059993744,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21727359294891357,
+      "step": 11446
+    },
+    {
+      "epoch": 0.22896,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0092193603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4312,
+      "loss/crossentropy": 2.1311055421829224,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19499187916517258,
+      "step": 11448
+    },
+    {
+      "epoch": 0.229,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.01021728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2043,
+      "loss/crossentropy": 2.169051766395569,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22266974300146103,
+      "step": 11450
+    },
+    {
+      "epoch": 0.22904,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.013727823893229166,
+      "learning_rate": 0.0001,
+      "loss": 4.1784,
+      "loss/crossentropy": 2.022417426109314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20541484653949738,
+      "step": 11452
+    },
+    {
+      "epoch": 0.22908,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.015458170572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.2921,
+      "loss/crossentropy": 2.2652071714401245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21639510244131088,
+      "step": 11454
+    },
+    {
+      "epoch": 0.22912,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.01343994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1575,
+      "loss/crossentropy": 1.912036418914795,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2204061597585678,
+      "step": 11456
+    },
+    {
+      "epoch": 0.22916,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.014188639322916667,
+      "learning_rate": 0.0001,
+      "loss": 4.4497,
+      "loss/crossentropy": 2.029780328273773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2032444253563881,
+      "step": 11458
+    },
+    {
+      "epoch": 0.2292,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.016039021809895835,
+      "learning_rate": 0.0001,
+      "loss": 4.0627,
+      "loss/crossentropy": 1.954946756362915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2028508484363556,
+      "step": 11460
+    },
+    {
+      "epoch": 0.22924,
+      "grad_norm": 2.8125,
+      "grad_norm_var": 0.049494425455729164,
+      "learning_rate": 0.0001,
+      "loss": 4.5787,
+      "loss/crossentropy": 2.3707855939865112,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2460392713546753,
+      "step": 11462
+    },
+    {
+      "epoch": 0.22928,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.047972615559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1509,
+      "loss/crossentropy": 2.410157322883606,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24578960239887238,
+      "step": 11464
+    },
+    {
+      "epoch": 0.22932,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.049117024739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.0935,
+      "loss/crossentropy": 2.1403380036354065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20563968271017075,
+      "step": 11466
+    },
+    {
+      "epoch": 0.22936,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.0471588134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0991,
+      "loss/crossentropy": 2.1596190333366394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24405791610479355,
+      "step": 11468
+    },
+    {
+      "epoch": 0.2294,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.0506744384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3973,
+      "loss/crossentropy": 1.9427857398986816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22412577271461487,
+      "step": 11470
+    },
+    {
+      "epoch": 0.22944,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.05306574503580729,
+      "learning_rate": 0.0001,
+      "loss": 4.1093,
+      "loss/crossentropy": 2.078735053539276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20844466239213943,
+      "step": 11472
+    },
+    {
+      "epoch": 0.22948,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.054323069254557294,
+      "learning_rate": 0.0001,
+      "loss": 4.0749,
+      "loss/crossentropy": 2.351656198501587,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24430027604103088,
+      "step": 11474
+    },
+    {
+      "epoch": 0.22952,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.05269139607747396,
+      "learning_rate": 0.0001,
+      "loss": 3.9647,
+      "loss/crossentropy": 2.057854652404785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21224451810121536,
+      "step": 11476
+    },
+    {
+      "epoch": 0.22956,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.018534088134765626,
+      "learning_rate": 0.0001,
+      "loss": 4.1451,
+      "loss/crossentropy": 2.207979917526245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22855235636234283,
+      "step": 11478
+    },
+    {
+      "epoch": 0.2296,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.01870905558268229,
+      "learning_rate": 0.0001,
+      "loss": 4.3748,
+      "loss/crossentropy": 2.08840012550354,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21988992393016815,
+      "step": 11480
+    },
+    {
+      "epoch": 0.22964,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.017144521077473957,
+      "learning_rate": 0.0001,
+      "loss": 4.2314,
+      "loss/crossentropy": 2.1339274644851685,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2278972491621971,
+      "step": 11482
+    },
+    {
+      "epoch": 0.22968,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.015386708577473958,
+      "learning_rate": 0.0001,
+      "loss": 4.2013,
+      "loss/crossentropy": 1.970679223537445,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20286859571933746,
+      "step": 11484
+    },
+    {
+      "epoch": 0.22972,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.005295562744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2443,
+      "loss/crossentropy": 2.1023008823394775,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21906304359436035,
+      "step": 11486
+    },
+    {
+      "epoch": 0.22976,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.004870351155598958,
+      "learning_rate": 0.0001,
+      "loss": 4.2774,
+      "loss/crossentropy": 2.1415608525276184,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21857701241970062,
+      "step": 11488
+    },
+    {
+      "epoch": 0.2298,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.005191802978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0697,
+      "loss/crossentropy": 1.9523325562477112,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19859597831964493,
+      "step": 11490
+    },
+    {
+      "epoch": 0.22984,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0046770731608072914,
+      "learning_rate": 0.0001,
+      "loss": 4.2433,
+      "loss/crossentropy": 2.1532927751541138,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2239024043083191,
+      "step": 11492
+    },
+    {
+      "epoch": 0.22988,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.006705474853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3556,
+      "loss/crossentropy": 2.160528779029846,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23207154870033264,
+      "step": 11494
+    },
+    {
+      "epoch": 0.22992,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.006528472900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.277,
+      "loss/crossentropy": 2.07854962348938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2189328968524933,
+      "step": 11496
+    },
+    {
+      "epoch": 0.22996,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.009069569905598958,
+      "learning_rate": 0.0001,
+      "loss": 4.4237,
+      "loss/crossentropy": 2.2270501852035522,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23237968981266022,
+      "step": 11498
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.0125396728515625,
+      "learning_rate": 0.0001,
+      "loss": 3.96,
+      "loss/crossentropy": 1.9641701579093933,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2063433900475502,
+      "step": 11500
+    },
+    {
+      "epoch": 0.23004,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.016866048177083332,
+      "learning_rate": 0.0001,
+      "loss": 4.3002,
+      "loss/crossentropy": 1.9243032932281494,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20478814095258713,
+      "step": 11502
+    },
+    {
+      "epoch": 0.23008,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.01587092081705729,
+      "learning_rate": 0.0001,
+      "loss": 4.329,
+      "loss/crossentropy": 2.295292854309082,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23184886574745178,
+      "step": 11504
+    },
+    {
+      "epoch": 0.23012,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.01654052734375,
+      "learning_rate": 0.0001,
+      "loss": 4.0697,
+      "loss/crossentropy": 2.134859561920166,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21923956274986267,
+      "step": 11506
+    },
+    {
+      "epoch": 0.23016,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0158355712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2523,
+      "loss/crossentropy": 1.905085265636444,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22443564236164093,
+      "step": 11508
+    },
+    {
+      "epoch": 0.2302,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.014892323811848959,
+      "learning_rate": 0.0001,
+      "loss": 4.0296,
+      "loss/crossentropy": 1.6967324614524841,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19518497586250305,
+      "step": 11510
+    },
+    {
+      "epoch": 0.23024,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.015364583333333333,
+      "learning_rate": 0.0001,
+      "loss": 4.225,
+      "loss/crossentropy": 1.638957679271698,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19098830223083496,
+      "step": 11512
+    },
+    {
+      "epoch": 0.23028,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.013732655843098959,
+      "learning_rate": 0.0001,
+      "loss": 4.0479,
+      "loss/crossentropy": 1.8760477900505066,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.202397421002388,
+      "step": 11514
+    },
+    {
+      "epoch": 0.23032,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.012189737955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.2265,
+      "loss/crossentropy": 2.201690912246704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22841450572013855,
+      "step": 11516
+    },
+    {
+      "epoch": 0.23036,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.007987467447916667,
+      "learning_rate": 0.0001,
+      "loss": 4.5451,
+      "loss/crossentropy": 2.5022183656692505,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25180216133594513,
+      "step": 11518
+    },
+    {
+      "epoch": 0.2304,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.007682291666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.2848,
+      "loss/crossentropy": 2.4634610414505005,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23911123722791672,
+      "step": 11520
+    },
+    {
+      "epoch": 0.23044,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.007503000895182291,
+      "learning_rate": 0.0001,
+      "loss": 4.2919,
+      "loss/crossentropy": 2.1176512241363525,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20835164189338684,
+      "step": 11522
+    },
+    {
+      "epoch": 0.23048,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.007252756754557292,
+      "learning_rate": 0.0001,
+      "loss": 4.0803,
+      "loss/crossentropy": 1.7708171606063843,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18085652589797974,
+      "step": 11524
+    },
+    {
+      "epoch": 0.23052,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.008504231770833334,
+      "learning_rate": 0.0001,
+      "loss": 4.1159,
+      "loss/crossentropy": 1.6740695238113403,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17464587092399597,
+      "step": 11526
+    },
+    {
+      "epoch": 0.23056,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.008135732014973958,
+      "learning_rate": 0.0001,
+      "loss": 4.1829,
+      "loss/crossentropy": 1.8153178691864014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19107046723365784,
+      "step": 11528
+    },
+    {
+      "epoch": 0.2306,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.006231435139973958,
+      "learning_rate": 0.0001,
+      "loss": 4.2074,
+      "loss/crossentropy": 2.2833873629570007,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19900934398174286,
+      "step": 11530
+    },
+    {
+      "epoch": 0.23064,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008593495686848958,
+      "learning_rate": 0.0001,
+      "loss": 4.3154,
+      "loss/crossentropy": 1.9106029272079468,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1996043100953102,
+      "step": 11532
+    },
+    {
+      "epoch": 0.23068,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.006030019124348958,
+      "learning_rate": 0.0001,
+      "loss": 4.3849,
+      "loss/crossentropy": 2.049258530139923,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21438845992088318,
+      "step": 11534
+    },
+    {
+      "epoch": 0.23072,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.006029256184895833,
+      "learning_rate": 0.0001,
+      "loss": 3.9236,
+      "loss/crossentropy": 1.9124351739883423,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20111311972141266,
+      "step": 11536
+    },
+    {
+      "epoch": 0.23076,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0059397379557291664,
+      "learning_rate": 0.0001,
+      "loss": 4.0781,
+      "loss/crossentropy": 1.8045400381088257,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20846854895353317,
+      "step": 11538
+    },
+    {
+      "epoch": 0.2308,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.007297515869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1804,
+      "loss/crossentropy": 1.9970109462738037,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20165999233722687,
+      "step": 11540
+    },
+    {
+      "epoch": 0.23084,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.013065338134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1756,
+      "loss/crossentropy": 2.0300097465515137,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24561113119125366,
+      "step": 11542
+    },
+    {
+      "epoch": 0.23088,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.025402577718098958,
+      "learning_rate": 0.0001,
+      "loss": 4.4841,
+      "loss/crossentropy": 2.194391131401062,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24075081944465637,
+      "step": 11544
+    },
+    {
+      "epoch": 0.23092,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0239898681640625,
+      "learning_rate": 0.0001,
+      "loss": 3.9177,
+      "loss/crossentropy": 2.0229761600494385,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21907109022140503,
+      "step": 11546
+    },
+    {
+      "epoch": 0.23096,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.022606404622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3089,
+      "loss/crossentropy": 1.831793487071991,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18943443894386292,
+      "step": 11548
+    },
+    {
+      "epoch": 0.231,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.023395792643229166,
+      "learning_rate": 0.0001,
+      "loss": 4.19,
+      "loss/crossentropy": 2.4099985361099243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22612392157316208,
+      "step": 11550
+    },
+    {
+      "epoch": 0.23104,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.022946929931640624,
+      "learning_rate": 0.0001,
+      "loss": 4.2512,
+      "loss/crossentropy": 2.0496281385421753,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20928414165973663,
+      "step": 11552
+    },
+    {
+      "epoch": 0.23108,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.025986480712890624,
+      "learning_rate": 0.0001,
+      "loss": 4.196,
+      "loss/crossentropy": 1.9366755485534668,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19255827367305756,
+      "step": 11554
+    },
+    {
+      "epoch": 0.23112,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.02516454060872396,
+      "learning_rate": 0.0001,
+      "loss": 4.3758,
+      "loss/crossentropy": 1.9077317714691162,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20340882241725922,
+      "step": 11556
+    },
+    {
+      "epoch": 0.23116,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.02020848592122396,
+      "learning_rate": 0.0001,
+      "loss": 4.091,
+      "loss/crossentropy": 1.9167283773422241,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21566946804523468,
+      "step": 11558
+    },
+    {
+      "epoch": 0.2312,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0059506734212239586,
+      "learning_rate": 0.0001,
+      "loss": 4.5758,
+      "loss/crossentropy": 2.4604564905166626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22104239463806152,
+      "step": 11560
+    },
+    {
+      "epoch": 0.23124,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.009124501546223959,
+      "learning_rate": 0.0001,
+      "loss": 4.0398,
+      "loss/crossentropy": 2.359586775302887,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2064266949892044,
+      "step": 11562
+    },
+    {
+      "epoch": 0.23128,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.009456125895182292,
+      "learning_rate": 0.0001,
+      "loss": 4.0065,
+      "loss/crossentropy": 1.9757064580917358,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19224581122398376,
+      "step": 11564
+    },
+    {
+      "epoch": 0.23132,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.03843561808268229,
+      "learning_rate": 0.0001,
+      "loss": 4.5544,
+      "loss/crossentropy": 1.9888432025909424,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20799735933542252,
+      "step": 11566
+    },
+    {
+      "epoch": 0.23136,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.03802057902018229,
+      "learning_rate": 0.0001,
+      "loss": 4.2281,
+      "loss/crossentropy": 2.0829046964645386,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22345586121082306,
+      "step": 11568
+    },
+    {
+      "epoch": 0.2314,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.036649322509765624,
+      "learning_rate": 0.0001,
+      "loss": 4.3421,
+      "loss/crossentropy": 1.964626431465149,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21068184822797775,
+      "step": 11570
+    },
+    {
+      "epoch": 0.23144,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.034993489583333336,
+      "learning_rate": 0.0001,
+      "loss": 4.4206,
+      "loss/crossentropy": 2.313928008079529,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23291154205799103,
+      "step": 11572
+    },
+    {
+      "epoch": 0.23148,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.038386027018229164,
+      "learning_rate": 0.0001,
+      "loss": 3.9609,
+      "loss/crossentropy": 2.221343159675598,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20674917846918106,
+      "step": 11574
+    },
+    {
+      "epoch": 0.23152,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.03853759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4097,
+      "loss/crossentropy": 2.070296823978424,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22503886371850967,
+      "step": 11576
+    },
+    {
+      "epoch": 0.23156,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.03483784993489583,
+      "learning_rate": 0.0001,
+      "loss": 3.9417,
+      "loss/crossentropy": 2.21540367603302,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2334480583667755,
+      "step": 11578
+    },
+    {
+      "epoch": 0.2316,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.03614679972330729,
+      "learning_rate": 0.0001,
+      "loss": 4.2491,
+      "loss/crossentropy": 1.9438464641571045,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20312584936618805,
+      "step": 11580
+    },
+    {
+      "epoch": 0.23164,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.010994211832682291,
+      "learning_rate": 0.0001,
+      "loss": 4.0622,
+      "loss/crossentropy": 1.4970324039459229,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17415452748537064,
+      "step": 11582
+    },
+    {
+      "epoch": 0.23168,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.026244099934895834,
+      "learning_rate": 0.0001,
+      "loss": 4.2293,
+      "loss/crossentropy": 2.2618579864501953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23066900670528412,
+      "step": 11584
+    },
+    {
+      "epoch": 0.23172,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.02603123982747396,
+      "learning_rate": 0.0001,
+      "loss": 4.0974,
+      "loss/crossentropy": 2.2342761754989624,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20567959547042847,
+      "step": 11586
+    },
+    {
+      "epoch": 0.23176,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.026151275634765624,
+      "learning_rate": 0.0001,
+      "loss": 4.0417,
+      "loss/crossentropy": 2.144785463809967,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19473369419574738,
+      "step": 11588
+    },
+    {
+      "epoch": 0.2318,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.02851130167643229,
+      "learning_rate": 0.0001,
+      "loss": 4.1996,
+      "loss/crossentropy": 1.9032491445541382,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1956048086285591,
+      "step": 11590
+    },
+    {
+      "epoch": 0.23184,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.03227717081705729,
+      "learning_rate": 0.0001,
+      "loss": 4.4689,
+      "loss/crossentropy": 1.934233546257019,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22426530718803406,
+      "step": 11592
+    },
+    {
+      "epoch": 0.23188,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.03166071573893229,
+      "learning_rate": 0.0001,
+      "loss": 4.2321,
+      "loss/crossentropy": 2.1792030930519104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22469744831323624,
+      "step": 11594
+    },
+    {
+      "epoch": 0.23192,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.027705891927083334,
+      "learning_rate": 0.0001,
+      "loss": 4.3779,
+      "loss/crossentropy": 2.3242534399032593,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23015478998422623,
+      "step": 11596
+    },
+    {
+      "epoch": 0.23196,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.036622873942057294,
+      "learning_rate": 0.0001,
+      "loss": 4.3553,
+      "loss/crossentropy": 2.2343804836273193,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2392263486981392,
+      "step": 11598
+    },
+    {
+      "epoch": 0.232,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.024559529622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3019,
+      "loss/crossentropy": 1.9622855186462402,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22550886124372482,
+      "step": 11600
+    },
+    {
+      "epoch": 0.23204,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.023981730143229168,
+      "learning_rate": 0.0001,
+      "loss": 4.4955,
+      "loss/crossentropy": 2.2274144887924194,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21090717613697052,
+      "step": 11602
+    },
+    {
+      "epoch": 0.23208,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.021683756510416666,
+      "learning_rate": 0.0001,
+      "loss": 4.3011,
+      "loss/crossentropy": 2.092648506164551,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21386967599391937,
+      "step": 11604
+    },
+    {
+      "epoch": 0.23212,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.020643870035807293,
+      "learning_rate": 0.0001,
+      "loss": 4.2739,
+      "loss/crossentropy": 2.3358936309814453,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22385042905807495,
+      "step": 11606
+    },
+    {
+      "epoch": 0.23216,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.01883112589518229,
+      "learning_rate": 0.0001,
+      "loss": 4.1968,
+      "loss/crossentropy": 1.9800407886505127,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20614364743232727,
+      "step": 11608
+    },
+    {
+      "epoch": 0.2322,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.018536122639973958,
+      "learning_rate": 0.0001,
+      "loss": 4.2385,
+      "loss/crossentropy": 1.9646947979927063,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21902073919773102,
+      "step": 11610
+    },
+    {
+      "epoch": 0.23224,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.018930816650390626,
+      "learning_rate": 0.0001,
+      "loss": 4.1564,
+      "loss/crossentropy": 2.0113691687583923,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21262314170598984,
+      "step": 11612
+    },
+    {
+      "epoch": 0.23228,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.0254791259765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3179,
+      "loss/crossentropy": 1.8359833359718323,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2125518098473549,
+      "step": 11614
+    },
+    {
+      "epoch": 0.23232,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.02577489217122396,
+      "learning_rate": 0.0001,
+      "loss": 4.0954,
+      "loss/crossentropy": 1.934333622455597,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2148296758532524,
+      "step": 11616
+    },
+    {
+      "epoch": 0.23236,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.026192220052083333,
+      "learning_rate": 0.0001,
+      "loss": 4.3556,
+      "loss/crossentropy": 1.9486380815505981,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.201569102704525,
+      "step": 11618
+    },
+    {
+      "epoch": 0.2324,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0251708984375,
+      "learning_rate": 0.0001,
+      "loss": 4.3761,
+      "loss/crossentropy": 1.9379103183746338,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2109408900141716,
+      "step": 11620
+    },
+    {
+      "epoch": 0.23244,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.02165705362955729,
+      "learning_rate": 0.0001,
+      "loss": 4.3957,
+      "loss/crossentropy": 2.1635884046554565,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23330900818109512,
+      "step": 11622
+    },
+    {
+      "epoch": 0.23248,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.02851130167643229,
+      "learning_rate": 0.0001,
+      "loss": 4.1035,
+      "loss/crossentropy": 2.3183244466781616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21685285866260529,
+      "step": 11624
+    },
+    {
+      "epoch": 0.23252,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.03022028605143229,
+      "learning_rate": 0.0001,
+      "loss": 3.9465,
+      "loss/crossentropy": 2.2932451367378235,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22147410362958908,
+      "step": 11626
+    },
+    {
+      "epoch": 0.23256,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.03144505818684896,
+      "learning_rate": 0.0001,
+      "loss": 4.0598,
+      "loss/crossentropy": 2.1384140253067017,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22141631692647934,
+      "step": 11628
+    },
+    {
+      "epoch": 0.2326,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0146728515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9313,
+      "loss/crossentropy": 1.7338963747024536,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18959754705429077,
+      "step": 11630
+    },
+    {
+      "epoch": 0.23264,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.014943186442057292,
+      "learning_rate": 0.0001,
+      "loss": 4.3293,
+      "loss/crossentropy": 2.186310887336731,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19735413044691086,
+      "step": 11632
+    },
+    {
+      "epoch": 0.23268,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.01092529296875,
+      "learning_rate": 0.0001,
+      "loss": 3.9636,
+      "loss/crossentropy": 2.0566734075546265,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21930715441703796,
+      "step": 11634
+    },
+    {
+      "epoch": 0.23272,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.011631011962890625,
+      "learning_rate": 0.0001,
+      "loss": 3.8382,
+      "loss/crossentropy": 1.9993655681610107,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18999667465686798,
+      "step": 11636
+    },
+    {
+      "epoch": 0.23276,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008316802978515624,
+      "learning_rate": 0.0001,
+      "loss": 4.2309,
+      "loss/crossentropy": 2.2788418531417847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21202587336301804,
+      "step": 11638
+    },
+    {
+      "epoch": 0.2328,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.00792236328125,
+      "learning_rate": 0.0001,
+      "loss": 3.7837,
+      "loss/crossentropy": 1.7248046398162842,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21424376964569092,
+      "step": 11640
+    },
+    {
+      "epoch": 0.23284,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.008235422770182292,
+      "learning_rate": 0.0001,
+      "loss": 4.0023,
+      "loss/crossentropy": 1.9053270816802979,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19046200066804886,
+      "step": 11642
+    },
+    {
+      "epoch": 0.23288,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0118408203125,
+      "learning_rate": 0.0001,
+      "loss": 4.3561,
+      "loss/crossentropy": 2.3588117361068726,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23569310456514359,
+      "step": 11644
+    },
+    {
+      "epoch": 0.23292,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.012271881103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4722,
+      "loss/crossentropy": 2.15751576423645,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22107885777950287,
+      "step": 11646
+    },
+    {
+      "epoch": 0.23296,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009720611572265624,
+      "learning_rate": 0.0001,
+      "loss": 4.4075,
+      "loss/crossentropy": 2.094591200351715,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20681703090667725,
+      "step": 11648
+    },
+    {
+      "epoch": 0.233,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009523264567057292,
+      "learning_rate": 0.0001,
+      "loss": 3.7832,
+      "loss/crossentropy": 1.7536470890045166,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18886109441518784,
+      "step": 11650
+    },
+    {
+      "epoch": 0.23304,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0082427978515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9905,
+      "loss/crossentropy": 1.8172362446784973,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18359588831663132,
+      "step": 11652
+    },
+    {
+      "epoch": 0.23308,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.00841064453125,
+      "learning_rate": 0.0001,
+      "loss": 4.2404,
+      "loss/crossentropy": 2.2346678376197815,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2154158428311348,
+      "step": 11654
+    },
+    {
+      "epoch": 0.23312,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008040110270182291,
+      "learning_rate": 0.0001,
+      "loss": 4.1559,
+      "loss/crossentropy": 2.0573307275772095,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2160855233669281,
+      "step": 11656
+    },
+    {
+      "epoch": 0.23316,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.007621256510416666,
+      "learning_rate": 0.0001,
+      "loss": 4.0669,
+      "loss/crossentropy": 2.4358904361724854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22637036442756653,
+      "step": 11658
+    },
+    {
+      "epoch": 0.2332,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.004644521077473958,
+      "learning_rate": 0.0001,
+      "loss": 3.9938,
+      "loss/crossentropy": 2.387966513633728,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2385111078619957,
+      "step": 11660
+    },
+    {
+      "epoch": 0.23324,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.004709625244140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3371,
+      "loss/crossentropy": 2.30223548412323,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2149546965956688,
+      "step": 11662
+    },
+    {
+      "epoch": 0.23328,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.005494944254557292,
+      "learning_rate": 0.0001,
+      "loss": 4.3242,
+      "loss/crossentropy": 2.001839280128479,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2001148834824562,
+      "step": 11664
+    },
+    {
+      "epoch": 0.23332,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.00693359375,
+      "learning_rate": 0.0001,
+      "loss": 4.4932,
+      "loss/crossentropy": 2.387674927711487,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22714952379465103,
+      "step": 11666
+    },
+    {
+      "epoch": 0.23336,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.007470448811848958,
+      "learning_rate": 0.0001,
+      "loss": 4.1333,
+      "loss/crossentropy": 2.133235454559326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22263716161251068,
+      "step": 11668
+    },
+    {
+      "epoch": 0.2334,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.008874257405598959,
+      "learning_rate": 0.0001,
+      "loss": 4.1392,
+      "loss/crossentropy": 2.1041141748428345,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20474642515182495,
+      "step": 11670
+    },
+    {
+      "epoch": 0.23344,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.010796864827473959,
+      "learning_rate": 0.0001,
+      "loss": 4.4075,
+      "loss/crossentropy": 2.0425861477851868,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23144961893558502,
+      "step": 11672
+    },
+    {
+      "epoch": 0.23348,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.009883626302083334,
+      "learning_rate": 0.0001,
+      "loss": 4.0878,
+      "loss/crossentropy": 2.0170212388038635,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2134404480457306,
+      "step": 11674
+    },
+    {
+      "epoch": 0.23352,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010489908854166667,
+      "learning_rate": 0.0001,
+      "loss": 4.2275,
+      "loss/crossentropy": 2.205981433391571,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.220575213432312,
+      "step": 11676
+    },
+    {
+      "epoch": 0.23356,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.011429595947265624,
+      "learning_rate": 0.0001,
+      "loss": 4.1845,
+      "loss/crossentropy": 1.9121403694152832,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23312295228242874,
+      "step": 11678
+    },
+    {
+      "epoch": 0.2336,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.0156890869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1112,
+      "loss/crossentropy": 1.6793898940086365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19598360359668732,
+      "step": 11680
+    },
+    {
+      "epoch": 0.23364,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.0222564697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4814,
+      "loss/crossentropy": 2.1657907962799072,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2281472533941269,
+      "step": 11682
+    },
+    {
+      "epoch": 0.23368,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.022027333577473957,
+      "learning_rate": 0.0001,
+      "loss": 4.0869,
+      "loss/crossentropy": 2.1918715238571167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22837525606155396,
+      "step": 11684
+    },
+    {
+      "epoch": 0.23372,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.020699055989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.27,
+      "loss/crossentropy": 2.296495795249939,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23491691797971725,
+      "step": 11686
+    },
+    {
+      "epoch": 0.23376,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.019559733072916665,
+      "learning_rate": 0.0001,
+      "loss": 4.1422,
+      "loss/crossentropy": 1.9278987646102905,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1960090771317482,
+      "step": 11688
+    },
+    {
+      "epoch": 0.2338,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.019461822509765626,
+      "learning_rate": 0.0001,
+      "loss": 4.3393,
+      "loss/crossentropy": 2.0491825938224792,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21345915645360947,
+      "step": 11690
+    },
+    {
+      "epoch": 0.23384,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.02127863566080729,
+      "learning_rate": 0.0001,
+      "loss": 3.9209,
+      "loss/crossentropy": 1.7594041228294373,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19004464149475098,
+      "step": 11692
+    },
+    {
+      "epoch": 0.23388,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.020961252848307292,
+      "learning_rate": 0.0001,
+      "loss": 4.3645,
+      "loss/crossentropy": 2.0313411951065063,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20986144989728928,
+      "step": 11694
+    },
+    {
+      "epoch": 0.23392,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.016068522135416666,
+      "learning_rate": 0.0001,
+      "loss": 4.4644,
+      "loss/crossentropy": 2.3580493927001953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2253532111644745,
+      "step": 11696
+    },
+    {
+      "epoch": 0.23396,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.008548990885416666,
+      "learning_rate": 0.0001,
+      "loss": 4.3954,
+      "loss/crossentropy": 2.2036253213882446,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22294757515192032,
+      "step": 11698
+    },
+    {
+      "epoch": 0.234,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.007972971598307291,
+      "learning_rate": 0.0001,
+      "loss": 4.2602,
+      "loss/crossentropy": 1.8575093150138855,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19865535199642181,
+      "step": 11700
+    },
+    {
+      "epoch": 0.23404,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.008571116129557292,
+      "learning_rate": 0.0001,
+      "loss": 4.2619,
+      "loss/crossentropy": 2.2169028520584106,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22326484322547913,
+      "step": 11702
+    },
+    {
+      "epoch": 0.23408,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007500966389973958,
+      "learning_rate": 0.0001,
+      "loss": 4.4045,
+      "loss/crossentropy": 2.1897542476654053,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2097000628709793,
+      "step": 11704
+    },
+    {
+      "epoch": 0.23412,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0063168843587239586,
+      "learning_rate": 0.0001,
+      "loss": 4.271,
+      "loss/crossentropy": 2.214607834815979,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24112706631422043,
+      "step": 11706
+    },
+    {
+      "epoch": 0.23416,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.004780832926432292,
+      "learning_rate": 0.0001,
+      "loss": 4.1278,
+      "loss/crossentropy": 2.124355912208557,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2222435548901558,
+      "step": 11708
+    },
+    {
+      "epoch": 0.2342,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0051025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3118,
+      "loss/crossentropy": 2.35421621799469,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26120802760124207,
+      "step": 11710
+    },
+    {
+      "epoch": 0.23424,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.00521240234375,
+      "learning_rate": 0.0001,
+      "loss": 4.1781,
+      "loss/crossentropy": 1.9134620428085327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21830002963542938,
+      "step": 11712
+    },
+    {
+      "epoch": 0.23428,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.003763580322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2445,
+      "loss/crossentropy": 1.9336887001991272,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2025599479675293,
+      "step": 11714
+    },
+    {
+      "epoch": 0.23432,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.004133097330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0775,
+      "loss/crossentropy": 1.9964489936828613,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2020488828420639,
+      "step": 11716
+    },
+    {
+      "epoch": 0.23436,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.00423583984375,
+      "learning_rate": 0.0001,
+      "loss": 4.264,
+      "loss/crossentropy": 2.108368992805481,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20985107123851776,
+      "step": 11718
+    },
+    {
+      "epoch": 0.2344,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.005008697509765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9458,
+      "loss/crossentropy": 2.0314669013023376,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20942936092615128,
+      "step": 11720
+    },
+    {
+      "epoch": 0.23444,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0054107666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1098,
+      "loss/crossentropy": 2.343130350112915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.237278014421463,
+      "step": 11722
+    },
+    {
+      "epoch": 0.23448,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.010658518473307291,
+      "learning_rate": 0.0001,
+      "loss": 4.2497,
+      "loss/crossentropy": 2.1542173624038696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22762110829353333,
+      "step": 11724
+    },
+    {
+      "epoch": 0.23452,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.009395090738932292,
+      "learning_rate": 0.0001,
+      "loss": 3.9056,
+      "loss/crossentropy": 1.638447105884552,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1707247570157051,
+      "step": 11726
+    },
+    {
+      "epoch": 0.23456,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.010573069254557291,
+      "learning_rate": 0.0001,
+      "loss": 4.0939,
+      "loss/crossentropy": 2.1509228944778442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2091827318072319,
+      "step": 11728
+    },
+    {
+      "epoch": 0.2346,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.011677805582682292,
+      "learning_rate": 0.0001,
+      "loss": 4.325,
+      "loss/crossentropy": 2.193585455417633,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2697590962052345,
+      "step": 11730
+    },
+    {
+      "epoch": 0.23464,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012648264567057291,
+      "learning_rate": 0.0001,
+      "loss": 4.2239,
+      "loss/crossentropy": 2.2058298587799072,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21958298981189728,
+      "step": 11732
+    },
+    {
+      "epoch": 0.23468,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.012894439697265624,
+      "learning_rate": 0.0001,
+      "loss": 4.3269,
+      "loss/crossentropy": 2.0816246271133423,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20916947722434998,
+      "step": 11734
+    },
+    {
+      "epoch": 0.23472,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0150543212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1495,
+      "loss/crossentropy": 1.838355541229248,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20002590864896774,
+      "step": 11736
+    },
+    {
+      "epoch": 0.23476,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.016795857747395834,
+      "learning_rate": 0.0001,
+      "loss": 4.0559,
+      "loss/crossentropy": 2.1602721214294434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21018436551094055,
+      "step": 11738
+    },
+    {
+      "epoch": 0.2348,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.013181304931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.5376,
+      "loss/crossentropy": 2.6967735290527344,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23233920335769653,
+      "step": 11740
+    },
+    {
+      "epoch": 0.23484,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.012715403238932292,
+      "learning_rate": 0.0001,
+      "loss": 3.9772,
+      "loss/crossentropy": 2.033313810825348,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21240226924419403,
+      "step": 11742
+    },
+    {
+      "epoch": 0.23488,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.012245432535807291,
+      "learning_rate": 0.0001,
+      "loss": 4.21,
+      "loss/crossentropy": 2.123607873916626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.207147017121315,
+      "step": 11744
+    },
+    {
+      "epoch": 0.23492,
+      "grad_norm": 1.8125,
+      "grad_norm_var": 0.013719685872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1121,
+      "loss/crossentropy": 2.1173813343048096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21387682110071182,
+      "step": 11746
+    },
+    {
+      "epoch": 0.23496,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.011456044514973958,
+      "learning_rate": 0.0001,
+      "loss": 4.1219,
+      "loss/crossentropy": 2.27209734916687,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2254154533147812,
+      "step": 11748
+    },
+    {
+      "epoch": 0.235,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.011156209309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.3138,
+      "loss/crossentropy": 1.844546616077423,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1740722581744194,
+      "step": 11750
+    },
+    {
+      "epoch": 0.23504,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009346516927083333,
+      "learning_rate": 0.0001,
+      "loss": 4.4615,
+      "loss/crossentropy": 2.341967821121216,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23638595640659332,
+      "step": 11752
+    },
+    {
+      "epoch": 0.23508,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.009354400634765624,
+      "learning_rate": 0.0001,
+      "loss": 4.4113,
+      "loss/crossentropy": 2.2561213970184326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23169096559286118,
+      "step": 11754
+    },
+    {
+      "epoch": 0.23512,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.007972971598307291,
+      "learning_rate": 0.0001,
+      "loss": 4.1603,
+      "loss/crossentropy": 1.8693158030509949,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19997497648000717,
+      "step": 11756
+    },
+    {
+      "epoch": 0.23516,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.0085845947265625,
+      "learning_rate": 0.0001,
+      "loss": 3.9721,
+      "loss/crossentropy": 2.2228434085845947,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19779722392559052,
+      "step": 11758
+    },
+    {
+      "epoch": 0.2352,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.007899729410807292,
+      "learning_rate": 0.0001,
+      "loss": 4.2572,
+      "loss/crossentropy": 2.177064299583435,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21091558039188385,
+      "step": 11760
+    },
+    {
+      "epoch": 0.23524,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.005020904541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.398,
+      "loss/crossentropy": 2.2667617797851562,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2120228409767151,
+      "step": 11762
+    },
+    {
+      "epoch": 0.23528,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.004713694254557292,
+      "learning_rate": 0.0001,
+      "loss": 4.198,
+      "loss/crossentropy": 2.0310307145118713,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2194480448961258,
+      "step": 11764
+    },
+    {
+      "epoch": 0.23532,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.005033111572265625,
+      "learning_rate": 0.0001,
+      "loss": 3.9906,
+      "loss/crossentropy": 1.8481100797653198,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20181456953287125,
+      "step": 11766
+    },
+    {
+      "epoch": 0.23536,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.008642323811848958,
+      "learning_rate": 0.0001,
+      "loss": 4.3486,
+      "loss/crossentropy": 2.254691958427429,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26685942709445953,
+      "step": 11768
+    },
+    {
+      "epoch": 0.2354,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.007993316650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.144,
+      "loss/crossentropy": 2.112728714942932,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19724663347005844,
+      "step": 11770
+    },
+    {
+      "epoch": 0.23544,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.011195627848307292,
+      "learning_rate": 0.0001,
+      "loss": 4.089,
+      "loss/crossentropy": 2.0724143981933594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20518633723258972,
+      "step": 11772
+    },
+    {
+      "epoch": 0.23548,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010064442952473959,
+      "learning_rate": 0.0001,
+      "loss": 4.1882,
+      "loss/crossentropy": 2.0743810534477234,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2181471511721611,
+      "step": 11774
+    },
+    {
+      "epoch": 0.23552,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.011740875244140626,
+      "learning_rate": 0.0001,
+      "loss": 4.1121,
+      "loss/crossentropy": 2.02871835231781,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22081361711025238,
+      "step": 11776
+    },
+    {
+      "epoch": 0.23556,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.013185373942057292,
+      "learning_rate": 0.0001,
+      "loss": 4.5607,
+      "loss/crossentropy": 2.1898428201675415,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22084421664476395,
+      "step": 11778
+    },
+    {
+      "epoch": 0.2356,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.012748209635416667,
+      "learning_rate": 0.0001,
+      "loss": 4.2404,
+      "loss/crossentropy": 2.0893847346305847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21616562455892563,
+      "step": 11780
+    },
+    {
+      "epoch": 0.23564,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.011769358317057292,
+      "learning_rate": 0.0001,
+      "loss": 4.215,
+      "loss/crossentropy": 1.8590916991233826,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18936381489038467,
+      "step": 11782
+    },
+    {
+      "epoch": 0.23568,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.012741851806640624,
+      "learning_rate": 0.0001,
+      "loss": 3.9544,
+      "loss/crossentropy": 2.105339765548706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22275932878255844,
+      "step": 11784
+    },
+    {
+      "epoch": 0.23572,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.011987050374348959,
+      "learning_rate": 0.0001,
+      "loss": 4.2585,
+      "loss/crossentropy": 2.1720080375671387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20734921097755432,
+      "step": 11786
+    },
+    {
+      "epoch": 0.23576,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.012640126546223958,
+      "learning_rate": 0.0001,
+      "loss": 4.2715,
+      "loss/crossentropy": 2.2301958799362183,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21865415573120117,
+      "step": 11788
+    },
+    {
+      "epoch": 0.2358,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.013626861572265624,
+      "learning_rate": 0.0001,
+      "loss": 3.7518,
+      "loss/crossentropy": 1.588155210018158,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19334131479263306,
+      "step": 11790
+    },
+    {
+      "epoch": 0.23584,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.025248209635416668,
+      "learning_rate": 0.0001,
+      "loss": 4.4052,
+      "loss/crossentropy": 2.208239734172821,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21055688709020615,
+      "step": 11792
+    },
+    {
+      "epoch": 0.23588,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.030368804931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1996,
+      "loss/crossentropy": 2.46909761428833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.225817009806633,
+      "step": 11794
+    },
+    {
+      "epoch": 0.23592,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.03367691040039063,
+      "learning_rate": 0.0001,
+      "loss": 4.2792,
+      "loss/crossentropy": 2.022938549518585,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21232923865318298,
+      "step": 11796
+    },
+    {
+      "epoch": 0.23596,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.03394953409830729,
+      "learning_rate": 0.0001,
+      "loss": 4.3949,
+      "loss/crossentropy": 2.333058714866638,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23842789232730865,
+      "step": 11798
+    },
+    {
+      "epoch": 0.236,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0366363525390625,
+      "learning_rate": 0.0001,
+      "loss": 3.8698,
+      "loss/crossentropy": 1.9570570588111877,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20660093426704407,
+      "step": 11800
+    },
+    {
+      "epoch": 0.23604,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0335601806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3649,
+      "loss/crossentropy": 2.2074697017669678,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23766764998435974,
+      "step": 11802
+    },
+    {
+      "epoch": 0.23608,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.03026301066080729,
+      "learning_rate": 0.0001,
+      "loss": 4.4127,
+      "loss/crossentropy": 2.327863335609436,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22967173904180527,
+      "step": 11804
+    },
+    {
+      "epoch": 0.23612,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.02802734375,
+      "learning_rate": 0.0001,
+      "loss": 4.3367,
+      "loss/crossentropy": 2.135041356086731,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21708428114652634,
+      "step": 11806
+    },
+    {
+      "epoch": 0.23616,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.00665283203125,
+      "learning_rate": 0.0001,
+      "loss": 3.9805,
+      "loss/crossentropy": 1.9111011624336243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20122328400611877,
+      "step": 11808
+    },
+    {
+      "epoch": 0.2362,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.006436920166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.19,
+      "loss/crossentropy": 2.0823878049850464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23593349009752274,
+      "step": 11810
+    },
+    {
+      "epoch": 0.23624,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0069000244140625,
+      "learning_rate": 0.0001,
+      "loss": 3.9554,
+      "loss/crossentropy": 2.298948645591736,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22076356410980225,
+      "step": 11812
+    },
+    {
+      "epoch": 0.23628,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.006483713785807292,
+      "learning_rate": 0.0001,
+      "loss": 4.1119,
+      "loss/crossentropy": 2.383033037185669,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20850034803152084,
+      "step": 11814
+    },
+    {
+      "epoch": 0.23632,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.10857645670572917,
+      "learning_rate": 0.0001,
+      "loss": 4.3648,
+      "loss/crossentropy": 2.004193425178528,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27853623032569885,
+      "step": 11816
+    },
+    {
+      "epoch": 0.23636,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.11877339680989583,
+      "learning_rate": 0.0001,
+      "loss": 4.4454,
+      "loss/crossentropy": 1.9110660552978516,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21773407608270645,
+      "step": 11818
+    },
+    {
+      "epoch": 0.2364,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.11894505818684896,
+      "learning_rate": 0.0001,
+      "loss": 4.5428,
+      "loss/crossentropy": 2.300672471523285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23710736632347107,
+      "step": 11820
+    },
+    {
+      "epoch": 0.23644,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.11943333943684896,
+      "learning_rate": 0.0001,
+      "loss": 4.1413,
+      "loss/crossentropy": 2.270769238471985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2182588130235672,
+      "step": 11822
+    },
+    {
+      "epoch": 0.23648,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.11885960896809895,
+      "learning_rate": 0.0001,
+      "loss": 4.2744,
+      "loss/crossentropy": 2.085016667842865,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19494660943746567,
+      "step": 11824
+    },
+    {
+      "epoch": 0.23652,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.11644261678059896,
+      "learning_rate": 0.0001,
+      "loss": 4.5554,
+      "loss/crossentropy": 2.384607672691345,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22075054794549942,
+      "step": 11826
+    },
+    {
+      "epoch": 0.23656,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.11926167805989583,
+      "learning_rate": 0.0001,
+      "loss": 3.9076,
+      "loss/crossentropy": 2.1217936277389526,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2126227766275406,
+      "step": 11828
+    },
+    {
+      "epoch": 0.2366,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.12241185506184896,
+      "learning_rate": 0.0001,
+      "loss": 3.8285,
+      "loss/crossentropy": 2.1690168380737305,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20450890809297562,
+      "step": 11830
+    },
+    {
+      "epoch": 0.23664,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.031998697916666666,
+      "learning_rate": 0.0001,
+      "loss": 4.2817,
+      "loss/crossentropy": 2.2614429593086243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22817185521125793,
+      "step": 11832
+    },
+    {
+      "epoch": 0.23668,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.023545074462890624,
+      "learning_rate": 0.0001,
+      "loss": 4.3074,
+      "loss/crossentropy": 2.177332043647766,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.226592555642128,
+      "step": 11834
+    },
+    {
+      "epoch": 0.23672,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.01765925089518229,
+      "learning_rate": 0.0001,
+      "loss": 4.0471,
+      "loss/crossentropy": 1.947661578655243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1762884557247162,
+      "step": 11836
+    },
+    {
+      "epoch": 0.23676,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.018184153238932292,
+      "learning_rate": 0.0001,
+      "loss": 3.9014,
+      "loss/crossentropy": 1.7112661004066467,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18696465343236923,
+      "step": 11838
+    },
+    {
+      "epoch": 0.2368,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.017439524332682293,
+      "learning_rate": 0.0001,
+      "loss": 4.1761,
+      "loss/crossentropy": 2.0478790402412415,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21436551213264465,
+      "step": 11840
+    },
+    {
+      "epoch": 0.23684,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.01718724568684896,
+      "learning_rate": 0.0001,
+      "loss": 4.1702,
+      "loss/crossentropy": 2.0541738867759705,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23121927678585052,
+      "step": 11842
+    },
+    {
+      "epoch": 0.23688,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.01876805623372396,
+      "learning_rate": 0.0001,
+      "loss": 4.208,
+      "loss/crossentropy": 2.04353004693985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22237974405288696,
+      "step": 11844
+    },
+    {
+      "epoch": 0.23692,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.017488606770833335,
+      "learning_rate": 0.0001,
+      "loss": 4.4641,
+      "loss/crossentropy": 2.110148549079895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23324476927518845,
+      "step": 11846
+    },
+    {
+      "epoch": 0.23696,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.018623860677083333,
+      "learning_rate": 0.0001,
+      "loss": 4.0831,
+      "loss/crossentropy": 2.207589864730835,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.209608756005764,
+      "step": 11848
+    },
+    {
+      "epoch": 0.237,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.012180328369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1896,
+      "loss/crossentropy": 2.2447429895401,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2016080766916275,
+      "step": 11850
+    },
+    {
+      "epoch": 0.23704,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009771474202473958,
+      "learning_rate": 0.0001,
+      "loss": 4.3133,
+      "loss/crossentropy": 2.3349474668502808,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22105249762535095,
+      "step": 11852
+    },
+    {
+      "epoch": 0.23708,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.008024088541666667,
+      "learning_rate": 0.0001,
+      "loss": 4.1091,
+      "loss/crossentropy": 1.8444748520851135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1993526816368103,
+      "step": 11854
+    },
+    {
+      "epoch": 0.23712,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009323883056640624,
+      "learning_rate": 0.0001,
+      "loss": 4.0758,
+      "loss/crossentropy": 1.9181615710258484,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18698236346244812,
+      "step": 11856
+    },
+    {
+      "epoch": 0.23716,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.009627024332682291,
+      "learning_rate": 0.0001,
+      "loss": 4.1454,
+      "loss/crossentropy": 2.1930960416793823,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2122444212436676,
+      "step": 11858
+    },
+    {
+      "epoch": 0.2372,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.009439849853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3177,
+      "loss/crossentropy": 1.7835432887077332,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20900961011648178,
+      "step": 11860
+    },
+    {
+      "epoch": 0.23724,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.011775461832682292,
+      "learning_rate": 0.0001,
+      "loss": 4.6362,
+      "loss/crossentropy": 2.1839439868927,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21574077755212784,
+      "step": 11862
+    },
+    {
+      "epoch": 0.23728,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.010791015625,
+      "learning_rate": 0.0001,
+      "loss": 3.8916,
+      "loss/crossentropy": 1.9817007184028625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21773921698331833,
+      "step": 11864
+    },
+    {
+      "epoch": 0.23732,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.008923085530598958,
+      "learning_rate": 0.0001,
+      "loss": 4.2065,
+      "loss/crossentropy": 2.017501652240753,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2247873619198799,
+      "step": 11866
+    },
+    {
+      "epoch": 0.23736,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.010992177327473958,
+      "learning_rate": 0.0001,
+      "loss": 3.8813,
+      "loss/crossentropy": 2.0779114961624146,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21136770397424698,
+      "step": 11868
+    },
+    {
+      "epoch": 0.2374,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.010497029622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.341,
+      "loss/crossentropy": 2.3987231254577637,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2326417714357376,
+      "step": 11870
+    },
+    {
+      "epoch": 0.23744,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.008975982666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0466,
+      "loss/crossentropy": 1.7145346999168396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19222228974103928,
+      "step": 11872
+    },
+    {
+      "epoch": 0.23748,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008829498291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2683,
+      "loss/crossentropy": 2.378560423851013,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21374844759702682,
+      "step": 11874
+    },
+    {
+      "epoch": 0.23752,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0073626200358072914,
+      "learning_rate": 0.0001,
+      "loss": 4.2548,
+      "loss/crossentropy": 1.951455295085907,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2217455804347992,
+      "step": 11876
+    },
+    {
+      "epoch": 0.23756,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.00513916015625,
+      "learning_rate": 0.0001,
+      "loss": 3.8867,
+      "loss/crossentropy": 1.9309074878692627,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2100483626127243,
+      "step": 11878
+    },
+    {
+      "epoch": 0.2376,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.00445556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.102,
+      "loss/crossentropy": 2.2295292615890503,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2196313813328743,
+      "step": 11880
+    },
+    {
+      "epoch": 0.23764,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.004964192708333333,
+      "learning_rate": 0.0001,
+      "loss": 4.0514,
+      "loss/crossentropy": 2.0665449500083923,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20752248913049698,
+      "step": 11882
+    },
+    {
+      "epoch": 0.23768,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.00560302734375,
+      "learning_rate": 0.0001,
+      "loss": 4.4022,
+      "loss/crossentropy": 1.9645958542823792,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20391641557216644,
+      "step": 11884
+    },
+    {
+      "epoch": 0.23772,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.005956013997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1698,
+      "loss/crossentropy": 2.1280174255371094,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21947231888771057,
+      "step": 11886
+    },
+    {
+      "epoch": 0.23776,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0073964436848958336,
+      "learning_rate": 0.0001,
+      "loss": 4.2043,
+      "loss/crossentropy": 1.9462851285934448,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1961178332567215,
+      "step": 11888
+    },
+    {
+      "epoch": 0.2378,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.009422810872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.4846,
+      "loss/crossentropy": 2.249086618423462,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22473402321338654,
+      "step": 11890
+    },
+    {
+      "epoch": 0.23784,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.010472615559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1368,
+      "loss/crossentropy": 2.1972378492355347,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2164871245622635,
+      "step": 11892
+    },
+    {
+      "epoch": 0.23788,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.009409332275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.228,
+      "loss/crossentropy": 2.0067209601402283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20195162296295166,
+      "step": 11894
+    },
+    {
+      "epoch": 0.23792,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009673817952473959,
+      "learning_rate": 0.0001,
+      "loss": 4.1767,
+      "loss/crossentropy": 1.9103696942329407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19616805016994476,
+      "step": 11896
+    },
+    {
+      "epoch": 0.23796,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.011595662434895833,
+      "learning_rate": 0.0001,
+      "loss": 4.228,
+      "loss/crossentropy": 2.1524049639701843,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21107815951108932,
+      "step": 11898
+    },
+    {
+      "epoch": 0.238,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0110260009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.183,
+      "loss/crossentropy": 2.2112287878990173,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2099093720316887,
+      "step": 11900
+    },
+    {
+      "epoch": 0.23804,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.8914347330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.647,
+      "loss/crossentropy": 2.2945204973220825,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2713918164372444,
+      "step": 11902
+    },
+    {
+      "epoch": 0.23808,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.8781575520833333,
+      "learning_rate": 0.0001,
+      "loss": 4.4699,
+      "loss/crossentropy": 2.1838968992233276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22640438377857208,
+      "step": 11904
+    },
+    {
+      "epoch": 0.23812,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.89010009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0843,
+      "loss/crossentropy": 1.9089699983596802,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19553899765014648,
+      "step": 11906
+    },
+    {
+      "epoch": 0.23816,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.8873443603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2573,
+      "loss/crossentropy": 2.105385661125183,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21528839319944382,
+      "step": 11908
+    },
+    {
+      "epoch": 0.2382,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.8921946207682292,
+      "learning_rate": 0.0001,
+      "loss": 4.0381,
+      "loss/crossentropy": 1.565223515033722,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17765602469444275,
+      "step": 11910
+    },
+    {
+      "epoch": 0.23824,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.8919016520182291,
+      "learning_rate": 0.0001,
+      "loss": 4.3014,
+      "loss/crossentropy": 2.220748543739319,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21904101967811584,
+      "step": 11912
+    },
+    {
+      "epoch": 0.23828,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.889306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2965,
+      "loss/crossentropy": 2.294031500816345,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21700909733772278,
+      "step": 11914
+    },
+    {
+      "epoch": 0.23832,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.8795237223307292,
+      "learning_rate": 0.0001,
+      "loss": 4.3439,
+      "loss/crossentropy": 2.072917103767395,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20462248474359512,
+      "step": 11916
+    },
+    {
+      "epoch": 0.23836,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.00533447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1843,
+      "loss/crossentropy": 2.0541720390319824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2127074897289276,
+      "step": 11918
+    },
+    {
+      "epoch": 0.2384,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0046539306640625,
+      "learning_rate": 0.0001,
+      "loss": 3.8922,
+      "loss/crossentropy": 1.9873629808425903,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19215020537376404,
+      "step": 11920
+    },
+    {
+      "epoch": 0.23844,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0032135009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1943,
+      "loss/crossentropy": 2.447067141532898,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2318919375538826,
+      "step": 11922
+    },
+    {
+      "epoch": 0.23848,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0038157145182291666,
+      "learning_rate": 0.0001,
+      "loss": 3.8756,
+      "loss/crossentropy": 1.9918025732040405,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20865648984909058,
+      "step": 11924
+    },
+    {
+      "epoch": 0.23852,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0026041666666666665,
+      "learning_rate": 0.0001,
+      "loss": 4.3243,
+      "loss/crossentropy": 2.0803070068359375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21753622591495514,
+      "step": 11926
+    },
+    {
+      "epoch": 0.23856,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.002976226806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1828,
+      "loss/crossentropy": 2.118411421775818,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21741919964551926,
+      "step": 11928
+    },
+    {
+      "epoch": 0.2386,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0034075419108072916,
+      "learning_rate": 0.0001,
+      "loss": 4.0056,
+      "loss/crossentropy": 1.8924900889396667,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20609601587057114,
+      "step": 11930
+    },
+    {
+      "epoch": 0.23864,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.002418772379557292,
+      "learning_rate": 0.0001,
+      "loss": 4.0012,
+      "loss/crossentropy": 1.743731439113617,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19873473048210144,
+      "step": 11932
+    },
+    {
+      "epoch": 0.23868,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.002929433186848958,
+      "learning_rate": 0.0001,
+      "loss": 4.2564,
+      "loss/crossentropy": 2.291381061077118,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21248165518045425,
+      "step": 11934
+    },
+    {
+      "epoch": 0.23872,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.007452138264973958,
+      "learning_rate": 0.0001,
+      "loss": 4.3938,
+      "loss/crossentropy": 1.7672501802444458,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19201595336198807,
+      "step": 11936
+    },
+    {
+      "epoch": 0.23876,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.008162434895833333,
+      "learning_rate": 0.0001,
+      "loss": 3.8693,
+      "loss/crossentropy": 1.8719280362129211,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1960783526301384,
+      "step": 11938
+    },
+    {
+      "epoch": 0.2388,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.010978190104166667,
+      "learning_rate": 0.0001,
+      "loss": 4.3345,
+      "loss/crossentropy": 1.8103876113891602,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1964937001466751,
+      "step": 11940
+    },
+    {
+      "epoch": 0.23884,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011901601155598959,
+      "learning_rate": 0.0001,
+      "loss": 4.2134,
+      "loss/crossentropy": 1.8743855953216553,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19404225796461105,
+      "step": 11942
+    },
+    {
+      "epoch": 0.23888,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.012562815348307292,
+      "learning_rate": 0.0001,
+      "loss": 4.2379,
+      "loss/crossentropy": 2.3096803426742554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23256508260965347,
+      "step": 11944
+    },
+    {
+      "epoch": 0.23892,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012123362223307291,
+      "learning_rate": 0.0001,
+      "loss": 4.5089,
+      "loss/crossentropy": 2.207027554512024,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20778407156467438,
+      "step": 11946
+    },
+    {
+      "epoch": 0.23896,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.011671702067057291,
+      "learning_rate": 0.0001,
+      "loss": 4.0423,
+      "loss/crossentropy": 1.845999002456665,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19381015002727509,
+      "step": 11948
+    },
+    {
+      "epoch": 0.239,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011628977457682292,
+      "learning_rate": 0.0001,
+      "loss": 4.1054,
+      "loss/crossentropy": 1.677983045578003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17493421584367752,
+      "step": 11950
+    },
+    {
+      "epoch": 0.23904,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.008107248942057292,
+      "learning_rate": 0.0001,
+      "loss": 4.1729,
+      "loss/crossentropy": 2.1798466444015503,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22395183145999908,
+      "step": 11952
+    },
+    {
+      "epoch": 0.23908,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.013525390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0162,
+      "loss/crossentropy": 2.008498191833496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21069375425577164,
+      "step": 11954
+    },
+    {
+      "epoch": 0.23912,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.014546712239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.3736,
+      "loss/crossentropy": 2.1712071895599365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21842175722122192,
+      "step": 11956
+    },
+    {
+      "epoch": 0.23916,
+      "grad_norm": 2.578125,
+      "grad_norm_var": 0.03242365519205729,
+      "learning_rate": 0.0001,
+      "loss": 4.1338,
+      "loss/crossentropy": 2.2769562005996704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22064895182847977,
+      "step": 11958
+    },
+    {
+      "epoch": 0.2392,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.032572428385416664,
+      "learning_rate": 0.0001,
+      "loss": 4.3401,
+      "loss/crossentropy": 2.127632260322571,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2139139398932457,
+      "step": 11960
+    },
+    {
+      "epoch": 0.23924,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.03223368326822917,
+      "learning_rate": 0.0001,
+      "loss": 4.3547,
+      "loss/crossentropy": 2.2687970399856567,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23024404793977737,
+      "step": 11962
+    },
+    {
+      "epoch": 0.23928,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.032136027018229166,
+      "learning_rate": 0.0001,
+      "loss": 4.3532,
+      "loss/crossentropy": 2.2271196246147156,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22528529912233353,
+      "step": 11964
+    },
+    {
+      "epoch": 0.23932,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.032136027018229166,
+      "learning_rate": 0.0001,
+      "loss": 4.1898,
+      "loss/crossentropy": 2.3209941387176514,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21150042116641998,
+      "step": 11966
+    },
+    {
+      "epoch": 0.23936,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.03208719889322917,
+      "learning_rate": 0.0001,
+      "loss": 4.197,
+      "loss/crossentropy": 1.9936136603355408,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22582116723060608,
+      "step": 11968
+    },
+    {
+      "epoch": 0.2394,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.024055989583333333,
+      "learning_rate": 0.0001,
+      "loss": 3.9045,
+      "loss/crossentropy": 1.7509311437606812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19028881192207336,
+      "step": 11970
+    },
+    {
+      "epoch": 0.23944,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.02213134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2365,
+      "loss/crossentropy": 2.2120620012283325,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23357221484184265,
+      "step": 11972
+    },
+    {
+      "epoch": 0.23948,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.005686187744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0129,
+      "loss/crossentropy": 1.794329285621643,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17914240062236786,
+      "step": 11974
+    },
+    {
+      "epoch": 0.23952,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.010550689697265626,
+      "learning_rate": 0.0001,
+      "loss": 4.1347,
+      "loss/crossentropy": 2.1647136211395264,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21009670197963715,
+      "step": 11976
+    },
+    {
+      "epoch": 0.23956,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.010660552978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2365,
+      "loss/crossentropy": 1.674091637134552,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19617585837841034,
+      "step": 11978
+    },
+    {
+      "epoch": 0.2396,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.011739095052083334,
+      "learning_rate": 0.0001,
+      "loss": 4.2474,
+      "loss/crossentropy": 1.9591755867004395,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18309858441352844,
+      "step": 11980
+    },
+    {
+      "epoch": 0.23964,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.017862955729166668,
+      "learning_rate": 0.0001,
+      "loss": 4.3642,
+      "loss/crossentropy": 2.0550093054771423,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22483228147029877,
+      "step": 11982
+    },
+    {
+      "epoch": 0.23968,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.019291178385416666,
+      "learning_rate": 0.0001,
+      "loss": 4.3937,
+      "loss/crossentropy": 2.264032781124115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2101321816444397,
+      "step": 11984
+    },
+    {
+      "epoch": 0.23972,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.020213826497395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4341,
+      "loss/crossentropy": 2.06082820892334,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2090815082192421,
+      "step": 11986
+    },
+    {
+      "epoch": 0.23976,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0197662353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4159,
+      "loss/crossentropy": 2.248009443283081,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22642739117145538,
+      "step": 11988
+    },
+    {
+      "epoch": 0.2398,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.02211278279622396,
+      "learning_rate": 0.0001,
+      "loss": 3.9601,
+      "loss/crossentropy": 1.797426462173462,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18600185215473175,
+      "step": 11990
+    },
+    {
+      "epoch": 0.23984,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.01573460896809896,
+      "learning_rate": 0.0001,
+      "loss": 4.1666,
+      "loss/crossentropy": 2.5529314279556274,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23957456648349762,
+      "step": 11992
+    },
+    {
+      "epoch": 0.23988,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.015933990478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2971,
+      "loss/crossentropy": 1.9974916577339172,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23130090534687042,
+      "step": 11994
+    },
+    {
+      "epoch": 0.23992,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.014679972330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0358,
+      "loss/crossentropy": 1.9714577794075012,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1969093456864357,
+      "step": 11996
+    },
+    {
+      "epoch": 0.23996,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.011055501302083333,
+      "learning_rate": 0.0001,
+      "loss": 4.282,
+      "loss/crossentropy": 1.8047285079956055,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19191773235797882,
+      "step": 11998
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.009187825520833333,
+      "learning_rate": 0.0001,
+      "loss": 4.4078,
+      "loss/crossentropy": 1.9240365028381348,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19021467864513397,
+      "step": 12000
+    },
+    {
+      "epoch": 0.24004,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.009041086832682291,
+      "learning_rate": 0.0001,
+      "loss": 4.1657,
+      "loss/crossentropy": 1.6503748297691345,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18838192522525787,
+      "step": 12002
+    },
+    {
+      "epoch": 0.24008,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009368642171223959,
+      "learning_rate": 0.0001,
+      "loss": 4.3451,
+      "loss/crossentropy": 2.238506555557251,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2139800265431404,
+      "step": 12004
+    },
+    {
+      "epoch": 0.24012,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.008695220947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0978,
+      "loss/crossentropy": 1.8623422384262085,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19039735198020935,
+      "step": 12006
+    },
+    {
+      "epoch": 0.24016,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0096588134765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9741,
+      "loss/crossentropy": 1.975899577140808,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20808710902929306,
+      "step": 12008
+    },
+    {
+      "epoch": 0.2402,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0108642578125,
+      "learning_rate": 0.0001,
+      "loss": 4.4332,
+      "loss/crossentropy": 2.2959831953048706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23725779354572296,
+      "step": 12010
+    },
+    {
+      "epoch": 0.24024,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.015130360921223959,
+      "learning_rate": 0.0001,
+      "loss": 3.9117,
+      "loss/crossentropy": 2.1626380681991577,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2068287953734398,
+      "step": 12012
+    },
+    {
+      "epoch": 0.24028,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.013474273681640624,
+      "learning_rate": 0.0001,
+      "loss": 4.1614,
+      "loss/crossentropy": 1.925924837589264,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2097126841545105,
+      "step": 12014
+    },
+    {
+      "epoch": 0.24032,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.018302154541015626,
+      "learning_rate": 0.0001,
+      "loss": 4.5545,
+      "loss/crossentropy": 2.147459626197815,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2345646619796753,
+      "step": 12016
+    },
+    {
+      "epoch": 0.24036,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.021418253580729168,
+      "learning_rate": 0.0001,
+      "loss": 4.0219,
+      "loss/crossentropy": 2.052124857902527,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2058626338839531,
+      "step": 12018
+    },
+    {
+      "epoch": 0.2404,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.021329752604166665,
+      "learning_rate": 0.0001,
+      "loss": 3.8619,
+      "loss/crossentropy": 2.377007842063904,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2124926745891571,
+      "step": 12020
+    },
+    {
+      "epoch": 0.24044,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.020173136393229166,
+      "learning_rate": 0.0001,
+      "loss": 3.9542,
+      "loss/crossentropy": 1.9117819666862488,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1976870447397232,
+      "step": 12022
+    },
+    {
+      "epoch": 0.24048,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.021478017171223957,
+      "learning_rate": 0.0001,
+      "loss": 4.5771,
+      "loss/crossentropy": 2.439339756965637,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22178436815738678,
+      "step": 12024
+    },
+    {
+      "epoch": 0.24052,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.02029596964518229,
+      "learning_rate": 0.0001,
+      "loss": 4.1182,
+      "loss/crossentropy": 1.6923209428787231,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19288206100463867,
+      "step": 12026
+    },
+    {
+      "epoch": 0.24056,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.016434478759765624,
+      "learning_rate": 0.0001,
+      "loss": 4.1733,
+      "loss/crossentropy": 2.071919083595276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21092405915260315,
+      "step": 12028
+    },
+    {
+      "epoch": 0.2406,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.016993967692057292,
+      "learning_rate": 0.0001,
+      "loss": 3.9688,
+      "loss/crossentropy": 1.994953691959381,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19953829050064087,
+      "step": 12030
+    },
+    {
+      "epoch": 0.24064,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011201731363932292,
+      "learning_rate": 0.0001,
+      "loss": 4.3252,
+      "loss/crossentropy": 2.0545560121536255,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20861412584781647,
+      "step": 12032
+    },
+    {
+      "epoch": 0.24068,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.0103424072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0859,
+      "loss/crossentropy": 2.0211291909217834,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2146567404270172,
+      "step": 12034
+    },
+    {
+      "epoch": 0.24072,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.012312825520833333,
+      "learning_rate": 0.0001,
+      "loss": 4.3156,
+      "loss/crossentropy": 2.165773868560791,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21522508561611176,
+      "step": 12036
+    },
+    {
+      "epoch": 0.24076,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.013216145833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.2253,
+      "loss/crossentropy": 2.0272024273872375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20012138038873672,
+      "step": 12038
+    },
+    {
+      "epoch": 0.2408,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.011946360270182291,
+      "learning_rate": 0.0001,
+      "loss": 4.2419,
+      "loss/crossentropy": 1.9311429262161255,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19755128771066666,
+      "step": 12040
+    },
+    {
+      "epoch": 0.24084,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.012086741129557292,
+      "learning_rate": 0.0001,
+      "loss": 4.1857,
+      "loss/crossentropy": 2.0740894079208374,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20651141554117203,
+      "step": 12042
+    },
+    {
+      "epoch": 0.24088,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.012876129150390625,
+      "learning_rate": 0.0001,
+      "loss": 3.9882,
+      "loss/crossentropy": 2.215203881263733,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21735452860593796,
+      "step": 12044
+    },
+    {
+      "epoch": 0.24092,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.02371190388997396,
+      "learning_rate": 0.0001,
+      "loss": 4.3755,
+      "loss/crossentropy": 1.9994693994522095,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22001128643751144,
+      "step": 12046
+    },
+    {
+      "epoch": 0.24096,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.025187174479166668,
+      "learning_rate": 0.0001,
+      "loss": 4.1145,
+      "loss/crossentropy": 2.195701003074646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20777013897895813,
+      "step": 12048
+    },
+    {
+      "epoch": 0.241,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.02474950154622396,
+      "learning_rate": 0.0001,
+      "loss": 4.3486,
+      "loss/crossentropy": 2.1975715160369873,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21987678855657578,
+      "step": 12050
+    },
+    {
+      "epoch": 0.24104,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.02188695271809896,
+      "learning_rate": 0.0001,
+      "loss": 4.1337,
+      "loss/crossentropy": 2.225023865699768,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24454358220100403,
+      "step": 12052
+    },
+    {
+      "epoch": 0.24108,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.022200520833333334,
+      "learning_rate": 0.0001,
+      "loss": 4.0404,
+      "loss/crossentropy": 1.892149806022644,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2110423520207405,
+      "step": 12054
+    },
+    {
+      "epoch": 0.24112,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.023579915364583332,
+      "learning_rate": 0.0001,
+      "loss": 4.1604,
+      "loss/crossentropy": 1.9996158480644226,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1821269765496254,
+      "step": 12056
+    },
+    {
+      "epoch": 0.24116,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.02569580078125,
+      "learning_rate": 0.0001,
+      "loss": 4.0591,
+      "loss/crossentropy": 2.0965115427970886,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2140883505344391,
+      "step": 12058
+    },
+    {
+      "epoch": 0.2412,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.0256500244140625,
+      "learning_rate": 0.0001,
+      "loss": 3.9494,
+      "loss/crossentropy": 1.9169449210166931,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2034926936030388,
+      "step": 12060
+    },
+    {
+      "epoch": 0.24124,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.015135701497395833,
+      "learning_rate": 0.0001,
+      "loss": 4.5,
+      "loss/crossentropy": 2.3266680240631104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24314038455486298,
+      "step": 12062
+    },
+    {
+      "epoch": 0.24128,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.013509114583333334,
+      "learning_rate": 0.0001,
+      "loss": 4.1009,
+      "loss/crossentropy": 2.0337759256362915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20842785388231277,
+      "step": 12064
+    },
+    {
+      "epoch": 0.24132,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.013618977864583333,
+      "learning_rate": 0.0001,
+      "loss": 3.9186,
+      "loss/crossentropy": 1.9502894878387451,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1812899112701416,
+      "step": 12066
+    },
+    {
+      "epoch": 0.24136,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.013822174072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.269,
+      "loss/crossentropy": 2.13326895236969,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21872679144144058,
+      "step": 12068
+    },
+    {
+      "epoch": 0.2414,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.013304646809895833,
+      "learning_rate": 0.0001,
+      "loss": 4.3594,
+      "loss/crossentropy": 2.0836809873580933,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.212929405272007,
+      "step": 12070
+    },
+    {
+      "epoch": 0.24144,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.012516276041666666,
+      "learning_rate": 0.0001,
+      "loss": 4.0467,
+      "loss/crossentropy": 2.207913398742676,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20640508085489273,
+      "step": 12072
+    },
+    {
+      "epoch": 0.24148,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.01080322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1877,
+      "loss/crossentropy": 2.0770451426506042,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20633003860712051,
+      "step": 12074
+    },
+    {
+      "epoch": 0.24152,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010786946614583333,
+      "learning_rate": 0.0001,
+      "loss": 4.2237,
+      "loss/crossentropy": 2.0042858719825745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18523608148097992,
+      "step": 12076
+    },
+    {
+      "epoch": 0.24156,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.010141754150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2755,
+      "loss/crossentropy": 1.8673237562179565,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19267796725034714,
+      "step": 12078
+    },
+    {
+      "epoch": 0.2416,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.013793690999348959,
+      "learning_rate": 0.0001,
+      "loss": 3.9766,
+      "loss/crossentropy": 2.0686238408088684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19344990700483322,
+      "step": 12080
+    },
+    {
+      "epoch": 0.24164,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009504954020182291,
+      "learning_rate": 0.0001,
+      "loss": 4.0366,
+      "loss/crossentropy": 2.0069685578346252,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1971682757139206,
+      "step": 12082
+    },
+    {
+      "epoch": 0.24168,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009468587239583333,
+      "learning_rate": 0.0001,
+      "loss": 4.1975,
+      "loss/crossentropy": 2.2052754163742065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20646511763334274,
+      "step": 12084
+    },
+    {
+      "epoch": 0.24172,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.008983357747395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1362,
+      "loss/crossentropy": 2.0343621373176575,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19575881958007812,
+      "step": 12086
+    },
+    {
+      "epoch": 0.24176,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.009894816080729167,
+      "learning_rate": 0.0001,
+      "loss": 3.9695,
+      "loss/crossentropy": 2.2500641345977783,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23151995986700058,
+      "step": 12088
+    },
+    {
+      "epoch": 0.2418,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.0375640869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.4361,
+      "loss/crossentropy": 2.2596821784973145,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22305716574192047,
+      "step": 12090
+    },
+    {
+      "epoch": 0.24184,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.03715184529622396,
+      "learning_rate": 0.0001,
+      "loss": 4.0381,
+      "loss/crossentropy": 2.0940088033676147,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23409207165241241,
+      "step": 12092
+    },
+    {
+      "epoch": 0.24188,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.03586018880208333,
+      "learning_rate": 0.0001,
+      "loss": 3.9715,
+      "loss/crossentropy": 1.6980834603309631,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19339510053396225,
+      "step": 12094
+    },
+    {
+      "epoch": 0.24192,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.03047459920247396,
+      "learning_rate": 0.0001,
+      "loss": 4.192,
+      "loss/crossentropy": 2.1921679973602295,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21085387468338013,
+      "step": 12096
+    },
+    {
+      "epoch": 0.24196,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.03551839192708333,
+      "learning_rate": 0.0001,
+      "loss": 4.3551,
+      "loss/crossentropy": 2.18610817193985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2332247570157051,
+      "step": 12098
+    },
+    {
+      "epoch": 0.242,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.038852691650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1216,
+      "loss/crossentropy": 2.08358097076416,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21169763058423996,
+      "step": 12100
+    },
+    {
+      "epoch": 0.24204,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.038913726806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2339,
+      "loss/crossentropy": 2.1001542806625366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2132650762796402,
+      "step": 12102
+    },
+    {
+      "epoch": 0.24208,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.033614095052083334,
+      "learning_rate": 0.0001,
+      "loss": 4.2051,
+      "loss/crossentropy": 1.9031851887702942,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2042345404624939,
+      "step": 12104
+    },
+    {
+      "epoch": 0.24212,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0127838134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0341,
+      "loss/crossentropy": 1.670085072517395,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1878880336880684,
+      "step": 12106
+    },
+    {
+      "epoch": 0.24216,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.012914784749348958,
+      "learning_rate": 0.0001,
+      "loss": 4.3395,
+      "loss/crossentropy": 1.9557109475135803,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2211412936449051,
+      "step": 12108
+    },
+    {
+      "epoch": 0.2422,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.013637034098307292,
+      "learning_rate": 0.0001,
+      "loss": 4.0107,
+      "loss/crossentropy": 2.085852086544037,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21336784213781357,
+      "step": 12110
+    },
+    {
+      "epoch": 0.24224,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.015721638997395832,
+      "learning_rate": 0.0001,
+      "loss": 4.0321,
+      "loss/crossentropy": 2.1064602732658386,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2096766158938408,
+      "step": 12112
+    },
+    {
+      "epoch": 0.24228,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009325917561848958,
+      "learning_rate": 0.0001,
+      "loss": 4.2998,
+      "loss/crossentropy": 2.381577968597412,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23788201808929443,
+      "step": 12114
+    },
+    {
+      "epoch": 0.24232,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.006758626302083333,
+      "learning_rate": 0.0001,
+      "loss": 4.0687,
+      "loss/crossentropy": 1.7200234532356262,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18620850890874863,
+      "step": 12116
+    },
+    {
+      "epoch": 0.24236,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.007697550455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2672,
+      "loss/crossentropy": 1.9106165170669556,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21456392109394073,
+      "step": 12118
+    },
+    {
+      "epoch": 0.2424,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.007106272379557291,
+      "learning_rate": 0.0001,
+      "loss": 4.04,
+      "loss/crossentropy": 1.9852410554885864,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1927378550171852,
+      "step": 12120
+    },
+    {
+      "epoch": 0.24244,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.007991282145182292,
+      "learning_rate": 0.0001,
+      "loss": 4.1278,
+      "loss/crossentropy": 1.4948370456695557,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1698828637599945,
+      "step": 12122
+    },
+    {
+      "epoch": 0.24248,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0078857421875,
+      "learning_rate": 0.0001,
+      "loss": 4.2874,
+      "loss/crossentropy": 1.956885814666748,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19851476699113846,
+      "step": 12124
+    },
+    {
+      "epoch": 0.24252,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.009422810872395833,
+      "learning_rate": 0.0001,
+      "loss": 3.8709,
+      "loss/crossentropy": 1.9052257537841797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18438522517681122,
+      "step": 12126
+    },
+    {
+      "epoch": 0.24256,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.008430735270182291,
+      "learning_rate": 0.0001,
+      "loss": 4.4347,
+      "loss/crossentropy": 2.353670358657837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22648276388645172,
+      "step": 12128
+    },
+    {
+      "epoch": 0.2426,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.01263427734375,
+      "learning_rate": 0.0001,
+      "loss": 3.9896,
+      "loss/crossentropy": 1.714030683040619,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19275055080652237,
+      "step": 12130
+    },
+    {
+      "epoch": 0.24264,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.012898763020833334,
+      "learning_rate": 0.0001,
+      "loss": 4.2981,
+      "loss/crossentropy": 2.0048200488090515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20478381216526031,
+      "step": 12132
+    },
+    {
+      "epoch": 0.24268,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.011766560872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0497,
+      "loss/crossentropy": 1.7151115536689758,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19872941821813583,
+      "step": 12134
+    },
+    {
+      "epoch": 0.24272,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011818186442057291,
+      "learning_rate": 0.0001,
+      "loss": 4.2753,
+      "loss/crossentropy": 2.1646838188171387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22345983982086182,
+      "step": 12136
+    },
+    {
+      "epoch": 0.24276,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.010807037353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.157,
+      "loss/crossentropy": 2.053311765193939,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2143322378396988,
+      "step": 12138
+    },
+    {
+      "epoch": 0.2428,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.009993235270182291,
+      "learning_rate": 0.0001,
+      "loss": 3.8805,
+      "loss/crossentropy": 2.0471617579460144,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20450014621019363,
+      "step": 12140
+    },
+    {
+      "epoch": 0.24284,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.008737945556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2966,
+      "loss/crossentropy": 2.1727033853530884,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22905820608139038,
+      "step": 12142
+    },
+    {
+      "epoch": 0.24288,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.008245595296223958,
+      "learning_rate": 0.0001,
+      "loss": 4.3861,
+      "loss/crossentropy": 2.197450280189514,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22059273719787598,
+      "step": 12144
+    },
+    {
+      "epoch": 0.24292,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.003413899739583333,
+      "learning_rate": 0.0001,
+      "loss": 4.4841,
+      "loss/crossentropy": 1.9954137206077576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22441548854112625,
+      "step": 12146
+    },
+    {
+      "epoch": 0.24296,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.005110677083333333,
+      "learning_rate": 0.0001,
+      "loss": 4.2113,
+      "loss/crossentropy": 2.3393132090568542,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22949577867984772,
+      "step": 12148
+    },
+    {
+      "epoch": 0.243,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.0067291259765625,
+      "learning_rate": 0.0001,
+      "loss": 4.146,
+      "loss/crossentropy": 2.11602646112442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19181709736585617,
+      "step": 12150
+    },
+    {
+      "epoch": 0.24304,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.005597941080729167,
+      "learning_rate": 0.0001,
+      "loss": 4.1381,
+      "loss/crossentropy": 2.130657136440277,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20904043316841125,
+      "step": 12152
+    },
+    {
+      "epoch": 0.24308,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0053293863932291664,
+      "learning_rate": 0.0001,
+      "loss": 4.2983,
+      "loss/crossentropy": 2.3923556804656982,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23029828071594238,
+      "step": 12154
+    },
+    {
+      "epoch": 0.24312,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.007209269205729166,
+      "learning_rate": 0.0001,
+      "loss": 3.9697,
+      "loss/crossentropy": 2.024892747402191,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19879258424043655,
+      "step": 12156
+    },
+    {
+      "epoch": 0.24316,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.007826487223307291,
+      "learning_rate": 0.0001,
+      "loss": 3.8761,
+      "loss/crossentropy": 1.7833393812179565,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18542324006557465,
+      "step": 12158
+    },
+    {
+      "epoch": 0.2432,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.007328033447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1896,
+      "loss/crossentropy": 2.253599762916565,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2148476019501686,
+      "step": 12160
+    },
+    {
+      "epoch": 0.24324,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.005177561442057292,
+      "learning_rate": 0.0001,
+      "loss": 4.0253,
+      "loss/crossentropy": 1.7218471765518188,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1923334002494812,
+      "step": 12162
+    },
+    {
+      "epoch": 0.24328,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.007575480143229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2947,
+      "loss/crossentropy": 2.210463523864746,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2203536182641983,
+      "step": 12164
+    },
+    {
+      "epoch": 0.24332,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0076487223307291664,
+      "learning_rate": 0.0001,
+      "loss": 4.4164,
+      "loss/crossentropy": 2.452837347984314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22888235747814178,
+      "step": 12166
+    },
+    {
+      "epoch": 0.24336,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.007657877604166667,
+      "learning_rate": 0.0001,
+      "loss": 4.2157,
+      "loss/crossentropy": 2.3511279821395874,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2431751787662506,
+      "step": 12168
+    },
+    {
+      "epoch": 0.2434,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.007819620768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0911,
+      "loss/crossentropy": 2.0734334588050842,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21137738972902298,
+      "step": 12170
+    },
+    {
+      "epoch": 0.24344,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0065582275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0987,
+      "loss/crossentropy": 1.9492397904396057,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22894760966300964,
+      "step": 12172
+    },
+    {
+      "epoch": 0.24348,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.005936431884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2316,
+      "loss/crossentropy": 2.03458708524704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20981666445732117,
+      "step": 12174
+    },
+    {
+      "epoch": 0.24352,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.007059478759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3396,
+      "loss/crossentropy": 2.1464394330978394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20741496980190277,
+      "step": 12176
+    },
+    {
+      "epoch": 0.24356,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0064999898274739586,
+      "learning_rate": 0.0001,
+      "loss": 3.9902,
+      "loss/crossentropy": 1.8449034094810486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21950078010559082,
+      "step": 12178
+    },
+    {
+      "epoch": 0.2436,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0054433186848958336,
+      "learning_rate": 0.0001,
+      "loss": 4.0129,
+      "loss/crossentropy": 1.8646993041038513,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2117064744234085,
+      "step": 12180
+    },
+    {
+      "epoch": 0.24364,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.008003743489583333,
+      "learning_rate": 0.0001,
+      "loss": 3.9347,
+      "loss/crossentropy": 1.9643146991729736,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1946149319410324,
+      "step": 12182
+    },
+    {
+      "epoch": 0.24368,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.008141835530598959,
+      "learning_rate": 0.0001,
+      "loss": 4.1468,
+      "loss/crossentropy": 1.867002248764038,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19111115485429764,
+      "step": 12184
+    },
+    {
+      "epoch": 0.24372,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.008129628499348958,
+      "learning_rate": 0.0001,
+      "loss": 4.0443,
+      "loss/crossentropy": 1.9431232810020447,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19612180441617966,
+      "step": 12186
+    },
+    {
+      "epoch": 0.24376,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0105224609375,
+      "learning_rate": 0.0001,
+      "loss": 4.2931,
+      "loss/crossentropy": 2.197216033935547,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22340577840805054,
+      "step": 12188
+    },
+    {
+      "epoch": 0.2438,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.010469563802083333,
+      "learning_rate": 0.0001,
+      "loss": 4.2499,
+      "loss/crossentropy": 1.916576623916626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20536810904741287,
+      "step": 12190
+    },
+    {
+      "epoch": 0.24384,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.010497029622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3838,
+      "loss/crossentropy": 2.0369369983673096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26386965811252594,
+      "step": 12192
+    },
+    {
+      "epoch": 0.24388,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.01639404296875,
+      "learning_rate": 0.0001,
+      "loss": 4.2817,
+      "loss/crossentropy": 2.086443066596985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21351207792758942,
+      "step": 12194
+    },
+    {
+      "epoch": 0.24392,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.01619873046875,
+      "learning_rate": 0.0001,
+      "loss": 4.3649,
+      "loss/crossentropy": 2.0969839096069336,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21489758789539337,
+      "step": 12196
+    },
+    {
+      "epoch": 0.24396,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.016463216145833334,
+      "learning_rate": 0.0001,
+      "loss": 4.2481,
+      "loss/crossentropy": 2.086832642555237,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20651830732822418,
+      "step": 12198
+    },
+    {
+      "epoch": 0.244,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.017146809895833334,
+      "learning_rate": 0.0001,
+      "loss": 4.1397,
+      "loss/crossentropy": 1.9104391932487488,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1963956654071808,
+      "step": 12200
+    },
+    {
+      "epoch": 0.24404,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0191314697265625,
+      "learning_rate": 0.0001,
+      "loss": 3.8558,
+      "loss/crossentropy": 1.9419523477554321,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19279541075229645,
+      "step": 12202
+    },
+    {
+      "epoch": 0.24408,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.01697998046875,
+      "learning_rate": 0.0001,
+      "loss": 4.3239,
+      "loss/crossentropy": 2.2867754697799683,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23261378705501556,
+      "step": 12204
+    },
+    {
+      "epoch": 0.24412,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.017634073893229168,
+      "learning_rate": 0.0001,
+      "loss": 4.0786,
+      "loss/crossentropy": 1.941315233707428,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20962880551815033,
+      "step": 12206
+    },
+    {
+      "epoch": 0.24416,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.01541748046875,
+      "learning_rate": 0.0001,
+      "loss": 4.2044,
+      "loss/crossentropy": 2.0638798475265503,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21499283611774445,
+      "step": 12208
+    },
+    {
+      "epoch": 0.2442,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.008454386393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1889,
+      "loss/crossentropy": 2.3167499899864197,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21805942803621292,
+      "step": 12210
+    },
+    {
+      "epoch": 0.24424,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.00848388671875,
+      "learning_rate": 0.0001,
+      "loss": 4.0956,
+      "loss/crossentropy": 2.1785646080970764,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21259736269712448,
+      "step": 12212
+    },
+    {
+      "epoch": 0.24428,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.006615193684895834,
+      "learning_rate": 0.0001,
+      "loss": 4.2054,
+      "loss/crossentropy": 1.983469545841217,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21091710776090622,
+      "step": 12214
+    },
+    {
+      "epoch": 0.24432,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.006937408447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2283,
+      "loss/crossentropy": 1.8501896858215332,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20537292212247849,
+      "step": 12216
+    },
+    {
+      "epoch": 0.24436,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0072934468587239586,
+      "learning_rate": 0.0001,
+      "loss": 4.182,
+      "loss/crossentropy": 1.7293490767478943,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1916719302535057,
+      "step": 12218
+    },
+    {
+      "epoch": 0.2444,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0069488525390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3247,
+      "loss/crossentropy": 2.0931429862976074,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22081860899925232,
+      "step": 12220
+    },
+    {
+      "epoch": 0.24444,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.007657623291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0619,
+      "loss/crossentropy": 2.157875657081604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20259525626897812,
+      "step": 12222
+    },
+    {
+      "epoch": 0.24448,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.007248687744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2712,
+      "loss/crossentropy": 2.2028552889823914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2431233748793602,
+      "step": 12224
+    },
+    {
+      "epoch": 0.24452,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.006473541259765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1184,
+      "loss/crossentropy": 2.0906929969787598,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2051575481891632,
+      "step": 12226
+    },
+    {
+      "epoch": 0.24456,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.006091054280598958,
+      "learning_rate": 0.0001,
+      "loss": 4.5601,
+      "loss/crossentropy": 2.5020272731781006,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2485819309949875,
+      "step": 12228
+    },
+    {
+      "epoch": 0.2446,
+      "grad_norm": 8.0625,
+      "grad_norm_var": 2.244887034098307,
+      "learning_rate": 0.0001,
+      "loss": 4.1465,
+      "loss/crossentropy": 1.502736508846283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1839340552687645,
+      "step": 12230
+    },
+    {
+      "epoch": 0.24464,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 2.2357358296712238,
+      "learning_rate": 0.0001,
+      "loss": 4.3882,
+      "loss/crossentropy": 2.1888676285743713,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.226266011595726,
+      "step": 12232
+    },
+    {
+      "epoch": 0.24468,
+      "grad_norm": 2.125,
+      "grad_norm_var": 2.243033599853516,
+      "learning_rate": 0.0001,
+      "loss": 4.3887,
+      "loss/crossentropy": 2.377061367034912,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23195213079452515,
+      "step": 12234
+    },
+    {
+      "epoch": 0.24472,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 2.257559967041016,
+      "learning_rate": 0.0001,
+      "loss": 3.9354,
+      "loss/crossentropy": 2.0567076206207275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20983586460351944,
+      "step": 12236
+    },
+    {
+      "epoch": 0.24476,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 2.246906534830729,
+      "learning_rate": 0.0001,
+      "loss": 4.1978,
+      "loss/crossentropy": 2.007763922214508,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21983467787504196,
+      "step": 12238
+    },
+    {
+      "epoch": 0.2448,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 2.247749837239583,
+      "learning_rate": 0.0001,
+      "loss": 4.1411,
+      "loss/crossentropy": 2.3309481143951416,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23099908232688904,
+      "step": 12240
+    },
+    {
+      "epoch": 0.24484,
+      "grad_norm": 2.5,
+      "grad_norm_var": 2.23668212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2159,
+      "loss/crossentropy": 1.8231184482574463,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22009263187646866,
+      "step": 12242
+    },
+    {
+      "epoch": 0.24488,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 2.2464637756347656,
+      "learning_rate": 0.0001,
+      "loss": 4.1131,
+      "loss/crossentropy": 2.140045642852783,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22202204167842865,
+      "step": 12244
+    },
+    {
+      "epoch": 0.24492,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.02835057576497396,
+      "learning_rate": 0.0001,
+      "loss": 4.2209,
+      "loss/crossentropy": 1.8495931029319763,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19803623855113983,
+      "step": 12246
+    },
+    {
+      "epoch": 0.24496,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0185943603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1624,
+      "loss/crossentropy": 2.2141982913017273,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23418182879686356,
+      "step": 12248
+    },
+    {
+      "epoch": 0.245,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.01844482421875,
+      "learning_rate": 0.0001,
+      "loss": 4.4104,
+      "loss/crossentropy": 2.14465594291687,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22683896869421005,
+      "step": 12250
+    },
+    {
+      "epoch": 0.24504,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.01610107421875,
+      "learning_rate": 0.0001,
+      "loss": 4.1913,
+      "loss/crossentropy": 2.0804443359375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2265845239162445,
+      "step": 12252
+    },
+    {
+      "epoch": 0.24508,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0160308837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.485,
+      "loss/crossentropy": 2.2451776266098022,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22881492972373962,
+      "step": 12254
+    },
+    {
+      "epoch": 0.24512,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.016947428385416668,
+      "learning_rate": 0.0001,
+      "loss": 4.0869,
+      "loss/crossentropy": 1.6861794590950012,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18126793205738068,
+      "step": 12256
+    },
+    {
+      "epoch": 0.24516,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0059773763020833336,
+      "learning_rate": 0.0001,
+      "loss": 4.2343,
+      "loss/crossentropy": 1.87660551071167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20990663766860962,
+      "step": 12258
+    },
+    {
+      "epoch": 0.2452,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.006799062093098958,
+      "learning_rate": 0.0001,
+      "loss": 4.2019,
+      "loss/crossentropy": 2.103124976158142,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19615671038627625,
+      "step": 12260
+    },
+    {
+      "epoch": 0.24524,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0069048563639322914,
+      "learning_rate": 0.0001,
+      "loss": 4.2509,
+      "loss/crossentropy": 2.139270842075348,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21861301362514496,
+      "step": 12262
+    },
+    {
+      "epoch": 0.24528,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.00791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.4081,
+      "loss/crossentropy": 2.108114778995514,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20996354520320892,
+      "step": 12264
+    },
+    {
+      "epoch": 0.24532,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.008847808837890625,
+      "learning_rate": 0.0001,
+      "loss": 3.9495,
+      "loss/crossentropy": 2.1205111145973206,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21352755278348923,
+      "step": 12266
+    },
+    {
+      "epoch": 0.24536,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009248860677083333,
+      "learning_rate": 0.0001,
+      "loss": 3.9946,
+      "loss/crossentropy": 2.20136821269989,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2277800738811493,
+      "step": 12268
+    },
+    {
+      "epoch": 0.2454,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.011336008707682291,
+      "learning_rate": 0.0001,
+      "loss": 4.1116,
+      "loss/crossentropy": 2.1149147748947144,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2062554806470871,
+      "step": 12270
+    },
+    {
+      "epoch": 0.24544,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.01932347615559896,
+      "learning_rate": 0.0001,
+      "loss": 4.5226,
+      "loss/crossentropy": 2.1119648218154907,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2922344133257866,
+      "step": 12272
+    },
+    {
+      "epoch": 0.24548,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.017964680989583332,
+      "learning_rate": 0.0001,
+      "loss": 4.1766,
+      "loss/crossentropy": 2.2631434202194214,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2044210433959961,
+      "step": 12274
+    },
+    {
+      "epoch": 0.24552,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.06928609212239584,
+      "learning_rate": 0.0001,
+      "loss": 4.0956,
+      "loss/crossentropy": 2.264181971549988,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23541904985904694,
+      "step": 12276
+    },
+    {
+      "epoch": 0.24556,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.06836649576822916,
+      "learning_rate": 0.0001,
+      "loss": 4.2083,
+      "loss/crossentropy": 2.256329298019409,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21784210950136185,
+      "step": 12278
+    },
+    {
+      "epoch": 0.2456,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.06968765258789063,
+      "learning_rate": 0.0001,
+      "loss": 4.267,
+      "loss/crossentropy": 2.0962833166122437,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.211236834526062,
+      "step": 12280
+    },
+    {
+      "epoch": 0.24564,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0683990478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4767,
+      "loss/crossentropy": 2.3430999517440796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25170228630304337,
+      "step": 12282
+    },
+    {
+      "epoch": 0.24568,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.06921361287434896,
+      "learning_rate": 0.0001,
+      "loss": 4.1492,
+      "loss/crossentropy": 1.8256065845489502,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20632392168045044,
+      "step": 12284
+    },
+    {
+      "epoch": 0.24572,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.06643778483072917,
+      "learning_rate": 0.0001,
+      "loss": 4.294,
+      "loss/crossentropy": 2.481971561908722,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2121504619717598,
+      "step": 12286
+    },
+    {
+      "epoch": 0.24576,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.060469563802083334,
+      "learning_rate": 0.0001,
+      "loss": 4.2136,
+      "loss/crossentropy": 2.2815581560134888,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22004567831754684,
+      "step": 12288
+    },
+    {
+      "epoch": 0.2458,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.060323079427083336,
+      "learning_rate": 0.0001,
+      "loss": 4.2426,
+      "loss/crossentropy": 1.7125394940376282,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19618325680494308,
+      "step": 12290
+    },
+    {
+      "epoch": 0.24584,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.008347320556640624,
+      "learning_rate": 0.0001,
+      "loss": 3.9676,
+      "loss/crossentropy": 2.0333253145217896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20251524448394775,
+      "step": 12292
+    },
+    {
+      "epoch": 0.24588,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.008939361572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0958,
+      "loss/crossentropy": 2.359953284263611,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23697812855243683,
+      "step": 12294
+    },
+    {
+      "epoch": 0.24592,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0065348307291666664,
+      "learning_rate": 0.0001,
+      "loss": 3.9914,
+      "loss/crossentropy": 1.8465049266815186,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18830078095197678,
+      "step": 12296
+    },
+    {
+      "epoch": 0.24596,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.00662841796875,
+      "learning_rate": 0.0001,
+      "loss": 4.1669,
+      "loss/crossentropy": 2.407967984676361,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20054074376821518,
+      "step": 12298
+    },
+    {
+      "epoch": 0.246,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007005818684895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1943,
+      "loss/crossentropy": 2.306265115737915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2286146879196167,
+      "step": 12300
+    },
+    {
+      "epoch": 0.24604,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0076416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.108,
+      "loss/crossentropy": 1.9776363968849182,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2015514224767685,
+      "step": 12302
+    },
+    {
+      "epoch": 0.24608,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.010041300455729167,
+      "learning_rate": 0.0001,
+      "loss": 3.7023,
+      "loss/crossentropy": 1.6589386463165283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16657201945781708,
+      "step": 12304
+    },
+    {
+      "epoch": 0.24612,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.009968058268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2749,
+      "loss/crossentropy": 2.263510227203369,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21220041066408157,
+      "step": 12306
+    },
+    {
+      "epoch": 0.24616,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008583323160807291,
+      "learning_rate": 0.0001,
+      "loss": 4.1735,
+      "loss/crossentropy": 2.0802704095840454,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20156628638505936,
+      "step": 12308
+    },
+    {
+      "epoch": 0.2462,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.015193430582682292,
+      "learning_rate": 0.0001,
+      "loss": 4.2597,
+      "loss/crossentropy": 2.0921266674995422,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22044362872838974,
+      "step": 12310
+    },
+    {
+      "epoch": 0.24624,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.0164703369140625,
+      "learning_rate": 0.0001,
+      "loss": 3.8587,
+      "loss/crossentropy": 1.7366862297058105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1722203940153122,
+      "step": 12312
+    },
+    {
+      "epoch": 0.24628,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.016377766927083332,
+      "learning_rate": 0.0001,
+      "loss": 4.1408,
+      "loss/crossentropy": 2.0741729140281677,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1942591667175293,
+      "step": 12314
+    },
+    {
+      "epoch": 0.24632,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0145751953125,
+      "learning_rate": 0.0001,
+      "loss": 4.1343,
+      "loss/crossentropy": 2.0611414909362793,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22244945168495178,
+      "step": 12316
+    },
+    {
+      "epoch": 0.24636,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.014196523030598958,
+      "learning_rate": 0.0001,
+      "loss": 3.8586,
+      "loss/crossentropy": 1.9337337017059326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20655318349599838,
+      "step": 12318
+    },
+    {
+      "epoch": 0.2464,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.012141672770182292,
+      "learning_rate": 0.0001,
+      "loss": 3.9197,
+      "loss/crossentropy": 1.8766502737998962,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1851816102862358,
+      "step": 12320
+    },
+    {
+      "epoch": 0.24644,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.012552897135416666,
+      "learning_rate": 0.0001,
+      "loss": 4.2407,
+      "loss/crossentropy": 2.121203899383545,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2177201583981514,
+      "step": 12322
+    },
+    {
+      "epoch": 0.24648,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0136138916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1441,
+      "loss/crossentropy": 1.8102782368659973,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19694476574659348,
+      "step": 12324
+    },
+    {
+      "epoch": 0.24652,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0062164306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2526,
+      "loss/crossentropy": 2.323423147201538,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20612536370754242,
+      "step": 12326
+    },
+    {
+      "epoch": 0.24656,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.007814280192057292,
+      "learning_rate": 0.0001,
+      "loss": 4.4862,
+      "loss/crossentropy": 2.2835768461227417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22697383165359497,
+      "step": 12328
+    },
+    {
+      "epoch": 0.2466,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.006192779541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2868,
+      "loss/crossentropy": 2.197639048099518,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22446971386671066,
+      "step": 12330
+    },
+    {
+      "epoch": 0.24664,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.006761678059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0758,
+      "loss/crossentropy": 2.1731618642807007,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21908622235059738,
+      "step": 12332
+    },
+    {
+      "epoch": 0.24668,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.009291330973307291,
+      "learning_rate": 0.0001,
+      "loss": 3.9108,
+      "loss/crossentropy": 1.9236284494400024,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1810280755162239,
+      "step": 12334
+    },
+    {
+      "epoch": 0.24672,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.008318837483723958,
+      "learning_rate": 0.0001,
+      "loss": 4.3975,
+      "loss/crossentropy": 2.2902809381484985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23192713409662247,
+      "step": 12336
+    },
+    {
+      "epoch": 0.24676,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008367665608723958,
+      "learning_rate": 0.0001,
+      "loss": 4.1681,
+      "loss/crossentropy": 1.9808775186538696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19344759732484818,
+      "step": 12338
+    },
+    {
+      "epoch": 0.2468,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.008337148030598958,
+      "learning_rate": 0.0001,
+      "loss": 4.1581,
+      "loss/crossentropy": 2.148995041847229,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19994106143712997,
+      "step": 12340
+    },
+    {
+      "epoch": 0.24684,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.008528391520182291,
+      "learning_rate": 0.0001,
+      "loss": 4.3927,
+      "loss/crossentropy": 2.374568462371826,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24350540339946747,
+      "step": 12342
+    },
+    {
+      "epoch": 0.24688,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.006959788004557292,
+      "learning_rate": 0.0001,
+      "loss": 4.258,
+      "loss/crossentropy": 1.9702014923095703,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2038111537694931,
+      "step": 12344
+    },
+    {
+      "epoch": 0.24692,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.007008616129557292,
+      "learning_rate": 0.0001,
+      "loss": 4.0126,
+      "loss/crossentropy": 1.8438855409622192,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1822836771607399,
+      "step": 12346
+    },
+    {
+      "epoch": 0.24696,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.010754140218098958,
+      "learning_rate": 0.0001,
+      "loss": 4.1352,
+      "loss/crossentropy": 2.005755662918091,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21399007737636566,
+      "step": 12348
+    },
+    {
+      "epoch": 0.247,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008819325764973959,
+      "learning_rate": 0.0001,
+      "loss": 4.0988,
+      "loss/crossentropy": 1.8640305399894714,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19055305421352386,
+      "step": 12350
+    },
+    {
+      "epoch": 0.24704,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.010453033447265624,
+      "learning_rate": 0.0001,
+      "loss": 4.1981,
+      "loss/crossentropy": 2.1293725967407227,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2117801234126091,
+      "step": 12352
+    },
+    {
+      "epoch": 0.24708,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.02337621053059896,
+      "learning_rate": 0.0001,
+      "loss": 4.2666,
+      "loss/crossentropy": 1.7655459642410278,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2085946872830391,
+      "step": 12354
+    },
+    {
+      "epoch": 0.24712,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.021897125244140624,
+      "learning_rate": 0.0001,
+      "loss": 4.3175,
+      "loss/crossentropy": 2.184986114501953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22597889602184296,
+      "step": 12356
+    },
+    {
+      "epoch": 0.24716,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.021897125244140624,
+      "learning_rate": 0.0001,
+      "loss": 4.1295,
+      "loss/crossentropy": 2.368021607398987,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21957845985889435,
+      "step": 12358
+    },
+    {
+      "epoch": 0.2472,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.020645904541015624,
+      "learning_rate": 0.0001,
+      "loss": 4.16,
+      "loss/crossentropy": 2.297884225845337,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2155354768037796,
+      "step": 12360
+    },
+    {
+      "epoch": 0.24724,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.023273722330729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2181,
+      "loss/crossentropy": 2.121878147125244,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22192668169736862,
+      "step": 12362
+    },
+    {
+      "epoch": 0.24728,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.020442454020182292,
+      "learning_rate": 0.0001,
+      "loss": 4.3233,
+      "loss/crossentropy": 2.130228877067566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20712029188871384,
+      "step": 12364
+    },
+    {
+      "epoch": 0.24732,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.02367121378580729,
+      "learning_rate": 0.0001,
+      "loss": 3.8542,
+      "loss/crossentropy": 1.6520383954048157,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16511157900094986,
+      "step": 12366
+    },
+    {
+      "epoch": 0.24736,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.02220637003580729,
+      "learning_rate": 0.0001,
+      "loss": 4.1894,
+      "loss/crossentropy": 1.9993118047714233,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2112603336572647,
+      "step": 12368
+    },
+    {
+      "epoch": 0.2474,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.011494700113932292,
+      "learning_rate": 0.0001,
+      "loss": 4.0522,
+      "loss/crossentropy": 1.9502257108688354,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1944574937224388,
+      "step": 12370
+    },
+    {
+      "epoch": 0.24744,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.011872355143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1098,
+      "loss/crossentropy": 2.1219520568847656,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22872696816921234,
+      "step": 12372
+    },
+    {
+      "epoch": 0.24748,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0502593994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.172,
+      "loss/crossentropy": 2.0668599605560303,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21508124470710754,
+      "step": 12374
+    },
+    {
+      "epoch": 0.24752,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.05098241170247396,
+      "learning_rate": 0.0001,
+      "loss": 4.1901,
+      "loss/crossentropy": 2.242877721786499,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2308686003088951,
+      "step": 12376
+    },
+    {
+      "epoch": 0.24756,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.047304026285807294,
+      "learning_rate": 0.0001,
+      "loss": 3.8779,
+      "loss/crossentropy": 1.8258161544799805,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1852153167128563,
+      "step": 12378
+    },
+    {
+      "epoch": 0.2476,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.04918390909830729,
+      "learning_rate": 0.0001,
+      "loss": 4.1241,
+      "loss/crossentropy": 2.0654167532920837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20526036620140076,
+      "step": 12380
+    },
+    {
+      "epoch": 0.24764,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.04582087198893229,
+      "learning_rate": 0.0001,
+      "loss": 4.1601,
+      "loss/crossentropy": 2.125216484069824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.216343455016613,
+      "step": 12382
+    },
+    {
+      "epoch": 0.24768,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0473052978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1756,
+      "loss/crossentropy": 2.285245180130005,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22941745072603226,
+      "step": 12384
+    },
+    {
+      "epoch": 0.24772,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.04720637003580729,
+      "learning_rate": 0.0001,
+      "loss": 4.1097,
+      "loss/crossentropy": 2.0148350596427917,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1874203458428383,
+      "step": 12386
+    },
+    {
+      "epoch": 0.24776,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.04812825520833333,
+      "learning_rate": 0.0001,
+      "loss": 4.3127,
+      "loss/crossentropy": 2.2349741458892822,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23187313228845596,
+      "step": 12388
+    },
+    {
+      "epoch": 0.2478,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009178670247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1875,
+      "loss/crossentropy": 2.195721983909607,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2119324654340744,
+      "step": 12390
+    },
+    {
+      "epoch": 0.24784,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.012676747639973958,
+      "learning_rate": 0.0001,
+      "loss": 4.5732,
+      "loss/crossentropy": 2.1261669397354126,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21143332868814468,
+      "step": 12392
+    },
+    {
+      "epoch": 0.24788,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.012798817952473958,
+      "learning_rate": 0.0001,
+      "loss": 3.9475,
+      "loss/crossentropy": 2.1640073657035828,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2192412167787552,
+      "step": 12394
+    },
+    {
+      "epoch": 0.24792,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.011230214436848959,
+      "learning_rate": 0.0001,
+      "loss": 4.2435,
+      "loss/crossentropy": 1.9555792808532715,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2110436111688614,
+      "step": 12396
+    },
+    {
+      "epoch": 0.24796,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.011156972249348958,
+      "learning_rate": 0.0001,
+      "loss": 4.0616,
+      "loss/crossentropy": 1.948053002357483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19957780838012695,
+      "step": 12398
+    },
+    {
+      "epoch": 0.248,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.011156972249348958,
+      "learning_rate": 0.0001,
+      "loss": 4.0865,
+      "loss/crossentropy": 2.355304718017578,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20301833003759384,
+      "step": 12400
+    },
+    {
+      "epoch": 0.24804,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.010944620768229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2537,
+      "loss/crossentropy": 2.172769784927368,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2157917320728302,
+      "step": 12402
+    },
+    {
+      "epoch": 0.24808,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.012109120686848959,
+      "learning_rate": 0.0001,
+      "loss": 4.2817,
+      "loss/crossentropy": 2.2186524868011475,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.221016563475132,
+      "step": 12404
+    },
+    {
+      "epoch": 0.24812,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.012113189697265625,
+      "learning_rate": 0.0001,
+      "loss": 3.9624,
+      "loss/crossentropy": 2.011409044265747,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19150879979133606,
+      "step": 12406
+    },
+    {
+      "epoch": 0.24816,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010984039306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.5039,
+      "loss/crossentropy": 1.9245591163635254,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2296570986509323,
+      "step": 12408
+    },
+    {
+      "epoch": 0.2482,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.016434733072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.5765,
+      "loss/crossentropy": 1.973130702972412,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20532477647066116,
+      "step": 12410
+    },
+    {
+      "epoch": 0.24824,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.01617609659830729,
+      "learning_rate": 0.0001,
+      "loss": 3.914,
+      "loss/crossentropy": 1.655932605266571,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1609746441245079,
+      "step": 12412
+    },
+    {
+      "epoch": 0.24828,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.02411677042643229,
+      "learning_rate": 0.0001,
+      "loss": 4.2172,
+      "loss/crossentropy": 2.104023277759552,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19898276031017303,
+      "step": 12414
+    },
+    {
+      "epoch": 0.24832,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.02210261027018229,
+      "learning_rate": 0.0001,
+      "loss": 4.0597,
+      "loss/crossentropy": 1.9362882375717163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20107803493738174,
+      "step": 12416
+    },
+    {
+      "epoch": 0.24836,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.021996815999348957,
+      "learning_rate": 0.0001,
+      "loss": 4.4181,
+      "loss/crossentropy": 2.1508368253707886,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2056911736726761,
+      "step": 12418
+    },
+    {
+      "epoch": 0.2484,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.02194391886393229,
+      "learning_rate": 0.0001,
+      "loss": 3.802,
+      "loss/crossentropy": 1.7712991833686829,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1908227875828743,
+      "step": 12420
+    },
+    {
+      "epoch": 0.24844,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.02060114542643229,
+      "learning_rate": 0.0001,
+      "loss": 4.3557,
+      "loss/crossentropy": 2.2705806493759155,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23912303894758224,
+      "step": 12422
+    },
+    {
+      "epoch": 0.24848,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.019791666666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.2696,
+      "loss/crossentropy": 2.4222676753997803,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21739919483661652,
+      "step": 12424
+    },
+    {
+      "epoch": 0.24852,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0145172119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1935,
+      "loss/crossentropy": 1.945872962474823,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19402413070201874,
+      "step": 12426
+    },
+    {
+      "epoch": 0.24856,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.01646728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0908,
+      "loss/crossentropy": 1.7793864011764526,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19452669471502304,
+      "step": 12428
+    },
+    {
+      "epoch": 0.2486,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.015941365559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1045,
+      "loss/crossentropy": 2.2827813625335693,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21572840213775635,
+      "step": 12430
+    },
+    {
+      "epoch": 0.24864,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.016544342041015625,
+      "learning_rate": 0.0001,
+      "loss": 3.749,
+      "loss/crossentropy": 1.9543398022651672,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20825360715389252,
+      "step": 12432
+    },
+    {
+      "epoch": 0.24868,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.017276763916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3727,
+      "loss/crossentropy": 2.223303198814392,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21834088116884232,
+      "step": 12434
+    },
+    {
+      "epoch": 0.24872,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.017071278889973958,
+      "learning_rate": 0.0001,
+      "loss": 3.9993,
+      "loss/crossentropy": 1.6987267136573792,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1834847405552864,
+      "step": 12436
+    },
+    {
+      "epoch": 0.24876,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.014522043863932292,
+      "learning_rate": 0.0001,
+      "loss": 4.4309,
+      "loss/crossentropy": 2.3947317600250244,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22872642427682877,
+      "step": 12438
+    },
+    {
+      "epoch": 0.2488,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.014435831705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4858,
+      "loss/crossentropy": 2.5481021404266357,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23773372173309326,
+      "step": 12440
+    },
+    {
+      "epoch": 0.24884,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0145904541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1621,
+      "loss/crossentropy": 2.2349241971969604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21584660559892654,
+      "step": 12442
+    },
+    {
+      "epoch": 0.24888,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.012988026936848958,
+      "learning_rate": 0.0001,
+      "loss": 4.2827,
+      "loss/crossentropy": 1.9308242201805115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19295598566532135,
+      "step": 12444
+    },
+    {
+      "epoch": 0.24892,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.008536783854166667,
+      "learning_rate": 0.0001,
+      "loss": 4.0778,
+      "loss/crossentropy": 2.3508042097091675,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22008787840604782,
+      "step": 12446
+    },
+    {
+      "epoch": 0.24896,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012593332926432292,
+      "learning_rate": 0.0001,
+      "loss": 4.0865,
+      "loss/crossentropy": 2.0870128870010376,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20749760419130325,
+      "step": 12448
+    },
+    {
+      "epoch": 0.249,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012971750895182292,
+      "learning_rate": 0.0001,
+      "loss": 4.2758,
+      "loss/crossentropy": 2.087821125984192,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21110248565673828,
+      "step": 12450
+    },
+    {
+      "epoch": 0.24904,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.011494954427083334,
+      "learning_rate": 0.0001,
+      "loss": 4.3075,
+      "loss/crossentropy": 2.131770372390747,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21426154673099518,
+      "step": 12452
+    },
+    {
+      "epoch": 0.24908,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.011092122395833333,
+      "learning_rate": 0.0001,
+      "loss": 4.2373,
+      "loss/crossentropy": 2.175555467605591,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20418058335781097,
+      "step": 12454
+    },
+    {
+      "epoch": 0.24912,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.012198893229166667,
+      "learning_rate": 0.0001,
+      "loss": 4.2788,
+      "loss/crossentropy": 2.205165147781372,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23265192657709122,
+      "step": 12456
+    },
+    {
+      "epoch": 0.24916,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.014534250895182291,
+      "learning_rate": 0.0001,
+      "loss": 3.9345,
+      "loss/crossentropy": 1.6410180926322937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18058068305253983,
+      "step": 12458
+    },
+    {
+      "epoch": 0.2492,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.014345041910807292,
+      "learning_rate": 0.0001,
+      "loss": 3.962,
+      "loss/crossentropy": 1.8249012231826782,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19164791703224182,
+      "step": 12460
+    },
+    {
+      "epoch": 0.24924,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0142242431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4945,
+      "loss/crossentropy": 2.0906582474708557,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23415963351726532,
+      "step": 12462
+    },
+    {
+      "epoch": 0.24928,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009285227457682291,
+      "learning_rate": 0.0001,
+      "loss": 4.4702,
+      "loss/crossentropy": 2.314555048942566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24343416839838028,
+      "step": 12464
+    },
+    {
+      "epoch": 0.24932,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.009105428059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3059,
+      "loss/crossentropy": 2.1258983612060547,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21851430088281631,
+      "step": 12466
+    },
+    {
+      "epoch": 0.24936,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009325917561848958,
+      "learning_rate": 0.0001,
+      "loss": 4.1642,
+      "loss/crossentropy": 1.977954626083374,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20545368641614914,
+      "step": 12468
+    },
+    {
+      "epoch": 0.2494,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.011258951822916667,
+      "learning_rate": 0.0001,
+      "loss": 4.2224,
+      "loss/crossentropy": 2.0212838649749756,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21720624715089798,
+      "step": 12470
+    },
+    {
+      "epoch": 0.24944,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010205078125,
+      "learning_rate": 0.0001,
+      "loss": 4.3526,
+      "loss/crossentropy": 1.964124321937561,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19991052150726318,
+      "step": 12472
+    },
+    {
+      "epoch": 0.24948,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.008567047119140626,
+      "learning_rate": 0.0001,
+      "loss": 4.2633,
+      "loss/crossentropy": 1.9038777947425842,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19013714790344238,
+      "step": 12474
+    },
+    {
+      "epoch": 0.24952,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.008449045817057292,
+      "learning_rate": 0.0001,
+      "loss": 4.4785,
+      "loss/crossentropy": 2.1772372722625732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21470360457897186,
+      "step": 12476
+    },
+    {
+      "epoch": 0.24956,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.006605784098307292,
+      "learning_rate": 0.0001,
+      "loss": 4.3636,
+      "loss/crossentropy": 2.153541684150696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23221681267023087,
+      "step": 12478
+    },
+    {
+      "epoch": 0.2496,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.006668853759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4578,
+      "loss/crossentropy": 2.421363592147827,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20812640339136124,
+      "step": 12480
+    },
+    {
+      "epoch": 0.24964,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.008188629150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0628,
+      "loss/crossentropy": 1.8497061133384705,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18766893446445465,
+      "step": 12482
+    },
+    {
+      "epoch": 0.24968,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.008894602457682291,
+      "learning_rate": 0.0001,
+      "loss": 4.1011,
+      "loss/crossentropy": 1.9876007437705994,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1932743340730667,
+      "step": 12484
+    },
+    {
+      "epoch": 0.24972,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.007616170247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2722,
+      "loss/crossentropy": 2.4254921674728394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23315788805484772,
+      "step": 12486
+    },
+    {
+      "epoch": 0.24976,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0079010009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1012,
+      "loss/crossentropy": 1.983458697795868,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2190450206398964,
+      "step": 12488
+    },
+    {
+      "epoch": 0.2498,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.007111612955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0284,
+      "loss/crossentropy": 1.8051987886428833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19166412204504013,
+      "step": 12490
+    },
+    {
+      "epoch": 0.24984,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0052398681640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2385,
+      "loss/crossentropy": 2.1552056670188904,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2053346112370491,
+      "step": 12492
+    },
+    {
+      "epoch": 0.24988,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.004992421468098958,
+      "learning_rate": 0.0001,
+      "loss": 4.0157,
+      "loss/crossentropy": 1.9644648432731628,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17631876468658447,
+      "step": 12494
+    },
+    {
+      "epoch": 0.24992,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.006089019775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1626,
+      "loss/crossentropy": 2.027154862880707,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21671167761087418,
+      "step": 12496
+    },
+    {
+      "epoch": 0.24996,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.00628662109375,
+      "learning_rate": 0.0001,
+      "loss": 4.2485,
+      "loss/crossentropy": 2.2398791313171387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22507991641759872,
+      "step": 12498
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.006232706705729166,
+      "learning_rate": 0.0001,
+      "loss": 4.13,
+      "loss/crossentropy": 2.2342909574508667,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2192724049091339,
+      "step": 12500
+    },
+    {
+      "epoch": 0.25004,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0054443359375,
+      "learning_rate": 0.0001,
+      "loss": 4.2779,
+      "loss/crossentropy": 1.8621744513511658,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1882321536540985,
+      "step": 12502
+    },
+    {
+      "epoch": 0.25008,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.004988606770833333,
+      "learning_rate": 0.0001,
+      "loss": 4.4516,
+      "loss/crossentropy": 2.2012354135513306,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2128412127494812,
+      "step": 12504
+    },
+    {
+      "epoch": 0.25012,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.005873362223307292,
+      "learning_rate": 0.0001,
+      "loss": 4.0122,
+      "loss/crossentropy": 1.7629758715629578,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1928938776254654,
+      "step": 12506
+    },
+    {
+      "epoch": 0.25016,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.005730946858723958,
+      "learning_rate": 0.0001,
+      "loss": 4.1563,
+      "loss/crossentropy": 2.076514720916748,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20530561357736588,
+      "step": 12508
+    },
+    {
+      "epoch": 0.2502,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.006207021077473959,
+      "learning_rate": 0.0001,
+      "loss": 4.2005,
+      "loss/crossentropy": 2.004107654094696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21081873774528503,
+      "step": 12510
+    },
+    {
+      "epoch": 0.25024,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.008013661702473958,
+      "learning_rate": 0.0001,
+      "loss": 4.3323,
+      "loss/crossentropy": 2.232061505317688,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21327269077301025,
+      "step": 12512
+    },
+    {
+      "epoch": 0.25028,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0068662007649739586,
+      "learning_rate": 0.0001,
+      "loss": 4.1507,
+      "loss/crossentropy": 2.1506210565567017,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20797627419233322,
+      "step": 12514
+    },
+    {
+      "epoch": 0.25032,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008207194010416667,
+      "learning_rate": 0.0001,
+      "loss": 4.1999,
+      "loss/crossentropy": 2.084704279899597,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25306878983974457,
+      "step": 12516
+    },
+    {
+      "epoch": 0.25036,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.008137003580729166,
+      "learning_rate": 0.0001,
+      "loss": 4.3286,
+      "loss/crossentropy": 2.000536620616913,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20802763104438782,
+      "step": 12518
+    },
+    {
+      "epoch": 0.2504,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0088531494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1324,
+      "loss/crossentropy": 2.2391778230667114,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2171899899840355,
+      "step": 12520
+    },
+    {
+      "epoch": 0.25044,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.011375935872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.458,
+      "loss/crossentropy": 2.1465210914611816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2026364952325821,
+      "step": 12522
+    },
+    {
+      "epoch": 0.25048,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.011226145426432292,
+      "learning_rate": 0.0001,
+      "loss": 4.0804,
+      "loss/crossentropy": 1.8129625916481018,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17917531728744507,
+      "step": 12524
+    },
+    {
+      "epoch": 0.25052,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.010029856363932292,
+      "learning_rate": 0.0001,
+      "loss": 4.1832,
+      "loss/crossentropy": 1.98322331905365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20269985496997833,
+      "step": 12526
+    },
+    {
+      "epoch": 0.25056,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.008955637613932291,
+      "learning_rate": 0.0001,
+      "loss": 4.373,
+      "loss/crossentropy": 2.196588397026062,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23827942460775375,
+      "step": 12528
+    },
+    {
+      "epoch": 0.2506,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.009209950764973959,
+      "learning_rate": 0.0001,
+      "loss": 4.3811,
+      "loss/crossentropy": 2.183007001876831,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2071165144443512,
+      "step": 12530
+    },
+    {
+      "epoch": 0.25064,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0106353759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4205,
+      "loss/crossentropy": 2.2728021144866943,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21004344522953033,
+      "step": 12532
+    },
+    {
+      "epoch": 0.25068,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0111083984375,
+      "learning_rate": 0.0001,
+      "loss": 4.5478,
+      "loss/crossentropy": 2.280518889427185,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22367073595523834,
+      "step": 12534
+    },
+    {
+      "epoch": 0.25072,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.015038045247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.135,
+      "loss/crossentropy": 2.088103711605072,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2038320079445839,
+      "step": 12536
+    },
+    {
+      "epoch": 0.25076,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.012837727864583334,
+      "learning_rate": 0.0001,
+      "loss": 4.1798,
+      "loss/crossentropy": 1.646530568599701,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18982253968715668,
+      "step": 12538
+    },
+    {
+      "epoch": 0.2508,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.013622792561848958,
+      "learning_rate": 0.0001,
+      "loss": 4.2246,
+      "loss/crossentropy": 2.1743921041488647,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22835461795330048,
+      "step": 12540
+    },
+    {
+      "epoch": 0.25084,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0151123046875,
+      "learning_rate": 0.0001,
+      "loss": 4.3138,
+      "loss/crossentropy": 2.2216947078704834,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21978579461574554,
+      "step": 12542
+    },
+    {
+      "epoch": 0.25088,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.014890289306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.298,
+      "loss/crossentropy": 2.141101062297821,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21347828209400177,
+      "step": 12544
+    },
+    {
+      "epoch": 0.25092,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.014861806233723959,
+      "learning_rate": 0.0001,
+      "loss": 4.1534,
+      "loss/crossentropy": 2.1961969137191772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2192479968070984,
+      "step": 12546
+    },
+    {
+      "epoch": 0.25096,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.014274088541666667,
+      "learning_rate": 0.0001,
+      "loss": 3.9219,
+      "loss/crossentropy": 1.9632240533828735,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19182069599628448,
+      "step": 12548
+    },
+    {
+      "epoch": 0.251,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.02173639933268229,
+      "learning_rate": 0.0001,
+      "loss": 4.0594,
+      "loss/crossentropy": 1.7821694612503052,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24946660548448563,
+      "step": 12550
+    },
+    {
+      "epoch": 0.25104,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.016290028889973957,
+      "learning_rate": 0.0001,
+      "loss": 4.2423,
+      "loss/crossentropy": 2.261335611343384,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23269569873809814,
+      "step": 12552
+    },
+    {
+      "epoch": 0.25108,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.01762669881184896,
+      "learning_rate": 0.0001,
+      "loss": 3.9858,
+      "loss/crossentropy": 2.2365309596061707,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2119913324713707,
+      "step": 12554
+    },
+    {
+      "epoch": 0.25112,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0176025390625,
+      "learning_rate": 0.0001,
+      "loss": 3.9798,
+      "loss/crossentropy": 2.0685681104660034,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19985978305339813,
+      "step": 12556
+    },
+    {
+      "epoch": 0.25116,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0144683837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0231,
+      "loss/crossentropy": 2.0773105025291443,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2006489858031273,
+      "step": 12558
+    },
+    {
+      "epoch": 0.2512,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.01561279296875,
+      "learning_rate": 0.0001,
+      "loss": 4.1948,
+      "loss/crossentropy": 2.1145309805870056,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20449287444353104,
+      "step": 12560
+    },
+    {
+      "epoch": 0.25124,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.017097981770833333,
+      "learning_rate": 0.0001,
+      "loss": 4.0543,
+      "loss/crossentropy": 2.109993577003479,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20761344581842422,
+      "step": 12562
+    },
+    {
+      "epoch": 0.25128,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.016755167643229166,
+      "learning_rate": 0.0001,
+      "loss": 4.1661,
+      "loss/crossentropy": 1.9653338193893433,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2069985270500183,
+      "step": 12564
+    },
+    {
+      "epoch": 0.25132,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007045237223307291,
+      "learning_rate": 0.0001,
+      "loss": 4.101,
+      "loss/crossentropy": 1.7084832191467285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18922459334135056,
+      "step": 12566
+    },
+    {
+      "epoch": 0.25136,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.006494140625,
+      "learning_rate": 0.0001,
+      "loss": 3.9288,
+      "loss/crossentropy": 1.8007041215896606,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2122679501771927,
+      "step": 12568
+    },
+    {
+      "epoch": 0.2514,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.007972971598307291,
+      "learning_rate": 0.0001,
+      "loss": 4.3199,
+      "loss/crossentropy": 2.041845440864563,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2219957411289215,
+      "step": 12570
+    },
+    {
+      "epoch": 0.25144,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.008577219645182292,
+      "learning_rate": 0.0001,
+      "loss": 4.3703,
+      "loss/crossentropy": 2.2668861150741577,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2117021307349205,
+      "step": 12572
+    },
+    {
+      "epoch": 0.25148,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.010235341389973958,
+      "learning_rate": 0.0001,
+      "loss": 3.8462,
+      "loss/crossentropy": 1.8246251940727234,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1822521835565567,
+      "step": 12574
+    },
+    {
+      "epoch": 0.25152,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.010553995768229166,
+      "learning_rate": 0.0001,
+      "loss": 3.8137,
+      "loss/crossentropy": 2.0739742517471313,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2114427089691162,
+      "step": 12576
+    },
+    {
+      "epoch": 0.25156,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008906809488932292,
+      "learning_rate": 0.0001,
+      "loss": 3.9261,
+      "loss/crossentropy": 1.6614344120025635,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18680214881896973,
+      "step": 12578
+    },
+    {
+      "epoch": 0.2516,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.01279296875,
+      "learning_rate": 0.0001,
+      "loss": 4.4042,
+      "loss/crossentropy": 2.5629080533981323,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26097629219293594,
+      "step": 12580
+    },
+    {
+      "epoch": 0.25164,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.01337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1583,
+      "loss/crossentropy": 2.058123230934143,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2273598164319992,
+      "step": 12582
+    },
+    {
+      "epoch": 0.25168,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.014314524332682292,
+      "learning_rate": 0.0001,
+      "loss": 4.2802,
+      "loss/crossentropy": 1.9959335327148438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1991373971104622,
+      "step": 12584
+    },
+    {
+      "epoch": 0.25172,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.018293253580729165,
+      "learning_rate": 0.0001,
+      "loss": 3.9794,
+      "loss/crossentropy": 1.8165839314460754,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1888013407588005,
+      "step": 12586
+    },
+    {
+      "epoch": 0.25176,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.015221913655598959,
+      "learning_rate": 0.0001,
+      "loss": 4.0687,
+      "loss/crossentropy": 1.9918802976608276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18641646206378937,
+      "step": 12588
+    },
+    {
+      "epoch": 0.2518,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.013378651936848958,
+      "learning_rate": 0.0001,
+      "loss": 4.3443,
+      "loss/crossentropy": 2.0256036520004272,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19628985226154327,
+      "step": 12590
+    },
+    {
+      "epoch": 0.25184,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.01236572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1317,
+      "loss/crossentropy": 1.948347806930542,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19971590489149094,
+      "step": 12592
+    },
+    {
+      "epoch": 0.25188,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.012544759114583333,
+      "learning_rate": 0.0001,
+      "loss": 3.8528,
+      "loss/crossentropy": 2.2025747299194336,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2437238171696663,
+      "step": 12594
+    },
+    {
+      "epoch": 0.25192,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010872141520182291,
+      "learning_rate": 0.0001,
+      "loss": 4.2102,
+      "loss/crossentropy": 2.3209575414657593,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2366873100399971,
+      "step": 12596
+    },
+    {
+      "epoch": 0.25196,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.010846964518229167,
+      "learning_rate": 0.0001,
+      "loss": 3.8028,
+      "loss/crossentropy": 1.7978705763816833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19252559542655945,
+      "step": 12598
+    },
+    {
+      "epoch": 0.252,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010359446207682291,
+      "learning_rate": 0.0001,
+      "loss": 3.9269,
+      "loss/crossentropy": 1.964760184288025,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1943562552332878,
+      "step": 12600
+    },
+    {
+      "epoch": 0.25204,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.005399322509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2205,
+      "loss/crossentropy": 2.1420929431915283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.206790953874588,
+      "step": 12602
+    },
+    {
+      "epoch": 0.25208,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.008084869384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.5232,
+      "loss/crossentropy": 2.2422659397125244,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22513067722320557,
+      "step": 12604
+    },
+    {
+      "epoch": 0.25212,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.04026667277018229,
+      "learning_rate": 0.0001,
+      "loss": 3.8394,
+      "loss/crossentropy": 1.587006688117981,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1833646520972252,
+      "step": 12606
+    },
+    {
+      "epoch": 0.25216,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.03873291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.4461,
+      "loss/crossentropy": 2.426867365837097,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2435208261013031,
+      "step": 12608
+    },
+    {
+      "epoch": 0.2522,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.040185546875,
+      "learning_rate": 0.0001,
+      "loss": 4.0039,
+      "loss/crossentropy": 1.9733251333236694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21772438287734985,
+      "step": 12610
+    },
+    {
+      "epoch": 0.25224,
+      "grad_norm": 2.84375,
+      "grad_norm_var": 0.0764312744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2712,
+      "loss/crossentropy": 1.4835429191589355,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17013566195964813,
+      "step": 12612
+    },
+    {
+      "epoch": 0.25228,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.07500178019205729,
+      "learning_rate": 0.0001,
+      "loss": 3.8985,
+      "loss/crossentropy": 2.165693759918213,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21315700560808182,
+      "step": 12614
+    },
+    {
+      "epoch": 0.25232,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.07396647135416666,
+      "learning_rate": 0.0001,
+      "loss": 3.9856,
+      "loss/crossentropy": 2.206323266029358,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19913922995328903,
+      "step": 12616
+    },
+    {
+      "epoch": 0.25236,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.07515360514322916,
+      "learning_rate": 0.0001,
+      "loss": 4.0788,
+      "loss/crossentropy": 2.2520995140075684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22997721284627914,
+      "step": 12618
+    },
+    {
+      "epoch": 0.2524,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.07444559733072917,
+      "learning_rate": 0.0001,
+      "loss": 4.365,
+      "loss/crossentropy": 1.9322227239608765,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20600315928459167,
+      "step": 12620
+    },
+    {
+      "epoch": 0.25244,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.047459920247395836,
+      "learning_rate": 0.0001,
+      "loss": 4.0578,
+      "loss/crossentropy": 2.140220284461975,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21072514355182648,
+      "step": 12622
+    },
+    {
+      "epoch": 0.25248,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.04942118326822917,
+      "learning_rate": 0.0001,
+      "loss": 4.2316,
+      "loss/crossentropy": 2.061431884765625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2259845808148384,
+      "step": 12624
+    },
+    {
+      "epoch": 0.25252,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.047055816650390624,
+      "learning_rate": 0.0001,
+      "loss": 4.14,
+      "loss/crossentropy": 2.323551654815674,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23168490827083588,
+      "step": 12626
+    },
+    {
+      "epoch": 0.25256,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.008335113525390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0036,
+      "loss/crossentropy": 1.93999582529068,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21139459311962128,
+      "step": 12628
+    },
+    {
+      "epoch": 0.2526,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.011173248291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.207,
+      "loss/crossentropy": 1.9334313869476318,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21537292003631592,
+      "step": 12630
+    },
+    {
+      "epoch": 0.25264,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.010477447509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1674,
+      "loss/crossentropy": 2.145465135574341,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22080931067466736,
+      "step": 12632
+    },
+    {
+      "epoch": 0.25268,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.011668904622395834,
+      "learning_rate": 0.0001,
+      "loss": 4.0058,
+      "loss/crossentropy": 1.9708059430122375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2007170245051384,
+      "step": 12634
+    },
+    {
+      "epoch": 0.25272,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0115631103515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9649,
+      "loss/crossentropy": 1.9522746801376343,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20030274242162704,
+      "step": 12636
+    },
+    {
+      "epoch": 0.25276,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.012035878499348958,
+      "learning_rate": 0.0001,
+      "loss": 4.0416,
+      "loss/crossentropy": 1.856387436389923,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19685833156108856,
+      "step": 12638
+    },
+    {
+      "epoch": 0.2528,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.013787587483723959,
+      "learning_rate": 0.0001,
+      "loss": 4.475,
+      "loss/crossentropy": 1.9449518322944641,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20622673630714417,
+      "step": 12640
+    },
+    {
+      "epoch": 0.25284,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.017533365885416666,
+      "learning_rate": 0.0001,
+      "loss": 4.4683,
+      "loss/crossentropy": 2.4559473991394043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21476459503173828,
+      "step": 12642
+    },
+    {
+      "epoch": 0.25288,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.016778310139973957,
+      "learning_rate": 0.0001,
+      "loss": 4.3164,
+      "loss/crossentropy": 2.238897919654846,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22538839280605316,
+      "step": 12644
+    },
+    {
+      "epoch": 0.25292,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.013944498697916667,
+      "learning_rate": 0.0001,
+      "loss": 3.9801,
+      "loss/crossentropy": 1.7506417036056519,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18603645265102386,
+      "step": 12646
+    },
+    {
+      "epoch": 0.25296,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.014359283447265624,
+      "learning_rate": 0.0001,
+      "loss": 4.1238,
+      "loss/crossentropy": 2.162258505821228,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21168682724237442,
+      "step": 12648
+    },
+    {
+      "epoch": 0.253,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.015154774983723958,
+      "learning_rate": 0.0001,
+      "loss": 4.0441,
+      "loss/crossentropy": 1.9122841954231262,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20636393874883652,
+      "step": 12650
+    },
+    {
+      "epoch": 0.25304,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.015187327067057292,
+      "learning_rate": 0.0001,
+      "loss": 4.3566,
+      "loss/crossentropy": 2.1501541137695312,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22346170246601105,
+      "step": 12652
+    },
+    {
+      "epoch": 0.25308,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.01456298828125,
+      "learning_rate": 0.0001,
+      "loss": 4.0914,
+      "loss/crossentropy": 1.797228217124939,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17510685324668884,
+      "step": 12654
+    },
+    {
+      "epoch": 0.25312,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.010422515869140624,
+      "learning_rate": 0.0001,
+      "loss": 3.9232,
+      "loss/crossentropy": 1.7334046363830566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19618064910173416,
+      "step": 12656
+    },
+    {
+      "epoch": 0.25316,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0066912333170572914,
+      "learning_rate": 0.0001,
+      "loss": 4.3287,
+      "loss/crossentropy": 1.99091237783432,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21426931023597717,
+      "step": 12658
+    },
+    {
+      "epoch": 0.2532,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0069244384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.5036,
+      "loss/crossentropy": 2.1908310651779175,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20360572636127472,
+      "step": 12660
+    },
+    {
+      "epoch": 0.25324,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.00787353515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9746,
+      "loss/crossentropy": 1.7925593852996826,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18203188478946686,
+      "step": 12662
+    },
+    {
+      "epoch": 0.25328,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009006500244140625,
+      "learning_rate": 0.0001,
+      "loss": 3.8955,
+      "loss/crossentropy": 1.9464862942695618,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1896686628460884,
+      "step": 12664
+    },
+    {
+      "epoch": 0.25332,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.006379954020182292,
+      "learning_rate": 0.0001,
+      "loss": 3.9922,
+      "loss/crossentropy": 2.0207581520080566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20436276495456696,
+      "step": 12666
+    },
+    {
+      "epoch": 0.25336,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.005322011311848959,
+      "learning_rate": 0.0001,
+      "loss": 4.141,
+      "loss/crossentropy": 1.9915068745613098,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21265029907226562,
+      "step": 12668
+    },
+    {
+      "epoch": 0.2534,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.005204010009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2602,
+      "loss/crossentropy": 1.787190020084381,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20195094496011734,
+      "step": 12670
+    },
+    {
+      "epoch": 0.25344,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.005036417643229167,
+      "learning_rate": 0.0001,
+      "loss": 4.3554,
+      "loss/crossentropy": 2.2403881549835205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2313152700662613,
+      "step": 12672
+    },
+    {
+      "epoch": 0.25348,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.00540771484375,
+      "learning_rate": 0.0001,
+      "loss": 3.9895,
+      "loss/crossentropy": 1.6122660636901855,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19166506081819534,
+      "step": 12674
+    },
+    {
+      "epoch": 0.25352,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.005402628580729167,
+      "learning_rate": 0.0001,
+      "loss": 4.1966,
+      "loss/crossentropy": 2.0758888125419617,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19904160499572754,
+      "step": 12676
+    },
+    {
+      "epoch": 0.25356,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.0064656575520833336,
+      "learning_rate": 0.0001,
+      "loss": 4.2466,
+      "loss/crossentropy": 1.8765565156936646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17384758591651917,
+      "step": 12678
+    },
+    {
+      "epoch": 0.2536,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.0054705301920572914,
+      "learning_rate": 0.0001,
+      "loss": 4.0977,
+      "loss/crossentropy": 2.1322853565216064,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21987880766391754,
+      "step": 12680
+    },
+    {
+      "epoch": 0.25364,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.006196848551432292,
+      "learning_rate": 0.0001,
+      "loss": 3.9862,
+      "loss/crossentropy": 2.0303893089294434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20912615954875946,
+      "step": 12682
+    },
+    {
+      "epoch": 0.25368,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0065826416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0022,
+      "loss/crossentropy": 2.2314319610595703,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22904963046312332,
+      "step": 12684
+    },
+    {
+      "epoch": 0.25372,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.007438151041666666,
+      "learning_rate": 0.0001,
+      "loss": 4.1854,
+      "loss/crossentropy": 1.9751350283622742,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20683745294809341,
+      "step": 12686
+    },
+    {
+      "epoch": 0.25376,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0088043212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.112,
+      "loss/crossentropy": 1.9198334217071533,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19134660065174103,
+      "step": 12688
+    },
+    {
+      "epoch": 0.2538,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.009214019775390625,
+      "learning_rate": 0.0001,
+      "loss": 3.8783,
+      "loss/crossentropy": 1.945135474205017,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20948659628629684,
+      "step": 12690
+    },
+    {
+      "epoch": 0.25384,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0094879150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1018,
+      "loss/crossentropy": 2.146402955055237,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21819791197776794,
+      "step": 12692
+    },
+    {
+      "epoch": 0.25388,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.007806142171223958,
+      "learning_rate": 0.0001,
+      "loss": 4.3696,
+      "loss/crossentropy": 2.309106230735779,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2160014659166336,
+      "step": 12694
+    },
+    {
+      "epoch": 0.25392,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.007616170247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0637,
+      "loss/crossentropy": 2.1726107597351074,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20791994035243988,
+      "step": 12696
+    },
+    {
+      "epoch": 0.25396,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.007370758056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.252,
+      "loss/crossentropy": 2.0905996561050415,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2083379328250885,
+      "step": 12698
+    },
+    {
+      "epoch": 0.254,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.008185831705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.197,
+      "loss/crossentropy": 1.9190585017204285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1856430396437645,
+      "step": 12700
+    },
+    {
+      "epoch": 0.25404,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.007877604166666666,
+      "learning_rate": 0.0001,
+      "loss": 4.3906,
+      "loss/crossentropy": 2.2493419647216797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20197859406471252,
+      "step": 12702
+    },
+    {
+      "epoch": 0.25408,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.007731119791666667,
+      "learning_rate": 0.0001,
+      "loss": 4.1935,
+      "loss/crossentropy": 2.0205613374710083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2155362293124199,
+      "step": 12704
+    },
+    {
+      "epoch": 0.25412,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.006257120768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.188,
+      "loss/crossentropy": 1.8853323459625244,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20137012749910355,
+      "step": 12706
+    },
+    {
+      "epoch": 0.25416,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0061431884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.098,
+      "loss/crossentropy": 2.109869122505188,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2036309316754341,
+      "step": 12708
+    },
+    {
+      "epoch": 0.2542,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.005890909830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.1782,
+      "loss/crossentropy": 2.0598954558372498,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22793744504451752,
+      "step": 12710
+    },
+    {
+      "epoch": 0.25424,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0055735270182291664,
+      "learning_rate": 0.0001,
+      "loss": 4.2984,
+      "loss/crossentropy": 2.169256567955017,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20931441336870193,
+      "step": 12712
+    },
+    {
+      "epoch": 0.25428,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.005197906494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2715,
+      "loss/crossentropy": 1.8178632855415344,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20344894379377365,
+      "step": 12714
+    },
+    {
+      "epoch": 0.25432,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.004327138264973958,
+      "learning_rate": 0.0001,
+      "loss": 4.0897,
+      "loss/crossentropy": 1.6468743085861206,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16709627211093903,
+      "step": 12716
+    },
+    {
+      "epoch": 0.25436,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.004748280843098958,
+      "learning_rate": 0.0001,
+      "loss": 4.1702,
+      "loss/crossentropy": 1.8353837728500366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20837673544883728,
+      "step": 12718
+    },
+    {
+      "epoch": 0.2544,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.005492146809895833,
+      "learning_rate": 0.0001,
+      "loss": 3.8056,
+      "loss/crossentropy": 1.7881666421890259,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18411727994680405,
+      "step": 12720
+    },
+    {
+      "epoch": 0.25444,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0065958658854166664,
+      "learning_rate": 0.0001,
+      "loss": 4.5369,
+      "loss/crossentropy": 2.19934618473053,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22607161849737167,
+      "step": 12722
+    },
+    {
+      "epoch": 0.25448,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.008821360270182292,
+      "learning_rate": 0.0001,
+      "loss": 4.5237,
+      "loss/crossentropy": 1.9789779782295227,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2156950756907463,
+      "step": 12724
+    },
+    {
+      "epoch": 0.25452,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.010778554280598958,
+      "learning_rate": 0.0001,
+      "loss": 4.1059,
+      "loss/crossentropy": 2.0605525970458984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20622699707746506,
+      "step": 12726
+    },
+    {
+      "epoch": 0.25456,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.011336263020833333,
+      "learning_rate": 0.0001,
+      "loss": 4.2862,
+      "loss/crossentropy": 2.1131407022476196,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21070127934217453,
+      "step": 12728
+    },
+    {
+      "epoch": 0.2546,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.011336263020833333,
+      "learning_rate": 0.0001,
+      "loss": 4.3789,
+      "loss/crossentropy": 1.9708096981048584,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22219926118850708,
+      "step": 12730
+    },
+    {
+      "epoch": 0.25464,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.011918131510416667,
+      "learning_rate": 0.0001,
+      "loss": 4.2985,
+      "loss/crossentropy": 2.220608353614807,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2271113097667694,
+      "step": 12732
+    },
+    {
+      "epoch": 0.25468,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.01146240234375,
+      "learning_rate": 0.0001,
+      "loss": 4.2097,
+      "loss/crossentropy": 2.535509705543518,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24400310218334198,
+      "step": 12734
+    },
+    {
+      "epoch": 0.25472,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.015952301025390626,
+      "learning_rate": 0.0001,
+      "loss": 4.6837,
+      "loss/crossentropy": 2.3619974851608276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2268039956688881,
+      "step": 12736
+    },
+    {
+      "epoch": 0.25476,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0168212890625,
+      "learning_rate": 0.0001,
+      "loss": 3.9929,
+      "loss/crossentropy": 2.0095953941345215,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20850248634815216,
+      "step": 12738
+    },
+    {
+      "epoch": 0.2548,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.012276204427083333,
+      "learning_rate": 0.0001,
+      "loss": 4.1466,
+      "loss/crossentropy": 1.934277892112732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2116999626159668,
+      "step": 12740
+    },
+    {
+      "epoch": 0.25484,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011701456705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.336,
+      "loss/crossentropy": 2.237201452255249,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22638197988271713,
+      "step": 12742
+    },
+    {
+      "epoch": 0.25488,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.011464182535807292,
+      "learning_rate": 0.0001,
+      "loss": 4.0354,
+      "loss/crossentropy": 2.302059292793274,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22285569459199905,
+      "step": 12744
+    },
+    {
+      "epoch": 0.25492,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.013944244384765625,
+      "learning_rate": 0.0001,
+      "loss": 3.7097,
+      "loss/crossentropy": 1.7690886855125427,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19209770113229752,
+      "step": 12746
+    },
+    {
+      "epoch": 0.25496,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.013944244384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2324,
+      "loss/crossentropy": 2.4372475147247314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2192872166633606,
+      "step": 12748
+    },
+    {
+      "epoch": 0.255,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.016405232747395835,
+      "learning_rate": 0.0001,
+      "loss": 4.2627,
+      "loss/crossentropy": 2.12698757648468,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21698793768882751,
+      "step": 12750
+    },
+    {
+      "epoch": 0.25504,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.007710520426432292,
+      "learning_rate": 0.0001,
+      "loss": 3.9991,
+      "loss/crossentropy": 1.989953875541687,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1969183310866356,
+      "step": 12752
+    },
+    {
+      "epoch": 0.25508,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.034501139322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.5505,
+      "loss/crossentropy": 2.2911970615386963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21863195300102234,
+      "step": 12754
+    },
+    {
+      "epoch": 0.25512,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.03460286458333333,
+      "learning_rate": 0.0001,
+      "loss": 4.0915,
+      "loss/crossentropy": 1.8447460532188416,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20167672634124756,
+      "step": 12756
+    },
+    {
+      "epoch": 0.25516,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.035471343994140626,
+      "learning_rate": 0.0001,
+      "loss": 3.8488,
+      "loss/crossentropy": 2.043856382369995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2075670212507248,
+      "step": 12758
+    },
+    {
+      "epoch": 0.2552,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.037536366780598955,
+      "learning_rate": 0.0001,
+      "loss": 4.0893,
+      "loss/crossentropy": 1.735123872756958,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19347237050533295,
+      "step": 12760
+    },
+    {
+      "epoch": 0.25524,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.034407552083333334,
+      "learning_rate": 0.0001,
+      "loss": 4.0586,
+      "loss/crossentropy": 2.1058340072631836,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21859320253133774,
+      "step": 12762
+    },
+    {
+      "epoch": 0.25528,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.03444722493489583,
+      "learning_rate": 0.0001,
+      "loss": 4.2518,
+      "loss/crossentropy": 2.0583395957946777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.224592886865139,
+      "step": 12764
+    },
+    {
+      "epoch": 0.25532,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.032206217447916664,
+      "learning_rate": 0.0001,
+      "loss": 4.3562,
+      "loss/crossentropy": 2.2550116777420044,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23644013702869415,
+      "step": 12766
+    },
+    {
+      "epoch": 0.25536,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.032293446858723956,
+      "learning_rate": 0.0001,
+      "loss": 4.2338,
+      "loss/crossentropy": 2.463193655014038,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2229352444410324,
+      "step": 12768
+    },
+    {
+      "epoch": 0.2554,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.006400299072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.016,
+      "loss/crossentropy": 2.010310709476471,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20725534856319427,
+      "step": 12770
+    },
+    {
+      "epoch": 0.25544,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.006219228108723958,
+      "learning_rate": 0.0001,
+      "loss": 4.1988,
+      "loss/crossentropy": 2.2002042531967163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21686428785324097,
+      "step": 12772
+    },
+    {
+      "epoch": 0.25548,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.009690093994140624,
+      "learning_rate": 0.0001,
+      "loss": 4.2974,
+      "loss/crossentropy": 2.072916865348816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2210928201675415,
+      "step": 12774
+    },
+    {
+      "epoch": 0.25552,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010149892171223958,
+      "learning_rate": 0.0001,
+      "loss": 4.3168,
+      "loss/crossentropy": 2.068341016769409,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20776809751987457,
+      "step": 12776
+    },
+    {
+      "epoch": 0.25556,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.010949452718098959,
+      "learning_rate": 0.0001,
+      "loss": 3.8552,
+      "loss/crossentropy": 1.6145030856132507,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2056947946548462,
+      "step": 12778
+    },
+    {
+      "epoch": 0.2556,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011311848958333334,
+      "learning_rate": 0.0001,
+      "loss": 4.0132,
+      "loss/crossentropy": 2.1529496908187866,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21802741289138794,
+      "step": 12780
+    },
+    {
+      "epoch": 0.25564,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011226399739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.1708,
+      "loss/crossentropy": 2.255640387535095,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21794818341732025,
+      "step": 12782
+    },
+    {
+      "epoch": 0.25568,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011546834309895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1102,
+      "loss/crossentropy": 1.9934669137001038,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21103205531835556,
+      "step": 12784
+    },
+    {
+      "epoch": 0.25572,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.011864217122395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2924,
+      "loss/crossentropy": 1.9404807090759277,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22771108150482178,
+      "step": 12786
+    },
+    {
+      "epoch": 0.25576,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.012084706624348959,
+      "learning_rate": 0.0001,
+      "loss": 4.0444,
+      "loss/crossentropy": 1.7803818583488464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19245180487632751,
+      "step": 12788
+    },
+    {
+      "epoch": 0.2558,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009423828125,
+      "learning_rate": 0.0001,
+      "loss": 4.2033,
+      "loss/crossentropy": 2.108216881752014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2239757925271988,
+      "step": 12790
+    },
+    {
+      "epoch": 0.25584,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0059478759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1999,
+      "loss/crossentropy": 2.056196451187134,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20573420077562332,
+      "step": 12792
+    },
+    {
+      "epoch": 0.25588,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.007275390625,
+      "learning_rate": 0.0001,
+      "loss": 3.9749,
+      "loss/crossentropy": 2.2215099334716797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21107079833745956,
+      "step": 12794
+    },
+    {
+      "epoch": 0.25592,
+      "grad_norm": 1.796875,
+      "grad_norm_var": 0.009281158447265625,
+      "learning_rate": 0.0001,
+      "loss": 3.9527,
+      "loss/crossentropy": 1.947695553302765,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19738437235355377,
+      "step": 12796
+    },
+    {
+      "epoch": 0.25596,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010204823811848958,
+      "learning_rate": 0.0001,
+      "loss": 4.1072,
+      "loss/crossentropy": 2.3351560831069946,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21714143455028534,
+      "step": 12798
+    },
+    {
+      "epoch": 0.256,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009798177083333333,
+      "learning_rate": 0.0001,
+      "loss": 4.3055,
+      "loss/crossentropy": 2.399898648262024,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25758983194828033,
+      "step": 12800
+    },
+    {
+      "epoch": 0.25604,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.03814697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.5632,
+      "loss/crossentropy": 2.3537880182266235,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2976065129041672,
+      "step": 12802
+    },
+    {
+      "epoch": 0.25608,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.03857421875,
+      "learning_rate": 0.0001,
+      "loss": 3.8394,
+      "loss/crossentropy": 1.7765586972236633,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1823679357767105,
+      "step": 12804
+    },
+    {
+      "epoch": 0.25612,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.039793904622395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4023,
+      "loss/crossentropy": 2.119332432746887,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23052836954593658,
+      "step": 12806
+    },
+    {
+      "epoch": 0.25616,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0414947509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0251,
+      "loss/crossentropy": 1.8325288891792297,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19811426103115082,
+      "step": 12808
+    },
+    {
+      "epoch": 0.2562,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.039896392822265626,
+      "learning_rate": 0.0001,
+      "loss": 4.1089,
+      "loss/crossentropy": 2.079294800758362,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21122989803552628,
+      "step": 12810
+    },
+    {
+      "epoch": 0.25624,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.03581110636393229,
+      "learning_rate": 0.0001,
+      "loss": 4.0105,
+      "loss/crossentropy": 2.029142141342163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2008090615272522,
+      "step": 12812
+    },
+    {
+      "epoch": 0.25628,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.03578058878580729,
+      "learning_rate": 0.0001,
+      "loss": 4.1743,
+      "loss/crossentropy": 1.8697097301483154,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19589021801948547,
+      "step": 12814
+    },
+    {
+      "epoch": 0.25632,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.038331858317057294,
+      "learning_rate": 0.0001,
+      "loss": 3.8837,
+      "loss/crossentropy": 1.8488793969154358,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19999422132968903,
+      "step": 12816
+    },
+    {
+      "epoch": 0.25636,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011946360270182291,
+      "learning_rate": 0.0001,
+      "loss": 4.236,
+      "loss/crossentropy": 2.352132201194763,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23569129407405853,
+      "step": 12818
+    },
+    {
+      "epoch": 0.2564,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010155232747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2247,
+      "loss/crossentropy": 2.213895559310913,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19782276451587677,
+      "step": 12820
+    },
+    {
+      "epoch": 0.25644,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.008876291910807292,
+      "learning_rate": 0.0001,
+      "loss": 4.205,
+      "loss/crossentropy": 2.328965425491333,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2138378769159317,
+      "step": 12822
+    },
+    {
+      "epoch": 0.25648,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.007407379150390625,
+      "learning_rate": 0.0001,
+      "loss": 3.9178,
+      "loss/crossentropy": 2.148880124092102,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20276722311973572,
+      "step": 12824
+    },
+    {
+      "epoch": 0.25652,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.007124582926432292,
+      "learning_rate": 0.0001,
+      "loss": 4.2742,
+      "loss/crossentropy": 2.1533660888671875,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22211932390928268,
+      "step": 12826
+    },
+    {
+      "epoch": 0.25656,
+      "grad_norm": 2.75,
+      "grad_norm_var": 0.04098078409830729,
+      "learning_rate": 0.0001,
+      "loss": 4.0928,
+      "loss/crossentropy": 1.7518101930618286,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17689460515975952,
+      "step": 12828
+    },
+    {
+      "epoch": 0.2566,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0395660400390625,
+      "learning_rate": 0.0001,
+      "loss": 3.9745,
+      "loss/crossentropy": 2.346145749092102,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22157911956310272,
+      "step": 12830
+    },
+    {
+      "epoch": 0.25664,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0358062744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1689,
+      "loss/crossentropy": 1.8355774283409119,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20288754999637604,
+      "step": 12832
+    },
+    {
+      "epoch": 0.25668,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.03906962076822917,
+      "learning_rate": 0.0001,
+      "loss": 3.9947,
+      "loss/crossentropy": 1.949233889579773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19860410690307617,
+      "step": 12834
+    },
+    {
+      "epoch": 0.25672,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.04248046875,
+      "learning_rate": 0.0001,
+      "loss": 4.0903,
+      "loss/crossentropy": 2.0141645669937134,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19531898200511932,
+      "step": 12836
+    },
+    {
+      "epoch": 0.25676,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.04277725219726562,
+      "learning_rate": 0.0001,
+      "loss": 3.918,
+      "loss/crossentropy": 1.6573863625526428,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.163545623421669,
+      "step": 12838
+    },
+    {
+      "epoch": 0.2568,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.04273656209309896,
+      "learning_rate": 0.0001,
+      "loss": 4.3252,
+      "loss/crossentropy": 2.2174651622772217,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23083333671092987,
+      "step": 12840
+    },
+    {
+      "epoch": 0.25684,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.045967356363932295,
+      "learning_rate": 0.0001,
+      "loss": 3.7593,
+      "loss/crossentropy": 1.8067168593406677,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1963481903076172,
+      "step": 12842
+    },
+    {
+      "epoch": 0.25688,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010536448160807291,
+      "learning_rate": 0.0001,
+      "loss": 3.936,
+      "loss/crossentropy": 1.928274691104889,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21783485263586044,
+      "step": 12844
+    },
+    {
+      "epoch": 0.25692,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.013498687744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0984,
+      "loss/crossentropy": 2.183669090270996,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2040836587548256,
+      "step": 12846
+    },
+    {
+      "epoch": 0.25696,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.012894439697265624,
+      "learning_rate": 0.0001,
+      "loss": 4.0624,
+      "loss/crossentropy": 1.7395422458648682,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19434035569429398,
+      "step": 12848
+    },
+    {
+      "epoch": 0.257,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.012772623697916667,
+      "learning_rate": 0.0001,
+      "loss": 3.9914,
+      "loss/crossentropy": 1.9124428629875183,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19825156033039093,
+      "step": 12850
+    },
+    {
+      "epoch": 0.25704,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.011606597900390625,
+      "learning_rate": 0.0001,
+      "loss": 3.9325,
+      "loss/crossentropy": 1.8064388036727905,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19374487549066544,
+      "step": 12852
+    },
+    {
+      "epoch": 0.25708,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0111724853515625,
+      "learning_rate": 0.0001,
+      "loss": 3.8722,
+      "loss/crossentropy": 2.0943931341171265,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20997963845729828,
+      "step": 12854
+    },
+    {
+      "epoch": 0.25712,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.00948486328125,
+      "learning_rate": 0.0001,
+      "loss": 3.9974,
+      "loss/crossentropy": 1.9252876043319702,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20608004182577133,
+      "step": 12856
+    },
+    {
+      "epoch": 0.25716,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007575480143229166,
+      "learning_rate": 0.0001,
+      "loss": 3.9536,
+      "loss/crossentropy": 1.813286304473877,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2107577547430992,
+      "step": 12858
+    },
+    {
+      "epoch": 0.2572,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.007405598958333333,
+      "learning_rate": 0.0001,
+      "loss": 4.1571,
+      "loss/crossentropy": 1.985919713973999,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18724270164966583,
+      "step": 12860
+    },
+    {
+      "epoch": 0.25724,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.004654693603515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9377,
+      "loss/crossentropy": 1.912338137626648,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19602636992931366,
+      "step": 12862
+    },
+    {
+      "epoch": 0.25728,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.004743448893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0823,
+      "loss/crossentropy": 1.962310791015625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1975855752825737,
+      "step": 12864
+    },
+    {
+      "epoch": 0.25732,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0050046284993489586,
+      "learning_rate": 0.0001,
+      "loss": 4.285,
+      "loss/crossentropy": 2.063133656978607,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20012972503900528,
+      "step": 12866
+    },
+    {
+      "epoch": 0.25736,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.005751291910807292,
+      "learning_rate": 0.0001,
+      "loss": 4.1443,
+      "loss/crossentropy": 2.1755728721618652,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21089357137680054,
+      "step": 12868
+    },
+    {
+      "epoch": 0.2574,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.006148274739583333,
+      "learning_rate": 0.0001,
+      "loss": 4.3854,
+      "loss/crossentropy": 2.21665620803833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22525488585233688,
+      "step": 12870
+    },
+    {
+      "epoch": 0.25744,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.006959788004557292,
+      "learning_rate": 0.0001,
+      "loss": 4.2465,
+      "loss/crossentropy": 2.2826790809631348,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22989312559366226,
+      "step": 12872
+    },
+    {
+      "epoch": 0.25748,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.008432769775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0753,
+      "loss/crossentropy": 2.3187366724014282,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22048770636320114,
+      "step": 12874
+    },
+    {
+      "epoch": 0.25752,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0093017578125,
+      "learning_rate": 0.0001,
+      "loss": 4.2469,
+      "loss/crossentropy": 2.234878957271576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19875259697437286,
+      "step": 12876
+    },
+    {
+      "epoch": 0.25756,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.007897694905598959,
+      "learning_rate": 0.0001,
+      "loss": 4.0981,
+      "loss/crossentropy": 1.8919751644134521,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20682695508003235,
+      "step": 12878
+    },
+    {
+      "epoch": 0.2576,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.008861287434895834,
+      "learning_rate": 0.0001,
+      "loss": 3.9564,
+      "loss/crossentropy": 2.135833740234375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2100691795349121,
+      "step": 12880
+    },
+    {
+      "epoch": 0.25764,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.008421834309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.163,
+      "loss/crossentropy": 2.043276846408844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20327290892601013,
+      "step": 12882
+    },
+    {
+      "epoch": 0.25768,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.0077512105305989586,
+      "learning_rate": 0.0001,
+      "loss": 4.2439,
+      "loss/crossentropy": 1.996088445186615,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20499806851148605,
+      "step": 12884
+    },
+    {
+      "epoch": 0.25772,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.006648508707682291,
+      "learning_rate": 0.0001,
+      "loss": 3.9821,
+      "loss/crossentropy": 1.7864345908164978,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19354674220085144,
+      "step": 12886
+    },
+    {
+      "epoch": 0.25776,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0051310221354166664,
+      "learning_rate": 0.0001,
+      "loss": 3.961,
+      "loss/crossentropy": 1.7890866994857788,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1806325614452362,
+      "step": 12888
+    },
+    {
+      "epoch": 0.2578,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.004938761393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1836,
+      "loss/crossentropy": 1.9287649989128113,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19555142521858215,
+      "step": 12890
+    },
+    {
+      "epoch": 0.25784,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.005081939697265625,
+      "learning_rate": 0.0001,
+      "loss": 3.9151,
+      "loss/crossentropy": 1.930963397026062,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19455663859844208,
+      "step": 12892
+    },
+    {
+      "epoch": 0.25788,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.007551829020182292,
+      "learning_rate": 0.0001,
+      "loss": 4.3216,
+      "loss/crossentropy": 2.1105018854141235,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22564146667718887,
+      "step": 12894
+    },
+    {
+      "epoch": 0.25792,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008365885416666666,
+      "learning_rate": 0.0001,
+      "loss": 4.1924,
+      "loss/crossentropy": 2.0535677671432495,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2083517387509346,
+      "step": 12896
+    },
+    {
+      "epoch": 0.25796,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007966105143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2951,
+      "loss/crossentropy": 2.062114655971527,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20135939121246338,
+      "step": 12898
+    },
+    {
+      "epoch": 0.258,
+      "grad_norm": 1.7734375,
+      "grad_norm_var": 0.012482706705729167,
+      "learning_rate": 0.0001,
+      "loss": 3.8187,
+      "loss/crossentropy": 1.968269407749176,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20238329470157623,
+      "step": 12900
+    },
+    {
+      "epoch": 0.25804,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0598541259765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1899,
+      "loss/crossentropy": 2.0783804655075073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23015306890010834,
+      "step": 12902
+    },
+    {
+      "epoch": 0.25808,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.05815836588541667,
+      "learning_rate": 0.0001,
+      "loss": 4.1245,
+      "loss/crossentropy": 1.8616145253181458,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19802623987197876,
+      "step": 12904
+    },
+    {
+      "epoch": 0.25812,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.05834147135416667,
+      "learning_rate": 0.0001,
+      "loss": 4.2293,
+      "loss/crossentropy": 2.0005027651786804,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21621856093406677,
+      "step": 12906
+    },
+    {
+      "epoch": 0.25816,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.06181208292643229,
+      "learning_rate": 0.0001,
+      "loss": 4.0563,
+      "loss/crossentropy": 1.796087920665741,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18594232201576233,
+      "step": 12908
+    },
+    {
+      "epoch": 0.2582,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.06258316040039062,
+      "learning_rate": 0.0001,
+      "loss": 4.3161,
+      "loss/crossentropy": 2.003196358680725,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19296320527791977,
+      "step": 12910
+    },
+    {
+      "epoch": 0.25824,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.06313654581705729,
+      "learning_rate": 0.0001,
+      "loss": 4.0489,
+      "loss/crossentropy": 1.8252119421958923,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1879737451672554,
+      "step": 12912
+    },
+    {
+      "epoch": 0.25828,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0629900614420573,
+      "learning_rate": 0.0001,
+      "loss": 4.2053,
+      "loss/crossentropy": 1.9801989793777466,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20867998152971268,
+      "step": 12914
+    },
+    {
+      "epoch": 0.25832,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.05583902994791667,
+      "learning_rate": 0.0001,
+      "loss": 3.9754,
+      "loss/crossentropy": 2.04214608669281,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21134068816900253,
+      "step": 12916
+    },
+    {
+      "epoch": 0.25836,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.009159342447916666,
+      "learning_rate": 0.0001,
+      "loss": 4.2202,
+      "loss/crossentropy": 2.0669824481010437,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20436694473028183,
+      "step": 12918
+    },
+    {
+      "epoch": 0.2584,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.008771769205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3616,
+      "loss/crossentropy": 1.951616883277893,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19572293758392334,
+      "step": 12920
+    },
+    {
+      "epoch": 0.25844,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.010074869791666666,
+      "learning_rate": 0.0001,
+      "loss": 3.8476,
+      "loss/crossentropy": 1.7768954634666443,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1930573582649231,
+      "step": 12922
+    },
+    {
+      "epoch": 0.25848,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.020458984375,
+      "learning_rate": 0.0001,
+      "loss": 3.8775,
+      "loss/crossentropy": 1.8527125716209412,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18682826310396194,
+      "step": 12924
+    },
+    {
+      "epoch": 0.25852,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.018990071614583333,
+      "learning_rate": 0.0001,
+      "loss": 4.3814,
+      "loss/crossentropy": 2.3325828313827515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21535523980855942,
+      "step": 12926
+    },
+    {
+      "epoch": 0.25856,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.02063166300455729,
+      "learning_rate": 0.0001,
+      "loss": 3.7701,
+      "loss/crossentropy": 1.759222686290741,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17470692098140717,
+      "step": 12928
+    },
+    {
+      "epoch": 0.2586,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.021247355143229167,
+      "learning_rate": 0.0001,
+      "loss": 3.8065,
+      "loss/crossentropy": 1.944493055343628,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20259656757116318,
+      "step": 12930
+    },
+    {
+      "epoch": 0.25864,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.05238825480143229,
+      "learning_rate": 0.0001,
+      "loss": 4.133,
+      "loss/crossentropy": 2.086555302143097,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2180738002061844,
+      "step": 12932
+    },
+    {
+      "epoch": 0.25868,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.052779134114583334,
+      "learning_rate": 0.0001,
+      "loss": 4.1414,
+      "loss/crossentropy": 1.9723011255264282,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22814874351024628,
+      "step": 12934
+    },
+    {
+      "epoch": 0.25872,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.05356038411458333,
+      "learning_rate": 0.0001,
+      "loss": 4.418,
+      "loss/crossentropy": 1.5646896958351135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18182098120450974,
+      "step": 12936
+    },
+    {
+      "epoch": 0.25876,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.054870351155598955,
+      "learning_rate": 0.0001,
+      "loss": 3.8356,
+      "loss/crossentropy": 1.8396940231323242,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19390598684549332,
+      "step": 12938
+    },
+    {
+      "epoch": 0.2588,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.041112263997395836,
+      "learning_rate": 0.0001,
+      "loss": 4.5902,
+      "loss/crossentropy": 2.086738705635071,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.209588885307312,
+      "step": 12940
+    },
+    {
+      "epoch": 0.25884,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.04136530558268229,
+      "learning_rate": 0.0001,
+      "loss": 4.2611,
+      "loss/crossentropy": 2.0965282917022705,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21021046489477158,
+      "step": 12942
+    },
+    {
+      "epoch": 0.25888,
+      "grad_norm": 1.8125,
+      "grad_norm_var": 0.04311421712239583,
+      "learning_rate": 0.0001,
+      "loss": 4.0702,
+      "loss/crossentropy": 2.2878576517105103,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2203715667128563,
+      "step": 12944
+    },
+    {
+      "epoch": 0.25892,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.045446523030598956,
+      "learning_rate": 0.0001,
+      "loss": 4.419,
+      "loss/crossentropy": 2.1753373742103577,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21585986018180847,
+      "step": 12946
+    },
+    {
+      "epoch": 0.25896,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.036232248942057295,
+      "learning_rate": 0.0001,
+      "loss": 4.3735,
+      "loss/crossentropy": 1.963489055633545,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20374132692813873,
+      "step": 12948
+    },
+    {
+      "epoch": 0.259,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.037751261393229166,
+      "learning_rate": 0.0001,
+      "loss": 4.4073,
+      "loss/crossentropy": 2.3734259605407715,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2375379502773285,
+      "step": 12950
+    },
+    {
+      "epoch": 0.25904,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.040710194905598955,
+      "learning_rate": 0.0001,
+      "loss": 4.184,
+      "loss/crossentropy": 2.266390085220337,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22274205088615417,
+      "step": 12952
+    },
+    {
+      "epoch": 0.25908,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.03729654947916667,
+      "learning_rate": 0.0001,
+      "loss": 3.9735,
+      "loss/crossentropy": 1.9015297293663025,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20370282977819443,
+      "step": 12954
+    },
+    {
+      "epoch": 0.25912,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.03886617024739583,
+      "learning_rate": 0.0001,
+      "loss": 4.3269,
+      "loss/crossentropy": 2.1192378997802734,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26288172602653503,
+      "step": 12956
+    },
+    {
+      "epoch": 0.25916,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.03792088826497396,
+      "learning_rate": 0.0001,
+      "loss": 4.0982,
+      "loss/crossentropy": 1.713826835155487,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18557386100292206,
+      "step": 12958
+    },
+    {
+      "epoch": 0.2592,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.029545084635416666,
+      "learning_rate": 0.0001,
+      "loss": 4.2002,
+      "loss/crossentropy": 2.0365039706230164,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21086978912353516,
+      "step": 12960
+    },
+    {
+      "epoch": 0.25924,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.027852376302083332,
+      "learning_rate": 0.0001,
+      "loss": 4.2422,
+      "loss/crossentropy": 1.6235400438308716,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18112218379974365,
+      "step": 12962
+    },
+    {
+      "epoch": 0.25928,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0116455078125,
+      "learning_rate": 0.0001,
+      "loss": 4.2996,
+      "loss/crossentropy": 2.042620003223419,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21463964879512787,
+      "step": 12964
+    },
+    {
+      "epoch": 0.25932,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.009611002604166667,
+      "learning_rate": 0.0001,
+      "loss": 4.0222,
+      "loss/crossentropy": 1.6238983273506165,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2171783745288849,
+      "step": 12966
+    },
+    {
+      "epoch": 0.25936,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.014070383707682292,
+      "learning_rate": 0.0001,
+      "loss": 4.0098,
+      "loss/crossentropy": 2.3150339126586914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22038520872592926,
+      "step": 12968
+    },
+    {
+      "epoch": 0.2594,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.013301595052083334,
+      "learning_rate": 0.0001,
+      "loss": 4.2202,
+      "loss/crossentropy": 2.3907227516174316,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2269512265920639,
+      "step": 12970
+    },
+    {
+      "epoch": 0.25944,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.014378865559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1032,
+      "loss/crossentropy": 2.1069165468215942,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21007738262414932,
+      "step": 12972
+    },
+    {
+      "epoch": 0.25948,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.017942047119140624,
+      "learning_rate": 0.0001,
+      "loss": 3.9475,
+      "loss/crossentropy": 2.277982234954834,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21795185655355453,
+      "step": 12974
+    },
+    {
+      "epoch": 0.25952,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.017775217692057293,
+      "learning_rate": 0.0001,
+      "loss": 4.0159,
+      "loss/crossentropy": 1.7227254509925842,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19050486385822296,
+      "step": 12976
+    },
+    {
+      "epoch": 0.25956,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0184722900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0281,
+      "loss/crossentropy": 2.0112340450286865,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21032612770795822,
+      "step": 12978
+    },
+    {
+      "epoch": 0.2596,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.019530232747395834,
+      "learning_rate": 0.0001,
+      "loss": 4.3041,
+      "loss/crossentropy": 2.4241796731948853,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22474834322929382,
+      "step": 12980
+    },
+    {
+      "epoch": 0.25964,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.019710286458333334,
+      "learning_rate": 0.0001,
+      "loss": 4.3814,
+      "loss/crossentropy": 1.893187701702118,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20330534130334854,
+      "step": 12982
+    },
+    {
+      "epoch": 0.25968,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.008467356363932291,
+      "learning_rate": 0.0001,
+      "loss": 4.1527,
+      "loss/crossentropy": 2.045100212097168,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20636773109436035,
+      "step": 12984
+    },
+    {
+      "epoch": 0.25972,
+      "grad_norm": 1.8125,
+      "grad_norm_var": 0.010035959879557292,
+      "learning_rate": 0.0001,
+      "loss": 3.864,
+      "loss/crossentropy": 1.819455087184906,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1775035709142685,
+      "step": 12986
+    },
+    {
+      "epoch": 0.25976,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.00985107421875,
+      "learning_rate": 0.0001,
+      "loss": 4.0553,
+      "loss/crossentropy": 2.1358631253242493,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2027900367975235,
+      "step": 12988
+    },
+    {
+      "epoch": 0.2598,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.010902659098307291,
+      "learning_rate": 0.0001,
+      "loss": 4.1552,
+      "loss/crossentropy": 1.7847145199775696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21203257888555527,
+      "step": 12990
+    },
+    {
+      "epoch": 0.25984,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011979166666666667,
+      "learning_rate": 0.0001,
+      "loss": 4.325,
+      "loss/crossentropy": 2.2848687171936035,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22337154299020767,
+      "step": 12992
+    },
+    {
+      "epoch": 0.25988,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.012889607747395834,
+      "learning_rate": 0.0001,
+      "loss": 3.975,
+      "loss/crossentropy": 1.9629716277122498,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20660604536533356,
+      "step": 12994
+    },
+    {
+      "epoch": 0.25992,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.010461171468098959,
+      "learning_rate": 0.0001,
+      "loss": 4.1521,
+      "loss/crossentropy": 2.0321004390716553,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21755626797676086,
+      "step": 12996
+    },
+    {
+      "epoch": 0.25996,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.0105712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1208,
+      "loss/crossentropy": 1.950038492679596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1987697035074234,
+      "step": 12998
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010544586181640624,
+      "learning_rate": 0.0001,
+      "loss": 4.3383,
+      "loss/crossentropy": 2.2752585411071777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20959648489952087,
+      "step": 13000
+    },
+    {
+      "epoch": 0.26004,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.008138020833333334,
+      "learning_rate": 0.0001,
+      "loss": 4.3132,
+      "loss/crossentropy": 2.162104368209839,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21731911599636078,
+      "step": 13002
+    },
+    {
+      "epoch": 0.26008,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.009102121988932291,
+      "learning_rate": 0.0001,
+      "loss": 4.0138,
+      "loss/crossentropy": 2.3739081025123596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22912397980690002,
+      "step": 13004
+    },
+    {
+      "epoch": 0.26012,
+      "grad_norm": 3.0625,
+      "grad_norm_var": 0.07608413696289062,
+      "learning_rate": 0.0001,
+      "loss": 4.3052,
+      "loss/crossentropy": 2.161367416381836,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22413842380046844,
+      "step": 13006
+    },
+    {
+      "epoch": 0.26016,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.07587661743164062,
+      "learning_rate": 0.0001,
+      "loss": 4.1435,
+      "loss/crossentropy": 1.9589285850524902,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21598373353481293,
+      "step": 13008
+    },
+    {
+      "epoch": 0.2602,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0731842041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3268,
+      "loss/crossentropy": 2.3472602367401123,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22626091539859772,
+      "step": 13010
+    },
+    {
+      "epoch": 0.26024,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.07296727498372396,
+      "learning_rate": 0.0001,
+      "loss": 3.8455,
+      "loss/crossentropy": 1.9968576431274414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20777586847543716,
+      "step": 13012
+    },
+    {
+      "epoch": 0.26028,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0729400634765625,
+      "learning_rate": 0.0001,
+      "loss": 3.8219,
+      "loss/crossentropy": 1.7597955465316772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19565816968679428,
+      "step": 13014
+    },
+    {
+      "epoch": 0.26032,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0735992431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1446,
+      "loss/crossentropy": 1.8115187287330627,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19675451517105103,
+      "step": 13016
+    },
+    {
+      "epoch": 0.26036,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.07507909138997396,
+      "learning_rate": 0.0001,
+      "loss": 4.2018,
+      "loss/crossentropy": 1.9728147983551025,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19999082386493683,
+      "step": 13018
+    },
+    {
+      "epoch": 0.2604,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.07516988118489583,
+      "learning_rate": 0.0001,
+      "loss": 4.1434,
+      "loss/crossentropy": 2.155986785888672,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20160435140132904,
+      "step": 13020
+    },
+    {
+      "epoch": 0.26044,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0081298828125,
+      "learning_rate": 0.0001,
+      "loss": 4.0647,
+      "loss/crossentropy": 2.119171440601349,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2231176346540451,
+      "step": 13022
+    },
+    {
+      "epoch": 0.26048,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.010884348551432292,
+      "learning_rate": 0.0001,
+      "loss": 3.8185,
+      "loss/crossentropy": 2.0395994186401367,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21075501292943954,
+      "step": 13024
+    },
+    {
+      "epoch": 0.26052,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.014808909098307291,
+      "learning_rate": 0.0001,
+      "loss": 4.4988,
+      "loss/crossentropy": 2.42897891998291,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24882060289382935,
+      "step": 13026
+    },
+    {
+      "epoch": 0.26056,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.018202463785807293,
+      "learning_rate": 0.0001,
+      "loss": 4.4995,
+      "loss/crossentropy": 2.367077350616455,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21996428817510605,
+      "step": 13028
+    },
+    {
+      "epoch": 0.2606,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.017207590738932292,
+      "learning_rate": 0.0001,
+      "loss": 4.0349,
+      "loss/crossentropy": 2.071397542953491,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20548687875270844,
+      "step": 13030
+    },
+    {
+      "epoch": 0.26064,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.016943359375,
+      "learning_rate": 0.0001,
+      "loss": 4.0425,
+      "loss/crossentropy": 1.764617681503296,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19541263580322266,
+      "step": 13032
+    },
+    {
+      "epoch": 0.26068,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.016916656494140626,
+      "learning_rate": 0.0001,
+      "loss": 4.2657,
+      "loss/crossentropy": 2.2966678142547607,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2187207117676735,
+      "step": 13034
+    },
+    {
+      "epoch": 0.26072,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0159820556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2959,
+      "loss/crossentropy": 1.923275649547577,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20180264115333557,
+      "step": 13036
+    },
+    {
+      "epoch": 0.26076,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.017207845052083334,
+      "learning_rate": 0.0001,
+      "loss": 4.3138,
+      "loss/crossentropy": 2.246683716773987,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22133737802505493,
+      "step": 13038
+    },
+    {
+      "epoch": 0.2608,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.011755116780598958,
+      "learning_rate": 0.0001,
+      "loss": 4.3685,
+      "loss/crossentropy": 2.271313190460205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23316439241170883,
+      "step": 13040
+    },
+    {
+      "epoch": 0.26084,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.008652496337890624,
+      "learning_rate": 0.0001,
+      "loss": 4.4451,
+      "loss/crossentropy": 2.2649654150009155,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.211005300283432,
+      "step": 13042
+    },
+    {
+      "epoch": 0.26088,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.008006795247395834,
+      "learning_rate": 0.0001,
+      "loss": 3.9505,
+      "loss/crossentropy": 1.7872197031974792,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17795034497976303,
+      "step": 13044
+    },
+    {
+      "epoch": 0.26092,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.007875315348307292,
+      "learning_rate": 0.0001,
+      "loss": 3.8693,
+      "loss/crossentropy": 1.793078064918518,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18739672750234604,
+      "step": 13046
+    },
+    {
+      "epoch": 0.26096,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0111572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4178,
+      "loss/crossentropy": 2.0676932334899902,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2156887650489807,
+      "step": 13048
+    },
+    {
+      "epoch": 0.261,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.013484700520833334,
+      "learning_rate": 0.0001,
+      "loss": 4.2851,
+      "loss/crossentropy": 2.2497498989105225,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21906063705682755,
+      "step": 13050
+    },
+    {
+      "epoch": 0.26104,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.014994303385416666,
+      "learning_rate": 0.0001,
+      "loss": 4.0203,
+      "loss/crossentropy": 1.901672899723053,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2070387825369835,
+      "step": 13052
+    },
+    {
+      "epoch": 0.26108,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.013887532552083333,
+      "learning_rate": 0.0001,
+      "loss": 4.3772,
+      "loss/crossentropy": 2.2334693670272827,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21539244055747986,
+      "step": 13054
+    },
+    {
+      "epoch": 0.26112,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.014288075764973958,
+      "learning_rate": 0.0001,
+      "loss": 4.2916,
+      "loss/crossentropy": 2.0297399759292603,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18854249268770218,
+      "step": 13056
+    },
+    {
+      "epoch": 0.26116,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.014452107747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1115,
+      "loss/crossentropy": 2.0964863896369934,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21821416169404984,
+      "step": 13058
+    },
+    {
+      "epoch": 0.2612,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.013702138264973959,
+      "learning_rate": 0.0001,
+      "loss": 4.4918,
+      "loss/crossentropy": 2.1526511907577515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22606930136680603,
+      "step": 13060
+    },
+    {
+      "epoch": 0.26124,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0101470947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2079,
+      "loss/crossentropy": 2.2275509238243103,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21959830820560455,
+      "step": 13062
+    },
+    {
+      "epoch": 0.26128,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0086181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4246,
+      "loss/crossentropy": 2.542263627052307,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22889885306358337,
+      "step": 13064
+    },
+    {
+      "epoch": 0.26132,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.008585611979166666,
+      "learning_rate": 0.0001,
+      "loss": 4.0292,
+      "loss/crossentropy": 1.7915868163108826,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20242593437433243,
+      "step": 13066
+    },
+    {
+      "epoch": 0.26136,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.006453450520833333,
+      "learning_rate": 0.0001,
+      "loss": 4.1415,
+      "loss/crossentropy": 2.183286130428314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.227259561419487,
+      "step": 13068
+    },
+    {
+      "epoch": 0.2614,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0063720703125,
+      "learning_rate": 0.0001,
+      "loss": 4.3211,
+      "loss/crossentropy": 2.2707515954971313,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2300432324409485,
+      "step": 13070
+    },
+    {
+      "epoch": 0.26144,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0066487630208333336,
+      "learning_rate": 0.0001,
+      "loss": 4.0237,
+      "loss/crossentropy": 1.822485864162445,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19380664825439453,
+      "step": 13072
+    },
+    {
+      "epoch": 0.26148,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.008349355061848958,
+      "learning_rate": 0.0001,
+      "loss": 4.0915,
+      "loss/crossentropy": 2.089439034461975,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22531338036060333,
+      "step": 13074
+    },
+    {
+      "epoch": 0.26152,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.008318837483723958,
+      "learning_rate": 0.0001,
+      "loss": 4.3872,
+      "loss/crossentropy": 2.0921221375465393,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22256288677453995,
+      "step": 13076
+    },
+    {
+      "epoch": 0.26156,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.009065500895182292,
+      "learning_rate": 0.0001,
+      "loss": 4.4644,
+      "loss/crossentropy": 1.8978914022445679,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2677152305841446,
+      "step": 13078
+    },
+    {
+      "epoch": 0.2616,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.011643218994140624,
+      "learning_rate": 0.0001,
+      "loss": 4.405,
+      "loss/crossentropy": 2.2540348768234253,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24195441603660583,
+      "step": 13080
+    },
+    {
+      "epoch": 0.26164,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.01092529296875,
+      "learning_rate": 0.0001,
+      "loss": 4.0965,
+      "loss/crossentropy": 1.9546288847923279,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19464369863271713,
+      "step": 13082
+    },
+    {
+      "epoch": 0.26168,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.012457021077473958,
+      "learning_rate": 0.0001,
+      "loss": 4.0103,
+      "loss/crossentropy": 2.1527568101882935,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22238580137491226,
+      "step": 13084
+    },
+    {
+      "epoch": 0.26172,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0128814697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3957,
+      "loss/crossentropy": 2.111591637134552,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21360620856285095,
+      "step": 13086
+    },
+    {
+      "epoch": 0.26176,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.01339111328125,
+      "learning_rate": 0.0001,
+      "loss": 3.99,
+      "loss/crossentropy": 2.006071925163269,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19746285676956177,
+      "step": 13088
+    },
+    {
+      "epoch": 0.2618,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0128814697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2073,
+      "loss/crossentropy": 2.276778221130371,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21222709119319916,
+      "step": 13090
+    },
+    {
+      "epoch": 0.26184,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.013346354166666666,
+      "learning_rate": 0.0001,
+      "loss": 4.0679,
+      "loss/crossentropy": 2.102404534816742,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2049730271100998,
+      "step": 13092
+    },
+    {
+      "epoch": 0.26188,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.013719685872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.439,
+      "loss/crossentropy": 2.204727053642273,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21260947734117508,
+      "step": 13094
+    },
+    {
+      "epoch": 0.26192,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.010184478759765626,
+      "learning_rate": 0.0001,
+      "loss": 4.2801,
+      "loss/crossentropy": 2.3762032985687256,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22091203182935715,
+      "step": 13096
+    },
+    {
+      "epoch": 0.26196,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.010286458333333333,
+      "learning_rate": 0.0001,
+      "loss": 4.3472,
+      "loss/crossentropy": 2.0710391998291016,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20782632380723953,
+      "step": 13098
+    },
+    {
+      "epoch": 0.262,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.009319814046223958,
+      "learning_rate": 0.0001,
+      "loss": 4.3257,
+      "loss/crossentropy": 2.193024158477783,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21109049022197723,
+      "step": 13100
+    },
+    {
+      "epoch": 0.26204,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009308878580729167,
+      "learning_rate": 0.0001,
+      "loss": 4.1863,
+      "loss/crossentropy": 1.9668392539024353,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19249404221773148,
+      "step": 13102
+    },
+    {
+      "epoch": 0.26208,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.00858154296875,
+      "learning_rate": 0.0001,
+      "loss": 4.3122,
+      "loss/crossentropy": 2.2675795555114746,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26278799772262573,
+      "step": 13104
+    },
+    {
+      "epoch": 0.26212,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.007857004801432291,
+      "learning_rate": 0.0001,
+      "loss": 3.9962,
+      "loss/crossentropy": 2.0119330883026123,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20821017771959305,
+      "step": 13106
+    },
+    {
+      "epoch": 0.26216,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.006237538655598959,
+      "learning_rate": 0.0001,
+      "loss": 4.329,
+      "loss/crossentropy": 2.1259734630584717,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22454438358545303,
+      "step": 13108
+    },
+    {
+      "epoch": 0.2622,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0043413798014322914,
+      "learning_rate": 0.0001,
+      "loss": 4.2558,
+      "loss/crossentropy": 1.8091979622840881,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19353152066469193,
+      "step": 13110
+    },
+    {
+      "epoch": 0.26224,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0038655598958333335,
+      "learning_rate": 0.0001,
+      "loss": 4.1847,
+      "loss/crossentropy": 1.9463382363319397,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20932665467262268,
+      "step": 13112
+    },
+    {
+      "epoch": 0.26228,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.004366048177083333,
+      "learning_rate": 0.0001,
+      "loss": 4.3975,
+      "loss/crossentropy": 2.475601077079773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23451963067054749,
+      "step": 13114
+    },
+    {
+      "epoch": 0.26232,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.004412587483723958,
+      "learning_rate": 0.0001,
+      "loss": 4.4397,
+      "loss/crossentropy": 2.0833849906921387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22380392253398895,
+      "step": 13116
+    },
+    {
+      "epoch": 0.26236,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.005301920572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.0408,
+      "loss/crossentropy": 1.9876453876495361,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20823375135660172,
+      "step": 13118
+    },
+    {
+      "epoch": 0.2624,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.006750233968098958,
+      "learning_rate": 0.0001,
+      "loss": 4.5321,
+      "loss/crossentropy": 2.184763789176941,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24357233941555023,
+      "step": 13120
+    },
+    {
+      "epoch": 0.26244,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0061187744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3552,
+      "loss/crossentropy": 2.217886805534363,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21944888681173325,
+      "step": 13122
+    },
+    {
+      "epoch": 0.26248,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0052886962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2443,
+      "loss/crossentropy": 2.3299126625061035,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2158309668302536,
+      "step": 13124
+    },
+    {
+      "epoch": 0.26252,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0058349609375,
+      "learning_rate": 0.0001,
+      "loss": 4.1303,
+      "loss/crossentropy": 2.0423209071159363,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20835570991039276,
+      "step": 13126
+    },
+    {
+      "epoch": 0.26256,
+      "grad_norm": 1.796875,
+      "grad_norm_var": 0.012325032552083334,
+      "learning_rate": 0.0001,
+      "loss": 4.31,
+      "loss/crossentropy": 2.227811813354492,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2043633684515953,
+      "step": 13128
+    },
+    {
+      "epoch": 0.2626,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.011153157552083333,
+      "learning_rate": 0.0001,
+      "loss": 4.1312,
+      "loss/crossentropy": 2.105396568775177,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2212589681148529,
+      "step": 13130
+    },
+    {
+      "epoch": 0.26264,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.012737782796223958,
+      "learning_rate": 0.0001,
+      "loss": 4.2244,
+      "loss/crossentropy": 2.1711814999580383,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21528291702270508,
+      "step": 13132
+    },
+    {
+      "epoch": 0.26268,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.011937459309895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1288,
+      "loss/crossentropy": 2.0086284279823303,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1974012851715088,
+      "step": 13134
+    },
+    {
+      "epoch": 0.26272,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.011205037434895834,
+      "learning_rate": 0.0001,
+      "loss": 3.895,
+      "loss/crossentropy": 1.7025322914123535,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1878160759806633,
+      "step": 13136
+    },
+    {
+      "epoch": 0.26276,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.015950520833333332,
+      "learning_rate": 0.0001,
+      "loss": 4.0134,
+      "loss/crossentropy": 1.821410596370697,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.197972871363163,
+      "step": 13138
+    },
+    {
+      "epoch": 0.2628,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.018993886311848958,
+      "learning_rate": 0.0001,
+      "loss": 4.0875,
+      "loss/crossentropy": 2.0485053658485413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20401400327682495,
+      "step": 13140
+    },
+    {
+      "epoch": 0.26284,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.018808746337890626,
+      "learning_rate": 0.0001,
+      "loss": 4.4344,
+      "loss/crossentropy": 2.350088357925415,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2638252303004265,
+      "step": 13142
+    },
+    {
+      "epoch": 0.26288,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.012308502197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.5756,
+      "loss/crossentropy": 2.3423168659210205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22735413163900375,
+      "step": 13144
+    },
+    {
+      "epoch": 0.26292,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.012033843994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3848,
+      "loss/crossentropy": 2.134036064147949,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20407382398843765,
+      "step": 13146
+    },
+    {
+      "epoch": 0.26296,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.009748331705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.105,
+      "loss/crossentropy": 2.0744638442993164,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2107524871826172,
+      "step": 13148
+    },
+    {
+      "epoch": 0.263,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.012604777018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0507,
+      "loss/crossentropy": 2.1522003412246704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2046833261847496,
+      "step": 13150
+    },
+    {
+      "epoch": 0.26304,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.013224283854166666,
+      "learning_rate": 0.0001,
+      "loss": 4.2754,
+      "loss/crossentropy": 2.328645348548889,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24340055882930756,
+      "step": 13152
+    },
+    {
+      "epoch": 0.26308,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.009161122639973958,
+      "learning_rate": 0.0001,
+      "loss": 4.0502,
+      "loss/crossentropy": 2.0878008008003235,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2110893502831459,
+      "step": 13154
+    },
+    {
+      "epoch": 0.26312,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0074155171712239586,
+      "learning_rate": 0.0001,
+      "loss": 4.3065,
+      "loss/crossentropy": 1.9381126761436462,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18893951922655106,
+      "step": 13156
+    },
+    {
+      "epoch": 0.26316,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.009364573160807292,
+      "learning_rate": 0.0001,
+      "loss": 4.0254,
+      "loss/crossentropy": 2.153562545776367,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1973574310541153,
+      "step": 13158
+    },
+    {
+      "epoch": 0.2632,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.011370595296223958,
+      "learning_rate": 0.0001,
+      "loss": 3.998,
+      "loss/crossentropy": 2.0583202242851257,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19830025732517242,
+      "step": 13160
+    },
+    {
+      "epoch": 0.26324,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.012117258707682292,
+      "learning_rate": 0.0001,
+      "loss": 4.4843,
+      "loss/crossentropy": 2.1951998472213745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21916336566209793,
+      "step": 13162
+    },
+    {
+      "epoch": 0.26328,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.011888631184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4944,
+      "loss/crossentropy": 2.1612678170204163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21040990203619003,
+      "step": 13164
+    },
+    {
+      "epoch": 0.26332,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009415690104166667,
+      "learning_rate": 0.0001,
+      "loss": 4.2012,
+      "loss/crossentropy": 1.9372112154960632,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24868668615818024,
+      "step": 13166
+    },
+    {
+      "epoch": 0.26336,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.009968058268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.3605,
+      "loss/crossentropy": 2.0220513939857483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22029200196266174,
+      "step": 13168
+    },
+    {
+      "epoch": 0.2634,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.010355631510416666,
+      "learning_rate": 0.0001,
+      "loss": 3.7251,
+      "loss/crossentropy": 1.6749334335327148,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18502884358167648,
+      "step": 13170
+    },
+    {
+      "epoch": 0.26344,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.010096995035807292,
+      "learning_rate": 0.0001,
+      "loss": 4.1394,
+      "loss/crossentropy": 2.1463050842285156,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21098337322473526,
+      "step": 13172
+    },
+    {
+      "epoch": 0.26348,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.009089914957682292,
+      "learning_rate": 0.0001,
+      "loss": 4.3261,
+      "loss/crossentropy": 2.20473051071167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23006527870893478,
+      "step": 13174
+    },
+    {
+      "epoch": 0.26352,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.012068684895833333,
+      "learning_rate": 0.0001,
+      "loss": 4.3289,
+      "loss/crossentropy": 2.097456157207489,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2077179104089737,
+      "step": 13176
+    },
+    {
+      "epoch": 0.26356,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0124176025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.402,
+      "loss/crossentropy": 2.136400580406189,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2152952179312706,
+      "step": 13178
+    },
+    {
+      "epoch": 0.2636,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.013451131184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.23,
+      "loss/crossentropy": 2.0791231393814087,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19688992202281952,
+      "step": 13180
+    },
+    {
+      "epoch": 0.26364,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0126617431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1233,
+      "loss/crossentropy": 2.135149598121643,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22163429111242294,
+      "step": 13182
+    },
+    {
+      "epoch": 0.26368,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.013936360677083334,
+      "learning_rate": 0.0001,
+      "loss": 4.4215,
+      "loss/crossentropy": 2.3207671642303467,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24321655184030533,
+      "step": 13184
+    },
+    {
+      "epoch": 0.26372,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.013069407145182291,
+      "learning_rate": 0.0001,
+      "loss": 4.0995,
+      "loss/crossentropy": 2.0136974453926086,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19274070858955383,
+      "step": 13186
+    },
+    {
+      "epoch": 0.26376,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.013610585530598959,
+      "learning_rate": 0.0001,
+      "loss": 4.5094,
+      "loss/crossentropy": 2.3542726039886475,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24547156691551208,
+      "step": 13188
+    },
+    {
+      "epoch": 0.2638,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.016007486979166666,
+      "learning_rate": 0.0001,
+      "loss": 4.068,
+      "loss/crossentropy": 1.905708134174347,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18261709809303284,
+      "step": 13190
+    },
+    {
+      "epoch": 0.26384,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.012147776285807292,
+      "learning_rate": 0.0001,
+      "loss": 4.1426,
+      "loss/crossentropy": 2.137321710586548,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2052912563085556,
+      "step": 13192
+    },
+    {
+      "epoch": 0.26388,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010827382405598959,
+      "learning_rate": 0.0001,
+      "loss": 4.2252,
+      "loss/crossentropy": 2.083053410053253,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2038491740822792,
+      "step": 13194
+    },
+    {
+      "epoch": 0.26392,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.011346181233723959,
+      "learning_rate": 0.0001,
+      "loss": 3.8349,
+      "loss/crossentropy": 1.8006438612937927,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1910744607448578,
+      "step": 13196
+    },
+    {
+      "epoch": 0.26396,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.012902577718098959,
+      "learning_rate": 0.0001,
+      "loss": 4.4104,
+      "loss/crossentropy": 2.323628544807434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2280128449201584,
+      "step": 13198
+    },
+    {
+      "epoch": 0.264,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.010137685139973958,
+      "learning_rate": 0.0001,
+      "loss": 4.1231,
+      "loss/crossentropy": 1.9763594269752502,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19776062667369843,
+      "step": 13200
+    },
+    {
+      "epoch": 0.26404,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.010601552327473958,
+      "learning_rate": 0.0001,
+      "loss": 4.0175,
+      "loss/crossentropy": 1.8776894807815552,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19732315093278885,
+      "step": 13202
+    },
+    {
+      "epoch": 0.26408,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.007478841145833333,
+      "learning_rate": 0.0001,
+      "loss": 4.0713,
+      "loss/crossentropy": 1.826314091682434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19153253734111786,
+      "step": 13204
+    },
+    {
+      "epoch": 0.26412,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.007673136393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1503,
+      "loss/crossentropy": 2.435014486312866,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22149913012981415,
+      "step": 13206
+    },
+    {
+      "epoch": 0.26416,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.01866633097330729,
+      "learning_rate": 0.0001,
+      "loss": 4.0829,
+      "loss/crossentropy": 2.055279493331909,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21625792980194092,
+      "step": 13208
+    },
+    {
+      "epoch": 0.2642,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.018641916910807292,
+      "learning_rate": 0.0001,
+      "loss": 4.216,
+      "loss/crossentropy": 2.4019049406051636,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22626115381717682,
+      "step": 13210
+    },
+    {
+      "epoch": 0.26424,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.017508697509765626,
+      "learning_rate": 0.0001,
+      "loss": 3.9822,
+      "loss/crossentropy": 2.336918354034424,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22773776203393936,
+      "step": 13212
+    },
+    {
+      "epoch": 0.26428,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.01608454386393229,
+      "learning_rate": 0.0001,
+      "loss": 4.3929,
+      "loss/crossentropy": 2.0398528575897217,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20827088505029678,
+      "step": 13214
+    },
+    {
+      "epoch": 0.26432,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.015419260660807291,
+      "learning_rate": 0.0001,
+      "loss": 4.026,
+      "loss/crossentropy": 2.166857123374939,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2136707603931427,
+      "step": 13216
+    },
+    {
+      "epoch": 0.26436,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.015095011393229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2282,
+      "loss/crossentropy": 1.8454258441925049,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20678912848234177,
+      "step": 13218
+    },
+    {
+      "epoch": 0.2644,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.014642079671223959,
+      "learning_rate": 0.0001,
+      "loss": 4.2966,
+      "loss/crossentropy": 2.249597668647766,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22785719484090805,
+      "step": 13220
+    },
+    {
+      "epoch": 0.26444,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.014134724934895834,
+      "learning_rate": 0.0001,
+      "loss": 3.875,
+      "loss/crossentropy": 1.7729946374893188,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2014644593000412,
+      "step": 13222
+    },
+    {
+      "epoch": 0.26448,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.004976145426432292,
+      "learning_rate": 0.0001,
+      "loss": 3.9144,
+      "loss/crossentropy": 1.7968116998672485,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1889382004737854,
+      "step": 13224
+    },
+    {
+      "epoch": 0.26452,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.006510162353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2015,
+      "loss/crossentropy": 2.0765512585639954,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20551903545856476,
+      "step": 13226
+    },
+    {
+      "epoch": 0.26456,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.007933553059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2603,
+      "loss/crossentropy": 2.2007554173469543,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2250915989279747,
+      "step": 13228
+    },
+    {
+      "epoch": 0.2646,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.008184560139973958,
+      "learning_rate": 0.0001,
+      "loss": 4.2014,
+      "loss/crossentropy": 2.319425046443939,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22913866490125656,
+      "step": 13230
+    },
+    {
+      "epoch": 0.26464,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.008365631103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1395,
+      "loss/crossentropy": 1.9845139980316162,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20769847929477692,
+      "step": 13232
+    },
+    {
+      "epoch": 0.26468,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.008674112955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0985,
+      "loss/crossentropy": 1.94157075881958,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2314443141222,
+      "step": 13234
+    },
+    {
+      "epoch": 0.26472,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.007470703125,
+      "learning_rate": 0.0001,
+      "loss": 4.3428,
+      "loss/crossentropy": 2.0699294209480286,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22432812303304672,
+      "step": 13236
+    },
+    {
+      "epoch": 0.26476,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.006841786702473958,
+      "learning_rate": 0.0001,
+      "loss": 3.983,
+      "loss/crossentropy": 1.7656533122062683,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20837824791669846,
+      "step": 13238
+    },
+    {
+      "epoch": 0.2648,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0060618082682291664,
+      "learning_rate": 0.0001,
+      "loss": 4.1799,
+      "loss/crossentropy": 1.9473227858543396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20151428878307343,
+      "step": 13240
+    },
+    {
+      "epoch": 0.26484,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.005387369791666667,
+      "learning_rate": 0.0001,
+      "loss": 4.284,
+      "loss/crossentropy": 2.060440719127655,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2041771411895752,
+      "step": 13242
+    },
+    {
+      "epoch": 0.26488,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.003922526041666667,
+      "learning_rate": 0.0001,
+      "loss": 3.7503,
+      "loss/crossentropy": 2.0440456867218018,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2041776031255722,
+      "step": 13244
+    },
+    {
+      "epoch": 0.26492,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.004622395833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.1921,
+      "loss/crossentropy": 1.9192551374435425,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18886462599039078,
+      "step": 13246
+    },
+    {
+      "epoch": 0.26496,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.0072100321451822914,
+      "learning_rate": 0.0001,
+      "loss": 4.1673,
+      "loss/crossentropy": 1.7112281918525696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18532422930002213,
+      "step": 13248
+    },
+    {
+      "epoch": 0.265,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.007637532552083334,
+      "learning_rate": 0.0001,
+      "loss": 4.0561,
+      "loss/crossentropy": 1.9225355386734009,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19994951784610748,
+      "step": 13250
+    },
+    {
+      "epoch": 0.26504,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.007950846354166667,
+      "learning_rate": 0.0001,
+      "loss": 3.6825,
+      "loss/crossentropy": 2.0243517756462097,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19785288721323013,
+      "step": 13252
+    },
+    {
+      "epoch": 0.26508,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.007413482666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2929,
+      "loss/crossentropy": 2.255920171737671,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21220286190509796,
+      "step": 13254
+    },
+    {
+      "epoch": 0.26512,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.008475748697916667,
+      "learning_rate": 0.0001,
+      "loss": 4.4145,
+      "loss/crossentropy": 2.3024203777313232,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22255335003137589,
+      "step": 13256
+    },
+    {
+      "epoch": 0.26516,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.00733642578125,
+      "learning_rate": 0.0001,
+      "loss": 4.1386,
+      "loss/crossentropy": 2.0467708110809326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20193494111299515,
+      "step": 13258
+    },
+    {
+      "epoch": 0.2652,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.006064605712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2188,
+      "loss/crossentropy": 2.1543976068496704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2257647141814232,
+      "step": 13260
+    },
+    {
+      "epoch": 0.26524,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.006373850504557291,
+      "learning_rate": 0.0001,
+      "loss": 4.2683,
+      "loss/crossentropy": 2.077743351459503,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21964918076992035,
+      "step": 13262
+    },
+    {
+      "epoch": 0.26528,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0044921875,
+      "learning_rate": 0.0001,
+      "loss": 3.963,
+      "loss/crossentropy": 1.6356236338615417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1847509667277336,
+      "step": 13264
+    },
+    {
+      "epoch": 0.26532,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.006925201416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3559,
+      "loss/crossentropy": 2.0876659750938416,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22080402821302414,
+      "step": 13266
+    },
+    {
+      "epoch": 0.26536,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.005747222900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0452,
+      "loss/crossentropy": 1.7057855129241943,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18772340565919876,
+      "step": 13268
+    },
+    {
+      "epoch": 0.2654,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.0078277587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0032,
+      "loss/crossentropy": 2.0550594329833984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18500665575265884,
+      "step": 13270
+    },
+    {
+      "epoch": 0.26544,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.011131795247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2837,
+      "loss/crossentropy": 2.027509331703186,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21377252787351608,
+      "step": 13272
+    },
+    {
+      "epoch": 0.26548,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.012393951416015625,
+      "learning_rate": 0.0001,
+      "loss": 3.9739,
+      "loss/crossentropy": 2.0517578125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21172185242176056,
+      "step": 13274
+    },
+    {
+      "epoch": 0.26552,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.01395263671875,
+      "learning_rate": 0.0001,
+      "loss": 4.1628,
+      "loss/crossentropy": 2.0550750494003296,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22374523431062698,
+      "step": 13276
+    },
+    {
+      "epoch": 0.26556,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.013399251302083333,
+      "learning_rate": 0.0001,
+      "loss": 4.1829,
+      "loss/crossentropy": 1.8055049777030945,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17029780894517899,
+      "step": 13278
+    },
+    {
+      "epoch": 0.2656,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.013114166259765626,
+      "learning_rate": 0.0001,
+      "loss": 4.3999,
+      "loss/crossentropy": 1.9650321006774902,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20327500253915787,
+      "step": 13280
+    },
+    {
+      "epoch": 0.26564,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.010209147135416667,
+      "learning_rate": 0.0001,
+      "loss": 4.1395,
+      "loss/crossentropy": 2.164494276046753,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21579623967409134,
+      "step": 13282
+    },
+    {
+      "epoch": 0.26568,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.009810129801432291,
+      "learning_rate": 0.0001,
+      "loss": 4.0676,
+      "loss/crossentropy": 2.0488376021385193,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20213299989700317,
+      "step": 13284
+    },
+    {
+      "epoch": 0.26572,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.010308583577473959,
+      "learning_rate": 0.0001,
+      "loss": 3.7442,
+      "loss/crossentropy": 1.948447048664093,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19875720143318176,
+      "step": 13286
+    },
+    {
+      "epoch": 0.26576,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.008487701416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2312,
+      "loss/crossentropy": 2.115865170955658,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19675103574991226,
+      "step": 13288
+    },
+    {
+      "epoch": 0.2658,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0105377197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2289,
+      "loss/crossentropy": 1.9720736145973206,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20975899696350098,
+      "step": 13290
+    },
+    {
+      "epoch": 0.26584,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.009154256184895833,
+      "learning_rate": 0.0001,
+      "loss": 3.9258,
+      "loss/crossentropy": 1.9105132222175598,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20582708716392517,
+      "step": 13292
+    },
+    {
+      "epoch": 0.26588,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.018369293212890624,
+      "learning_rate": 0.0001,
+      "loss": 4.2629,
+      "loss/crossentropy": 2.089789867401123,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21264012902975082,
+      "step": 13294
+    },
+    {
+      "epoch": 0.26592,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.018358357747395835,
+      "learning_rate": 0.0001,
+      "loss": 4.41,
+      "loss/crossentropy": 2.271156430244446,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21146492660045624,
+      "step": 13296
+    },
+    {
+      "epoch": 0.26596,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0196929931640625,
+      "learning_rate": 0.0001,
+      "loss": 3.8554,
+      "loss/crossentropy": 1.930766224861145,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1959189623594284,
+      "step": 13298
+    },
+    {
+      "epoch": 0.266,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.023374176025390624,
+      "learning_rate": 0.0001,
+      "loss": 4.3551,
+      "loss/crossentropy": 1.9945995807647705,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20959854125976562,
+      "step": 13300
+    },
+    {
+      "epoch": 0.26604,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.020637003580729167,
+      "learning_rate": 0.0001,
+      "loss": 4.5445,
+      "loss/crossentropy": 2.2128443717956543,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21266601234674454,
+      "step": 13302
+    },
+    {
+      "epoch": 0.26608,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.018049112955729165,
+      "learning_rate": 0.0001,
+      "loss": 4.2578,
+      "loss/crossentropy": 2.076206088066101,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2037106677889824,
+      "step": 13304
+    },
+    {
+      "epoch": 0.26612,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.018724568684895835,
+      "learning_rate": 0.0001,
+      "loss": 4.0512,
+      "loss/crossentropy": 2.077200174331665,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.228537917137146,
+      "step": 13306
+    },
+    {
+      "epoch": 0.26616,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.01874567667643229,
+      "learning_rate": 0.0001,
+      "loss": 3.9641,
+      "loss/crossentropy": 2.0368794202804565,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2150697112083435,
+      "step": 13308
+    },
+    {
+      "epoch": 0.2662,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.016556803385416666,
+      "learning_rate": 0.0001,
+      "loss": 3.8516,
+      "loss/crossentropy": 1.9108307361602783,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1903306469321251,
+      "step": 13310
+    },
+    {
+      "epoch": 0.26624,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.016218058268229165,
+      "learning_rate": 0.0001,
+      "loss": 3.9928,
+      "loss/crossentropy": 1.8118465542793274,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18557358533143997,
+      "step": 13312
+    },
+    {
+      "epoch": 0.26628,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.01736424763997396,
+      "learning_rate": 0.0001,
+      "loss": 3.897,
+      "loss/crossentropy": 1.7392275929450989,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19214990735054016,
+      "step": 13314
+    },
+    {
+      "epoch": 0.26632,
+      "grad_norm": 3.359375,
+      "grad_norm_var": 0.12773412068684895,
+      "learning_rate": 0.0001,
+      "loss": 3.9728,
+      "loss/crossentropy": 2.1361005306243896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22042304277420044,
+      "step": 13316
+    },
+    {
+      "epoch": 0.26636,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.12786026000976564,
+      "learning_rate": 0.0001,
+      "loss": 4.0055,
+      "loss/crossentropy": 2.0087279677391052,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20278790593147278,
+      "step": 13318
+    },
+    {
+      "epoch": 0.2664,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.12776692708333334,
+      "learning_rate": 0.0001,
+      "loss": 4.0781,
+      "loss/crossentropy": 2.307933807373047,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24929606914520264,
+      "step": 13320
+    },
+    {
+      "epoch": 0.26644,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.12690022786458333,
+      "learning_rate": 0.0001,
+      "loss": 4.3451,
+      "loss/crossentropy": 2.5539438724517822,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2219071164727211,
+      "step": 13322
+    },
+    {
+      "epoch": 0.26648,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.12910334269205728,
+      "learning_rate": 0.0001,
+      "loss": 3.9984,
+      "loss/crossentropy": 1.9069242477416992,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19276316463947296,
+      "step": 13324
+    },
+    {
+      "epoch": 0.26652,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.12194010416666666,
+      "learning_rate": 0.0001,
+      "loss": 4.535,
+      "loss/crossentropy": 2.0715484619140625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20079267024993896,
+      "step": 13326
+    },
+    {
+      "epoch": 0.26656,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.12164713541666666,
+      "learning_rate": 0.0001,
+      "loss": 4.46,
+      "loss/crossentropy": 2.1134172677993774,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20564979314804077,
+      "step": 13328
+    },
+    {
+      "epoch": 0.2666,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.11581929524739583,
+      "learning_rate": 0.0001,
+      "loss": 4.3992,
+      "loss/crossentropy": 2.103445053100586,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19361437857151031,
+      "step": 13330
+    },
+    {
+      "epoch": 0.26664,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.014499664306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0169,
+      "loss/crossentropy": 2.1882529258728027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21402588486671448,
+      "step": 13332
+    },
+    {
+      "epoch": 0.26668,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.015498606363932292,
+      "learning_rate": 0.0001,
+      "loss": 4.3247,
+      "loss/crossentropy": 2.2531981468200684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2019234597682953,
+      "step": 13334
+    },
+    {
+      "epoch": 0.26672,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.021954091389973958,
+      "learning_rate": 0.0001,
+      "loss": 3.8596,
+      "loss/crossentropy": 1.8077877759933472,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19121356308460236,
+      "step": 13336
+    },
+    {
+      "epoch": 0.26676,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.02600072224934896,
+      "learning_rate": 0.0001,
+      "loss": 4.3195,
+      "loss/crossentropy": 1.7387139797210693,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20326050370931625,
+      "step": 13338
+    },
+    {
+      "epoch": 0.2668,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.022712198893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.065,
+      "loss/crossentropy": 2.2061930894851685,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22271326184272766,
+      "step": 13340
+    },
+    {
+      "epoch": 0.26684,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.02191162109375,
+      "learning_rate": 0.0001,
+      "loss": 4.4018,
+      "loss/crossentropy": 2.096329092979431,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2201632410287857,
+      "step": 13342
+    },
+    {
+      "epoch": 0.26688,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0232574462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0783,
+      "loss/crossentropy": 2.2015284299850464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22004209458827972,
+      "step": 13344
+    },
+    {
+      "epoch": 0.26692,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.023957316080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.5227,
+      "loss/crossentropy": 2.256517231464386,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21762817353010178,
+      "step": 13346
+    },
+    {
+      "epoch": 0.26696,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.022874959309895835,
+      "learning_rate": 0.0001,
+      "loss": 3.8578,
+      "loss/crossentropy": 1.8551252484321594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18950944393873215,
+      "step": 13348
+    },
+    {
+      "epoch": 0.267,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.020967610677083335,
+      "learning_rate": 0.0001,
+      "loss": 4.3051,
+      "loss/crossentropy": 1.8834841847419739,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17950639128684998,
+      "step": 13350
+    },
+    {
+      "epoch": 0.26704,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012872060139973959,
+      "learning_rate": 0.0001,
+      "loss": 4.0234,
+      "loss/crossentropy": 2.0059815645217896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2279331535100937,
+      "step": 13352
+    },
+    {
+      "epoch": 0.26708,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.00594482421875,
+      "learning_rate": 0.0001,
+      "loss": 4.0102,
+      "loss/crossentropy": 2.2327537536621094,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22790935635566711,
+      "step": 13354
+    },
+    {
+      "epoch": 0.26712,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0066650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0535,
+      "loss/crossentropy": 1.7623894214630127,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17889687418937683,
+      "step": 13356
+    },
+    {
+      "epoch": 0.26716,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.006205240885416667,
+      "learning_rate": 0.0001,
+      "loss": 4.1595,
+      "loss/crossentropy": 1.9240076541900635,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18552126735448837,
+      "step": 13358
+    },
+    {
+      "epoch": 0.2672,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.007387034098307292,
+      "learning_rate": 0.0001,
+      "loss": 3.9162,
+      "loss/crossentropy": 2.301589012145996,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20251981914043427,
+      "step": 13360
+    },
+    {
+      "epoch": 0.26724,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.007671864827473959,
+      "learning_rate": 0.0001,
+      "loss": 4.1575,
+      "loss/crossentropy": 2.144057512283325,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22049692273139954,
+      "step": 13362
+    },
+    {
+      "epoch": 0.26728,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.006209309895833333,
+      "learning_rate": 0.0001,
+      "loss": 4.2381,
+      "loss/crossentropy": 1.9329981207847595,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2015545517206192,
+      "step": 13364
+    },
+    {
+      "epoch": 0.26732,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0062896728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2544,
+      "loss/crossentropy": 2.057462990283966,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19826926290988922,
+      "step": 13366
+    },
+    {
+      "epoch": 0.26736,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.006670888264973958,
+      "learning_rate": 0.0001,
+      "loss": 4.085,
+      "loss/crossentropy": 2.097291588783264,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18807460367679596,
+      "step": 13368
+    },
+    {
+      "epoch": 0.2674,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007281239827473958,
+      "learning_rate": 0.0001,
+      "loss": 4.1204,
+      "loss/crossentropy": 2.154082179069519,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22133710980415344,
+      "step": 13370
+    },
+    {
+      "epoch": 0.26744,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.017561848958333334,
+      "learning_rate": 0.0001,
+      "loss": 4.1412,
+      "loss/crossentropy": 2.208779454231262,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2139016017317772,
+      "step": 13372
+    },
+    {
+      "epoch": 0.26748,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.018355305989583334,
+      "learning_rate": 0.0001,
+      "loss": 3.8929,
+      "loss/crossentropy": 2.251901865005493,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22264431416988373,
+      "step": 13374
+    },
+    {
+      "epoch": 0.26752,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.016857655843098958,
+      "learning_rate": 0.0001,
+      "loss": 4.3243,
+      "loss/crossentropy": 2.4881285429000854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.250653475522995,
+      "step": 13376
+    },
+    {
+      "epoch": 0.26756,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.017427571614583335,
+      "learning_rate": 0.0001,
+      "loss": 4.2806,
+      "loss/crossentropy": 2.104843556880951,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19004888087511063,
+      "step": 13378
+    },
+    {
+      "epoch": 0.2676,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.022899373372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.0708,
+      "loss/crossentropy": 1.913047194480896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20572267472743988,
+      "step": 13380
+    },
+    {
+      "epoch": 0.26764,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.02463353474934896,
+      "learning_rate": 0.0001,
+      "loss": 4.3632,
+      "loss/crossentropy": 2.1043838262557983,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21029697358608246,
+      "step": 13382
+    },
+    {
+      "epoch": 0.26768,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.02617162068684896,
+      "learning_rate": 0.0001,
+      "loss": 4.2429,
+      "loss/crossentropy": 2.055271625518799,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2113131284713745,
+      "step": 13384
+    },
+    {
+      "epoch": 0.26772,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.0488433837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4956,
+      "loss/crossentropy": 1.8407886624336243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2729990780353546,
+      "step": 13386
+    },
+    {
+      "epoch": 0.26776,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.04177017211914062,
+      "learning_rate": 0.0001,
+      "loss": 4.0097,
+      "loss/crossentropy": 2.0368717908859253,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21415862441062927,
+      "step": 13388
+    },
+    {
+      "epoch": 0.2678,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.040897369384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1836,
+      "loss/crossentropy": 1.8235292434692383,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2062123641371727,
+      "step": 13390
+    },
+    {
+      "epoch": 0.26784,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.04222183227539063,
+      "learning_rate": 0.0001,
+      "loss": 3.8392,
+      "loss/crossentropy": 1.733048439025879,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20651907473802567,
+      "step": 13392
+    },
+    {
+      "epoch": 0.26788,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.043454742431640624,
+      "learning_rate": 0.0001,
+      "loss": 3.9004,
+      "loss/crossentropy": 2.1339367628097534,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21730080246925354,
+      "step": 13394
+    },
+    {
+      "epoch": 0.26792,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.0452789306640625,
+      "learning_rate": 0.0001,
+      "loss": 3.8421,
+      "loss/crossentropy": 1.885707974433899,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20833835750818253,
+      "step": 13396
+    },
+    {
+      "epoch": 0.26796,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.04429423014322917,
+      "learning_rate": 0.0001,
+      "loss": 3.9714,
+      "loss/crossentropy": 1.9896257519721985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1938292756676674,
+      "step": 13398
+    },
+    {
+      "epoch": 0.268,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.04457575480143229,
+      "learning_rate": 0.0001,
+      "loss": 4.2962,
+      "loss/crossentropy": 2.0359573364257812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21572312712669373,
+      "step": 13400
+    },
+    {
+      "epoch": 0.26804,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.014095052083333334,
+      "learning_rate": 0.0001,
+      "loss": 4.3483,
+      "loss/crossentropy": 2.241006851196289,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2361859679222107,
+      "step": 13402
+    },
+    {
+      "epoch": 0.26808,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.013852691650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0518,
+      "loss/crossentropy": 2.0988917350769043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20220966637134552,
+      "step": 13404
+    },
+    {
+      "epoch": 0.26812,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.014802805582682292,
+      "learning_rate": 0.0001,
+      "loss": 4.3664,
+      "loss/crossentropy": 2.0803651213645935,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22251859307289124,
+      "step": 13406
+    },
+    {
+      "epoch": 0.26816,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.015827433268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.0473,
+      "loss/crossentropy": 1.9151242971420288,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19792036712169647,
+      "step": 13408
+    },
+    {
+      "epoch": 0.2682,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.07619196573893229,
+      "learning_rate": 0.0001,
+      "loss": 4.0321,
+      "loss/crossentropy": 1.908549964427948,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21568355709314346,
+      "step": 13410
+    },
+    {
+      "epoch": 0.26824,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.06887919108072917,
+      "learning_rate": 0.0001,
+      "loss": 4.3844,
+      "loss/crossentropy": 2.3226535320281982,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2332221046090126,
+      "step": 13412
+    },
+    {
+      "epoch": 0.26828,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.06696548461914062,
+      "learning_rate": 0.0001,
+      "loss": 4.24,
+      "loss/crossentropy": 1.9767839312553406,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2038547545671463,
+      "step": 13414
+    },
+    {
+      "epoch": 0.26832,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.06800028483072916,
+      "learning_rate": 0.0001,
+      "loss": 3.9592,
+      "loss/crossentropy": 2.2013003826141357,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2076154574751854,
+      "step": 13416
+    },
+    {
+      "epoch": 0.26836,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.06597671508789063,
+      "learning_rate": 0.0001,
+      "loss": 4.1128,
+      "loss/crossentropy": 2.0941338539123535,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20538556575775146,
+      "step": 13418
+    },
+    {
+      "epoch": 0.2684,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.06494954427083334,
+      "learning_rate": 0.0001,
+      "loss": 4.3656,
+      "loss/crossentropy": 2.0206560492515564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20901528000831604,
+      "step": 13420
+    },
+    {
+      "epoch": 0.26844,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.06690648396809896,
+      "learning_rate": 0.0001,
+      "loss": 4.2606,
+      "loss/crossentropy": 2.1317169070243835,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21660470962524414,
+      "step": 13422
+    },
+    {
+      "epoch": 0.26848,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.07109273274739583,
+      "learning_rate": 0.0001,
+      "loss": 3.6026,
+      "loss/crossentropy": 1.631429135799408,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1729787141084671,
+      "step": 13424
+    },
+    {
+      "epoch": 0.26852,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.012491607666015625,
+      "learning_rate": 0.0001,
+      "loss": 3.923,
+      "loss/crossentropy": 1.8428707122802734,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18231894075870514,
+      "step": 13426
+    },
+    {
+      "epoch": 0.26856,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.011104075113932292,
+      "learning_rate": 0.0001,
+      "loss": 4.2351,
+      "loss/crossentropy": 1.860277235507965,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18769516795873642,
+      "step": 13428
+    },
+    {
+      "epoch": 0.2686,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.012542470296223959,
+      "learning_rate": 0.0001,
+      "loss": 4.0199,
+      "loss/crossentropy": 2.02046799659729,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2160683423280716,
+      "step": 13430
+    },
+    {
+      "epoch": 0.26864,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.021491495768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.5105,
+      "loss/crossentropy": 2.094564437866211,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22155898064374924,
+      "step": 13432
+    },
+    {
+      "epoch": 0.26868,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.02215550740559896,
+      "learning_rate": 0.0001,
+      "loss": 4.0382,
+      "loss/crossentropy": 2.013366401195526,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2126912623643875,
+      "step": 13434
+    },
+    {
+      "epoch": 0.26872,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.02212702433268229,
+      "learning_rate": 0.0001,
+      "loss": 4.5826,
+      "loss/crossentropy": 2.360998272895813,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2355523630976677,
+      "step": 13436
+    },
+    {
+      "epoch": 0.26876,
+      "grad_norm": 1.7890625,
+      "grad_norm_var": 0.02289606730143229,
+      "learning_rate": 0.0001,
+      "loss": 3.7129,
+      "loss/crossentropy": 1.754651427268982,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18666986376047134,
+      "step": 13438
+    },
+    {
+      "epoch": 0.2688,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.01812922159830729,
+      "learning_rate": 0.0001,
+      "loss": 4.1024,
+      "loss/crossentropy": 1.9176424741744995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19241741299629211,
+      "step": 13440
+    },
+    {
+      "epoch": 0.26884,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0154449462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2548,
+      "loss/crossentropy": 2.324304223060608,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22302107512950897,
+      "step": 13442
+    },
+    {
+      "epoch": 0.26888,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.028758748372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.0693,
+      "loss/crossentropy": 1.8527624011039734,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18644367158412933,
+      "step": 13444
+    },
+    {
+      "epoch": 0.26892,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.027570597330729165,
+      "learning_rate": 0.0001,
+      "loss": 4.4437,
+      "loss/crossentropy": 2.0999260544776917,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21172572672367096,
+      "step": 13446
+    },
+    {
+      "epoch": 0.26896,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.026627349853515624,
+      "learning_rate": 0.0001,
+      "loss": 3.8125,
+      "loss/crossentropy": 1.8577081561088562,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17302225530147552,
+      "step": 13448
+    },
+    {
+      "epoch": 0.269,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.026956939697265626,
+      "learning_rate": 0.0001,
+      "loss": 4.2182,
+      "loss/crossentropy": 2.11561119556427,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20249811559915543,
+      "step": 13450
+    },
+    {
+      "epoch": 0.26904,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.026656087239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.2854,
+      "loss/crossentropy": 2.3036913871765137,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20101947337388992,
+      "step": 13452
+    },
+    {
+      "epoch": 0.26908,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.02341283162434896,
+      "learning_rate": 0.0001,
+      "loss": 3.9013,
+      "loss/crossentropy": 2.0498175621032715,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19935546070337296,
+      "step": 13454
+    },
+    {
+      "epoch": 0.26912,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.08842137654622396,
+      "learning_rate": 0.0001,
+      "loss": 4.2601,
+      "loss/crossentropy": 2.27053964138031,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31231266260147095,
+      "step": 13456
+    },
+    {
+      "epoch": 0.26916,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.0908953348795573,
+      "learning_rate": 0.0001,
+      "loss": 4.0035,
+      "loss/crossentropy": 2.191486120223999,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2058379054069519,
+      "step": 13458
+    },
+    {
+      "epoch": 0.2692,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.07780939737955729,
+      "learning_rate": 0.0001,
+      "loss": 4.1193,
+      "loss/crossentropy": 2.008154332637787,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20862725377082825,
+      "step": 13460
+    },
+    {
+      "epoch": 0.26924,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.07779947916666667,
+      "learning_rate": 0.0001,
+      "loss": 4.3191,
+      "loss/crossentropy": 2.3374814987182617,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24534277617931366,
+      "step": 13462
+    },
+    {
+      "epoch": 0.26928,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.07355855305989584,
+      "learning_rate": 0.0001,
+      "loss": 4.0558,
+      "loss/crossentropy": 2.2654502391815186,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21406329423189163,
+      "step": 13464
+    },
+    {
+      "epoch": 0.26932,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.07412821451822917,
+      "learning_rate": 0.0001,
+      "loss": 4.2078,
+      "loss/crossentropy": 2.1021856665611267,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19653601199388504,
+      "step": 13466
+    },
+    {
+      "epoch": 0.26936,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0727068583170573,
+      "learning_rate": 0.0001,
+      "loss": 4.4466,
+      "loss/crossentropy": 1.9482329487800598,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20822357386350632,
+      "step": 13468
+    },
+    {
+      "epoch": 0.2694,
+      "grad_norm": 1.8125,
+      "grad_norm_var": 0.07814915974934895,
+      "learning_rate": 0.0001,
+      "loss": 4.1036,
+      "loss/crossentropy": 2.122725486755371,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2132444903254509,
+      "step": 13470
+    },
+    {
+      "epoch": 0.26944,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.015044911702473959,
+      "learning_rate": 0.0001,
+      "loss": 4.0337,
+      "loss/crossentropy": 1.760430932044983,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1835266426205635,
+      "step": 13472
+    },
+    {
+      "epoch": 0.26948,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.014847564697265624,
+      "learning_rate": 0.0001,
+      "loss": 4.0338,
+      "loss/crossentropy": 1.610491931438446,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19731693714857101,
+      "step": 13474
+    },
+    {
+      "epoch": 0.26952,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.015313466389973959,
+      "learning_rate": 0.0001,
+      "loss": 4.213,
+      "loss/crossentropy": 2.1966941356658936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22074176371097565,
+      "step": 13476
+    },
+    {
+      "epoch": 0.26956,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.029904937744140624,
+      "learning_rate": 0.0001,
+      "loss": 4.1291,
+      "loss/crossentropy": 2.045714259147644,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2193753719329834,
+      "step": 13478
+    },
+    {
+      "epoch": 0.2696,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.029886881510416668,
+      "learning_rate": 0.0001,
+      "loss": 4.0127,
+      "loss/crossentropy": 1.860603928565979,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1900448128581047,
+      "step": 13480
+    },
+    {
+      "epoch": 0.26964,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.028449503580729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0643,
+      "loss/crossentropy": 2.171591639518738,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21908972412347794,
+      "step": 13482
+    },
+    {
+      "epoch": 0.26968,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.030265299479166667,
+      "learning_rate": 0.0001,
+      "loss": 4.1351,
+      "loss/crossentropy": 2.271330237388611,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21823390573263168,
+      "step": 13484
+    },
+    {
+      "epoch": 0.26972,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.026875813802083332,
+      "learning_rate": 0.0001,
+      "loss": 4.192,
+      "loss/crossentropy": 2.155557870864868,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20512574911117554,
+      "step": 13486
+    },
+    {
+      "epoch": 0.26976,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.027164459228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2561,
+      "loss/crossentropy": 2.3827039003372192,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22485698014497757,
+      "step": 13488
+    },
+    {
+      "epoch": 0.2698,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.028636678059895834,
+      "learning_rate": 0.0001,
+      "loss": 3.9937,
+      "loss/crossentropy": 2.225351929664612,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22235903143882751,
+      "step": 13490
+    },
+    {
+      "epoch": 0.26984,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.027717081705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0144,
+      "loss/crossentropy": 2.1125447750091553,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20206717401742935,
+      "step": 13492
+    },
+    {
+      "epoch": 0.26988,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.013639068603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0318,
+      "loss/crossentropy": 2.30017626285553,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22321031987667084,
+      "step": 13494
+    },
+    {
+      "epoch": 0.26992,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.0145904541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3979,
+      "loss/crossentropy": 2.416603446006775,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2413351908326149,
+      "step": 13496
+    },
+    {
+      "epoch": 0.26996,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.013185373942057292,
+      "learning_rate": 0.0001,
+      "loss": 4.0436,
+      "loss/crossentropy": 2.06734299659729,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23090296238660812,
+      "step": 13498
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0050961812337239586,
+      "learning_rate": 0.0001,
+      "loss": 4.0,
+      "loss/crossentropy": 1.8088473677635193,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1988995596766472,
+      "step": 13500
+    },
+    {
+      "epoch": 0.27004,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.005092112223307291,
+      "learning_rate": 0.0001,
+      "loss": 4.0205,
+      "loss/crossentropy": 1.7075524926185608,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18463176488876343,
+      "step": 13502
+    },
+    {
+      "epoch": 0.27008,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0083160400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2291,
+      "loss/crossentropy": 2.1184898018836975,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20941460877656937,
+      "step": 13504
+    },
+    {
+      "epoch": 0.27012,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.007736968994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0139,
+      "loss/crossentropy": 1.8351567387580872,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18913155794143677,
+      "step": 13506
+    },
+    {
+      "epoch": 0.27016,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.00914306640625,
+      "learning_rate": 0.0001,
+      "loss": 3.935,
+      "loss/crossentropy": 2.090391516685486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21020452678203583,
+      "step": 13508
+    },
+    {
+      "epoch": 0.2702,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0078521728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1805,
+      "loss/crossentropy": 2.0680218935012817,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20998934656381607,
+      "step": 13510
+    },
+    {
+      "epoch": 0.27024,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.007236480712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0074,
+      "loss/crossentropy": 2.258071780204773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22634688764810562,
+      "step": 13512
+    },
+    {
+      "epoch": 0.27028,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.03350397745768229,
+      "learning_rate": 0.0001,
+      "loss": 4.775,
+      "loss/crossentropy": 1.931319773197174,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29265259206295013,
+      "step": 13514
+    },
+    {
+      "epoch": 0.27032,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.03398844401041667,
+      "learning_rate": 0.0001,
+      "loss": 4.3408,
+      "loss/crossentropy": 2.45454478263855,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22465243190526962,
+      "step": 13516
+    },
+    {
+      "epoch": 0.27036,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.034063466389973956,
+      "learning_rate": 0.0001,
+      "loss": 4.2279,
+      "loss/crossentropy": 2.005887746810913,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19818827509880066,
+      "step": 13518
+    },
+    {
+      "epoch": 0.2704,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.031107330322265626,
+      "learning_rate": 0.0001,
+      "loss": 4.318,
+      "loss/crossentropy": 2.2329577207565308,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21015001833438873,
+      "step": 13520
+    },
+    {
+      "epoch": 0.27044,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.032373046875,
+      "learning_rate": 0.0001,
+      "loss": 4.1026,
+      "loss/crossentropy": 2.1716688871383667,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20930524915456772,
+      "step": 13522
+    },
+    {
+      "epoch": 0.27048,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.030651601155598958,
+      "learning_rate": 0.0001,
+      "loss": 3.957,
+      "loss/crossentropy": 1.659675419330597,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1963459849357605,
+      "step": 13524
+    },
+    {
+      "epoch": 0.27052,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.035359700520833336,
+      "learning_rate": 0.0001,
+      "loss": 4.2214,
+      "loss/crossentropy": 1.9285706877708435,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1708521470427513,
+      "step": 13526
+    },
+    {
+      "epoch": 0.27056,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.03286031087239583,
+      "learning_rate": 0.0001,
+      "loss": 4.1771,
+      "loss/crossentropy": 2.0428953170776367,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20893662422895432,
+      "step": 13528
+    },
+    {
+      "epoch": 0.2706,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.011655426025390625,
+      "learning_rate": 0.0001,
+      "loss": 3.9885,
+      "loss/crossentropy": 1.979922592639923,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19091727584600449,
+      "step": 13530
+    },
+    {
+      "epoch": 0.27064,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.015922037760416667,
+      "learning_rate": 0.0001,
+      "loss": 4.3176,
+      "loss/crossentropy": 2.1879321336746216,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24813223630189896,
+      "step": 13532
+    },
+    {
+      "epoch": 0.27068,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.01631647745768229,
+      "learning_rate": 0.0001,
+      "loss": 4.1852,
+      "loss/crossentropy": 2.0450875759124756,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21059003472328186,
+      "step": 13534
+    },
+    {
+      "epoch": 0.27072,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.01673151652018229,
+      "learning_rate": 0.0001,
+      "loss": 4.3549,
+      "loss/crossentropy": 2.050394892692566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21235206723213196,
+      "step": 13536
+    },
+    {
+      "epoch": 0.27076,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.014793904622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0128,
+      "loss/crossentropy": 2.0687233805656433,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22125782817602158,
+      "step": 13538
+    },
+    {
+      "epoch": 0.2708,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.014647420247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1247,
+      "loss/crossentropy": 1.8587198853492737,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19718395173549652,
+      "step": 13540
+    },
+    {
+      "epoch": 0.27084,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010223134358723959,
+      "learning_rate": 0.0001,
+      "loss": 4.2992,
+      "loss/crossentropy": 2.2416387796401978,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23797060549259186,
+      "step": 13542
+    },
+    {
+      "epoch": 0.27088,
+      "grad_norm": 4.8125,
+      "grad_norm_var": 0.495751953125,
+      "learning_rate": 0.0001,
+      "loss": 4.1875,
+      "loss/crossentropy": 2.251328468322754,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23915010690689087,
+      "step": 13544
+    },
+    {
+      "epoch": 0.27092,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.4896074930826823,
+      "learning_rate": 0.0001,
+      "loss": 4.0601,
+      "loss/crossentropy": 1.990403652191162,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21710015833377838,
+      "step": 13546
+    },
+    {
+      "epoch": 0.27096,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.4871070861816406,
+      "learning_rate": 0.0001,
+      "loss": 4.38,
+      "loss/crossentropy": 2.176861047744751,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2153458073735237,
+      "step": 13548
+    },
+    {
+      "epoch": 0.271,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.4867286682128906,
+      "learning_rate": 0.0001,
+      "loss": 4.0311,
+      "loss/crossentropy": 1.581967830657959,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1787950098514557,
+      "step": 13550
+    },
+    {
+      "epoch": 0.27104,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.4920183817545573,
+      "learning_rate": 0.0001,
+      "loss": 3.9803,
+      "loss/crossentropy": 1.827072560787201,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18847094476222992,
+      "step": 13552
+    },
+    {
+      "epoch": 0.27108,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.49279683430989585,
+      "learning_rate": 0.0001,
+      "loss": 3.9381,
+      "loss/crossentropy": 1.5748514533042908,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18339257687330246,
+      "step": 13554
+    },
+    {
+      "epoch": 0.27112,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.4897989908854167,
+      "learning_rate": 0.0001,
+      "loss": 4.2545,
+      "loss/crossentropy": 2.3187586069107056,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2535529136657715,
+      "step": 13556
+    },
+    {
+      "epoch": 0.27116,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.4885660807291667,
+      "learning_rate": 0.0001,
+      "loss": 4.4848,
+      "loss/crossentropy": 2.3675668239593506,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23080945760011673,
+      "step": 13558
+    },
+    {
+      "epoch": 0.2712,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0066912333170572914,
+      "learning_rate": 0.0001,
+      "loss": 4.2032,
+      "loss/crossentropy": 2.3976542949676514,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2418106645345688,
+      "step": 13560
+    },
+    {
+      "epoch": 0.27124,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.006221262613932291,
+      "learning_rate": 0.0001,
+      "loss": 4.1877,
+      "loss/crossentropy": 2.1341328024864197,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21845046430826187,
+      "step": 13562
+    },
+    {
+      "epoch": 0.27128,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.006154123942057292,
+      "learning_rate": 0.0001,
+      "loss": 4.3067,
+      "loss/crossentropy": 2.0519689321517944,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21991144120693207,
+      "step": 13564
+    },
+    {
+      "epoch": 0.27132,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.005812327067057292,
+      "learning_rate": 0.0001,
+      "loss": 4.106,
+      "loss/crossentropy": 1.9690070748329163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20945511013269424,
+      "step": 13566
+    },
+    {
+      "epoch": 0.27136,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.004609934488932292,
+      "learning_rate": 0.0001,
+      "loss": 4.0929,
+      "loss/crossentropy": 1.9553492069244385,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2299523577094078,
+      "step": 13568
+    },
+    {
+      "epoch": 0.2714,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0055501302083333336,
+      "learning_rate": 0.0001,
+      "loss": 4.158,
+      "loss/crossentropy": 1.8214278817176819,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1801338642835617,
+      "step": 13570
+    },
+    {
+      "epoch": 0.27144,
+      "grad_norm": 1.7578125,
+      "grad_norm_var": 0.011822255452473958,
+      "learning_rate": 0.0001,
+      "loss": 3.8658,
+      "loss/crossentropy": 1.7834638953208923,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18421506136655807,
+      "step": 13572
+    },
+    {
+      "epoch": 0.27148,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.013097890218098958,
+      "learning_rate": 0.0001,
+      "loss": 3.8404,
+      "loss/crossentropy": 1.7703429460525513,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18263264745473862,
+      "step": 13574
+    },
+    {
+      "epoch": 0.27152,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0155181884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1489,
+      "loss/crossentropy": 2.228934168815613,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21034922450780869,
+      "step": 13576
+    },
+    {
+      "epoch": 0.27156,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.015379842122395833,
+      "learning_rate": 0.0001,
+      "loss": 4.4665,
+      "loss/crossentropy": 2.160655975341797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22300932556390762,
+      "step": 13578
+    },
+    {
+      "epoch": 0.2716,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0150634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2042,
+      "loss/crossentropy": 2.1582624912261963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21843481063842773,
+      "step": 13580
+    },
+    {
+      "epoch": 0.27164,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0150146484375,
+      "learning_rate": 0.0001,
+      "loss": 4.2513,
+      "loss/crossentropy": 2.0660162568092346,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1985679790377617,
+      "step": 13582
+    },
+    {
+      "epoch": 0.27168,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.015433502197265626,
+      "learning_rate": 0.0001,
+      "loss": 3.873,
+      "loss/crossentropy": 1.839695692062378,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1772506907582283,
+      "step": 13584
+    },
+    {
+      "epoch": 0.27172,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.012345123291015624,
+      "learning_rate": 0.0001,
+      "loss": 3.9649,
+      "loss/crossentropy": 2.0837132930755615,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19461007416248322,
+      "step": 13586
+    },
+    {
+      "epoch": 0.27176,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.00750732421875,
+      "learning_rate": 0.0001,
+      "loss": 4.1986,
+      "loss/crossentropy": 1.910473346710205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20272061973810196,
+      "step": 13588
+    },
+    {
+      "epoch": 0.2718,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.054402669270833336,
+      "learning_rate": 0.0001,
+      "loss": 4.2259,
+      "loss/crossentropy": 2.2427467107772827,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21372541785240173,
+      "step": 13590
+    },
+    {
+      "epoch": 0.27184,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.055272420247395836,
+      "learning_rate": 0.0001,
+      "loss": 4.1123,
+      "loss/crossentropy": 2.165425181388855,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2552504763007164,
+      "step": 13592
+    },
+    {
+      "epoch": 0.27188,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.05620930989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.0993,
+      "loss/crossentropy": 1.9684009552001953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19894393533468246,
+      "step": 13594
+    },
+    {
+      "epoch": 0.27192,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.055863444010416666,
+      "learning_rate": 0.0001,
+      "loss": 4.2421,
+      "loss/crossentropy": 2.0572606325149536,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20954644680023193,
+      "step": 13596
+    },
+    {
+      "epoch": 0.27196,
+      "grad_norm": 1.78125,
+      "grad_norm_var": 0.0611724853515625,
+      "learning_rate": 0.0001,
+      "loss": 3.5511,
+      "loss/crossentropy": 1.8215171694755554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1905388981103897,
+      "step": 13598
+    },
+    {
+      "epoch": 0.272,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.06299209594726562,
+      "learning_rate": 0.0001,
+      "loss": 4.2407,
+      "loss/crossentropy": 2.114215850830078,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21045731008052826,
+      "step": 13600
+    },
+    {
+      "epoch": 0.27204,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.06243057250976562,
+      "learning_rate": 0.0001,
+      "loss": 4.1409,
+      "loss/crossentropy": 2.301407814025879,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2148626372218132,
+      "step": 13602
+    },
+    {
+      "epoch": 0.27208,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.06350809733072917,
+      "learning_rate": 0.0001,
+      "loss": 4.4367,
+      "loss/crossentropy": 2.0508424639701843,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21292225271463394,
+      "step": 13604
+    },
+    {
+      "epoch": 0.27212,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.014826456705729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0024,
+      "loss/crossentropy": 2.095071792602539,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23258864879608154,
+      "step": 13606
+    },
+    {
+      "epoch": 0.27216,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.013802083333333333,
+      "learning_rate": 0.0001,
+      "loss": 4.1722,
+      "loss/crossentropy": 2.3613405227661133,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20384693890810013,
+      "step": 13608
+    },
+    {
+      "epoch": 0.2722,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0142578125,
+      "learning_rate": 0.0001,
+      "loss": 4.1397,
+      "loss/crossentropy": 2.1618664264678955,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21691302955150604,
+      "step": 13610
+    },
+    {
+      "epoch": 0.27224,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.015620676676432292,
+      "learning_rate": 0.0001,
+      "loss": 3.8335,
+      "loss/crossentropy": 1.6352717280387878,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18613358587026596,
+      "step": 13612
+    },
+    {
+      "epoch": 0.27228,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.012851715087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2405,
+      "loss/crossentropy": 2.304627537727356,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21715758740901947,
+      "step": 13614
+    },
+    {
+      "epoch": 0.27232,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009357706705729166,
+      "learning_rate": 0.0001,
+      "loss": 4.1845,
+      "loss/crossentropy": 2.232232451438904,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20997462421655655,
+      "step": 13616
+    },
+    {
+      "epoch": 0.27236,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011503092447916667,
+      "learning_rate": 0.0001,
+      "loss": 4.1273,
+      "loss/crossentropy": 2.0350120663642883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20753738284111023,
+      "step": 13618
+    },
+    {
+      "epoch": 0.2724,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.015746053059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1955,
+      "loss/crossentropy": 1.892416536808014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1920388638973236,
+      "step": 13620
+    },
+    {
+      "epoch": 0.27244,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0128814697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3283,
+      "loss/crossentropy": 2.1694198846817017,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.217724971473217,
+      "step": 13622
+    },
+    {
+      "epoch": 0.27248,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.013444010416666667,
+      "learning_rate": 0.0001,
+      "loss": 3.9786,
+      "loss/crossentropy": 2.1418001651763916,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2016705498099327,
+      "step": 13624
+    },
+    {
+      "epoch": 0.27252,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.014240519205729166,
+      "learning_rate": 0.0001,
+      "loss": 4.3992,
+      "loss/crossentropy": 2.32223117351532,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21616849303245544,
+      "step": 13626
+    },
+    {
+      "epoch": 0.27256,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.013044230143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1847,
+      "loss/crossentropy": 1.9991823434829712,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2077011615037918,
+      "step": 13628
+    },
+    {
+      "epoch": 0.2726,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.012434641520182291,
+      "learning_rate": 0.0001,
+      "loss": 4.1331,
+      "loss/crossentropy": 2.1381043195724487,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20748750865459442,
+      "step": 13630
+    },
+    {
+      "epoch": 0.27264,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.012303670247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.2215,
+      "loss/crossentropy": 2.018588602542877,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20897169411182404,
+      "step": 13632
+    },
+    {
+      "epoch": 0.27268,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.01114501953125,
+      "learning_rate": 0.0001,
+      "loss": 4.3528,
+      "loss/crossentropy": 2.1859233379364014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21695572137832642,
+      "step": 13634
+    },
+    {
+      "epoch": 0.27272,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0060198465983072914,
+      "learning_rate": 0.0001,
+      "loss": 4.2106,
+      "loss/crossentropy": 2.17002010345459,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21605493128299713,
+      "step": 13636
+    },
+    {
+      "epoch": 0.27276,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.006058502197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.261,
+      "loss/crossentropy": 2.186649441719055,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2172158733010292,
+      "step": 13638
+    },
+    {
+      "epoch": 0.2728,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.007306925455729167,
+      "learning_rate": 0.0001,
+      "loss": 3.8394,
+      "loss/crossentropy": 1.7324808835983276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17334696650505066,
+      "step": 13640
+    },
+    {
+      "epoch": 0.27284,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0049550374348958336,
+      "learning_rate": 0.0001,
+      "loss": 3.9805,
+      "loss/crossentropy": 1.7768054008483887,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1750154346227646,
+      "step": 13642
+    },
+    {
+      "epoch": 0.27288,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0047515869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2292,
+      "loss/crossentropy": 1.9629738330841064,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19489944726228714,
+      "step": 13644
+    },
+    {
+      "epoch": 0.27292,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.006882476806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4096,
+      "loss/crossentropy": 2.144750416278839,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19367601722478867,
+      "step": 13646
+    },
+    {
+      "epoch": 0.27296,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0067291259765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1301,
+      "loss/crossentropy": 1.9281827211380005,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19894402474164963,
+      "step": 13648
+    },
+    {
+      "epoch": 0.273,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.007132720947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1998,
+      "loss/crossentropy": 1.8092535138130188,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18536384403705597,
+      "step": 13650
+    },
+    {
+      "epoch": 0.27304,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.007726796468098958,
+      "learning_rate": 0.0001,
+      "loss": 4.1697,
+      "loss/crossentropy": 2.5707184076309204,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23067744076251984,
+      "step": 13652
+    },
+    {
+      "epoch": 0.27308,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008434804280598958,
+      "learning_rate": 0.0001,
+      "loss": 4.0189,
+      "loss/crossentropy": 2.108555316925049,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2052021473646164,
+      "step": 13654
+    },
+    {
+      "epoch": 0.27312,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.007228342692057291,
+      "learning_rate": 0.0001,
+      "loss": 4.3412,
+      "loss/crossentropy": 1.836738109588623,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20734501630067825,
+      "step": 13656
+    },
+    {
+      "epoch": 0.27316,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.007990519205729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0783,
+      "loss/crossentropy": 2.223781406879425,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20618071407079697,
+      "step": 13658
+    },
+    {
+      "epoch": 0.2732,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.007207997639973958,
+      "learning_rate": 0.0001,
+      "loss": 4.1722,
+      "loss/crossentropy": 2.028991222381592,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20579702407121658,
+      "step": 13660
+    },
+    {
+      "epoch": 0.27324,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.006915028889973958,
+      "learning_rate": 0.0001,
+      "loss": 4.0294,
+      "loss/crossentropy": 2.1003119349479675,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21900994330644608,
+      "step": 13662
+    },
+    {
+      "epoch": 0.27328,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.007289377848307291,
+      "learning_rate": 0.0001,
+      "loss": 4.1695,
+      "loss/crossentropy": 2.1813069581985474,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2117675319314003,
+      "step": 13664
+    },
+    {
+      "epoch": 0.27332,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0105621337890625,
+      "learning_rate": 0.0001,
+      "loss": 3.8666,
+      "loss/crossentropy": 2.2203832864761353,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20088640600442886,
+      "step": 13666
+    },
+    {
+      "epoch": 0.27336,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0103759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1084,
+      "loss/crossentropy": 1.8810867071151733,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20646882802248,
+      "step": 13668
+    },
+    {
+      "epoch": 0.2734,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.010060373942057292,
+      "learning_rate": 0.0001,
+      "loss": 4.0746,
+      "loss/crossentropy": 2.2147200107574463,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2038613259792328,
+      "step": 13670
+    },
+    {
+      "epoch": 0.27344,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.008408355712890624,
+      "learning_rate": 0.0001,
+      "loss": 4.2151,
+      "loss/crossentropy": 2.3654199838638306,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22163932025432587,
+      "step": 13672
+    },
+    {
+      "epoch": 0.27348,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0086669921875,
+      "learning_rate": 0.0001,
+      "loss": 4.1584,
+      "loss/crossentropy": 2.422420859336853,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23172692954540253,
+      "step": 13674
+    },
+    {
+      "epoch": 0.27352,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008634440104166667,
+      "learning_rate": 0.0001,
+      "loss": 4.391,
+      "loss/crossentropy": 2.0776702165603638,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2030162662267685,
+      "step": 13676
+    },
+    {
+      "epoch": 0.27356,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.009611002604166667,
+      "learning_rate": 0.0001,
+      "loss": 3.9108,
+      "loss/crossentropy": 1.7956212162971497,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1932467818260193,
+      "step": 13678
+    },
+    {
+      "epoch": 0.2736,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.010573069254557291,
+      "learning_rate": 0.0001,
+      "loss": 4.1768,
+      "loss/crossentropy": 2.176727533340454,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20450271666049957,
+      "step": 13680
+    },
+    {
+      "epoch": 0.27364,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0062255859375,
+      "learning_rate": 0.0001,
+      "loss": 4.061,
+      "loss/crossentropy": 2.0534290075302124,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21150045841932297,
+      "step": 13682
+    },
+    {
+      "epoch": 0.27368,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0061920166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1682,
+      "loss/crossentropy": 1.726151466369629,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19256682693958282,
+      "step": 13684
+    },
+    {
+      "epoch": 0.27372,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.005956013997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2666,
+      "loss/crossentropy": 2.2024354934692383,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2222209945321083,
+      "step": 13686
+    },
+    {
+      "epoch": 0.27376,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.007347615559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3252,
+      "loss/crossentropy": 2.095518469810486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2092512547969818,
+      "step": 13688
+    },
+    {
+      "epoch": 0.2738,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0074460347493489586,
+      "learning_rate": 0.0001,
+      "loss": 4.0389,
+      "loss/crossentropy": 1.9978103637695312,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20576970279216766,
+      "step": 13690
+    },
+    {
+      "epoch": 0.27384,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.008599599202473959,
+      "learning_rate": 0.0001,
+      "loss": 4.1023,
+      "loss/crossentropy": 2.233831286430359,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20514021068811417,
+      "step": 13692
+    },
+    {
+      "epoch": 0.27388,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.010251617431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0479,
+      "loss/crossentropy": 1.6729156970977783,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18054980039596558,
+      "step": 13694
+    },
+    {
+      "epoch": 0.27392,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.010910797119140624,
+      "learning_rate": 0.0001,
+      "loss": 4.1911,
+      "loss/crossentropy": 2.0113388895988464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21645111590623856,
+      "step": 13696
+    },
+    {
+      "epoch": 0.27396,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.010469563802083333,
+      "learning_rate": 0.0001,
+      "loss": 4.3487,
+      "loss/crossentropy": 1.9311461448669434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23587358742952347,
+      "step": 13698
+    },
+    {
+      "epoch": 0.274,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.01889012654622396,
+      "learning_rate": 0.0001,
+      "loss": 4.1099,
+      "loss/crossentropy": 2.1205984354019165,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20603877305984497,
+      "step": 13700
+    },
+    {
+      "epoch": 0.27404,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.01889012654622396,
+      "learning_rate": 0.0001,
+      "loss": 4.2947,
+      "loss/crossentropy": 2.1700201630592346,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21855031698942184,
+      "step": 13702
+    },
+    {
+      "epoch": 0.27408,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.022484334309895833,
+      "learning_rate": 0.0001,
+      "loss": 3.7717,
+      "loss/crossentropy": 1.5643411874771118,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.161862351000309,
+      "step": 13704
+    },
+    {
+      "epoch": 0.27412,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.024881998697916668,
+      "learning_rate": 0.0001,
+      "loss": 3.7292,
+      "loss/crossentropy": 2.032666802406311,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20749235153198242,
+      "step": 13706
+    },
+    {
+      "epoch": 0.27416,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.024738566080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0899,
+      "loss/crossentropy": 1.9313859939575195,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19065556675195694,
+      "step": 13708
+    },
+    {
+      "epoch": 0.2742,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.02225519816080729,
+      "learning_rate": 0.0001,
+      "loss": 4.3295,
+      "loss/crossentropy": 2.1200226545333862,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1967960000038147,
+      "step": 13710
+    },
+    {
+      "epoch": 0.27424,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.020765940348307293,
+      "learning_rate": 0.0001,
+      "loss": 4.0655,
+      "loss/crossentropy": 2.144526958465576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20019693672657013,
+      "step": 13712
+    },
+    {
+      "epoch": 0.27428,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.01809666951497396,
+      "learning_rate": 0.0001,
+      "loss": 4.2039,
+      "loss/crossentropy": 1.9802407622337341,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20123689621686935,
+      "step": 13714
+    },
+    {
+      "epoch": 0.27432,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0062334696451822914,
+      "learning_rate": 0.0001,
+      "loss": 3.9415,
+      "loss/crossentropy": 2.096527099609375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20282836258411407,
+      "step": 13716
+    },
+    {
+      "epoch": 0.27436,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.005558013916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1829,
+      "loss/crossentropy": 2.1037757992744446,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2039530873298645,
+      "step": 13718
+    },
+    {
+      "epoch": 0.2744,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.004752349853515625,
+      "learning_rate": 0.0001,
+      "loss": 3.8416,
+      "loss/crossentropy": 1.5415751934051514,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18478797376155853,
+      "step": 13720
+    },
+    {
+      "epoch": 0.27444,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.001859283447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.071,
+      "loss/crossentropy": 2.0357913970947266,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20176240801811218,
+      "step": 13722
+    },
+    {
+      "epoch": 0.27448,
+      "grad_norm": 1.8125,
+      "grad_norm_var": 0.0034543355305989582,
+      "learning_rate": 0.0001,
+      "loss": 3.8472,
+      "loss/crossentropy": 1.9013367891311646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20012690126895905,
+      "step": 13724
+    },
+    {
+      "epoch": 0.27452,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.00343017578125,
+      "learning_rate": 0.0001,
+      "loss": 3.8072,
+      "loss/crossentropy": 1.5798682570457458,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18169572949409485,
+      "step": 13726
+    },
+    {
+      "epoch": 0.27456,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0034624735514322915,
+      "learning_rate": 0.0001,
+      "loss": 4.163,
+      "loss/crossentropy": 2.196288228034973,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22311393916606903,
+      "step": 13728
+    },
+    {
+      "epoch": 0.2746,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.003928375244140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0172,
+      "loss/crossentropy": 1.714774489402771,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17624646425247192,
+      "step": 13730
+    },
+    {
+      "epoch": 0.27464,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0053484598795572914,
+      "learning_rate": 0.0001,
+      "loss": 4.3264,
+      "loss/crossentropy": 1.9853646159172058,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18827050924301147,
+      "step": 13732
+    },
+    {
+      "epoch": 0.27468,
+      "grad_norm": 1.7890625,
+      "grad_norm_var": 0.0074045817057291664,
+      "learning_rate": 0.0001,
+      "loss": 4.0074,
+      "loss/crossentropy": 1.7830212116241455,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1801598072052002,
+      "step": 13734
+    },
+    {
+      "epoch": 0.27472,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.008046213785807292,
+      "learning_rate": 0.0001,
+      "loss": 3.7923,
+      "loss/crossentropy": 1.7400763034820557,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1930072009563446,
+      "step": 13736
+    },
+    {
+      "epoch": 0.27476,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.008421834309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0992,
+      "loss/crossentropy": 2.2292455434799194,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20573973655700684,
+      "step": 13738
+    },
+    {
+      "epoch": 0.2748,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.009056599934895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4305,
+      "loss/crossentropy": 2.1399097442626953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2265445813536644,
+      "step": 13740
+    },
+    {
+      "epoch": 0.27484,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.009366861979166667,
+      "learning_rate": 0.0001,
+      "loss": 4.0596,
+      "loss/crossentropy": 2.049125075340271,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20623435080051422,
+      "step": 13742
+    },
+    {
+      "epoch": 0.27488,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010689036051432291,
+      "learning_rate": 0.0001,
+      "loss": 4.2678,
+      "loss/crossentropy": 2.388680577278137,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23650053143501282,
+      "step": 13744
+    },
+    {
+      "epoch": 0.27492,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.019577789306640624,
+      "learning_rate": 0.0001,
+      "loss": 4.5091,
+      "loss/crossentropy": 2.05421245098114,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19932237267494202,
+      "step": 13746
+    },
+    {
+      "epoch": 0.27496,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.01969172159830729,
+      "learning_rate": 0.0001,
+      "loss": 3.8274,
+      "loss/crossentropy": 1.892092227935791,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1907898187637329,
+      "step": 13748
+    },
+    {
+      "epoch": 0.275,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.018507639567057293,
+      "learning_rate": 0.0001,
+      "loss": 4.2196,
+      "loss/crossentropy": 2.160037875175476,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24229361861944199,
+      "step": 13750
+    },
+    {
+      "epoch": 0.27504,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.017451985677083334,
+      "learning_rate": 0.0001,
+      "loss": 4.1278,
+      "loss/crossentropy": 2.5185710191726685,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2394627332687378,
+      "step": 13752
+    },
+    {
+      "epoch": 0.27508,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.01754735310872396,
+      "learning_rate": 0.0001,
+      "loss": 4.1551,
+      "loss/crossentropy": 2.0613635778427124,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19758722186088562,
+      "step": 13754
+    },
+    {
+      "epoch": 0.27512,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.019496409098307292,
+      "learning_rate": 0.0001,
+      "loss": 4.1905,
+      "loss/crossentropy": 1.9780999422073364,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21773972362279892,
+      "step": 13756
+    },
+    {
+      "epoch": 0.27516,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.01987482706705729,
+      "learning_rate": 0.0001,
+      "loss": 4.3297,
+      "loss/crossentropy": 2.1116772890090942,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20809265226125717,
+      "step": 13758
+    },
+    {
+      "epoch": 0.2752,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.021201324462890626,
+      "learning_rate": 0.0001,
+      "loss": 4.2079,
+      "loss/crossentropy": 2.1072784662246704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21189533174037933,
+      "step": 13760
+    },
+    {
+      "epoch": 0.27524,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.014902496337890625,
+      "learning_rate": 0.0001,
+      "loss": 3.9102,
+      "loss/crossentropy": 2.261967897415161,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2058071494102478,
+      "step": 13762
+    },
+    {
+      "epoch": 0.27528,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.01571044921875,
+      "learning_rate": 0.0001,
+      "loss": 4.0684,
+      "loss/crossentropy": 2.240622043609619,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2305934578180313,
+      "step": 13764
+    },
+    {
+      "epoch": 0.27532,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.013874308268229166,
+      "learning_rate": 0.0001,
+      "loss": 3.8268,
+      "loss/crossentropy": 1.6203233003616333,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17164954543113708,
+      "step": 13766
+    },
+    {
+      "epoch": 0.27536,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.013521067301432292,
+      "learning_rate": 0.0001,
+      "loss": 3.928,
+      "loss/crossentropy": 2.171602725982666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20083081722259521,
+      "step": 13768
+    },
+    {
+      "epoch": 0.2754,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.012645467122395834,
+      "learning_rate": 0.0001,
+      "loss": 4.2738,
+      "loss/crossentropy": 2.2198195457458496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21113458275794983,
+      "step": 13770
+    },
+    {
+      "epoch": 0.27544,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.012296549479166667,
+      "learning_rate": 0.0001,
+      "loss": 4.2833,
+      "loss/crossentropy": 2.1037912368774414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21217171102762222,
+      "step": 13772
+    },
+    {
+      "epoch": 0.27548,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.012015533447265626,
+      "learning_rate": 0.0001,
+      "loss": 3.9824,
+      "loss/crossentropy": 2.252090096473694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20365406572818756,
+      "step": 13774
+    },
+    {
+      "epoch": 0.27552,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.011701456705729167,
+      "learning_rate": 0.0001,
+      "loss": 3.9513,
+      "loss/crossentropy": 1.8044906258583069,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1854797750711441,
+      "step": 13776
+    },
+    {
+      "epoch": 0.27556,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.012284342447916667,
+      "learning_rate": 0.0001,
+      "loss": 4.5229,
+      "loss/crossentropy": 2.2397952675819397,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.214948832988739,
+      "step": 13778
+    },
+    {
+      "epoch": 0.2756,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.0164215087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3569,
+      "loss/crossentropy": 1.776510238647461,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19995201379060745,
+      "step": 13780
+    },
+    {
+      "epoch": 0.27564,
+      "grad_norm": 4.0625,
+      "grad_norm_var": 0.26768290201822914,
+      "learning_rate": 0.0001,
+      "loss": 4.3061,
+      "loss/crossentropy": 2.470343589782715,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2194378450512886,
+      "step": 13782
+    },
+    {
+      "epoch": 0.27568,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.26253433227539064,
+      "learning_rate": 0.0001,
+      "loss": 4.0708,
+      "loss/crossentropy": 2.1340490579605103,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20280296355485916,
+      "step": 13784
+    },
+    {
+      "epoch": 0.27572,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.26368179321289065,
+      "learning_rate": 0.0001,
+      "loss": 4.1072,
+      "loss/crossentropy": 1.6801128387451172,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16750500351190567,
+      "step": 13786
+    },
+    {
+      "epoch": 0.27576,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.26387939453125,
+      "learning_rate": 0.0001,
+      "loss": 3.9902,
+      "loss/crossentropy": 1.9632562398910522,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19115211814641953,
+      "step": 13788
+    },
+    {
+      "epoch": 0.2758,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.25916519165039065,
+      "learning_rate": 0.0001,
+      "loss": 4.2139,
+      "loss/crossentropy": 2.0817149877548218,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21430402994155884,
+      "step": 13790
+    },
+    {
+      "epoch": 0.27584,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.253088124593099,
+      "learning_rate": 0.0001,
+      "loss": 4.3244,
+      "loss/crossentropy": 2.2736687660217285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22024155408143997,
+      "step": 13792
+    },
+    {
+      "epoch": 0.27588,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.25625712076822915,
+      "learning_rate": 0.0001,
+      "loss": 4.2548,
+      "loss/crossentropy": 2.2632944583892822,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.210355743765831,
+      "step": 13794
+    },
+    {
+      "epoch": 0.27592,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.26166966756184895,
+      "learning_rate": 0.0001,
+      "loss": 4.2799,
+      "loss/crossentropy": 2.5457879304885864,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22530706226825714,
+      "step": 13796
+    },
+    {
+      "epoch": 0.27596,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.008410390218098958,
+      "learning_rate": 0.0001,
+      "loss": 3.8969,
+      "loss/crossentropy": 2.0878910422325134,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19901857525110245,
+      "step": 13798
+    },
+    {
+      "epoch": 0.276,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008348592122395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3213,
+      "loss/crossentropy": 1.803489863872528,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19378191232681274,
+      "step": 13800
+    },
+    {
+      "epoch": 0.27604,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.009596506754557291,
+      "learning_rate": 0.0001,
+      "loss": 4.1916,
+      "loss/crossentropy": 2.156951904296875,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2091423124074936,
+      "step": 13802
+    },
+    {
+      "epoch": 0.27608,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.010493977864583334,
+      "learning_rate": 0.0001,
+      "loss": 3.9346,
+      "loss/crossentropy": 1.7714723944664001,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20779500901699066,
+      "step": 13804
+    },
+    {
+      "epoch": 0.27612,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.009590403238932291,
+      "learning_rate": 0.0001,
+      "loss": 4.0499,
+      "loss/crossentropy": 1.8808923363685608,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19429123401641846,
+      "step": 13806
+    },
+    {
+      "epoch": 0.27616,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.005516560872395834,
+      "learning_rate": 0.0001,
+      "loss": 3.706,
+      "loss/crossentropy": 1.6658846735954285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17177510261535645,
+      "step": 13808
+    },
+    {
+      "epoch": 0.2762,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.018697102864583332,
+      "learning_rate": 0.0001,
+      "loss": 4.0909,
+      "loss/crossentropy": 2.052769422531128,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20889723300933838,
+      "step": 13810
+    },
+    {
+      "epoch": 0.27624,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.18883031209309895,
+      "learning_rate": 0.0001,
+      "loss": 4.0225,
+      "loss/crossentropy": 1.5800148844718933,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24654322117567062,
+      "step": 13812
+    },
+    {
+      "epoch": 0.27628,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.5619504292805989,
+      "learning_rate": 0.0001,
+      "loss": 4.1059,
+      "loss/crossentropy": 1.836085557937622,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19563084840774536,
+      "step": 13814
+    },
+    {
+      "epoch": 0.27632,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.5601722717285156,
+      "learning_rate": 0.0001,
+      "loss": 4.0077,
+      "loss/crossentropy": 1.627321183681488,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17698387801647186,
+      "step": 13816
+    },
+    {
+      "epoch": 0.27636,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.55164794921875,
+      "learning_rate": 0.0001,
+      "loss": 4.2038,
+      "loss/crossentropy": 2.0546024441719055,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19832595437765121,
+      "step": 13818
+    },
+    {
+      "epoch": 0.2764,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.5499501546223958,
+      "learning_rate": 0.0001,
+      "loss": 4.064,
+      "loss/crossentropy": 2.112219452857971,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19605965912342072,
+      "step": 13820
+    },
+    {
+      "epoch": 0.27644,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.5405006408691406,
+      "learning_rate": 0.0001,
+      "loss": 4.3825,
+      "loss/crossentropy": 1.9582284688949585,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2015855312347412,
+      "step": 13822
+    },
+    {
+      "epoch": 0.27648,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.5307573954264323,
+      "learning_rate": 0.0001,
+      "loss": 4.2699,
+      "loss/crossentropy": 2.54653799533844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23546704649925232,
+      "step": 13824
+    },
+    {
+      "epoch": 0.27652,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.5336090087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0729,
+      "loss/crossentropy": 1.630593478679657,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16478916257619858,
+      "step": 13826
+    },
+    {
+      "epoch": 0.27656,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.40966796875,
+      "learning_rate": 0.0001,
+      "loss": 4.0123,
+      "loss/crossentropy": 1.7995671033859253,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19886638224124908,
+      "step": 13828
+    },
+    {
+      "epoch": 0.2766,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0177886962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4547,
+      "loss/crossentropy": 2.5423338413238525,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28057335317134857,
+      "step": 13830
+    },
+    {
+      "epoch": 0.27664,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.018631744384765624,
+      "learning_rate": 0.0001,
+      "loss": 4.0213,
+      "loss/crossentropy": 1.8524783849716187,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20141787081956863,
+      "step": 13832
+    },
+    {
+      "epoch": 0.27668,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.019440714518229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2155,
+      "loss/crossentropy": 2.242451548576355,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21003998070955276,
+      "step": 13834
+    },
+    {
+      "epoch": 0.27672,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0192047119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2665,
+      "loss/crossentropy": 2.2434345483779907,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21679577976465225,
+      "step": 13836
+    },
+    {
+      "epoch": 0.27676,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.019551595052083332,
+      "learning_rate": 0.0001,
+      "loss": 4.014,
+      "loss/crossentropy": 1.6854392290115356,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19326400756835938,
+      "step": 13838
+    },
+    {
+      "epoch": 0.2768,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.026668294270833334,
+      "learning_rate": 0.0001,
+      "loss": 4.2837,
+      "loss/crossentropy": 2.0702155232429504,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2043909877538681,
+      "step": 13840
+    },
+    {
+      "epoch": 0.27684,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.024738566080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0732,
+      "loss/crossentropy": 1.964758813381195,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19540227204561234,
+      "step": 13842
+    },
+    {
+      "epoch": 0.27688,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0151611328125,
+      "learning_rate": 0.0001,
+      "loss": 4.1681,
+      "loss/crossentropy": 2.0383411645889282,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21456278860569,
+      "step": 13844
+    },
+    {
+      "epoch": 0.27692,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.015461222330729166,
+      "learning_rate": 0.0001,
+      "loss": 4.1407,
+      "loss/crossentropy": 2.020030975341797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20521440356969833,
+      "step": 13846
+    },
+    {
+      "epoch": 0.27696,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.016110992431640624,
+      "learning_rate": 0.0001,
+      "loss": 4.1142,
+      "loss/crossentropy": 1.7514970302581787,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19116375595331192,
+      "step": 13848
+    },
+    {
+      "epoch": 0.277,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.016437784830729166,
+      "learning_rate": 0.0001,
+      "loss": 3.9209,
+      "loss/crossentropy": 1.8194025754928589,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18198709934949875,
+      "step": 13850
+    },
+    {
+      "epoch": 0.27704,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.017223866780598958,
+      "learning_rate": 0.0001,
+      "loss": 3.7911,
+      "loss/crossentropy": 1.878225862979889,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18159592151641846,
+      "step": 13852
+    },
+    {
+      "epoch": 0.27708,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0178619384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.008,
+      "loss/crossentropy": 1.8619664311408997,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18098927289247513,
+      "step": 13854
+    },
+    {
+      "epoch": 0.27712,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.003474934895833333,
+      "learning_rate": 0.0001,
+      "loss": 4.2336,
+      "loss/crossentropy": 2.0475016832351685,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19999271631240845,
+      "step": 13856
+    },
+    {
+      "epoch": 0.27716,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.004443105061848958,
+      "learning_rate": 0.0001,
+      "loss": 4.2412,
+      "loss/crossentropy": 2.2279679775238037,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21767108887434006,
+      "step": 13858
+    },
+    {
+      "epoch": 0.2772,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0039670308430989586,
+      "learning_rate": 0.0001,
+      "loss": 3.8087,
+      "loss/crossentropy": 1.801176130771637,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18873406946659088,
+      "step": 13860
+    },
+    {
+      "epoch": 0.27724,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.005303700764973958,
+      "learning_rate": 0.0001,
+      "loss": 3.7958,
+      "loss/crossentropy": 1.948195457458496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19519966840744019,
+      "step": 13862
+    },
+    {
+      "epoch": 0.27728,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0050432840983072914,
+      "learning_rate": 0.0001,
+      "loss": 4.0754,
+      "loss/crossentropy": 2.1044042110443115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20454693585634232,
+      "step": 13864
+    },
+    {
+      "epoch": 0.27732,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0051910400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1338,
+      "loss/crossentropy": 2.383382737636566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2179596871137619,
+      "step": 13866
+    },
+    {
+      "epoch": 0.27736,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.004621378580729167,
+      "learning_rate": 0.0001,
+      "loss": 3.9475,
+      "loss/crossentropy": 1.7798078656196594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17871373891830444,
+      "step": 13868
+    },
+    {
+      "epoch": 0.2774,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.004756418863932291,
+      "learning_rate": 0.0001,
+      "loss": 3.9937,
+      "loss/crossentropy": 2.1839531660079956,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20030800253152847,
+      "step": 13870
+    },
+    {
+      "epoch": 0.27744,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.005028279622395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1637,
+      "loss/crossentropy": 1.990351676940918,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1951240971684456,
+      "step": 13872
+    },
+    {
+      "epoch": 0.27748,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.006573232014973959,
+      "learning_rate": 0.0001,
+      "loss": 4.2919,
+      "loss/crossentropy": 2.191170334815979,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.230004720389843,
+      "step": 13874
+    },
+    {
+      "epoch": 0.27752,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.015726470947265626,
+      "learning_rate": 0.0001,
+      "loss": 4.1432,
+      "loss/crossentropy": 1.8893607258796692,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2138589546084404,
+      "step": 13876
+    },
+    {
+      "epoch": 0.27756,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.012483723958333333,
+      "learning_rate": 0.0001,
+      "loss": 4.2749,
+      "loss/crossentropy": 2.059523820877075,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20147833973169327,
+      "step": 13878
+    },
+    {
+      "epoch": 0.2776,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.013206990559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4959,
+      "loss/crossentropy": 2.0867209434509277,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23967817425727844,
+      "step": 13880
+    },
+    {
+      "epoch": 0.27764,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.028393300374348958,
+      "learning_rate": 0.0001,
+      "loss": 4.5704,
+      "loss/crossentropy": 2.3017314672470093,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23674342036247253,
+      "step": 13882
+    },
+    {
+      "epoch": 0.27768,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0261138916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3316,
+      "loss/crossentropy": 2.241922974586487,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21731412410736084,
+      "step": 13884
+    },
+    {
+      "epoch": 0.27772,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.025724283854166665,
+      "learning_rate": 0.0001,
+      "loss": 4.3317,
+      "loss/crossentropy": 2.3800796270370483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21916473656892776,
+      "step": 13886
+    },
+    {
+      "epoch": 0.27776,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.02467625935872396,
+      "learning_rate": 0.0001,
+      "loss": 4.2972,
+      "loss/crossentropy": 1.966018259525299,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21862812340259552,
+      "step": 13888
+    },
+    {
+      "epoch": 0.2778,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.030104319254557293,
+      "learning_rate": 0.0001,
+      "loss": 4.1955,
+      "loss/crossentropy": 1.920684039592743,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2026856392621994,
+      "step": 13890
+    },
+    {
+      "epoch": 0.27784,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.02601318359375,
+      "learning_rate": 0.0001,
+      "loss": 3.9662,
+      "loss/crossentropy": 2.1939095854759216,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20429420471191406,
+      "step": 13892
+    },
+    {
+      "epoch": 0.27788,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.026292928059895835,
+      "learning_rate": 0.0001,
+      "loss": 4.2335,
+      "loss/crossentropy": 2.1600992679595947,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20445890724658966,
+      "step": 13894
+    },
+    {
+      "epoch": 0.27792,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.026805623372395834,
+      "learning_rate": 0.0001,
+      "loss": 3.9973,
+      "loss/crossentropy": 1.9898765683174133,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19056615233421326,
+      "step": 13896
+    },
+    {
+      "epoch": 0.27796,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.008861287434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3037,
+      "loss/crossentropy": 2.0497539043426514,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1878463476896286,
+      "step": 13898
+    },
+    {
+      "epoch": 0.278,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.007373046875,
+      "learning_rate": 0.0001,
+      "loss": 4.2299,
+      "loss/crossentropy": 1.894934356212616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.183290496468544,
+      "step": 13900
+    },
+    {
+      "epoch": 0.27804,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.010094960530598959,
+      "learning_rate": 0.0001,
+      "loss": 3.858,
+      "loss/crossentropy": 1.808231770992279,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19447006285190582,
+      "step": 13902
+    },
+    {
+      "epoch": 0.27808,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009688059488932291,
+      "learning_rate": 0.0001,
+      "loss": 4.0553,
+      "loss/crossentropy": 1.9384279251098633,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20293861627578735,
+      "step": 13904
+    },
+    {
+      "epoch": 0.27812,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.012835439046223958,
+      "learning_rate": 0.0001,
+      "loss": 4.3092,
+      "loss/crossentropy": 2.4291017055511475,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24332769960165024,
+      "step": 13906
+    },
+    {
+      "epoch": 0.27816,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.012669881184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.331,
+      "loss/crossentropy": 2.3018234968185425,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21657106280326843,
+      "step": 13908
+    },
+    {
+      "epoch": 0.2782,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.0204742431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1756,
+      "loss/crossentropy": 2.211503267288208,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19193635880947113,
+      "step": 13910
+    },
+    {
+      "epoch": 0.27824,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.02191162109375,
+      "learning_rate": 0.0001,
+      "loss": 4.372,
+      "loss/crossentropy": 2.17062246799469,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21989908069372177,
+      "step": 13912
+    },
+    {
+      "epoch": 0.27828,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.018839518229166668,
+      "learning_rate": 0.0001,
+      "loss": 4.0895,
+      "loss/crossentropy": 2.1446024775505066,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2036902904510498,
+      "step": 13914
+    },
+    {
+      "epoch": 0.27832,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0177642822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0729,
+      "loss/crossentropy": 1.8176022171974182,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19358078390359879,
+      "step": 13916
+    },
+    {
+      "epoch": 0.27836,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.01946996053059896,
+      "learning_rate": 0.0001,
+      "loss": 4.2376,
+      "loss/crossentropy": 2.341936469078064,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22822833061218262,
+      "step": 13918
+    },
+    {
+      "epoch": 0.2784,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.01697565714518229,
+      "learning_rate": 0.0001,
+      "loss": 4.0943,
+      "loss/crossentropy": 1.9315840601921082,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17218464612960815,
+      "step": 13920
+    },
+    {
+      "epoch": 0.27844,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.01615168253580729,
+      "learning_rate": 0.0001,
+      "loss": 4.1507,
+      "loss/crossentropy": 1.9442673921585083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19567463546991348,
+      "step": 13922
+    },
+    {
+      "epoch": 0.27848,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.016673787434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1611,
+      "loss/crossentropy": 2.3023592233657837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21824805438518524,
+      "step": 13924
+    },
+    {
+      "epoch": 0.27852,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.01170654296875,
+      "learning_rate": 0.0001,
+      "loss": 3.7557,
+      "loss/crossentropy": 2.116863250732422,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2087356001138687,
+      "step": 13926
+    },
+    {
+      "epoch": 0.27856,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.010936482747395834,
+      "learning_rate": 0.0001,
+      "loss": 4.3746,
+      "loss/crossentropy": 2.0485963821411133,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18910974264144897,
+      "step": 13928
+    },
+    {
+      "epoch": 0.2786,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.024857584635416666,
+      "learning_rate": 0.0001,
+      "loss": 4.1309,
+      "loss/crossentropy": 2.031753659248352,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23851437866687775,
+      "step": 13930
+    },
+    {
+      "epoch": 0.27864,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.025585683186848958,
+      "learning_rate": 0.0001,
+      "loss": 4.0688,
+      "loss/crossentropy": 1.8019705414772034,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18659329414367676,
+      "step": 13932
+    },
+    {
+      "epoch": 0.27868,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.025243123372395832,
+      "learning_rate": 0.0001,
+      "loss": 3.9609,
+      "loss/crossentropy": 1.9191861152648926,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19393808394670486,
+      "step": 13934
+    },
+    {
+      "epoch": 0.27872,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.025763956705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.1765,
+      "loss/crossentropy": 2.242986798286438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.213445246219635,
+      "step": 13936
+    },
+    {
+      "epoch": 0.27876,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0300689697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1767,
+      "loss/crossentropy": 2.1652570962905884,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2070690169930458,
+      "step": 13938
+    },
+    {
+      "epoch": 0.2788,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.031583404541015624,
+      "learning_rate": 0.0001,
+      "loss": 3.9643,
+      "loss/crossentropy": 2.0970187187194824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20118873566389084,
+      "step": 13940
+    },
+    {
+      "epoch": 0.27884,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.02823053995768229,
+      "learning_rate": 0.0001,
+      "loss": 4.1116,
+      "loss/crossentropy": 1.9957542419433594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2377041131258011,
+      "step": 13942
+    },
+    {
+      "epoch": 0.27888,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.02522761027018229,
+      "learning_rate": 0.0001,
+      "loss": 4.0718,
+      "loss/crossentropy": 1.8680259585380554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1919926255941391,
+      "step": 13944
+    },
+    {
+      "epoch": 0.27892,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.015516916910807291,
+      "learning_rate": 0.0001,
+      "loss": 4.301,
+      "loss/crossentropy": 2.369240164756775,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24805612862110138,
+      "step": 13946
+    },
+    {
+      "epoch": 0.27896,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.014530436197916666,
+      "learning_rate": 0.0001,
+      "loss": 4.3082,
+      "loss/crossentropy": 2.0923795104026794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20823675394058228,
+      "step": 13948
+    },
+    {
+      "epoch": 0.279,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.013197580973307291,
+      "learning_rate": 0.0001,
+      "loss": 4.3049,
+      "loss/crossentropy": 2.1777398586273193,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21319355070590973,
+      "step": 13950
+    },
+    {
+      "epoch": 0.27904,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.014869944254557291,
+      "learning_rate": 0.0001,
+      "loss": 4.0918,
+      "loss/crossentropy": 2.2155595421791077,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2029823139309883,
+      "step": 13952
+    },
+    {
+      "epoch": 0.27908,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.011435699462890626,
+      "learning_rate": 0.0001,
+      "loss": 3.7195,
+      "loss/crossentropy": 1.5916873812675476,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18207873404026031,
+      "step": 13954
+    },
+    {
+      "epoch": 0.27912,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.009791819254557292,
+      "learning_rate": 0.0001,
+      "loss": 4.3336,
+      "loss/crossentropy": 2.165639281272888,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22917281091213226,
+      "step": 13956
+    },
+    {
+      "epoch": 0.27916,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.012851715087890625,
+      "learning_rate": 0.0001,
+      "loss": 3.9586,
+      "loss/crossentropy": 2.0245776772499084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19862860441207886,
+      "step": 13958
+    },
+    {
+      "epoch": 0.2792,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.013646443684895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1593,
+      "loss/crossentropy": 2.0179941654205322,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20749768614768982,
+      "step": 13960
+    },
+    {
+      "epoch": 0.27924,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.011966705322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3978,
+      "loss/crossentropy": 1.998136818408966,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1992153376340866,
+      "step": 13962
+    },
+    {
+      "epoch": 0.27928,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.015346018473307292,
+      "learning_rate": 0.0001,
+      "loss": 4.1801,
+      "loss/crossentropy": 2.041996657848358,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22224828600883484,
+      "step": 13964
+    },
+    {
+      "epoch": 0.27932,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.019406890869140624,
+      "learning_rate": 0.0001,
+      "loss": 4.4133,
+      "loss/crossentropy": 2.0259117484092712,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2001984491944313,
+      "step": 13966
+    },
+    {
+      "epoch": 0.27936,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.018436686197916666,
+      "learning_rate": 0.0001,
+      "loss": 4.0439,
+      "loss/crossentropy": 1.8219285607337952,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19637393951416016,
+      "step": 13968
+    },
+    {
+      "epoch": 0.2794,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.017073567708333334,
+      "learning_rate": 0.0001,
+      "loss": 4.0294,
+      "loss/crossentropy": 2.1451289653778076,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21846695989370346,
+      "step": 13970
+    },
+    {
+      "epoch": 0.27944,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.016877237955729166,
+      "learning_rate": 0.0001,
+      "loss": 4.349,
+      "loss/crossentropy": 2.0527132749557495,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2147705778479576,
+      "step": 13972
+    },
+    {
+      "epoch": 0.27948,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.015143839518229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0188,
+      "loss/crossentropy": 1.9238123893737793,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21083690226078033,
+      "step": 13974
+    },
+    {
+      "epoch": 0.27952,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.015057118733723958,
+      "learning_rate": 0.0001,
+      "loss": 4.0072,
+      "loss/crossentropy": 1.894473671913147,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20077680051326752,
+      "step": 13976
+    },
+    {
+      "epoch": 0.27956,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.013871256510416667,
+      "learning_rate": 0.0001,
+      "loss": 4.2344,
+      "loss/crossentropy": 2.139513611793518,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2138260081410408,
+      "step": 13978
+    },
+    {
+      "epoch": 0.2796,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.013468424479166666,
+      "learning_rate": 0.0001,
+      "loss": 4.4519,
+      "loss/crossentropy": 2.4868407249450684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23635651916265488,
+      "step": 13980
+    },
+    {
+      "epoch": 0.27964,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.007940419514973958,
+      "learning_rate": 0.0001,
+      "loss": 3.7428,
+      "loss/crossentropy": 1.426945686340332,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16342981159687042,
+      "step": 13982
+    },
+    {
+      "epoch": 0.27968,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.007885487874348958,
+      "learning_rate": 0.0001,
+      "loss": 4.0457,
+      "loss/crossentropy": 2.0577683448791504,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20394782721996307,
+      "step": 13984
+    },
+    {
+      "epoch": 0.27972,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.008063761393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0045,
+      "loss/crossentropy": 1.884658396244049,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18946699798107147,
+      "step": 13986
+    },
+    {
+      "epoch": 0.27976,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0070383707682291664,
+      "learning_rate": 0.0001,
+      "loss": 3.9423,
+      "loss/crossentropy": 1.7344964742660522,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1923774629831314,
+      "step": 13988
+    },
+    {
+      "epoch": 0.2798,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.00888671875,
+      "learning_rate": 0.0001,
+      "loss": 3.9541,
+      "loss/crossentropy": 1.9660141468048096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18560642004013062,
+      "step": 13990
+    },
+    {
+      "epoch": 0.27984,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.008998362223307292,
+      "learning_rate": 0.0001,
+      "loss": 3.9308,
+      "loss/crossentropy": 1.7832527160644531,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19922740012407303,
+      "step": 13992
+    },
+    {
+      "epoch": 0.27988,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009987131754557291,
+      "learning_rate": 0.0001,
+      "loss": 4.3919,
+      "loss/crossentropy": 1.9763087630271912,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2111397087574005,
+      "step": 13994
+    },
+    {
+      "epoch": 0.27992,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.005252838134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1331,
+      "loss/crossentropy": 2.262092709541321,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21212221682071686,
+      "step": 13996
+    },
+    {
+      "epoch": 0.27996,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0054595947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1946,
+      "loss/crossentropy": 2.237086296081543,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20895393192768097,
+      "step": 13998
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.0061337788899739586,
+      "learning_rate": 0.0001,
+      "loss": 4.063,
+      "loss/crossentropy": 1.9579973816871643,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18024132400751114,
+      "step": 14000
+    },
+    {
+      "epoch": 0.28004,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0055844624837239586,
+      "learning_rate": 0.0001,
+      "loss": 4.1463,
+      "loss/crossentropy": 2.2264128923416138,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2181634083390236,
+      "step": 14002
+    },
+    {
+      "epoch": 0.28008,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0058013916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0081,
+      "loss/crossentropy": 1.836763322353363,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1983085498213768,
+      "step": 14004
+    },
+    {
+      "epoch": 0.28012,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.005448404947916667,
+      "learning_rate": 0.0001,
+      "loss": 4.1192,
+      "loss/crossentropy": 2.0163257718086243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.214015431702137,
+      "step": 14006
+    },
+    {
+      "epoch": 0.28016,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0070383707682291664,
+      "learning_rate": 0.0001,
+      "loss": 4.139,
+      "loss/crossentropy": 2.31567645072937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22832776606082916,
+      "step": 14008
+    },
+    {
+      "epoch": 0.2802,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.006685129801432292,
+      "learning_rate": 0.0001,
+      "loss": 4.0753,
+      "loss/crossentropy": 2.3515210151672363,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21946918964385986,
+      "step": 14010
+    },
+    {
+      "epoch": 0.28024,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.006845855712890625,
+      "learning_rate": 0.0001,
+      "loss": 3.9761,
+      "loss/crossentropy": 2.0987448692321777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2180740088224411,
+      "step": 14012
+    },
+    {
+      "epoch": 0.28028,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.00662841796875,
+      "learning_rate": 0.0001,
+      "loss": 4.0909,
+      "loss/crossentropy": 2.2203436493873596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2157250940799713,
+      "step": 14014
+    },
+    {
+      "epoch": 0.28032,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.006241861979166667,
+      "learning_rate": 0.0001,
+      "loss": 4.3152,
+      "loss/crossentropy": 2.083233594894409,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2187238186597824,
+      "step": 14016
+    },
+    {
+      "epoch": 0.28036,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0061075846354166664,
+      "learning_rate": 0.0001,
+      "loss": 4.1311,
+      "loss/crossentropy": 2.1261327266693115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2063112109899521,
+      "step": 14018
+    },
+    {
+      "epoch": 0.2804,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.006648508707682291,
+      "learning_rate": 0.0001,
+      "loss": 4.0901,
+      "loss/crossentropy": 2.012390434741974,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20225800573825836,
+      "step": 14020
+    },
+    {
+      "epoch": 0.28044,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0047271728515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9985,
+      "loss/crossentropy": 1.8605966567993164,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2003621682524681,
+      "step": 14022
+    },
+    {
+      "epoch": 0.28048,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0028157552083333333,
+      "learning_rate": 0.0001,
+      "loss": 3.9774,
+      "loss/crossentropy": 1.9208670258522034,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20056191086769104,
+      "step": 14024
+    },
+    {
+      "epoch": 0.28052,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.003951009114583333,
+      "learning_rate": 0.0001,
+      "loss": 3.7462,
+      "loss/crossentropy": 1.924518644809723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19410791993141174,
+      "step": 14026
+    },
+    {
+      "epoch": 0.28056,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0035113016764322918,
+      "learning_rate": 0.0001,
+      "loss": 4.0319,
+      "loss/crossentropy": 2.103038251399994,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2054024636745453,
+      "step": 14028
+    },
+    {
+      "epoch": 0.2806,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.004303995768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0526,
+      "loss/crossentropy": 1.8974853157997131,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18592579662799835,
+      "step": 14030
+    },
+    {
+      "epoch": 0.28064,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.004447428385416666,
+      "learning_rate": 0.0001,
+      "loss": 4.1735,
+      "loss/crossentropy": 2.0207908749580383,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.216909758746624,
+      "step": 14032
+    },
+    {
+      "epoch": 0.28068,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.006119537353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5759,
+      "loss/crossentropy": 2.0010873079299927,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20815415680408478,
+      "step": 14034
+    },
+    {
+      "epoch": 0.28072,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.006961822509765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9162,
+      "loss/crossentropy": 2.0840484499931335,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21481210738420486,
+      "step": 14036
+    },
+    {
+      "epoch": 0.28076,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.006880442301432292,
+      "learning_rate": 0.0001,
+      "loss": 4.1298,
+      "loss/crossentropy": 2.0026179552078247,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1904347836971283,
+      "step": 14038
+    },
+    {
+      "epoch": 0.2808,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.006880442301432292,
+      "learning_rate": 0.0001,
+      "loss": 4.0618,
+      "loss/crossentropy": 2.1660486459732056,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21236423403024673,
+      "step": 14040
+    },
+    {
+      "epoch": 0.28084,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.007669830322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0474,
+      "loss/crossentropy": 2.312765598297119,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20535308122634888,
+      "step": 14042
+    },
+    {
+      "epoch": 0.28088,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0077789306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1959,
+      "loss/crossentropy": 1.8449203372001648,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1986582651734352,
+      "step": 14044
+    },
+    {
+      "epoch": 0.28092,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.008894602457682291,
+      "learning_rate": 0.0001,
+      "loss": 4.016,
+      "loss/crossentropy": 2.0159433484077454,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19724585115909576,
+      "step": 14046
+    },
+    {
+      "epoch": 0.28096,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0086090087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0691,
+      "loss/crossentropy": 2.138873815536499,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21438196301460266,
+      "step": 14048
+    },
+    {
+      "epoch": 0.281,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.011372629801432292,
+      "learning_rate": 0.0001,
+      "loss": 4.0985,
+      "loss/crossentropy": 1.6896708607673645,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19260118901729584,
+      "step": 14050
+    },
+    {
+      "epoch": 0.28104,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.014656321207682291,
+      "learning_rate": 0.0001,
+      "loss": 3.8383,
+      "loss/crossentropy": 1.8495243191719055,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18072611093521118,
+      "step": 14052
+    },
+    {
+      "epoch": 0.28108,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.017354329427083332,
+      "learning_rate": 0.0001,
+      "loss": 4.3581,
+      "loss/crossentropy": 2.193492293357849,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2247578352689743,
+      "step": 14054
+    },
+    {
+      "epoch": 0.28112,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.01761652628580729,
+      "learning_rate": 0.0001,
+      "loss": 4.1252,
+      "loss/crossentropy": 2.154744803905487,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20149191468954086,
+      "step": 14056
+    },
+    {
+      "epoch": 0.28116,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.016022745768229166,
+      "learning_rate": 0.0001,
+      "loss": 4.0832,
+      "loss/crossentropy": 2.137023687362671,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20409516990184784,
+      "step": 14058
+    },
+    {
+      "epoch": 0.2812,
+      "grad_norm": 2.96875,
+      "grad_norm_var": 0.0767242431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.6582,
+      "loss/crossentropy": 2.4530670642852783,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24031317979097366,
+      "step": 14060
+    },
+    {
+      "epoch": 0.28124,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.07713394165039063,
+      "learning_rate": 0.0001,
+      "loss": 4.2973,
+      "loss/crossentropy": 2.3285664319992065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.223993182182312,
+      "step": 14062
+    },
+    {
+      "epoch": 0.28128,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0784088134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0899,
+      "loss/crossentropy": 2.126828193664551,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21428050100803375,
+      "step": 14064
+    },
+    {
+      "epoch": 0.28132,
+      "grad_norm": 1.796875,
+      "grad_norm_var": 0.08012288411458333,
+      "learning_rate": 0.0001,
+      "loss": 3.8083,
+      "loss/crossentropy": 2.0194268226623535,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20124298334121704,
+      "step": 14066
+    },
+    {
+      "epoch": 0.28136,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.07445882161458334,
+      "learning_rate": 0.0001,
+      "loss": 4.0002,
+      "loss/crossentropy": 1.796358585357666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1902359575033188,
+      "step": 14068
+    },
+    {
+      "epoch": 0.2814,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.07318700154622396,
+      "learning_rate": 0.0001,
+      "loss": 4.2,
+      "loss/crossentropy": 1.8975006341934204,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19528520107269287,
+      "step": 14070
+    },
+    {
+      "epoch": 0.28144,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.07691141764322916,
+      "learning_rate": 0.0001,
+      "loss": 3.9885,
+      "loss/crossentropy": 1.628948986530304,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1823427528142929,
+      "step": 14072
+    },
+    {
+      "epoch": 0.28148,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.07718480428059896,
+      "learning_rate": 0.0001,
+      "loss": 4.0404,
+      "loss/crossentropy": 1.9894697070121765,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1945328414440155,
+      "step": 14074
+    },
+    {
+      "epoch": 0.28152,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.022489166259765624,
+      "learning_rate": 0.0001,
+      "loss": 4.0927,
+      "loss/crossentropy": 1.9978017210960388,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21055076271295547,
+      "step": 14076
+    },
+    {
+      "epoch": 0.28156,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.01697565714518229,
+      "learning_rate": 0.0001,
+      "loss": 3.9224,
+      "loss/crossentropy": 2.087389588356018,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19179877638816833,
+      "step": 14078
+    },
+    {
+      "epoch": 0.2816,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.018308258056640624,
+      "learning_rate": 0.0001,
+      "loss": 4.3356,
+      "loss/crossentropy": 2.3281190395355225,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2292996421456337,
+      "step": 14080
+    },
+    {
+      "epoch": 0.28164,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.015215810139973958,
+      "learning_rate": 0.0001,
+      "loss": 3.8584,
+      "loss/crossentropy": 1.8850311040878296,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19652675837278366,
+      "step": 14082
+    },
+    {
+      "epoch": 0.28168,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.015604654947916666,
+      "learning_rate": 0.0001,
+      "loss": 3.9013,
+      "loss/crossentropy": 1.757595181465149,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17358998954296112,
+      "step": 14084
+    },
+    {
+      "epoch": 0.28172,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.01387939453125,
+      "learning_rate": 0.0001,
+      "loss": 4.1367,
+      "loss/crossentropy": 2.173685908317566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22043757140636444,
+      "step": 14086
+    },
+    {
+      "epoch": 0.28176,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.010758463541666667,
+      "learning_rate": 0.0001,
+      "loss": 4.0381,
+      "loss/crossentropy": 2.0290868282318115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19695669412612915,
+      "step": 14088
+    },
+    {
+      "epoch": 0.2818,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010609690348307292,
+      "learning_rate": 0.0001,
+      "loss": 4.3497,
+      "loss/crossentropy": 2.320943236351013,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21747629344463348,
+      "step": 14090
+    },
+    {
+      "epoch": 0.28184,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.011205037434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.0918,
+      "loss/crossentropy": 2.0311750173568726,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19410807639360428,
+      "step": 14092
+    },
+    {
+      "epoch": 0.28188,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011207834879557291,
+      "learning_rate": 0.0001,
+      "loss": 4.1303,
+      "loss/crossentropy": 2.164630174636841,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1962611824274063,
+      "step": 14094
+    },
+    {
+      "epoch": 0.28192,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0093658447265625,
+      "learning_rate": 0.0001,
+      "loss": 3.9681,
+      "loss/crossentropy": 2.170566439628601,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24082274734973907,
+      "step": 14096
+    },
+    {
+      "epoch": 0.28196,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.009285227457682291,
+      "learning_rate": 0.0001,
+      "loss": 4.209,
+      "loss/crossentropy": 2.3137707710266113,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22475039958953857,
+      "step": 14098
+    },
+    {
+      "epoch": 0.282,
+      "grad_norm": 1.7421875,
+      "grad_norm_var": 0.010794830322265626,
+      "learning_rate": 0.0001,
+      "loss": 3.7432,
+      "loss/crossentropy": 2.0448675751686096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19814980030059814,
+      "step": 14100
+    },
+    {
+      "epoch": 0.28204,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.011423492431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1729,
+      "loss/crossentropy": 2.095334231853485,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19743670523166656,
+      "step": 14102
+    },
+    {
+      "epoch": 0.28208,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.008737945556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0362,
+      "loss/crossentropy": 2.280429720878601,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21895557641983032,
+      "step": 14104
+    },
+    {
+      "epoch": 0.28212,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.008292388916015626,
+      "learning_rate": 0.0001,
+      "loss": 4.1166,
+      "loss/crossentropy": 1.8548901677131653,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19965855032205582,
+      "step": 14106
+    },
+    {
+      "epoch": 0.28216,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.01163330078125,
+      "learning_rate": 0.0001,
+      "loss": 4.2589,
+      "loss/crossentropy": 2.3574371337890625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23160286247730255,
+      "step": 14108
+    },
+    {
+      "epoch": 0.2822,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.012800089518229167,
+      "learning_rate": 0.0001,
+      "loss": 3.932,
+      "loss/crossentropy": 2.112724542617798,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19597014784812927,
+      "step": 14110
+    },
+    {
+      "epoch": 0.28224,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.013068644205729167,
+      "learning_rate": 0.0001,
+      "loss": 3.8792,
+      "loss/crossentropy": 1.9877798557281494,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21120092272758484,
+      "step": 14112
+    },
+    {
+      "epoch": 0.28228,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.013199869791666667,
+      "learning_rate": 0.0001,
+      "loss": 4.1761,
+      "loss/crossentropy": 2.253028154373169,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2046835869550705,
+      "step": 14114
+    },
+    {
+      "epoch": 0.28232,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.008561197916666667,
+      "learning_rate": 0.0001,
+      "loss": 4.1711,
+      "loss/crossentropy": 2.322582960128784,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22178008407354355,
+      "step": 14116
+    },
+    {
+      "epoch": 0.28236,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008204905192057292,
+      "learning_rate": 0.0001,
+      "loss": 4.1584,
+      "loss/crossentropy": 2.470620036125183,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23360955715179443,
+      "step": 14118
+    },
+    {
+      "epoch": 0.2824,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0086334228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0206,
+      "loss/crossentropy": 1.9218884110450745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1978505551815033,
+      "step": 14120
+    },
+    {
+      "epoch": 0.28244,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0092437744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2045,
+      "loss/crossentropy": 1.9114753007888794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19295217841863632,
+      "step": 14122
+    },
+    {
+      "epoch": 0.28248,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.498583984375,
+      "learning_rate": 0.0001,
+      "loss": 4.2259,
+      "loss/crossentropy": 2.1196082830429077,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2068253606557846,
+      "step": 14124
+    },
+    {
+      "epoch": 0.28252,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.492419179280599,
+      "learning_rate": 0.0001,
+      "loss": 4.2616,
+      "loss/crossentropy": 1.9869969487190247,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1950489580631256,
+      "step": 14126
+    },
+    {
+      "epoch": 0.28256,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.488177235921224,
+      "learning_rate": 0.0001,
+      "loss": 4.1635,
+      "loss/crossentropy": 2.09599232673645,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22184181958436966,
+      "step": 14128
+    },
+    {
+      "epoch": 0.2826,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.489172108968099,
+      "learning_rate": 0.0001,
+      "loss": 4.2238,
+      "loss/crossentropy": 2.1891895532608032,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2229236587882042,
+      "step": 14130
+    },
+    {
+      "epoch": 0.28264,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.48787816365559894,
+      "learning_rate": 0.0001,
+      "loss": 4.12,
+      "loss/crossentropy": 2.5070972442626953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21862925589084625,
+      "step": 14132
+    },
+    {
+      "epoch": 0.28268,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.48812662760416664,
+      "learning_rate": 0.0001,
+      "loss": 4.0993,
+      "loss/crossentropy": 2.100346863269806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19969436526298523,
+      "step": 14134
+    },
+    {
+      "epoch": 0.28272,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.48713150024414065,
+      "learning_rate": 0.0001,
+      "loss": 4.2607,
+      "loss/crossentropy": 2.0019801259040833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21319210529327393,
+      "step": 14136
+    },
+    {
+      "epoch": 0.28276,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.4855323791503906,
+      "learning_rate": 0.0001,
+      "loss": 4.7069,
+      "loss/crossentropy": 2.2734099626541138,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24008513987064362,
+      "step": 14138
+    },
+    {
+      "epoch": 0.2828,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.02581965128580729,
+      "learning_rate": 0.0001,
+      "loss": 4.1085,
+      "loss/crossentropy": 2.2589277029037476,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2224387675523758,
+      "step": 14140
+    },
+    {
+      "epoch": 0.28284,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.02667236328125,
+      "learning_rate": 0.0001,
+      "loss": 4.2749,
+      "loss/crossentropy": 1.713346004486084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19133105129003525,
+      "step": 14142
+    },
+    {
+      "epoch": 0.28288,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.02616144816080729,
+      "learning_rate": 0.0001,
+      "loss": 4.1792,
+      "loss/crossentropy": 2.212652564048767,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21326624602079391,
+      "step": 14144
+    },
+    {
+      "epoch": 0.28292,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.02902399698893229,
+      "learning_rate": 0.0001,
+      "loss": 3.901,
+      "loss/crossentropy": 1.8900847434997559,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1882227212190628,
+      "step": 14146
+    },
+    {
+      "epoch": 0.28296,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.028913370768229165,
+      "learning_rate": 0.0001,
+      "loss": 4.3775,
+      "loss/crossentropy": 2.171455979347229,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20516864210367203,
+      "step": 14148
+    },
+    {
+      "epoch": 0.283,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.027608235677083332,
+      "learning_rate": 0.0001,
+      "loss": 4.3466,
+      "loss/crossentropy": 2.0112149715423584,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20043446868658066,
+      "step": 14150
+    },
+    {
+      "epoch": 0.28304,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.026596832275390624,
+      "learning_rate": 0.0001,
+      "loss": 4.2879,
+      "loss/crossentropy": 2.0091559886932373,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21023409068584442,
+      "step": 14152
+    },
+    {
+      "epoch": 0.28308,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.023045857747395832,
+      "learning_rate": 0.0001,
+      "loss": 3.9234,
+      "loss/crossentropy": 2.122502863407135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2023375853896141,
+      "step": 14154
+    },
+    {
+      "epoch": 0.28312,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.011462148030598958,
+      "learning_rate": 0.0001,
+      "loss": 4.3187,
+      "loss/crossentropy": 2.2842462062835693,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24532422423362732,
+      "step": 14156
+    },
+    {
+      "epoch": 0.28316,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010990397135416666,
+      "learning_rate": 0.0001,
+      "loss": 4.5248,
+      "loss/crossentropy": 2.247039318084717,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24580712616443634,
+      "step": 14158
+    },
+    {
+      "epoch": 0.2832,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.011432902018229166,
+      "learning_rate": 0.0001,
+      "loss": 3.9486,
+      "loss/crossentropy": 2.182216167449951,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21095345169305801,
+      "step": 14160
+    },
+    {
+      "epoch": 0.28324,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.008648427327473958,
+      "learning_rate": 0.0001,
+      "loss": 4.4582,
+      "loss/crossentropy": 2.219786763191223,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21747471392154694,
+      "step": 14162
+    },
+    {
+      "epoch": 0.28328,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.005830637613932292,
+      "learning_rate": 0.0001,
+      "loss": 3.9483,
+      "loss/crossentropy": 1.8589028716087341,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19629907608032227,
+      "step": 14164
+    },
+    {
+      "epoch": 0.28332,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0138092041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0217,
+      "loss/crossentropy": 1.9820821285247803,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20433218777179718,
+      "step": 14166
+    },
+    {
+      "epoch": 0.28336,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.014216868082682292,
+      "learning_rate": 0.0001,
+      "loss": 4.0081,
+      "loss/crossentropy": 1.8423896431922913,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18210452795028687,
+      "step": 14168
+    },
+    {
+      "epoch": 0.2834,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0158203125,
+      "learning_rate": 0.0001,
+      "loss": 4.1244,
+      "loss/crossentropy": 2.0365665555000305,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21229179203510284,
+      "step": 14170
+    },
+    {
+      "epoch": 0.28344,
+      "grad_norm": 1.7578125,
+      "grad_norm_var": 0.019212849934895835,
+      "learning_rate": 0.0001,
+      "loss": 3.7894,
+      "loss/crossentropy": 1.6819360256195068,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20171043276786804,
+      "step": 14172
+    },
+    {
+      "epoch": 0.28348,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.019846343994140626,
+      "learning_rate": 0.0001,
+      "loss": 4.1247,
+      "loss/crossentropy": 1.9932513236999512,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21374855190515518,
+      "step": 14174
+    },
+    {
+      "epoch": 0.28352,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.01975072224934896,
+      "learning_rate": 0.0001,
+      "loss": 3.8874,
+      "loss/crossentropy": 1.7137236595153809,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17538649588823318,
+      "step": 14176
+    },
+    {
+      "epoch": 0.28356,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.01977717081705729,
+      "learning_rate": 0.0001,
+      "loss": 3.8762,
+      "loss/crossentropy": 1.7120450735092163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19030355662107468,
+      "step": 14178
+    },
+    {
+      "epoch": 0.2836,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.019962565104166666,
+      "learning_rate": 0.0001,
+      "loss": 4.4603,
+      "loss/crossentropy": 2.346954107284546,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24070723354816437,
+      "step": 14180
+    },
+    {
+      "epoch": 0.28364,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.013755035400390626,
+      "learning_rate": 0.0001,
+      "loss": 4.2238,
+      "loss/crossentropy": 2.241390824317932,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2247970700263977,
+      "step": 14182
+    },
+    {
+      "epoch": 0.28368,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.02080078125,
+      "learning_rate": 0.0001,
+      "loss": 4.206,
+      "loss/crossentropy": 2.234646439552307,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21947349607944489,
+      "step": 14184
+    },
+    {
+      "epoch": 0.28372,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.019010416666666665,
+      "learning_rate": 0.0001,
+      "loss": 4.0446,
+      "loss/crossentropy": 1.9331231117248535,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20428168773651123,
+      "step": 14186
+    },
+    {
+      "epoch": 0.28376,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.015315500895182292,
+      "learning_rate": 0.0001,
+      "loss": 4.0184,
+      "loss/crossentropy": 1.8166351318359375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21285146474838257,
+      "step": 14188
+    },
+    {
+      "epoch": 0.2838,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.015077463785807292,
+      "learning_rate": 0.0001,
+      "loss": 4.313,
+      "loss/crossentropy": 2.181701421737671,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20924244076013565,
+      "step": 14190
+    },
+    {
+      "epoch": 0.28384,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.015558878580729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0866,
+      "loss/crossentropy": 2.1605955958366394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24021095037460327,
+      "step": 14192
+    },
+    {
+      "epoch": 0.28388,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.015455881754557291,
+      "learning_rate": 0.0001,
+      "loss": 4.1617,
+      "loss/crossentropy": 2.3546024560928345,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24228987097740173,
+      "step": 14194
+    },
+    {
+      "epoch": 0.28392,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.015314737955729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0791,
+      "loss/crossentropy": 2.2335511445999146,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21917300671339035,
+      "step": 14196
+    },
+    {
+      "epoch": 0.28396,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.012760416666666666,
+      "learning_rate": 0.0001,
+      "loss": 3.9915,
+      "loss/crossentropy": 1.8429189324378967,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18830660730600357,
+      "step": 14198
+    },
+    {
+      "epoch": 0.284,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0058746337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2671,
+      "loss/crossentropy": 2.4578241109848022,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23527230322360992,
+      "step": 14200
+    },
+    {
+      "epoch": 0.28404,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.00592041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3027,
+      "loss/crossentropy": 2.0944234132766724,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2211083546280861,
+      "step": 14202
+    },
+    {
+      "epoch": 0.28408,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0066070556640625,
+      "learning_rate": 0.0001,
+      "loss": 3.9341,
+      "loss/crossentropy": 1.8876588344573975,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1908133327960968,
+      "step": 14204
+    },
+    {
+      "epoch": 0.28412,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.006190745035807291,
+      "learning_rate": 0.0001,
+      "loss": 4.368,
+      "loss/crossentropy": 2.0882590413093567,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21159439533948898,
+      "step": 14206
+    },
+    {
+      "epoch": 0.28416,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.005296834309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.3005,
+      "loss/crossentropy": 2.219560742378235,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2282530516386032,
+      "step": 14208
+    },
+    {
+      "epoch": 0.2842,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.004759724934895833,
+      "learning_rate": 0.0001,
+      "loss": 3.9308,
+      "loss/crossentropy": 1.8383984565734863,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1868659406900406,
+      "step": 14210
+    },
+    {
+      "epoch": 0.28424,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.006892649332682291,
+      "learning_rate": 0.0001,
+      "loss": 3.9613,
+      "loss/crossentropy": 1.7509828209877014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1862550526857376,
+      "step": 14212
+    },
+    {
+      "epoch": 0.28428,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.007513173421223958,
+      "learning_rate": 0.0001,
+      "loss": 4.1301,
+      "loss/crossentropy": 2.1544927954673767,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19378253817558289,
+      "step": 14214
+    },
+    {
+      "epoch": 0.28432,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.007533518473307291,
+      "learning_rate": 0.0001,
+      "loss": 4.1261,
+      "loss/crossentropy": 2.498751997947693,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22881630063056946,
+      "step": 14216
+    },
+    {
+      "epoch": 0.28436,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.006780751546223958,
+      "learning_rate": 0.0001,
+      "loss": 4.3414,
+      "loss/crossentropy": 2.2638392448425293,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21407127380371094,
+      "step": 14218
+    },
+    {
+      "epoch": 0.2844,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.006380208333333333,
+      "learning_rate": 0.0001,
+      "loss": 4.0469,
+      "loss/crossentropy": 2.097207546234131,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21202068030834198,
+      "step": 14220
+    },
+    {
+      "epoch": 0.28444,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.007661946614583333,
+      "learning_rate": 0.0001,
+      "loss": 4.211,
+      "loss/crossentropy": 2.0577695965766907,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21462847292423248,
+      "step": 14222
+    },
+    {
+      "epoch": 0.28448,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.008373769124348958,
+      "learning_rate": 0.0001,
+      "loss": 4.2646,
+      "loss/crossentropy": 2.3061472177505493,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23122046887874603,
+      "step": 14224
+    },
+    {
+      "epoch": 0.28452,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0068511962890625,
+      "learning_rate": 0.0001,
+      "loss": 3.947,
+      "loss/crossentropy": 1.850829005241394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19229399412870407,
+      "step": 14226
+    },
+    {
+      "epoch": 0.28456,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0051348368326822914,
+      "learning_rate": 0.0001,
+      "loss": 4.0891,
+      "loss/crossentropy": 1.9520751237869263,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19159140437841415,
+      "step": 14228
+    },
+    {
+      "epoch": 0.2846,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.005145009358723958,
+      "learning_rate": 0.0001,
+      "loss": 4.2161,
+      "loss/crossentropy": 2.507380247116089,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23365024477243423,
+      "step": 14230
+    },
+    {
+      "epoch": 0.28464,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.006261952718098958,
+      "learning_rate": 0.0001,
+      "loss": 4.0787,
+      "loss/crossentropy": 1.9148982763290405,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19785508513450623,
+      "step": 14232
+    },
+    {
+      "epoch": 0.28468,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.007769521077473958,
+      "learning_rate": 0.0001,
+      "loss": 4.1882,
+      "loss/crossentropy": 2.004162549972534,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.192252054810524,
+      "step": 14234
+    },
+    {
+      "epoch": 0.28472,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0078277587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1318,
+      "loss/crossentropy": 2.1455901861190796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19151784479618073,
+      "step": 14236
+    },
+    {
+      "epoch": 0.28476,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.005580393473307291,
+      "learning_rate": 0.0001,
+      "loss": 3.9548,
+      "loss/crossentropy": 2.2556833028793335,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20640526711940765,
+      "step": 14238
+    },
+    {
+      "epoch": 0.2848,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.013698069254557292,
+      "learning_rate": 0.0001,
+      "loss": 4.3805,
+      "loss/crossentropy": 2.1121991872787476,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21154547482728958,
+      "step": 14240
+    },
+    {
+      "epoch": 0.28484,
+      "grad_norm": 1.7890625,
+      "grad_norm_var": 0.01586888631184896,
+      "learning_rate": 0.0001,
+      "loss": 3.7502,
+      "loss/crossentropy": 1.9072884321212769,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19220874458551407,
+      "step": 14242
+    },
+    {
+      "epoch": 0.28488,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0154052734375,
+      "learning_rate": 0.0001,
+      "loss": 4.1925,
+      "loss/crossentropy": 1.907566249370575,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20355048030614853,
+      "step": 14244
+    },
+    {
+      "epoch": 0.28492,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.0149169921875,
+      "learning_rate": 0.0001,
+      "loss": 3.8873,
+      "loss/crossentropy": 2.1140421628952026,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20535241812467575,
+      "step": 14246
+    },
+    {
+      "epoch": 0.28496,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.016022745768229166,
+      "learning_rate": 0.0001,
+      "loss": 4.1704,
+      "loss/crossentropy": 2.192967176437378,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20353975892066956,
+      "step": 14248
+    },
+    {
+      "epoch": 0.285,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.015213775634765624,
+      "learning_rate": 0.0001,
+      "loss": 3.9064,
+      "loss/crossentropy": 1.6658200025558472,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17274750024080276,
+      "step": 14250
+    },
+    {
+      "epoch": 0.28504,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.017411041259765624,
+      "learning_rate": 0.0001,
+      "loss": 4.2015,
+      "loss/crossentropy": 2.014496326446533,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.203952856361866,
+      "step": 14252
+    },
+    {
+      "epoch": 0.28508,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0249908447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4607,
+      "loss/crossentropy": 2.1534151434898376,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23526588827371597,
+      "step": 14254
+    },
+    {
+      "epoch": 0.28512,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.01701838175455729,
+      "learning_rate": 0.0001,
+      "loss": 4.1263,
+      "loss/crossentropy": 2.08686763048172,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20443058758974075,
+      "step": 14256
+    },
+    {
+      "epoch": 0.28516,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0135406494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.314,
+      "loss/crossentropy": 1.9428821802139282,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20955658704042435,
+      "step": 14258
+    },
+    {
+      "epoch": 0.2852,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.015710194905598957,
+      "learning_rate": 0.0001,
+      "loss": 3.9157,
+      "loss/crossentropy": 1.8959371447563171,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20648592710494995,
+      "step": 14260
+    },
+    {
+      "epoch": 0.28524,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.017081705729166667,
+      "learning_rate": 0.0001,
+      "loss": 3.8246,
+      "loss/crossentropy": 1.7842467427253723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17766441404819489,
+      "step": 14262
+    },
+    {
+      "epoch": 0.28528,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.018805948893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0581,
+      "loss/crossentropy": 1.989040195941925,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20939016342163086,
+      "step": 14264
+    },
+    {
+      "epoch": 0.28532,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.018656158447265626,
+      "learning_rate": 0.0001,
+      "loss": 3.9216,
+      "loss/crossentropy": 2.0441301465034485,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20979472994804382,
+      "step": 14266
+    },
+    {
+      "epoch": 0.28536,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.017970530192057292,
+      "learning_rate": 0.0001,
+      "loss": 4.2181,
+      "loss/crossentropy": 1.9348166584968567,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1930733323097229,
+      "step": 14268
+    },
+    {
+      "epoch": 0.2854,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.009650675455729167,
+      "learning_rate": 0.0001,
+      "loss": 4.2039,
+      "loss/crossentropy": 2.056081175804138,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20383870601654053,
+      "step": 14270
+    },
+    {
+      "epoch": 0.28544,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.011271158854166666,
+      "learning_rate": 0.0001,
+      "loss": 4.4046,
+      "loss/crossentropy": 2.2752946615219116,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2458929866552353,
+      "step": 14272
+    },
+    {
+      "epoch": 0.28548,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010456339518229166,
+      "learning_rate": 0.0001,
+      "loss": 4.1693,
+      "loss/crossentropy": 2.093926787376404,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19211938977241516,
+      "step": 14274
+    },
+    {
+      "epoch": 0.28552,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.009913889567057292,
+      "learning_rate": 0.0001,
+      "loss": 4.1841,
+      "loss/crossentropy": 2.448614239692688,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23177827894687653,
+      "step": 14276
+    },
+    {
+      "epoch": 0.28556,
+      "grad_norm": 1.765625,
+      "grad_norm_var": 0.012401326497395834,
+      "learning_rate": 0.0001,
+      "loss": 4.0002,
+      "loss/crossentropy": 2.0047106742858887,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2162417769432068,
+      "step": 14278
+    },
+    {
+      "epoch": 0.2856,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.017986806233723958,
+      "learning_rate": 0.0001,
+      "loss": 4.1669,
+      "loss/crossentropy": 1.982455313205719,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22751964628696442,
+      "step": 14280
+    },
+    {
+      "epoch": 0.28564,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.017411295572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.0355,
+      "loss/crossentropy": 1.872545838356018,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19046122580766678,
+      "step": 14282
+    },
+    {
+      "epoch": 0.28568,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.016388956705729166,
+      "learning_rate": 0.0001,
+      "loss": 4.1266,
+      "loss/crossentropy": 1.8999648690223694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2155570462346077,
+      "step": 14284
+    },
+    {
+      "epoch": 0.28572,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.014129384358723959,
+      "learning_rate": 0.0001,
+      "loss": 3.9649,
+      "loss/crossentropy": 1.8433392643928528,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18906012177467346,
+      "step": 14286
+    },
+    {
+      "epoch": 0.28576,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.014697011311848958,
+      "learning_rate": 0.0001,
+      "loss": 4.085,
+      "loss/crossentropy": 1.817060947418213,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20473239570856094,
+      "step": 14288
+    },
+    {
+      "epoch": 0.2858,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.01796239217122396,
+      "learning_rate": 0.0001,
+      "loss": 4.1856,
+      "loss/crossentropy": 2.0344348549842834,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19991667568683624,
+      "step": 14290
+    },
+    {
+      "epoch": 0.28584,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.01784032185872396,
+      "learning_rate": 0.0001,
+      "loss": 4.2334,
+      "loss/crossentropy": 2.3413926362991333,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2365436628460884,
+      "step": 14292
+    },
+    {
+      "epoch": 0.28588,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.015464019775390626,
+      "learning_rate": 0.0001,
+      "loss": 3.9437,
+      "loss/crossentropy": 2.0959852933883667,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19703902304172516,
+      "step": 14294
+    },
+    {
+      "epoch": 0.28592,
+      "grad_norm": 7.625,
+      "grad_norm_var": 1.9304239908854166,
+      "learning_rate": 0.0001,
+      "loss": 4.2144,
+      "loss/crossentropy": 2.220101237297058,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21996742486953735,
+      "step": 14296
+    },
+    {
+      "epoch": 0.28596,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 1.915691884358724,
+      "learning_rate": 0.0001,
+      "loss": 4.1806,
+      "loss/crossentropy": 2.073485493659973,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2020936980843544,
+      "step": 14298
+    },
+    {
+      "epoch": 0.286,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 1.9216265360514322,
+      "learning_rate": 0.0001,
+      "loss": 3.9176,
+      "loss/crossentropy": 2.228920817375183,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22304438799619675,
+      "step": 14300
+    },
+    {
+      "epoch": 0.28604,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 1.933642323811849,
+      "learning_rate": 0.0001,
+      "loss": 4.0831,
+      "loss/crossentropy": 1.724998950958252,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1592913344502449,
+      "step": 14302
+    },
+    {
+      "epoch": 0.28608,
+      "grad_norm": 2.0,
+      "grad_norm_var": 1.9375221252441406,
+      "learning_rate": 0.0001,
+      "loss": 4.1738,
+      "loss/crossentropy": 1.8970724940299988,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18843455612659454,
+      "step": 14304
+    },
+    {
+      "epoch": 0.28612,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 1.9384295145670574,
+      "learning_rate": 0.0001,
+      "loss": 4.0808,
+      "loss/crossentropy": 1.9957273602485657,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18272704631090164,
+      "step": 14306
+    },
+    {
+      "epoch": 0.28616,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 1.9298177083333334,
+      "learning_rate": 0.0001,
+      "loss": 4.3861,
+      "loss/crossentropy": 2.1424412727355957,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21485400944948196,
+      "step": 14308
+    },
+    {
+      "epoch": 0.2862,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 1.9411211649576823,
+      "learning_rate": 0.0001,
+      "loss": 3.9622,
+      "loss/crossentropy": 1.7729334235191345,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20436155050992966,
+      "step": 14310
+    },
+    {
+      "epoch": 0.28624,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.013952382405598958,
+      "learning_rate": 0.0001,
+      "loss": 4.2495,
+      "loss/crossentropy": 2.5863327980041504,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24598564952611923,
+      "step": 14312
+    },
+    {
+      "epoch": 0.28628,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.01014404296875,
+      "learning_rate": 0.0001,
+      "loss": 3.962,
+      "loss/crossentropy": 1.9089832305908203,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1905580461025238,
+      "step": 14314
+    },
+    {
+      "epoch": 0.28632,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.010228474934895834,
+      "learning_rate": 0.0001,
+      "loss": 4.2167,
+      "loss/crossentropy": 2.0153337121009827,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21556401252746582,
+      "step": 14316
+    },
+    {
+      "epoch": 0.28636,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008776601155598958,
+      "learning_rate": 0.0001,
+      "loss": 4.4037,
+      "loss/crossentropy": 1.9692147970199585,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17990782111883163,
+      "step": 14318
+    },
+    {
+      "epoch": 0.2864,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.010969034830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.1049,
+      "loss/crossentropy": 2.0404393076896667,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2108697146177292,
+      "step": 14320
+    },
+    {
+      "epoch": 0.28644,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.009488932291666667,
+      "learning_rate": 0.0001,
+      "loss": 4.2809,
+      "loss/crossentropy": 2.133711099624634,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1942032128572464,
+      "step": 14322
+    },
+    {
+      "epoch": 0.28648,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.004878489176432291,
+      "learning_rate": 0.0001,
+      "loss": 4.1605,
+      "loss/crossentropy": 1.9240726232528687,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18174728006124496,
+      "step": 14324
+    },
+    {
+      "epoch": 0.28652,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.00850830078125,
+      "learning_rate": 0.0001,
+      "loss": 4.2671,
+      "loss/crossentropy": 2.0952632427215576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20621410757303238,
+      "step": 14326
+    },
+    {
+      "epoch": 0.28656,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.008326975504557292,
+      "learning_rate": 0.0001,
+      "loss": 4.0683,
+      "loss/crossentropy": 2.2604600191116333,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2310045212507248,
+      "step": 14328
+    },
+    {
+      "epoch": 0.2866,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.00777587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2551,
+      "loss/crossentropy": 2.0915167331695557,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23344147205352783,
+      "step": 14330
+    },
+    {
+      "epoch": 0.28664,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.00804443359375,
+      "learning_rate": 0.0001,
+      "loss": 3.9575,
+      "loss/crossentropy": 2.2928651571273804,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23840509355068207,
+      "step": 14332
+    },
+    {
+      "epoch": 0.28668,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.008685048421223958,
+      "learning_rate": 0.0001,
+      "loss": 4.4469,
+      "loss/crossentropy": 2.261072874069214,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22056513279676437,
+      "step": 14334
+    },
+    {
+      "epoch": 0.28672,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.007883453369140625,
+      "learning_rate": 0.0001,
+      "loss": 3.9218,
+      "loss/crossentropy": 1.7591394186019897,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1900082305073738,
+      "step": 14336
+    },
+    {
+      "epoch": 0.28676,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.008426666259765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0457,
+      "loss/crossentropy": 2.129893183708191,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19578705728054047,
+      "step": 14338
+    },
+    {
+      "epoch": 0.2868,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.008451334635416667,
+      "learning_rate": 0.0001,
+      "loss": 4.4884,
+      "loss/crossentropy": 2.0376795530319214,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2088106870651245,
+      "step": 14340
+    },
+    {
+      "epoch": 0.28684,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.005018870035807292,
+      "learning_rate": 0.0001,
+      "loss": 4.282,
+      "loss/crossentropy": 2.1484848260879517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2113661915063858,
+      "step": 14342
+    },
+    {
+      "epoch": 0.28688,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.00496826171875,
+      "learning_rate": 0.0001,
+      "loss": 4.2713,
+      "loss/crossentropy": 2.204781651496887,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20908530056476593,
+      "step": 14344
+    },
+    {
+      "epoch": 0.28692,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0052154541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1075,
+      "loss/crossentropy": 1.8096813559532166,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18912722170352936,
+      "step": 14346
+    },
+    {
+      "epoch": 0.28696,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.005580393473307291,
+      "learning_rate": 0.0001,
+      "loss": 4.0057,
+      "loss/crossentropy": 1.8014967441558838,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18832575529813766,
+      "step": 14348
+    },
+    {
+      "epoch": 0.287,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.00789794921875,
+      "learning_rate": 0.0001,
+      "loss": 4.3211,
+      "loss/crossentropy": 2.1443156003952026,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.232828289270401,
+      "step": 14350
+    },
+    {
+      "epoch": 0.28704,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.008259073893229166,
+      "learning_rate": 0.0001,
+      "loss": 4.0984,
+      "loss/crossentropy": 2.1773927211761475,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19964880496263504,
+      "step": 14352
+    },
+    {
+      "epoch": 0.28708,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.007795206705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.1432,
+      "loss/crossentropy": 2.157313585281372,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2149301916360855,
+      "step": 14354
+    },
+    {
+      "epoch": 0.28712,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.007184855143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2938,
+      "loss/crossentropy": 2.3431146144866943,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2297225296497345,
+      "step": 14356
+    },
+    {
+      "epoch": 0.28716,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.007940419514973958,
+      "learning_rate": 0.0001,
+      "loss": 4.2026,
+      "loss/crossentropy": 1.894954264163971,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19812744110822678,
+      "step": 14358
+    },
+    {
+      "epoch": 0.2872,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.008784993489583334,
+      "learning_rate": 0.0001,
+      "loss": 4.0654,
+      "loss/crossentropy": 2.1701435446739197,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21183644235134125,
+      "step": 14360
+    },
+    {
+      "epoch": 0.28724,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007975006103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2457,
+      "loss/crossentropy": 2.202796459197998,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21229087561368942,
+      "step": 14362
+    },
+    {
+      "epoch": 0.28728,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.009789784749348959,
+      "learning_rate": 0.0001,
+      "loss": 4.3692,
+      "loss/crossentropy": 2.225682258605957,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22512775659561157,
+      "step": 14364
+    },
+    {
+      "epoch": 0.28732,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.007088216145833334,
+      "learning_rate": 0.0001,
+      "loss": 4.0938,
+      "loss/crossentropy": 1.9562655687332153,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19557978957891464,
+      "step": 14366
+    },
+    {
+      "epoch": 0.28736,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0062945048014322914,
+      "learning_rate": 0.0001,
+      "loss": 4.2922,
+      "loss/crossentropy": 1.794309377670288,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1851721778512001,
+      "step": 14368
+    },
+    {
+      "epoch": 0.2874,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.006660970052083334,
+      "learning_rate": 0.0001,
+      "loss": 4.1198,
+      "loss/crossentropy": 2.276759445667267,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2071758210659027,
+      "step": 14370
+    },
+    {
+      "epoch": 0.28744,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.006696573893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1162,
+      "loss/crossentropy": 2.332028031349182,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22525647282600403,
+      "step": 14372
+    },
+    {
+      "epoch": 0.28748,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.005997721354166667,
+      "learning_rate": 0.0001,
+      "loss": 4.2594,
+      "loss/crossentropy": 1.975858986377716,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18405582010746002,
+      "step": 14374
+    },
+    {
+      "epoch": 0.28752,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.006453196207682292,
+      "learning_rate": 0.0001,
+      "loss": 4.0615,
+      "loss/crossentropy": 2.002572774887085,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19543466717004776,
+      "step": 14376
+    },
+    {
+      "epoch": 0.28756,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.006703440348307292,
+      "learning_rate": 0.0001,
+      "loss": 4.1837,
+      "loss/crossentropy": 2.400219678878784,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23009777069091797,
+      "step": 14378
+    },
+    {
+      "epoch": 0.2876,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.003352864583333333,
+      "learning_rate": 0.0001,
+      "loss": 4.0071,
+      "loss/crossentropy": 1.6437376737594604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17785440385341644,
+      "step": 14380
+    },
+    {
+      "epoch": 0.28764,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.007225545247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1494,
+      "loss/crossentropy": 1.8833640813827515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22255511581897736,
+      "step": 14382
+    },
+    {
+      "epoch": 0.28768,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.007100423177083333,
+      "learning_rate": 0.0001,
+      "loss": 4.0881,
+      "loss/crossentropy": 1.9113904237747192,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2050003483891487,
+      "step": 14384
+    },
+    {
+      "epoch": 0.28772,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.007765452067057292,
+      "learning_rate": 0.0001,
+      "loss": 4.1114,
+      "loss/crossentropy": 2.236249327659607,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21349424868822098,
+      "step": 14386
+    },
+    {
+      "epoch": 0.28776,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.008090972900390625,
+      "learning_rate": 0.0001,
+      "loss": 3.9282,
+      "loss/crossentropy": 1.8709489703178406,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19309207051992416,
+      "step": 14388
+    },
+    {
+      "epoch": 0.2878,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.007933553059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0871,
+      "loss/crossentropy": 1.926324725151062,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19902244210243225,
+      "step": 14390
+    },
+    {
+      "epoch": 0.28784,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.00810546875,
+      "learning_rate": 0.0001,
+      "loss": 4.085,
+      "loss/crossentropy": 1.898009479045868,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18698234856128693,
+      "step": 14392
+    },
+    {
+      "epoch": 0.28788,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0122711181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2367,
+      "loss/crossentropy": 1.9728660583496094,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2033083364367485,
+      "step": 14394
+    },
+    {
+      "epoch": 0.28792,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.01282958984375,
+      "learning_rate": 0.0001,
+      "loss": 3.7965,
+      "loss/crossentropy": 1.957255482673645,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20039385557174683,
+      "step": 14396
+    },
+    {
+      "epoch": 0.28796,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009897613525390625,
+      "learning_rate": 0.0001,
+      "loss": 3.8525,
+      "loss/crossentropy": 1.8001562356948853,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2059212401509285,
+      "step": 14398
+    },
+    {
+      "epoch": 0.288,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.011177571614583333,
+      "learning_rate": 0.0001,
+      "loss": 4.1749,
+      "loss/crossentropy": 1.9166946411132812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19456689059734344,
+      "step": 14400
+    },
+    {
+      "epoch": 0.28804,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.01221923828125,
+      "learning_rate": 0.0001,
+      "loss": 4.3302,
+      "loss/crossentropy": 2.033670485019684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24448300898075104,
+      "step": 14402
+    },
+    {
+      "epoch": 0.28808,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.012580362955729167,
+      "learning_rate": 0.0001,
+      "loss": 3.8679,
+      "loss/crossentropy": 2.2668023705482483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20903322845697403,
+      "step": 14404
+    },
+    {
+      "epoch": 0.28812,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.012532297770182292,
+      "learning_rate": 0.0001,
+      "loss": 4.0638,
+      "loss/crossentropy": 2.1076024770736694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20488093793392181,
+      "step": 14406
+    },
+    {
+      "epoch": 0.28816,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.012117258707682292,
+      "learning_rate": 0.0001,
+      "loss": 4.0707,
+      "loss/crossentropy": 1.996046781539917,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20725766569375992,
+      "step": 14408
+    },
+    {
+      "epoch": 0.2882,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.008365885416666666,
+      "learning_rate": 0.0001,
+      "loss": 4.2302,
+      "loss/crossentropy": 2.2031015157699585,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19491394609212875,
+      "step": 14410
+    },
+    {
+      "epoch": 0.28824,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.008714803059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.3962,
+      "loss/crossentropy": 2.0339081287384033,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21979261189699173,
+      "step": 14412
+    },
+    {
+      "epoch": 0.28828,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.011641438802083333,
+      "learning_rate": 0.0001,
+      "loss": 4.0042,
+      "loss/crossentropy": 1.9283286929130554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19970305263996124,
+      "step": 14414
+    },
+    {
+      "epoch": 0.28832,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.011156972249348958,
+      "learning_rate": 0.0001,
+      "loss": 4.3879,
+      "loss/crossentropy": 2.1092429161071777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2114599123597145,
+      "step": 14416
+    },
+    {
+      "epoch": 0.28836,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.01004638671875,
+      "learning_rate": 0.0001,
+      "loss": 4.1234,
+      "loss/crossentropy": 1.8102002143859863,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18407940864562988,
+      "step": 14418
+    },
+    {
+      "epoch": 0.2884,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.008906809488932292,
+      "learning_rate": 0.0001,
+      "loss": 4.0943,
+      "loss/crossentropy": 2.3192719221115112,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22208110988140106,
+      "step": 14420
+    },
+    {
+      "epoch": 0.28844,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0082183837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2268,
+      "loss/crossentropy": 2.2302430868148804,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21011962741613388,
+      "step": 14422
+    },
+    {
+      "epoch": 0.28848,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.008542633056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1235,
+      "loss/crossentropy": 2.3015987873077393,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2012081891298294,
+      "step": 14424
+    },
+    {
+      "epoch": 0.28852,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0077056884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3094,
+      "loss/crossentropy": 2.060473084449768,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20174618810415268,
+      "step": 14426
+    },
+    {
+      "epoch": 0.28856,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.014082845052083333,
+      "learning_rate": 0.0001,
+      "loss": 3.9594,
+      "loss/crossentropy": 1.8146299719810486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18100766092538834,
+      "step": 14428
+    },
+    {
+      "epoch": 0.2886,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.011277008056640624,
+      "learning_rate": 0.0001,
+      "loss": 3.8444,
+      "loss/crossentropy": 2.1504935026168823,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20493116229772568,
+      "step": 14430
+    },
+    {
+      "epoch": 0.28864,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.010796864827473959,
+      "learning_rate": 0.0001,
+      "loss": 4.1153,
+      "loss/crossentropy": 2.4947997331619263,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2339334562420845,
+      "step": 14432
+    },
+    {
+      "epoch": 0.28868,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.01612116495768229,
+      "learning_rate": 0.0001,
+      "loss": 4.3011,
+      "loss/crossentropy": 2.421883702278137,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23815617710351944,
+      "step": 14434
+    },
+    {
+      "epoch": 0.28872,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.016600545247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.2152,
+      "loss/crossentropy": 2.093776822090149,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20425530523061752,
+      "step": 14436
+    },
+    {
+      "epoch": 0.28876,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.01817601521809896,
+      "learning_rate": 0.0001,
+      "loss": 3.9126,
+      "loss/crossentropy": 2.02763295173645,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1972591131925583,
+      "step": 14438
+    },
+    {
+      "epoch": 0.2888,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0176422119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1348,
+      "loss/crossentropy": 1.9769265055656433,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.186862014234066,
+      "step": 14440
+    },
+    {
+      "epoch": 0.28884,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.019024403889973958,
+      "learning_rate": 0.0001,
+      "loss": 4.4997,
+      "loss/crossentropy": 2.34747314453125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22067507356405258,
+      "step": 14442
+    },
+    {
+      "epoch": 0.28888,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.0128570556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1497,
+      "loss/crossentropy": 2.1036205887794495,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2231181040406227,
+      "step": 14444
+    },
+    {
+      "epoch": 0.28892,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.01638768513997396,
+      "learning_rate": 0.0001,
+      "loss": 4.251,
+      "loss/crossentropy": 2.2350860834121704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23475627601146698,
+      "step": 14446
+    },
+    {
+      "epoch": 0.28896,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.01778132120768229,
+      "learning_rate": 0.0001,
+      "loss": 4.2095,
+      "loss/crossentropy": 2.2336645126342773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20643477141857147,
+      "step": 14448
+    },
+    {
+      "epoch": 0.289,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.012923177083333333,
+      "learning_rate": 0.0001,
+      "loss": 4.0793,
+      "loss/crossentropy": 2.345365524291992,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20650553703308105,
+      "step": 14450
+    },
+    {
+      "epoch": 0.28904,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0120025634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.131,
+      "loss/crossentropy": 2.307543635368347,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21354079991579056,
+      "step": 14452
+    },
+    {
+      "epoch": 0.28908,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.012001291910807291,
+      "learning_rate": 0.0001,
+      "loss": 4.1006,
+      "loss/crossentropy": 1.8094390034675598,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1850133240222931,
+      "step": 14454
+    },
+    {
+      "epoch": 0.28912,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.011424763997395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1326,
+      "loss/crossentropy": 1.859872817993164,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.226328507065773,
+      "step": 14456
+    },
+    {
+      "epoch": 0.28916,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.009340159098307292,
+      "learning_rate": 0.0001,
+      "loss": 3.9997,
+      "loss/crossentropy": 1.9507490396499634,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18718113750219345,
+      "step": 14458
+    },
+    {
+      "epoch": 0.2892,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.01126708984375,
+      "learning_rate": 0.0001,
+      "loss": 3.7868,
+      "loss/crossentropy": 1.6642532348632812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1674812287092209,
+      "step": 14460
+    },
+    {
+      "epoch": 0.28924,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.007039133707682292,
+      "learning_rate": 0.0001,
+      "loss": 3.9523,
+      "loss/crossentropy": 1.9071928262710571,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18480891734361649,
+      "step": 14462
+    },
+    {
+      "epoch": 0.28928,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.005718739827473959,
+      "learning_rate": 0.0001,
+      "loss": 4.2876,
+      "loss/crossentropy": 2.0758568048477173,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2002219259738922,
+      "step": 14464
+    },
+    {
+      "epoch": 0.28932,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.006769816080729167,
+      "learning_rate": 0.0001,
+      "loss": 3.9438,
+      "loss/crossentropy": 1.9777529835700989,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19538866728544235,
+      "step": 14466
+    },
+    {
+      "epoch": 0.28936,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0062652587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1262,
+      "loss/crossentropy": 2.008077323436737,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20578377693891525,
+      "step": 14468
+    },
+    {
+      "epoch": 0.2894,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.006673177083333333,
+      "learning_rate": 0.0001,
+      "loss": 3.9341,
+      "loss/crossentropy": 2.011266529560089,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20298370718955994,
+      "step": 14470
+    },
+    {
+      "epoch": 0.28944,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0057769775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.201,
+      "loss/crossentropy": 2.231672167778015,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21847998350858688,
+      "step": 14472
+    },
+    {
+      "epoch": 0.28948,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.007279205322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.382,
+      "loss/crossentropy": 2.2268325090408325,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22168640047311783,
+      "step": 14474
+    },
+    {
+      "epoch": 0.28952,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.012105305989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.2193,
+      "loss/crossentropy": 2.2705942392349243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2110700011253357,
+      "step": 14476
+    },
+    {
+      "epoch": 0.28956,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.012566884358723959,
+      "learning_rate": 0.0001,
+      "loss": 3.8754,
+      "loss/crossentropy": 1.893052339553833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18865809589624405,
+      "step": 14478
+    },
+    {
+      "epoch": 0.2896,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.013732655843098959,
+      "learning_rate": 0.0001,
+      "loss": 4.0989,
+      "loss/crossentropy": 2.020704984664917,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18822921812534332,
+      "step": 14480
+    },
+    {
+      "epoch": 0.28964,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.014632161458333333,
+      "learning_rate": 0.0001,
+      "loss": 4.1651,
+      "loss/crossentropy": 1.762313961982727,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.204545259475708,
+      "step": 14482
+    },
+    {
+      "epoch": 0.28968,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.013158162434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.0261,
+      "loss/crossentropy": 1.8793032765388489,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17913836985826492,
+      "step": 14484
+    },
+    {
+      "epoch": 0.28972,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.013372548421223958,
+      "learning_rate": 0.0001,
+      "loss": 4.2731,
+      "loss/crossentropy": 2.428762197494507,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23766332119703293,
+      "step": 14486
+    },
+    {
+      "epoch": 0.28976,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.014969635009765624,
+      "learning_rate": 0.0001,
+      "loss": 4.1134,
+      "loss/crossentropy": 2.2160138487815857,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21619082987308502,
+      "step": 14488
+    },
+    {
+      "epoch": 0.2898,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0150787353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1049,
+      "loss/crossentropy": 1.8169404864311218,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1981067657470703,
+      "step": 14490
+    },
+    {
+      "epoch": 0.28984,
+      "grad_norm": 1.7890625,
+      "grad_norm_var": 0.0131591796875,
+      "learning_rate": 0.0001,
+      "loss": 4.1456,
+      "loss/crossentropy": 2.147166609764099,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1993183195590973,
+      "step": 14492
+    },
+    {
+      "epoch": 0.28988,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.011362457275390625,
+      "learning_rate": 0.0001,
+      "loss": 3.8507,
+      "loss/crossentropy": 1.7740440964698792,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18381935358047485,
+      "step": 14494
+    },
+    {
+      "epoch": 0.28992,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.010428619384765626,
+      "learning_rate": 0.0001,
+      "loss": 4.0327,
+      "loss/crossentropy": 2.206219792366028,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21052061766386032,
+      "step": 14496
+    },
+    {
+      "epoch": 0.28996,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008600870768229166,
+      "learning_rate": 0.0001,
+      "loss": 4.4421,
+      "loss/crossentropy": 2.4030661582946777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22094716131687164,
+      "step": 14498
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.016633097330729166,
+      "learning_rate": 0.0001,
+      "loss": 4.4593,
+      "loss/crossentropy": 2.142518997192383,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2111787647008896,
+      "step": 14500
+    },
+    {
+      "epoch": 0.29004,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.015633138020833333,
+      "learning_rate": 0.0001,
+      "loss": 4.3679,
+      "loss/crossentropy": 1.9228865504264832,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19389715045690536,
+      "step": 14502
+    },
+    {
+      "epoch": 0.29008,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.014141591389973958,
+      "learning_rate": 0.0001,
+      "loss": 4.011,
+      "loss/crossentropy": 1.789183259010315,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18835698068141937,
+      "step": 14504
+    },
+    {
+      "epoch": 0.29012,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.013798014322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.2862,
+      "loss/crossentropy": 2.3876765966415405,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23156627267599106,
+      "step": 14506
+    },
+    {
+      "epoch": 0.29016,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010151926676432292,
+      "learning_rate": 0.0001,
+      "loss": 4.2024,
+      "loss/crossentropy": 2.055350124835968,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1995788812637329,
+      "step": 14508
+    },
+    {
+      "epoch": 0.2902,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.010007476806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.343,
+      "loss/crossentropy": 2.2045267820358276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2197813093662262,
+      "step": 14510
+    },
+    {
+      "epoch": 0.29024,
+      "grad_norm": 1.765625,
+      "grad_norm_var": 0.014503733317057291,
+      "learning_rate": 0.0001,
+      "loss": 3.7136,
+      "loss/crossentropy": 1.9596665501594543,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17329587787389755,
+      "step": 14512
+    },
+    {
+      "epoch": 0.29028,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.017465972900390626,
+      "learning_rate": 0.0001,
+      "loss": 3.9546,
+      "loss/crossentropy": 1.8273005485534668,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17768846452236176,
+      "step": 14514
+    },
+    {
+      "epoch": 0.29032,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.011739095052083334,
+      "learning_rate": 0.0001,
+      "loss": 3.9924,
+      "loss/crossentropy": 2.0428807735443115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2104019746184349,
+      "step": 14516
+    },
+    {
+      "epoch": 0.29036,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.014216868082682292,
+      "learning_rate": 0.0001,
+      "loss": 4.3591,
+      "loss/crossentropy": 2.1242733001708984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2153768166899681,
+      "step": 14518
+    },
+    {
+      "epoch": 0.2904,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.013944244384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0831,
+      "loss/crossentropy": 1.9223615527153015,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18382133543491364,
+      "step": 14520
+    },
+    {
+      "epoch": 0.29044,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.013993326822916667,
+      "learning_rate": 0.0001,
+      "loss": 3.9527,
+      "loss/crossentropy": 2.12492972612381,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19818739593029022,
+      "step": 14522
+    },
+    {
+      "epoch": 0.29048,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.013665517171223959,
+      "learning_rate": 0.0001,
+      "loss": 4.1539,
+      "loss/crossentropy": 2.0251020789146423,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19633153080940247,
+      "step": 14524
+    },
+    {
+      "epoch": 0.29052,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.014949544270833334,
+      "learning_rate": 0.0001,
+      "loss": 4.192,
+      "loss/crossentropy": 2.177084445953369,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22618898749351501,
+      "step": 14526
+    },
+    {
+      "epoch": 0.29056,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.013997141520182292,
+      "learning_rate": 0.0001,
+      "loss": 3.9215,
+      "loss/crossentropy": 1.718012809753418,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18632279336452484,
+      "step": 14528
+    },
+    {
+      "epoch": 0.2906,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.017850494384765624,
+      "learning_rate": 0.0001,
+      "loss": 4.2869,
+      "loss/crossentropy": 1.7026863098144531,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.211147278547287,
+      "step": 14530
+    },
+    {
+      "epoch": 0.29064,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.017682902018229165,
+      "learning_rate": 0.0001,
+      "loss": 3.9369,
+      "loss/crossentropy": 1.6652680039405823,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16967280209064484,
+      "step": 14532
+    },
+    {
+      "epoch": 0.29068,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.013996378580729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0815,
+      "loss/crossentropy": 2.191527247428894,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20215752720832825,
+      "step": 14534
+    },
+    {
+      "epoch": 0.29072,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.015927886962890624,
+      "learning_rate": 0.0001,
+      "loss": 3.8817,
+      "loss/crossentropy": 1.8385645747184753,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19906776398420334,
+      "step": 14536
+    },
+    {
+      "epoch": 0.29076,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.01665013631184896,
+      "learning_rate": 0.0001,
+      "loss": 3.9186,
+      "loss/crossentropy": 2.0456249117851257,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2057228460907936,
+      "step": 14538
+    },
+    {
+      "epoch": 0.2908,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.018651326497395832,
+      "learning_rate": 0.0001,
+      "loss": 4.0636,
+      "loss/crossentropy": 2.0413625836372375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20096366852521896,
+      "step": 14540
+    },
+    {
+      "epoch": 0.29084,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.016755167643229166,
+      "learning_rate": 0.0001,
+      "loss": 3.7819,
+      "loss/crossentropy": 1.723297357559204,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.180135078728199,
+      "step": 14542
+    },
+    {
+      "epoch": 0.29088,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.01578343709309896,
+      "learning_rate": 0.0001,
+      "loss": 4.2115,
+      "loss/crossentropy": 2.3143200874328613,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2289530113339424,
+      "step": 14544
+    },
+    {
+      "epoch": 0.29092,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.010017903645833333,
+      "learning_rate": 0.0001,
+      "loss": 4.408,
+      "loss/crossentropy": 2.1612138748168945,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21224550902843475,
+      "step": 14546
+    },
+    {
+      "epoch": 0.29096,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.009227498372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1101,
+      "loss/crossentropy": 2.0591673851013184,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18913161754608154,
+      "step": 14548
+    },
+    {
+      "epoch": 0.291,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.009505208333333333,
+      "learning_rate": 0.0001,
+      "loss": 4.132,
+      "loss/crossentropy": 2.0901013016700745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19315654039382935,
+      "step": 14550
+    },
+    {
+      "epoch": 0.29104,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.006833648681640625,
+      "learning_rate": 0.0001,
+      "loss": 3.9351,
+      "loss/crossentropy": 2.0716471672058105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1951451674103737,
+      "step": 14552
+    },
+    {
+      "epoch": 0.29108,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.008186848958333333,
+      "learning_rate": 0.0001,
+      "loss": 4.3656,
+      "loss/crossentropy": 2.084408760070801,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21354945749044418,
+      "step": 14554
+    },
+    {
+      "epoch": 0.29112,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.006486002604166667,
+      "learning_rate": 0.0001,
+      "loss": 4.0342,
+      "loss/crossentropy": 2.154771566390991,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20277246832847595,
+      "step": 14556
+    },
+    {
+      "epoch": 0.29116,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0069353739420572914,
+      "learning_rate": 0.0001,
+      "loss": 4.259,
+      "loss/crossentropy": 2.24834144115448,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23204077035188675,
+      "step": 14558
+    },
+    {
+      "epoch": 0.2912,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.006951649983723958,
+      "learning_rate": 0.0001,
+      "loss": 3.8054,
+      "loss/crossentropy": 1.8239850401878357,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18208947032690048,
+      "step": 14560
+    },
+    {
+      "epoch": 0.29124,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.006701405843098958,
+      "learning_rate": 0.0001,
+      "loss": 3.9465,
+      "loss/crossentropy": 1.969277262687683,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2093023955821991,
+      "step": 14562
+    },
+    {
+      "epoch": 0.29128,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.007670084635416667,
+      "learning_rate": 0.0001,
+      "loss": 3.7176,
+      "loss/crossentropy": 1.561119556427002,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1653747707605362,
+      "step": 14564
+    },
+    {
+      "epoch": 0.29132,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.008154042561848958,
+      "learning_rate": 0.0001,
+      "loss": 4.3307,
+      "loss/crossentropy": 2.2341216802597046,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23967482149600983,
+      "step": 14566
+    },
+    {
+      "epoch": 0.29136,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.008227284749348958,
+      "learning_rate": 0.0001,
+      "loss": 3.9899,
+      "loss/crossentropy": 1.999854326248169,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19598717242479324,
+      "step": 14568
+    },
+    {
+      "epoch": 0.2914,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.006898752848307292,
+      "learning_rate": 0.0001,
+      "loss": 4.1262,
+      "loss/crossentropy": 1.8079020380973816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1919228658080101,
+      "step": 14570
+    },
+    {
+      "epoch": 0.29144,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009970855712890626,
+      "learning_rate": 0.0001,
+      "loss": 3.9946,
+      "loss/crossentropy": 2.044901430606842,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21737617254257202,
+      "step": 14572
+    },
+    {
+      "epoch": 0.29148,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009806060791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0498,
+      "loss/crossentropy": 1.972197949886322,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20828764885663986,
+      "step": 14574
+    },
+    {
+      "epoch": 0.29152,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010131581624348959,
+      "learning_rate": 0.0001,
+      "loss": 4.0011,
+      "loss/crossentropy": 1.9129782915115356,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17108886688947678,
+      "step": 14576
+    },
+    {
+      "epoch": 0.29156,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.01068115234375,
+      "learning_rate": 0.0001,
+      "loss": 4.1458,
+      "loss/crossentropy": 2.0545085072517395,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20631857216358185,
+      "step": 14578
+    },
+    {
+      "epoch": 0.2916,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.012654368082682292,
+      "learning_rate": 0.0001,
+      "loss": 4.355,
+      "loss/crossentropy": 2.0274637937545776,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2099764049053192,
+      "step": 14580
+    },
+    {
+      "epoch": 0.29164,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.013765207926432292,
+      "learning_rate": 0.0001,
+      "loss": 3.9014,
+      "loss/crossentropy": 1.8154139518737793,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19045238196849823,
+      "step": 14582
+    },
+    {
+      "epoch": 0.29168,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.014937082926432291,
+      "learning_rate": 0.0001,
+      "loss": 4.0327,
+      "loss/crossentropy": 2.061666965484619,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21272550523281097,
+      "step": 14584
+    },
+    {
+      "epoch": 0.29172,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.01580785115559896,
+      "learning_rate": 0.0001,
+      "loss": 4.6019,
+      "loss/crossentropy": 2.3831146955490112,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22327116876840591,
+      "step": 14586
+    },
+    {
+      "epoch": 0.29176,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.013090006510416667,
+      "learning_rate": 0.0001,
+      "loss": 4.2142,
+      "loss/crossentropy": 1.8024229407310486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1871618777513504,
+      "step": 14588
+    },
+    {
+      "epoch": 0.2918,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.014967600504557291,
+      "learning_rate": 0.0001,
+      "loss": 4.1611,
+      "loss/crossentropy": 2.289118528366089,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22116435319185257,
+      "step": 14590
+    },
+    {
+      "epoch": 0.29184,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.013277180989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.1358,
+      "loss/crossentropy": 2.11151522397995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22029711306095123,
+      "step": 14592
+    },
+    {
+      "epoch": 0.29188,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.013288370768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.3103,
+      "loss/crossentropy": 1.7856897711753845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1815958321094513,
+      "step": 14594
+    },
+    {
+      "epoch": 0.29192,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0101959228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1085,
+      "loss/crossentropy": 1.9745690822601318,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2074003368616104,
+      "step": 14596
+    },
+    {
+      "epoch": 0.29196,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.0091461181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3034,
+      "loss/crossentropy": 1.9199401140213013,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19998926669359207,
+      "step": 14598
+    },
+    {
+      "epoch": 0.292,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.007743072509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0842,
+      "loss/crossentropy": 1.9938844442367554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20138052105903625,
+      "step": 14600
+    },
+    {
+      "epoch": 0.29204,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0059397379557291664,
+      "learning_rate": 0.0001,
+      "loss": 4.0025,
+      "loss/crossentropy": 1.6290993094444275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18716775625944138,
+      "step": 14602
+    },
+    {
+      "epoch": 0.29208,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.006115468343098959,
+      "learning_rate": 0.0001,
+      "loss": 3.9953,
+      "loss/crossentropy": 2.116807520389557,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2113799750804901,
+      "step": 14604
+    },
+    {
+      "epoch": 0.29212,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.006392415364583333,
+      "learning_rate": 0.0001,
+      "loss": 4.0527,
+      "loss/crossentropy": 1.6732030510902405,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19776474684476852,
+      "step": 14606
+    },
+    {
+      "epoch": 0.29216,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.006762440999348958,
+      "learning_rate": 0.0001,
+      "loss": 4.2572,
+      "loss/crossentropy": 1.9403682351112366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18560070544481277,
+      "step": 14608
+    },
+    {
+      "epoch": 0.2922,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.011641438802083333,
+      "learning_rate": 0.0001,
+      "loss": 4.2238,
+      "loss/crossentropy": 1.9337440729141235,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2172246277332306,
+      "step": 14610
+    },
+    {
+      "epoch": 0.29224,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.011836751302083334,
+      "learning_rate": 0.0001,
+      "loss": 4.1839,
+      "loss/crossentropy": 2.1603177785873413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2076597660779953,
+      "step": 14612
+    },
+    {
+      "epoch": 0.29228,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.0118804931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1103,
+      "loss/crossentropy": 1.979960322380066,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2007075995206833,
+      "step": 14614
+    },
+    {
+      "epoch": 0.29232,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0134429931640625,
+      "learning_rate": 0.0001,
+      "loss": 3.9901,
+      "loss/crossentropy": 2.1262378096580505,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2033269852399826,
+      "step": 14616
+    },
+    {
+      "epoch": 0.29236,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.013327789306640626,
+      "learning_rate": 0.0001,
+      "loss": 3.9001,
+      "loss/crossentropy": 1.836608648300171,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17032578587532043,
+      "step": 14618
+    },
+    {
+      "epoch": 0.2924,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.012473297119140626,
+      "learning_rate": 0.0001,
+      "loss": 4.0165,
+      "loss/crossentropy": 1.784467339515686,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.187791109085083,
+      "step": 14620
+    },
+    {
+      "epoch": 0.29244,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.010196940104166666,
+      "learning_rate": 0.0001,
+      "loss": 4.2963,
+      "loss/crossentropy": 2.5023289918899536,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2469349130988121,
+      "step": 14622
+    },
+    {
+      "epoch": 0.29248,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0142974853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3748,
+      "loss/crossentropy": 2.244086742401123,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20690800249576569,
+      "step": 14624
+    },
+    {
+      "epoch": 0.29252,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.008829752604166666,
+      "learning_rate": 0.0001,
+      "loss": 4.2594,
+      "loss/crossentropy": 2.167649209499359,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23074676096439362,
+      "step": 14626
+    },
+    {
+      "epoch": 0.29256,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0092437744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1837,
+      "loss/crossentropy": 1.934467613697052,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2041843980550766,
+      "step": 14628
+    },
+    {
+      "epoch": 0.2926,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0152740478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3277,
+      "loss/crossentropy": 2.457708954811096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24325969815254211,
+      "step": 14630
+    },
+    {
+      "epoch": 0.29264,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.014446767171223958,
+      "learning_rate": 0.0001,
+      "loss": 4.0523,
+      "loss/crossentropy": 2.165565609931946,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20386752486228943,
+      "step": 14632
+    },
+    {
+      "epoch": 0.29268,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.013677724202473958,
+      "learning_rate": 0.0001,
+      "loss": 4.0767,
+      "loss/crossentropy": 1.9475921988487244,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2017974555492401,
+      "step": 14634
+    },
+    {
+      "epoch": 0.29272,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.015819295247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1435,
+      "loss/crossentropy": 1.95538991689682,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19920305162668228,
+      "step": 14636
+    },
+    {
+      "epoch": 0.29276,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.01803766886393229,
+      "learning_rate": 0.0001,
+      "loss": 4.4073,
+      "loss/crossentropy": 2.1813108921051025,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2130545824766159,
+      "step": 14638
+    },
+    {
+      "epoch": 0.2928,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.01672337849934896,
+      "learning_rate": 0.0001,
+      "loss": 4.1841,
+      "loss/crossentropy": 2.198926568031311,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21660766005516052,
+      "step": 14640
+    },
+    {
+      "epoch": 0.29284,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.01692682902018229,
+      "learning_rate": 0.0001,
+      "loss": 4.2872,
+      "loss/crossentropy": 2.244659662246704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22478342056274414,
+      "step": 14642
+    },
+    {
+      "epoch": 0.29288,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.01608861287434896,
+      "learning_rate": 0.0001,
+      "loss": 4.234,
+      "loss/crossentropy": 2.1159621477127075,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22023546695709229,
+      "step": 14644
+    },
+    {
+      "epoch": 0.29292,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.011067708333333334,
+      "learning_rate": 0.0001,
+      "loss": 3.9166,
+      "loss/crossentropy": 2.2336114645004272,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20683540403842926,
+      "step": 14646
+    },
+    {
+      "epoch": 0.29296,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.011128743489583334,
+      "learning_rate": 0.0001,
+      "loss": 3.8647,
+      "loss/crossentropy": 1.9015604257583618,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2791958376765251,
+      "step": 14648
+    },
+    {
+      "epoch": 0.293,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.011180623372395834,
+      "learning_rate": 0.0001,
+      "loss": 3.7325,
+      "loss/crossentropy": 1.6337950229644775,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17279411852359772,
+      "step": 14650
+    },
+    {
+      "epoch": 0.29304,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009618123372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.111,
+      "loss/crossentropy": 1.920238435268402,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18912465870380402,
+      "step": 14652
+    },
+    {
+      "epoch": 0.29308,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.006075032552083333,
+      "learning_rate": 0.0001,
+      "loss": 4.4346,
+      "loss/crossentropy": 2.5529074668884277,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23994534462690353,
+      "step": 14654
+    },
+    {
+      "epoch": 0.29312,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.0100982666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3154,
+      "loss/crossentropy": 2.1320537328720093,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21480896323919296,
+      "step": 14656
+    },
+    {
+      "epoch": 0.29316,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.013205718994140626,
+      "learning_rate": 0.0001,
+      "loss": 3.867,
+      "loss/crossentropy": 1.8921156525611877,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1749916449189186,
+      "step": 14658
+    },
+    {
+      "epoch": 0.2932,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.015433502197265626,
+      "learning_rate": 0.0001,
+      "loss": 3.7405,
+      "loss/crossentropy": 2.2475300431251526,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20675917714834213,
+      "step": 14660
+    },
+    {
+      "epoch": 0.29324,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.015384928385416666,
+      "learning_rate": 0.0001,
+      "loss": 4.2091,
+      "loss/crossentropy": 2.291227698326111,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2244400456547737,
+      "step": 14662
+    },
+    {
+      "epoch": 0.29328,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.017976888020833335,
+      "learning_rate": 0.0001,
+      "loss": 4.1468,
+      "loss/crossentropy": 1.8830837607383728,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21137318760156631,
+      "step": 14664
+    },
+    {
+      "epoch": 0.29332,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.019742838541666665,
+      "learning_rate": 0.0001,
+      "loss": 4.3424,
+      "loss/crossentropy": 2.366846203804016,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22360631078481674,
+      "step": 14666
+    },
+    {
+      "epoch": 0.29336,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.021720123291015626,
+      "learning_rate": 0.0001,
+      "loss": 4.1778,
+      "loss/crossentropy": 2.165378987789154,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2088456228375435,
+      "step": 14668
+    },
+    {
+      "epoch": 0.2934,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.021809895833333332,
+      "learning_rate": 0.0001,
+      "loss": 4.178,
+      "loss/crossentropy": 1.782981276512146,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17654509842395782,
+      "step": 14670
+    },
+    {
+      "epoch": 0.29344,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.021581013997395832,
+      "learning_rate": 0.0001,
+      "loss": 4.0796,
+      "loss/crossentropy": 1.9891030192375183,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18961766362190247,
+      "step": 14672
+    },
+    {
+      "epoch": 0.29348,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.015738932291666667,
+      "learning_rate": 0.0001,
+      "loss": 4.0359,
+      "loss/crossentropy": 1.9350959062576294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2124071717262268,
+      "step": 14674
+    },
+    {
+      "epoch": 0.29352,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.01131591796875,
+      "learning_rate": 0.0001,
+      "loss": 4.2651,
+      "loss/crossentropy": 2.1763634085655212,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22027570754289627,
+      "step": 14676
+    },
+    {
+      "epoch": 0.29356,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.012499745686848958,
+      "learning_rate": 0.0001,
+      "loss": 4.1502,
+      "loss/crossentropy": 1.741381287574768,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17308862507343292,
+      "step": 14678
+    },
+    {
+      "epoch": 0.2936,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011633046468098958,
+      "learning_rate": 0.0001,
+      "loss": 4.2012,
+      "loss/crossentropy": 2.352774977684021,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20929308235645294,
+      "step": 14680
+    },
+    {
+      "epoch": 0.29364,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.012679036458333333,
+      "learning_rate": 0.0001,
+      "loss": 3.9799,
+      "loss/crossentropy": 1.6342085003852844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17740759253501892,
+      "step": 14682
+    },
+    {
+      "epoch": 0.29368,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0110260009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1968,
+      "loss/crossentropy": 2.315647602081299,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2202264443039894,
+      "step": 14684
+    },
+    {
+      "epoch": 0.29372,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.011860911051432292,
+      "learning_rate": 0.0001,
+      "loss": 4.0276,
+      "loss/crossentropy": 2.1667529344558716,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20557944476604462,
+      "step": 14686
+    },
+    {
+      "epoch": 0.29376,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.012970987955729167,
+      "learning_rate": 0.0001,
+      "loss": 3.7212,
+      "loss/crossentropy": 1.9426026940345764,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17865119129419327,
+      "step": 14688
+    },
+    {
+      "epoch": 0.2938,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.013720703125,
+      "learning_rate": 0.0001,
+      "loss": 3.9345,
+      "loss/crossentropy": 2.012951970100403,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1918969601392746,
+      "step": 14690
+    },
+    {
+      "epoch": 0.29384,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0097808837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2131,
+      "loss/crossentropy": 2.2038668394088745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21520675718784332,
+      "step": 14692
+    },
+    {
+      "epoch": 0.29388,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.009659576416015624,
+      "learning_rate": 0.0001,
+      "loss": 3.9577,
+      "loss/crossentropy": 2.284528613090515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2084675058722496,
+      "step": 14694
+    },
+    {
+      "epoch": 0.29392,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.009991200764973958,
+      "learning_rate": 0.0001,
+      "loss": 4.1159,
+      "loss/crossentropy": 2.3619518280029297,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23107902705669403,
+      "step": 14696
+    },
+    {
+      "epoch": 0.29396,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009870402018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1488,
+      "loss/crossentropy": 2.279863119125366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21235084533691406,
+      "step": 14698
+    },
+    {
+      "epoch": 0.294,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.009639231363932292,
+      "learning_rate": 0.0001,
+      "loss": 4.2216,
+      "loss/crossentropy": 2.128455936908722,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21381118893623352,
+      "step": 14700
+    },
+    {
+      "epoch": 0.29404,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.009014638264973958,
+      "learning_rate": 0.0001,
+      "loss": 4.0147,
+      "loss/crossentropy": 2.265346884727478,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2086036428809166,
+      "step": 14702
+    },
+    {
+      "epoch": 0.29408,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.008421834309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2778,
+      "loss/crossentropy": 2.4323991537094116,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21795648336410522,
+      "step": 14704
+    },
+    {
+      "epoch": 0.29412,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.30881729125976565,
+      "learning_rate": 0.0001,
+      "loss": 4.3864,
+      "loss/crossentropy": 2.2948319911956787,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23302766680717468,
+      "step": 14706
+    },
+    {
+      "epoch": 0.29416,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.3063140869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.121,
+      "loss/crossentropy": 2.1273980140686035,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2240317091345787,
+      "step": 14708
+    },
+    {
+      "epoch": 0.2942,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.304357655843099,
+      "learning_rate": 0.0001,
+      "loss": 3.7468,
+      "loss/crossentropy": 2.0593321323394775,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19727306067943573,
+      "step": 14710
+    },
+    {
+      "epoch": 0.29424,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.3060373942057292,
+      "learning_rate": 0.0001,
+      "loss": 3.9979,
+      "loss/crossentropy": 2.339016914367676,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2313404530286789,
+      "step": 14712
+    },
+    {
+      "epoch": 0.29428,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.30822652180989585,
+      "learning_rate": 0.0001,
+      "loss": 3.9406,
+      "loss/crossentropy": 2.1587395668029785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20085373520851135,
+      "step": 14714
+    },
+    {
+      "epoch": 0.29432,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.3103912353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0616,
+      "loss/crossentropy": 1.8317759037017822,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20150133967399597,
+      "step": 14716
+    },
+    {
+      "epoch": 0.29436,
+      "grad_norm": 1.7890625,
+      "grad_norm_var": 0.31324437459309895,
+      "learning_rate": 0.0001,
+      "loss": 3.9291,
+      "loss/crossentropy": 1.849327266216278,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17959115654230118,
+      "step": 14718
+    },
+    {
+      "epoch": 0.2944,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.32099507649739584,
+      "learning_rate": 0.0001,
+      "loss": 3.7933,
+      "loss/crossentropy": 1.652997612953186,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17243028432130814,
+      "step": 14720
+    },
+    {
+      "epoch": 0.29444,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.012995402018229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2919,
+      "loss/crossentropy": 1.968604028224945,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20564967393875122,
+      "step": 14722
+    },
+    {
+      "epoch": 0.29448,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.007466634114583333,
+      "learning_rate": 0.0001,
+      "loss": 4.1211,
+      "loss/crossentropy": 2.349083185195923,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24004054814577103,
+      "step": 14724
+    },
+    {
+      "epoch": 0.29452,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.006200917561848958,
+      "learning_rate": 0.0001,
+      "loss": 4.2078,
+      "loss/crossentropy": 2.1629676818847656,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2260306179523468,
+      "step": 14726
+    },
+    {
+      "epoch": 0.29456,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.012572224934895833,
+      "learning_rate": 0.0001,
+      "loss": 4.5995,
+      "loss/crossentropy": 2.221126079559326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21054977923631668,
+      "step": 14728
+    },
+    {
+      "epoch": 0.2946,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.3638987223307292,
+      "learning_rate": 0.0001,
+      "loss": 4.2588,
+      "loss/crossentropy": 2.0333986282348633,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27580036222934723,
+      "step": 14730
+    },
+    {
+      "epoch": 0.29464,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.36292699178059895,
+      "learning_rate": 0.0001,
+      "loss": 4.2601,
+      "loss/crossentropy": 2.0025470852851868,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20166245102882385,
+      "step": 14732
+    },
+    {
+      "epoch": 0.29468,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.35410868326822914,
+      "learning_rate": 0.0001,
+      "loss": 4.1044,
+      "loss/crossentropy": 2.2882933616638184,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1972086951136589,
+      "step": 14734
+    },
+    {
+      "epoch": 0.29472,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.3395342508951823,
+      "learning_rate": 0.0001,
+      "loss": 4.1616,
+      "loss/crossentropy": 2.2404085397720337,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22015579044818878,
+      "step": 14736
+    },
+    {
+      "epoch": 0.29476,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.33505223592122396,
+      "learning_rate": 0.0001,
+      "loss": 4.0132,
+      "loss/crossentropy": 1.9309356808662415,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2082076519727707,
+      "step": 14738
+    },
+    {
+      "epoch": 0.2948,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.33227437337239585,
+      "learning_rate": 0.0001,
+      "loss": 4.1652,
+      "loss/crossentropy": 2.2118934988975525,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20175088196992874,
+      "step": 14740
+    },
+    {
+      "epoch": 0.29484,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.36171875,
+      "learning_rate": 0.0001,
+      "loss": 4.1204,
+      "loss/crossentropy": 2.19997900724411,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22470968216657639,
+      "step": 14742
+    },
+    {
+      "epoch": 0.29488,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.365966796875,
+      "learning_rate": 0.0001,
+      "loss": 4.2159,
+      "loss/crossentropy": 1.8140272498130798,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20032335817813873,
+      "step": 14744
+    },
+    {
+      "epoch": 0.29492,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.07648086547851562,
+      "learning_rate": 0.0001,
+      "loss": 3.957,
+      "loss/crossentropy": 2.0233620405197144,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20464400947093964,
+      "step": 14746
+    },
+    {
+      "epoch": 0.29496,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.07643407185872396,
+      "learning_rate": 0.0001,
+      "loss": 4.3946,
+      "loss/crossentropy": 2.3900705575942993,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23763202875852585,
+      "step": 14748
+    },
+    {
+      "epoch": 0.295,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.07768325805664063,
+      "learning_rate": 0.0001,
+      "loss": 4.205,
+      "loss/crossentropy": 1.9895538687705994,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20736829191446304,
+      "step": 14750
+    },
+    {
+      "epoch": 0.29504,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0857421875,
+      "learning_rate": 0.0001,
+      "loss": 4.1454,
+      "loss/crossentropy": 1.7315555810928345,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1839754432439804,
+      "step": 14752
+    },
+    {
+      "epoch": 0.29508,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.08105367024739583,
+      "learning_rate": 0.0001,
+      "loss": 4.5482,
+      "loss/crossentropy": 2.44241464138031,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2505839467048645,
+      "step": 14754
+    },
+    {
+      "epoch": 0.29512,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0849029541015625,
+      "learning_rate": 0.0001,
+      "loss": 3.782,
+      "loss/crossentropy": 1.7948896884918213,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1987960934638977,
+      "step": 14756
+    },
+    {
+      "epoch": 0.29516,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.020334625244140626,
+      "learning_rate": 0.0001,
+      "loss": 4.0898,
+      "loss/crossentropy": 1.710760235786438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18624412268400192,
+      "step": 14758
+    },
+    {
+      "epoch": 0.2952,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.027205149332682293,
+      "learning_rate": 0.0001,
+      "loss": 4.4651,
+      "loss/crossentropy": 2.0127750635147095,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20178033411502838,
+      "step": 14760
+    },
+    {
+      "epoch": 0.29524,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.027042388916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1152,
+      "loss/crossentropy": 1.9764790534973145,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19000716507434845,
+      "step": 14762
+    },
+    {
+      "epoch": 0.29528,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.020930735270182292,
+      "learning_rate": 0.0001,
+      "loss": 3.9634,
+      "loss/crossentropy": 2.0221771597862244,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1980113685131073,
+      "step": 14764
+    },
+    {
+      "epoch": 0.29532,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.019972483317057293,
+      "learning_rate": 0.0001,
+      "loss": 4.1721,
+      "loss/crossentropy": 2.278168559074402,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.231141597032547,
+      "step": 14766
+    },
+    {
+      "epoch": 0.29536,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.0168853759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1308,
+      "loss/crossentropy": 2.1837246417999268,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2026391476392746,
+      "step": 14768
+    },
+    {
+      "epoch": 0.2954,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.019795735677083332,
+      "learning_rate": 0.0001,
+      "loss": 4.0694,
+      "loss/crossentropy": 2.1170668601989746,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1871805638074875,
+      "step": 14770
+    },
+    {
+      "epoch": 0.29544,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.01784032185872396,
+      "learning_rate": 0.0001,
+      "loss": 4.1003,
+      "loss/crossentropy": 1.9905366897583008,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1951729580760002,
+      "step": 14772
+    },
+    {
+      "epoch": 0.29548,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.017146809895833334,
+      "learning_rate": 0.0001,
+      "loss": 3.9811,
+      "loss/crossentropy": 1.781678318977356,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19464702904224396,
+      "step": 14774
+    },
+    {
+      "epoch": 0.29552,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.012303670247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1944,
+      "loss/crossentropy": 2.2671462297439575,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22269655019044876,
+      "step": 14776
+    },
+    {
+      "epoch": 0.29556,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.012511952718098959,
+      "learning_rate": 0.0001,
+      "loss": 4.2128,
+      "loss/crossentropy": 2.0294137001037598,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21939906477928162,
+      "step": 14778
+    },
+    {
+      "epoch": 0.2956,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.011264801025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1295,
+      "loss/crossentropy": 1.9174134731292725,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19592813402414322,
+      "step": 14780
+    },
+    {
+      "epoch": 0.29564,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.009528605143229167,
+      "learning_rate": 0.0001,
+      "loss": 3.9877,
+      "loss/crossentropy": 2.0033875703811646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2121451273560524,
+      "step": 14782
+    },
+    {
+      "epoch": 0.29568,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.09626439412434896,
+      "learning_rate": 0.0001,
+      "loss": 4.0263,
+      "loss/crossentropy": 2.209542691707611,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18373841792345047,
+      "step": 14784
+    },
+    {
+      "epoch": 0.29572,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.09244155883789062,
+      "learning_rate": 0.0001,
+      "loss": 4.149,
+      "loss/crossentropy": 2.0651984214782715,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20486845821142197,
+      "step": 14786
+    },
+    {
+      "epoch": 0.29576,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.09123433430989583,
+      "learning_rate": 0.0001,
+      "loss": 4.1697,
+      "loss/crossentropy": 2.311215043067932,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22553718090057373,
+      "step": 14788
+    },
+    {
+      "epoch": 0.2958,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.09251708984375,
+      "learning_rate": 0.0001,
+      "loss": 4.3291,
+      "loss/crossentropy": 2.0282764434814453,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20748327672481537,
+      "step": 14790
+    },
+    {
+      "epoch": 0.29584,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.09575093587239583,
+      "learning_rate": 0.0001,
+      "loss": 3.764,
+      "loss/crossentropy": 2.04198157787323,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19687633216381073,
+      "step": 14792
+    },
+    {
+      "epoch": 0.29588,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.09601236979166666,
+      "learning_rate": 0.0001,
+      "loss": 4.389,
+      "loss/crossentropy": 2.049328565597534,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20265275985002518,
+      "step": 14794
+    },
+    {
+      "epoch": 0.29592,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0992876688639323,
+      "learning_rate": 0.0001,
+      "loss": 3.9118,
+      "loss/crossentropy": 1.905173659324646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.193840891122818,
+      "step": 14796
+    },
+    {
+      "epoch": 0.29596,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.09698486328125,
+      "learning_rate": 0.0001,
+      "loss": 3.941,
+      "loss/crossentropy": 1.8715303540229797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18632445484399796,
+      "step": 14798
+    },
+    {
+      "epoch": 0.296,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.013765207926432292,
+      "learning_rate": 0.0001,
+      "loss": 4.039,
+      "loss/crossentropy": 2.0428889989852905,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19538581371307373,
+      "step": 14800
+    },
+    {
+      "epoch": 0.29604,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.015553538004557292,
+      "learning_rate": 0.0001,
+      "loss": 4.0834,
+      "loss/crossentropy": 2.2959046363830566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2372349202632904,
+      "step": 14802
+    },
+    {
+      "epoch": 0.29608,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.016190592447916666,
+      "learning_rate": 0.0001,
+      "loss": 4.0399,
+      "loss/crossentropy": 2.083451807498932,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20915375649929047,
+      "step": 14804
+    },
+    {
+      "epoch": 0.29612,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0132232666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.293,
+      "loss/crossentropy": 2.31546950340271,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20648062229156494,
+      "step": 14806
+    },
+    {
+      "epoch": 0.29616,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.013019816080729166,
+      "learning_rate": 0.0001,
+      "loss": 3.7273,
+      "loss/crossentropy": 1.9833638072013855,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19132772833108902,
+      "step": 14808
+    },
+    {
+      "epoch": 0.2962,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.013655344645182291,
+      "learning_rate": 0.0001,
+      "loss": 3.8247,
+      "loss/crossentropy": 1.784917414188385,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19032159447669983,
+      "step": 14810
+    },
+    {
+      "epoch": 0.29624,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.011263020833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.0011,
+      "loss/crossentropy": 1.843966782093048,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22127597779035568,
+      "step": 14812
+    },
+    {
+      "epoch": 0.29628,
+      "grad_norm": 2.921875,
+      "grad_norm_var": 0.06729100545247396,
+      "learning_rate": 0.0001,
+      "loss": 4.0535,
+      "loss/crossentropy": 1.8107360005378723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18275444209575653,
+      "step": 14814
+    },
+    {
+      "epoch": 0.29632,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.06665827433268229,
+      "learning_rate": 0.0001,
+      "loss": 4.2711,
+      "loss/crossentropy": 2.147618055343628,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23800157755613327,
+      "step": 14816
+    },
+    {
+      "epoch": 0.29636,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.06475804646809896,
+      "learning_rate": 0.0001,
+      "loss": 4.267,
+      "loss/crossentropy": 2.4430278539657593,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2401380091905594,
+      "step": 14818
+    },
+    {
+      "epoch": 0.2964,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.06377665201822917,
+      "learning_rate": 0.0001,
+      "loss": 4.3204,
+      "loss/crossentropy": 2.181501626968384,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22491848468780518,
+      "step": 14820
+    },
+    {
+      "epoch": 0.29644,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.06398111979166667,
+      "learning_rate": 0.0001,
+      "loss": 4.0375,
+      "loss/crossentropy": 2.101313889026642,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21598663926124573,
+      "step": 14822
+    },
+    {
+      "epoch": 0.29648,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.059020741780598955,
+      "learning_rate": 0.0001,
+      "loss": 3.9629,
+      "loss/crossentropy": 1.6834549307823181,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19430063664913177,
+      "step": 14824
+    },
+    {
+      "epoch": 0.29652,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0566162109375,
+      "learning_rate": 0.0001,
+      "loss": 4.0087,
+      "loss/crossentropy": 1.8400229215621948,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18764454871416092,
+      "step": 14826
+    },
+    {
+      "epoch": 0.29656,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.05609944661458333,
+      "learning_rate": 0.0001,
+      "loss": 4.1427,
+      "loss/crossentropy": 1.9417667388916016,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21304062008857727,
+      "step": 14828
+    },
+    {
+      "epoch": 0.2966,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.003639475504557292,
+      "learning_rate": 0.0001,
+      "loss": 4.125,
+      "loss/crossentropy": 2.227192521095276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23011230677366257,
+      "step": 14830
+    },
+    {
+      "epoch": 0.29664,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0029436747233072915,
+      "learning_rate": 0.0001,
+      "loss": 4.3496,
+      "loss/crossentropy": 2.2136365175247192,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21694285422563553,
+      "step": 14832
+    },
+    {
+      "epoch": 0.29668,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.004137929280598958,
+      "learning_rate": 0.0001,
+      "loss": 3.9464,
+      "loss/crossentropy": 1.9179469347000122,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1996636986732483,
+      "step": 14834
+    },
+    {
+      "epoch": 0.29672,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0043718973795572914,
+      "learning_rate": 0.0001,
+      "loss": 3.9001,
+      "loss/crossentropy": 1.9639176726341248,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17738713324069977,
+      "step": 14836
+    },
+    {
+      "epoch": 0.29676,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.004564412434895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2665,
+      "loss/crossentropy": 1.7748695611953735,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18663641810417175,
+      "step": 14838
+    },
+    {
+      "epoch": 0.2968,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.007054646809895833,
+      "learning_rate": 0.0001,
+      "loss": 3.9879,
+      "loss/crossentropy": 2.094521999359131,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18323734402656555,
+      "step": 14840
+    },
+    {
+      "epoch": 0.29684,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.01407470703125,
+      "learning_rate": 0.0001,
+      "loss": 4.5969,
+      "loss/crossentropy": 2.488257050514221,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23173442482948303,
+      "step": 14842
+    },
+    {
+      "epoch": 0.29688,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.013444010416666667,
+      "learning_rate": 0.0001,
+      "loss": 4.2015,
+      "loss/crossentropy": 2.2317086458206177,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2046324908733368,
+      "step": 14844
+    },
+    {
+      "epoch": 0.29692,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.014134724934895834,
+      "learning_rate": 0.0001,
+      "loss": 3.9917,
+      "loss/crossentropy": 2.275767207145691,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20739831775426865,
+      "step": 14846
+    },
+    {
+      "epoch": 0.29696,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.014501698811848958,
+      "learning_rate": 0.0001,
+      "loss": 3.9728,
+      "loss/crossentropy": 2.0288257002830505,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18860980868339539,
+      "step": 14848
+    },
+    {
+      "epoch": 0.297,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.015449778238932291,
+      "learning_rate": 0.0001,
+      "loss": 4.3509,
+      "loss/crossentropy": 2.0609896183013916,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20585624873638153,
+      "step": 14850
+    },
+    {
+      "epoch": 0.29704,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.018214670817057292,
+      "learning_rate": 0.0001,
+      "loss": 3.5485,
+      "loss/crossentropy": 1.9250158667564392,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1843906193971634,
+      "step": 14852
+    },
+    {
+      "epoch": 0.29708,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.018070475260416666,
+      "learning_rate": 0.0001,
+      "loss": 3.9761,
+      "loss/crossentropy": 2.118358612060547,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2035902589559555,
+      "step": 14854
+    },
+    {
+      "epoch": 0.29712,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.015610504150390624,
+      "learning_rate": 0.0001,
+      "loss": 4.135,
+      "loss/crossentropy": 1.8381852507591248,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1859317123889923,
+      "step": 14856
+    },
+    {
+      "epoch": 0.29716,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.009513346354166667,
+      "learning_rate": 0.0001,
+      "loss": 4.1235,
+      "loss/crossentropy": 2.078063726425171,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21035100519657135,
+      "step": 14858
+    },
+    {
+      "epoch": 0.2972,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010701497395833334,
+      "learning_rate": 0.0001,
+      "loss": 4.0549,
+      "loss/crossentropy": 2.1546566486358643,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19998939335346222,
+      "step": 14860
+    },
+    {
+      "epoch": 0.29724,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.010164133707682292,
+      "learning_rate": 0.0001,
+      "loss": 4.2264,
+      "loss/crossentropy": 2.019156754016876,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1957200989127159,
+      "step": 14862
+    },
+    {
+      "epoch": 0.29728,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.013138834635416667,
+      "learning_rate": 0.0001,
+      "loss": 4.0808,
+      "loss/crossentropy": 1.8542814254760742,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19331540167331696,
+      "step": 14864
+    },
+    {
+      "epoch": 0.29732,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.010895792643229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2488,
+      "loss/crossentropy": 2.241651773452759,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22117780148983002,
+      "step": 14866
+    },
+    {
+      "epoch": 0.29736,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.011226145426432292,
+      "learning_rate": 0.0001,
+      "loss": 4.2139,
+      "loss/crossentropy": 1.8856277465820312,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19262754172086716,
+      "step": 14868
+    },
+    {
+      "epoch": 0.2974,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.013388824462890626,
+      "learning_rate": 0.0001,
+      "loss": 4.3315,
+      "loss/crossentropy": 2.0159581899642944,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21892660111188889,
+      "step": 14870
+    },
+    {
+      "epoch": 0.29744,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.014562733968098958,
+      "learning_rate": 0.0001,
+      "loss": 3.8965,
+      "loss/crossentropy": 1.8794240355491638,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18827088177204132,
+      "step": 14872
+    },
+    {
+      "epoch": 0.29748,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.013726552327473959,
+      "learning_rate": 0.0001,
+      "loss": 4.2176,
+      "loss/crossentropy": 2.3741602897644043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2292410209774971,
+      "step": 14874
+    },
+    {
+      "epoch": 0.29752,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.014025624593098958,
+      "learning_rate": 0.0001,
+      "loss": 3.8625,
+      "loss/crossentropy": 1.9994327425956726,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21266867220401764,
+      "step": 14876
+    },
+    {
+      "epoch": 0.29756,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.014143880208333333,
+      "learning_rate": 0.0001,
+      "loss": 4.1337,
+      "loss/crossentropy": 2.0914021730422974,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2071390450000763,
+      "step": 14878
+    },
+    {
+      "epoch": 0.2976,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.012798817952473958,
+      "learning_rate": 0.0001,
+      "loss": 4.0463,
+      "loss/crossentropy": 1.8055935502052307,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19375848025083542,
+      "step": 14880
+    },
+    {
+      "epoch": 0.29764,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.012896474202473958,
+      "learning_rate": 0.0001,
+      "loss": 4.3662,
+      "loss/crossentropy": 2.614544630050659,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22156962752342224,
+      "step": 14882
+    },
+    {
+      "epoch": 0.29768,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.013205718994140626,
+      "learning_rate": 0.0001,
+      "loss": 3.6502,
+      "loss/crossentropy": 1.7306513786315918,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17159338295459747,
+      "step": 14884
+    },
+    {
+      "epoch": 0.29772,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.010060373942057292,
+      "learning_rate": 0.0001,
+      "loss": 4.0683,
+      "loss/crossentropy": 2.060949981212616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20615212619304657,
+      "step": 14886
+    },
+    {
+      "epoch": 0.29776,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.008756510416666667,
+      "learning_rate": 0.0001,
+      "loss": 3.9903,
+      "loss/crossentropy": 1.8338764309883118,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20274019241333008,
+      "step": 14888
+    },
+    {
+      "epoch": 0.2978,
+      "grad_norm": 1.78125,
+      "grad_norm_var": 0.013683827718098958,
+      "learning_rate": 0.0001,
+      "loss": 4.0335,
+      "loss/crossentropy": 2.2182517051696777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22446341067552567,
+      "step": 14890
+    },
+    {
+      "epoch": 0.29784,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.012951405843098958,
+      "learning_rate": 0.0001,
+      "loss": 4.3924,
+      "loss/crossentropy": 2.222801446914673,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22065874934196472,
+      "step": 14892
+    },
+    {
+      "epoch": 0.29788,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.012808990478515626,
+      "learning_rate": 0.0001,
+      "loss": 4.1315,
+      "loss/crossentropy": 2.2181414365768433,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21953265368938446,
+      "step": 14894
+    },
+    {
+      "epoch": 0.29792,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.013166300455729167,
+      "learning_rate": 0.0001,
+      "loss": 3.654,
+      "loss/crossentropy": 1.9304096102714539,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20096193999052048,
+      "step": 14896
+    },
+    {
+      "epoch": 0.29796,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.012043253580729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2074,
+      "loss/crossentropy": 2.2288341522216797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24108773469924927,
+      "step": 14898
+    },
+    {
+      "epoch": 0.298,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.012550608317057291,
+      "learning_rate": 0.0001,
+      "loss": 4.191,
+      "loss/crossentropy": 2.1711790561676025,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2395533323287964,
+      "step": 14900
+    },
+    {
+      "epoch": 0.29804,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.019482167561848958,
+      "learning_rate": 0.0001,
+      "loss": 4.2686,
+      "loss/crossentropy": 2.0533857345581055,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21801679581403732,
+      "step": 14902
+    },
+    {
+      "epoch": 0.29808,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.01883519490559896,
+      "learning_rate": 0.0001,
+      "loss": 3.8678,
+      "loss/crossentropy": 1.634634256362915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17861492931842804,
+      "step": 14904
+    },
+    {
+      "epoch": 0.29812,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0149322509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2152,
+      "loss/crossentropy": 2.3151358366012573,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22694342583417892,
+      "step": 14906
+    },
+    {
+      "epoch": 0.29816,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.01585871378580729,
+      "learning_rate": 0.0001,
+      "loss": 4.085,
+      "loss/crossentropy": 1.9165552258491516,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18231045454740524,
+      "step": 14908
+    },
+    {
+      "epoch": 0.2982,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.01480712890625,
+      "learning_rate": 0.0001,
+      "loss": 3.9693,
+      "loss/crossentropy": 2.0094637274742126,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19547566026449203,
+      "step": 14910
+    },
+    {
+      "epoch": 0.29824,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.014753214518229167,
+      "learning_rate": 0.0001,
+      "loss": 3.9091,
+      "loss/crossentropy": 1.8764418959617615,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18009892851114273,
+      "step": 14912
+    },
+    {
+      "epoch": 0.29828,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.015868123372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.429,
+      "loss/crossentropy": 2.1255921721458435,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20359216630458832,
+      "step": 14914
+    },
+    {
+      "epoch": 0.29832,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.014891560872395833,
+      "learning_rate": 0.0001,
+      "loss": 3.7889,
+      "loss/crossentropy": 2.1316330432891846,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1970125362277031,
+      "step": 14916
+    },
+    {
+      "epoch": 0.29836,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0089996337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0305,
+      "loss/crossentropy": 1.7825700640678406,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1902567446231842,
+      "step": 14918
+    },
+    {
+      "epoch": 0.2984,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.008565012613932292,
+      "learning_rate": 0.0001,
+      "loss": 3.9577,
+      "loss/crossentropy": 1.982073962688446,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20216460525989532,
+      "step": 14920
+    },
+    {
+      "epoch": 0.29844,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.0060943603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.04,
+      "loss/crossentropy": 2.3210418224334717,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2224876508116722,
+      "step": 14922
+    },
+    {
+      "epoch": 0.29848,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.005535634358723959,
+      "learning_rate": 0.0001,
+      "loss": 4.0509,
+      "loss/crossentropy": 1.7990338206291199,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17401983588933945,
+      "step": 14924
+    },
+    {
+      "epoch": 0.29852,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.010545857747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.4979,
+      "loss/crossentropy": 1.8772737979888916,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21376933157444,
+      "step": 14926
+    },
+    {
+      "epoch": 0.29856,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.009716542561848958,
+      "learning_rate": 0.0001,
+      "loss": 4.0544,
+      "loss/crossentropy": 1.6914128065109253,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17860179394483566,
+      "step": 14928
+    },
+    {
+      "epoch": 0.2986,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.009691365559895833,
+      "learning_rate": 0.0001,
+      "loss": 3.9737,
+      "loss/crossentropy": 1.9545430541038513,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2015315517783165,
+      "step": 14930
+    },
+    {
+      "epoch": 0.29864,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.009175618489583334,
+      "learning_rate": 0.0001,
+      "loss": 3.9181,
+      "loss/crossentropy": 2.0585074424743652,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2029297649860382,
+      "step": 14932
+    },
+    {
+      "epoch": 0.29868,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.009679921468098958,
+      "learning_rate": 0.0001,
+      "loss": 3.8638,
+      "loss/crossentropy": 1.9196518063545227,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1970217153429985,
+      "step": 14934
+    },
+    {
+      "epoch": 0.29872,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0097564697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1366,
+      "loss/crossentropy": 2.1077409982681274,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22366851568222046,
+      "step": 14936
+    },
+    {
+      "epoch": 0.29876,
+      "grad_norm": 3.921875,
+      "grad_norm_var": 0.24059829711914063,
+      "learning_rate": 0.0001,
+      "loss": 3.8552,
+      "loss/crossentropy": 1.7660444974899292,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23386957496404648,
+      "step": 14938
+    },
+    {
+      "epoch": 0.2988,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.23871841430664062,
+      "learning_rate": 0.0001,
+      "loss": 4.2936,
+      "loss/crossentropy": 1.8436493873596191,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19589588046073914,
+      "step": 14940
+    },
+    {
+      "epoch": 0.29884,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.24443333943684895,
+      "learning_rate": 0.0001,
+      "loss": 4.3465,
+      "loss/crossentropy": 2.2123888731002808,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26607823371887207,
+      "step": 14942
+    },
+    {
+      "epoch": 0.29888,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.2546119689941406,
+      "learning_rate": 0.0001,
+      "loss": 4.3234,
+      "loss/crossentropy": 2.70485258102417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2529008686542511,
+      "step": 14944
+    },
+    {
+      "epoch": 0.29892,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.25155843098958336,
+      "learning_rate": 0.0001,
+      "loss": 4.1379,
+      "loss/crossentropy": 1.932865023612976,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22141847014427185,
+      "step": 14946
+    },
+    {
+      "epoch": 0.29896,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.25643310546875,
+      "learning_rate": 0.0001,
+      "loss": 4.0119,
+      "loss/crossentropy": 2.213751196861267,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21382954716682434,
+      "step": 14948
+    },
+    {
+      "epoch": 0.299,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.251073964436849,
+      "learning_rate": 0.0001,
+      "loss": 3.9673,
+      "loss/crossentropy": 1.679059088230133,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16484958678483963,
+      "step": 14950
+    },
+    {
+      "epoch": 0.29904,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.24972508748372396,
+      "learning_rate": 0.0001,
+      "loss": 3.9873,
+      "loss/crossentropy": 1.748594582080841,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20150119066238403,
+      "step": 14952
+    },
+    {
+      "epoch": 0.29908,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.05215657552083333,
+      "learning_rate": 0.0001,
+      "loss": 4.8053,
+      "loss/crossentropy": 2.4588215351104736,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2513396218419075,
+      "step": 14954
+    },
+    {
+      "epoch": 0.29912,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.05468114217122396,
+      "learning_rate": 0.0001,
+      "loss": 4.0782,
+      "loss/crossentropy": 2.1104516983032227,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20738063752651215,
+      "step": 14956
+    },
+    {
+      "epoch": 0.29916,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.04905776977539063,
+      "learning_rate": 0.0001,
+      "loss": 4.2351,
+      "loss/crossentropy": 1.7171857953071594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27423302084207535,
+      "step": 14958
+    },
+    {
+      "epoch": 0.2992,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.03264058430989583,
+      "learning_rate": 0.0001,
+      "loss": 3.7507,
+      "loss/crossentropy": 1.5607159733772278,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1616373285651207,
+      "step": 14960
+    },
+    {
+      "epoch": 0.29924,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.033394114176432295,
+      "learning_rate": 0.0001,
+      "loss": 4.0128,
+      "loss/crossentropy": 1.9257365465164185,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18031761050224304,
+      "step": 14962
+    },
+    {
+      "epoch": 0.29928,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.031062825520833334,
+      "learning_rate": 0.0001,
+      "loss": 3.9406,
+      "loss/crossentropy": 1.876187801361084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19531002640724182,
+      "step": 14964
+    },
+    {
+      "epoch": 0.29932,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0317047119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2028,
+      "loss/crossentropy": 1.9325042963027954,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19496598094701767,
+      "step": 14966
+    },
+    {
+      "epoch": 0.29936,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.03322652180989583,
+      "learning_rate": 0.0001,
+      "loss": 4.1293,
+      "loss/crossentropy": 2.148952007293701,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18833627551794052,
+      "step": 14968
+    },
+    {
+      "epoch": 0.2994,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008099110921223958,
+      "learning_rate": 0.0001,
+      "loss": 4.0954,
+      "loss/crossentropy": 2.3639365434646606,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21851496398448944,
+      "step": 14970
+    },
+    {
+      "epoch": 0.29944,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.008371734619140625,
+      "learning_rate": 0.0001,
+      "loss": 3.8026,
+      "loss/crossentropy": 1.941165030002594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19548548012971878,
+      "step": 14972
+    },
+    {
+      "epoch": 0.29948,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0082916259765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1203,
+      "loss/crossentropy": 2.030466139316559,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1922580599784851,
+      "step": 14974
+    },
+    {
+      "epoch": 0.29952,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.008487955729166666,
+      "learning_rate": 0.0001,
+      "loss": 4.2854,
+      "loss/crossentropy": 2.2279865741729736,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20831073075532913,
+      "step": 14976
+    },
+    {
+      "epoch": 0.29956,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.007047271728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1708,
+      "loss/crossentropy": 2.0749244689941406,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18032152205705643,
+      "step": 14978
+    },
+    {
+      "epoch": 0.2996,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.04270731608072917,
+      "learning_rate": 0.0001,
+      "loss": 4.199,
+      "loss/crossentropy": 2.3190104961395264,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24735668301582336,
+      "step": 14980
+    },
+    {
+      "epoch": 0.29964,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.044406890869140625,
+      "learning_rate": 0.0001,
+      "loss": 3.9549,
+      "loss/crossentropy": 1.970013439655304,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19323202222585678,
+      "step": 14982
+    },
+    {
+      "epoch": 0.29968,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.045169830322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.4762,
+      "loss/crossentropy": 2.052343726158142,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20909954607486725,
+      "step": 14984
+    },
+    {
+      "epoch": 0.29972,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.043268839518229164,
+      "learning_rate": 0.0001,
+      "loss": 4.1776,
+      "loss/crossentropy": 2.2586673498153687,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22640614211559296,
+      "step": 14986
+    },
+    {
+      "epoch": 0.29976,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.042557779947916666,
+      "learning_rate": 0.0001,
+      "loss": 3.8465,
+      "loss/crossentropy": 1.8843002319335938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19117384403944016,
+      "step": 14988
+    },
+    {
+      "epoch": 0.2998,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.042740885416666666,
+      "learning_rate": 0.0001,
+      "loss": 3.7343,
+      "loss/crossentropy": 1.542019009590149,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18295453488826752,
+      "step": 14990
+    },
+    {
+      "epoch": 0.29984,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.043369293212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1335,
+      "loss/crossentropy": 2.160573959350586,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22916415333747864,
+      "step": 14992
+    },
+    {
+      "epoch": 0.29988,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.062168121337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1898,
+      "loss/crossentropy": 2.1296870708465576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20498204976320267,
+      "step": 14994
+    },
+    {
+      "epoch": 0.29992,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.0336669921875,
+      "learning_rate": 0.0001,
+      "loss": 4.0554,
+      "loss/crossentropy": 2.0356279015541077,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21798957884311676,
+      "step": 14996
+    },
+    {
+      "epoch": 0.29996,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.035796864827473955,
+      "learning_rate": 0.0001,
+      "loss": 4.6195,
+      "loss/crossentropy": 2.466023027896881,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22107623517513275,
+      "step": 14998
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.03408203125,
+      "learning_rate": 0.0001,
+      "loss": 4.1455,
+      "loss/crossentropy": 1.8457531332969666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21306351572275162,
+      "step": 15000
+    },
+    {
+      "epoch": 0.30004,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.034211222330729166,
+      "learning_rate": 0.0001,
+      "loss": 4.5128,
+      "loss/crossentropy": 2.202435612678528,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21542339771986008,
+      "step": 15002
+    },
+    {
+      "epoch": 0.30008,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.032613118489583336,
+      "learning_rate": 0.0001,
+      "loss": 4.1489,
+      "loss/crossentropy": 2.142255425453186,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21016352623701096,
+      "step": 15004
+    },
+    {
+      "epoch": 0.30012,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.030631256103515626,
+      "learning_rate": 0.0001,
+      "loss": 4.1443,
+      "loss/crossentropy": 1.949910044670105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19256845861673355,
+      "step": 15006
+    },
+    {
+      "epoch": 0.30016,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.029750315348307292,
+      "learning_rate": 0.0001,
+      "loss": 4.34,
+      "loss/crossentropy": 2.257867217063904,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22471557557582855,
+      "step": 15008
+    },
+    {
+      "epoch": 0.3002,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.011358388264973958,
+      "learning_rate": 0.0001,
+      "loss": 4.3457,
+      "loss/crossentropy": 2.1541898250579834,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23054643720388412,
+      "step": 15010
+    },
+    {
+      "epoch": 0.30024,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.01165771484375,
+      "learning_rate": 0.0001,
+      "loss": 4.2537,
+      "loss/crossentropy": 2.0029123425483704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21815954893827438,
+      "step": 15012
+    },
+    {
+      "epoch": 0.30028,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0076416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0093,
+      "loss/crossentropy": 2.0617172718048096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19382908195257187,
+      "step": 15014
+    },
+    {
+      "epoch": 0.30032,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.00921630859375,
+      "learning_rate": 0.0001,
+      "loss": 4.3123,
+      "loss/crossentropy": 2.1493303775787354,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21762489527463913,
+      "step": 15016
+    },
+    {
+      "epoch": 0.30036,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0093505859375,
+      "learning_rate": 0.0001,
+      "loss": 4.1282,
+      "loss/crossentropy": 2.081854462623596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22617685049772263,
+      "step": 15018
+    },
+    {
+      "epoch": 0.3004,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.009842681884765624,
+      "learning_rate": 0.0001,
+      "loss": 4.0352,
+      "loss/crossentropy": 1.8565402626991272,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18223749846220016,
+      "step": 15020
+    },
+    {
+      "epoch": 0.30044,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.010815175374348958,
+      "learning_rate": 0.0001,
+      "loss": 4.1779,
+      "loss/crossentropy": 2.129835605621338,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21263620257377625,
+      "step": 15022
+    },
+    {
+      "epoch": 0.30048,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.011189524332682292,
+      "learning_rate": 0.0001,
+      "loss": 4.5131,
+      "loss/crossentropy": 2.113425612449646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22930295765399933,
+      "step": 15024
+    },
+    {
+      "epoch": 0.30052,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.010109202067057291,
+      "learning_rate": 0.0001,
+      "loss": 4.1557,
+      "loss/crossentropy": 2.254178762435913,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21644078195095062,
+      "step": 15026
+    },
+    {
+      "epoch": 0.30056,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.011966959635416666,
+      "learning_rate": 0.0001,
+      "loss": 4.0863,
+      "loss/crossentropy": 2.233784854412079,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22132302820682526,
+      "step": 15028
+    },
+    {
+      "epoch": 0.3006,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.0118804931640625,
+      "learning_rate": 0.0001,
+      "loss": 3.8657,
+      "loss/crossentropy": 2.2234549522399902,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2144894078373909,
+      "step": 15030
+    },
+    {
+      "epoch": 0.30064,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0108306884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1315,
+      "loss/crossentropy": 2.0401915907859802,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20804419368505478,
+      "step": 15032
+    },
+    {
+      "epoch": 0.30068,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.010978190104166667,
+      "learning_rate": 0.0001,
+      "loss": 4.0939,
+      "loss/crossentropy": 2.058899462223053,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2222386747598648,
+      "step": 15034
+    },
+    {
+      "epoch": 0.30072,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.020694986979166666,
+      "learning_rate": 0.0001,
+      "loss": 4.5452,
+      "loss/crossentropy": 2.396425485610962,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24734684079885483,
+      "step": 15036
+    },
+    {
+      "epoch": 0.30076,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.02011693318684896,
+      "learning_rate": 0.0001,
+      "loss": 4.1124,
+      "loss/crossentropy": 1.9825797080993652,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20504453778266907,
+      "step": 15038
+    },
+    {
+      "epoch": 0.3008,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.02068049112955729,
+      "learning_rate": 0.0001,
+      "loss": 4.0287,
+      "loss/crossentropy": 1.9550088047981262,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2051691859960556,
+      "step": 15040
+    },
+    {
+      "epoch": 0.30084,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.02102839152018229,
+      "learning_rate": 0.0001,
+      "loss": 4.1926,
+      "loss/crossentropy": 2.166532874107361,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20486898720264435,
+      "step": 15042
+    },
+    {
+      "epoch": 0.30088,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.0158355712890625,
+      "learning_rate": 0.0001,
+      "loss": 3.9905,
+      "loss/crossentropy": 1.8752986192703247,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20286446809768677,
+      "step": 15044
+    },
+    {
+      "epoch": 0.30092,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.016649373372395835,
+      "learning_rate": 0.0001,
+      "loss": 4.3217,
+      "loss/crossentropy": 1.9599428176879883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2061413824558258,
+      "step": 15046
+    },
+    {
+      "epoch": 0.30096,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.015428670247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2914,
+      "loss/crossentropy": 2.5601563453674316,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2123739868402481,
+      "step": 15048
+    },
+    {
+      "epoch": 0.301,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.015062459309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2784,
+      "loss/crossentropy": 2.075629949569702,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20791151374578476,
+      "step": 15050
+    },
+    {
+      "epoch": 0.30104,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.008532460530598958,
+      "learning_rate": 0.0001,
+      "loss": 4.1578,
+      "loss/crossentropy": 1.8346505165100098,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16620153188705444,
+      "step": 15052
+    },
+    {
+      "epoch": 0.30108,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009382120768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.003,
+      "loss/crossentropy": 1.9715532660484314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20792889595031738,
+      "step": 15054
+    },
+    {
+      "epoch": 0.30112,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.0090728759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.06,
+      "loss/crossentropy": 2.135376811027527,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.218796044588089,
+      "step": 15056
+    },
+    {
+      "epoch": 0.30116,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.00941162109375,
+      "learning_rate": 0.0001,
+      "loss": 4.4339,
+      "loss/crossentropy": 2.095982074737549,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2120692878961563,
+      "step": 15058
+    },
+    {
+      "epoch": 0.3012,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.00911865234375,
+      "learning_rate": 0.0001,
+      "loss": 4.3395,
+      "loss/crossentropy": 2.2836159467697144,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23748520761728287,
+      "step": 15060
+    },
+    {
+      "epoch": 0.30124,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.010163370768229167,
+      "learning_rate": 0.0001,
+      "loss": 3.8877,
+      "loss/crossentropy": 2.073283314704895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19927848130464554,
+      "step": 15062
+    },
+    {
+      "epoch": 0.30128,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010196940104166666,
+      "learning_rate": 0.0001,
+      "loss": 4.1908,
+      "loss/crossentropy": 2.2401102781295776,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22298122942447662,
+      "step": 15064
+    },
+    {
+      "epoch": 0.30132,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.012650299072265624,
+      "learning_rate": 0.0001,
+      "loss": 4.1878,
+      "loss/crossentropy": 2.063979387283325,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19714537262916565,
+      "step": 15066
+    },
+    {
+      "epoch": 0.30136,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.012894694010416667,
+      "learning_rate": 0.0001,
+      "loss": 3.6407,
+      "loss/crossentropy": 2.0346400141716003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18405266851186752,
+      "step": 15068
+    },
+    {
+      "epoch": 0.3014,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.013437652587890625,
+      "learning_rate": 0.0001,
+      "loss": 3.8641,
+      "loss/crossentropy": 2.1211341619491577,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19659722596406937,
+      "step": 15070
+    },
+    {
+      "epoch": 0.30144,
+      "grad_norm": 3.484375,
+      "grad_norm_var": 0.1490966796875,
+      "learning_rate": 0.0001,
+      "loss": 3.9967,
+      "loss/crossentropy": 1.937787652015686,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18829380720853806,
+      "step": 15072
+    },
+    {
+      "epoch": 0.30148,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.1507219950358073,
+      "learning_rate": 0.0001,
+      "loss": 3.922,
+      "loss/crossentropy": 1.9047453999519348,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18311911821365356,
+      "step": 15074
+    },
+    {
+      "epoch": 0.30152,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.15110651652018228,
+      "learning_rate": 0.0001,
+      "loss": 4.3423,
+      "loss/crossentropy": 2.1409407258033752,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21341674029827118,
+      "step": 15076
+    },
+    {
+      "epoch": 0.30156,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.15178197224934895,
+      "learning_rate": 0.0001,
+      "loss": 4.2815,
+      "loss/crossentropy": 2.188783288002014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2406207174062729,
+      "step": 15078
+    },
+    {
+      "epoch": 0.3016,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.15162938435872395,
+      "learning_rate": 0.0001,
+      "loss": 4.368,
+      "loss/crossentropy": 2.088913679122925,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2055455967783928,
+      "step": 15080
+    },
+    {
+      "epoch": 0.30164,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.15350316365559896,
+      "learning_rate": 0.0001,
+      "loss": 3.9039,
+      "loss/crossentropy": 1.9477434754371643,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19234874844551086,
+      "step": 15082
+    },
+    {
+      "epoch": 0.30168,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.1490069071451823,
+      "learning_rate": 0.0001,
+      "loss": 4.2653,
+      "loss/crossentropy": 2.2611928582191467,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2250279188156128,
+      "step": 15084
+    },
+    {
+      "epoch": 0.30172,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.14735514322916668,
+      "learning_rate": 0.0001,
+      "loss": 4.043,
+      "loss/crossentropy": 2.340996742248535,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2201436087489128,
+      "step": 15086
+    },
+    {
+      "epoch": 0.30176,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.01718724568684896,
+      "learning_rate": 0.0001,
+      "loss": 3.9232,
+      "loss/crossentropy": 2.045587956905365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2049652263522148,
+      "step": 15088
+    },
+    {
+      "epoch": 0.3018,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.015265909830729167,
+      "learning_rate": 0.0001,
+      "loss": 3.9266,
+      "loss/crossentropy": 1.6079826951026917,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1584620103240013,
+      "step": 15090
+    },
+    {
+      "epoch": 0.30184,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.013494618733723958,
+      "learning_rate": 0.0001,
+      "loss": 4.1049,
+      "loss/crossentropy": 2.122701048851013,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2095094472169876,
+      "step": 15092
+    },
+    {
+      "epoch": 0.30188,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.008243560791015625,
+      "learning_rate": 0.0001,
+      "loss": 3.8602,
+      "loss/crossentropy": 1.8888922929763794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19282665848731995,
+      "step": 15094
+    },
+    {
+      "epoch": 0.30192,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0063873291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1279,
+      "loss/crossentropy": 2.121204972267151,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1894393339753151,
+      "step": 15096
+    },
+    {
+      "epoch": 0.30196,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.005991363525390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0433,
+      "loss/crossentropy": 1.836807131767273,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1851387470960617,
+      "step": 15098
+    },
+    {
+      "epoch": 0.302,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.005246734619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1723,
+      "loss/crossentropy": 2.170082688331604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20964065939188004,
+      "step": 15100
+    },
+    {
+      "epoch": 0.30204,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0037595113118489582,
+      "learning_rate": 0.0001,
+      "loss": 4.0642,
+      "loss/crossentropy": 2.128955125808716,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20737067610025406,
+      "step": 15102
+    },
+    {
+      "epoch": 0.30208,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.0062744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1561,
+      "loss/crossentropy": 1.970711886882782,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19101983308792114,
+      "step": 15104
+    },
+    {
+      "epoch": 0.30212,
+      "grad_norm": 2.40625,
+      "grad_norm_var": 0.018040974934895832,
+      "learning_rate": 0.0001,
+      "loss": 4.5089,
+      "loss/crossentropy": 2.105665922164917,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23741377890110016,
+      "step": 15106
+    },
+    {
+      "epoch": 0.30216,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.01962865193684896,
+      "learning_rate": 0.0001,
+      "loss": 4.3766,
+      "loss/crossentropy": 2.274402379989624,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20633316040039062,
+      "step": 15108
+    },
+    {
+      "epoch": 0.3022,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.017996978759765626,
+      "learning_rate": 0.0001,
+      "loss": 4.0393,
+      "loss/crossentropy": 2.134031891822815,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20932414382696152,
+      "step": 15110
+    },
+    {
+      "epoch": 0.30224,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.01930720011393229,
+      "learning_rate": 0.0001,
+      "loss": 3.9084,
+      "loss/crossentropy": 2.015448212623596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18043135851621628,
+      "step": 15112
+    },
+    {
+      "epoch": 0.30228,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.018232981363932293,
+      "learning_rate": 0.0001,
+      "loss": 4.2025,
+      "loss/crossentropy": 2.3585838079452515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2323882058262825,
+      "step": 15114
+    },
+    {
+      "epoch": 0.30232,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.018607584635416667,
+      "learning_rate": 0.0001,
+      "loss": 3.8586,
+      "loss/crossentropy": 1.790964961051941,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18150723725557327,
+      "step": 15116
+    },
+    {
+      "epoch": 0.30236,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.018802642822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1433,
+      "loss/crossentropy": 2.0206486582756042,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2102368101477623,
+      "step": 15118
+    },
+    {
+      "epoch": 0.3024,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.016169230143229168,
+      "learning_rate": 0.0001,
+      "loss": 4.1773,
+      "loss/crossentropy": 2.21474289894104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22742661088705063,
+      "step": 15120
+    },
+    {
+      "epoch": 0.30244,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.009781901041666667,
+      "learning_rate": 0.0001,
+      "loss": 3.9536,
+      "loss/crossentropy": 2.0452207922935486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19045353680849075,
+      "step": 15122
+    },
+    {
+      "epoch": 0.30248,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.0105865478515625,
+      "learning_rate": 0.0001,
+      "loss": 3.7914,
+      "loss/crossentropy": 2.019612729549408,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2032184973359108,
+      "step": 15124
+    },
+    {
+      "epoch": 0.30252,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.010221099853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0303,
+      "loss/crossentropy": 2.1821396350860596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22832991182804108,
+      "step": 15126
+    },
+    {
+      "epoch": 0.30256,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.010088857014973958,
+      "learning_rate": 0.0001,
+      "loss": 3.7838,
+      "loss/crossentropy": 2.0395348072052,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18255837261676788,
+      "step": 15128
+    },
+    {
+      "epoch": 0.3026,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.009081013997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.177,
+      "loss/crossentropy": 2.1724050045013428,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19672775268554688,
+      "step": 15130
+    },
+    {
+      "epoch": 0.30264,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.008470662434895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2438,
+      "loss/crossentropy": 2.107967436313629,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19564508646726608,
+      "step": 15132
+    },
+    {
+      "epoch": 0.30268,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.007355753580729167,
+      "learning_rate": 0.0001,
+      "loss": 3.8674,
+      "loss/crossentropy": 1.7179370522499084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17135357856750488,
+      "step": 15134
+    },
+    {
+      "epoch": 0.30272,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.008119455973307292,
+      "learning_rate": 0.0001,
+      "loss": 4.104,
+      "loss/crossentropy": 1.9004579186439514,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19902431219816208,
+      "step": 15136
+    },
+    {
+      "epoch": 0.30276,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0052487691243489586,
+      "learning_rate": 0.0001,
+      "loss": 4.013,
+      "loss/crossentropy": 1.9554911255836487,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19999338686466217,
+      "step": 15138
+    },
+    {
+      "epoch": 0.3028,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.004793294270833333,
+      "learning_rate": 0.0001,
+      "loss": 4.1051,
+      "loss/crossentropy": 1.8687474131584167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19250088185071945,
+      "step": 15140
+    },
+    {
+      "epoch": 0.30284,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.004929351806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1217,
+      "loss/crossentropy": 2.1360538005828857,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22316065430641174,
+      "step": 15142
+    },
+    {
+      "epoch": 0.30288,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.006776682535807292,
+      "learning_rate": 0.0001,
+      "loss": 3.7681,
+      "loss/crossentropy": 1.6100040078163147,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16926150023937225,
+      "step": 15144
+    },
+    {
+      "epoch": 0.30292,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0070302327473958336,
+      "learning_rate": 0.0001,
+      "loss": 4.0459,
+      "loss/crossentropy": 1.983083188533783,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19075944274663925,
+      "step": 15146
+    },
+    {
+      "epoch": 0.30296,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.009110514322916667,
+      "learning_rate": 0.0001,
+      "loss": 3.8722,
+      "loss/crossentropy": 1.9027757048606873,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2069152221083641,
+      "step": 15148
+    },
+    {
+      "epoch": 0.303,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.009144846598307292,
+      "learning_rate": 0.0001,
+      "loss": 3.996,
+      "loss/crossentropy": 1.9141475558280945,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19106873124837875,
+      "step": 15150
+    },
+    {
+      "epoch": 0.30304,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009821573893229166,
+      "learning_rate": 0.0001,
+      "loss": 4.0872,
+      "loss/crossentropy": 1.924220085144043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19870451837778091,
+      "step": 15152
+    },
+    {
+      "epoch": 0.30308,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.009169260660807291,
+      "learning_rate": 0.0001,
+      "loss": 4.1611,
+      "loss/crossentropy": 2.18874990940094,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2040283977985382,
+      "step": 15154
+    },
+    {
+      "epoch": 0.30312,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0096588134765625,
+      "learning_rate": 0.0001,
+      "loss": 3.6905,
+      "loss/crossentropy": 2.0296109914779663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20384181290864944,
+      "step": 15156
+    },
+    {
+      "epoch": 0.30316,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0091949462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2475,
+      "loss/crossentropy": 2.1908326148986816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20775151252746582,
+      "step": 15158
+    },
+    {
+      "epoch": 0.3032,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.007523345947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.191,
+      "loss/crossentropy": 2.1348751187324524,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21342730522155762,
+      "step": 15160
+    },
+    {
+      "epoch": 0.30324,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0075103759765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9642,
+      "loss/crossentropy": 2.03944593667984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1934272050857544,
+      "step": 15162
+    },
+    {
+      "epoch": 0.30328,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.005475870768229167,
+      "learning_rate": 0.0001,
+      "loss": 3.9692,
+      "loss/crossentropy": 1.8574647307395935,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1847086101770401,
+      "step": 15164
+    },
+    {
+      "epoch": 0.30332,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.005671183268229167,
+      "learning_rate": 0.0001,
+      "loss": 4.3025,
+      "loss/crossentropy": 2.219611406326294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1982950195670128,
+      "step": 15166
+    },
+    {
+      "epoch": 0.30336,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.005159250895182292,
+      "learning_rate": 0.0001,
+      "loss": 4.1903,
+      "loss/crossentropy": 2.024726688861847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.217759907245636,
+      "step": 15168
+    },
+    {
+      "epoch": 0.3034,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.005353800455729167,
+      "learning_rate": 0.0001,
+      "loss": 4.2273,
+      "loss/crossentropy": 2.196849226951599,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21290963143110275,
+      "step": 15170
+    },
+    {
+      "epoch": 0.30344,
+      "grad_norm": 1.7421875,
+      "grad_norm_var": 0.00750732421875,
+      "learning_rate": 0.0001,
+      "loss": 4.0096,
+      "loss/crossentropy": 1.9023171067237854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19199151545763016,
+      "step": 15172
+    },
+    {
+      "epoch": 0.30348,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008644358317057291,
+      "learning_rate": 0.0001,
+      "loss": 4.0798,
+      "loss/crossentropy": 2.1286264657974243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20934996753931046,
+      "step": 15174
+    },
+    {
+      "epoch": 0.30352,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.008007558186848958,
+      "learning_rate": 0.0001,
+      "loss": 4.2621,
+      "loss/crossentropy": 2.2262455224990845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2091647908091545,
+      "step": 15176
+    },
+    {
+      "epoch": 0.30356,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.04183349609375,
+      "learning_rate": 0.0001,
+      "loss": 4.2439,
+      "loss/crossentropy": 2.4126710891723633,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22972645610570908,
+      "step": 15178
+    },
+    {
+      "epoch": 0.3036,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.042569732666015624,
+      "learning_rate": 0.0001,
+      "loss": 4.4268,
+      "loss/crossentropy": 2.4708153009414673,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23798953741788864,
+      "step": 15180
+    },
+    {
+      "epoch": 0.30364,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.04053929646809896,
+      "learning_rate": 0.0001,
+      "loss": 4.1841,
+      "loss/crossentropy": 1.9924429655075073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19149084389209747,
+      "step": 15182
+    },
+    {
+      "epoch": 0.30368,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.04182103474934896,
+      "learning_rate": 0.0001,
+      "loss": 4.1335,
+      "loss/crossentropy": 1.8018346428871155,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1883997619152069,
+      "step": 15184
+    },
+    {
+      "epoch": 0.30372,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.04639867146809896,
+      "learning_rate": 0.0001,
+      "loss": 3.96,
+      "loss/crossentropy": 2.001536011695862,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2008758783340454,
+      "step": 15186
+    },
+    {
+      "epoch": 0.30376,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.06099828084309896,
+      "learning_rate": 0.0001,
+      "loss": 4.2308,
+      "loss/crossentropy": 1.9239726066589355,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2068573236465454,
+      "step": 15188
+    },
+    {
+      "epoch": 0.3038,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.059845987955729166,
+      "learning_rate": 0.0001,
+      "loss": 4.303,
+      "loss/crossentropy": 2.252953827381134,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22697407007217407,
+      "step": 15190
+    },
+    {
+      "epoch": 0.30384,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.0631011962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.174,
+      "loss/crossentropy": 2.135131001472473,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2088206186890602,
+      "step": 15192
+    },
+    {
+      "epoch": 0.30388,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0391754150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.22,
+      "loss/crossentropy": 2.394876003265381,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22017298638820648,
+      "step": 15194
+    },
+    {
+      "epoch": 0.30392,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.03862711588541667,
+      "learning_rate": 0.0001,
+      "loss": 4.3951,
+      "loss/crossentropy": 2.3517301082611084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21459681540727615,
+      "step": 15196
+    },
+    {
+      "epoch": 0.30396,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.038914998372395836,
+      "learning_rate": 0.0001,
+      "loss": 4.3068,
+      "loss/crossentropy": 1.956869900226593,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19348695874214172,
+      "step": 15198
+    },
+    {
+      "epoch": 0.304,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.045660146077473956,
+      "learning_rate": 0.0001,
+      "loss": 3.5481,
+      "loss/crossentropy": 1.8349076509475708,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19203926622867584,
+      "step": 15200
+    },
+    {
+      "epoch": 0.30404,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.041265614827473956,
+      "learning_rate": 0.0001,
+      "loss": 4.1749,
+      "loss/crossentropy": 2.253283977508545,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24825416505336761,
+      "step": 15202
+    },
+    {
+      "epoch": 0.30408,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.0156646728515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9585,
+      "loss/crossentropy": 1.7219101190567017,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18302495777606964,
+      "step": 15204
+    },
+    {
+      "epoch": 0.30412,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.010309855143229166,
+      "learning_rate": 0.0001,
+      "loss": 4.4398,
+      "loss/crossentropy": 2.2931004762649536,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21909544616937637,
+      "step": 15206
+    },
+    {
+      "epoch": 0.30416,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.009883626302083334,
+      "learning_rate": 0.0001,
+      "loss": 4.0707,
+      "loss/crossentropy": 1.9331820011138916,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19565977901220322,
+      "step": 15208
+    },
+    {
+      "epoch": 0.3042,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008568318684895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0936,
+      "loss/crossentropy": 1.9863171577453613,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20089460909366608,
+      "step": 15210
+    },
+    {
+      "epoch": 0.30424,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.010945383707682292,
+      "learning_rate": 0.0001,
+      "loss": 4.3762,
+      "loss/crossentropy": 2.326872229576111,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20695248246192932,
+      "step": 15212
+    },
+    {
+      "epoch": 0.30428,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.012679036458333333,
+      "learning_rate": 0.0001,
+      "loss": 4.1684,
+      "loss/crossentropy": 2.0216678380966187,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19882218539714813,
+      "step": 15214
+    },
+    {
+      "epoch": 0.30432,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0074503580729166664,
+      "learning_rate": 0.0001,
+      "loss": 4.3682,
+      "loss/crossentropy": 2.170402765274048,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22156177461147308,
+      "step": 15216
+    },
+    {
+      "epoch": 0.30436,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.007608795166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1655,
+      "loss/crossentropy": 1.9652912616729736,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20658399909734726,
+      "step": 15218
+    },
+    {
+      "epoch": 0.3044,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0090972900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1234,
+      "loss/crossentropy": 2.4518260955810547,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20796608924865723,
+      "step": 15220
+    },
+    {
+      "epoch": 0.30444,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.009952545166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0179,
+      "loss/crossentropy": 1.9519163370132446,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19269376248121262,
+      "step": 15222
+    },
+    {
+      "epoch": 0.30448,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.010758209228515624,
+      "learning_rate": 0.0001,
+      "loss": 3.9411,
+      "loss/crossentropy": 2.1493905782699585,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19713661074638367,
+      "step": 15224
+    },
+    {
+      "epoch": 0.30452,
+      "grad_norm": 2.5,
+      "grad_norm_var": 0.026878865559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2124,
+      "loss/crossentropy": 2.2856662273406982,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23240232467651367,
+      "step": 15226
+    },
+    {
+      "epoch": 0.30456,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0249176025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1723,
+      "loss/crossentropy": 2.3631181716918945,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22731658816337585,
+      "step": 15228
+    },
+    {
+      "epoch": 0.3046,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.027469635009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0952,
+      "loss/crossentropy": 2.1270273327827454,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20736727863550186,
+      "step": 15230
+    },
+    {
+      "epoch": 0.30464,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.029320271809895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4801,
+      "loss/crossentropy": 2.1448079347610474,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20350141823291779,
+      "step": 15232
+    },
+    {
+      "epoch": 0.30468,
+      "grad_norm": 2.71875,
+      "grad_norm_var": 0.059576161702473956,
+      "learning_rate": 0.0001,
+      "loss": 4.295,
+      "loss/crossentropy": 2.0582846999168396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19898280501365662,
+      "step": 15234
+    },
+    {
+      "epoch": 0.30472,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0597076416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0946,
+      "loss/crossentropy": 2.2282591462135315,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22299205511808395,
+      "step": 15236
+    },
+    {
+      "epoch": 0.30476,
+      "grad_norm": 4.34375,
+      "grad_norm_var": 0.3757484436035156,
+      "learning_rate": 0.0001,
+      "loss": 4.4031,
+      "loss/crossentropy": 1.902436077594757,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19752992689609528,
+      "step": 15238
+    },
+    {
+      "epoch": 0.3048,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.3626177469889323,
+      "learning_rate": 0.0001,
+      "loss": 4.5476,
+      "loss/crossentropy": 2.5741195678710938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25522060692310333,
+      "step": 15240
+    },
+    {
+      "epoch": 0.30484,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.3576942443847656,
+      "learning_rate": 0.0001,
+      "loss": 4.2319,
+      "loss/crossentropy": 2.261489987373352,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2379501387476921,
+      "step": 15242
+    },
+    {
+      "epoch": 0.30488,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.3581451416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1728,
+      "loss/crossentropy": 2.3075523376464844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21830804646015167,
+      "step": 15244
+    },
+    {
+      "epoch": 0.30492,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.36470133463541665,
+      "learning_rate": 0.0001,
+      "loss": 3.9585,
+      "loss/crossentropy": 2.096464157104492,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2073582485318184,
+      "step": 15246
+    },
+    {
+      "epoch": 0.30496,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.41076558430989585,
+      "learning_rate": 0.0001,
+      "loss": 4.0121,
+      "loss/crossentropy": 1.9624161124229431,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19424450397491455,
+      "step": 15248
+    },
+    {
+      "epoch": 0.305,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.38983154296875,
+      "learning_rate": 0.0001,
+      "loss": 4.1351,
+      "loss/crossentropy": 2.0838447213172913,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22090255469083786,
+      "step": 15250
+    },
+    {
+      "epoch": 0.30504,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.379644521077474,
+      "learning_rate": 0.0001,
+      "loss": 4.4244,
+      "loss/crossentropy": 2.349258542060852,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20451825857162476,
+      "step": 15252
+    },
+    {
+      "epoch": 0.30508,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.06938451131184896,
+      "learning_rate": 0.0001,
+      "loss": 4.328,
+      "loss/crossentropy": 2.3996351957321167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24550051987171173,
+      "step": 15254
+    },
+    {
+      "epoch": 0.30512,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.07430191040039062,
+      "learning_rate": 0.0001,
+      "loss": 4.0098,
+      "loss/crossentropy": 2.325650215148926,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2189955934882164,
+      "step": 15256
+    },
+    {
+      "epoch": 0.30516,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.07815348307291667,
+      "learning_rate": 0.0001,
+      "loss": 4.0063,
+      "loss/crossentropy": 1.8568453788757324,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18639766424894333,
+      "step": 15258
+    },
+    {
+      "epoch": 0.3052,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.08263346354166666,
+      "learning_rate": 0.0001,
+      "loss": 3.8955,
+      "loss/crossentropy": 1.8268811106681824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18669788539409637,
+      "step": 15260
+    },
+    {
+      "epoch": 0.30524,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.08479715983072916,
+      "learning_rate": 0.0001,
+      "loss": 4.1124,
+      "loss/crossentropy": 2.0555724501609802,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2101765051484108,
+      "step": 15262
+    },
+    {
+      "epoch": 0.30528,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.018070475260416666,
+      "learning_rate": 0.0001,
+      "loss": 4.1075,
+      "loss/crossentropy": 2.080612599849701,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18955913186073303,
+      "step": 15264
+    },
+    {
+      "epoch": 0.30532,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.0137939453125,
+      "learning_rate": 0.0001,
+      "loss": 4.2678,
+      "loss/crossentropy": 2.352793037891388,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21857291460037231,
+      "step": 15266
+    },
+    {
+      "epoch": 0.30536,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.010434722900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0871,
+      "loss/crossentropy": 2.0258968472480774,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18990560621023178,
+      "step": 15268
+    },
+    {
+      "epoch": 0.3054,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.010871378580729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3126,
+      "loss/crossentropy": 2.228433310985565,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20862557739019394,
+      "step": 15270
+    },
+    {
+      "epoch": 0.30544,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.012330881754557292,
+      "learning_rate": 0.0001,
+      "loss": 3.7731,
+      "loss/crossentropy": 1.9475300312042236,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20644760876893997,
+      "step": 15272
+    },
+    {
+      "epoch": 0.30548,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.012166341145833334,
+      "learning_rate": 0.0001,
+      "loss": 4.1685,
+      "loss/crossentropy": 2.3502081632614136,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22096982598304749,
+      "step": 15274
+    },
+    {
+      "epoch": 0.30552,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.012064615885416666,
+      "learning_rate": 0.0001,
+      "loss": 4.294,
+      "loss/crossentropy": 2.2581464052200317,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21158118546009064,
+      "step": 15276
+    },
+    {
+      "epoch": 0.30556,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.014054361979166667,
+      "learning_rate": 0.0001,
+      "loss": 4.0301,
+      "loss/crossentropy": 1.9583097696304321,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19379162788391113,
+      "step": 15278
+    },
+    {
+      "epoch": 0.3056,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.016527303059895835,
+      "learning_rate": 0.0001,
+      "loss": 4.4464,
+      "loss/crossentropy": 1.988980233669281,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19495020806789398,
+      "step": 15280
+    },
+    {
+      "epoch": 0.30564,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.017350260416666666,
+      "learning_rate": 0.0001,
+      "loss": 4.1733,
+      "loss/crossentropy": 2.1816195249557495,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2196531891822815,
+      "step": 15282
+    },
+    {
+      "epoch": 0.30568,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.015264638264973958,
+      "learning_rate": 0.0001,
+      "loss": 3.7994,
+      "loss/crossentropy": 2.180557370185852,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2106764316558838,
+      "step": 15284
+    },
+    {
+      "epoch": 0.30572,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.011027018229166666,
+      "learning_rate": 0.0001,
+      "loss": 4.0655,
+      "loss/crossentropy": 1.901601493358612,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18202877044677734,
+      "step": 15286
+    },
+    {
+      "epoch": 0.30576,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.019301096598307293,
+      "learning_rate": 0.0001,
+      "loss": 4.1989,
+      "loss/crossentropy": 2.259449601173401,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23166191577911377,
+      "step": 15288
+    },
+    {
+      "epoch": 0.3058,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.020514933268229167,
+      "learning_rate": 0.0001,
+      "loss": 3.9866,
+      "loss/crossentropy": 1.786646544933319,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19156906753778458,
+      "step": 15290
+    },
+    {
+      "epoch": 0.30584,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.02205174763997396,
+      "learning_rate": 0.0001,
+      "loss": 3.9258,
+      "loss/crossentropy": 1.8522255420684814,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1828666776418686,
+      "step": 15292
+    },
+    {
+      "epoch": 0.30588,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.02127685546875,
+      "learning_rate": 0.0001,
+      "loss": 4.2543,
+      "loss/crossentropy": 2.195667266845703,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21043212711811066,
+      "step": 15294
+    },
+    {
+      "epoch": 0.30592,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.020918528238932293,
+      "learning_rate": 0.0001,
+      "loss": 3.8389,
+      "loss/crossentropy": 1.8748087882995605,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20358053594827652,
+      "step": 15296
+    },
+    {
+      "epoch": 0.30596,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.017536417643229166,
+      "learning_rate": 0.0001,
+      "loss": 4.0789,
+      "loss/crossentropy": 2.2658244371414185,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2206317037343979,
+      "step": 15298
+    },
+    {
+      "epoch": 0.306,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.018778483072916668,
+      "learning_rate": 0.0001,
+      "loss": 4.2882,
+      "loss/crossentropy": 1.7914378643035889,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20813927054405212,
+      "step": 15300
+    },
+    {
+      "epoch": 0.30604,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.01895319620768229,
+      "learning_rate": 0.0001,
+      "loss": 4.1428,
+      "loss/crossentropy": 2.046573519706726,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20681843161582947,
+      "step": 15302
+    },
+    {
+      "epoch": 0.30608,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.005960845947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2538,
+      "loss/crossentropy": 2.0068886280059814,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.201521098613739,
+      "step": 15304
+    },
+    {
+      "epoch": 0.30612,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.005509440104166667,
+      "learning_rate": 0.0001,
+      "loss": 4.0672,
+      "loss/crossentropy": 2.2020740509033203,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20413114875555038,
+      "step": 15306
+    },
+    {
+      "epoch": 0.30616,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0057769775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3654,
+      "loss/crossentropy": 2.3031119108200073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20878810435533524,
+      "step": 15308
+    },
+    {
+      "epoch": 0.3062,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.006204986572265625,
+      "learning_rate": 0.0001,
+      "loss": 3.7786,
+      "loss/crossentropy": 1.6445855498313904,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1618219092488289,
+      "step": 15310
+    },
+    {
+      "epoch": 0.30624,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.004572550455729167,
+      "learning_rate": 0.0001,
+      "loss": 4.1378,
+      "loss/crossentropy": 1.995127022266388,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21076080948114395,
+      "step": 15312
+    },
+    {
+      "epoch": 0.30628,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.005411783854166667,
+      "learning_rate": 0.0001,
+      "loss": 3.6278,
+      "loss/crossentropy": 1.8340229392051697,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19418048858642578,
+      "step": 15314
+    },
+    {
+      "epoch": 0.30632,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.007106272379557291,
+      "learning_rate": 0.0001,
+      "loss": 3.9496,
+      "loss/crossentropy": 1.713787317276001,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17608212679624557,
+      "step": 15316
+    },
+    {
+      "epoch": 0.30636,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.006967926025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0223,
+      "loss/crossentropy": 1.7105762362480164,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1760871261358261,
+      "step": 15318
+    },
+    {
+      "epoch": 0.3064,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0071441650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0326,
+      "loss/crossentropy": 2.337005376815796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2222321778535843,
+      "step": 15320
+    },
+    {
+      "epoch": 0.30644,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.008868153889973958,
+      "learning_rate": 0.0001,
+      "loss": 4.3269,
+      "loss/crossentropy": 1.8426868915557861,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1880369558930397,
+      "step": 15322
+    },
+    {
+      "epoch": 0.30648,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.009162394205729167,
+      "learning_rate": 0.0001,
+      "loss": 3.7798,
+      "loss/crossentropy": 2.300028443336487,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22529813647270203,
+      "step": 15324
+    },
+    {
+      "epoch": 0.30652,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.009193674723307291,
+      "learning_rate": 0.0001,
+      "loss": 3.9838,
+      "loss/crossentropy": 2.3558120727539062,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2000514343380928,
+      "step": 15326
+    },
+    {
+      "epoch": 0.30656,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.007879384358723958,
+      "learning_rate": 0.0001,
+      "loss": 3.6921,
+      "loss/crossentropy": 1.681145966053009,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1756032332777977,
+      "step": 15328
+    },
+    {
+      "epoch": 0.3066,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.010298665364583333,
+      "learning_rate": 0.0001,
+      "loss": 4.41,
+      "loss/crossentropy": 2.1566712260246277,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21067694574594498,
+      "step": 15330
+    },
+    {
+      "epoch": 0.30664,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0075347900390625,
+      "learning_rate": 0.0001,
+      "loss": 3.993,
+      "loss/crossentropy": 2.125000774860382,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20768606662750244,
+      "step": 15332
+    },
+    {
+      "epoch": 0.30668,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.007328033447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0805,
+      "loss/crossentropy": 2.168649673461914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20393794029951096,
+      "step": 15334
+    },
+    {
+      "epoch": 0.30672,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.008072662353515624,
+      "learning_rate": 0.0001,
+      "loss": 3.9154,
+      "loss/crossentropy": 1.657529592514038,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19351129233837128,
+      "step": 15336
+    },
+    {
+      "epoch": 0.30676,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.006780751546223958,
+      "learning_rate": 0.0001,
+      "loss": 4.2166,
+      "loss/crossentropy": 2.115979850292206,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20223377645015717,
+      "step": 15338
+    },
+    {
+      "epoch": 0.3068,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.005833943684895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1288,
+      "loss/crossentropy": 1.699661135673523,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17160461097955704,
+      "step": 15340
+    },
+    {
+      "epoch": 0.30684,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.008180491129557292,
+      "learning_rate": 0.0001,
+      "loss": 3.8418,
+      "loss/crossentropy": 2.141602098941803,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19209860265254974,
+      "step": 15342
+    },
+    {
+      "epoch": 0.30688,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.007806142171223958,
+      "learning_rate": 0.0001,
+      "loss": 4.3619,
+      "loss/crossentropy": 2.1502809524536133,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21510163694620132,
+      "step": 15344
+    },
+    {
+      "epoch": 0.30692,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.006078847249348958,
+      "learning_rate": 0.0001,
+      "loss": 4.0684,
+      "loss/crossentropy": 2.1354172825813293,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19959942996501923,
+      "step": 15346
+    },
+    {
+      "epoch": 0.30696,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.006095377604166666,
+      "learning_rate": 0.0001,
+      "loss": 4.1196,
+      "loss/crossentropy": 2.0634223222732544,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2073688805103302,
+      "step": 15348
+    },
+    {
+      "epoch": 0.307,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.008385976155598959,
+      "learning_rate": 0.0001,
+      "loss": 4.1489,
+      "loss/crossentropy": 1.9851733446121216,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2060396894812584,
+      "step": 15350
+    },
+    {
+      "epoch": 0.30704,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.007490793863932292,
+      "learning_rate": 0.0001,
+      "loss": 4.1196,
+      "loss/crossentropy": 2.3790550231933594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23205619305372238,
+      "step": 15352
+    },
+    {
+      "epoch": 0.30708,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.008552042643229167,
+      "learning_rate": 0.0001,
+      "loss": 3.9085,
+      "loss/crossentropy": 1.943642258644104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18238085508346558,
+      "step": 15354
+    },
+    {
+      "epoch": 0.30712,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.008988189697265624,
+      "learning_rate": 0.0001,
+      "loss": 4.112,
+      "loss/crossentropy": 1.9680217504501343,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18599005788564682,
+      "step": 15356
+    },
+    {
+      "epoch": 0.30716,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.0077898661295572914,
+      "learning_rate": 0.0001,
+      "loss": 3.7402,
+      "loss/crossentropy": 1.9464871287345886,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18814007937908173,
+      "step": 15358
+    },
+    {
+      "epoch": 0.3072,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.013179524739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.0231,
+      "loss/crossentropy": 2.0038596987724304,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19203800708055496,
+      "step": 15360
+    },
+    {
+      "epoch": 0.30724,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.013071441650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0843,
+      "loss/crossentropy": 1.93240225315094,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20595912635326385,
+      "step": 15362
+    },
+    {
+      "epoch": 0.30728,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.012938435872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2042,
+      "loss/crossentropy": 2.067903220653534,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21503040939569473,
+      "step": 15364
+    },
+    {
+      "epoch": 0.30732,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.012674713134765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9056,
+      "loss/crossentropy": 2.179181218147278,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20675741881132126,
+      "step": 15366
+    },
+    {
+      "epoch": 0.30736,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.02471923828125,
+      "learning_rate": 0.0001,
+      "loss": 4.4657,
+      "loss/crossentropy": 2.106004774570465,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23683416098356247,
+      "step": 15368
+    },
+    {
+      "epoch": 0.3074,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.023266347249348958,
+      "learning_rate": 0.0001,
+      "loss": 4.135,
+      "loss/crossentropy": 2.2843246459960938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24139763414859772,
+      "step": 15370
+    },
+    {
+      "epoch": 0.30744,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.028484853108723958,
+      "learning_rate": 0.0001,
+      "loss": 4.122,
+      "loss/crossentropy": 2.2482924461364746,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2242894545197487,
+      "step": 15372
+    },
+    {
+      "epoch": 0.30748,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.025321451822916667,
+      "learning_rate": 0.0001,
+      "loss": 4.221,
+      "loss/crossentropy": 2.1573593616485596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22176649421453476,
+      "step": 15374
+    },
+    {
+      "epoch": 0.30752,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.022705078125,
+      "learning_rate": 0.0001,
+      "loss": 4.219,
+      "loss/crossentropy": 1.9243710041046143,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19137389957904816,
+      "step": 15376
+    },
+    {
+      "epoch": 0.30756,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.02211481730143229,
+      "learning_rate": 0.0001,
+      "loss": 4.0903,
+      "loss/crossentropy": 2.004511833190918,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20657775551080704,
+      "step": 15378
+    },
+    {
+      "epoch": 0.3076,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.022956339518229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0174,
+      "loss/crossentropy": 1.8409560918807983,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19139550626277924,
+      "step": 15380
+    },
+    {
+      "epoch": 0.30764,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.019974772135416666,
+      "learning_rate": 0.0001,
+      "loss": 4.2722,
+      "loss/crossentropy": 2.1442220211029053,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23278377205133438,
+      "step": 15382
+    },
+    {
+      "epoch": 0.30768,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.013651275634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2915,
+      "loss/crossentropy": 2.086996078491211,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20602282136678696,
+      "step": 15384
+    },
+    {
+      "epoch": 0.30772,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.014404042561848959,
+      "learning_rate": 0.0001,
+      "loss": 4.4884,
+      "loss/crossentropy": 2.479863405227661,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23815791308879852,
+      "step": 15386
+    },
+    {
+      "epoch": 0.30776,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.009034983317057292,
+      "learning_rate": 0.0001,
+      "loss": 4.0648,
+      "loss/crossentropy": 1.9218324422836304,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22571419924497604,
+      "step": 15388
+    },
+    {
+      "epoch": 0.3078,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009004720052083333,
+      "learning_rate": 0.0001,
+      "loss": 4.0315,
+      "loss/crossentropy": 1.9488168954849243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2004484310746193,
+      "step": 15390
+    },
+    {
+      "epoch": 0.30784,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.006585439046223958,
+      "learning_rate": 0.0001,
+      "loss": 4.3221,
+      "loss/crossentropy": 2.307602286338806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23881366103887558,
+      "step": 15392
+    },
+    {
+      "epoch": 0.30788,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.007283528645833333,
+      "learning_rate": 0.0001,
+      "loss": 4.1025,
+      "loss/crossentropy": 2.1380842328071594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2019854038953781,
+      "step": 15394
+    },
+    {
+      "epoch": 0.30792,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0072509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1245,
+      "loss/crossentropy": 2.0551159977912903,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18669020384550095,
+      "step": 15396
+    },
+    {
+      "epoch": 0.30796,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.009250640869140625,
+      "learning_rate": 0.0001,
+      "loss": 3.959,
+      "loss/crossentropy": 1.9213807582855225,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1843273714184761,
+      "step": 15398
+    },
+    {
+      "epoch": 0.308,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008628082275390626,
+      "learning_rate": 0.0001,
+      "loss": 3.9724,
+      "loss/crossentropy": 2.0028855204582214,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2215728610754013,
+      "step": 15400
+    },
+    {
+      "epoch": 0.30804,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0064776102701822914,
+      "learning_rate": 0.0001,
+      "loss": 4.0287,
+      "loss/crossentropy": 1.957477331161499,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20727651566267014,
+      "step": 15402
+    },
+    {
+      "epoch": 0.30808,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.0080718994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0255,
+      "loss/crossentropy": 2.2241835594177246,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25234874337911606,
+      "step": 15404
+    },
+    {
+      "epoch": 0.30812,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.006852213541666667,
+      "learning_rate": 0.0001,
+      "loss": 4.2677,
+      "loss/crossentropy": 2.108901560306549,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20304366201162338,
+      "step": 15406
+    },
+    {
+      "epoch": 0.30816,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0075032552083333336,
+      "learning_rate": 0.0001,
+      "loss": 4.302,
+      "loss/crossentropy": 2.1005085110664368,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21205321699380875,
+      "step": 15408
+    },
+    {
+      "epoch": 0.3082,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008139801025390626,
+      "learning_rate": 0.0001,
+      "loss": 4.2505,
+      "loss/crossentropy": 2.1565412282943726,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22271832078695297,
+      "step": 15410
+    },
+    {
+      "epoch": 0.30824,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.006241607666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1236,
+      "loss/crossentropy": 1.906798243522644,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21911519020795822,
+      "step": 15412
+    },
+    {
+      "epoch": 0.30828,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.006089019775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1679,
+      "loss/crossentropy": 2.0200547575950623,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19892341643571854,
+      "step": 15414
+    },
+    {
+      "epoch": 0.30832,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.006037394205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.302,
+      "loss/crossentropy": 2.14735209941864,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21503761410713196,
+      "step": 15416
+    },
+    {
+      "epoch": 0.30836,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.007405344645182292,
+      "learning_rate": 0.0001,
+      "loss": 4.2943,
+      "loss/crossentropy": 2.0479459166526794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21268057823181152,
+      "step": 15418
+    },
+    {
+      "epoch": 0.3084,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.006831614176432291,
+      "learning_rate": 0.0001,
+      "loss": 4.1969,
+      "loss/crossentropy": 1.4835808873176575,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16378673166036606,
+      "step": 15420
+    },
+    {
+      "epoch": 0.30844,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.007802073160807292,
+      "learning_rate": 0.0001,
+      "loss": 4.058,
+      "loss/crossentropy": 1.7942256331443787,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18387632071971893,
+      "step": 15422
+    },
+    {
+      "epoch": 0.30848,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.006534576416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1352,
+      "loss/crossentropy": 1.8660341501235962,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20189601182937622,
+      "step": 15424
+    },
+    {
+      "epoch": 0.30852,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.006257120768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0359,
+      "loss/crossentropy": 1.8545112013816833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21422524750232697,
+      "step": 15426
+    },
+    {
+      "epoch": 0.30856,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.0069163004557291664,
+      "learning_rate": 0.0001,
+      "loss": 4.1411,
+      "loss/crossentropy": 2.204701781272888,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21733464300632477,
+      "step": 15428
+    },
+    {
+      "epoch": 0.3086,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.006087239583333333,
+      "learning_rate": 0.0001,
+      "loss": 4.0865,
+      "loss/crossentropy": 1.9295769929885864,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20931877195835114,
+      "step": 15430
+    },
+    {
+      "epoch": 0.30864,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.008534495035807292,
+      "learning_rate": 0.0001,
+      "loss": 3.9165,
+      "loss/crossentropy": 2.0090243220329285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20928232371807098,
+      "step": 15432
+    },
+    {
+      "epoch": 0.30868,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009326171875,
+      "learning_rate": 0.0001,
+      "loss": 4.0517,
+      "loss/crossentropy": 2.1884257793426514,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20887935161590576,
+      "step": 15434
+    },
+    {
+      "epoch": 0.30872,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.009553019205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0101,
+      "loss/crossentropy": 2.023981511592865,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1893448904156685,
+      "step": 15436
+    },
+    {
+      "epoch": 0.30876,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.00909423828125,
+      "learning_rate": 0.0001,
+      "loss": 4.0168,
+      "loss/crossentropy": 2.427197813987732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21649108827114105,
+      "step": 15438
+    },
+    {
+      "epoch": 0.3088,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.015478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1243,
+      "loss/crossentropy": 2.2578816413879395,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21297947317361832,
+      "step": 15440
+    },
+    {
+      "epoch": 0.30884,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.015248362223307292,
+      "learning_rate": 0.0001,
+      "loss": 4.2977,
+      "loss/crossentropy": 1.8909358382225037,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1924509033560753,
+      "step": 15442
+    },
+    {
+      "epoch": 0.30888,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.016228993733723957,
+      "learning_rate": 0.0001,
+      "loss": 4.0142,
+      "loss/crossentropy": 2.344139575958252,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20414643734693527,
+      "step": 15444
+    },
+    {
+      "epoch": 0.30892,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.01624755859375,
+      "learning_rate": 0.0001,
+      "loss": 4.2373,
+      "loss/crossentropy": 2.268660068511963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22332928329706192,
+      "step": 15446
+    },
+    {
+      "epoch": 0.30896,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0148193359375,
+      "learning_rate": 0.0001,
+      "loss": 4.0944,
+      "loss/crossentropy": 2.1276373267173767,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2123691812157631,
+      "step": 15448
+    },
+    {
+      "epoch": 0.309,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.012975819905598958,
+      "learning_rate": 0.0001,
+      "loss": 4.1025,
+      "loss/crossentropy": 2.2471970319747925,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2228669673204422,
+      "step": 15450
+    },
+    {
+      "epoch": 0.30904,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0124908447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2153,
+      "loss/crossentropy": 2.122064530849457,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22981297969818115,
+      "step": 15452
+    },
+    {
+      "epoch": 0.30908,
+      "grad_norm": 1.7734375,
+      "grad_norm_var": 0.014823150634765626,
+      "learning_rate": 0.0001,
+      "loss": 3.815,
+      "loss/crossentropy": 2.016224443912506,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19105292856693268,
+      "step": 15454
+    },
+    {
+      "epoch": 0.30912,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.016193644205729166,
+      "learning_rate": 0.0001,
+      "loss": 4.1518,
+      "loss/crossentropy": 1.9021474719047546,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23519159853458405,
+      "step": 15456
+    },
+    {
+      "epoch": 0.30916,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.016617838541666666,
+      "learning_rate": 0.0001,
+      "loss": 4.0971,
+      "loss/crossentropy": 2.067293703556061,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20525048673152924,
+      "step": 15458
+    },
+    {
+      "epoch": 0.3092,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.021930948893229166,
+      "learning_rate": 0.0001,
+      "loss": 4.3147,
+      "loss/crossentropy": 2.0720648169517517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20188772678375244,
+      "step": 15460
+    },
+    {
+      "epoch": 0.30924,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.030890909830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.212,
+      "loss/crossentropy": 2.1377363204956055,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20943745225667953,
+      "step": 15462
+    },
+    {
+      "epoch": 0.30928,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.028595987955729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0799,
+      "loss/crossentropy": 1.9101733565330505,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18436391651630402,
+      "step": 15464
+    },
+    {
+      "epoch": 0.30932,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.026775868733723958,
+      "learning_rate": 0.0001,
+      "loss": 4.0945,
+      "loss/crossentropy": 2.205660581588745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2431517392396927,
+      "step": 15466
+    },
+    {
+      "epoch": 0.30936,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.02670262654622396,
+      "learning_rate": 0.0001,
+      "loss": 4.502,
+      "loss/crossentropy": 2.3487337827682495,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22333069890737534,
+      "step": 15468
+    },
+    {
+      "epoch": 0.3094,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.021971638997395834,
+      "learning_rate": 0.0001,
+      "loss": 3.9717,
+      "loss/crossentropy": 1.953830897808075,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20369011163711548,
+      "step": 15470
+    },
+    {
+      "epoch": 0.30944,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0156402587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2282,
+      "loss/crossentropy": 2.1175807118415833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23428738862276077,
+      "step": 15472
+    },
+    {
+      "epoch": 0.30948,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.016585032145182293,
+      "learning_rate": 0.0001,
+      "loss": 4.0257,
+      "loss/crossentropy": 2.114426612854004,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20113752037286758,
+      "step": 15474
+    },
+    {
+      "epoch": 0.30952,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.01676203409830729,
+      "learning_rate": 0.0001,
+      "loss": 3.8239,
+      "loss/crossentropy": 2.121155083179474,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23428727686405182,
+      "step": 15476
+    },
+    {
+      "epoch": 0.30956,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.007061513264973959,
+      "learning_rate": 0.0001,
+      "loss": 3.9982,
+      "loss/crossentropy": 2.1271342635154724,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1993299275636673,
+      "step": 15478
+    },
+    {
+      "epoch": 0.3096,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.007502237955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0211,
+      "loss/crossentropy": 2.132619023323059,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20428313314914703,
+      "step": 15480
+    },
+    {
+      "epoch": 0.30964,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.013720703125,
+      "learning_rate": 0.0001,
+      "loss": 4.2644,
+      "loss/crossentropy": 1.8355163931846619,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1969365030527115,
+      "step": 15482
+    },
+    {
+      "epoch": 0.30968,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.016695149739583335,
+      "learning_rate": 0.0001,
+      "loss": 4.0001,
+      "loss/crossentropy": 1.6765839457511902,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19648748636245728,
+      "step": 15484
+    },
+    {
+      "epoch": 0.30972,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0168609619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0806,
+      "loss/crossentropy": 1.7628436088562012,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1850176900625229,
+      "step": 15486
+    },
+    {
+      "epoch": 0.30976,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.017014312744140624,
+      "learning_rate": 0.0001,
+      "loss": 4.5902,
+      "loss/crossentropy": 2.407894253730774,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2243446335196495,
+      "step": 15488
+    },
+    {
+      "epoch": 0.3098,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.017409006754557293,
+      "learning_rate": 0.0001,
+      "loss": 4.0274,
+      "loss/crossentropy": 2.0359743237495422,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2140478640794754,
+      "step": 15490
+    },
+    {
+      "epoch": 0.30984,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.012111155192057292,
+      "learning_rate": 0.0001,
+      "loss": 4.086,
+      "loss/crossentropy": 2.1705892086029053,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2108132392168045,
+      "step": 15492
+    },
+    {
+      "epoch": 0.30988,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.011126454671223958,
+      "learning_rate": 0.0001,
+      "loss": 3.9782,
+      "loss/crossentropy": 1.836803376674652,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19436977803707123,
+      "step": 15494
+    },
+    {
+      "epoch": 0.30992,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010432688395182292,
+      "learning_rate": 0.0001,
+      "loss": 4.1018,
+      "loss/crossentropy": 2.1445683240890503,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21950577944517136,
+      "step": 15496
+    },
+    {
+      "epoch": 0.30996,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.007299550374348958,
+      "learning_rate": 0.0001,
+      "loss": 4.1382,
+      "loss/crossentropy": 2.258531093597412,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.212137870490551,
+      "step": 15498
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.004052480061848958,
+      "learning_rate": 0.0001,
+      "loss": 4.2473,
+      "loss/crossentropy": 1.9310460686683655,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18316251039505005,
+      "step": 15500
+    },
+    {
+      "epoch": 0.31004,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.004788970947265625,
+      "learning_rate": 0.0001,
+      "loss": 3.8656,
+      "loss/crossentropy": 1.867807149887085,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18478462100028992,
+      "step": 15502
+    },
+    {
+      "epoch": 0.31008,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.008042144775390624,
+      "learning_rate": 0.0001,
+      "loss": 4.4691,
+      "loss/crossentropy": 2.0724986791610718,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20986932516098022,
+      "step": 15504
+    },
+    {
+      "epoch": 0.31012,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.010138956705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.5188,
+      "loss/crossentropy": 2.1599953174591064,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2250317633152008,
+      "step": 15506
+    },
+    {
+      "epoch": 0.31016,
+      "grad_norm": 1.8125,
+      "grad_norm_var": 0.015608469645182291,
+      "learning_rate": 0.0001,
+      "loss": 3.7842,
+      "loss/crossentropy": 1.975223183631897,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2104134038090706,
+      "step": 15508
+    },
+    {
+      "epoch": 0.3102,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.019539388020833333,
+      "learning_rate": 0.0001,
+      "loss": 4.0028,
+      "loss/crossentropy": 2.133773148059845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20153696089982986,
+      "step": 15510
+    },
+    {
+      "epoch": 0.31024,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.02014948527018229,
+      "learning_rate": 0.0001,
+      "loss": 3.9948,
+      "loss/crossentropy": 1.994078278541565,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20573429763317108,
+      "step": 15512
+    },
+    {
+      "epoch": 0.31028,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.019280751546223957,
+      "learning_rate": 0.0001,
+      "loss": 3.8209,
+      "loss/crossentropy": 1.7885381579399109,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1889629364013672,
+      "step": 15514
+    },
+    {
+      "epoch": 0.31032,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.019191233317057292,
+      "learning_rate": 0.0001,
+      "loss": 4.2765,
+      "loss/crossentropy": 2.495113968849182,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22110097110271454,
+      "step": 15516
+    },
+    {
+      "epoch": 0.31036,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.019160715738932292,
+      "learning_rate": 0.0001,
+      "loss": 4.0465,
+      "loss/crossentropy": 2.0757648944854736,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20948036015033722,
+      "step": 15518
+    },
+    {
+      "epoch": 0.3104,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.015793609619140624,
+      "learning_rate": 0.0001,
+      "loss": 4.1369,
+      "loss/crossentropy": 2.151356875896454,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22598887234926224,
+      "step": 15520
+    },
+    {
+      "epoch": 0.31044,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.010933176676432291,
+      "learning_rate": 0.0001,
+      "loss": 4.2193,
+      "loss/crossentropy": 2.1210484504699707,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23365220427513123,
+      "step": 15522
+    },
+    {
+      "epoch": 0.31048,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.008414459228515626,
+      "learning_rate": 0.0001,
+      "loss": 4.0584,
+      "loss/crossentropy": 2.3766754865646362,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22161870449781418,
+      "step": 15524
+    },
+    {
+      "epoch": 0.31052,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0043853759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0968,
+      "loss/crossentropy": 2.33975088596344,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21598009765148163,
+      "step": 15526
+    },
+    {
+      "epoch": 0.31056,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.006468709309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.4643,
+      "loss/crossentropy": 2.3024542331695557,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20656803250312805,
+      "step": 15528
+    },
+    {
+      "epoch": 0.3106,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.007666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1137,
+      "loss/crossentropy": 2.009307861328125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19054396450519562,
+      "step": 15530
+    },
+    {
+      "epoch": 0.31064,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.008902994791666667,
+      "learning_rate": 0.0001,
+      "loss": 3.9413,
+      "loss/crossentropy": 1.8119140267372131,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19199557602405548,
+      "step": 15532
+    },
+    {
+      "epoch": 0.31068,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.009614817301432292,
+      "learning_rate": 0.0001,
+      "loss": 3.8753,
+      "loss/crossentropy": 1.9586234092712402,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18981865793466568,
+      "step": 15534
+    },
+    {
+      "epoch": 0.31072,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.010798899332682292,
+      "learning_rate": 0.0001,
+      "loss": 4.4319,
+      "loss/crossentropy": 2.1208658814430237,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20590776205062866,
+      "step": 15536
+    },
+    {
+      "epoch": 0.31076,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.012336222330729167,
+      "learning_rate": 0.0001,
+      "loss": 3.7806,
+      "loss/crossentropy": 1.539306402206421,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1615876704454422,
+      "step": 15538
+    },
+    {
+      "epoch": 0.3108,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.013768513997395834,
+      "learning_rate": 0.0001,
+      "loss": 4.0686,
+      "loss/crossentropy": 2.1376689672470093,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21377786993980408,
+      "step": 15540
+    },
+    {
+      "epoch": 0.31084,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.016249338785807293,
+      "learning_rate": 0.0001,
+      "loss": 4.368,
+      "loss/crossentropy": 1.954556941986084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19270780682563782,
+      "step": 15542
+    },
+    {
+      "epoch": 0.31088,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.013575998942057292,
+      "learning_rate": 0.0001,
+      "loss": 4.1313,
+      "loss/crossentropy": 2.0225048661231995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1937834918498993,
+      "step": 15544
+    },
+    {
+      "epoch": 0.31092,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.011310831705729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0919,
+      "loss/crossentropy": 2.149065852165222,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21962979435920715,
+      "step": 15546
+    },
+    {
+      "epoch": 0.31096,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.013073476155598958,
+      "learning_rate": 0.0001,
+      "loss": 4.0192,
+      "loss/crossentropy": 1.9546958804130554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20022059231996536,
+      "step": 15548
+    },
+    {
+      "epoch": 0.311,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.013728841145833334,
+      "learning_rate": 0.0001,
+      "loss": 4.2815,
+      "loss/crossentropy": 2.3048110008239746,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2232206016778946,
+      "step": 15550
+    },
+    {
+      "epoch": 0.31104,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.014525349934895833,
+      "learning_rate": 0.0001,
+      "loss": 3.9263,
+      "loss/crossentropy": 2.0329134464263916,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20285522937774658,
+      "step": 15552
+    },
+    {
+      "epoch": 0.31108,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.013138580322265624,
+      "learning_rate": 0.0001,
+      "loss": 4.1654,
+      "loss/crossentropy": 2.0910086631774902,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20308854430913925,
+      "step": 15554
+    },
+    {
+      "epoch": 0.31112,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.012361399332682292,
+      "learning_rate": 0.0001,
+      "loss": 4.1922,
+      "loss/crossentropy": 2.3473485708236694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2138630449771881,
+      "step": 15556
+    },
+    {
+      "epoch": 0.31116,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0098297119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3456,
+      "loss/crossentropy": 2.4176318645477295,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23624707758426666,
+      "step": 15558
+    },
+    {
+      "epoch": 0.3112,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.009943644205729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0898,
+      "loss/crossentropy": 2.190505266189575,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21235676854848862,
+      "step": 15560
+    },
+    {
+      "epoch": 0.31124,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.1935198465983073,
+      "learning_rate": 0.0001,
+      "loss": 4.3688,
+      "loss/crossentropy": 2.1774531602859497,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2568993717432022,
+      "step": 15562
+    },
+    {
+      "epoch": 0.31128,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.1934832255045573,
+      "learning_rate": 0.0001,
+      "loss": 4.004,
+      "loss/crossentropy": 2.0320950150489807,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21587081998586655,
+      "step": 15564
+    },
+    {
+      "epoch": 0.31132,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.19068781534830728,
+      "learning_rate": 0.0001,
+      "loss": 4.3037,
+      "loss/crossentropy": 2.0477761030197144,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21909283846616745,
+      "step": 15566
+    },
+    {
+      "epoch": 0.31136,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.18454996744791666,
+      "learning_rate": 0.0001,
+      "loss": 4.2288,
+      "loss/crossentropy": 2.3907222747802734,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21334309875965118,
+      "step": 15568
+    },
+    {
+      "epoch": 0.3114,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.1816973368326823,
+      "learning_rate": 0.0001,
+      "loss": 4.28,
+      "loss/crossentropy": 2.2539732456207275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21502045542001724,
+      "step": 15570
+    },
+    {
+      "epoch": 0.31144,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.17766520182291667,
+      "learning_rate": 0.0001,
+      "loss": 3.9675,
+      "loss/crossentropy": 2.036223292350769,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1959664672613144,
+      "step": 15572
+    },
+    {
+      "epoch": 0.31148,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.17722880045572917,
+      "learning_rate": 0.0001,
+      "loss": 4.0387,
+      "loss/crossentropy": 1.7659193873405457,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19535844773054123,
+      "step": 15574
+    },
+    {
+      "epoch": 0.31152,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.17946751912434897,
+      "learning_rate": 0.0001,
+      "loss": 3.8892,
+      "loss/crossentropy": 2.0053369402885437,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24516403675079346,
+      "step": 15576
+    },
+    {
+      "epoch": 0.31156,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0060384114583333336,
+      "learning_rate": 0.0001,
+      "loss": 3.9851,
+      "loss/crossentropy": 2.1804131269454956,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2349890023469925,
+      "step": 15578
+    },
+    {
+      "epoch": 0.3116,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.008552042643229167,
+      "learning_rate": 0.0001,
+      "loss": 3.9369,
+      "loss/crossentropy": 2.314085602760315,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21073052287101746,
+      "step": 15580
+    },
+    {
+      "epoch": 0.31164,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.008973948160807292,
+      "learning_rate": 0.0001,
+      "loss": 3.7951,
+      "loss/crossentropy": 2.1868577003479004,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21520060300827026,
+      "step": 15582
+    },
+    {
+      "epoch": 0.31168,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.007957967122395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1618,
+      "loss/crossentropy": 1.9785407185554504,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18188606202602386,
+      "step": 15584
+    },
+    {
+      "epoch": 0.31172,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009598795572916667,
+      "learning_rate": 0.0001,
+      "loss": 3.9173,
+      "loss/crossentropy": 1.9320110082626343,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2022753208875656,
+      "step": 15586
+    },
+    {
+      "epoch": 0.31176,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.015653483072916665,
+      "learning_rate": 0.0001,
+      "loss": 4.0995,
+      "loss/crossentropy": 1.7925443649291992,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17203447967767715,
+      "step": 15588
+    },
+    {
+      "epoch": 0.3118,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.016556803385416666,
+      "learning_rate": 0.0001,
+      "loss": 3.8876,
+      "loss/crossentropy": 2.048543393611908,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18696243315935135,
+      "step": 15590
+    },
+    {
+      "epoch": 0.31184,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.01658935546875,
+      "learning_rate": 0.0001,
+      "loss": 4.125,
+      "loss/crossentropy": 2.376060724258423,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20351988822221756,
+      "step": 15592
+    },
+    {
+      "epoch": 0.31188,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.0164947509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1778,
+      "loss/crossentropy": 1.5962707996368408,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1538907140493393,
+      "step": 15594
+    },
+    {
+      "epoch": 0.31192,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.016281890869140624,
+      "learning_rate": 0.0001,
+      "loss": 3.7529,
+      "loss/crossentropy": 1.7845428586006165,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18107154965400696,
+      "step": 15596
+    },
+    {
+      "epoch": 0.31196,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.01536865234375,
+      "learning_rate": 0.0001,
+      "loss": 4.0385,
+      "loss/crossentropy": 1.8557100296020508,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2010527327656746,
+      "step": 15598
+    },
+    {
+      "epoch": 0.312,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.015372721354166667,
+      "learning_rate": 0.0001,
+      "loss": 3.9735,
+      "loss/crossentropy": 1.8659257888793945,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19198190420866013,
+      "step": 15600
+    },
+    {
+      "epoch": 0.31204,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.014078521728515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9842,
+      "loss/crossentropy": 2.0340747833251953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19963373243808746,
+      "step": 15602
+    },
+    {
+      "epoch": 0.31208,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.00931396484375,
+      "learning_rate": 0.0001,
+      "loss": 4.1192,
+      "loss/crossentropy": 2.0676557421684265,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20130064338445663,
+      "step": 15604
+    },
+    {
+      "epoch": 0.31212,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.008397420247395834,
+      "learning_rate": 0.0001,
+      "loss": 3.8069,
+      "loss/crossentropy": 2.038145124912262,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2034953236579895,
+      "step": 15606
+    },
+    {
+      "epoch": 0.31216,
+      "grad_norm": 2.65625,
+      "grad_norm_var": 0.03449605305989583,
+      "learning_rate": 0.0001,
+      "loss": 4.3615,
+      "loss/crossentropy": 2.3371706008911133,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21727421879768372,
+      "step": 15608
+    },
+    {
+      "epoch": 0.3122,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.03327611287434896,
+      "learning_rate": 0.0001,
+      "loss": 4.1988,
+      "loss/crossentropy": 2.3314318656921387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2064831778407097,
+      "step": 15610
+    },
+    {
+      "epoch": 0.31224,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.03206558227539062,
+      "learning_rate": 0.0001,
+      "loss": 3.9634,
+      "loss/crossentropy": 1.8074566721916199,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1755329668521881,
+      "step": 15612
+    },
+    {
+      "epoch": 0.31228,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.032138824462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1849,
+      "loss/crossentropy": 2.1219228506088257,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23183195292949677,
+      "step": 15614
+    },
+    {
+      "epoch": 0.31232,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.031434885660807294,
+      "learning_rate": 0.0001,
+      "loss": 4.1418,
+      "loss/crossentropy": 2.3573983907699585,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20713336020708084,
+      "step": 15616
+    },
+    {
+      "epoch": 0.31236,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.033933258056640624,
+      "learning_rate": 0.0001,
+      "loss": 4.048,
+      "loss/crossentropy": 2.0748316049575806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19384868443012238,
+      "step": 15618
+    },
+    {
+      "epoch": 0.3124,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.033882395426432295,
+      "learning_rate": 0.0001,
+      "loss": 4.1834,
+      "loss/crossentropy": 1.7933568358421326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19711802154779434,
+      "step": 15620
+    },
+    {
+      "epoch": 0.31244,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.03115819295247396,
+      "learning_rate": 0.0001,
+      "loss": 4.2023,
+      "loss/crossentropy": 1.8922778367996216,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19847699999809265,
+      "step": 15622
+    },
+    {
+      "epoch": 0.31248,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0048004150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2326,
+      "loss/crossentropy": 2.3432679176330566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22804179042577744,
+      "step": 15624
+    },
+    {
+      "epoch": 0.31252,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.007469685872395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1215,
+      "loss/crossentropy": 2.340356707572937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22454006224870682,
+      "step": 15626
+    },
+    {
+      "epoch": 0.31256,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.010737864176432292,
+      "learning_rate": 0.0001,
+      "loss": 4.0874,
+      "loss/crossentropy": 2.037777841091156,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20374249666929245,
+      "step": 15628
+    },
+    {
+      "epoch": 0.3126,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010562896728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3193,
+      "loss/crossentropy": 2.2576274275779724,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19910124689340591,
+      "step": 15630
+    },
+    {
+      "epoch": 0.31264,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.011302693684895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1561,
+      "loss/crossentropy": 2.1354891061782837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20646820217370987,
+      "step": 15632
+    },
+    {
+      "epoch": 0.31268,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.011136627197265625,
+      "learning_rate": 0.0001,
+      "loss": 3.906,
+      "loss/crossentropy": 1.8888733386993408,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20730097591876984,
+      "step": 15634
+    },
+    {
+      "epoch": 0.31272,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.012474568684895833,
+      "learning_rate": 0.0001,
+      "loss": 3.9522,
+      "loss/crossentropy": 2.1037773489952087,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20308485627174377,
+      "step": 15636
+    },
+    {
+      "epoch": 0.31276,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.017085774739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.244,
+      "loss/crossentropy": 1.996969223022461,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2080031782388687,
+      "step": 15638
+    },
+    {
+      "epoch": 0.3128,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.0469146728515625,
+      "learning_rate": 0.0001,
+      "loss": 3.915,
+      "loss/crossentropy": 2.0528018474578857,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2248053327202797,
+      "step": 15640
+    },
+    {
+      "epoch": 0.31284,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.044864654541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1718,
+      "loss/crossentropy": 2.270516335964203,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19578035175800323,
+      "step": 15642
+    },
+    {
+      "epoch": 0.31288,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0400299072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1463,
+      "loss/crossentropy": 2.146915912628174,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21282948553562164,
+      "step": 15644
+    },
+    {
+      "epoch": 0.31292,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.041562652587890624,
+      "learning_rate": 0.0001,
+      "loss": 4.3275,
+      "loss/crossentropy": 2.058235287666321,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20604285597801208,
+      "step": 15646
+    },
+    {
+      "epoch": 0.31296,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0414215087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4304,
+      "loss/crossentropy": 2.3217705488204956,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24202678352594376,
+      "step": 15648
+    },
+    {
+      "epoch": 0.313,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.041257476806640624,
+      "learning_rate": 0.0001,
+      "loss": 4.1372,
+      "loss/crossentropy": 2.108501672744751,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20358111709356308,
+      "step": 15650
+    },
+    {
+      "epoch": 0.31304,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.041025543212890626,
+      "learning_rate": 0.0001,
+      "loss": 4.1224,
+      "loss/crossentropy": 2.0875802636146545,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20186328142881393,
+      "step": 15652
+    },
+    {
+      "epoch": 0.31308,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.036321767171223956,
+      "learning_rate": 0.0001,
+      "loss": 4.1994,
+      "loss/crossentropy": 2.0159415006637573,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2053827941417694,
+      "step": 15654
+    },
+    {
+      "epoch": 0.31312,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008402252197265625,
+      "learning_rate": 0.0001,
+      "loss": 3.9711,
+      "loss/crossentropy": 1.8664127588272095,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18666206300258636,
+      "step": 15656
+    },
+    {
+      "epoch": 0.31316,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.008548736572265625,
+      "learning_rate": 0.0001,
+      "loss": 3.9399,
+      "loss/crossentropy": 1.9351357221603394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.189142145216465,
+      "step": 15658
+    },
+    {
+      "epoch": 0.3132,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.010872141520182291,
+      "learning_rate": 0.0001,
+      "loss": 4.3288,
+      "loss/crossentropy": 1.9937176704406738,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18212904781103134,
+      "step": 15660
+    },
+    {
+      "epoch": 0.31324,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009801991780598958,
+      "learning_rate": 0.0001,
+      "loss": 4.1238,
+      "loss/crossentropy": 2.0626447796821594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20908795297145844,
+      "step": 15662
+    },
+    {
+      "epoch": 0.31328,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.008506011962890626,
+      "learning_rate": 0.0001,
+      "loss": 4.0436,
+      "loss/crossentropy": 1.8491488695144653,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18241358548402786,
+      "step": 15664
+    },
+    {
+      "epoch": 0.31332,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.0084625244140625,
+      "learning_rate": 0.0001,
+      "loss": 3.9885,
+      "loss/crossentropy": 2.2744827270507812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2295297086238861,
+      "step": 15666
+    },
+    {
+      "epoch": 0.31336,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.0078521728515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9213,
+      "loss/crossentropy": 2.30733585357666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23149366676807404,
+      "step": 15668
+    },
+    {
+      "epoch": 0.3134,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.00892333984375,
+      "learning_rate": 0.0001,
+      "loss": 4.1357,
+      "loss/crossentropy": 2.038204550743103,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19608831405639648,
+      "step": 15670
+    },
+    {
+      "epoch": 0.31344,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.008676910400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1235,
+      "loss/crossentropy": 2.076035261154175,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21801196038722992,
+      "step": 15672
+    },
+    {
+      "epoch": 0.31348,
+      "grad_norm": 1.7734375,
+      "grad_norm_var": 0.011310831705729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0923,
+      "loss/crossentropy": 1.9234278202056885,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18380460888147354,
+      "step": 15674
+    },
+    {
+      "epoch": 0.31352,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.00931396484375,
+      "learning_rate": 0.0001,
+      "loss": 4.1002,
+      "loss/crossentropy": 1.9312421679496765,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18474044650793076,
+      "step": 15676
+    },
+    {
+      "epoch": 0.31356,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009716542561848958,
+      "learning_rate": 0.0001,
+      "loss": 4.079,
+      "loss/crossentropy": 2.014132261276245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2045312151312828,
+      "step": 15678
+    },
+    {
+      "epoch": 0.3136,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0099365234375,
+      "learning_rate": 0.0001,
+      "loss": 4.2325,
+      "loss/crossentropy": 1.925381362438202,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18306680023670197,
+      "step": 15680
+    },
+    {
+      "epoch": 0.31364,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.009346516927083333,
+      "learning_rate": 0.0001,
+      "loss": 4.0766,
+      "loss/crossentropy": 1.932526171207428,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20741897821426392,
+      "step": 15682
+    },
+    {
+      "epoch": 0.31368,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008963775634765626,
+      "learning_rate": 0.0001,
+      "loss": 4.2234,
+      "loss/crossentropy": 2.1144750714302063,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19398797303438187,
+      "step": 15684
+    },
+    {
+      "epoch": 0.31372,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.0076983133951822914,
+      "learning_rate": 0.0001,
+      "loss": 3.8411,
+      "loss/crossentropy": 1.8115296363830566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1869400143623352,
+      "step": 15686
+    },
+    {
+      "epoch": 0.31376,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.014686838785807291,
+      "learning_rate": 0.0001,
+      "loss": 4.0935,
+      "loss/crossentropy": 2.0550562739372253,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2093418911099434,
+      "step": 15688
+    },
+    {
+      "epoch": 0.3138,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.013541412353515626,
+      "learning_rate": 0.0001,
+      "loss": 3.8549,
+      "loss/crossentropy": 2.144737482070923,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20025025308132172,
+      "step": 15690
+    },
+    {
+      "epoch": 0.31384,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.012629191080729166,
+      "learning_rate": 0.0001,
+      "loss": 3.9911,
+      "loss/crossentropy": 2.2495052814483643,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21178698539733887,
+      "step": 15692
+    },
+    {
+      "epoch": 0.31388,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.012740071614583333,
+      "learning_rate": 0.0001,
+      "loss": 4.2208,
+      "loss/crossentropy": 2.1858155727386475,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20891498774290085,
+      "step": 15694
+    },
+    {
+      "epoch": 0.31392,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.0131744384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1841,
+      "loss/crossentropy": 2.1480127573013306,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2149479240179062,
+      "step": 15696
+    },
+    {
+      "epoch": 0.31396,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.013329060872395833,
+      "learning_rate": 0.0001,
+      "loss": 3.9218,
+      "loss/crossentropy": 2.1585733294487,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20317788422107697,
+      "step": 15698
+    },
+    {
+      "epoch": 0.314,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.2103179931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.06,
+      "loss/crossentropy": 2.1281010508537292,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2032032608985901,
+      "step": 15700
+    },
+    {
+      "epoch": 0.31404,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.2096588134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0189,
+      "loss/crossentropy": 2.160776972770691,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20999448001384735,
+      "step": 15702
+    },
+    {
+      "epoch": 0.31408,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.20640869140625,
+      "learning_rate": 0.0001,
+      "loss": 3.961,
+      "loss/crossentropy": 2.1508986949920654,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21675271540880203,
+      "step": 15704
+    },
+    {
+      "epoch": 0.31412,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.2011871337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1573,
+      "loss/crossentropy": 2.083084225654602,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21110595017671585,
+      "step": 15706
+    },
+    {
+      "epoch": 0.31416,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.20275472005208334,
+      "learning_rate": 0.0001,
+      "loss": 3.9785,
+      "loss/crossentropy": 2.1268805265426636,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20728368312120438,
+      "step": 15708
+    },
+    {
+      "epoch": 0.3142,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.20265706380208334,
+      "learning_rate": 0.0001,
+      "loss": 4.4291,
+      "loss/crossentropy": 2.156775116920471,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22175253927707672,
+      "step": 15710
+    },
+    {
+      "epoch": 0.31424,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.20224583943684896,
+      "learning_rate": 0.0001,
+      "loss": 4.1537,
+      "loss/crossentropy": 2.2541415691375732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22773633152246475,
+      "step": 15712
+    },
+    {
+      "epoch": 0.31428,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.20350723266601561,
+      "learning_rate": 0.0001,
+      "loss": 3.9166,
+      "loss/crossentropy": 1.6463716626167297,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18172992020845413,
+      "step": 15714
+    },
+    {
+      "epoch": 0.31432,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011669921875,
+      "learning_rate": 0.0001,
+      "loss": 4.0144,
+      "loss/crossentropy": 1.9325580596923828,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19147639721632004,
+      "step": 15716
+    },
+    {
+      "epoch": 0.31436,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.014115142822265624,
+      "learning_rate": 0.0001,
+      "loss": 3.9952,
+      "loss/crossentropy": 2.4016847610473633,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2058410421013832,
+      "step": 15718
+    },
+    {
+      "epoch": 0.3144,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.014338175455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.1847,
+      "loss/crossentropy": 2.3456791639328003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22646376490592957,
+      "step": 15720
+    },
+    {
+      "epoch": 0.31444,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.013785807291666667,
+      "learning_rate": 0.0001,
+      "loss": 4.0551,
+      "loss/crossentropy": 1.6703909635543823,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1818019300699234,
+      "step": 15722
+    },
+    {
+      "epoch": 0.31448,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.011065419514973958,
+      "learning_rate": 0.0001,
+      "loss": 4.0561,
+      "loss/crossentropy": 2.0004162192344666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2033773437142372,
+      "step": 15724
+    },
+    {
+      "epoch": 0.31452,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.009905751546223958,
+      "learning_rate": 0.0001,
+      "loss": 4.1799,
+      "loss/crossentropy": 2.018653154373169,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20320426672697067,
+      "step": 15726
+    },
+    {
+      "epoch": 0.31456,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.009592437744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0991,
+      "loss/crossentropy": 2.0881033539772034,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2126169577240944,
+      "step": 15728
+    },
+    {
+      "epoch": 0.3146,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0067942301432291664,
+      "learning_rate": 0.0001,
+      "loss": 4.2159,
+      "loss/crossentropy": 2.1103954911231995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2163369059562683,
+      "step": 15730
+    },
+    {
+      "epoch": 0.31464,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007868448893229166,
+      "learning_rate": 0.0001,
+      "loss": 4.142,
+      "loss/crossentropy": 1.9757474660873413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21398526430130005,
+      "step": 15732
+    },
+    {
+      "epoch": 0.31468,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.006304677327473958,
+      "learning_rate": 0.0001,
+      "loss": 3.8965,
+      "loss/crossentropy": 2.114749312400818,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21620750427246094,
+      "step": 15734
+    },
+    {
+      "epoch": 0.31472,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.007163238525390625,
+      "learning_rate": 0.0001,
+      "loss": 3.9688,
+      "loss/crossentropy": 1.9022989869117737,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1895657405257225,
+      "step": 15736
+    },
+    {
+      "epoch": 0.31476,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.013059234619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2254,
+      "loss/crossentropy": 2.0457635521888733,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21281428635120392,
+      "step": 15738
+    },
+    {
+      "epoch": 0.3148,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.011954752604166667,
+      "learning_rate": 0.0001,
+      "loss": 3.8985,
+      "loss/crossentropy": 1.9585834741592407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2003040909767151,
+      "step": 15740
+    },
+    {
+      "epoch": 0.31484,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.014745076497395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1682,
+      "loss/crossentropy": 2.06454074382782,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22020584344863892,
+      "step": 15742
+    },
+    {
+      "epoch": 0.31488,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.015665690104166668,
+      "learning_rate": 0.0001,
+      "loss": 4.2384,
+      "loss/crossentropy": 2.2651820182800293,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22998705506324768,
+      "step": 15744
+    },
+    {
+      "epoch": 0.31492,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0172607421875,
+      "learning_rate": 0.0001,
+      "loss": 4.258,
+      "loss/crossentropy": 2.269286036491394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21012098342180252,
+      "step": 15746
+    },
+    {
+      "epoch": 0.31496,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.017891438802083333,
+      "learning_rate": 0.0001,
+      "loss": 4.0604,
+      "loss/crossentropy": 1.6778026223182678,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17330823838710785,
+      "step": 15748
+    },
+    {
+      "epoch": 0.315,
+      "grad_norm": 1.78125,
+      "grad_norm_var": 0.020775349934895833,
+      "learning_rate": 0.0001,
+      "loss": 3.9785,
+      "loss/crossentropy": 1.8929405808448792,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18170948326587677,
+      "step": 15750
+    },
+    {
+      "epoch": 0.31504,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.01980768839518229,
+      "learning_rate": 0.0001,
+      "loss": 4.0874,
+      "loss/crossentropy": 1.8814340233802795,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2097821682691574,
+      "step": 15752
+    },
+    {
+      "epoch": 0.31508,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.014465077718098959,
+      "learning_rate": 0.0001,
+      "loss": 3.8622,
+      "loss/crossentropy": 1.7101264595985413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1834770292043686,
+      "step": 15754
+    },
+    {
+      "epoch": 0.31512,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.01607233683268229,
+      "learning_rate": 0.0001,
+      "loss": 3.8498,
+      "loss/crossentropy": 2.0287395119667053,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19603880494832993,
+      "step": 15756
+    },
+    {
+      "epoch": 0.31516,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.010467274983723959,
+      "learning_rate": 0.0001,
+      "loss": 3.7485,
+      "loss/crossentropy": 2.0119568705558777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.212313212454319,
+      "step": 15758
+    },
+    {
+      "epoch": 0.3152,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.011441802978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.5141,
+      "loss/crossentropy": 1.8937278985977173,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1774902269244194,
+      "step": 15760
+    },
+    {
+      "epoch": 0.31524,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.009104156494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3562,
+      "loss/crossentropy": 2.557780146598816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22111114859580994,
+      "step": 15762
+    },
+    {
+      "epoch": 0.31528,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.011262003580729167,
+      "learning_rate": 0.0001,
+      "loss": 4.1646,
+      "loss/crossentropy": 2.1467400789260864,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2134222611784935,
+      "step": 15764
+    },
+    {
+      "epoch": 0.31532,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.00814208984375,
+      "learning_rate": 0.0001,
+      "loss": 3.9631,
+      "loss/crossentropy": 1.9758468270301819,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1853625252842903,
+      "step": 15766
+    },
+    {
+      "epoch": 0.31536,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.009699503580729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2144,
+      "loss/crossentropy": 1.8620794415473938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18791437149047852,
+      "step": 15768
+    },
+    {
+      "epoch": 0.3154,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008893839518229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2066,
+      "loss/crossentropy": 2.187020480632782,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2102060467004776,
+      "step": 15770
+    },
+    {
+      "epoch": 0.31544,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.006522369384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1828,
+      "loss/crossentropy": 2.0049954652786255,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19090547412633896,
+      "step": 15772
+    },
+    {
+      "epoch": 0.31548,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.010357411702473958,
+      "learning_rate": 0.0001,
+      "loss": 4.195,
+      "loss/crossentropy": 2.0030736327171326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20134840160608292,
+      "step": 15774
+    },
+    {
+      "epoch": 0.31552,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.010687001546223958,
+      "learning_rate": 0.0001,
+      "loss": 4.3092,
+      "loss/crossentropy": 1.8941256999969482,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20826192945241928,
+      "step": 15776
+    },
+    {
+      "epoch": 0.31556,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.015596516927083333,
+      "learning_rate": 0.0001,
+      "loss": 4.0767,
+      "loss/crossentropy": 2.0645129680633545,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19137012213468552,
+      "step": 15778
+    },
+    {
+      "epoch": 0.3156,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.014240519205729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0349,
+      "loss/crossentropy": 2.198352813720703,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21606199443340302,
+      "step": 15780
+    },
+    {
+      "epoch": 0.31564,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.014020792643229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1544,
+      "loss/crossentropy": 1.9041627049446106,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19438787549734116,
+      "step": 15782
+    },
+    {
+      "epoch": 0.31568,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.027741495768229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2615,
+      "loss/crossentropy": 2.110167443752289,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2026640698313713,
+      "step": 15784
+    },
+    {
+      "epoch": 0.31572,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.02655029296875,
+      "learning_rate": 0.0001,
+      "loss": 4.1932,
+      "loss/crossentropy": 2.5422832369804382,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20274986326694489,
+      "step": 15786
+    },
+    {
+      "epoch": 0.31576,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.02840576171875,
+      "learning_rate": 0.0001,
+      "loss": 4.2038,
+      "loss/crossentropy": 2.106445074081421,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20275159180164337,
+      "step": 15788
+    },
+    {
+      "epoch": 0.3158,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.029857381184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.2718,
+      "loss/crossentropy": 2.341952681541443,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22970512509346008,
+      "step": 15790
+    },
+    {
+      "epoch": 0.31584,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.030228678385416666,
+      "learning_rate": 0.0001,
+      "loss": 4.238,
+      "loss/crossentropy": 2.120057761669159,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21171308308839798,
+      "step": 15792
+    },
+    {
+      "epoch": 0.31588,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.027644856770833334,
+      "learning_rate": 0.0001,
+      "loss": 3.8819,
+      "loss/crossentropy": 1.5601414442062378,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20278310775756836,
+      "step": 15794
+    },
+    {
+      "epoch": 0.31592,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.027596028645833333,
+      "learning_rate": 0.0001,
+      "loss": 4.0743,
+      "loss/crossentropy": 2.0634138584136963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1815958246588707,
+      "step": 15796
+    },
+    {
+      "epoch": 0.31596,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.0312652587890625,
+      "learning_rate": 0.0001,
+      "loss": 3.8183,
+      "loss/crossentropy": 1.8437300324440002,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18099966645240784,
+      "step": 15798
+    },
+    {
+      "epoch": 0.316,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.015897369384765624,
+      "learning_rate": 0.0001,
+      "loss": 4.1102,
+      "loss/crossentropy": 1.9323074221611023,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1884680539369583,
+      "step": 15800
+    },
+    {
+      "epoch": 0.31604,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.018027496337890626,
+      "learning_rate": 0.0001,
+      "loss": 4.3145,
+      "loss/crossentropy": 2.158499598503113,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23394601047039032,
+      "step": 15802
+    },
+    {
+      "epoch": 0.31608,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.018062337239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.0894,
+      "loss/crossentropy": 1.9635908007621765,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20692527294158936,
+      "step": 15804
+    },
+    {
+      "epoch": 0.31612,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.010503896077473958,
+      "learning_rate": 0.0001,
+      "loss": 4.1912,
+      "loss/crossentropy": 1.96099454164505,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1971232146024704,
+      "step": 15806
+    },
+    {
+      "epoch": 0.31616,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.010643513997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2205,
+      "loss/crossentropy": 1.9690999388694763,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19965435564517975,
+      "step": 15808
+    },
+    {
+      "epoch": 0.3162,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.011354319254557292,
+      "learning_rate": 0.0001,
+      "loss": 4.113,
+      "loss/crossentropy": 2.3924723863601685,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2192596197128296,
+      "step": 15810
+    },
+    {
+      "epoch": 0.31624,
+      "grad_norm": 1.7890625,
+      "grad_norm_var": 0.013999176025390626,
+      "learning_rate": 0.0001,
+      "loss": 3.8794,
+      "loss/crossentropy": 2.137382209300995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20949169248342514,
+      "step": 15812
+    },
+    {
+      "epoch": 0.31628,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.013103993733723958,
+      "learning_rate": 0.0001,
+      "loss": 4.0187,
+      "loss/crossentropy": 1.986245334148407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19678276032209396,
+      "step": 15814
+    },
+    {
+      "epoch": 0.31632,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.013492584228515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9623,
+      "loss/crossentropy": 1.9210307598114014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18748773634433746,
+      "step": 15816
+    },
+    {
+      "epoch": 0.31636,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0087066650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1328,
+      "loss/crossentropy": 1.9330076575279236,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21187862008810043,
+      "step": 15818
+    },
+    {
+      "epoch": 0.3164,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007306925455729167,
+      "learning_rate": 0.0001,
+      "loss": 4.2475,
+      "loss/crossentropy": 2.275822639465332,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2303369864821434,
+      "step": 15820
+    },
+    {
+      "epoch": 0.31644,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0082275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3641,
+      "loss/crossentropy": 1.9317327737808228,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19543734192848206,
+      "step": 15822
+    },
+    {
+      "epoch": 0.31648,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.008715565999348958,
+      "learning_rate": 0.0001,
+      "loss": 4.3209,
+      "loss/crossentropy": 2.2146100997924805,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22172697633504868,
+      "step": 15824
+    },
+    {
+      "epoch": 0.31652,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.007173411051432292,
+      "learning_rate": 0.0001,
+      "loss": 4.0226,
+      "loss/crossentropy": 2.296873092651367,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22240595519542694,
+      "step": 15826
+    },
+    {
+      "epoch": 0.31656,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.005900065104166667,
+      "learning_rate": 0.0001,
+      "loss": 4.2914,
+      "loss/crossentropy": 2.130843997001648,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21288827806711197,
+      "step": 15828
+    },
+    {
+      "epoch": 0.3166,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.017856597900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1067,
+      "loss/crossentropy": 2.042823553085327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1972874104976654,
+      "step": 15830
+    },
+    {
+      "epoch": 0.31664,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.01701838175455729,
+      "learning_rate": 0.0001,
+      "loss": 3.8878,
+      "loss/crossentropy": 1.8040228486061096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19943677634000778,
+      "step": 15832
+    },
+    {
+      "epoch": 0.31668,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.018400065104166665,
+      "learning_rate": 0.0001,
+      "loss": 4.2647,
+      "loss/crossentropy": 2.085720181465149,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2592329904437065,
+      "step": 15834
+    },
+    {
+      "epoch": 0.31672,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.022334798177083334,
+      "learning_rate": 0.0001,
+      "loss": 3.7694,
+      "loss/crossentropy": 1.9271164536476135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1766722872853279,
+      "step": 15836
+    },
+    {
+      "epoch": 0.31676,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.022106679280598958,
+      "learning_rate": 0.0001,
+      "loss": 4.0742,
+      "loss/crossentropy": 2.188960611820221,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19409415870904922,
+      "step": 15838
+    },
+    {
+      "epoch": 0.3168,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.020949045817057293,
+      "learning_rate": 0.0001,
+      "loss": 3.7821,
+      "loss/crossentropy": 1.5765752792358398,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16517025232315063,
+      "step": 15840
+    },
+    {
+      "epoch": 0.31684,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.020804595947265626,
+      "learning_rate": 0.0001,
+      "loss": 4.3356,
+      "loss/crossentropy": 1.9913761019706726,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20838972181081772,
+      "step": 15842
+    },
+    {
+      "epoch": 0.31688,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.034366607666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3525,
+      "loss/crossentropy": 1.8618363738059998,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19304241985082626,
+      "step": 15844
+    },
+    {
+      "epoch": 0.31692,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.026911417643229168,
+      "learning_rate": 0.0001,
+      "loss": 3.8087,
+      "loss/crossentropy": 1.9099775552749634,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2251715511083603,
+      "step": 15846
+    },
+    {
+      "epoch": 0.31696,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.026486968994140624,
+      "learning_rate": 0.0001,
+      "loss": 4.0781,
+      "loss/crossentropy": 2.3534491062164307,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19920790195465088,
+      "step": 15848
+    },
+    {
+      "epoch": 0.317,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.02804743448893229,
+      "learning_rate": 0.0001,
+      "loss": 4.1676,
+      "loss/crossentropy": 2.1890534162521362,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2140251249074936,
+      "step": 15850
+    },
+    {
+      "epoch": 0.31704,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.0298980712890625,
+      "learning_rate": 0.0001,
+      "loss": 3.6303,
+      "loss/crossentropy": 1.880380094051361,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18094944953918457,
+      "step": 15852
+    },
+    {
+      "epoch": 0.31708,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.03880106608072917,
+      "learning_rate": 0.0001,
+      "loss": 4.3276,
+      "loss/crossentropy": 1.8723257184028625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2101321741938591,
+      "step": 15854
+    },
+    {
+      "epoch": 0.31712,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.039582316080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.077,
+      "loss/crossentropy": 2.0442943572998047,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1947987675666809,
+      "step": 15856
+    },
+    {
+      "epoch": 0.31716,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.03819071451822917,
+      "learning_rate": 0.0001,
+      "loss": 4.2969,
+      "loss/crossentropy": 2.0415892601013184,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2022911012172699,
+      "step": 15858
+    },
+    {
+      "epoch": 0.3172,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.02176513671875,
+      "learning_rate": 0.0001,
+      "loss": 3.8801,
+      "loss/crossentropy": 1.8290876150131226,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1914537101984024,
+      "step": 15860
+    },
+    {
+      "epoch": 0.31724,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.022904459635416666,
+      "learning_rate": 0.0001,
+      "loss": 3.8257,
+      "loss/crossentropy": 1.6169148087501526,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21117067337036133,
+      "step": 15862
+    },
+    {
+      "epoch": 0.31728,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.022648111979166666,
+      "learning_rate": 0.0001,
+      "loss": 4.2553,
+      "loss/crossentropy": 2.3701289892196655,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2557762861251831,
+      "step": 15864
+    },
+    {
+      "epoch": 0.31732,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.019657135009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2299,
+      "loss/crossentropy": 2.2519643306732178,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21390919387340546,
+      "step": 15866
+    },
+    {
+      "epoch": 0.31736,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.015409088134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.266,
+      "loss/crossentropy": 2.3287068009376526,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22211260348558426,
+      "step": 15868
+    },
+    {
+      "epoch": 0.3174,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.004401652018229166,
+      "learning_rate": 0.0001,
+      "loss": 3.9622,
+      "loss/crossentropy": 2.144728899002075,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2148151397705078,
+      "step": 15870
+    },
+    {
+      "epoch": 0.31744,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.004847971598307291,
+      "learning_rate": 0.0001,
+      "loss": 3.9042,
+      "loss/crossentropy": 2.120876669883728,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21692734956741333,
+      "step": 15872
+    },
+    {
+      "epoch": 0.31748,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.004703521728515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9254,
+      "loss/crossentropy": 2.2827231884002686,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20714152604341507,
+      "step": 15874
+    },
+    {
+      "epoch": 0.31752,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.004788970947265625,
+      "learning_rate": 0.0001,
+      "loss": 3.8742,
+      "loss/crossentropy": 1.8011438846588135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19294942915439606,
+      "step": 15876
+    },
+    {
+      "epoch": 0.31756,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.0048258463541666664,
+      "learning_rate": 0.0001,
+      "loss": 3.9082,
+      "loss/crossentropy": 1.9359464049339294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19418440759181976,
+      "step": 15878
+    },
+    {
+      "epoch": 0.3176,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.006371053059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.4035,
+      "loss/crossentropy": 2.220693826675415,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2583991438150406,
+      "step": 15880
+    },
+    {
+      "epoch": 0.31764,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.006254069010416667,
+      "learning_rate": 0.0001,
+      "loss": 3.9762,
+      "loss/crossentropy": 1.7859990000724792,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1756511777639389,
+      "step": 15882
+    },
+    {
+      "epoch": 0.31768,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.00587158203125,
+      "learning_rate": 0.0001,
+      "loss": 3.7008,
+      "loss/crossentropy": 1.9461398720741272,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18634959310293198,
+      "step": 15884
+    },
+    {
+      "epoch": 0.31772,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.010833485921223959,
+      "learning_rate": 0.0001,
+      "loss": 4.3732,
+      "loss/crossentropy": 2.1066328287124634,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2553827613592148,
+      "step": 15886
+    },
+    {
+      "epoch": 0.31776,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011087799072265625,
+      "learning_rate": 0.0001,
+      "loss": 3.8751,
+      "loss/crossentropy": 1.7809287309646606,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19362767785787582,
+      "step": 15888
+    },
+    {
+      "epoch": 0.3178,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.01077880859375,
+      "learning_rate": 0.0001,
+      "loss": 4.0424,
+      "loss/crossentropy": 1.9554992318153381,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18544971197843552,
+      "step": 15890
+    },
+    {
+      "epoch": 0.31784,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.0327392578125,
+      "learning_rate": 0.0001,
+      "loss": 3.9083,
+      "loss/crossentropy": 1.5843109488487244,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15569136291742325,
+      "step": 15892
+    },
+    {
+      "epoch": 0.31788,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0307525634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1213,
+      "loss/crossentropy": 2.2000880241394043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2167157679796219,
+      "step": 15894
+    },
+    {
+      "epoch": 0.31792,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.031870269775390626,
+      "learning_rate": 0.0001,
+      "loss": 4.265,
+      "loss/crossentropy": 2.2597590684890747,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21171507239341736,
+      "step": 15896
+    },
+    {
+      "epoch": 0.31796,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.03181330362955729,
+      "learning_rate": 0.0001,
+      "loss": 3.9516,
+      "loss/crossentropy": 2.0513627529144287,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21095634251832962,
+      "step": 15898
+    },
+    {
+      "epoch": 0.318,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.029271443684895832,
+      "learning_rate": 0.0001,
+      "loss": 4.0667,
+      "loss/crossentropy": 2.1822222471237183,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21117551624774933,
+      "step": 15900
+    },
+    {
+      "epoch": 0.31804,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.027784983317057293,
+      "learning_rate": 0.0001,
+      "loss": 4.3526,
+      "loss/crossentropy": 1.9500460624694824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20769241452217102,
+      "step": 15902
+    },
+    {
+      "epoch": 0.31808,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.02684326171875,
+      "learning_rate": 0.0001,
+      "loss": 4.2862,
+      "loss/crossentropy": 2.179791212081909,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20147693157196045,
+      "step": 15904
+    },
+    {
+      "epoch": 0.31812,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.025217437744140626,
+      "learning_rate": 0.0001,
+      "loss": 4.0485,
+      "loss/crossentropy": 1.939364731311798,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18659047782421112,
+      "step": 15906
+    },
+    {
+      "epoch": 0.31816,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0059478759765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9782,
+      "loss/crossentropy": 2.219409167766571,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19967788457870483,
+      "step": 15908
+    },
+    {
+      "epoch": 0.3182,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.007741038004557292,
+      "learning_rate": 0.0001,
+      "loss": 3.829,
+      "loss/crossentropy": 1.8363550901412964,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18773943930864334,
+      "step": 15910
+    },
+    {
+      "epoch": 0.31824,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.006151326497395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2196,
+      "loss/crossentropy": 1.779226541519165,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20676184445619583,
+      "step": 15912
+    },
+    {
+      "epoch": 0.31828,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.006172688802083334,
+      "learning_rate": 0.0001,
+      "loss": 3.9532,
+      "loss/crossentropy": 2.1175760626792908,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20498017221689224,
+      "step": 15914
+    },
+    {
+      "epoch": 0.31832,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.005893707275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1601,
+      "loss/crossentropy": 2.0964329838752747,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1898122876882553,
+      "step": 15916
+    },
+    {
+      "epoch": 0.31836,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.005680084228515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9093,
+      "loss/crossentropy": 1.9687228798866272,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1931748390197754,
+      "step": 15918
+    },
+    {
+      "epoch": 0.3184,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.005197906494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0003,
+      "loss/crossentropy": 1.8344767689704895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19926829636096954,
+      "step": 15920
+    },
+    {
+      "epoch": 0.31844,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.004443359375,
+      "learning_rate": 0.0001,
+      "loss": 4.0426,
+      "loss/crossentropy": 2.154146194458008,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21284651011228561,
+      "step": 15922
+    },
+    {
+      "epoch": 0.31848,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.004223378499348959,
+      "learning_rate": 0.0001,
+      "loss": 4.2112,
+      "loss/crossentropy": 1.877986490726471,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18935512751340866,
+      "step": 15924
+    },
+    {
+      "epoch": 0.31852,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.009297434488932292,
+      "learning_rate": 0.0001,
+      "loss": 4.4296,
+      "loss/crossentropy": 2.361938714981079,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23210977017879486,
+      "step": 15926
+    },
+    {
+      "epoch": 0.31856,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.01043701171875,
+      "learning_rate": 0.0001,
+      "loss": 4.1209,
+      "loss/crossentropy": 2.3886083364486694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22222661972045898,
+      "step": 15928
+    },
+    {
+      "epoch": 0.3186,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.014806874593098958,
+      "learning_rate": 0.0001,
+      "loss": 4.1164,
+      "loss/crossentropy": 2.073238492012024,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20130325853824615,
+      "step": 15930
+    },
+    {
+      "epoch": 0.31864,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.015569814046223958,
+      "learning_rate": 0.0001,
+      "loss": 4.1692,
+      "loss/crossentropy": 1.9937713742256165,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.199419766664505,
+      "step": 15932
+    },
+    {
+      "epoch": 0.31868,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0135894775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0594,
+      "loss/crossentropy": 2.003056764602661,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18682510405778885,
+      "step": 15934
+    },
+    {
+      "epoch": 0.31872,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0137359619140625,
+      "learning_rate": 0.0001,
+      "loss": 3.9555,
+      "loss/crossentropy": 1.7363762259483337,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17377281934022903,
+      "step": 15936
+    },
+    {
+      "epoch": 0.31876,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.014412180582682291,
+      "learning_rate": 0.0001,
+      "loss": 4.1321,
+      "loss/crossentropy": 1.9791433811187744,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24057063460350037,
+      "step": 15938
+    },
+    {
+      "epoch": 0.3188,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.011378733317057292,
+      "learning_rate": 0.0001,
+      "loss": 3.9662,
+      "loss/crossentropy": 1.9962583780288696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19619113206863403,
+      "step": 15940
+    },
+    {
+      "epoch": 0.31884,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.01053466796875,
+      "learning_rate": 0.0001,
+      "loss": 4.2093,
+      "loss/crossentropy": 1.8568945527076721,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1979837343096733,
+      "step": 15942
+    },
+    {
+      "epoch": 0.31888,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.008991495768229166,
+      "learning_rate": 0.0001,
+      "loss": 4.0148,
+      "loss/crossentropy": 2.031070291996002,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1951117366552353,
+      "step": 15944
+    },
+    {
+      "epoch": 0.31892,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.007916005452473958,
+      "learning_rate": 0.0001,
+      "loss": 4.0139,
+      "loss/crossentropy": 2.2226544618606567,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21066214889287949,
+      "step": 15946
+    },
+    {
+      "epoch": 0.31896,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.007783762613932292,
+      "learning_rate": 0.0001,
+      "loss": 4.0275,
+      "loss/crossentropy": 2.090229034423828,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.225972518324852,
+      "step": 15948
+    },
+    {
+      "epoch": 0.319,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0080078125,
+      "learning_rate": 0.0001,
+      "loss": 3.9792,
+      "loss/crossentropy": 1.8562734723091125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17633049190044403,
+      "step": 15950
+    },
+    {
+      "epoch": 0.31904,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.00992431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0227,
+      "loss/crossentropy": 1.6677230596542358,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16521989554166794,
+      "step": 15952
+    },
+    {
+      "epoch": 0.31908,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.008939361572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2391,
+      "loss/crossentropy": 2.1918782591819763,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21977706253528595,
+      "step": 15954
+    },
+    {
+      "epoch": 0.31912,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.008701324462890625,
+      "learning_rate": 0.0001,
+      "loss": 3.8932,
+      "loss/crossentropy": 1.8623137474060059,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1937999278306961,
+      "step": 15956
+    },
+    {
+      "epoch": 0.31916,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.014839426676432291,
+      "learning_rate": 0.0001,
+      "loss": 4.2378,
+      "loss/crossentropy": 2.262889266014099,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19410346448421478,
+      "step": 15958
+    },
+    {
+      "epoch": 0.3192,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.014654286702473958,
+      "learning_rate": 0.0001,
+      "loss": 4.0487,
+      "loss/crossentropy": 1.9620846509933472,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1978064849972725,
+      "step": 15960
+    },
+    {
+      "epoch": 0.31924,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.014207967122395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2746,
+      "loss/crossentropy": 2.453945517539978,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21578459441661835,
+      "step": 15962
+    },
+    {
+      "epoch": 0.31928,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.013529459635416666,
+      "learning_rate": 0.0001,
+      "loss": 3.9692,
+      "loss/crossentropy": 1.9565781354904175,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17789364606142044,
+      "step": 15964
+    },
+    {
+      "epoch": 0.31932,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.012589518229166667,
+      "learning_rate": 0.0001,
+      "loss": 4.3504,
+      "loss/crossentropy": 2.1175976991653442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2239348292350769,
+      "step": 15966
+    },
+    {
+      "epoch": 0.31936,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.010554758707682292,
+      "learning_rate": 0.0001,
+      "loss": 4.0815,
+      "loss/crossentropy": 1.8919953107833862,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19832338392734528,
+      "step": 15968
+    },
+    {
+      "epoch": 0.3194,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0126220703125,
+      "learning_rate": 0.0001,
+      "loss": 4.1282,
+      "loss/crossentropy": 1.9757861495018005,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19513867795467377,
+      "step": 15970
+    },
+    {
+      "epoch": 0.31944,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.012630208333333334,
+      "learning_rate": 0.0001,
+      "loss": 4.1045,
+      "loss/crossentropy": 1.8716764450073242,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18571900576353073,
+      "step": 15972
+    },
+    {
+      "epoch": 0.31948,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.006575520833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.2572,
+      "loss/crossentropy": 2.251496374607086,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2199217826128006,
+      "step": 15974
+    },
+    {
+      "epoch": 0.31952,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.008424631754557292,
+      "learning_rate": 0.0001,
+      "loss": 4.1033,
+      "loss/crossentropy": 1.8639826774597168,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18889226019382477,
+      "step": 15976
+    },
+    {
+      "epoch": 0.31956,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.008499908447265624,
+      "learning_rate": 0.0001,
+      "loss": 3.8386,
+      "loss/crossentropy": 1.824375331401825,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1809096783399582,
+      "step": 15978
+    },
+    {
+      "epoch": 0.3196,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010361480712890624,
+      "learning_rate": 0.0001,
+      "loss": 4.4576,
+      "loss/crossentropy": 2.357889413833618,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21762560307979584,
+      "step": 15980
+    },
+    {
+      "epoch": 0.31964,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.011527252197265626,
+      "learning_rate": 0.0001,
+      "loss": 4.0706,
+      "loss/crossentropy": 2.0944892168045044,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18413915485143661,
+      "step": 15982
+    },
+    {
+      "epoch": 0.31968,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.011736806233723958,
+      "learning_rate": 0.0001,
+      "loss": 3.9458,
+      "loss/crossentropy": 1.8600184321403503,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18580162525177002,
+      "step": 15984
+    },
+    {
+      "epoch": 0.31972,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008697255452473959,
+      "learning_rate": 0.0001,
+      "loss": 4.1152,
+      "loss/crossentropy": 1.9213852882385254,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20599720627069473,
+      "step": 15986
+    },
+    {
+      "epoch": 0.31976,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0119537353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2187,
+      "loss/crossentropy": 2.0295584201812744,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2125483900308609,
+      "step": 15988
+    },
+    {
+      "epoch": 0.3198,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.012198638916015626,
+      "learning_rate": 0.0001,
+      "loss": 4.2448,
+      "loss/crossentropy": 2.148836612701416,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22027160972356796,
+      "step": 15990
+    },
+    {
+      "epoch": 0.31984,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.010201009114583333,
+      "learning_rate": 0.0001,
+      "loss": 4.122,
+      "loss/crossentropy": 1.9927499294281006,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20413245260715485,
+      "step": 15992
+    },
+    {
+      "epoch": 0.31988,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0113433837890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2675,
+      "loss/crossentropy": 2.090071678161621,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21670149266719818,
+      "step": 15994
+    },
+    {
+      "epoch": 0.31992,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.010074869791666666,
+      "learning_rate": 0.0001,
+      "loss": 4.1333,
+      "loss/crossentropy": 1.8375160098075867,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1938413679599762,
+      "step": 15996
+    },
+    {
+      "epoch": 0.31996,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.010498046875,
+      "learning_rate": 0.0001,
+      "loss": 3.8915,
+      "loss/crossentropy": 1.5196507573127747,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16407200694084167,
+      "step": 15998
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.011344146728515626,
+      "learning_rate": 0.0001,
+      "loss": 4.0965,
+      "loss/crossentropy": 2.0738461017608643,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21532364934682846,
+      "step": 16000
+    },
+    {
+      "epoch": 0.32004,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.011948394775390624,
+      "learning_rate": 0.0001,
+      "loss": 4.0687,
+      "loss/crossentropy": 2.3243744373321533,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23544494807720184,
+      "step": 16002
+    },
+    {
+      "epoch": 0.32008,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.009862263997395834,
+      "learning_rate": 0.0001,
+      "loss": 3.9637,
+      "loss/crossentropy": 1.7145931124687195,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1816524863243103,
+      "step": 16004
+    },
+    {
+      "epoch": 0.32012,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010506184895833333,
+      "learning_rate": 0.0001,
+      "loss": 4.2317,
+      "loss/crossentropy": 2.369017481803894,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21978877484798431,
+      "step": 16006
+    },
+    {
+      "epoch": 0.32016,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.010276031494140626,
+      "learning_rate": 0.0001,
+      "loss": 4.2283,
+      "loss/crossentropy": 2.2246848344802856,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21983492374420166,
+      "step": 16008
+    },
+    {
+      "epoch": 0.3202,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.03774185180664062,
+      "learning_rate": 0.0001,
+      "loss": 4.4077,
+      "loss/crossentropy": 2.0235647559165955,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22780708968639374,
+      "step": 16010
+    },
+    {
+      "epoch": 0.32024,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.03724543253580729,
+      "learning_rate": 0.0001,
+      "loss": 4.1692,
+      "loss/crossentropy": 2.3185853958129883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22308620065450668,
+      "step": 16012
+    },
+    {
+      "epoch": 0.32028,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1314,
+      "loss/crossentropy": 1.8504603505134583,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1909484937787056,
+      "step": 16014
+    },
+    {
+      "epoch": 0.32032,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.03502604166666667,
+      "learning_rate": 0.0001,
+      "loss": 3.8247,
+      "loss/crossentropy": 1.7095224857330322,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18588167428970337,
+      "step": 16016
+    },
+    {
+      "epoch": 0.32036,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.03749974568684896,
+      "learning_rate": 0.0001,
+      "loss": 4.1015,
+      "loss/crossentropy": 1.8398523330688477,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18085161596536636,
+      "step": 16018
+    },
+    {
+      "epoch": 0.3204,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.03569717407226562,
+      "learning_rate": 0.0001,
+      "loss": 4.1463,
+      "loss/crossentropy": 2.297620415687561,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20937249064445496,
+      "step": 16020
+    },
+    {
+      "epoch": 0.32044,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.03618545532226562,
+      "learning_rate": 0.0001,
+      "loss": 4.3518,
+      "loss/crossentropy": 1.9124106764793396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2072141021490097,
+      "step": 16022
+    },
+    {
+      "epoch": 0.32048,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.03752415974934896,
+      "learning_rate": 0.0001,
+      "loss": 4.2695,
+      "loss/crossentropy": 2.119450092315674,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20875975489616394,
+      "step": 16024
+    },
+    {
+      "epoch": 0.32052,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.008125559488932291,
+      "learning_rate": 0.0001,
+      "loss": 4.35,
+      "loss/crossentropy": 1.722611904144287,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1751066967844963,
+      "step": 16026
+    },
+    {
+      "epoch": 0.32056,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.008760579427083333,
+      "learning_rate": 0.0001,
+      "loss": 4.0689,
+      "loss/crossentropy": 2.246767997741699,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1999046877026558,
+      "step": 16028
+    },
+    {
+      "epoch": 0.3206,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0083404541015625,
+      "learning_rate": 0.0001,
+      "loss": 3.9949,
+      "loss/crossentropy": 2.0449349880218506,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1969577819108963,
+      "step": 16030
+    },
+    {
+      "epoch": 0.32064,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.010050201416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1257,
+      "loss/crossentropy": 2.125577926635742,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22252507507801056,
+      "step": 16032
+    },
+    {
+      "epoch": 0.32068,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.009079742431640624,
+      "learning_rate": 0.0001,
+      "loss": 3.639,
+      "loss/crossentropy": 1.8161216974258423,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17801867425441742,
+      "step": 16034
+    },
+    {
+      "epoch": 0.32072,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.009422810872395833,
+      "learning_rate": 0.0001,
+      "loss": 3.9475,
+      "loss/crossentropy": 2.0116894841194153,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18890849500894547,
+      "step": 16036
+    },
+    {
+      "epoch": 0.32076,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.0100982666015625,
+      "learning_rate": 0.0001,
+      "loss": 3.8274,
+      "loss/crossentropy": 1.505588710308075,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17151623219251633,
+      "step": 16038
+    },
+    {
+      "epoch": 0.3208,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.008585357666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2162,
+      "loss/crossentropy": 2.1405014991760254,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2036641389131546,
+      "step": 16040
+    },
+    {
+      "epoch": 0.32084,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.006628163655598958,
+      "learning_rate": 0.0001,
+      "loss": 4.0743,
+      "loss/crossentropy": 2.1562809348106384,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21301954984664917,
+      "step": 16042
+    },
+    {
+      "epoch": 0.32088,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.006628163655598958,
+      "learning_rate": 0.0001,
+      "loss": 3.9773,
+      "loss/crossentropy": 1.9317167401313782,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19167031347751617,
+      "step": 16044
+    },
+    {
+      "epoch": 0.32092,
+      "grad_norm": 1.796875,
+      "grad_norm_var": 0.007330067952473958,
+      "learning_rate": 0.0001,
+      "loss": 4.0523,
+      "loss/crossentropy": 1.6257455348968506,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1676744669675827,
+      "step": 16046
+    },
+    {
+      "epoch": 0.32096,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.006640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2454,
+      "loss/crossentropy": 2.0711347460746765,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2268730029463768,
+      "step": 16048
+    },
+    {
+      "epoch": 0.321,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0071489969889322914,
+      "learning_rate": 0.0001,
+      "loss": 4.3333,
+      "loss/crossentropy": 2.173452377319336,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1981828361749649,
+      "step": 16050
+    },
+    {
+      "epoch": 0.32104,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.009593709309895834,
+      "learning_rate": 0.0001,
+      "loss": 4.4601,
+      "loss/crossentropy": 2.348356604576111,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2538425847887993,
+      "step": 16052
+    },
+    {
+      "epoch": 0.32108,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0089263916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3408,
+      "loss/crossentropy": 2.5441235303878784,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22638535499572754,
+      "step": 16054
+    },
+    {
+      "epoch": 0.32112,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.0098876953125,
+      "learning_rate": 0.0001,
+      "loss": 4.08,
+      "loss/crossentropy": 2.129339337348938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20430771261453629,
+      "step": 16056
+    },
+    {
+      "epoch": 0.32116,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.01142578125,
+      "learning_rate": 0.0001,
+      "loss": 3.9048,
+      "loss/crossentropy": 2.1599501371383667,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20718347281217575,
+      "step": 16058
+    },
+    {
+      "epoch": 0.3212,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.010296376546223958,
+      "learning_rate": 0.0001,
+      "loss": 4.2746,
+      "loss/crossentropy": 2.2704890966415405,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22025802731513977,
+      "step": 16060
+    },
+    {
+      "epoch": 0.32124,
+      "grad_norm": 1.796875,
+      "grad_norm_var": 0.011702473958333333,
+      "learning_rate": 0.0001,
+      "loss": 3.7282,
+      "loss/crossentropy": 1.7704021334648132,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17811128497123718,
+      "step": 16062
+    },
+    {
+      "epoch": 0.32128,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.015705362955729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2366,
+      "loss/crossentropy": 1.8030991554260254,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20078134536743164,
+      "step": 16064
+    },
+    {
+      "epoch": 0.32132,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0161041259765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9832,
+      "loss/crossentropy": 2.14484703540802,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21000836789608002,
+      "step": 16066
+    },
+    {
+      "epoch": 0.32136,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.016521962483723958,
+      "learning_rate": 0.0001,
+      "loss": 4.0812,
+      "loss/crossentropy": 2.1680833101272583,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22287416458129883,
+      "step": 16068
+    },
+    {
+      "epoch": 0.3214,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.01708958943684896,
+      "learning_rate": 0.0001,
+      "loss": 4.1355,
+      "loss/crossentropy": 1.9971441626548767,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20709602534770966,
+      "step": 16070
+    },
+    {
+      "epoch": 0.32144,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.016462198893229165,
+      "learning_rate": 0.0001,
+      "loss": 4.2735,
+      "loss/crossentropy": 2.180538833141327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2097538262605667,
+      "step": 16072
+    },
+    {
+      "epoch": 0.32148,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.013529459635416666,
+      "learning_rate": 0.0001,
+      "loss": 4.0725,
+      "loss/crossentropy": 2.074296534061432,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20961299538612366,
+      "step": 16074
+    },
+    {
+      "epoch": 0.32152,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.014998372395833333,
+      "learning_rate": 0.0001,
+      "loss": 4.3,
+      "loss/crossentropy": 2.0936360359191895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18253444135189056,
+      "step": 16076
+    },
+    {
+      "epoch": 0.32156,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.0121337890625,
+      "learning_rate": 0.0001,
+      "loss": 3.9992,
+      "loss/crossentropy": 2.1265164613723755,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2127028927206993,
+      "step": 16078
+    },
+    {
+      "epoch": 0.3216,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.008410390218098958,
+      "learning_rate": 0.0001,
+      "loss": 4.1907,
+      "loss/crossentropy": 2.0804547667503357,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23104286193847656,
+      "step": 16080
+    },
+    {
+      "epoch": 0.32164,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.007759602864583334,
+      "learning_rate": 0.0001,
+      "loss": 3.9653,
+      "loss/crossentropy": 1.9458459615707397,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18746595829725266,
+      "step": 16082
+    },
+    {
+      "epoch": 0.32168,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0056793212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0581,
+      "loss/crossentropy": 2.186868667602539,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20350559055805206,
+      "step": 16084
+    },
+    {
+      "epoch": 0.32172,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.007340494791666667,
+      "learning_rate": 0.0001,
+      "loss": 4.2364,
+      "loss/crossentropy": 1.5468108654022217,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17710693180561066,
+      "step": 16086
+    },
+    {
+      "epoch": 0.32176,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.006180572509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2635,
+      "loss/crossentropy": 2.222296118736267,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22016968578100204,
+      "step": 16088
+    },
+    {
+      "epoch": 0.3218,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.005619049072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0659,
+      "loss/crossentropy": 1.7950996160507202,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1892552226781845,
+      "step": 16090
+    },
+    {
+      "epoch": 0.32184,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.007120768229166667,
+      "learning_rate": 0.0001,
+      "loss": 4.2899,
+      "loss/crossentropy": 2.293117642402649,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2351122871041298,
+      "step": 16092
+    },
+    {
+      "epoch": 0.32188,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.006208292643229167,
+      "learning_rate": 0.0001,
+      "loss": 3.8235,
+      "loss/crossentropy": 1.6602438688278198,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18052196502685547,
+      "step": 16094
+    },
+    {
+      "epoch": 0.32192,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.008501942952473958,
+      "learning_rate": 0.0001,
+      "loss": 4.114,
+      "loss/crossentropy": 1.9335945844650269,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20585428178310394,
+      "step": 16096
+    },
+    {
+      "epoch": 0.32196,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.011378733317057292,
+      "learning_rate": 0.0001,
+      "loss": 4.3911,
+      "loss/crossentropy": 2.0107831358909607,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19297368824481964,
+      "step": 16098
+    },
+    {
+      "epoch": 0.322,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.011801910400390626,
+      "learning_rate": 0.0001,
+      "loss": 4.0153,
+      "loss/crossentropy": 1.9804013967514038,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17699319869279861,
+      "step": 16100
+    },
+    {
+      "epoch": 0.32204,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.01048583984375,
+      "learning_rate": 0.0001,
+      "loss": 4.1069,
+      "loss/crossentropy": 1.9009913206100464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1931428462266922,
+      "step": 16102
+    },
+    {
+      "epoch": 0.32208,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.011358388264973958,
+      "learning_rate": 0.0001,
+      "loss": 4.5134,
+      "loss/crossentropy": 1.9558063745498657,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2139815017580986,
+      "step": 16104
+    },
+    {
+      "epoch": 0.32212,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.013323720296223958,
+      "learning_rate": 0.0001,
+      "loss": 4.0989,
+      "loss/crossentropy": 2.0626373887062073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2022782266139984,
+      "step": 16106
+    },
+    {
+      "epoch": 0.32216,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.013814036051432292,
+      "learning_rate": 0.0001,
+      "loss": 3.9408,
+      "loss/crossentropy": 2.166835308074951,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22070696204900742,
+      "step": 16108
+    },
+    {
+      "epoch": 0.3222,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.012963612874348959,
+      "learning_rate": 0.0001,
+      "loss": 4.1382,
+      "loss/crossentropy": 2.2970025539398193,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23291311413049698,
+      "step": 16110
+    },
+    {
+      "epoch": 0.32224,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.012174224853515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9577,
+      "loss/crossentropy": 1.9400970935821533,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.199058435857296,
+      "step": 16112
+    },
+    {
+      "epoch": 0.32228,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.009110260009765624,
+      "learning_rate": 0.0001,
+      "loss": 3.9659,
+      "loss/crossentropy": 1.6467864513397217,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19500216841697693,
+      "step": 16114
+    },
+    {
+      "epoch": 0.32232,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.010546875,
+      "learning_rate": 0.0001,
+      "loss": 4.1791,
+      "loss/crossentropy": 2.477790355682373,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22540342807769775,
+      "step": 16116
+    },
+    {
+      "epoch": 0.32236,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.011066691080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0661,
+      "loss/crossentropy": 1.946933627128601,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17866672575473785,
+      "step": 16118
+    },
+    {
+      "epoch": 0.3224,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.01153564453125,
+      "learning_rate": 0.0001,
+      "loss": 4.1416,
+      "loss/crossentropy": 2.4484145641326904,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21314629167318344,
+      "step": 16120
+    },
+    {
+      "epoch": 0.32244,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.009155019124348959,
+      "learning_rate": 0.0001,
+      "loss": 3.9754,
+      "loss/crossentropy": 1.9663755893707275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19814299046993256,
+      "step": 16122
+    },
+    {
+      "epoch": 0.32248,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.5887278238932292,
+      "learning_rate": 0.0001,
+      "loss": 3.7417,
+      "loss/crossentropy": 1.9786240458488464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19666218757629395,
+      "step": 16124
+    },
+    {
+      "epoch": 0.32252,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.5901079813639323,
+      "learning_rate": 0.0001,
+      "loss": 3.9504,
+      "loss/crossentropy": 1.90617835521698,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1803947538137436,
+      "step": 16126
+    },
+    {
+      "epoch": 0.32256,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.5877764383951823,
+      "learning_rate": 0.0001,
+      "loss": 4.3267,
+      "loss/crossentropy": 2.1199004650115967,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21409574151039124,
+      "step": 16128
+    },
+    {
+      "epoch": 0.3226,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.5875445048014323,
+      "learning_rate": 0.0001,
+      "loss": 4.184,
+      "loss/crossentropy": 2.081290364265442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20103489607572556,
+      "step": 16130
+    },
+    {
+      "epoch": 0.32264,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.5902565002441407,
+      "learning_rate": 0.0001,
+      "loss": 4.216,
+      "loss/crossentropy": 1.9839922785758972,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20679790526628494,
+      "step": 16132
+    },
+    {
+      "epoch": 0.32268,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.5830800374348958,
+      "learning_rate": 0.0001,
+      "loss": 4.3441,
+      "loss/crossentropy": 2.2765486240386963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21012597531080246,
+      "step": 16134
+    },
+    {
+      "epoch": 0.32272,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.57388916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0003,
+      "loss/crossentropy": 1.8353520035743713,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19367212802171707,
+      "step": 16136
+    },
+    {
+      "epoch": 0.32276,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.5781572977701823,
+      "learning_rate": 0.0001,
+      "loss": 3.8104,
+      "loss/crossentropy": 1.8009640574455261,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18030469119548798,
+      "step": 16138
+    },
+    {
+      "epoch": 0.3228,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.020975494384765626,
+      "learning_rate": 0.0001,
+      "loss": 4.1914,
+      "loss/crossentropy": 1.865959644317627,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20606407523155212,
+      "step": 16140
+    },
+    {
+      "epoch": 0.32284,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.021445465087890626,
+      "learning_rate": 0.0001,
+      "loss": 4.0253,
+      "loss/crossentropy": 2.238165020942688,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22822897136211395,
+      "step": 16142
+    },
+    {
+      "epoch": 0.32288,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.02393366495768229,
+      "learning_rate": 0.0001,
+      "loss": 4.0425,
+      "loss/crossentropy": 2.175424814224243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2260037586092949,
+      "step": 16144
+    },
+    {
+      "epoch": 0.32292,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.025050608317057292,
+      "learning_rate": 0.0001,
+      "loss": 4.1493,
+      "loss/crossentropy": 2.2071722745895386,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23026321083307266,
+      "step": 16146
+    },
+    {
+      "epoch": 0.32296,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.024468739827473957,
+      "learning_rate": 0.0001,
+      "loss": 4.2384,
+      "loss/crossentropy": 2.211812973022461,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20882735401391983,
+      "step": 16148
+    },
+    {
+      "epoch": 0.323,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.02444636027018229,
+      "learning_rate": 0.0001,
+      "loss": 4.0356,
+      "loss/crossentropy": 1.6503818035125732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1640765517950058,
+      "step": 16150
+    },
+    {
+      "epoch": 0.32304,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.02520726521809896,
+      "learning_rate": 0.0001,
+      "loss": 4.0692,
+      "loss/crossentropy": 1.9449399709701538,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1876515969634056,
+      "step": 16152
+    },
+    {
+      "epoch": 0.32308,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.022581990559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1665,
+      "loss/crossentropy": 2.001387894153595,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21191075444221497,
+      "step": 16154
+    },
+    {
+      "epoch": 0.32312,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0143218994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2115,
+      "loss/crossentropy": 2.1146541833877563,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2692951038479805,
+      "step": 16156
+    },
+    {
+      "epoch": 0.32316,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.015602366129557291,
+      "learning_rate": 0.0001,
+      "loss": 4.1534,
+      "loss/crossentropy": 2.0765844583511353,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19518104195594788,
+      "step": 16158
+    },
+    {
+      "epoch": 0.3232,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0128814697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1432,
+      "loss/crossentropy": 2.5394046306610107,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22651594877243042,
+      "step": 16160
+    },
+    {
+      "epoch": 0.32324,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.011702473958333333,
+      "learning_rate": 0.0001,
+      "loss": 4.0412,
+      "loss/crossentropy": 2.0140721797943115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18245946615934372,
+      "step": 16162
+    },
+    {
+      "epoch": 0.32328,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.012303670247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.3029,
+      "loss/crossentropy": 2.472638249397278,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23127944022417068,
+      "step": 16164
+    },
+    {
+      "epoch": 0.32332,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.014924112955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.1974,
+      "loss/crossentropy": 2.1551883220672607,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2093031033873558,
+      "step": 16166
+    },
+    {
+      "epoch": 0.32336,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.01641845703125,
+      "learning_rate": 0.0001,
+      "loss": 4.1221,
+      "loss/crossentropy": 2.042950928211212,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2049141526222229,
+      "step": 16168
+    },
+    {
+      "epoch": 0.3234,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.014849599202473958,
+      "learning_rate": 0.0001,
+      "loss": 4.4795,
+      "loss/crossentropy": 2.0262559056282043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18649922311306,
+      "step": 16170
+    },
+    {
+      "epoch": 0.32344,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.011675771077473958,
+      "learning_rate": 0.0001,
+      "loss": 4.0477,
+      "loss/crossentropy": 2.3014276027679443,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21770215779542923,
+      "step": 16172
+    },
+    {
+      "epoch": 0.32348,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009349568684895834,
+      "learning_rate": 0.0001,
+      "loss": 4.2802,
+      "loss/crossentropy": 2.213370680809021,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18508044630289078,
+      "step": 16174
+    },
+    {
+      "epoch": 0.32352,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009075673421223958,
+      "learning_rate": 0.0001,
+      "loss": 4.2529,
+      "loss/crossentropy": 1.973215639591217,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1937488242983818,
+      "step": 16176
+    },
+    {
+      "epoch": 0.32356,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.018553670247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.3534,
+      "loss/crossentropy": 2.1585731506347656,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2029733881354332,
+      "step": 16178
+    },
+    {
+      "epoch": 0.3236,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.021647135416666668,
+      "learning_rate": 0.0001,
+      "loss": 4.3324,
+      "loss/crossentropy": 2.3878796100616455,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19985762238502502,
+      "step": 16180
+    },
+    {
+      "epoch": 0.32364,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.027860514322916665,
+      "learning_rate": 0.0001,
+      "loss": 3.8043,
+      "loss/crossentropy": 2.07839834690094,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1994347795844078,
+      "step": 16182
+    },
+    {
+      "epoch": 0.32368,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.02607421875,
+      "learning_rate": 0.0001,
+      "loss": 4.3147,
+      "loss/crossentropy": 2.1800806522369385,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21444538235664368,
+      "step": 16184
+    },
+    {
+      "epoch": 0.32372,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.029352823893229168,
+      "learning_rate": 0.0001,
+      "loss": 3.9927,
+      "loss/crossentropy": 2.0900917053222656,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20476030558347702,
+      "step": 16186
+    },
+    {
+      "epoch": 0.32376,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.02899169921875,
+      "learning_rate": 0.0001,
+      "loss": 4.209,
+      "loss/crossentropy": 2.1523303985595703,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20984850078821182,
+      "step": 16188
+    },
+    {
+      "epoch": 0.3238,
+      "grad_norm": 1.765625,
+      "grad_norm_var": 0.03593648274739583,
+      "learning_rate": 0.0001,
+      "loss": 3.799,
+      "loss/crossentropy": 1.8808646202087402,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18587414175271988,
+      "step": 16190
+    },
+    {
+      "epoch": 0.32384,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.03593648274739583,
+      "learning_rate": 0.0001,
+      "loss": 4.1002,
+      "loss/crossentropy": 1.981968104839325,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20611849427223206,
+      "step": 16192
+    },
+    {
+      "epoch": 0.32388,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.020951334635416666,
+      "learning_rate": 0.0001,
+      "loss": 4.2811,
+      "loss/crossentropy": 2.1454352140426636,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21696189790964127,
+      "step": 16194
+    },
+    {
+      "epoch": 0.32392,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.011905924479166666,
+      "learning_rate": 0.0001,
+      "loss": 4.159,
+      "loss/crossentropy": 2.026822328567505,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2112257331609726,
+      "step": 16196
+    },
+    {
+      "epoch": 0.32396,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.009706370035807292,
+      "learning_rate": 0.0001,
+      "loss": 4.0838,
+      "loss/crossentropy": 2.332329034805298,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20129209011793137,
+      "step": 16198
+    },
+    {
+      "epoch": 0.324,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.007260894775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.11,
+      "loss/crossentropy": 2.271657705307007,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2460309863090515,
+      "step": 16200
+    },
+    {
+      "epoch": 0.32404,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0064084370930989586,
+      "learning_rate": 0.0001,
+      "loss": 4.0835,
+      "loss/crossentropy": 2.0890920162200928,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2132192626595497,
+      "step": 16202
+    },
+    {
+      "epoch": 0.32408,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.007226308186848958,
+      "learning_rate": 0.0001,
+      "loss": 4.0271,
+      "loss/crossentropy": 1.8368538618087769,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1891050636768341,
+      "step": 16204
+    },
+    {
+      "epoch": 0.32412,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.004375966389973959,
+      "learning_rate": 0.0001,
+      "loss": 4.2859,
+      "loss/crossentropy": 2.3123772144317627,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22660605609416962,
+      "step": 16206
+    },
+    {
+      "epoch": 0.32416,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.005411529541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0989,
+      "loss/crossentropy": 2.21794331073761,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21767260879278183,
+      "step": 16208
+    },
+    {
+      "epoch": 0.3242,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010591379801432292,
+      "learning_rate": 0.0001,
+      "loss": 4.5407,
+      "loss/crossentropy": 2.169134736061096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2172672376036644,
+      "step": 16210
+    },
+    {
+      "epoch": 0.32424,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.011331939697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0327,
+      "loss/crossentropy": 2.2830699682235718,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22920189797878265,
+      "step": 16212
+    },
+    {
+      "epoch": 0.32428,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.012770334879557291,
+      "learning_rate": 0.0001,
+      "loss": 3.9278,
+      "loss/crossentropy": 2.024571657180786,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21096136420965195,
+      "step": 16214
+    },
+    {
+      "epoch": 0.32432,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.013057200113932292,
+      "learning_rate": 0.0001,
+      "loss": 4.0244,
+      "loss/crossentropy": 2.060720980167389,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21043655276298523,
+      "step": 16216
+    },
+    {
+      "epoch": 0.32436,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.014134724934895834,
+      "learning_rate": 0.0001,
+      "loss": 4.074,
+      "loss/crossentropy": 2.0332056283950806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19506454467773438,
+      "step": 16218
+    },
+    {
+      "epoch": 0.3244,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.014021555582682291,
+      "learning_rate": 0.0001,
+      "loss": 3.8173,
+      "loss/crossentropy": 1.811396062374115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19359800219535828,
+      "step": 16220
+    },
+    {
+      "epoch": 0.32444,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.013117472330729166,
+      "learning_rate": 0.0001,
+      "loss": 4.1846,
+      "loss/crossentropy": 2.074379801750183,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21763048321008682,
+      "step": 16222
+    },
+    {
+      "epoch": 0.32448,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.014682769775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1984,
+      "loss/crossentropy": 2.1836538314819336,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20814575254917145,
+      "step": 16224
+    },
+    {
+      "epoch": 0.32452,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.010794830322265626,
+      "learning_rate": 0.0001,
+      "loss": 4.2866,
+      "loss/crossentropy": 2.365849494934082,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24219170212745667,
+      "step": 16226
+    },
+    {
+      "epoch": 0.32456,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010343424479166667,
+      "learning_rate": 0.0001,
+      "loss": 4.1976,
+      "loss/crossentropy": 1.8738153576850891,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22152718901634216,
+      "step": 16228
+    },
+    {
+      "epoch": 0.3246,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.010209147135416667,
+      "learning_rate": 0.0001,
+      "loss": 4.1458,
+      "loss/crossentropy": 2.2568124532699585,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21668314933776855,
+      "step": 16230
+    },
+    {
+      "epoch": 0.32464,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.010900624593098958,
+      "learning_rate": 0.0001,
+      "loss": 4.2889,
+      "loss/crossentropy": 2.4037723541259766,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2225075364112854,
+      "step": 16232
+    },
+    {
+      "epoch": 0.32468,
+      "grad_norm": 1.765625,
+      "grad_norm_var": 0.01375732421875,
+      "learning_rate": 0.0001,
+      "loss": 3.7288,
+      "loss/crossentropy": 2.2305864095687866,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19381006807088852,
+      "step": 16234
+    },
+    {
+      "epoch": 0.32472,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.013862864176432291,
+      "learning_rate": 0.0001,
+      "loss": 4.186,
+      "loss/crossentropy": 2.1115931272506714,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20033665746450424,
+      "step": 16236
+    },
+    {
+      "epoch": 0.32476,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.014054361979166667,
+      "learning_rate": 0.0001,
+      "loss": 4.1602,
+      "loss/crossentropy": 2.0450429916381836,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2080780565738678,
+      "step": 16238
+    },
+    {
+      "epoch": 0.3248,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.011888631184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.056,
+      "loss/crossentropy": 2.003769636154175,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20560404658317566,
+      "step": 16240
+    },
+    {
+      "epoch": 0.32484,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010383860270182291,
+      "learning_rate": 0.0001,
+      "loss": 4.3504,
+      "loss/crossentropy": 1.9452934265136719,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2063707485795021,
+      "step": 16242
+    },
+    {
+      "epoch": 0.32488,
+      "grad_norm": 1.7734375,
+      "grad_norm_var": 0.012497711181640624,
+      "learning_rate": 0.0001,
+      "loss": 3.9172,
+      "loss/crossentropy": 2.1412742137908936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20292682200670242,
+      "step": 16244
+    },
+    {
+      "epoch": 0.32492,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.011140950520833333,
+      "learning_rate": 0.0001,
+      "loss": 4.3025,
+      "loss/crossentropy": 2.2285404205322266,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22571790218353271,
+      "step": 16246
+    },
+    {
+      "epoch": 0.32496,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.010713704427083333,
+      "learning_rate": 0.0001,
+      "loss": 3.6582,
+      "loss/crossentropy": 1.82416570186615,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1771545708179474,
+      "step": 16248
+    },
+    {
+      "epoch": 0.325,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.010422515869140624,
+      "learning_rate": 0.0001,
+      "loss": 4.2533,
+      "loss/crossentropy": 2.1432559490203857,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20037438720464706,
+      "step": 16250
+    },
+    {
+      "epoch": 0.32504,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.015337880452473958,
+      "learning_rate": 0.0001,
+      "loss": 4.162,
+      "loss/crossentropy": 2.1962517499923706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22743911296129227,
+      "step": 16252
+    },
+    {
+      "epoch": 0.32508,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.01636530558268229,
+      "learning_rate": 0.0001,
+      "loss": 4.2223,
+      "loss/crossentropy": 1.932455599308014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2043299823999405,
+      "step": 16254
+    },
+    {
+      "epoch": 0.32512,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.015824127197265624,
+      "learning_rate": 0.0001,
+      "loss": 3.9024,
+      "loss/crossentropy": 1.9702956676483154,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19528628885746002,
+      "step": 16256
+    },
+    {
+      "epoch": 0.32516,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.020475260416666665,
+      "learning_rate": 0.0001,
+      "loss": 4.0628,
+      "loss/crossentropy": 2.0042436718940735,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19861597567796707,
+      "step": 16258
+    },
+    {
+      "epoch": 0.3252,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.019573720296223958,
+      "learning_rate": 0.0001,
+      "loss": 4.1876,
+      "loss/crossentropy": 2.3544296622276306,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22385136783123016,
+      "step": 16260
+    },
+    {
+      "epoch": 0.32524,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.01942723592122396,
+      "learning_rate": 0.0001,
+      "loss": 4.1435,
+      "loss/crossentropy": 2.1482596397399902,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21662700176239014,
+      "step": 16262
+    },
+    {
+      "epoch": 0.32528,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.01395263671875,
+      "learning_rate": 0.0001,
+      "loss": 4.3441,
+      "loss/crossentropy": 2.243022322654724,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22729884833097458,
+      "step": 16264
+    },
+    {
+      "epoch": 0.32532,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.018656158447265626,
+      "learning_rate": 0.0001,
+      "loss": 3.9013,
+      "loss/crossentropy": 2.2499197721481323,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20558901876211166,
+      "step": 16266
+    },
+    {
+      "epoch": 0.32536,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.022907511393229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2313,
+      "loss/crossentropy": 1.6232356429100037,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18864237517118454,
+      "step": 16268
+    },
+    {
+      "epoch": 0.3254,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.02402521769205729,
+      "learning_rate": 0.0001,
+      "loss": 4.0435,
+      "loss/crossentropy": 1.559517502784729,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1704516038298607,
+      "step": 16270
+    },
+    {
+      "epoch": 0.32544,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.02484308878580729,
+      "learning_rate": 0.0001,
+      "loss": 3.9018,
+      "loss/crossentropy": 1.7346046566963196,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17380736768245697,
+      "step": 16272
+    },
+    {
+      "epoch": 0.32548,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.02237523396809896,
+      "learning_rate": 0.0001,
+      "loss": 4.2894,
+      "loss/crossentropy": 2.1451315879821777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21778792887926102,
+      "step": 16274
+    },
+    {
+      "epoch": 0.32552,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.022684733072916668,
+      "learning_rate": 0.0001,
+      "loss": 3.8737,
+      "loss/crossentropy": 1.8013367056846619,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18631768971681595,
+      "step": 16276
+    },
+    {
+      "epoch": 0.32556,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.025655110677083332,
+      "learning_rate": 0.0001,
+      "loss": 3.8573,
+      "loss/crossentropy": 1.8858280181884766,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1940329447388649,
+      "step": 16278
+    },
+    {
+      "epoch": 0.3256,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0259429931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1268,
+      "loss/crossentropy": 1.8475046157836914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18596403300762177,
+      "step": 16280
+    },
+    {
+      "epoch": 0.32564,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.023636881510416666,
+      "learning_rate": 0.0001,
+      "loss": 4.2602,
+      "loss/crossentropy": 2.2585566639900208,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22430463135242462,
+      "step": 16282
+    },
+    {
+      "epoch": 0.32568,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.015900675455729166,
+      "learning_rate": 0.0001,
+      "loss": 3.7224,
+      "loss/crossentropy": 1.7233783602714539,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17318468540906906,
+      "step": 16284
+    },
+    {
+      "epoch": 0.32572,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.022172037760416666,
+      "learning_rate": 0.0001,
+      "loss": 4.3042,
+      "loss/crossentropy": 2.174088716506958,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21329404413700104,
+      "step": 16286
+    },
+    {
+      "epoch": 0.32576,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.022345987955729167,
+      "learning_rate": 0.0001,
+      "loss": 3.8932,
+      "loss/crossentropy": 1.8016705513000488,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20395102351903915,
+      "step": 16288
+    },
+    {
+      "epoch": 0.3258,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.020182037353515626,
+      "learning_rate": 0.0001,
+      "loss": 3.8634,
+      "loss/crossentropy": 2.166901111602783,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21237251162528992,
+      "step": 16290
+    },
+    {
+      "epoch": 0.32584,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.014766184488932292,
+      "learning_rate": 0.0001,
+      "loss": 4.07,
+      "loss/crossentropy": 2.024773359298706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1973796784877777,
+      "step": 16292
+    },
+    {
+      "epoch": 0.32588,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.015466054280598959,
+      "learning_rate": 0.0001,
+      "loss": 4.173,
+      "loss/crossentropy": 2.457033157348633,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21896487474441528,
+      "step": 16294
+    },
+    {
+      "epoch": 0.32592,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.014180501302083334,
+      "learning_rate": 0.0001,
+      "loss": 4.11,
+      "loss/crossentropy": 2.4872595071792603,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2348102256655693,
+      "step": 16296
+    },
+    {
+      "epoch": 0.32596,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.014249420166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0756,
+      "loss/crossentropy": 2.110987067222595,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21079359203577042,
+      "step": 16298
+    },
+    {
+      "epoch": 0.326,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0133941650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2334,
+      "loss/crossentropy": 2.2166510820388794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19656993448734283,
+      "step": 16300
+    },
+    {
+      "epoch": 0.32604,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.008125813802083333,
+      "learning_rate": 0.0001,
+      "loss": 4.0212,
+      "loss/crossentropy": 2.014153838157654,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1956234946846962,
+      "step": 16302
+    },
+    {
+      "epoch": 0.32608,
+      "grad_norm": 2.671875,
+      "grad_norm_var": 0.03588231404622396,
+      "learning_rate": 0.0001,
+      "loss": 4.3045,
+      "loss/crossentropy": 2.298485517501831,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21710477769374847,
+      "step": 16304
+    },
+    {
+      "epoch": 0.32612,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.033599599202473955,
+      "learning_rate": 0.0001,
+      "loss": 4.3134,
+      "loss/crossentropy": 2.233761191368103,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21609390527009964,
+      "step": 16306
+    },
+    {
+      "epoch": 0.32616,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.03675918579101563,
+      "learning_rate": 0.0001,
+      "loss": 3.9141,
+      "loss/crossentropy": 1.9988956451416016,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1924634352326393,
+      "step": 16308
+    },
+    {
+      "epoch": 0.3262,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.035194651285807295,
+      "learning_rate": 0.0001,
+      "loss": 3.9726,
+      "loss/crossentropy": 1.9185590147972107,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20295168459415436,
+      "step": 16310
+    },
+    {
+      "epoch": 0.32624,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.03476155598958333,
+      "learning_rate": 0.0001,
+      "loss": 4.0805,
+      "loss/crossentropy": 2.219870448112488,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21425887942314148,
+      "step": 16312
+    },
+    {
+      "epoch": 0.32628,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.03612442016601562,
+      "learning_rate": 0.0001,
+      "loss": 4.5119,
+      "loss/crossentropy": 2.219307541847229,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22380203753709793,
+      "step": 16314
+    },
+    {
+      "epoch": 0.32632,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.03746515909830729,
+      "learning_rate": 0.0001,
+      "loss": 3.9652,
+      "loss/crossentropy": 1.7907224893569946,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18037232011556625,
+      "step": 16316
+    },
+    {
+      "epoch": 0.32636,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.03876927693684896,
+      "learning_rate": 0.0001,
+      "loss": 4.1337,
+      "loss/crossentropy": 1.9702708721160889,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18266448378562927,
+      "step": 16318
+    },
+    {
+      "epoch": 0.3264,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.012481435139973959,
+      "learning_rate": 0.0001,
+      "loss": 4.0551,
+      "loss/crossentropy": 2.1748844385147095,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20725325495004654,
+      "step": 16320
+    },
+    {
+      "epoch": 0.32644,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.0119781494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2882,
+      "loss/crossentropy": 2.2458006143569946,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21462788432836533,
+      "step": 16322
+    },
+    {
+      "epoch": 0.32648,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.009276326497395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1231,
+      "loss/crossentropy": 1.923665463924408,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.223393976688385,
+      "step": 16324
+    },
+    {
+      "epoch": 0.32652,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.010129547119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1643,
+      "loss/crossentropy": 2.36092209815979,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22316965460777283,
+      "step": 16326
+    },
+    {
+      "epoch": 0.32656,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.010456339518229166,
+      "learning_rate": 0.0001,
+      "loss": 4.0778,
+      "loss/crossentropy": 1.7765440344810486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1736016422510147,
+      "step": 16328
+    },
+    {
+      "epoch": 0.3266,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.006638336181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.4885,
+      "loss/crossentropy": 2.2588138580322266,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21762139350175858,
+      "step": 16330
+    },
+    {
+      "epoch": 0.32664,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.0064084370930989586,
+      "learning_rate": 0.0001,
+      "loss": 4.2476,
+      "loss/crossentropy": 2.177221417427063,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19594799727201462,
+      "step": 16332
+    },
+    {
+      "epoch": 0.32668,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0051422119140625,
+      "learning_rate": 0.0001,
+      "loss": 3.941,
+      "loss/crossentropy": 1.8424673676490784,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19562739878892899,
+      "step": 16334
+    },
+    {
+      "epoch": 0.32672,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.006473541259765625,
+      "learning_rate": 0.0001,
+      "loss": 3.861,
+      "loss/crossentropy": 1.8677524328231812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19708558171987534,
+      "step": 16336
+    },
+    {
+      "epoch": 0.32676,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.007081858317057292,
+      "learning_rate": 0.0001,
+      "loss": 3.9222,
+      "loss/crossentropy": 1.9559081196784973,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19585052132606506,
+      "step": 16338
+    },
+    {
+      "epoch": 0.3268,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.006892649332682291,
+      "learning_rate": 0.0001,
+      "loss": 4.144,
+      "loss/crossentropy": 1.979565978050232,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2045053392648697,
+      "step": 16340
+    },
+    {
+      "epoch": 0.32684,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.00714111328125,
+      "learning_rate": 0.0001,
+      "loss": 4.2517,
+      "loss/crossentropy": 2.296669840812683,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.207671657204628,
+      "step": 16342
+    },
+    {
+      "epoch": 0.32688,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0065305074055989586,
+      "learning_rate": 0.0001,
+      "loss": 4.0279,
+      "loss/crossentropy": 1.9962339401245117,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19833384454250336,
+      "step": 16344
+    },
+    {
+      "epoch": 0.32692,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.006078084309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1779,
+      "loss/crossentropy": 2.020018517971039,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19721734523773193,
+      "step": 16346
+    },
+    {
+      "epoch": 0.32696,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.006815592447916667,
+      "learning_rate": 0.0001,
+      "loss": 3.7614,
+      "loss/crossentropy": 2.112026810646057,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20460353046655655,
+      "step": 16348
+    },
+    {
+      "epoch": 0.327,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.006540679931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1509,
+      "loss/crossentropy": 2.2069387435913086,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2233232483267784,
+      "step": 16350
+    },
+    {
+      "epoch": 0.32704,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.004792277018229167,
+      "learning_rate": 0.0001,
+      "loss": 3.8675,
+      "loss/crossentropy": 1.9365113377571106,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19625376164913177,
+      "step": 16352
+    },
+    {
+      "epoch": 0.32708,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0052154541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1831,
+      "loss/crossentropy": 2.2316179871559143,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22712747752666473,
+      "step": 16354
+    },
+    {
+      "epoch": 0.32712,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.035672760009765624,
+      "learning_rate": 0.0001,
+      "loss": 3.9148,
+      "loss/crossentropy": 1.801742434501648,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1831187978386879,
+      "step": 16356
+    },
+    {
+      "epoch": 0.32716,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.03590087890625,
+      "learning_rate": 0.0001,
+      "loss": 3.9819,
+      "loss/crossentropy": 2.157936453819275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21520215272903442,
+      "step": 16358
+    },
+    {
+      "epoch": 0.3272,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.035835520426432295,
+      "learning_rate": 0.0001,
+      "loss": 4.0743,
+      "loss/crossentropy": 1.6054654717445374,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1769917830824852,
+      "step": 16360
+    },
+    {
+      "epoch": 0.32724,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.039613596598307294,
+      "learning_rate": 0.0001,
+      "loss": 4.0751,
+      "loss/crossentropy": 2.0573782324790955,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22550390660762787,
+      "step": 16362
+    },
+    {
+      "epoch": 0.32728,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.035131581624348956,
+      "learning_rate": 0.0001,
+      "loss": 4.281,
+      "loss/crossentropy": 2.0643117427825928,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21060281991958618,
+      "step": 16364
+    },
+    {
+      "epoch": 0.32732,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.042525227864583334,
+      "learning_rate": 0.0001,
+      "loss": 4.0884,
+      "loss/crossentropy": 2.1980225443840027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19246292114257812,
+      "step": 16366
+    },
+    {
+      "epoch": 0.32736,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.042577107747395836,
+      "learning_rate": 0.0001,
+      "loss": 4.05,
+      "loss/crossentropy": 1.8917307257652283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17911095917224884,
+      "step": 16368
+    },
+    {
+      "epoch": 0.3274,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0429595947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1415,
+      "loss/crossentropy": 2.0520957708358765,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19265015423297882,
+      "step": 16370
+    },
+    {
+      "epoch": 0.32744,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.019510904947916668,
+      "learning_rate": 0.0001,
+      "loss": 4.3026,
+      "loss/crossentropy": 2.4500341415405273,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21713324636220932,
+      "step": 16372
+    },
+    {
+      "epoch": 0.32748,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.02032648722330729,
+      "learning_rate": 0.0001,
+      "loss": 4.0754,
+      "loss/crossentropy": 1.8100037574768066,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16967355459928513,
+      "step": 16374
+    },
+    {
+      "epoch": 0.32752,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.021174875895182292,
+      "learning_rate": 0.0001,
+      "loss": 4.0244,
+      "loss/crossentropy": 2.092597723007202,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1998964473605156,
+      "step": 16376
+    },
+    {
+      "epoch": 0.32756,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.016076405843098957,
+      "learning_rate": 0.0001,
+      "loss": 4.3627,
+      "loss/crossentropy": 2.151292622089386,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19120946526527405,
+      "step": 16378
+    },
+    {
+      "epoch": 0.3276,
+      "grad_norm": 1.796875,
+      "grad_norm_var": 0.020906575520833335,
+      "learning_rate": 0.0001,
+      "loss": 3.6949,
+      "loss/crossentropy": 1.7384315729141235,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18261852860450745,
+      "step": 16380
+    },
+    {
+      "epoch": 0.32764,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.019147745768229165,
+      "learning_rate": 0.0001,
+      "loss": 3.9341,
+      "loss/crossentropy": 1.662436068058014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16703163087368011,
+      "step": 16382
+    },
+    {
+      "epoch": 0.32768,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.018871053059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3202,
+      "loss/crossentropy": 2.2057151794433594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20989537239074707,
+      "step": 16384
+    },
+    {
+      "epoch": 0.32772,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.068359375,
+      "learning_rate": 0.0001,
+      "loss": 4.3033,
+      "loss/crossentropy": 2.1689382791519165,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20804189145565033,
+      "step": 16386
+    },
+    {
+      "epoch": 0.32776,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.06450169881184896,
+      "learning_rate": 0.0001,
+      "loss": 4.2241,
+      "loss/crossentropy": 2.3121412992477417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20552029460668564,
+      "step": 16388
+    },
+    {
+      "epoch": 0.3278,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.06318333943684896,
+      "learning_rate": 0.0001,
+      "loss": 4.1396,
+      "loss/crossentropy": 2.0784353017807007,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20427466928958893,
+      "step": 16390
+    },
+    {
+      "epoch": 0.32784,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.06282145182291667,
+      "learning_rate": 0.0001,
+      "loss": 4.1132,
+      "loss/crossentropy": 2.147130608558655,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2181989550590515,
+      "step": 16392
+    },
+    {
+      "epoch": 0.32788,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.06468505859375,
+      "learning_rate": 0.0001,
+      "loss": 3.812,
+      "loss/crossentropy": 1.7986091375350952,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18294771760702133,
+      "step": 16394
+    },
+    {
+      "epoch": 0.32792,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.059004720052083334,
+      "learning_rate": 0.0001,
+      "loss": 4.037,
+      "loss/crossentropy": 2.0563793778419495,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19180169701576233,
+      "step": 16396
+    },
+    {
+      "epoch": 0.32796,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.05506591796875,
+      "learning_rate": 0.0001,
+      "loss": 4.3111,
+      "loss/crossentropy": 2.462133765220642,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21563740074634552,
+      "step": 16398
+    },
+    {
+      "epoch": 0.328,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.059427897135416664,
+      "learning_rate": 0.0001,
+      "loss": 3.9743,
+      "loss/crossentropy": 1.8900938630104065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1845797374844551,
+      "step": 16400
+    },
+    {
+      "epoch": 0.32804,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.0195068359375,
+      "learning_rate": 0.0001,
+      "loss": 4.1653,
+      "loss/crossentropy": 2.0627527832984924,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20268447697162628,
+      "step": 16402
+    },
+    {
+      "epoch": 0.32808,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.030147043863932292,
+      "learning_rate": 0.0001,
+      "loss": 4.3042,
+      "loss/crossentropy": 2.058929443359375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19984640181064606,
+      "step": 16404
+    },
+    {
+      "epoch": 0.32812,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.030564117431640624,
+      "learning_rate": 0.0001,
+      "loss": 4.1162,
+      "loss/crossentropy": 1.816510558128357,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1754368245601654,
+      "step": 16406
+    },
+    {
+      "epoch": 0.32816,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0302886962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.263,
+      "loss/crossentropy": 1.9079806208610535,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18917685747146606,
+      "step": 16408
+    },
+    {
+      "epoch": 0.3282,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.028612263997395835,
+      "learning_rate": 0.0001,
+      "loss": 3.9264,
+      "loss/crossentropy": 2.0913302898406982,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2088220864534378,
+      "step": 16410
+    },
+    {
+      "epoch": 0.32824,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.028547159830729165,
+      "learning_rate": 0.0001,
+      "loss": 3.9029,
+      "loss/crossentropy": 2.1609301567077637,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2010297030210495,
+      "step": 16412
+    },
+    {
+      "epoch": 0.32828,
+      "grad_norm": 3.078125,
+      "grad_norm_var": 0.0982666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2563,
+      "loss/crossentropy": 1.8449677228927612,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23620254546403885,
+      "step": 16414
+    },
+    {
+      "epoch": 0.32832,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.09134699503580729,
+      "learning_rate": 0.0001,
+      "loss": 4.1053,
+      "loss/crossentropy": 2.059410274028778,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20259736478328705,
+      "step": 16416
+    },
+    {
+      "epoch": 0.32836,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.08761571248372396,
+      "learning_rate": 0.0001,
+      "loss": 3.8718,
+      "loss/crossentropy": 1.6095005869865417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17026344686746597,
+      "step": 16418
+    },
+    {
+      "epoch": 0.3284,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.07932510375976562,
+      "learning_rate": 0.0001,
+      "loss": 4.2623,
+      "loss/crossentropy": 2.200170636177063,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2259170413017273,
+      "step": 16420
+    },
+    {
+      "epoch": 0.32844,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.08116226196289063,
+      "learning_rate": 0.0001,
+      "loss": 4.0724,
+      "loss/crossentropy": 1.970030963420868,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19053932279348373,
+      "step": 16422
+    },
+    {
+      "epoch": 0.32848,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.08263753255208334,
+      "learning_rate": 0.0001,
+      "loss": 4.106,
+      "loss/crossentropy": 2.2675901651382446,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22277145832777023,
+      "step": 16424
+    },
+    {
+      "epoch": 0.32852,
+      "grad_norm": 1.78125,
+      "grad_norm_var": 0.08910725911458334,
+      "learning_rate": 0.0001,
+      "loss": 3.723,
+      "loss/crossentropy": 1.6446372866630554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15234287828207016,
+      "step": 16426
+    },
+    {
+      "epoch": 0.32856,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0875689188639323,
+      "learning_rate": 0.0001,
+      "loss": 4.3421,
+      "loss/crossentropy": 2.350424647331238,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2190476581454277,
+      "step": 16428
+    },
+    {
+      "epoch": 0.3286,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.009279123942057292,
+      "learning_rate": 0.0001,
+      "loss": 3.9663,
+      "loss/crossentropy": 1.8809763193130493,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19404269009828568,
+      "step": 16430
+    },
+    {
+      "epoch": 0.32864,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.008257802327473958,
+      "learning_rate": 0.0001,
+      "loss": 4.1731,
+      "loss/crossentropy": 2.1159849166870117,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21232837438583374,
+      "step": 16432
+    },
+    {
+      "epoch": 0.32868,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.02133763631184896,
+      "learning_rate": 0.0001,
+      "loss": 4.2875,
+      "loss/crossentropy": 2.3252066373825073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24867044389247894,
+      "step": 16434
+    },
+    {
+      "epoch": 0.32872,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.020186360677083334,
+      "learning_rate": 0.0001,
+      "loss": 4.2462,
+      "loss/crossentropy": 2.0642696619033813,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18409788608551025,
+      "step": 16436
+    },
+    {
+      "epoch": 0.32876,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.022725423177083332,
+      "learning_rate": 0.0001,
+      "loss": 4.3832,
+      "loss/crossentropy": 2.359580874443054,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.218842551112175,
+      "step": 16438
+    },
+    {
+      "epoch": 0.3288,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.02264378865559896,
+      "learning_rate": 0.0001,
+      "loss": 3.7996,
+      "loss/crossentropy": 2.0959436893463135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1993689462542534,
+      "step": 16440
+    },
+    {
+      "epoch": 0.32884,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.017679595947265626,
+      "learning_rate": 0.0001,
+      "loss": 4.0838,
+      "loss/crossentropy": 2.1507667303085327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20800111442804337,
+      "step": 16442
+    },
+    {
+      "epoch": 0.32888,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.020005035400390624,
+      "learning_rate": 0.0001,
+      "loss": 3.9428,
+      "loss/crossentropy": 1.8771533370018005,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20476838946342468,
+      "step": 16444
+    },
+    {
+      "epoch": 0.32892,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.020005035400390624,
+      "learning_rate": 0.0001,
+      "loss": 4.2689,
+      "loss/crossentropy": 2.365285038948059,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22512932121753693,
+      "step": 16446
+    },
+    {
+      "epoch": 0.32896,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.019701131184895835,
+      "learning_rate": 0.0001,
+      "loss": 4.2928,
+      "loss/crossentropy": 2.2584418058395386,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21187593042850494,
+      "step": 16448
+    },
+    {
+      "epoch": 0.329,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.010270182291666667,
+      "learning_rate": 0.0001,
+      "loss": 3.8186,
+      "loss/crossentropy": 1.9227403402328491,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1945207566022873,
+      "step": 16450
+    },
+    {
+      "epoch": 0.32904,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.011897786458333334,
+      "learning_rate": 0.0001,
+      "loss": 3.8512,
+      "loss/crossentropy": 2.062865734100342,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2064652517437935,
+      "step": 16452
+    },
+    {
+      "epoch": 0.32908,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.007505035400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0493,
+      "loss/crossentropy": 1.98399817943573,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22567245364189148,
+      "step": 16454
+    },
+    {
+      "epoch": 0.32912,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.007428995768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2888,
+      "loss/crossentropy": 2.0655510425567627,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20936425030231476,
+      "step": 16456
+    },
+    {
+      "epoch": 0.32916,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.01739501953125,
+      "learning_rate": 0.0001,
+      "loss": 4.1486,
+      "loss/crossentropy": 1.8360095024108887,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18788952380418777,
+      "step": 16458
+    },
+    {
+      "epoch": 0.3292,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.01672948201497396,
+      "learning_rate": 0.0001,
+      "loss": 4.2265,
+      "loss/crossentropy": 2.1088568568229675,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20265556871891022,
+      "step": 16460
+    },
+    {
+      "epoch": 0.32924,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.017601521809895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1004,
+      "loss/crossentropy": 2.3746429681777954,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21173583716154099,
+      "step": 16462
+    },
+    {
+      "epoch": 0.32928,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.01946996053059896,
+      "learning_rate": 0.0001,
+      "loss": 3.9577,
+      "loss/crossentropy": 2.0129969716072083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18141476809978485,
+      "step": 16464
+    },
+    {
+      "epoch": 0.32932,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.019160970052083334,
+      "learning_rate": 0.0001,
+      "loss": 4.2122,
+      "loss/crossentropy": 2.385707139968872,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22456685453653336,
+      "step": 16466
+    },
+    {
+      "epoch": 0.32936,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.015958658854166665,
+      "learning_rate": 0.0001,
+      "loss": 4.183,
+      "loss/crossentropy": 2.466804623603821,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23166514188051224,
+      "step": 16468
+    },
+    {
+      "epoch": 0.3294,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.018700154622395833,
+      "learning_rate": 0.0001,
+      "loss": 3.8552,
+      "loss/crossentropy": 1.9182489514350891,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1918247565627098,
+      "step": 16470
+    },
+    {
+      "epoch": 0.32944,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.021993001302083332,
+      "learning_rate": 0.0001,
+      "loss": 4.4364,
+      "loss/crossentropy": 2.27492892742157,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2237030565738678,
+      "step": 16472
+    },
+    {
+      "epoch": 0.32948,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0144927978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0022,
+      "loss/crossentropy": 1.856327474117279,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17718566209077835,
+      "step": 16474
+    },
+    {
+      "epoch": 0.32952,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.0145263671875,
+      "learning_rate": 0.0001,
+      "loss": 3.9994,
+      "loss/crossentropy": 2.014274477958679,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20536774396896362,
+      "step": 16476
+    },
+    {
+      "epoch": 0.32956,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.016383616129557292,
+      "learning_rate": 0.0001,
+      "loss": 4.4606,
+      "loss/crossentropy": 2.098844289779663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18847080320119858,
+      "step": 16478
+    },
+    {
+      "epoch": 0.3296,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.014647420247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1978,
+      "loss/crossentropy": 2.0402814149856567,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1929381936788559,
+      "step": 16480
+    },
+    {
+      "epoch": 0.32964,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.013315582275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2737,
+      "loss/crossentropy": 2.413077712059021,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2426547110080719,
+      "step": 16482
+    },
+    {
+      "epoch": 0.32968,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.013401031494140625,
+      "learning_rate": 0.0001,
+      "loss": 3.9544,
+      "loss/crossentropy": 2.081916332244873,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2190306931734085,
+      "step": 16484
+    },
+    {
+      "epoch": 0.32972,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.010765584309895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1093,
+      "loss/crossentropy": 2.1017117500305176,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19809769093990326,
+      "step": 16486
+    },
+    {
+      "epoch": 0.32976,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.008139801025390626,
+      "learning_rate": 0.0001,
+      "loss": 3.9585,
+      "loss/crossentropy": 2.289687156677246,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21503399312496185,
+      "step": 16488
+    },
+    {
+      "epoch": 0.3298,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.008642323811848958,
+      "learning_rate": 0.0001,
+      "loss": 4.1952,
+      "loss/crossentropy": 2.0613157749176025,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20428171753883362,
+      "step": 16490
+    },
+    {
+      "epoch": 0.32984,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0073964436848958336,
+      "learning_rate": 0.0001,
+      "loss": 4.3191,
+      "loss/crossentropy": 2.1484400033950806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20850211381912231,
+      "step": 16492
+    },
+    {
+      "epoch": 0.32988,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.0072629292805989586,
+      "learning_rate": 0.0001,
+      "loss": 3.875,
+      "loss/crossentropy": 1.6853107213974,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1736336275935173,
+      "step": 16494
+    },
+    {
+      "epoch": 0.32992,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.007389068603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2047,
+      "loss/crossentropy": 2.323481321334839,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22742202132940292,
+      "step": 16496
+    },
+    {
+      "epoch": 0.32996,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.00738525390625,
+      "learning_rate": 0.0001,
+      "loss": 4.124,
+      "loss/crossentropy": 2.1196956038475037,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19792834669351578,
+      "step": 16498
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.007417805989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.0013,
+      "loss/crossentropy": 1.947241187095642,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21067001670598984,
+      "step": 16500
+    },
+    {
+      "epoch": 0.33004,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.00758056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0631,
+      "loss/crossentropy": 2.256502628326416,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22300700098276138,
+      "step": 16502
+    },
+    {
+      "epoch": 0.33008,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.007574208577473958,
+      "learning_rate": 0.0001,
+      "loss": 3.7874,
+      "loss/crossentropy": 1.7947281002998352,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1860191375017166,
+      "step": 16504
+    },
+    {
+      "epoch": 0.33012,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.007995351155598959,
+      "learning_rate": 0.0001,
+      "loss": 4.0189,
+      "loss/crossentropy": 1.9331985712051392,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20170047879219055,
+      "step": 16506
+    },
+    {
+      "epoch": 0.33016,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.011030832926432291,
+      "learning_rate": 0.0001,
+      "loss": 4.2148,
+      "loss/crossentropy": 2.2274327278137207,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1977931559085846,
+      "step": 16508
+    },
+    {
+      "epoch": 0.3302,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007884724934895834,
+      "learning_rate": 0.0001,
+      "loss": 4.15,
+      "loss/crossentropy": 1.8204082250595093,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18195096403360367,
+      "step": 16510
+    },
+    {
+      "epoch": 0.33024,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.009633127848307292,
+      "learning_rate": 0.0001,
+      "loss": 4.2363,
+      "loss/crossentropy": 2.308950901031494,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22067535668611526,
+      "step": 16512
+    },
+    {
+      "epoch": 0.33028,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.014794921875,
+      "learning_rate": 0.0001,
+      "loss": 4.3895,
+      "loss/crossentropy": 2.1682082414627075,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2021712213754654,
+      "step": 16514
+    },
+    {
+      "epoch": 0.33032,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.016047159830729168,
+      "learning_rate": 0.0001,
+      "loss": 3.9213,
+      "loss/crossentropy": 1.9669402837753296,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19773626327514648,
+      "step": 16516
+    },
+    {
+      "epoch": 0.33036,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.016047159830729168,
+      "learning_rate": 0.0001,
+      "loss": 4.1723,
+      "loss/crossentropy": 2.0478034019470215,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19190094619989395,
+      "step": 16518
+    },
+    {
+      "epoch": 0.3304,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.014127349853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1399,
+      "loss/crossentropy": 1.8262990713119507,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19281092286109924,
+      "step": 16520
+    },
+    {
+      "epoch": 0.33044,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0139801025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.385,
+      "loss/crossentropy": 2.3943055868148804,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2286393940448761,
+      "step": 16522
+    },
+    {
+      "epoch": 0.33048,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0124664306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1117,
+      "loss/crossentropy": 1.7113690972328186,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20282406359910965,
+      "step": 16524
+    },
+    {
+      "epoch": 0.33052,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.012924957275390624,
+      "learning_rate": 0.0001,
+      "loss": 3.9182,
+      "loss/crossentropy": 1.9077526926994324,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18814775347709656,
+      "step": 16526
+    },
+    {
+      "epoch": 0.33056,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0106597900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3404,
+      "loss/crossentropy": 1.9100408554077148,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1897171437740326,
+      "step": 16528
+    },
+    {
+      "epoch": 0.3306,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.0066314697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1403,
+      "loss/crossentropy": 1.8145674467086792,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18221604079008102,
+      "step": 16530
+    },
+    {
+      "epoch": 0.33064,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0085113525390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0647,
+      "loss/crossentropy": 2.017480492591858,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19173581898212433,
+      "step": 16532
+    },
+    {
+      "epoch": 0.33068,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.008739980061848958,
+      "learning_rate": 0.0001,
+      "loss": 4.1918,
+      "loss/crossentropy": 1.8937708139419556,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1974407583475113,
+      "step": 16534
+    },
+    {
+      "epoch": 0.33072,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.012664540608723959,
+      "learning_rate": 0.0001,
+      "loss": 4.1354,
+      "loss/crossentropy": 1.824431598186493,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18114721775054932,
+      "step": 16536
+    },
+    {
+      "epoch": 0.33076,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.018464152018229166,
+      "learning_rate": 0.0001,
+      "loss": 4.034,
+      "loss/crossentropy": 1.6330693364143372,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17686167359352112,
+      "step": 16538
+    },
+    {
+      "epoch": 0.3308,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.028595987955729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2359,
+      "loss/crossentropy": 2.1948810815811157,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21977168321609497,
+      "step": 16540
+    },
+    {
+      "epoch": 0.33084,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.03220926920572917,
+      "learning_rate": 0.0001,
+      "loss": 4.1942,
+      "loss/crossentropy": 2.1222333908081055,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19481626898050308,
+      "step": 16542
+    },
+    {
+      "epoch": 0.33088,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.03241780598958333,
+      "learning_rate": 0.0001,
+      "loss": 3.9908,
+      "loss/crossentropy": 1.9703376293182373,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20912524312734604,
+      "step": 16544
+    },
+    {
+      "epoch": 0.33092,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.03206965128580729,
+      "learning_rate": 0.0001,
+      "loss": 4.0774,
+      "loss/crossentropy": 2.2693361043930054,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2137979120016098,
+      "step": 16546
+    },
+    {
+      "epoch": 0.33096,
+      "grad_norm": 3.703125,
+      "grad_norm_var": 0.19548746744791667,
+      "learning_rate": 0.0001,
+      "loss": 4.2372,
+      "loss/crossentropy": 2.029613673686981,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21167294681072235,
+      "step": 16548
+    },
+    {
+      "epoch": 0.331,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.19552586873372396,
+      "learning_rate": 0.0001,
+      "loss": 3.7223,
+      "loss/crossentropy": 2.0102853775024414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2100009247660637,
+      "step": 16550
+    },
+    {
+      "epoch": 0.33104,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.19806722005208333,
+      "learning_rate": 0.0001,
+      "loss": 4.6058,
+      "loss/crossentropy": 2.1570287942886353,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21619540452957153,
+      "step": 16552
+    },
+    {
+      "epoch": 0.33108,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.20053609212239584,
+      "learning_rate": 0.0001,
+      "loss": 4.0271,
+      "loss/crossentropy": 1.9157934188842773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.193254753947258,
+      "step": 16554
+    },
+    {
+      "epoch": 0.33112,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.2052642822265625,
+      "learning_rate": 0.0001,
+      "loss": 3.6593,
+      "loss/crossentropy": 1.7680367827415466,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19283122569322586,
+      "step": 16556
+    },
+    {
+      "epoch": 0.33116,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.2058013916015625,
+      "learning_rate": 0.0001,
+      "loss": 3.8828,
+      "loss/crossentropy": 1.8805989623069763,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20299813151359558,
+      "step": 16558
+    },
+    {
+      "epoch": 0.3312,
+      "grad_norm": 1.71875,
+      "grad_norm_var": 0.21938069661458334,
+      "learning_rate": 0.0001,
+      "loss": 3.7657,
+      "loss/crossentropy": 1.9145240187644958,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18402665108442307,
+      "step": 16560
+    },
+    {
+      "epoch": 0.33124,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.2167144775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1984,
+      "loss/crossentropy": 1.7958417534828186,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17038261890411377,
+      "step": 16562
+    },
+    {
+      "epoch": 0.33128,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.038919830322265626,
+      "learning_rate": 0.0001,
+      "loss": 3.8916,
+      "loss/crossentropy": 2.0782148838043213,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21738532930612564,
+      "step": 16564
+    },
+    {
+      "epoch": 0.33132,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.03910903930664063,
+      "learning_rate": 0.0001,
+      "loss": 4.3098,
+      "loss/crossentropy": 1.9767170548439026,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2127690687775612,
+      "step": 16566
+    },
+    {
+      "epoch": 0.33136,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.01185302734375,
+      "learning_rate": 0.0001,
+      "loss": 3.9133,
+      "loss/crossentropy": 2.086383044719696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19811449944972992,
+      "step": 16568
+    },
+    {
+      "epoch": 0.3314,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.011991119384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0358,
+      "loss/crossentropy": 2.0471088886260986,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20480120182037354,
+      "step": 16570
+    },
+    {
+      "epoch": 0.33144,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012485504150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1097,
+      "loss/crossentropy": 2.2637689113616943,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21112071722745895,
+      "step": 16572
+    },
+    {
+      "epoch": 0.33148,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.012018839518229166,
+      "learning_rate": 0.0001,
+      "loss": 4.023,
+      "loss/crossentropy": 1.8035425543785095,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18357180804014206,
+      "step": 16574
+    },
+    {
+      "epoch": 0.33152,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.007189687093098958,
+      "learning_rate": 0.0001,
+      "loss": 3.7453,
+      "loss/crossentropy": 1.647614598274231,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17512448877096176,
+      "step": 16576
+    },
+    {
+      "epoch": 0.33156,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0063873291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0452,
+      "loss/crossentropy": 2.0647078156471252,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21568534523248672,
+      "step": 16578
+    },
+    {
+      "epoch": 0.3316,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.00615234375,
+      "learning_rate": 0.0001,
+      "loss": 3.8656,
+      "loss/crossentropy": 1.7758954763412476,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1891157627105713,
+      "step": 16580
+    },
+    {
+      "epoch": 0.33164,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0060618082682291664,
+      "learning_rate": 0.0001,
+      "loss": 4.1022,
+      "loss/crossentropy": 2.0523850321769714,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21198877692222595,
+      "step": 16582
+    },
+    {
+      "epoch": 0.33168,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.006725819905598959,
+      "learning_rate": 0.0001,
+      "loss": 4.0584,
+      "loss/crossentropy": 1.99459570646286,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20555292069911957,
+      "step": 16584
+    },
+    {
+      "epoch": 0.33172,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.007771809895833333,
+      "learning_rate": 0.0001,
+      "loss": 4.4979,
+      "loss/crossentropy": 2.353522777557373,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23193839937448502,
+      "step": 16586
+    },
+    {
+      "epoch": 0.33176,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.020970662434895832,
+      "learning_rate": 0.0001,
+      "loss": 4.1303,
+      "loss/crossentropy": 2.0257768630981445,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20071294903755188,
+      "step": 16588
+    },
+    {
+      "epoch": 0.3318,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.020467122395833332,
+      "learning_rate": 0.0001,
+      "loss": 4.2135,
+      "loss/crossentropy": 2.2084985971450806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20876871049404144,
+      "step": 16590
+    },
+    {
+      "epoch": 0.33184,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.019791412353515624,
+      "learning_rate": 0.0001,
+      "loss": 4.3275,
+      "loss/crossentropy": 1.9118791818618774,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24547216296195984,
+      "step": 16592
+    },
+    {
+      "epoch": 0.33188,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.022188313802083335,
+      "learning_rate": 0.0001,
+      "loss": 4.0069,
+      "loss/crossentropy": 2.184453248977661,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1988314613699913,
+      "step": 16594
+    },
+    {
+      "epoch": 0.33192,
+      "grad_norm": 1.71875,
+      "grad_norm_var": 0.02801488240559896,
+      "learning_rate": 0.0001,
+      "loss": 3.8711,
+      "loss/crossentropy": 2.2105953097343445,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20877376198768616,
+      "step": 16596
+    },
+    {
+      "epoch": 0.33196,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.028586578369140626,
+      "learning_rate": 0.0001,
+      "loss": 4.2969,
+      "loss/crossentropy": 2.028349459171295,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20705370604991913,
+      "step": 16598
+    },
+    {
+      "epoch": 0.332,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.02982177734375,
+      "learning_rate": 0.0001,
+      "loss": 3.8968,
+      "loss/crossentropy": 2.0778703689575195,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19761346280574799,
+      "step": 16600
+    },
+    {
+      "epoch": 0.33204,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.029670206705729167,
+      "learning_rate": 0.0001,
+      "loss": 3.9786,
+      "loss/crossentropy": 2.046931743621826,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21325716376304626,
+      "step": 16602
+    },
+    {
+      "epoch": 0.33208,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.011486562093098958,
+      "learning_rate": 0.0001,
+      "loss": 4.1525,
+      "loss/crossentropy": 1.7907955050468445,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1822371482849121,
+      "step": 16604
+    },
+    {
+      "epoch": 0.33212,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.012108357747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1586,
+      "loss/crossentropy": 1.7620025277137756,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19196610152721405,
+      "step": 16606
+    },
+    {
+      "epoch": 0.33216,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010282135009765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9175,
+      "loss/crossentropy": 1.6695470213890076,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16451346129179,
+      "step": 16608
+    },
+    {
+      "epoch": 0.3322,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.012717437744140626,
+      "learning_rate": 0.0001,
+      "loss": 4.0304,
+      "loss/crossentropy": 2.2249897718429565,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21898073703050613,
+      "step": 16610
+    },
+    {
+      "epoch": 0.33224,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.009474436442057291,
+      "learning_rate": 0.0001,
+      "loss": 4.1589,
+      "loss/crossentropy": 2.0861976146698,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1993977203965187,
+      "step": 16612
+    },
+    {
+      "epoch": 0.33228,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.009527333577473958,
+      "learning_rate": 0.0001,
+      "loss": 4.1516,
+      "loss/crossentropy": 2.195378541946411,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2103184312582016,
+      "step": 16614
+    },
+    {
+      "epoch": 0.33232,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.009478505452473958,
+      "learning_rate": 0.0001,
+      "loss": 3.6947,
+      "loss/crossentropy": 1.9064915180206299,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17310689389705658,
+      "step": 16616
+    },
+    {
+      "epoch": 0.33236,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.009452311197916667,
+      "learning_rate": 0.0001,
+      "loss": 3.9698,
+      "loss/crossentropy": 1.7967488169670105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18099892884492874,
+      "step": 16618
+    },
+    {
+      "epoch": 0.3324,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.012565104166666667,
+      "learning_rate": 0.0001,
+      "loss": 4.1319,
+      "loss/crossentropy": 2.0776742696762085,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2005431056022644,
+      "step": 16620
+    },
+    {
+      "epoch": 0.33244,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0109375,
+      "learning_rate": 0.0001,
+      "loss": 4.1881,
+      "loss/crossentropy": 2.0590518712997437,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20676207542419434,
+      "step": 16622
+    },
+    {
+      "epoch": 0.33248,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.011628977457682292,
+      "learning_rate": 0.0001,
+      "loss": 4.2067,
+      "loss/crossentropy": 1.8861089944839478,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1860756129026413,
+      "step": 16624
+    },
+    {
+      "epoch": 0.33252,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.009308878580729167,
+      "learning_rate": 0.0001,
+      "loss": 3.9757,
+      "loss/crossentropy": 2.4207329750061035,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22214040160179138,
+      "step": 16626
+    },
+    {
+      "epoch": 0.33256,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.009295399983723958,
+      "learning_rate": 0.0001,
+      "loss": 3.934,
+      "loss/crossentropy": 2.089089274406433,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21007104218006134,
+      "step": 16628
+    },
+    {
+      "epoch": 0.3326,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.011380767822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3451,
+      "loss/crossentropy": 2.357658624649048,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22044895589351654,
+      "step": 16630
+    },
+    {
+      "epoch": 0.33264,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.010223134358723959,
+      "learning_rate": 0.0001,
+      "loss": 3.9142,
+      "loss/crossentropy": 2.1036278009414673,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2003040835261345,
+      "step": 16632
+    },
+    {
+      "epoch": 0.33268,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.010749308268229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0802,
+      "loss/crossentropy": 2.202688694000244,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21549009531736374,
+      "step": 16634
+    },
+    {
+      "epoch": 0.33272,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.008642323811848958,
+      "learning_rate": 0.0001,
+      "loss": 4.1642,
+      "loss/crossentropy": 2.261624753475189,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2209767997264862,
+      "step": 16636
+    },
+    {
+      "epoch": 0.33276,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.013570149739583334,
+      "learning_rate": 0.0001,
+      "loss": 4.0556,
+      "loss/crossentropy": 1.7452040910720825,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18558355420827866,
+      "step": 16638
+    },
+    {
+      "epoch": 0.3328,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.012797037760416666,
+      "learning_rate": 0.0001,
+      "loss": 3.9364,
+      "loss/crossentropy": 1.9910151362419128,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2115819901227951,
+      "step": 16640
+    },
+    {
+      "epoch": 0.33284,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.014583079020182292,
+      "learning_rate": 0.0001,
+      "loss": 3.9412,
+      "loss/crossentropy": 2.0230116844177246,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2001063972711563,
+      "step": 16642
+    },
+    {
+      "epoch": 0.33288,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.015818023681640626,
+      "learning_rate": 0.0001,
+      "loss": 4.0476,
+      "loss/crossentropy": 1.9575697183609009,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20243427157402039,
+      "step": 16644
+    },
+    {
+      "epoch": 0.33292,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.012890370686848958,
+      "learning_rate": 0.0001,
+      "loss": 4.052,
+      "loss/crossentropy": 1.9650630354881287,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.203716441988945,
+      "step": 16646
+    },
+    {
+      "epoch": 0.33296,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.0132476806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1345,
+      "loss/crossentropy": 1.497282326221466,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15893913060426712,
+      "step": 16648
+    },
+    {
+      "epoch": 0.333,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.014491526285807292,
+      "learning_rate": 0.0001,
+      "loss": 4.1273,
+      "loss/crossentropy": 1.8017843961715698,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18981081247329712,
+      "step": 16650
+    },
+    {
+      "epoch": 0.33304,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.014296213785807291,
+      "learning_rate": 0.0001,
+      "loss": 4.07,
+      "loss/crossentropy": 2.089366614818573,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2076069712638855,
+      "step": 16652
+    },
+    {
+      "epoch": 0.33308,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.008695475260416667,
+      "learning_rate": 0.0001,
+      "loss": 3.9662,
+      "loss/crossentropy": 1.9445012211799622,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18241490423679352,
+      "step": 16654
+    },
+    {
+      "epoch": 0.33312,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.008740234375,
+      "learning_rate": 0.0001,
+      "loss": 3.9922,
+      "loss/crossentropy": 1.844423532485962,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18664997816085815,
+      "step": 16656
+    },
+    {
+      "epoch": 0.33316,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.007085927327473958,
+      "learning_rate": 0.0001,
+      "loss": 4.2221,
+      "loss/crossentropy": 2.1095730662345886,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21053151786327362,
+      "step": 16658
+    },
+    {
+      "epoch": 0.3332,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.007477823893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.255,
+      "loss/crossentropy": 2.139304041862488,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19131766259670258,
+      "step": 16660
+    },
+    {
+      "epoch": 0.33324,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.007450103759765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9358,
+      "loss/crossentropy": 1.8872849345207214,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19200573861598969,
+      "step": 16662
+    },
+    {
+      "epoch": 0.33328,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.006695302327473959,
+      "learning_rate": 0.0001,
+      "loss": 4.0939,
+      "loss/crossentropy": 2.488931655883789,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24411997199058533,
+      "step": 16664
+    },
+    {
+      "epoch": 0.33332,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.006151326497395833,
+      "learning_rate": 0.0001,
+      "loss": 3.7294,
+      "loss/crossentropy": 1.8066997528076172,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19391655176877975,
+      "step": 16666
+    },
+    {
+      "epoch": 0.33336,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.005370076497395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1831,
+      "loss/crossentropy": 2.085490345954895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21987342834472656,
+      "step": 16668
+    },
+    {
+      "epoch": 0.3334,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.006459299723307292,
+      "learning_rate": 0.0001,
+      "loss": 3.7791,
+      "loss/crossentropy": 2.052145302295685,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21673783659934998,
+      "step": 16670
+    },
+    {
+      "epoch": 0.33344,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.00640869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1648,
+      "loss/crossentropy": 2.0924419164657593,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21523578464984894,
+      "step": 16672
+    },
+    {
+      "epoch": 0.33348,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.006525675455729167,
+      "learning_rate": 0.0001,
+      "loss": 4.125,
+      "loss/crossentropy": 2.0340868830680847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1785074919462204,
+      "step": 16674
+    },
+    {
+      "epoch": 0.33352,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.010990397135416666,
+      "learning_rate": 0.0001,
+      "loss": 4.2404,
+      "loss/crossentropy": 2.4421908855438232,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24263421446084976,
+      "step": 16676
+    },
+    {
+      "epoch": 0.33356,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.0108551025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1879,
+      "loss/crossentropy": 1.9141955971717834,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18666712939739227,
+      "step": 16678
+    },
+    {
+      "epoch": 0.3336,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.011336008707682291,
+      "learning_rate": 0.0001,
+      "loss": 4.0378,
+      "loss/crossentropy": 2.0019638538360596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17722105979919434,
+      "step": 16680
+    },
+    {
+      "epoch": 0.33364,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.009511057535807292,
+      "learning_rate": 0.0001,
+      "loss": 4.2047,
+      "loss/crossentropy": 2.0835896134376526,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.251879021525383,
+      "step": 16682
+    },
+    {
+      "epoch": 0.33368,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.015192667643229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2406,
+      "loss/crossentropy": 1.8940032720565796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19034714996814728,
+      "step": 16684
+    },
+    {
+      "epoch": 0.33372,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.014817047119140624,
+      "learning_rate": 0.0001,
+      "loss": 4.0845,
+      "loss/crossentropy": 1.863099992275238,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19675587117671967,
+      "step": 16686
+    },
+    {
+      "epoch": 0.33376,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.024836222330729168,
+      "learning_rate": 0.0001,
+      "loss": 4.2785,
+      "loss/crossentropy": 2.027459740638733,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19650273770093918,
+      "step": 16688
+    },
+    {
+      "epoch": 0.3338,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.026008097330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.2892,
+      "loss/crossentropy": 2.2324944734573364,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23016268759965897,
+      "step": 16690
+    },
+    {
+      "epoch": 0.33384,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.02474950154622396,
+      "learning_rate": 0.0001,
+      "loss": 4.155,
+      "loss/crossentropy": 2.213741898536682,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21159538626670837,
+      "step": 16692
+    },
+    {
+      "epoch": 0.33388,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.025340779622395834,
+      "learning_rate": 0.0001,
+      "loss": 4.4426,
+      "loss/crossentropy": 2.084823966026306,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22060541808605194,
+      "step": 16694
+    },
+    {
+      "epoch": 0.33392,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.02417780558268229,
+      "learning_rate": 0.0001,
+      "loss": 4.1847,
+      "loss/crossentropy": 2.2417017221450806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2050381377339363,
+      "step": 16696
+    },
+    {
+      "epoch": 0.33396,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.032134755452473955,
+      "learning_rate": 0.0001,
+      "loss": 4.3037,
+      "loss/crossentropy": 2.3696242570877075,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22559890151023865,
+      "step": 16698
+    },
+    {
+      "epoch": 0.334,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.030295562744140626,
+      "learning_rate": 0.0001,
+      "loss": 4.158,
+      "loss/crossentropy": 2.0899609327316284,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20470503717660904,
+      "step": 16700
+    },
+    {
+      "epoch": 0.33404,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.029386138916015624,
+      "learning_rate": 0.0001,
+      "loss": 3.924,
+      "loss/crossentropy": 1.8663234114646912,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1815531924366951,
+      "step": 16702
+    },
+    {
+      "epoch": 0.33408,
+      "grad_norm": 3.0,
+      "grad_norm_var": 0.08069432576497396,
+      "learning_rate": 0.0001,
+      "loss": 4.1651,
+      "loss/crossentropy": 2.030432403087616,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2063748836517334,
+      "step": 16704
+    },
+    {
+      "epoch": 0.33412,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.08033218383789062,
+      "learning_rate": 0.0001,
+      "loss": 4.152,
+      "loss/crossentropy": 2.12885981798172,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19536730647087097,
+      "step": 16706
+    },
+    {
+      "epoch": 0.33416,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.07674153645833333,
+      "learning_rate": 0.0001,
+      "loss": 3.909,
+      "loss/crossentropy": 2.0870607495307922,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20478558540344238,
+      "step": 16708
+    },
+    {
+      "epoch": 0.3342,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.07646382649739583,
+      "learning_rate": 0.0001,
+      "loss": 4.2058,
+      "loss/crossentropy": 2.215463638305664,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2118721902370453,
+      "step": 16710
+    },
+    {
+      "epoch": 0.33424,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.08145243326822917,
+      "learning_rate": 0.0001,
+      "loss": 3.8417,
+      "loss/crossentropy": 2.0292049646377563,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18440847843885422,
+      "step": 16712
+    },
+    {
+      "epoch": 0.33428,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.08017756144205729,
+      "learning_rate": 0.0001,
+      "loss": 3.7722,
+      "loss/crossentropy": 1.9214341640472412,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18510671705007553,
+      "step": 16714
+    },
+    {
+      "epoch": 0.33432,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.07769266764322917,
+      "learning_rate": 0.0001,
+      "loss": 4.3905,
+      "loss/crossentropy": 2.1761534214019775,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20527766644954681,
+      "step": 16716
+    },
+    {
+      "epoch": 0.33436,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.07639567057291667,
+      "learning_rate": 0.0001,
+      "loss": 3.9234,
+      "loss/crossentropy": 2.020824670791626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20723462104797363,
+      "step": 16718
+    },
+    {
+      "epoch": 0.3344,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.0129547119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.167,
+      "loss/crossentropy": 2.243465781211853,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21866197139024734,
+      "step": 16720
+    },
+    {
+      "epoch": 0.33444,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.013244374593098959,
+      "learning_rate": 0.0001,
+      "loss": 3.9963,
+      "loss/crossentropy": 1.9071390628814697,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20368033647537231,
+      "step": 16722
+    },
+    {
+      "epoch": 0.33448,
+      "grad_norm": 1.7890625,
+      "grad_norm_var": 0.011966959635416666,
+      "learning_rate": 0.0001,
+      "loss": 3.8974,
+      "loss/crossentropy": 1.72525554895401,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18200145661830902,
+      "step": 16724
+    },
+    {
+      "epoch": 0.33452,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0154052734375,
+      "learning_rate": 0.0001,
+      "loss": 4.101,
+      "loss/crossentropy": 2.174055576324463,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21181844174861908,
+      "step": 16726
+    },
+    {
+      "epoch": 0.33456,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.014251454671223959,
+      "learning_rate": 0.0001,
+      "loss": 3.8004,
+      "loss/crossentropy": 1.6175724864006042,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15338046848773956,
+      "step": 16728
+    },
+    {
+      "epoch": 0.3346,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.012353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1101,
+      "loss/crossentropy": 2.0030194520950317,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20353461802005768,
+      "step": 16730
+    },
+    {
+      "epoch": 0.33464,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.012092081705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3876,
+      "loss/crossentropy": 2.272015690803528,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22368235886096954,
+      "step": 16732
+    },
+    {
+      "epoch": 0.33468,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.012189737955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3365,
+      "loss/crossentropy": 2.149975538253784,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20634270459413528,
+      "step": 16734
+    },
+    {
+      "epoch": 0.33472,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.009869130452473958,
+      "learning_rate": 0.0001,
+      "loss": 3.9201,
+      "loss/crossentropy": 2.095462441444397,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20027077198028564,
+      "step": 16736
+    },
+    {
+      "epoch": 0.33476,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.008819325764973959,
+      "learning_rate": 0.0001,
+      "loss": 4.0859,
+      "loss/crossentropy": 1.988598644733429,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20503661036491394,
+      "step": 16738
+    },
+    {
+      "epoch": 0.3348,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0066569010416666664,
+      "learning_rate": 0.0001,
+      "loss": 4.2618,
+      "loss/crossentropy": 2.0615572333335876,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20363956689834595,
+      "step": 16740
+    },
+    {
+      "epoch": 0.33484,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0027903238932291668,
+      "learning_rate": 0.0001,
+      "loss": 3.9476,
+      "loss/crossentropy": 1.8172455430030823,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20091129839420319,
+      "step": 16742
+    },
+    {
+      "epoch": 0.33488,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0043853759765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9541,
+      "loss/crossentropy": 2.212782144546509,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19710610061883926,
+      "step": 16744
+    },
+    {
+      "epoch": 0.33492,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.008674875895182291,
+      "learning_rate": 0.0001,
+      "loss": 4.2629,
+      "loss/crossentropy": 2.374183773994446,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23687051236629486,
+      "step": 16746
+    },
+    {
+      "epoch": 0.33496,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.008902740478515626,
+      "learning_rate": 0.0001,
+      "loss": 4.3642,
+      "loss/crossentropy": 1.9374622702598572,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19821622967720032,
+      "step": 16748
+    },
+    {
+      "epoch": 0.335,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.008740234375,
+      "learning_rate": 0.0001,
+      "loss": 4.1289,
+      "loss/crossentropy": 1.6063715815544128,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16296840459108353,
+      "step": 16750
+    },
+    {
+      "epoch": 0.33504,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.0096832275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0315,
+      "loss/crossentropy": 2.1251469254493713,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17997874319553375,
+      "step": 16752
+    },
+    {
+      "epoch": 0.33508,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0097564697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.03,
+      "loss/crossentropy": 2.126043915748596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2283206582069397,
+      "step": 16754
+    },
+    {
+      "epoch": 0.33512,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.009901682535807291,
+      "learning_rate": 0.0001,
+      "loss": 4.2969,
+      "loss/crossentropy": 2.3409340381622314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21817704290151596,
+      "step": 16756
+    },
+    {
+      "epoch": 0.33516,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.01024169921875,
+      "learning_rate": 0.0001,
+      "loss": 4.0083,
+      "loss/crossentropy": 1.9816790223121643,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19743956625461578,
+      "step": 16758
+    },
+    {
+      "epoch": 0.3352,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.012967681884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3056,
+      "loss/crossentropy": 2.246092438697815,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22662658989429474,
+      "step": 16760
+    },
+    {
+      "epoch": 0.33524,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010794830322265626,
+      "learning_rate": 0.0001,
+      "loss": 4.2842,
+      "loss/crossentropy": 2.1571802496910095,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20506682991981506,
+      "step": 16762
+    },
+    {
+      "epoch": 0.33528,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010636138916015624,
+      "learning_rate": 0.0001,
+      "loss": 4.0372,
+      "loss/crossentropy": 1.966954231262207,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18581678718328476,
+      "step": 16764
+    },
+    {
+      "epoch": 0.33532,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.012961578369140626,
+      "learning_rate": 0.0001,
+      "loss": 3.9978,
+      "loss/crossentropy": 1.9360153079032898,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20846693962812424,
+      "step": 16766
+    },
+    {
+      "epoch": 0.33536,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.012644195556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0953,
+      "loss/crossentropy": 2.3978073596954346,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20406542718410492,
+      "step": 16768
+    },
+    {
+      "epoch": 0.3354,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.013639068603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0933,
+      "loss/crossentropy": 2.239229917526245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21536701172590256,
+      "step": 16770
+    },
+    {
+      "epoch": 0.33544,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.013600413004557292,
+      "learning_rate": 0.0001,
+      "loss": 4.2153,
+      "loss/crossentropy": 2.1333428621292114,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2062046378850937,
+      "step": 16772
+    },
+    {
+      "epoch": 0.33548,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.014362589518229166,
+      "learning_rate": 0.0001,
+      "loss": 4.0094,
+      "loss/crossentropy": 2.3503127098083496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20422626286745071,
+      "step": 16774
+    },
+    {
+      "epoch": 0.33552,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.009772745768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1466,
+      "loss/crossentropy": 1.9181209802627563,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16844520717859268,
+      "step": 16776
+    },
+    {
+      "epoch": 0.33556,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.009269205729166667,
+      "learning_rate": 0.0001,
+      "loss": 4.1975,
+      "loss/crossentropy": 2.022172212600708,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19443488866090775,
+      "step": 16778
+    },
+    {
+      "epoch": 0.3356,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.010951487223307292,
+      "learning_rate": 0.0001,
+      "loss": 3.9544,
+      "loss/crossentropy": 2.0416210293769836,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17731749266386032,
+      "step": 16780
+    },
+    {
+      "epoch": 0.33564,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009748331705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0757,
+      "loss/crossentropy": 2.1722596883773804,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19995421171188354,
+      "step": 16782
+    },
+    {
+      "epoch": 0.33568,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.009146881103515626,
+      "learning_rate": 0.0001,
+      "loss": 4.0942,
+      "loss/crossentropy": 2.4041545391082764,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22784388065338135,
+      "step": 16784
+    },
+    {
+      "epoch": 0.33572,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.007757314046223958,
+      "learning_rate": 0.0001,
+      "loss": 4.0692,
+      "loss/crossentropy": 2.189204216003418,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20079431682825089,
+      "step": 16786
+    },
+    {
+      "epoch": 0.33576,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.007130686442057292,
+      "learning_rate": 0.0001,
+      "loss": 3.9771,
+      "loss/crossentropy": 2.2308130860328674,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22352158278226852,
+      "step": 16788
+    },
+    {
+      "epoch": 0.3358,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.005248006184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0901,
+      "loss/crossentropy": 2.238118886947632,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21165399253368378,
+      "step": 16790
+    },
+    {
+      "epoch": 0.33584,
+      "grad_norm": 11.9375,
+      "grad_norm_var": 6.248060862223308,
+      "learning_rate": 0.0001,
+      "loss": 4.5633,
+      "loss/crossentropy": 2.452346444129944,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2192159816622734,
+      "step": 16792
+    },
+    {
+      "epoch": 0.33588,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 6.234175364176433,
+      "learning_rate": 0.0001,
+      "loss": 4.2032,
+      "loss/crossentropy": 2.115506410598755,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19228096306324005,
+      "step": 16794
+    },
+    {
+      "epoch": 0.33592,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 6.207706705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.313,
+      "loss/crossentropy": 2.1052395701408386,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24314512312412262,
+      "step": 16796
+    },
+    {
+      "epoch": 0.33596,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 6.191576131184896,
+      "learning_rate": 0.0001,
+      "loss": 4.201,
+      "loss/crossentropy": 1.7643597722053528,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17998731136322021,
+      "step": 16798
+    },
+    {
+      "epoch": 0.336,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 6.213846588134766,
+      "learning_rate": 0.0001,
+      "loss": 3.886,
+      "loss/crossentropy": 2.0926302671432495,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.199835367500782,
+      "step": 16800
+    },
+    {
+      "epoch": 0.33604,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 6.19762954711914,
+      "learning_rate": 0.0001,
+      "loss": 4.0106,
+      "loss/crossentropy": 1.9939849972724915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19728610664606094,
+      "step": 16802
+    },
+    {
+      "epoch": 0.33608,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 6.1746826171875,
+      "learning_rate": 0.0001,
+      "loss": 4.3381,
+      "loss/crossentropy": 2.154136300086975,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2071739211678505,
+      "step": 16804
+    },
+    {
+      "epoch": 0.33612,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 6.173281860351563,
+      "learning_rate": 0.0001,
+      "loss": 4.06,
+      "loss/crossentropy": 1.6244451403617859,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1825866624712944,
+      "step": 16806
+    },
+    {
+      "epoch": 0.33616,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.010945383707682292,
+      "learning_rate": 0.0001,
+      "loss": 3.6796,
+      "loss/crossentropy": 1.936405599117279,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20031629502773285,
+      "step": 16808
+    },
+    {
+      "epoch": 0.3362,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.011568196614583333,
+      "learning_rate": 0.0001,
+      "loss": 4.276,
+      "loss/crossentropy": 2.117881119251251,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21941083669662476,
+      "step": 16810
+    },
+    {
+      "epoch": 0.33624,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.0125152587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1104,
+      "loss/crossentropy": 2.0384327173233032,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2019578069448471,
+      "step": 16812
+    },
+    {
+      "epoch": 0.33628,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.011893463134765626,
+      "learning_rate": 0.0001,
+      "loss": 4.1383,
+      "loss/crossentropy": 2.058988571166992,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20959977805614471,
+      "step": 16814
+    },
+    {
+      "epoch": 0.33632,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.010680898030598959,
+      "learning_rate": 0.0001,
+      "loss": 4.0734,
+      "loss/crossentropy": 2.117241382598877,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2039024829864502,
+      "step": 16816
+    },
+    {
+      "epoch": 0.33636,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.009968058268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.157,
+      "loss/crossentropy": 1.8142318725585938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1877349317073822,
+      "step": 16818
+    },
+    {
+      "epoch": 0.3364,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.014440663655598958,
+      "learning_rate": 0.0001,
+      "loss": 4.2402,
+      "loss/crossentropy": 2.1766676902770996,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21788865327835083,
+      "step": 16820
+    },
+    {
+      "epoch": 0.33644,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0163726806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2953,
+      "loss/crossentropy": 2.2596739530563354,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23641519993543625,
+      "step": 16822
+    },
+    {
+      "epoch": 0.33648,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.012572987874348959,
+      "learning_rate": 0.0001,
+      "loss": 4.1689,
+      "loss/crossentropy": 2.1782987117767334,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20562626421451569,
+      "step": 16824
+    },
+    {
+      "epoch": 0.33652,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.01633275349934896,
+      "learning_rate": 0.0001,
+      "loss": 4.0693,
+      "loss/crossentropy": 1.9579410552978516,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20332375168800354,
+      "step": 16826
+    },
+    {
+      "epoch": 0.33656,
+      "grad_norm": 1.75,
+      "grad_norm_var": 0.020977528889973958,
+      "learning_rate": 0.0001,
+      "loss": 4.0193,
+      "loss/crossentropy": 1.917382538318634,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.188996322453022,
+      "step": 16828
+    },
+    {
+      "epoch": 0.3366,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.02045466105143229,
+      "learning_rate": 0.0001,
+      "loss": 3.9885,
+      "loss/crossentropy": 1.8045424222946167,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18587132543325424,
+      "step": 16830
+    },
+    {
+      "epoch": 0.33664,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.07043355305989583,
+      "learning_rate": 0.0001,
+      "loss": 4.0171,
+      "loss/crossentropy": 2.113844871520996,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20383527874946594,
+      "step": 16832
+    },
+    {
+      "epoch": 0.33668,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.06795654296875,
+      "learning_rate": 0.0001,
+      "loss": 4.1255,
+      "loss/crossentropy": 1.9635592699050903,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20447230339050293,
+      "step": 16834
+    },
+    {
+      "epoch": 0.33672,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.06646219889322917,
+      "learning_rate": 0.0001,
+      "loss": 4.1336,
+      "loss/crossentropy": 1.8918602466583252,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21786046028137207,
+      "step": 16836
+    },
+    {
+      "epoch": 0.33676,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.06734390258789062,
+      "learning_rate": 0.0001,
+      "loss": 4.3372,
+      "loss/crossentropy": 2.347909450531006,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23047995567321777,
+      "step": 16838
+    },
+    {
+      "epoch": 0.3368,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.06854248046875,
+      "learning_rate": 0.0001,
+      "loss": 4.0242,
+      "loss/crossentropy": 1.834633469581604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1915532797574997,
+      "step": 16840
+    },
+    {
+      "epoch": 0.33684,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.06948954264322917,
+      "learning_rate": 0.0001,
+      "loss": 3.6871,
+      "loss/crossentropy": 1.9172112345695496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18901804834604263,
+      "step": 16842
+    },
+    {
+      "epoch": 0.33688,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.06441141764322916,
+      "learning_rate": 0.0001,
+      "loss": 4.3911,
+      "loss/crossentropy": 2.37927508354187,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2328021451830864,
+      "step": 16844
+    },
+    {
+      "epoch": 0.33692,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.06558024088541667,
+      "learning_rate": 0.0001,
+      "loss": 3.8573,
+      "loss/crossentropy": 1.7887099385261536,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17989587783813477,
+      "step": 16846
+    },
+    {
+      "epoch": 0.33696,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0051513671875,
+      "learning_rate": 0.0001,
+      "loss": 4.0826,
+      "loss/crossentropy": 2.008473217487335,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20677915215492249,
+      "step": 16848
+    },
+    {
+      "epoch": 0.337,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.005362955729166666,
+      "learning_rate": 0.0001,
+      "loss": 4.1751,
+      "loss/crossentropy": 2.101797103881836,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19833290576934814,
+      "step": 16850
+    },
+    {
+      "epoch": 0.33704,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.005779774983723959,
+      "learning_rate": 0.0001,
+      "loss": 4.0087,
+      "loss/crossentropy": 1.9306662678718567,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1893247440457344,
+      "step": 16852
+    },
+    {
+      "epoch": 0.33708,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.005293782552083333,
+      "learning_rate": 0.0001,
+      "loss": 3.9986,
+      "loss/crossentropy": 2.1301704049110413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1975892335176468,
+      "step": 16854
+    },
+    {
+      "epoch": 0.33712,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.007173411051432292,
+      "learning_rate": 0.0001,
+      "loss": 3.9583,
+      "loss/crossentropy": 2.020972192287445,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19376973807811737,
+      "step": 16856
+    },
+    {
+      "epoch": 0.33716,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.006159464518229167,
+      "learning_rate": 0.0001,
+      "loss": 3.9632,
+      "loss/crossentropy": 1.8190429210662842,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1629476398229599,
+      "step": 16858
+    },
+    {
+      "epoch": 0.3372,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.007183583577473959,
+      "learning_rate": 0.0001,
+      "loss": 4.2147,
+      "loss/crossentropy": 2.360092878341675,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22874057292938232,
+      "step": 16860
+    },
+    {
+      "epoch": 0.33724,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.008853912353515625,
+      "learning_rate": 0.0001,
+      "loss": 4.232,
+      "loss/crossentropy": 2.3477792739868164,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2215813398361206,
+      "step": 16862
+    },
+    {
+      "epoch": 0.33728,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.009919230143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2551,
+      "loss/crossentropy": 2.474452257156372,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2185184359550476,
+      "step": 16864
+    },
+    {
+      "epoch": 0.33732,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010396067301432292,
+      "learning_rate": 0.0001,
+      "loss": 3.7782,
+      "loss/crossentropy": 1.8387269973754883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18137617409229279,
+      "step": 16866
+    },
+    {
+      "epoch": 0.33736,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.009924062093098958,
+      "learning_rate": 0.0001,
+      "loss": 3.9414,
+      "loss/crossentropy": 1.8537965416908264,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19480426609516144,
+      "step": 16868
+    },
+    {
+      "epoch": 0.3374,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.010057576497395833,
+      "learning_rate": 0.0001,
+      "loss": 3.9808,
+      "loss/crossentropy": 1.9274648427963257,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19545693695545197,
+      "step": 16870
+    },
+    {
+      "epoch": 0.33744,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.010273996988932292,
+      "learning_rate": 0.0001,
+      "loss": 4.1097,
+      "loss/crossentropy": 1.9180024862289429,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2180539146065712,
+      "step": 16872
+    },
+    {
+      "epoch": 0.33748,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.011016591389973959,
+      "learning_rate": 0.0001,
+      "loss": 4.4008,
+      "loss/crossentropy": 2.185365915298462,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2225860357284546,
+      "step": 16874
+    },
+    {
+      "epoch": 0.33752,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.011523183186848958,
+      "learning_rate": 0.0001,
+      "loss": 3.9882,
+      "loss/crossentropy": 1.9749634861946106,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19897626340389252,
+      "step": 16876
+    },
+    {
+      "epoch": 0.33756,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.010284169514973959,
+      "learning_rate": 0.0001,
+      "loss": 4.237,
+      "loss/crossentropy": 2.1505807638168335,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20921844244003296,
+      "step": 16878
+    },
+    {
+      "epoch": 0.3376,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.01002197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0692,
+      "loss/crossentropy": 1.7159577012062073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19649318605661392,
+      "step": 16880
+    },
+    {
+      "epoch": 0.33764,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.010251617431640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1615,
+      "loss/crossentropy": 2.187040388584137,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20251552015542984,
+      "step": 16882
+    },
+    {
+      "epoch": 0.33768,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.014609527587890626,
+      "learning_rate": 0.0001,
+      "loss": 4.0587,
+      "loss/crossentropy": 1.9247627258300781,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17710037529468536,
+      "step": 16884
+    },
+    {
+      "epoch": 0.33772,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0141845703125,
+      "learning_rate": 0.0001,
+      "loss": 4.001,
+      "loss/crossentropy": 1.8499276041984558,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18950388580560684,
+      "step": 16886
+    },
+    {
+      "epoch": 0.33776,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.0138824462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0998,
+      "loss/crossentropy": 2.0127341747283936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18379998207092285,
+      "step": 16888
+    },
+    {
+      "epoch": 0.3378,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.013602447509765626,
+      "learning_rate": 0.0001,
+      "loss": 3.9096,
+      "loss/crossentropy": 2.2561115026474,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20785757154226303,
+      "step": 16890
+    },
+    {
+      "epoch": 0.33784,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0131103515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9638,
+      "loss/crossentropy": 1.9391398429870605,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19481037557125092,
+      "step": 16892
+    },
+    {
+      "epoch": 0.33788,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.014827219645182292,
+      "learning_rate": 0.0001,
+      "loss": 4.2368,
+      "loss/crossentropy": 2.3129481077194214,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2065483033657074,
+      "step": 16894
+    },
+    {
+      "epoch": 0.33792,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.01984430948893229,
+      "learning_rate": 0.0001,
+      "loss": 4.0774,
+      "loss/crossentropy": 1.8482372760772705,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19291523844003677,
+      "step": 16896
+    },
+    {
+      "epoch": 0.33796,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.020031483968098958,
+      "learning_rate": 0.0001,
+      "loss": 4.3876,
+      "loss/crossentropy": 1.973912537097931,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22137422114610672,
+      "step": 16898
+    },
+    {
+      "epoch": 0.338,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.01654841105143229,
+      "learning_rate": 0.0001,
+      "loss": 3.9599,
+      "loss/crossentropy": 1.7987132668495178,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18385548144578934,
+      "step": 16900
+    },
+    {
+      "epoch": 0.33804,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.01727879842122396,
+      "learning_rate": 0.0001,
+      "loss": 4.3673,
+      "loss/crossentropy": 2.199060797691345,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2016109600663185,
+      "step": 16902
+    },
+    {
+      "epoch": 0.33808,
+      "grad_norm": 3.046875,
+      "grad_norm_var": 0.0830718994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.27,
+      "loss/crossentropy": 2.179081439971924,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21342043578624725,
+      "step": 16904
+    },
+    {
+      "epoch": 0.33812,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0810455322265625,
+      "learning_rate": 0.0001,
+      "loss": 3.9861,
+      "loss/crossentropy": 1.953968107700348,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20462830364704132,
+      "step": 16906
+    },
+    {
+      "epoch": 0.33816,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.08123270670572917,
+      "learning_rate": 0.0001,
+      "loss": 4.3432,
+      "loss/crossentropy": 2.3202494382858276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21390480548143387,
+      "step": 16908
+    },
+    {
+      "epoch": 0.3382,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.07872899373372395,
+      "learning_rate": 0.0001,
+      "loss": 4.3085,
+      "loss/crossentropy": 2.085767686367035,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20946675539016724,
+      "step": 16910
+    },
+    {
+      "epoch": 0.33824,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.07372945149739583,
+      "learning_rate": 0.0001,
+      "loss": 4.217,
+      "loss/crossentropy": 1.789705514907837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2247294932603836,
+      "step": 16912
+    },
+    {
+      "epoch": 0.33828,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.07440770467122396,
+      "learning_rate": 0.0001,
+      "loss": 4.0409,
+      "loss/crossentropy": 1.9354140758514404,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17958863079547882,
+      "step": 16914
+    },
+    {
+      "epoch": 0.33832,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.07529296875,
+      "learning_rate": 0.0001,
+      "loss": 3.6949,
+      "loss/crossentropy": 1.878045916557312,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1812344640493393,
+      "step": 16916
+    },
+    {
+      "epoch": 0.33836,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.0769488016764323,
+      "learning_rate": 0.0001,
+      "loss": 3.9392,
+      "loss/crossentropy": 1.7474132776260376,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19444099068641663,
+      "step": 16918
+    },
+    {
+      "epoch": 0.3384,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.007185618082682292,
+      "learning_rate": 0.0001,
+      "loss": 4.1524,
+      "loss/crossentropy": 2.0462751984596252,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20077970623970032,
+      "step": 16920
+    },
+    {
+      "epoch": 0.33844,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.006912994384765625,
+      "learning_rate": 0.0001,
+      "loss": 4.205,
+      "loss/crossentropy": 2.1104917526245117,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21799270808696747,
+      "step": 16922
+    },
+    {
+      "epoch": 0.33848,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.007083892822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1915,
+      "loss/crossentropy": 2.11525696516037,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20585542917251587,
+      "step": 16924
+    },
+    {
+      "epoch": 0.33852,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.005222320556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.3629,
+      "loss/crossentropy": 2.3451485633850098,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22672247886657715,
+      "step": 16926
+    },
+    {
+      "epoch": 0.33856,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.004906209309895834,
+      "learning_rate": 0.0001,
+      "loss": 4.0493,
+      "loss/crossentropy": 1.8872862458229065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1973365843296051,
+      "step": 16928
+    },
+    {
+      "epoch": 0.3386,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.00777587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1082,
+      "loss/crossentropy": 1.9465213418006897,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2297648787498474,
+      "step": 16930
+    },
+    {
+      "epoch": 0.33864,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0084625244140625,
+      "learning_rate": 0.0001,
+      "loss": 3.9397,
+      "loss/crossentropy": 1.7463516592979431,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.193784698843956,
+      "step": 16932
+    },
+    {
+      "epoch": 0.33868,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.006738026936848958,
+      "learning_rate": 0.0001,
+      "loss": 4.1561,
+      "loss/crossentropy": 2.2214877605438232,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2064037024974823,
+      "step": 16934
+    },
+    {
+      "epoch": 0.33872,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.007600657145182292,
+      "learning_rate": 0.0001,
+      "loss": 4.0355,
+      "loss/crossentropy": 1.896401822566986,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21860874444246292,
+      "step": 16936
+    },
+    {
+      "epoch": 0.33876,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.007209269205729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0701,
+      "loss/crossentropy": 2.508669376373291,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23076221346855164,
+      "step": 16938
+    },
+    {
+      "epoch": 0.3388,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.009266916910807292,
+      "learning_rate": 0.0001,
+      "loss": 3.8421,
+      "loss/crossentropy": 1.9412622451782227,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1833793818950653,
+      "step": 16940
+    },
+    {
+      "epoch": 0.33884,
+      "grad_norm": 2.484375,
+      "grad_norm_var": 0.025406646728515624,
+      "learning_rate": 0.0001,
+      "loss": 4.362,
+      "loss/crossentropy": 2.4757901430130005,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23629899322986603,
+      "step": 16942
+    },
+    {
+      "epoch": 0.33888,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.02580744425455729,
+      "learning_rate": 0.0001,
+      "loss": 4.1204,
+      "loss/crossentropy": 1.788071870803833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1894787922501564,
+      "step": 16944
+    },
+    {
+      "epoch": 0.33892,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.023436482747395834,
+      "learning_rate": 0.0001,
+      "loss": 4.0626,
+      "loss/crossentropy": 2.1902449131011963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21310994774103165,
+      "step": 16946
+    },
+    {
+      "epoch": 0.33896,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.021394856770833335,
+      "learning_rate": 0.0001,
+      "loss": 3.9828,
+      "loss/crossentropy": 2.0444132685661316,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1755325198173523,
+      "step": 16948
+    },
+    {
+      "epoch": 0.339,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.02200902303059896,
+      "learning_rate": 0.0001,
+      "loss": 4.1609,
+      "loss/crossentropy": 1.9242961406707764,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18970132619142532,
+      "step": 16950
+    },
+    {
+      "epoch": 0.33904,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0216552734375,
+      "learning_rate": 0.0001,
+      "loss": 3.9819,
+      "loss/crossentropy": 1.7046860456466675,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19807563722133636,
+      "step": 16952
+    },
+    {
+      "epoch": 0.33908,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.02184015909830729,
+      "learning_rate": 0.0001,
+      "loss": 3.8975,
+      "loss/crossentropy": 1.8960286974906921,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17571083456277847,
+      "step": 16954
+    },
+    {
+      "epoch": 0.33912,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.019481404622395834,
+      "learning_rate": 0.0001,
+      "loss": 3.922,
+      "loss/crossentropy": 1.9351304769515991,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1918834000825882,
+      "step": 16956
+    },
+    {
+      "epoch": 0.33916,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.017728424072265624,
+      "learning_rate": 0.0001,
+      "loss": 4.167,
+      "loss/crossentropy": 1.974304735660553,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20828718692064285,
+      "step": 16958
+    },
+    {
+      "epoch": 0.3392,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.017867024739583334,
+      "learning_rate": 0.0001,
+      "loss": 3.8506,
+      "loss/crossentropy": 2.3836190700531006,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2074722871184349,
+      "step": 16960
+    },
+    {
+      "epoch": 0.33924,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.018123372395833334,
+      "learning_rate": 0.0001,
+      "loss": 4.0827,
+      "loss/crossentropy": 2.1663198471069336,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20097027719020844,
+      "step": 16962
+    },
+    {
+      "epoch": 0.33928,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.01810277303059896,
+      "learning_rate": 0.0001,
+      "loss": 4.2466,
+      "loss/crossentropy": 2.275226354598999,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22942040860652924,
+      "step": 16964
+    },
+    {
+      "epoch": 0.33932,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.017796834309895832,
+      "learning_rate": 0.0001,
+      "loss": 3.9802,
+      "loss/crossentropy": 1.7741501331329346,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1822734773159027,
+      "step": 16966
+    },
+    {
+      "epoch": 0.33936,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.017215983072916666,
+      "learning_rate": 0.0001,
+      "loss": 3.9016,
+      "loss/crossentropy": 1.791014850139618,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2008860930800438,
+      "step": 16968
+    },
+    {
+      "epoch": 0.3394,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.016888173421223958,
+      "learning_rate": 0.0001,
+      "loss": 4.1941,
+      "loss/crossentropy": 1.9239555597305298,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.192842036485672,
+      "step": 16970
+    },
+    {
+      "epoch": 0.33944,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.016788482666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.108,
+      "loss/crossentropy": 2.1543636322021484,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20888526737689972,
+      "step": 16972
+    },
+    {
+      "epoch": 0.33948,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.005012003580729166,
+      "learning_rate": 0.0001,
+      "loss": 3.975,
+      "loss/crossentropy": 2.2322418093681335,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19024743884801865,
+      "step": 16974
+    },
+    {
+      "epoch": 0.33952,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012422434488932292,
+      "learning_rate": 0.0001,
+      "loss": 4.3208,
+      "loss/crossentropy": 1.9934163093566895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26193471997976303,
+      "step": 16976
+    },
+    {
+      "epoch": 0.33956,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.013244374593098959,
+      "learning_rate": 0.0001,
+      "loss": 4.2753,
+      "loss/crossentropy": 2.340088725090027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2110593616962433,
+      "step": 16978
+    },
+    {
+      "epoch": 0.3396,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.03021214803059896,
+      "learning_rate": 0.0001,
+      "loss": 4.2047,
+      "loss/crossentropy": 2.057206869125366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2071269005537033,
+      "step": 16980
+    },
+    {
+      "epoch": 0.33964,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.030210113525390624,
+      "learning_rate": 0.0001,
+      "loss": 4.0262,
+      "loss/crossentropy": 2.002032458782196,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19903026521205902,
+      "step": 16982
+    },
+    {
+      "epoch": 0.33968,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.030804189046223958,
+      "learning_rate": 0.0001,
+      "loss": 4.246,
+      "loss/crossentropy": 2.01567679643631,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20198361575603485,
+      "step": 16984
+    },
+    {
+      "epoch": 0.33972,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.032814280192057295,
+      "learning_rate": 0.0001,
+      "loss": 4.0606,
+      "loss/crossentropy": 1.661778211593628,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16601823270320892,
+      "step": 16986
+    },
+    {
+      "epoch": 0.33976,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.03467203776041667,
+      "learning_rate": 0.0001,
+      "loss": 3.9411,
+      "loss/crossentropy": 2.284485101699829,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20108629018068314,
+      "step": 16988
+    },
+    {
+      "epoch": 0.3398,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.03200861612955729,
+      "learning_rate": 0.0001,
+      "loss": 4.1066,
+      "loss/crossentropy": 2.1860578656196594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22195414453744888,
+      "step": 16990
+    },
+    {
+      "epoch": 0.33984,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.02719904581705729,
+      "learning_rate": 0.0001,
+      "loss": 4.0212,
+      "loss/crossentropy": 2.032013416290283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19930274784564972,
+      "step": 16992
+    },
+    {
+      "epoch": 0.33988,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.1898577372233073,
+      "learning_rate": 0.0001,
+      "loss": 4.2847,
+      "loss/crossentropy": 2.227185010910034,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19806954264640808,
+      "step": 16994
+    },
+    {
+      "epoch": 0.33992,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.17535807291666666,
+      "learning_rate": 0.0001,
+      "loss": 4.1526,
+      "loss/crossentropy": 2.222020983695984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2354518324136734,
+      "step": 16996
+    },
+    {
+      "epoch": 0.33996,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.1755767822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1766,
+      "loss/crossentropy": 2.336432456970215,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19555255770683289,
+      "step": 16998
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.1748443603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3312,
+      "loss/crossentropy": 2.17374986410141,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21119412034749985,
+      "step": 17000
+    },
+    {
+      "epoch": 0.34004,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.17195536295572916,
+      "learning_rate": 0.0001,
+      "loss": 4.1445,
+      "loss/crossentropy": 2.135382056236267,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2263253927230835,
+      "step": 17002
+    },
+    {
+      "epoch": 0.34008,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.20149917602539064,
+      "learning_rate": 0.0001,
+      "loss": 3.811,
+      "loss/crossentropy": 1.9646037220954895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18890808522701263,
+      "step": 17004
+    },
+    {
+      "epoch": 0.34012,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.20098368326822916,
+      "learning_rate": 0.0001,
+      "loss": 4.0889,
+      "loss/crossentropy": 1.8009583353996277,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1882813572883606,
+      "step": 17006
+    },
+    {
+      "epoch": 0.34016,
+      "grad_norm": 3.109375,
+      "grad_norm_var": 0.24433492024739584,
+      "learning_rate": 0.0001,
+      "loss": 4.4921,
+      "loss/crossentropy": 2.0362982153892517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.208044171333313,
+      "step": 17008
+    },
+    {
+      "epoch": 0.3402,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.11477432250976563,
+      "learning_rate": 0.0001,
+      "loss": 4.1781,
+      "loss/crossentropy": 2.2055057287216187,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21141493320465088,
+      "step": 17010
+    },
+    {
+      "epoch": 0.34024,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.11963475545247396,
+      "learning_rate": 0.0001,
+      "loss": 3.772,
+      "loss/crossentropy": 1.386088252067566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16025834530591965,
+      "step": 17012
+    },
+    {
+      "epoch": 0.34028,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.11599299112955729,
+      "learning_rate": 0.0001,
+      "loss": 4.2389,
+      "loss/crossentropy": 2.131182312965393,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20460083335638046,
+      "step": 17014
+    },
+    {
+      "epoch": 0.34032,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.11592992146809895,
+      "learning_rate": 0.0001,
+      "loss": 4.2681,
+      "loss/crossentropy": 2.2682281732559204,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22915159910917282,
+      "step": 17016
+    },
+    {
+      "epoch": 0.34036,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.12011693318684896,
+      "learning_rate": 0.0001,
+      "loss": 4.0521,
+      "loss/crossentropy": 2.0855059027671814,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19398894160985947,
+      "step": 17018
+    },
+    {
+      "epoch": 0.3404,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0842437744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0421,
+      "loss/crossentropy": 1.895095944404602,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2172701731324196,
+      "step": 17020
+    },
+    {
+      "epoch": 0.34044,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.08455403645833333,
+      "learning_rate": 0.0001,
+      "loss": 4.0814,
+      "loss/crossentropy": 2.10223788022995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2144509255886078,
+      "step": 17022
+    },
+    {
+      "epoch": 0.34048,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.014241282145182292,
+      "learning_rate": 0.0001,
+      "loss": 4.1211,
+      "loss/crossentropy": 2.1131081581115723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21176592260599136,
+      "step": 17024
+    },
+    {
+      "epoch": 0.34052,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.009991200764973958,
+      "learning_rate": 0.0001,
+      "loss": 4.1581,
+      "loss/crossentropy": 2.1125452518463135,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1931193768978119,
+      "step": 17026
+    },
+    {
+      "epoch": 0.34056,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.017775217692057293,
+      "learning_rate": 0.0001,
+      "loss": 4.2624,
+      "loss/crossentropy": 2.213385283946991,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2576863020658493,
+      "step": 17028
+    },
+    {
+      "epoch": 0.3406,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.017773183186848958,
+      "learning_rate": 0.0001,
+      "loss": 4.408,
+      "loss/crossentropy": 2.0793206095695496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20338336378335953,
+      "step": 17030
+    },
+    {
+      "epoch": 0.34064,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.01807225545247396,
+      "learning_rate": 0.0001,
+      "loss": 4.3102,
+      "loss/crossentropy": 1.8870239853858948,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18544750660657883,
+      "step": 17032
+    },
+    {
+      "epoch": 0.34068,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.017254384358723958,
+      "learning_rate": 0.0001,
+      "loss": 4.1538,
+      "loss/crossentropy": 1.9893989562988281,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19679997861385345,
+      "step": 17034
+    },
+    {
+      "epoch": 0.34072,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.01845270792643229,
+      "learning_rate": 0.0001,
+      "loss": 3.9447,
+      "loss/crossentropy": 2.0401668548583984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19242482632398605,
+      "step": 17036
+    },
+    {
+      "epoch": 0.34076,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.015472157796223959,
+      "learning_rate": 0.0001,
+      "loss": 4.0509,
+      "loss/crossentropy": 2.0786415934562683,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21781006455421448,
+      "step": 17038
+    },
+    {
+      "epoch": 0.3408,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.015533192952473959,
+      "learning_rate": 0.0001,
+      "loss": 3.9472,
+      "loss/crossentropy": 2.143572211265564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20608216524124146,
+      "step": 17040
+    },
+    {
+      "epoch": 0.34084,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0154693603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3305,
+      "loss/crossentropy": 2.0578572750091553,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22006108611822128,
+      "step": 17042
+    },
+    {
+      "epoch": 0.34088,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.0061920166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1295,
+      "loss/crossentropy": 2.1016032099723816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2059948667883873,
+      "step": 17044
+    },
+    {
+      "epoch": 0.34092,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.00438232421875,
+      "learning_rate": 0.0001,
+      "loss": 3.9293,
+      "loss/crossentropy": 1.814449965953827,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1946963220834732,
+      "step": 17046
+    },
+    {
+      "epoch": 0.34096,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.004233551025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0696,
+      "loss/crossentropy": 2.2421000599861145,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2114911824464798,
+      "step": 17048
+    },
+    {
+      "epoch": 0.341,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0070231119791666664,
+      "learning_rate": 0.0001,
+      "loss": 3.984,
+      "loss/crossentropy": 2.1200218200683594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.212583489716053,
+      "step": 17050
+    },
+    {
+      "epoch": 0.34104,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.007441965738932291,
+      "learning_rate": 0.0001,
+      "loss": 3.8656,
+      "loss/crossentropy": 1.9486305713653564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18631484359502792,
+      "step": 17052
+    },
+    {
+      "epoch": 0.34108,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.007721964518229167,
+      "learning_rate": 0.0001,
+      "loss": 4.3888,
+      "loss/crossentropy": 2.4838292598724365,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23271672427654266,
+      "step": 17054
+    },
+    {
+      "epoch": 0.34112,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.008137766520182292,
+      "learning_rate": 0.0001,
+      "loss": 3.9099,
+      "loss/crossentropy": 1.8606489896774292,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1955883800983429,
+      "step": 17056
+    },
+    {
+      "epoch": 0.34116,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.013199615478515624,
+      "learning_rate": 0.0001,
+      "loss": 4.2847,
+      "loss/crossentropy": 2.1702204942703247,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21619878709316254,
+      "step": 17058
+    },
+    {
+      "epoch": 0.3412,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.013158162434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1987,
+      "loss/crossentropy": 2.1011393070220947,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2120063453912735,
+      "step": 17060
+    },
+    {
+      "epoch": 0.34124,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.013142903645833334,
+      "learning_rate": 0.0001,
+      "loss": 4.1704,
+      "loss/crossentropy": 2.267427682876587,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20903929322957993,
+      "step": 17062
+    },
+    {
+      "epoch": 0.34128,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.01620457967122396,
+      "learning_rate": 0.0001,
+      "loss": 4.3312,
+      "loss/crossentropy": 2.125304937362671,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18540818989276886,
+      "step": 17064
+    },
+    {
+      "epoch": 0.34132,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.013925933837890625,
+      "learning_rate": 0.0001,
+      "loss": 3.9552,
+      "loss/crossentropy": 1.8741024136543274,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19243815541267395,
+      "step": 17066
+    },
+    {
+      "epoch": 0.34136,
+      "grad_norm": 2.46875,
+      "grad_norm_var": 0.023451487223307293,
+      "learning_rate": 0.0001,
+      "loss": 4.0252,
+      "loss/crossentropy": 2.12781822681427,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23974300920963287,
+      "step": 17068
+    },
+    {
+      "epoch": 0.3414,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.02448298136393229,
+      "learning_rate": 0.0001,
+      "loss": 4.2728,
+      "loss/crossentropy": 2.0786141753196716,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22311320155858994,
+      "step": 17070
+    },
+    {
+      "epoch": 0.34144,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.02408421834309896,
+      "learning_rate": 0.0001,
+      "loss": 4.1686,
+      "loss/crossentropy": 2.1569892168045044,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20391946285963058,
+      "step": 17072
+    },
+    {
+      "epoch": 0.34148,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.024607086181640626,
+      "learning_rate": 0.0001,
+      "loss": 3.8035,
+      "loss/crossentropy": 1.8860353231430054,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1937229335308075,
+      "step": 17074
+    },
+    {
+      "epoch": 0.34152,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.02448094685872396,
+      "learning_rate": 0.0001,
+      "loss": 4.0993,
+      "loss/crossentropy": 2.25645911693573,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2088249772787094,
+      "step": 17076
+    },
+    {
+      "epoch": 0.34156,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.02972997029622396,
+      "learning_rate": 0.0001,
+      "loss": 4.3486,
+      "loss/crossentropy": 2.379481792449951,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2341819554567337,
+      "step": 17078
+    },
+    {
+      "epoch": 0.3416,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.0273345947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0841,
+      "loss/crossentropy": 2.1564733386039734,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2162168025970459,
+      "step": 17080
+    },
+    {
+      "epoch": 0.34164,
+      "grad_norm": 3.015625,
+      "grad_norm_var": 0.0860260009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2526,
+      "loss/crossentropy": 1.8906886577606201,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18614360690116882,
+      "step": 17082
+    },
+    {
+      "epoch": 0.34168,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.07732747395833334,
+      "learning_rate": 0.0001,
+      "loss": 4.0135,
+      "loss/crossentropy": 2.093637704849243,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19867172092199326,
+      "step": 17084
+    },
+    {
+      "epoch": 0.34172,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0760210673014323,
+      "learning_rate": 0.0001,
+      "loss": 4.2364,
+      "loss/crossentropy": 2.2579731941223145,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2153376340866089,
+      "step": 17086
+    },
+    {
+      "epoch": 0.34176,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.07312393188476562,
+      "learning_rate": 0.0001,
+      "loss": 4.3275,
+      "loss/crossentropy": 2.250504732131958,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2759099751710892,
+      "step": 17088
+    },
+    {
+      "epoch": 0.3418,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.07190653483072916,
+      "learning_rate": 0.0001,
+      "loss": 3.8977,
+      "loss/crossentropy": 1.6062138676643372,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15309542417526245,
+      "step": 17090
+    },
+    {
+      "epoch": 0.34184,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.07587865193684896,
+      "learning_rate": 0.0001,
+      "loss": 4.2199,
+      "loss/crossentropy": 2.227494239807129,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20563329756259918,
+      "step": 17092
+    },
+    {
+      "epoch": 0.34188,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.07314631144205729,
+      "learning_rate": 0.0001,
+      "loss": 4.1095,
+      "loss/crossentropy": 2.1047890186309814,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19775691628456116,
+      "step": 17094
+    },
+    {
+      "epoch": 0.34192,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.07301813761393229,
+      "learning_rate": 0.0001,
+      "loss": 4.0508,
+      "loss/crossentropy": 2.044450581073761,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19867002964019775,
+      "step": 17096
+    },
+    {
+      "epoch": 0.34196,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.009511057535807292,
+      "learning_rate": 0.0001,
+      "loss": 4.2319,
+      "loss/crossentropy": 1.9139947891235352,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20143548399209976,
+      "step": 17098
+    },
+    {
+      "epoch": 0.342,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.008162180582682291,
+      "learning_rate": 0.0001,
+      "loss": 3.7159,
+      "loss/crossentropy": 1.6752634048461914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17691873013973236,
+      "step": 17100
+    },
+    {
+      "epoch": 0.34204,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0156402587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2159,
+      "loss/crossentropy": 2.277552843093872,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2049017697572708,
+      "step": 17102
+    },
+    {
+      "epoch": 0.34208,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.015608469645182291,
+      "learning_rate": 0.0001,
+      "loss": 4.0663,
+      "loss/crossentropy": 2.458711266517639,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21726765483617783,
+      "step": 17104
+    },
+    {
+      "epoch": 0.34212,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.01627197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0384,
+      "loss/crossentropy": 1.9302194714546204,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19446631520986557,
+      "step": 17106
+    },
+    {
+      "epoch": 0.34216,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.01622314453125,
+      "learning_rate": 0.0001,
+      "loss": 3.8935,
+      "loss/crossentropy": 2.021213114261627,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19394385814666748,
+      "step": 17108
+    },
+    {
+      "epoch": 0.3422,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.01689453125,
+      "learning_rate": 0.0001,
+      "loss": 4.4094,
+      "loss/crossentropy": 2.4856090545654297,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2156004160642624,
+      "step": 17110
+    },
+    {
+      "epoch": 0.34224,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.016869862874348957,
+      "learning_rate": 0.0001,
+      "loss": 4.4135,
+      "loss/crossentropy": 2.410070061683655,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21203476190567017,
+      "step": 17112
+    },
+    {
+      "epoch": 0.34228,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.019136555989583335,
+      "learning_rate": 0.0001,
+      "loss": 4.0484,
+      "loss/crossentropy": 2.026827871799469,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20053605735301971,
+      "step": 17114
+    },
+    {
+      "epoch": 0.34232,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.016283162434895835,
+      "learning_rate": 0.0001,
+      "loss": 4.2319,
+      "loss/crossentropy": 2.1833966970443726,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2193789780139923,
+      "step": 17116
+    },
+    {
+      "epoch": 0.34236,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.010798899332682292,
+      "learning_rate": 0.0001,
+      "loss": 4.0921,
+      "loss/crossentropy": 2.137472152709961,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21218669414520264,
+      "step": 17118
+    },
+    {
+      "epoch": 0.3424,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.013741048177083333,
+      "learning_rate": 0.0001,
+      "loss": 4.2064,
+      "loss/crossentropy": 2.0607420206069946,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2138473093509674,
+      "step": 17120
+    },
+    {
+      "epoch": 0.34244,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0123687744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1787,
+      "loss/crossentropy": 1.9109066724777222,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20130306482315063,
+      "step": 17122
+    },
+    {
+      "epoch": 0.34248,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.010944620768229166,
+      "learning_rate": 0.0001,
+      "loss": 3.7531,
+      "loss/crossentropy": 2.1049917936325073,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20878130197525024,
+      "step": 17124
+    },
+    {
+      "epoch": 0.34252,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.010008748372395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0951,
+      "loss/crossentropy": 1.9087567925453186,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18959704041481018,
+      "step": 17126
+    },
+    {
+      "epoch": 0.34256,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010033162434895833,
+      "learning_rate": 0.0001,
+      "loss": 4.3854,
+      "loss/crossentropy": 2.491095781326294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21739569306373596,
+      "step": 17128
+    },
+    {
+      "epoch": 0.3426,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.0081207275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0227,
+      "loss/crossentropy": 2.045976400375366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18868304044008255,
+      "step": 17130
+    },
+    {
+      "epoch": 0.34264,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009318033854166666,
+      "learning_rate": 0.0001,
+      "loss": 4.1221,
+      "loss/crossentropy": 2.021378517150879,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1955355852842331,
+      "step": 17132
+    },
+    {
+      "epoch": 0.34268,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.010247548421223959,
+      "learning_rate": 0.0001,
+      "loss": 4.0244,
+      "loss/crossentropy": 2.052547872066498,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18654850870370865,
+      "step": 17134
+    },
+    {
+      "epoch": 0.34272,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.008886464436848958,
+      "learning_rate": 0.0001,
+      "loss": 4.2536,
+      "loss/crossentropy": 2.1709023118019104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21335267275571823,
+      "step": 17136
+    },
+    {
+      "epoch": 0.34276,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.010477701822916666,
+      "learning_rate": 0.0001,
+      "loss": 4.1682,
+      "loss/crossentropy": 2.310550093650818,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18719495832920074,
+      "step": 17138
+    },
+    {
+      "epoch": 0.3428,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.011262003580729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0145,
+      "loss/crossentropy": 1.9357584714889526,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1914527416229248,
+      "step": 17140
+    },
+    {
+      "epoch": 0.34284,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.012684885660807292,
+      "learning_rate": 0.0001,
+      "loss": 3.9673,
+      "loss/crossentropy": 1.7561541199684143,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1711764633655548,
+      "step": 17142
+    },
+    {
+      "epoch": 0.34288,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.7546994527180989,
+      "learning_rate": 0.0001,
+      "loss": 4.562,
+      "loss/crossentropy": 2.4081480503082275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2405518889427185,
+      "step": 17144
+    },
+    {
+      "epoch": 0.34292,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.7456451416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.4023,
+      "loss/crossentropy": 2.44531512260437,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23487353324890137,
+      "step": 17146
+    },
+    {
+      "epoch": 0.34296,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.7445696512858073,
+      "learning_rate": 0.0001,
+      "loss": 4.0845,
+      "loss/crossentropy": 1.9712305068969727,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20045534521341324,
+      "step": 17148
+    },
+    {
+      "epoch": 0.343,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.7468706766764323,
+      "learning_rate": 0.0001,
+      "loss": 3.8479,
+      "loss/crossentropy": 1.9718485474586487,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18550659716129303,
+      "step": 17150
+    },
+    {
+      "epoch": 0.34304,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.7527414957682291,
+      "learning_rate": 0.0001,
+      "loss": 4.2708,
+      "loss/crossentropy": 2.1901930570602417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2082839384675026,
+      "step": 17152
+    },
+    {
+      "epoch": 0.34308,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.7463905334472656,
+      "learning_rate": 0.0001,
+      "loss": 4.2104,
+      "loss/crossentropy": 2.098864793777466,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19527551531791687,
+      "step": 17154
+    },
+    {
+      "epoch": 0.34312,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.7467437744140625,
+      "learning_rate": 0.0001,
+      "loss": 3.9873,
+      "loss/crossentropy": 1.7448402643203735,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1862412989139557,
+      "step": 17156
+    },
+    {
+      "epoch": 0.34316,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.7429603576660156,
+      "learning_rate": 0.0001,
+      "loss": 4.1016,
+      "loss/crossentropy": 1.9973859190940857,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19645331799983978,
+      "step": 17158
+    },
+    {
+      "epoch": 0.3432,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.016874186197916665,
+      "learning_rate": 0.0001,
+      "loss": 4.2022,
+      "loss/crossentropy": 2.2570079565048218,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20290197432041168,
+      "step": 17160
+    },
+    {
+      "epoch": 0.34324,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.008642323811848958,
+      "learning_rate": 0.0001,
+      "loss": 4.2304,
+      "loss/crossentropy": 1.7943695187568665,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1924503892660141,
+      "step": 17162
+    },
+    {
+      "epoch": 0.34328,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.008225250244140624,
+      "learning_rate": 0.0001,
+      "loss": 4.2142,
+      "loss/crossentropy": 2.1875799894332886,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1968172788619995,
+      "step": 17164
+    },
+    {
+      "epoch": 0.34332,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.007477823893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1844,
+      "loss/crossentropy": 2.242374360561371,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21981361508369446,
+      "step": 17166
+    },
+    {
+      "epoch": 0.34336,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.006754302978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0406,
+      "loss/crossentropy": 1.8542492985725403,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19412916153669357,
+      "step": 17168
+    },
+    {
+      "epoch": 0.3434,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0067860921223958336,
+      "learning_rate": 0.0001,
+      "loss": 3.8804,
+      "loss/crossentropy": 1.9377904534339905,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20015903562307358,
+      "step": 17170
+    },
+    {
+      "epoch": 0.34344,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.006319173177083333,
+      "learning_rate": 0.0001,
+      "loss": 4.1691,
+      "loss/crossentropy": 2.2250781655311584,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1943892240524292,
+      "step": 17172
+    },
+    {
+      "epoch": 0.34348,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0054094950358072914,
+      "learning_rate": 0.0001,
+      "loss": 4.0274,
+      "loss/crossentropy": 2.0463303923606873,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1931215077638626,
+      "step": 17174
+    },
+    {
+      "epoch": 0.34352,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.004984537760416667,
+      "learning_rate": 0.0001,
+      "loss": 3.826,
+      "loss/crossentropy": 2.0576277375221252,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24007725715637207,
+      "step": 17176
+    },
+    {
+      "epoch": 0.34356,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.003999582926432292,
+      "learning_rate": 0.0001,
+      "loss": 3.9865,
+      "loss/crossentropy": 1.8423307538032532,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17171106487512589,
+      "step": 17178
+    },
+    {
+      "epoch": 0.3436,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0053375244140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3089,
+      "loss/crossentropy": 2.05659818649292,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20058965682983398,
+      "step": 17180
+    },
+    {
+      "epoch": 0.34364,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0055620829264322914,
+      "learning_rate": 0.0001,
+      "loss": 3.8965,
+      "loss/crossentropy": 1.926946997642517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1770440638065338,
+      "step": 17182
+    },
+    {
+      "epoch": 0.34368,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.005098215738932292,
+      "learning_rate": 0.0001,
+      "loss": 3.9781,
+      "loss/crossentropy": 1.8181686401367188,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19718249142169952,
+      "step": 17184
+    },
+    {
+      "epoch": 0.34372,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.006624094645182292,
+      "learning_rate": 0.0001,
+      "loss": 4.0732,
+      "loss/crossentropy": 2.021128237247467,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1968582272529602,
+      "step": 17186
+    },
+    {
+      "epoch": 0.34376,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.008365631103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2861,
+      "loss/crossentropy": 2.04776668548584,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22718969732522964,
+      "step": 17188
+    },
+    {
+      "epoch": 0.3438,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.008957672119140624,
+      "learning_rate": 0.0001,
+      "loss": 4.2196,
+      "loss/crossentropy": 2.2225213050842285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20369096100330353,
+      "step": 17190
+    },
+    {
+      "epoch": 0.34384,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.008089192708333333,
+      "learning_rate": 0.0001,
+      "loss": 4.1137,
+      "loss/crossentropy": 1.8544179201126099,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20437808334827423,
+      "step": 17192
+    },
+    {
+      "epoch": 0.34388,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.006730143229166667,
+      "learning_rate": 0.0001,
+      "loss": 4.387,
+      "loss/crossentropy": 2.1270517110824585,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1997273936867714,
+      "step": 17194
+    },
+    {
+      "epoch": 0.34392,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0060791015625,
+      "learning_rate": 0.0001,
+      "loss": 3.8062,
+      "loss/crossentropy": 1.8699182271957397,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19382557272911072,
+      "step": 17196
+    },
+    {
+      "epoch": 0.34396,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.006941731770833333,
+      "learning_rate": 0.0001,
+      "loss": 3.9864,
+      "loss/crossentropy": 1.8734498023986816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18053626269102097,
+      "step": 17198
+    },
+    {
+      "epoch": 0.344,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.008097330729166666,
+      "learning_rate": 0.0001,
+      "loss": 3.9234,
+      "loss/crossentropy": 1.9386133551597595,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1851777583360672,
+      "step": 17200
+    },
+    {
+      "epoch": 0.34404,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0069163004557291664,
+      "learning_rate": 0.0001,
+      "loss": 3.9619,
+      "loss/crossentropy": 2.1074278354644775,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22176695615053177,
+      "step": 17202
+    },
+    {
+      "epoch": 0.34408,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.005895741780598958,
+      "learning_rate": 0.0001,
+      "loss": 4.1231,
+      "loss/crossentropy": 1.826387107372284,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19368141889572144,
+      "step": 17204
+    },
+    {
+      "epoch": 0.34412,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.005680084228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1367,
+      "loss/crossentropy": 1.9109328389167786,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2023227959871292,
+      "step": 17206
+    },
+    {
+      "epoch": 0.34416,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.006517537434895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1495,
+      "loss/crossentropy": 1.8752732872962952,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17867545038461685,
+      "step": 17208
+    },
+    {
+      "epoch": 0.3442,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.00972900390625,
+      "learning_rate": 0.0001,
+      "loss": 3.9998,
+      "loss/crossentropy": 1.95436429977417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19947312772274017,
+      "step": 17210
+    },
+    {
+      "epoch": 0.34424,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.00947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0074,
+      "loss/crossentropy": 1.9237809777259827,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2055458500981331,
+      "step": 17212
+    },
+    {
+      "epoch": 0.34428,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.007995351155598959,
+      "learning_rate": 0.0001,
+      "loss": 4.1351,
+      "loss/crossentropy": 2.4497138261795044,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23067011684179306,
+      "step": 17214
+    },
+    {
+      "epoch": 0.34432,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.007791900634765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9144,
+      "loss/crossentropy": 2.33384370803833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22732949256896973,
+      "step": 17216
+    },
+    {
+      "epoch": 0.34436,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0075439453125,
+      "learning_rate": 0.0001,
+      "loss": 3.9414,
+      "loss/crossentropy": 2.022156059741974,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2063901051878929,
+      "step": 17218
+    },
+    {
+      "epoch": 0.3444,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.007614898681640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0642,
+      "loss/crossentropy": 2.183963179588318,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21098726987838745,
+      "step": 17220
+    },
+    {
+      "epoch": 0.34444,
+      "grad_norm": 6.875,
+      "grad_norm_var": 1.501227823893229,
+      "learning_rate": 0.0001,
+      "loss": 4.3139,
+      "loss/crossentropy": 2.256633758544922,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21449647843837738,
+      "step": 17222
+    },
+    {
+      "epoch": 0.34448,
+      "grad_norm": 14.3125,
+      "grad_norm_var": 10.478612263997396,
+      "learning_rate": 0.0001,
+      "loss": 3.838,
+      "loss/crossentropy": 1.8303287625312805,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18412896245718002,
+      "step": 17224
+    },
+    {
+      "epoch": 0.34452,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 10.405972290039063,
+      "learning_rate": 0.0001,
+      "loss": 4.1852,
+      "loss/crossentropy": 2.3248833417892456,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23459070920944214,
+      "step": 17226
+    },
+    {
+      "epoch": 0.34456,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 10.418485514322917,
+      "learning_rate": 0.0001,
+      "loss": 4.062,
+      "loss/crossentropy": 2.018009066581726,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21266046166419983,
+      "step": 17228
+    },
+    {
+      "epoch": 0.3446,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 10.434609985351562,
+      "learning_rate": 0.0001,
+      "loss": 4.186,
+      "loss/crossentropy": 2.337665557861328,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.225304514169693,
+      "step": 17230
+    },
+    {
+      "epoch": 0.34464,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 10.453043365478516,
+      "learning_rate": 0.0001,
+      "loss": 3.86,
+      "loss/crossentropy": 1.8581790924072266,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18770471215248108,
+      "step": 17232
+    },
+    {
+      "epoch": 0.34468,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 10.462247721354167,
+      "learning_rate": 0.0001,
+      "loss": 3.9213,
+      "loss/crossentropy": 1.9208187460899353,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19106490910053253,
+      "step": 17234
+    },
+    {
+      "epoch": 0.34472,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 10.454630279541016,
+      "learning_rate": 0.0001,
+      "loss": 3.8284,
+      "loss/crossentropy": 1.8844050765037537,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17407642304897308,
+      "step": 17236
+    },
+    {
+      "epoch": 0.34476,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 9.497085571289062,
+      "learning_rate": 0.0001,
+      "loss": 3.7423,
+      "loss/crossentropy": 2.0443355441093445,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.198170468211174,
+      "step": 17238
+    },
+    {
+      "epoch": 0.3448,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.027342732747395834,
+      "learning_rate": 0.0001,
+      "loss": 4.0965,
+      "loss/crossentropy": 1.8563520908355713,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1958291083574295,
+      "step": 17240
+    },
+    {
+      "epoch": 0.34484,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0127349853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3028,
+      "loss/crossentropy": 2.1252214908599854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22386425733566284,
+      "step": 17242
+    },
+    {
+      "epoch": 0.34488,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.011848958333333333,
+      "learning_rate": 0.0001,
+      "loss": 3.926,
+      "loss/crossentropy": 2.1379681825637817,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19781935214996338,
+      "step": 17244
+    },
+    {
+      "epoch": 0.34492,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.013529205322265625,
+      "learning_rate": 0.0001,
+      "loss": 3.7403,
+      "loss/crossentropy": 1.9774981141090393,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18523120880126953,
+      "step": 17246
+    },
+    {
+      "epoch": 0.34496,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0120025634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2668,
+      "loss/crossentropy": 2.3234479427337646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21939057111740112,
+      "step": 17248
+    },
+    {
+      "epoch": 0.345,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011706288655598958,
+      "learning_rate": 0.0001,
+      "loss": 4.11,
+      "loss/crossentropy": 2.0521583557128906,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20204473286867142,
+      "step": 17250
+    },
+    {
+      "epoch": 0.34504,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.03452860514322917,
+      "learning_rate": 0.0001,
+      "loss": 4.2184,
+      "loss/crossentropy": 2.486477255821228,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21673013269901276,
+      "step": 17252
+    },
+    {
+      "epoch": 0.34508,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.035456339518229164,
+      "learning_rate": 0.0001,
+      "loss": 4.0215,
+      "loss/crossentropy": 2.285652995109558,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21322200447320938,
+      "step": 17254
+    },
+    {
+      "epoch": 0.34512,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.03436686197916667,
+      "learning_rate": 0.0001,
+      "loss": 4.1855,
+      "loss/crossentropy": 2.0946747064590454,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19791530817747116,
+      "step": 17256
+    },
+    {
+      "epoch": 0.34516,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.03250732421875,
+      "learning_rate": 0.0001,
+      "loss": 4.1129,
+      "loss/crossentropy": 1.7509565353393555,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16754039376974106,
+      "step": 17258
+    },
+    {
+      "epoch": 0.3452,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0325347900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3411,
+      "loss/crossentropy": 2.3828574419021606,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22555553913116455,
+      "step": 17260
+    },
+    {
+      "epoch": 0.34524,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.029060872395833333,
+      "learning_rate": 0.0001,
+      "loss": 4.215,
+      "loss/crossentropy": 1.997941255569458,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1815095767378807,
+      "step": 17262
+    },
+    {
+      "epoch": 0.34528,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.0296630859375,
+      "learning_rate": 0.0001,
+      "loss": 4.3814,
+      "loss/crossentropy": 2.3036177158355713,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2305641770362854,
+      "step": 17264
+    },
+    {
+      "epoch": 0.34532,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.03178075154622396,
+      "learning_rate": 0.0001,
+      "loss": 4.1668,
+      "loss/crossentropy": 2.202543616294861,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19594155251979828,
+      "step": 17266
+    },
+    {
+      "epoch": 0.34536,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.011199696858723959,
+      "learning_rate": 0.0001,
+      "loss": 4.1792,
+      "loss/crossentropy": 2.1162038445472717,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1945410966873169,
+      "step": 17268
+    },
+    {
+      "epoch": 0.3454,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.009883626302083334,
+      "learning_rate": 0.0001,
+      "loss": 4.1109,
+      "loss/crossentropy": 2.273344039916992,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24327096343040466,
+      "step": 17270
+    },
+    {
+      "epoch": 0.34544,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.011351521809895833,
+      "learning_rate": 0.0001,
+      "loss": 3.8254,
+      "loss/crossentropy": 1.851362407207489,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18494703620672226,
+      "step": 17272
+    },
+    {
+      "epoch": 0.34548,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.01114501953125,
+      "learning_rate": 0.0001,
+      "loss": 4.0846,
+      "loss/crossentropy": 1.9861729145050049,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20077189058065414,
+      "step": 17274
+    },
+    {
+      "epoch": 0.34552,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.011716461181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0692,
+      "loss/crossentropy": 2.0234099626541138,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20506415516138077,
+      "step": 17276
+    },
+    {
+      "epoch": 0.34556,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.012326812744140625,
+      "learning_rate": 0.0001,
+      "loss": 3.9843,
+      "loss/crossentropy": 2.1666045784950256,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18108145147562027,
+      "step": 17278
+    },
+    {
+      "epoch": 0.3456,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.0060536702473958336,
+      "learning_rate": 0.0001,
+      "loss": 4.0877,
+      "loss/crossentropy": 2.066905975341797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2112976461648941,
+      "step": 17280
+    },
+    {
+      "epoch": 0.34564,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0059722900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.224,
+      "loss/crossentropy": 2.1789294481277466,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20467212051153183,
+      "step": 17282
+    },
+    {
+      "epoch": 0.34568,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.005736287434895833,
+      "learning_rate": 0.0001,
+      "loss": 3.9716,
+      "loss/crossentropy": 1.8310245275497437,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1974945366382599,
+      "step": 17284
+    },
+    {
+      "epoch": 0.34572,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.006563059488932292,
+      "learning_rate": 0.0001,
+      "loss": 4.1832,
+      "loss/crossentropy": 2.033496856689453,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1989584118127823,
+      "step": 17286
+    },
+    {
+      "epoch": 0.34576,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0091705322265625,
+      "learning_rate": 0.0001,
+      "loss": 3.8748,
+      "loss/crossentropy": 1.8883211016654968,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1889789029955864,
+      "step": 17288
+    },
+    {
+      "epoch": 0.3458,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008495076497395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1637,
+      "loss/crossentropy": 2.4520593881607056,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2183564007282257,
+      "step": 17290
+    },
+    {
+      "epoch": 0.34584,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.008272298177083333,
+      "learning_rate": 0.0001,
+      "loss": 3.9375,
+      "loss/crossentropy": 2.030746340751648,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19382008165121078,
+      "step": 17292
+    },
+    {
+      "epoch": 0.34588,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.007295735677083333,
+      "learning_rate": 0.0001,
+      "loss": 4.2182,
+      "loss/crossentropy": 1.9897491931915283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21130450069904327,
+      "step": 17294
+    },
+    {
+      "epoch": 0.34592,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.006883748372395833,
+      "learning_rate": 0.0001,
+      "loss": 3.982,
+      "loss/crossentropy": 1.8000388145446777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20359987765550613,
+      "step": 17296
+    },
+    {
+      "epoch": 0.34596,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0065610249837239586,
+      "learning_rate": 0.0001,
+      "loss": 4.0341,
+      "loss/crossentropy": 2.0408164262771606,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18042601644992828,
+      "step": 17298
+    },
+    {
+      "epoch": 0.346,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.006742350260416667,
+      "learning_rate": 0.0001,
+      "loss": 4.0849,
+      "loss/crossentropy": 2.28415846824646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20799801498651505,
+      "step": 17300
+    },
+    {
+      "epoch": 0.34604,
+      "grad_norm": 2.25,
+      "grad_norm_var": 0.010925038655598959,
+      "learning_rate": 0.0001,
+      "loss": 4.3258,
+      "loss/crossentropy": 1.9420422315597534,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20019559562206268,
+      "step": 17302
+    },
+    {
+      "epoch": 0.34608,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0102691650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1892,
+      "loss/crossentropy": 2.100687026977539,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21129299700260162,
+      "step": 17304
+    },
+    {
+      "epoch": 0.34612,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.010970052083333333,
+      "learning_rate": 0.0001,
+      "loss": 4.0209,
+      "loss/crossentropy": 1.8657938241958618,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2081414759159088,
+      "step": 17306
+    },
+    {
+      "epoch": 0.34616,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.012853749593098958,
+      "learning_rate": 0.0001,
+      "loss": 4.025,
+      "loss/crossentropy": 2.0629169940948486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21117018163204193,
+      "step": 17308
+    },
+    {
+      "epoch": 0.3462,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0126129150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.159,
+      "loss/crossentropy": 2.2877084016799927,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1959577053785324,
+      "step": 17310
+    },
+    {
+      "epoch": 0.34624,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.012280019124348958,
+      "learning_rate": 0.0001,
+      "loss": 4.1135,
+      "loss/crossentropy": 2.2501026391983032,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2108026072382927,
+      "step": 17312
+    },
+    {
+      "epoch": 0.34628,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.012011464436848958,
+      "learning_rate": 0.0001,
+      "loss": 4.1498,
+      "loss/crossentropy": 1.9485750794410706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2126113697886467,
+      "step": 17314
+    },
+    {
+      "epoch": 0.34632,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.011500803629557292,
+      "learning_rate": 0.0001,
+      "loss": 4.13,
+      "loss/crossentropy": 2.193112373352051,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19680321961641312,
+      "step": 17316
+    },
+    {
+      "epoch": 0.34636,
+      "grad_norm": 1.796875,
+      "grad_norm_var": 0.010084788004557291,
+      "learning_rate": 0.0001,
+      "loss": 4.1181,
+      "loss/crossentropy": 2.30054771900177,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2211720421910286,
+      "step": 17318
+    },
+    {
+      "epoch": 0.3464,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.026869455973307293,
+      "learning_rate": 0.0001,
+      "loss": 4.2737,
+      "loss/crossentropy": 2.0091487169265747,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19040243327617645,
+      "step": 17320
+    },
+    {
+      "epoch": 0.34644,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.036628977457682295,
+      "learning_rate": 0.0001,
+      "loss": 4.2257,
+      "loss/crossentropy": 2.1841423511505127,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2112671434879303,
+      "step": 17322
+    },
+    {
+      "epoch": 0.34648,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.03555501302083333,
+      "learning_rate": 0.0001,
+      "loss": 3.8837,
+      "loss/crossentropy": 2.092822253704071,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2037336230278015,
+      "step": 17324
+    },
+    {
+      "epoch": 0.34652,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0335601806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0322,
+      "loss/crossentropy": 2.280580163002014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20753052085638046,
+      "step": 17326
+    },
+    {
+      "epoch": 0.34656,
+      "grad_norm": 1.7890625,
+      "grad_norm_var": 0.03884048461914062,
+      "learning_rate": 0.0001,
+      "loss": 4.1206,
+      "loss/crossentropy": 1.60361909866333,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15820877254009247,
+      "step": 17328
+    },
+    {
+      "epoch": 0.3466,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.04915949503580729,
+      "learning_rate": 0.0001,
+      "loss": 3.6214,
+      "loss/crossentropy": 1.608814001083374,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16808264702558517,
+      "step": 17330
+    },
+    {
+      "epoch": 0.34664,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.06932373046875,
+      "learning_rate": 0.0001,
+      "loss": 4.4983,
+      "loss/crossentropy": 2.3868669271469116,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23717772960662842,
+      "step": 17332
+    },
+    {
+      "epoch": 0.34668,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.06720759073893229,
+      "learning_rate": 0.0001,
+      "loss": 4.0862,
+      "loss/crossentropy": 2.1908692717552185,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21705647557973862,
+      "step": 17334
+    },
+    {
+      "epoch": 0.34672,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.05272191365559896,
+      "learning_rate": 0.0001,
+      "loss": 4.2436,
+      "loss/crossentropy": 2.459980010986328,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24167515337467194,
+      "step": 17336
+    },
+    {
+      "epoch": 0.34676,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.04353612263997396,
+      "learning_rate": 0.0001,
+      "loss": 4.219,
+      "loss/crossentropy": 2.221674919128418,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2097555324435234,
+      "step": 17338
+    },
+    {
+      "epoch": 0.3468,
+      "grad_norm": 1.78125,
+      "grad_norm_var": 0.04521458943684896,
+      "learning_rate": 0.0001,
+      "loss": 3.7565,
+      "loss/crossentropy": 2.2840858697891235,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2250964418053627,
+      "step": 17340
+    },
+    {
+      "epoch": 0.34684,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.04426854451497396,
+      "learning_rate": 0.0001,
+      "loss": 3.6803,
+      "loss/crossentropy": 2.0715824365615845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2149435505270958,
+      "step": 17342
+    },
+    {
+      "epoch": 0.34688,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0420806884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1017,
+      "loss/crossentropy": 2.2389899492263794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22377658635377884,
+      "step": 17344
+    },
+    {
+      "epoch": 0.34692,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.035278065999348955,
+      "learning_rate": 0.0001,
+      "loss": 3.9827,
+      "loss/crossentropy": 2.0308732986450195,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2074751853942871,
+      "step": 17346
+    },
+    {
+      "epoch": 0.34696,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.010864003499348959,
+      "learning_rate": 0.0001,
+      "loss": 3.7901,
+      "loss/crossentropy": 1.8123770356178284,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19080865383148193,
+      "step": 17348
+    },
+    {
+      "epoch": 0.347,
+      "grad_norm": 1.734375,
+      "grad_norm_var": 0.010367838541666667,
+      "learning_rate": 0.0001,
+      "loss": 3.8772,
+      "loss/crossentropy": 1.884379267692566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1799243837594986,
+      "step": 17350
+    },
+    {
+      "epoch": 0.34704,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.012442779541015626,
+      "learning_rate": 0.0001,
+      "loss": 4.4317,
+      "loss/crossentropy": 2.237663149833679,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20197343081235886,
+      "step": 17352
+    },
+    {
+      "epoch": 0.34708,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.013337961832682292,
+      "learning_rate": 0.0001,
+      "loss": 3.7444,
+      "loss/crossentropy": 1.9563414454460144,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18722663074731827,
+      "step": 17354
+    },
+    {
+      "epoch": 0.34712,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.020817057291666666,
+      "learning_rate": 0.0001,
+      "loss": 4.2799,
+      "loss/crossentropy": 2.1188591718673706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21183937788009644,
+      "step": 17356
+    },
+    {
+      "epoch": 0.34716,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0219146728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.4551,
+      "loss/crossentropy": 2.276149272918701,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2232571393251419,
+      "step": 17358
+    },
+    {
+      "epoch": 0.3472,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.02123998006184896,
+      "learning_rate": 0.0001,
+      "loss": 4.4329,
+      "loss/crossentropy": 2.360138177871704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22683367878198624,
+      "step": 17360
+    },
+    {
+      "epoch": 0.34724,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.020576985677083333,
+      "learning_rate": 0.0001,
+      "loss": 4.2762,
+      "loss/crossentropy": 1.99272620677948,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18479204177856445,
+      "step": 17362
+    },
+    {
+      "epoch": 0.34728,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.019760894775390624,
+      "learning_rate": 0.0001,
+      "loss": 4.2303,
+      "loss/crossentropy": 2.053771197795868,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20591863989830017,
+      "step": 17364
+    },
+    {
+      "epoch": 0.34732,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.016428375244140626,
+      "learning_rate": 0.0001,
+      "loss": 3.9703,
+      "loss/crossentropy": 2.1311055421829224,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2049081027507782,
+      "step": 17366
+    },
+    {
+      "epoch": 0.34736,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.01246337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1102,
+      "loss/crossentropy": 2.0712148547172546,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19809392094612122,
+      "step": 17368
+    },
+    {
+      "epoch": 0.3474,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.009468587239583333,
+      "learning_rate": 0.0001,
+      "loss": 4.0732,
+      "loss/crossentropy": 2.144728183746338,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21964795887470245,
+      "step": 17370
+    },
+    {
+      "epoch": 0.34744,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.003856404622395833,
+      "learning_rate": 0.0001,
+      "loss": 3.9145,
+      "loss/crossentropy": 2.024275004863739,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19414371997117996,
+      "step": 17372
+    },
+    {
+      "epoch": 0.34748,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.00400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0925,
+      "loss/crossentropy": 2.460733413696289,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2185375839471817,
+      "step": 17374
+    },
+    {
+      "epoch": 0.34752,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.010423787434895833,
+      "learning_rate": 0.0001,
+      "loss": 4.308,
+      "loss/crossentropy": 2.3502203226089478,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2140796035528183,
+      "step": 17376
+    },
+    {
+      "epoch": 0.34756,
+      "grad_norm": 1.78125,
+      "grad_norm_var": 0.012547810872395834,
+      "learning_rate": 0.0001,
+      "loss": 4.0238,
+      "loss/crossentropy": 2.2896264791488647,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2149006575345993,
+      "step": 17378
+    },
+    {
+      "epoch": 0.3476,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.013118489583333334,
+      "learning_rate": 0.0001,
+      "loss": 3.9757,
+      "loss/crossentropy": 2.101209044456482,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20842310786247253,
+      "step": 17380
+    },
+    {
+      "epoch": 0.34764,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0189361572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3287,
+      "loss/crossentropy": 1.8995551466941833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1851196512579918,
+      "step": 17382
+    },
+    {
+      "epoch": 0.34768,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.01887995402018229,
+      "learning_rate": 0.0001,
+      "loss": 3.9219,
+      "loss/crossentropy": 1.850643277168274,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16962846368551254,
+      "step": 17384
+    },
+    {
+      "epoch": 0.34772,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.019528961181640624,
+      "learning_rate": 0.0001,
+      "loss": 4.2478,
+      "loss/crossentropy": 2.1131063103675842,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19152897596359253,
+      "step": 17386
+    },
+    {
+      "epoch": 0.34776,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.020140584309895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1932,
+      "loss/crossentropy": 2.172673463821411,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20645174384117126,
+      "step": 17388
+    },
+    {
+      "epoch": 0.3478,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.019147745768229165,
+      "learning_rate": 0.0001,
+      "loss": 4.269,
+      "loss/crossentropy": 2.431947708129883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20513125509023666,
+      "step": 17390
+    },
+    {
+      "epoch": 0.34784,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.014461008707682292,
+      "learning_rate": 0.0001,
+      "loss": 3.9314,
+      "loss/crossentropy": 2.062328279018402,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1929447129368782,
+      "step": 17392
+    },
+    {
+      "epoch": 0.34788,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010609690348307292,
+      "learning_rate": 0.0001,
+      "loss": 4.2277,
+      "loss/crossentropy": 2.189133882522583,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20945163816213608,
+      "step": 17394
+    },
+    {
+      "epoch": 0.34792,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.010846964518229167,
+      "learning_rate": 0.0001,
+      "loss": 3.9426,
+      "loss/crossentropy": 2.070523977279663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20401575416326523,
+      "step": 17396
+    },
+    {
+      "epoch": 0.34796,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0075266520182291664,
+      "learning_rate": 0.0001,
+      "loss": 4.0668,
+      "loss/crossentropy": 2.3387876749038696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19960296899080276,
+      "step": 17398
+    },
+    {
+      "epoch": 0.348,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.00955810546875,
+      "learning_rate": 0.0001,
+      "loss": 4.0225,
+      "loss/crossentropy": 1.9857355952262878,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19195467978715897,
+      "step": 17400
+    },
+    {
+      "epoch": 0.34804,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.009041086832682291,
+      "learning_rate": 0.0001,
+      "loss": 4.0715,
+      "loss/crossentropy": 2.2523770332336426,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22015457600355148,
+      "step": 17402
+    },
+    {
+      "epoch": 0.34808,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.0074045817057291664,
+      "learning_rate": 0.0001,
+      "loss": 3.9181,
+      "loss/crossentropy": 1.7088012099266052,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1644933521747589,
+      "step": 17404
+    },
+    {
+      "epoch": 0.34812,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.006772613525390625,
+      "learning_rate": 0.0001,
+      "loss": 3.8759,
+      "loss/crossentropy": 2.022001802921295,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19237526506185532,
+      "step": 17406
+    },
+    {
+      "epoch": 0.34816,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.015669504801432293,
+      "learning_rate": 0.0001,
+      "loss": 4.2196,
+      "loss/crossentropy": 2.240332841873169,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2214636355638504,
+      "step": 17408
+    },
+    {
+      "epoch": 0.3482,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.015099843343098959,
+      "learning_rate": 0.0001,
+      "loss": 4.0211,
+      "loss/crossentropy": 2.1392452716827393,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21769221127033234,
+      "step": 17410
+    },
+    {
+      "epoch": 0.34824,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.016078440348307292,
+      "learning_rate": 0.0001,
+      "loss": 4.0243,
+      "loss/crossentropy": 1.9182460308074951,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.202327162027359,
+      "step": 17412
+    },
+    {
+      "epoch": 0.34828,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0153961181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0654,
+      "loss/crossentropy": 2.036426305770874,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19997400790452957,
+      "step": 17414
+    },
+    {
+      "epoch": 0.34832,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.014353179931640625,
+      "learning_rate": 0.0001,
+      "loss": 3.9086,
+      "loss/crossentropy": 2.0429354906082153,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20519915968179703,
+      "step": 17416
+    },
+    {
+      "epoch": 0.34836,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.013923136393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2742,
+      "loss/crossentropy": 2.3943710327148438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2106815129518509,
+      "step": 17418
+    },
+    {
+      "epoch": 0.3484,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0130615234375,
+      "learning_rate": 0.0001,
+      "loss": 4.1596,
+      "loss/crossentropy": 2.060115098953247,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20672088861465454,
+      "step": 17420
+    },
+    {
+      "epoch": 0.34844,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.012471516927083334,
+      "learning_rate": 0.0001,
+      "loss": 4.2223,
+      "loss/crossentropy": 2.2820589542388916,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23306988179683685,
+      "step": 17422
+    },
+    {
+      "epoch": 0.34848,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.008788045247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1665,
+      "loss/crossentropy": 2.13969624042511,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20909114927053452,
+      "step": 17424
+    },
+    {
+      "epoch": 0.34852,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.011214192708333333,
+      "learning_rate": 0.0001,
+      "loss": 3.9572,
+      "loss/crossentropy": 2.0822505950927734,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20429539680480957,
+      "step": 17426
+    },
+    {
+      "epoch": 0.34856,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.009759267171223959,
+      "learning_rate": 0.0001,
+      "loss": 3.9435,
+      "loss/crossentropy": 2.512156844139099,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21776874363422394,
+      "step": 17428
+    },
+    {
+      "epoch": 0.3486,
+      "grad_norm": 2.359375,
+      "grad_norm_var": 0.018700917561848957,
+      "learning_rate": 0.0001,
+      "loss": 4.1145,
+      "loss/crossentropy": 2.0379759669303894,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20926732569932938,
+      "step": 17430
+    },
+    {
+      "epoch": 0.34864,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.017295074462890626,
+      "learning_rate": 0.0001,
+      "loss": 4.1998,
+      "loss/crossentropy": 2.443945050239563,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22597475349903107,
+      "step": 17432
+    },
+    {
+      "epoch": 0.34868,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.019576009114583334,
+      "learning_rate": 0.0001,
+      "loss": 3.8442,
+      "loss/crossentropy": 1.8295652866363525,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18034228682518005,
+      "step": 17434
+    },
+    {
+      "epoch": 0.34872,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.020428212483723958,
+      "learning_rate": 0.0001,
+      "loss": 4.0614,
+      "loss/crossentropy": 1.9487475156784058,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18921684473752975,
+      "step": 17436
+    },
+    {
+      "epoch": 0.34876,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.018721262613932293,
+      "learning_rate": 0.0001,
+      "loss": 4.1153,
+      "loss/crossentropy": 2.099911689758301,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2079886943101883,
+      "step": 17438
+    },
+    {
+      "epoch": 0.3488,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.016739654541015624,
+      "learning_rate": 0.0001,
+      "loss": 4.1762,
+      "loss/crossentropy": 2.52177894115448,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22080882638692856,
+      "step": 17440
+    },
+    {
+      "epoch": 0.34884,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.014900461832682291,
+      "learning_rate": 0.0001,
+      "loss": 4.1076,
+      "loss/crossentropy": 2.222637891769409,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22301562875509262,
+      "step": 17442
+    },
+    {
+      "epoch": 0.34888,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.014725494384765624,
+      "learning_rate": 0.0001,
+      "loss": 4.3965,
+      "loss/crossentropy": 2.2042930126190186,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2051277905702591,
+      "step": 17444
+    },
+    {
+      "epoch": 0.34892,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0065093994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2452,
+      "loss/crossentropy": 2.0599029064178467,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20651433616876602,
+      "step": 17446
+    },
+    {
+      "epoch": 0.34896,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.007027180989583334,
+      "learning_rate": 0.0001,
+      "loss": 3.9174,
+      "loss/crossentropy": 1.9319151639938354,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18495241552591324,
+      "step": 17448
+    },
+    {
+      "epoch": 0.349,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.006197102864583333,
+      "learning_rate": 0.0001,
+      "loss": 4.0844,
+      "loss/crossentropy": 2.341706871986389,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20252344757318497,
+      "step": 17450
+    },
+    {
+      "epoch": 0.34904,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.004833984375,
+      "learning_rate": 0.0001,
+      "loss": 4.0301,
+      "loss/crossentropy": 2.0161609053611755,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2176668643951416,
+      "step": 17452
+    },
+    {
+      "epoch": 0.34908,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.0062334696451822914,
+      "learning_rate": 0.0001,
+      "loss": 3.6046,
+      "loss/crossentropy": 1.7961083054542542,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16384318470954895,
+      "step": 17454
+    },
+    {
+      "epoch": 0.34912,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.01791966756184896,
+      "learning_rate": 0.0001,
+      "loss": 4.6054,
+      "loss/crossentropy": 2.108873188495636,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21137713640928268,
+      "step": 17456
+    },
+    {
+      "epoch": 0.34916,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.017830403645833333,
+      "learning_rate": 0.0001,
+      "loss": 3.9543,
+      "loss/crossentropy": 2.158664584159851,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19957562536001205,
+      "step": 17458
+    },
+    {
+      "epoch": 0.3492,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.018660227457682293,
+      "learning_rate": 0.0001,
+      "loss": 3.9454,
+      "loss/crossentropy": 1.7799381017684937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.193112812936306,
+      "step": 17460
+    },
+    {
+      "epoch": 0.34924,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.019128163655598957,
+      "learning_rate": 0.0001,
+      "loss": 4.4072,
+      "loss/crossentropy": 2.372501015663147,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2180323675274849,
+      "step": 17462
+    },
+    {
+      "epoch": 0.34928,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0185699462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0343,
+      "loss/crossentropy": 1.9338968396186829,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19117196649312973,
+      "step": 17464
+    },
+    {
+      "epoch": 0.34932,
+      "grad_norm": 1.8125,
+      "grad_norm_var": 0.020774078369140626,
+      "learning_rate": 0.0001,
+      "loss": 3.6429,
+      "loss/crossentropy": 1.4602742195129395,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1677519753575325,
+      "step": 17466
+    },
+    {
+      "epoch": 0.34936,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.02080078125,
+      "learning_rate": 0.0001,
+      "loss": 4.2498,
+      "loss/crossentropy": 2.1464006304740906,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21201838552951813,
+      "step": 17468
+    },
+    {
+      "epoch": 0.3494,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.018529256184895832,
+      "learning_rate": 0.0001,
+      "loss": 4.1736,
+      "loss/crossentropy": 2.071315884590149,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19322700798511505,
+      "step": 17470
+    },
+    {
+      "epoch": 0.34944,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.010309855143229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2653,
+      "loss/crossentropy": 1.9166500568389893,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23960981518030167,
+      "step": 17472
+    },
+    {
+      "epoch": 0.34948,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.011457316080729167,
+      "learning_rate": 0.0001,
+      "loss": 3.9433,
+      "loss/crossentropy": 1.8666390180587769,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1875123456120491,
+      "step": 17474
+    },
+    {
+      "epoch": 0.34952,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.009736887613932292,
+      "learning_rate": 0.0001,
+      "loss": 4.0984,
+      "loss/crossentropy": 1.9197289943695068,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21075090020895004,
+      "step": 17476
+    },
+    {
+      "epoch": 0.34956,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.011018625895182292,
+      "learning_rate": 0.0001,
+      "loss": 4.0971,
+      "loss/crossentropy": 1.904780387878418,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20412997901439667,
+      "step": 17478
+    },
+    {
+      "epoch": 0.3496,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.012325032552083334,
+      "learning_rate": 0.0001,
+      "loss": 3.735,
+      "loss/crossentropy": 1.6775096654891968,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1669272631406784,
+      "step": 17480
+    },
+    {
+      "epoch": 0.34964,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0089111328125,
+      "learning_rate": 0.0001,
+      "loss": 4.1039,
+      "loss/crossentropy": 2.050463318824768,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19382070004940033,
+      "step": 17482
+    },
+    {
+      "epoch": 0.34968,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.009224192301432291,
+      "learning_rate": 0.0001,
+      "loss": 3.8739,
+      "loss/crossentropy": 1.8124673962593079,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20674846321344376,
+      "step": 17484
+    },
+    {
+      "epoch": 0.34972,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.008969879150390625,
+      "learning_rate": 0.0001,
+      "loss": 3.6797,
+      "loss/crossentropy": 1.6487451791763306,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17874807119369507,
+      "step": 17486
+    },
+    {
+      "epoch": 0.34976,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.009430948893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.4025,
+      "loss/crossentropy": 2.101687431335449,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24856076389551163,
+      "step": 17488
+    },
+    {
+      "epoch": 0.3498,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.009205881754557292,
+      "learning_rate": 0.0001,
+      "loss": 4.0236,
+      "loss/crossentropy": 1.9233632683753967,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21046236902475357,
+      "step": 17490
+    },
+    {
+      "epoch": 0.34984,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.010589345296223959,
+      "learning_rate": 0.0001,
+      "loss": 4.0376,
+      "loss/crossentropy": 2.2882679104804993,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23568424582481384,
+      "step": 17492
+    },
+    {
+      "epoch": 0.34988,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.022342681884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.4205,
+      "loss/crossentropy": 2.205874502658844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21892941743135452,
+      "step": 17494
+    },
+    {
+      "epoch": 0.34992,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.0232177734375,
+      "learning_rate": 0.0001,
+      "loss": 4.3628,
+      "loss/crossentropy": 2.278464913368225,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22409500926733017,
+      "step": 17496
+    },
+    {
+      "epoch": 0.34996,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.026082356770833332,
+      "learning_rate": 0.0001,
+      "loss": 4.0736,
+      "loss/crossentropy": 2.5003273487091064,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22478827834129333,
+      "step": 17498
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.04237848917643229,
+      "learning_rate": 0.0001,
+      "loss": 4.3658,
+      "loss/crossentropy": 2.034249722957611,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19176610559225082,
+      "step": 17500
+    },
+    {
+      "epoch": 0.35004,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0401031494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1586,
+      "loss/crossentropy": 2.271065592765808,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21523310244083405,
+      "step": 17502
+    },
+    {
+      "epoch": 0.35008,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.03870824178059896,
+      "learning_rate": 0.0001,
+      "loss": 4.1198,
+      "loss/crossentropy": 2.182866334915161,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21311558783054352,
+      "step": 17504
+    },
+    {
+      "epoch": 0.35012,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.03878173828125,
+      "learning_rate": 0.0001,
+      "loss": 4.1471,
+      "loss/crossentropy": 1.7134016752243042,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16884544491767883,
+      "step": 17506
+    },
+    {
+      "epoch": 0.35016,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.03687718709309896,
+      "learning_rate": 0.0001,
+      "loss": 4.2464,
+      "loss/crossentropy": 2.198198080062866,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1916792169213295,
+      "step": 17508
+    },
+    {
+      "epoch": 0.3502,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0272125244140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2651,
+      "loss/crossentropy": 1.9370547533035278,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23538394272327423,
+      "step": 17510
+    },
+    {
+      "epoch": 0.35024,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.02469456990559896,
+      "learning_rate": 0.0001,
+      "loss": 4.0664,
+      "loss/crossentropy": 2.1743874549865723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22514399886131287,
+      "step": 17512
+    },
+    {
+      "epoch": 0.35028,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.02174657185872396,
+      "learning_rate": 0.0001,
+      "loss": 4.2767,
+      "loss/crossentropy": 1.8546866178512573,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1772822067141533,
+      "step": 17514
+    },
+    {
+      "epoch": 0.35032,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.004154205322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2603,
+      "loss/crossentropy": 1.8439211249351501,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19603434205055237,
+      "step": 17516
+    },
+    {
+      "epoch": 0.35036,
+      "grad_norm": 1.8125,
+      "grad_norm_var": 0.006843058268229166,
+      "learning_rate": 0.0001,
+      "loss": 3.9506,
+      "loss/crossentropy": 2.2717851400375366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21598497033119202,
+      "step": 17518
+    },
+    {
+      "epoch": 0.3504,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.006843058268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.1807,
+      "loss/crossentropy": 2.2645692825317383,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2229163572192192,
+      "step": 17520
+    },
+    {
+      "epoch": 0.35044,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.0071604410807291664,
+      "learning_rate": 0.0001,
+      "loss": 4.09,
+      "loss/crossentropy": 2.0507450103759766,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19268949329853058,
+      "step": 17522
+    },
+    {
+      "epoch": 0.35048,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.005521647135416667,
+      "learning_rate": 0.0001,
+      "loss": 4.1468,
+      "loss/crossentropy": 2.3831188678741455,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21931231766939163,
+      "step": 17524
+    },
+    {
+      "epoch": 0.35052,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.006281534830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.116,
+      "loss/crossentropy": 1.8574120998382568,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17972451448440552,
+      "step": 17526
+    },
+    {
+      "epoch": 0.35056,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.006013997395833333,
+      "learning_rate": 0.0001,
+      "loss": 3.9063,
+      "loss/crossentropy": 2.141213893890381,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21798025816679,
+      "step": 17528
+    },
+    {
+      "epoch": 0.3506,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.006520334879557292,
+      "learning_rate": 0.0001,
+      "loss": 3.9454,
+      "loss/crossentropy": 2.2224762439727783,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2031020149588585,
+      "step": 17530
+    },
+    {
+      "epoch": 0.35064,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.006476847330729166,
+      "learning_rate": 0.0001,
+      "loss": 3.765,
+      "loss/crossentropy": 2.163232743740082,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22002054750919342,
+      "step": 17532
+    },
+    {
+      "epoch": 0.35068,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0055653889973958336,
+      "learning_rate": 0.0001,
+      "loss": 4.1397,
+      "loss/crossentropy": 2.1463611721992493,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20084905624389648,
+      "step": 17534
+    },
+    {
+      "epoch": 0.35072,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.006566365559895833,
+      "learning_rate": 0.0001,
+      "loss": 3.9751,
+      "loss/crossentropy": 2.1863406896591187,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2150622010231018,
+      "step": 17536
+    },
+    {
+      "epoch": 0.35076,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.005744425455729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0958,
+      "loss/crossentropy": 2.2162610292434692,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20196689665317535,
+      "step": 17538
+    },
+    {
+      "epoch": 0.3508,
+      "grad_norm": 10.6875,
+      "grad_norm_var": 4.776464589436849,
+      "learning_rate": 0.0001,
+      "loss": 4.1004,
+      "loss/crossentropy": 1.8908233642578125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23238417506217957,
+      "step": 17540
+    },
+    {
+      "epoch": 0.35084,
+      "grad_norm": 2.125,
+      "grad_norm_var": 4.763142903645833,
+      "learning_rate": 0.0001,
+      "loss": 4.416,
+      "loss/crossentropy": 2.3578076362609863,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21626722812652588,
+      "step": 17542
+    },
+    {
+      "epoch": 0.35088,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 4.736722819010416,
+      "learning_rate": 0.0001,
+      "loss": 4.3243,
+      "loss/crossentropy": 2.4079915285110474,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2292388305068016,
+      "step": 17544
+    },
+    {
+      "epoch": 0.35092,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 4.717746734619141,
+      "learning_rate": 0.0001,
+      "loss": 4.2647,
+      "loss/crossentropy": 2.1021666526794434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20725515484809875,
+      "step": 17546
+    },
+    {
+      "epoch": 0.35096,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 4.707061513264974,
+      "learning_rate": 0.0001,
+      "loss": 4.0905,
+      "loss/crossentropy": 2.039289176464081,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20614883303642273,
+      "step": 17548
+    },
+    {
+      "epoch": 0.351,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 4.68468017578125,
+      "learning_rate": 0.0001,
+      "loss": 4.269,
+      "loss/crossentropy": 1.9731826782226562,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2164870798587799,
+      "step": 17550
+    },
+    {
+      "epoch": 0.35104,
+      "grad_norm": 1.875,
+      "grad_norm_var": 4.69991455078125,
+      "learning_rate": 0.0001,
+      "loss": 4.0785,
+      "loss/crossentropy": 2.0166266560554504,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1977868676185608,
+      "step": 17552
+    },
+    {
+      "epoch": 0.35108,
+      "grad_norm": 2.0,
+      "grad_norm_var": 4.692333730061849,
+      "learning_rate": 0.0001,
+      "loss": 4.2823,
+      "loss/crossentropy": 2.284690737724304,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2204381301999092,
+      "step": 17554
+    },
+    {
+      "epoch": 0.35112,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.012977854410807291,
+      "learning_rate": 0.0001,
+      "loss": 4.2629,
+      "loss/crossentropy": 2.293992757797241,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21718238294124603,
+      "step": 17556
+    },
+    {
+      "epoch": 0.35116,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.013133748372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.0687,
+      "loss/crossentropy": 2.252563714981079,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23132775723934174,
+      "step": 17558
+    },
+    {
+      "epoch": 0.3512,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.0097564697265625,
+      "learning_rate": 0.0001,
+      "loss": 3.977,
+      "loss/crossentropy": 1.8454533219337463,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20084110647439957,
+      "step": 17560
+    },
+    {
+      "epoch": 0.35124,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009616851806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1526,
+      "loss/crossentropy": 2.3160746097564697,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21168464422225952,
+      "step": 17562
+    },
+    {
+      "epoch": 0.35128,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.00953369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.031,
+      "loss/crossentropy": 2.0470253229141235,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20728407055139542,
+      "step": 17564
+    },
+    {
+      "epoch": 0.35132,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0064389546712239586,
+      "learning_rate": 0.0001,
+      "loss": 4.404,
+      "loss/crossentropy": 2.098679304122925,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2133883535861969,
+      "step": 17566
+    },
+    {
+      "epoch": 0.35136,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.005615234375,
+      "learning_rate": 0.0001,
+      "loss": 4.0509,
+      "loss/crossentropy": 1.975411057472229,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19393378496170044,
+      "step": 17568
+    },
+    {
+      "epoch": 0.3514,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.005606842041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2861,
+      "loss/crossentropy": 2.594779372215271,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2499995082616806,
+      "step": 17570
+    },
+    {
+      "epoch": 0.35144,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.005915323893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1045,
+      "loss/crossentropy": 2.075734496116638,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.207140251994133,
+      "step": 17572
+    },
+    {
+      "epoch": 0.35148,
+      "grad_norm": 1.765625,
+      "grad_norm_var": 0.0062978108723958336,
+      "learning_rate": 0.0001,
+      "loss": 3.869,
+      "loss/crossentropy": 2.124872624874115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20058108121156693,
+      "step": 17574
+    },
+    {
+      "epoch": 0.35152,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.012056477864583333,
+      "learning_rate": 0.0001,
+      "loss": 4.1552,
+      "loss/crossentropy": 2.0275574922561646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2105395644903183,
+      "step": 17576
+    },
+    {
+      "epoch": 0.35156,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.011913045247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.248,
+      "loss/crossentropy": 2.0694713592529297,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21342340111732483,
+      "step": 17578
+    },
+    {
+      "epoch": 0.3516,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.012239583333333333,
+      "learning_rate": 0.0001,
+      "loss": 3.9729,
+      "loss/crossentropy": 1.8013625741004944,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1766640990972519,
+      "step": 17580
+    },
+    {
+      "epoch": 0.35164,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0123443603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0313,
+      "loss/crossentropy": 2.1990463733673096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19608185440301895,
+      "step": 17582
+    },
+    {
+      "epoch": 0.35168,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.013361612955729166,
+      "learning_rate": 0.0001,
+      "loss": 3.9041,
+      "loss/crossentropy": 2.0714540481567383,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20064443349838257,
+      "step": 17584
+    },
+    {
+      "epoch": 0.35172,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.015242258707682291,
+      "learning_rate": 0.0001,
+      "loss": 3.7431,
+      "loss/crossentropy": 1.7437097430229187,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17001167684793472,
+      "step": 17586
+    },
+    {
+      "epoch": 0.35176,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.020531972249348957,
+      "learning_rate": 0.0001,
+      "loss": 4.1987,
+      "loss/crossentropy": 2.072261691093445,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21019883453845978,
+      "step": 17588
+    },
+    {
+      "epoch": 0.3518,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.021922810872395834,
+      "learning_rate": 0.0001,
+      "loss": 4.0826,
+      "loss/crossentropy": 2.1804715394973755,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21589763462543488,
+      "step": 17590
+    },
+    {
+      "epoch": 0.35184,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.01734619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0964,
+      "loss/crossentropy": 2.0044930577278137,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20204917341470718,
+      "step": 17592
+    },
+    {
+      "epoch": 0.35188,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.017362467447916665,
+      "learning_rate": 0.0001,
+      "loss": 4.1394,
+      "loss/crossentropy": 1.8961025476455688,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1865018978714943,
+      "step": 17594
+    },
+    {
+      "epoch": 0.35192,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.016971588134765625,
+      "learning_rate": 0.0001,
+      "loss": 3.8503,
+      "loss/crossentropy": 1.7005563378334045,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18951866030693054,
+      "step": 17596
+    },
+    {
+      "epoch": 0.35196,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.017002105712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0044,
+      "loss/crossentropy": 2.002982437610626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2076285257935524,
+      "step": 17598
+    },
+    {
+      "epoch": 0.352,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.016950480143229165,
+      "learning_rate": 0.0001,
+      "loss": 4.0595,
+      "loss/crossentropy": 2.0283620357513428,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20175684988498688,
+      "step": 17600
+    },
+    {
+      "epoch": 0.35204,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.014289347330729167,
+      "learning_rate": 0.0001,
+      "loss": 3.9495,
+      "loss/crossentropy": 1.855184018611908,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19368328154087067,
+      "step": 17602
+    },
+    {
+      "epoch": 0.35208,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.009089152018229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2364,
+      "loss/crossentropy": 2.2288308143615723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2121831700205803,
+      "step": 17604
+    },
+    {
+      "epoch": 0.35212,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.002512359619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1398,
+      "loss/crossentropy": 1.9114368557929993,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17937570065259933,
+      "step": 17606
+    },
+    {
+      "epoch": 0.35216,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.0026751200358072916,
+      "learning_rate": 0.0001,
+      "loss": 3.9921,
+      "loss/crossentropy": 2.238133430480957,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23264692723751068,
+      "step": 17608
+    },
+    {
+      "epoch": 0.3522,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0024169921875,
+      "learning_rate": 0.0001,
+      "loss": 3.85,
+      "loss/crossentropy": 1.7219743728637695,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17108283936977386,
+      "step": 17610
+    },
+    {
+      "epoch": 0.35224,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.005037434895833333,
+      "learning_rate": 0.0001,
+      "loss": 3.8406,
+      "loss/crossentropy": 1.9613978862762451,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2042389214038849,
+      "step": 17612
+    },
+    {
+      "epoch": 0.35228,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.005132802327473958,
+      "learning_rate": 0.0001,
+      "loss": 3.8567,
+      "loss/crossentropy": 1.946107029914856,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19604258984327316,
+      "step": 17614
+    },
+    {
+      "epoch": 0.35232,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.005077870686848959,
+      "learning_rate": 0.0001,
+      "loss": 4.1404,
+      "loss/crossentropy": 2.255744218826294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2184707298874855,
+      "step": 17616
+    },
+    {
+      "epoch": 0.35236,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0058062235514322914,
+      "learning_rate": 0.0001,
+      "loss": 4.1852,
+      "loss/crossentropy": 2.339258551597595,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21353042870759964,
+      "step": 17618
+    },
+    {
+      "epoch": 0.3524,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0054595947265625,
+      "learning_rate": 0.0001,
+      "loss": 3.9494,
+      "loss/crossentropy": 1.8912869691848755,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19134657084941864,
+      "step": 17620
+    },
+    {
+      "epoch": 0.35244,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0058258056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2607,
+      "loss/crossentropy": 2.186914384365082,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20765449851751328,
+      "step": 17622
+    },
+    {
+      "epoch": 0.35248,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.005773671468098958,
+      "learning_rate": 0.0001,
+      "loss": 3.8567,
+      "loss/crossentropy": 1.6339558959007263,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18255474418401718,
+      "step": 17624
+    },
+    {
+      "epoch": 0.35252,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.005796051025390625,
+      "learning_rate": 0.0001,
+      "loss": 3.9593,
+      "loss/crossentropy": 2.030660629272461,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1943136677145958,
+      "step": 17626
+    },
+    {
+      "epoch": 0.35256,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.004173787434895834,
+      "learning_rate": 0.0001,
+      "loss": 3.9738,
+      "loss/crossentropy": 1.8040945529937744,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18390020728111267,
+      "step": 17628
+    },
+    {
+      "epoch": 0.3526,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0055539449055989586,
+      "learning_rate": 0.0001,
+      "loss": 4.4613,
+      "loss/crossentropy": 1.9993014335632324,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20794235169887543,
+      "step": 17630
+    },
+    {
+      "epoch": 0.35264,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.004881795247395833,
+      "learning_rate": 0.0001,
+      "loss": 3.8539,
+      "loss/crossentropy": 2.0751482248306274,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20385687798261642,
+      "step": 17632
+    },
+    {
+      "epoch": 0.35268,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0052073160807291664,
+      "learning_rate": 0.0001,
+      "loss": 4.0059,
+      "loss/crossentropy": 1.8923512697219849,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19879557192325592,
+      "step": 17634
+    },
+    {
+      "epoch": 0.35272,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.004935709635416666,
+      "learning_rate": 0.0001,
+      "loss": 4.2176,
+      "loss/crossentropy": 2.068901300430298,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21545641124248505,
+      "step": 17636
+    },
+    {
+      "epoch": 0.35276,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.005968983968098958,
+      "learning_rate": 0.0001,
+      "loss": 3.8627,
+      "loss/crossentropy": 1.8329379558563232,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1875568851828575,
+      "step": 17638
+    },
+    {
+      "epoch": 0.3528,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.007811482747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0878,
+      "loss/crossentropy": 2.2368232011795044,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22392207384109497,
+      "step": 17640
+    },
+    {
+      "epoch": 0.35284,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.014972941080729166,
+      "learning_rate": 0.0001,
+      "loss": 4.0397,
+      "loss/crossentropy": 2.2503843307495117,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21132144331932068,
+      "step": 17642
+    },
+    {
+      "epoch": 0.35288,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.01582819620768229,
+      "learning_rate": 0.0001,
+      "loss": 4.0459,
+      "loss/crossentropy": 2.1919764280319214,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.197752483189106,
+      "step": 17644
+    },
+    {
+      "epoch": 0.35292,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0148345947265625,
+      "learning_rate": 0.0001,
+      "loss": 3.9122,
+      "loss/crossentropy": 1.9867302775382996,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2115078568458557,
+      "step": 17646
+    },
+    {
+      "epoch": 0.35296,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.017899576822916666,
+      "learning_rate": 0.0001,
+      "loss": 3.8369,
+      "loss/crossentropy": 1.9015426635742188,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17188573628664017,
+      "step": 17648
+    },
+    {
+      "epoch": 0.353,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.019006093343098957,
+      "learning_rate": 0.0001,
+      "loss": 3.6023,
+      "loss/crossentropy": 1.8092533946037292,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1764475554227829,
+      "step": 17650
+    },
+    {
+      "epoch": 0.35304,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.01936620076497396,
+      "learning_rate": 0.0001,
+      "loss": 4.0542,
+      "loss/crossentropy": 2.0736488103866577,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20046666264533997,
+      "step": 17652
+    },
+    {
+      "epoch": 0.35308,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.019364166259765624,
+      "learning_rate": 0.0001,
+      "loss": 3.9716,
+      "loss/crossentropy": 2.1787428855895996,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23303276300430298,
+      "step": 17654
+    },
+    {
+      "epoch": 0.35312,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.017020416259765626,
+      "learning_rate": 0.0001,
+      "loss": 4.4889,
+      "loss/crossentropy": 2.419381618499756,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23756036162376404,
+      "step": 17656
+    },
+    {
+      "epoch": 0.35316,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.012111155192057292,
+      "learning_rate": 0.0001,
+      "loss": 4.4202,
+      "loss/crossentropy": 2.4017220735549927,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2373708263039589,
+      "step": 17658
+    },
+    {
+      "epoch": 0.3532,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.011919911702473958,
+      "learning_rate": 0.0001,
+      "loss": 3.8805,
+      "loss/crossentropy": 1.7796767354011536,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19661501049995422,
+      "step": 17660
+    },
+    {
+      "epoch": 0.35324,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.013158162434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.2215,
+      "loss/crossentropy": 2.1815608739852905,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20645780861377716,
+      "step": 17662
+    },
+    {
+      "epoch": 0.35328,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.010178375244140624,
+      "learning_rate": 0.0001,
+      "loss": 4.2148,
+      "loss/crossentropy": 2.056411921977997,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21581003069877625,
+      "step": 17664
+    },
+    {
+      "epoch": 0.35332,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.007824452718098958,
+      "learning_rate": 0.0001,
+      "loss": 3.8676,
+      "loss/crossentropy": 1.8901747465133667,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19498124718666077,
+      "step": 17666
+    },
+    {
+      "epoch": 0.35336,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.006493123372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.0884,
+      "loss/crossentropy": 2.0241716504096985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19405542314052582,
+      "step": 17668
+    },
+    {
+      "epoch": 0.3534,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.008241526285807292,
+      "learning_rate": 0.0001,
+      "loss": 4.1452,
+      "loss/crossentropy": 2.016151189804077,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1856973022222519,
+      "step": 17670
+    },
+    {
+      "epoch": 0.35344,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009096018473307292,
+      "learning_rate": 0.0001,
+      "loss": 4.3895,
+      "loss/crossentropy": 2.5066399574279785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23949292302131653,
+      "step": 17672
+    },
+    {
+      "epoch": 0.35348,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007236480712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1432,
+      "loss/crossentropy": 2.1384177207946777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21078043431043625,
+      "step": 17674
+    },
+    {
+      "epoch": 0.35352,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.0070709228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1732,
+      "loss/crossentropy": 2.1211976408958435,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2064737230539322,
+      "step": 17676
+    },
+    {
+      "epoch": 0.35356,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0072418212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3103,
+      "loss/crossentropy": 2.018574059009552,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20446718484163284,
+      "step": 17678
+    },
+    {
+      "epoch": 0.3536,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008186848958333333,
+      "learning_rate": 0.0001,
+      "loss": 3.9862,
+      "loss/crossentropy": 2.10149747133255,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1974436193704605,
+      "step": 17680
+    },
+    {
+      "epoch": 0.35364,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.009698232014973959,
+      "learning_rate": 0.0001,
+      "loss": 4.3133,
+      "loss/crossentropy": 2.288322687149048,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22535154223442078,
+      "step": 17682
+    },
+    {
+      "epoch": 0.35368,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.010400136311848959,
+      "learning_rate": 0.0001,
+      "loss": 4.3396,
+      "loss/crossentropy": 2.002479314804077,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20964795351028442,
+      "step": 17684
+    },
+    {
+      "epoch": 0.35372,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.0099029541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0116,
+      "loss/crossentropy": 2.2548930644989014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19946102052927017,
+      "step": 17686
+    },
+    {
+      "epoch": 0.35376,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.015006510416666667,
+      "learning_rate": 0.0001,
+      "loss": 4.1777,
+      "loss/crossentropy": 2.3636194467544556,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21626830101013184,
+      "step": 17688
+    },
+    {
+      "epoch": 0.3538,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.016471099853515626,
+      "learning_rate": 0.0001,
+      "loss": 4.1137,
+      "loss/crossentropy": 2.2401771545410156,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.202116459608078,
+      "step": 17690
+    },
+    {
+      "epoch": 0.35384,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.01602783203125,
+      "learning_rate": 0.0001,
+      "loss": 4.2248,
+      "loss/crossentropy": 2.192560911178589,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20161370187997818,
+      "step": 17692
+    },
+    {
+      "epoch": 0.35388,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.01495361328125,
+      "learning_rate": 0.0001,
+      "loss": 4.175,
+      "loss/crossentropy": 2.31100332736969,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23697054386138916,
+      "step": 17694
+    },
+    {
+      "epoch": 0.35392,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.014139811197916666,
+      "learning_rate": 0.0001,
+      "loss": 4.0894,
+      "loss/crossentropy": 2.018012821674347,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1954345926642418,
+      "step": 17696
+    },
+    {
+      "epoch": 0.35396,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.0168853759765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3128,
+      "loss/crossentropy": 2.0922133326530457,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2219238504767418,
+      "step": 17698
+    },
+    {
+      "epoch": 0.354,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.01591796875,
+      "learning_rate": 0.0001,
+      "loss": 4.3803,
+      "loss/crossentropy": 2.727385640144348,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21042727679014206,
+      "step": 17700
+    },
+    {
+      "epoch": 0.35404,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.014631144205729167,
+      "learning_rate": 0.0001,
+      "loss": 3.9535,
+      "loss/crossentropy": 1.8388070464134216,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18624315410852432,
+      "step": 17702
+    },
+    {
+      "epoch": 0.35408,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008719889322916667,
+      "learning_rate": 0.0001,
+      "loss": 3.9327,
+      "loss/crossentropy": 1.879045307636261,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1827290803194046,
+      "step": 17704
+    },
+    {
+      "epoch": 0.35412,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0075266520182291664,
+      "learning_rate": 0.0001,
+      "loss": 4.2715,
+      "loss/crossentropy": 2.2604658603668213,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22230257838964462,
+      "step": 17706
+    },
+    {
+      "epoch": 0.35416,
+      "grad_norm": 1.7890625,
+      "grad_norm_var": 0.01329345703125,
+      "learning_rate": 0.0001,
+      "loss": 3.7828,
+      "loss/crossentropy": 2.1200287342071533,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20524511486291885,
+      "step": 17708
+    },
+    {
+      "epoch": 0.3542,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.014135487874348958,
+      "learning_rate": 0.0001,
+      "loss": 3.9624,
+      "loss/crossentropy": 1.933307945728302,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19991742819547653,
+      "step": 17710
+    },
+    {
+      "epoch": 0.35424,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.013231404622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0673,
+      "loss/crossentropy": 1.9442221522331238,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1861218512058258,
+      "step": 17712
+    },
+    {
+      "epoch": 0.35428,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.008885701497395834,
+      "learning_rate": 0.0001,
+      "loss": 3.6592,
+      "loss/crossentropy": 1.7169482111930847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16892607510089874,
+      "step": 17714
+    },
+    {
+      "epoch": 0.35432,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.006891886393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.3472,
+      "loss/crossentropy": 2.105340003967285,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19026386737823486,
+      "step": 17716
+    },
+    {
+      "epoch": 0.35436,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0069976806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1666,
+      "loss/crossentropy": 2.0366504192352295,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19709386676549911,
+      "step": 17718
+    },
+    {
+      "epoch": 0.3544,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.007393391927083334,
+      "learning_rate": 0.0001,
+      "loss": 3.9829,
+      "loss/crossentropy": 1.9578897356987,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19842654466629028,
+      "step": 17720
+    },
+    {
+      "epoch": 0.35444,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.006888834635416666,
+      "learning_rate": 0.0001,
+      "loss": 3.8608,
+      "loss/crossentropy": 2.139336943626404,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19530436396598816,
+      "step": 17722
+    },
+    {
+      "epoch": 0.35448,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.009669748942057292,
+      "learning_rate": 0.0001,
+      "loss": 4.3803,
+      "loss/crossentropy": 2.4077011346817017,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2209417074918747,
+      "step": 17724
+    },
+    {
+      "epoch": 0.35452,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.009373982747395834,
+      "learning_rate": 0.0001,
+      "loss": 4.2412,
+      "loss/crossentropy": 2.1217586994171143,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2052367851138115,
+      "step": 17726
+    },
+    {
+      "epoch": 0.35456,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.009751129150390624,
+      "learning_rate": 0.0001,
+      "loss": 4.1872,
+      "loss/crossentropy": 2.078941583633423,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2033359408378601,
+      "step": 17728
+    },
+    {
+      "epoch": 0.3546,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0072265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3514,
+      "loss/crossentropy": 2.1763141751289368,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22765995562076569,
+      "step": 17730
+    },
+    {
+      "epoch": 0.35464,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.00784912109375,
+      "learning_rate": 0.0001,
+      "loss": 4.1094,
+      "loss/crossentropy": 1.83991938829422,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18603162467479706,
+      "step": 17732
+    },
+    {
+      "epoch": 0.35468,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.009132639567057291,
+      "learning_rate": 0.0001,
+      "loss": 3.93,
+      "loss/crossentropy": 2.021396040916443,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19647164642810822,
+      "step": 17734
+    },
+    {
+      "epoch": 0.35472,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.009297434488932292,
+      "learning_rate": 0.0001,
+      "loss": 4.2137,
+      "loss/crossentropy": 2.2339202165603638,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20721730589866638,
+      "step": 17736
+    },
+    {
+      "epoch": 0.35476,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.008017730712890626,
+      "learning_rate": 0.0001,
+      "loss": 4.1859,
+      "loss/crossentropy": 2.2280211448669434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21610353142023087,
+      "step": 17738
+    },
+    {
+      "epoch": 0.3548,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.0082672119140625,
+      "learning_rate": 0.0001,
+      "loss": 3.8836,
+      "loss/crossentropy": 1.8112387657165527,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18565281480550766,
+      "step": 17740
+    },
+    {
+      "epoch": 0.35484,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.009098307291666666,
+      "learning_rate": 0.0001,
+      "loss": 4.5429,
+      "loss/crossentropy": 2.1125651597976685,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1949760466814041,
+      "step": 17742
+    },
+    {
+      "epoch": 0.35488,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.009163411458333333,
+      "learning_rate": 0.0001,
+      "loss": 3.9613,
+      "loss/crossentropy": 2.1020091772079468,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2126646414399147,
+      "step": 17744
+    },
+    {
+      "epoch": 0.35492,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.010553995768229166,
+      "learning_rate": 0.0001,
+      "loss": 4.0522,
+      "loss/crossentropy": 1.8875654339790344,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1913822665810585,
+      "step": 17746
+    },
+    {
+      "epoch": 0.35496,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.011921946207682292,
+      "learning_rate": 0.0001,
+      "loss": 4.2408,
+      "loss/crossentropy": 2.1043163537979126,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18798817694187164,
+      "step": 17748
+    },
+    {
+      "epoch": 0.355,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.012373606363932291,
+      "learning_rate": 0.0001,
+      "loss": 4.0744,
+      "loss/crossentropy": 2.0751022696495056,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19110247492790222,
+      "step": 17750
+    },
+    {
+      "epoch": 0.35504,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.011815388997395834,
+      "learning_rate": 0.0001,
+      "loss": 4.2077,
+      "loss/crossentropy": 2.0335286259651184,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18653328716754913,
+      "step": 17752
+    },
+    {
+      "epoch": 0.35508,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.012230428059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2833,
+      "loss/crossentropy": 2.091266691684723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19435560703277588,
+      "step": 17754
+    },
+    {
+      "epoch": 0.35512,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.015702056884765624,
+      "learning_rate": 0.0001,
+      "loss": 4.4468,
+      "loss/crossentropy": 2.2889565229415894,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21760451793670654,
+      "step": 17756
+    },
+    {
+      "epoch": 0.35516,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0162017822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1216,
+      "loss/crossentropy": 1.818089485168457,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18759652972221375,
+      "step": 17758
+    },
+    {
+      "epoch": 0.3552,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.015531158447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0322,
+      "loss/crossentropy": 1.8579466938972473,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20318903028964996,
+      "step": 17760
+    },
+    {
+      "epoch": 0.35524,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0152008056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.058,
+      "loss/crossentropy": 2.0463815927505493,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20810134708881378,
+      "step": 17762
+    },
+    {
+      "epoch": 0.35528,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.0146392822265625,
+      "learning_rate": 0.0001,
+      "loss": 3.9702,
+      "loss/crossentropy": 1.8263658285140991,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1781037524342537,
+      "step": 17764
+    },
+    {
+      "epoch": 0.35532,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0123199462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0973,
+      "loss/crossentropy": 2.0662325620651245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2161395400762558,
+      "step": 17766
+    },
+    {
+      "epoch": 0.35536,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.013598378499348958,
+      "learning_rate": 0.0001,
+      "loss": 4.1028,
+      "loss/crossentropy": 1.9901453256607056,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20523327589035034,
+      "step": 17768
+    },
+    {
+      "epoch": 0.3554,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011766560872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1567,
+      "loss/crossentropy": 1.9877265095710754,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20279201865196228,
+      "step": 17770
+    },
+    {
+      "epoch": 0.35544,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.013695271809895833,
+      "learning_rate": 0.0001,
+      "loss": 4.5072,
+      "loss/crossentropy": 2.1376627683639526,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20117300748825073,
+      "step": 17772
+    },
+    {
+      "epoch": 0.35548,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.015327962239583333,
+      "learning_rate": 0.0001,
+      "loss": 3.8024,
+      "loss/crossentropy": 2.0174089074134827,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2042396143078804,
+      "step": 17774
+    },
+    {
+      "epoch": 0.35552,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.019090779622395835,
+      "learning_rate": 0.0001,
+      "loss": 3.7872,
+      "loss/crossentropy": 2.1650888919830322,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2062518149614334,
+      "step": 17776
+    },
+    {
+      "epoch": 0.35556,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.017836252848307293,
+      "learning_rate": 0.0001,
+      "loss": 4.275,
+      "loss/crossentropy": 2.078373670578003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.205182746052742,
+      "step": 17778
+    },
+    {
+      "epoch": 0.3556,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.017693837483723957,
+      "learning_rate": 0.0001,
+      "loss": 3.7109,
+      "loss/crossentropy": 1.7496679425239563,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17602672427892685,
+      "step": 17780
+    },
+    {
+      "epoch": 0.35564,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.01779963175455729,
+      "learning_rate": 0.0001,
+      "loss": 3.8767,
+      "loss/crossentropy": 2.3132810592651367,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21808438748121262,
+      "step": 17782
+    },
+    {
+      "epoch": 0.35568,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.017210896809895834,
+      "learning_rate": 0.0001,
+      "loss": 3.8465,
+      "loss/crossentropy": 1.9900661706924438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20264087617397308,
+      "step": 17784
+    },
+    {
+      "epoch": 0.35572,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.01665013631184896,
+      "learning_rate": 0.0001,
+      "loss": 4.2735,
+      "loss/crossentropy": 2.095883369445801,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19925827533006668,
+      "step": 17786
+    },
+    {
+      "epoch": 0.35576,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.008771769205729167,
+      "learning_rate": 0.0001,
+      "loss": 3.9558,
+      "loss/crossentropy": 1.9932443499565125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1864292174577713,
+      "step": 17788
+    },
+    {
+      "epoch": 0.3558,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.008455149332682292,
+      "learning_rate": 0.0001,
+      "loss": 4.0811,
+      "loss/crossentropy": 1.8468709588050842,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18770359456539154,
+      "step": 17790
+    },
+    {
+      "epoch": 0.35584,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.006502278645833333,
+      "learning_rate": 0.0001,
+      "loss": 4.1749,
+      "loss/crossentropy": 2.293928623199463,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20280499011278152,
+      "step": 17792
+    },
+    {
+      "epoch": 0.35588,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.00618896484375,
+      "learning_rate": 0.0001,
+      "loss": 4.0155,
+      "loss/crossentropy": 1.8333890438079834,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20070593804121017,
+      "step": 17794
+    },
+    {
+      "epoch": 0.35592,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.0061767578125,
+      "learning_rate": 0.0001,
+      "loss": 3.9903,
+      "loss/crossentropy": 1.7173805236816406,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17676317691802979,
+      "step": 17796
+    },
+    {
+      "epoch": 0.35596,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.0058977762858072914,
+      "learning_rate": 0.0001,
+      "loss": 3.8196,
+      "loss/crossentropy": 1.8198468685150146,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20926716923713684,
+      "step": 17798
+    },
+    {
+      "epoch": 0.356,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0054840087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2473,
+      "loss/crossentropy": 2.354674279689789,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23675478994846344,
+      "step": 17800
+    },
+    {
+      "epoch": 0.35604,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.005610911051432291,
+      "learning_rate": 0.0001,
+      "loss": 4.1092,
+      "loss/crossentropy": 2.055357277393341,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18840713798999786,
+      "step": 17802
+    },
+    {
+      "epoch": 0.35608,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.007089996337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1351,
+      "loss/crossentropy": 2.3461071252822876,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22149299830198288,
+      "step": 17804
+    },
+    {
+      "epoch": 0.35612,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0083892822265625,
+      "learning_rate": 0.0001,
+      "loss": 4.165,
+      "loss/crossentropy": 2.2675124406814575,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21928898990154266,
+      "step": 17806
+    },
+    {
+      "epoch": 0.35616,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.007346343994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2397,
+      "loss/crossentropy": 2.239398717880249,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20473385602235794,
+      "step": 17808
+    },
+    {
+      "epoch": 0.3562,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.009262847900390624,
+      "learning_rate": 0.0001,
+      "loss": 3.8152,
+      "loss/crossentropy": 2.022661864757538,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1910046860575676,
+      "step": 17810
+    },
+    {
+      "epoch": 0.35624,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.009236653645833334,
+      "learning_rate": 0.0001,
+      "loss": 4.053,
+      "loss/crossentropy": 1.874055802822113,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19950387626886368,
+      "step": 17812
+    },
+    {
+      "epoch": 0.35628,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.009178670247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.058,
+      "loss/crossentropy": 1.7601851224899292,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18487969785928726,
+      "step": 17814
+    },
+    {
+      "epoch": 0.35632,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.008348592122395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0428,
+      "loss/crossentropy": 1.8841391801834106,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17906954884529114,
+      "step": 17816
+    },
+    {
+      "epoch": 0.35636,
+      "grad_norm": 1.7890625,
+      "grad_norm_var": 0.011092122395833333,
+      "learning_rate": 0.0001,
+      "loss": 3.6292,
+      "loss/crossentropy": 1.7313326597213745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16394374519586563,
+      "step": 17818
+    },
+    {
+      "epoch": 0.3564,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0076901753743489586,
+      "learning_rate": 0.0001,
+      "loss": 3.7718,
+      "loss/crossentropy": 2.094825506210327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20041480660438538,
+      "step": 17820
+    },
+    {
+      "epoch": 0.35644,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.004854329427083333,
+      "learning_rate": 0.0001,
+      "loss": 4.075,
+      "loss/crossentropy": 2.222295045852661,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19912637770175934,
+      "step": 17822
+    },
+    {
+      "epoch": 0.35648,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.005980428059895833,
+      "learning_rate": 0.0001,
+      "loss": 3.971,
+      "loss/crossentropy": 2.1671608090400696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20070582628250122,
+      "step": 17824
+    },
+    {
+      "epoch": 0.35652,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.006884511311848958,
+      "learning_rate": 0.0001,
+      "loss": 4.0805,
+      "loss/crossentropy": 2.164097785949707,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2535782679915428,
+      "step": 17826
+    },
+    {
+      "epoch": 0.35656,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.008405558268229167,
+      "learning_rate": 0.0001,
+      "loss": 4.128,
+      "loss/crossentropy": 1.738038182258606,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17469938844442368,
+      "step": 17828
+    },
+    {
+      "epoch": 0.3566,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.010636393229166667,
+      "learning_rate": 0.0001,
+      "loss": 4.3157,
+      "loss/crossentropy": 2.3383474349975586,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2087705433368683,
+      "step": 17830
+    },
+    {
+      "epoch": 0.35664,
+      "grad_norm": 1.7734375,
+      "grad_norm_var": 0.012870279947916667,
+      "learning_rate": 0.0001,
+      "loss": 3.854,
+      "loss/crossentropy": 2.0826202034950256,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19418098032474518,
+      "step": 17832
+    },
+    {
+      "epoch": 0.35668,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011287434895833334,
+      "learning_rate": 0.0001,
+      "loss": 4.1733,
+      "loss/crossentropy": 1.943938970565796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20667240023612976,
+      "step": 17834
+    },
+    {
+      "epoch": 0.35672,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.011226145426432292,
+      "learning_rate": 0.0001,
+      "loss": 4.1312,
+      "loss/crossentropy": 1.9756113290786743,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1973763257265091,
+      "step": 17836
+    },
+    {
+      "epoch": 0.35676,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.012898763020833334,
+      "learning_rate": 0.0001,
+      "loss": 3.6895,
+      "loss/crossentropy": 1.8583598732948303,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17464379966259003,
+      "step": 17838
+    },
+    {
+      "epoch": 0.3568,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.011791737874348958,
+      "learning_rate": 0.0001,
+      "loss": 3.9421,
+      "loss/crossentropy": 1.9786911606788635,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20051777362823486,
+      "step": 17840
+    },
+    {
+      "epoch": 0.35684,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.010414377848307291,
+      "learning_rate": 0.0001,
+      "loss": 4.0804,
+      "loss/crossentropy": 2.1271785497665405,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21754157543182373,
+      "step": 17842
+    },
+    {
+      "epoch": 0.35688,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.009639485677083334,
+      "learning_rate": 0.0001,
+      "loss": 3.9739,
+      "loss/crossentropy": 2.264941096305847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20562931895256042,
+      "step": 17844
+    },
+    {
+      "epoch": 0.35692,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0083160400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.3865,
+      "loss/crossentropy": 2.141028881072998,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23258531093597412,
+      "step": 17846
+    },
+    {
+      "epoch": 0.35696,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.009781901041666667,
+      "learning_rate": 0.0001,
+      "loss": 4.1484,
+      "loss/crossentropy": 1.9527946710586548,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24948758631944656,
+      "step": 17848
+    },
+    {
+      "epoch": 0.357,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008854166666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.0745,
+      "loss/crossentropy": 1.9595746397972107,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19834068417549133,
+      "step": 17850
+    },
+    {
+      "epoch": 0.35704,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.010396321614583334,
+      "learning_rate": 0.0001,
+      "loss": 4.1407,
+      "loss/crossentropy": 1.8414466977119446,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19133952260017395,
+      "step": 17852
+    },
+    {
+      "epoch": 0.35708,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.010326894124348958,
+      "learning_rate": 0.0001,
+      "loss": 3.6912,
+      "loss/crossentropy": 1.6589071154594421,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17579880356788635,
+      "step": 17854
+    },
+    {
+      "epoch": 0.35712,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.011213175455729167,
+      "learning_rate": 0.0001,
+      "loss": 3.8164,
+      "loss/crossentropy": 1.8462252020835876,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19051598757505417,
+      "step": 17856
+    },
+    {
+      "epoch": 0.35716,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.015083567301432291,
+      "learning_rate": 0.0001,
+      "loss": 3.9453,
+      "loss/crossentropy": 2.0473897457122803,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21250004321336746,
+      "step": 17858
+    },
+    {
+      "epoch": 0.3572,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.014664459228515624,
+      "learning_rate": 0.0001,
+      "loss": 4.1836,
+      "loss/crossentropy": 2.220176875591278,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22177650034427643,
+      "step": 17860
+    },
+    {
+      "epoch": 0.35724,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.13489176432291666,
+      "learning_rate": 0.0001,
+      "loss": 3.5816,
+      "loss/crossentropy": 1.6774207949638367,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16900332272052765,
+      "step": 17862
+    },
+    {
+      "epoch": 0.35728,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.13516616821289062,
+      "learning_rate": 0.0001,
+      "loss": 3.8075,
+      "loss/crossentropy": 1.9064326286315918,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1925061270594597,
+      "step": 17864
+    },
+    {
+      "epoch": 0.35732,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.1366607666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1698,
+      "loss/crossentropy": 2.188231110572815,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20198386162519455,
+      "step": 17866
+    },
+    {
+      "epoch": 0.35736,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.13647359212239582,
+      "learning_rate": 0.0001,
+      "loss": 4.0141,
+      "loss/crossentropy": 1.9380639791488647,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19545384496450424,
+      "step": 17868
+    },
+    {
+      "epoch": 0.3574,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.13407389322916666,
+      "learning_rate": 0.0001,
+      "loss": 4.1965,
+      "loss/crossentropy": 1.995704710483551,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19666064530611038,
+      "step": 17870
+    },
+    {
+      "epoch": 0.35744,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.13017349243164061,
+      "learning_rate": 0.0001,
+      "loss": 4.0191,
+      "loss/crossentropy": 2.2148354053497314,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23265502601861954,
+      "step": 17872
+    },
+    {
+      "epoch": 0.35748,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.1243072509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0021,
+      "loss/crossentropy": 1.7091269493103027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21242853999137878,
+      "step": 17874
+    },
+    {
+      "epoch": 0.35752,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.12219950358072916,
+      "learning_rate": 0.0001,
+      "loss": 4.3081,
+      "loss/crossentropy": 2.2395507097244263,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21505478769540787,
+      "step": 17876
+    },
+    {
+      "epoch": 0.35756,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.008512369791666667,
+      "learning_rate": 0.0001,
+      "loss": 3.8636,
+      "loss/crossentropy": 2.014355480670929,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20956294238567352,
+      "step": 17878
+    },
+    {
+      "epoch": 0.3576,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.012383778889973959,
+      "learning_rate": 0.0001,
+      "loss": 4.3642,
+      "loss/crossentropy": 2.3606066703796387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22014011442661285,
+      "step": 17880
+    },
+    {
+      "epoch": 0.35764,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.011368560791015624,
+      "learning_rate": 0.0001,
+      "loss": 3.9368,
+      "loss/crossentropy": 1.6152977347373962,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1630660966038704,
+      "step": 17882
+    },
+    {
+      "epoch": 0.35768,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.011138661702473959,
+      "learning_rate": 0.0001,
+      "loss": 4.156,
+      "loss/crossentropy": 2.1977567076683044,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20369882881641388,
+      "step": 17884
+    },
+    {
+      "epoch": 0.35772,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.012359364827473959,
+      "learning_rate": 0.0001,
+      "loss": 3.9927,
+      "loss/crossentropy": 1.9582098126411438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20284517109394073,
+      "step": 17886
+    },
+    {
+      "epoch": 0.35776,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.013818105061848959,
+      "learning_rate": 0.0001,
+      "loss": 4.108,
+      "loss/crossentropy": 2.1300426721572876,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19376185536384583,
+      "step": 17888
+    },
+    {
+      "epoch": 0.3578,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.01883519490559896,
+      "learning_rate": 0.0001,
+      "loss": 4.3648,
+      "loss/crossentropy": 2.057590961456299,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21069013327360153,
+      "step": 17890
+    },
+    {
+      "epoch": 0.35784,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.01968994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0006,
+      "loss/crossentropy": 2.21635901927948,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19591110199689865,
+      "step": 17892
+    },
+    {
+      "epoch": 0.35788,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.019809722900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.4868,
+      "loss/crossentropy": 2.6530216932296753,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24127788096666336,
+      "step": 17894
+    },
+    {
+      "epoch": 0.35792,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.013256581624348958,
+      "learning_rate": 0.0001,
+      "loss": 4.0727,
+      "loss/crossentropy": 1.9931264519691467,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20592481642961502,
+      "step": 17896
+    },
+    {
+      "epoch": 0.35796,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.013305409749348959,
+      "learning_rate": 0.0001,
+      "loss": 4.1574,
+      "loss/crossentropy": 2.178193688392639,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2204412743449211,
+      "step": 17898
+    },
+    {
+      "epoch": 0.358,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.012878163655598959,
+      "learning_rate": 0.0001,
+      "loss": 4.1748,
+      "loss/crossentropy": 2.192083954811096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22313552349805832,
+      "step": 17900
+    },
+    {
+      "epoch": 0.35804,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.011281077067057292,
+      "learning_rate": 0.0001,
+      "loss": 4.1695,
+      "loss/crossentropy": 2.228816568851471,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2120378389954567,
+      "step": 17902
+    },
+    {
+      "epoch": 0.35808,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.010871378580729167,
+      "learning_rate": 0.0001,
+      "loss": 4.1369,
+      "loss/crossentropy": 2.0921813249588013,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19552303105592728,
+      "step": 17904
+    },
+    {
+      "epoch": 0.35812,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.004687245686848958,
+      "learning_rate": 0.0001,
+      "loss": 4.2206,
+      "loss/crossentropy": 2.0568217635154724,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19668184220790863,
+      "step": 17906
+    },
+    {
+      "epoch": 0.35816,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.0038937886555989584,
+      "learning_rate": 0.0001,
+      "loss": 3.9946,
+      "loss/crossentropy": 1.8241485357284546,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15895915031433105,
+      "step": 17908
+    },
+    {
+      "epoch": 0.3582,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.003082021077473958,
+      "learning_rate": 0.0001,
+      "loss": 3.9904,
+      "loss/crossentropy": 2.2571341395378113,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2065516859292984,
+      "step": 17910
+    },
+    {
+      "epoch": 0.35824,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.0032297770182291665,
+      "learning_rate": 0.0001,
+      "loss": 3.9289,
+      "loss/crossentropy": 2.0472288727760315,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21586833149194717,
+      "step": 17912
+    },
+    {
+      "epoch": 0.35828,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0035947163899739585,
+      "learning_rate": 0.0001,
+      "loss": 4.0652,
+      "loss/crossentropy": 2.0345569252967834,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19726714491844177,
+      "step": 17914
+    },
+    {
+      "epoch": 0.35832,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0037127176920572916,
+      "learning_rate": 0.0001,
+      "loss": 4.0294,
+      "loss/crossentropy": 2.15024471282959,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20401738584041595,
+      "step": 17916
+    },
+    {
+      "epoch": 0.35836,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.004874420166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3122,
+      "loss/crossentropy": 2.110986351966858,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2299470156431198,
+      "step": 17918
+    },
+    {
+      "epoch": 0.3584,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.003897857666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2147,
+      "loss/crossentropy": 1.7297720909118652,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18035240471363068,
+      "step": 17920
+    },
+    {
+      "epoch": 0.35844,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.01236572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3997,
+      "loss/crossentropy": 1.9483368396759033,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2060721442103386,
+      "step": 17922
+    },
+    {
+      "epoch": 0.35848,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0123443603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1527,
+      "loss/crossentropy": 1.9840999841690063,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20700833201408386,
+      "step": 17924
+    },
+    {
+      "epoch": 0.35852,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.011525217692057292,
+      "learning_rate": 0.0001,
+      "loss": 4.1754,
+      "loss/crossentropy": 2.4118131399154663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20695140957832336,
+      "step": 17926
+    },
+    {
+      "epoch": 0.35856,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.011195627848307292,
+      "learning_rate": 0.0001,
+      "loss": 3.9363,
+      "loss/crossentropy": 1.9709432721138,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18202026933431625,
+      "step": 17928
+    },
+    {
+      "epoch": 0.3586,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010206858317057291,
+      "learning_rate": 0.0001,
+      "loss": 4.2972,
+      "loss/crossentropy": 1.9450251460075378,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21454186737537384,
+      "step": 17930
+    },
+    {
+      "epoch": 0.35864,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0094390869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3317,
+      "loss/crossentropy": 2.0738271474838257,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21376954019069672,
+      "step": 17932
+    },
+    {
+      "epoch": 0.35868,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.010109202067057291,
+      "learning_rate": 0.0001,
+      "loss": 4.1613,
+      "loss/crossentropy": 2.1857110261917114,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2238282486796379,
+      "step": 17934
+    },
+    {
+      "epoch": 0.35872,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.010643513997395833,
+      "learning_rate": 0.0001,
+      "loss": 4.2208,
+      "loss/crossentropy": 2.2790093421936035,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1987098902463913,
+      "step": 17936
+    },
+    {
+      "epoch": 0.35876,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.0054280598958333336,
+      "learning_rate": 0.0001,
+      "loss": 3.9261,
+      "loss/crossentropy": 2.1101399064064026,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2034989446401596,
+      "step": 17938
+    },
+    {
+      "epoch": 0.3588,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.005793253580729167,
+      "learning_rate": 0.0001,
+      "loss": 3.7024,
+      "loss/crossentropy": 1.6923771500587463,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18680892139673233,
+      "step": 17940
+    },
+    {
+      "epoch": 0.35884,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.005342356363932292,
+      "learning_rate": 0.0001,
+      "loss": 3.9282,
+      "loss/crossentropy": 1.6772454977035522,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17377988994121552,
+      "step": 17942
+    },
+    {
+      "epoch": 0.35888,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.007722981770833333,
+      "learning_rate": 0.0001,
+      "loss": 4.141,
+      "loss/crossentropy": 2.2448233366012573,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21804184466600418,
+      "step": 17944
+    },
+    {
+      "epoch": 0.35892,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.006688435872395833,
+      "learning_rate": 0.0001,
+      "loss": 3.9984,
+      "loss/crossentropy": 1.934161365032196,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1834414005279541,
+      "step": 17946
+    },
+    {
+      "epoch": 0.35896,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.007344563802083333,
+      "learning_rate": 0.0001,
+      "loss": 3.9943,
+      "loss/crossentropy": 2.06991970539093,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21084094047546387,
+      "step": 17948
+    },
+    {
+      "epoch": 0.359,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.007916005452473958,
+      "learning_rate": 0.0001,
+      "loss": 4.0493,
+      "loss/crossentropy": 2.0942054986953735,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21652275323867798,
+      "step": 17950
+    },
+    {
+      "epoch": 0.35904,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009134928385416666,
+      "learning_rate": 0.0001,
+      "loss": 4.1125,
+      "loss/crossentropy": 2.0005252361297607,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.198953777551651,
+      "step": 17952
+    },
+    {
+      "epoch": 0.35908,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.0327301025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1639,
+      "loss/crossentropy": 2.1563133597373962,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2072513848543167,
+      "step": 17954
+    },
+    {
+      "epoch": 0.35912,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.0312896728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1789,
+      "loss/crossentropy": 2.3533977270126343,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21483703702688217,
+      "step": 17956
+    },
+    {
+      "epoch": 0.35916,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.029581705729166668,
+      "learning_rate": 0.0001,
+      "loss": 4.0921,
+      "loss/crossentropy": 2.1915602684020996,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2241670861840248,
+      "step": 17958
+    },
+    {
+      "epoch": 0.3592,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0314117431640625,
+      "learning_rate": 0.0001,
+      "loss": 3.9225,
+      "loss/crossentropy": 1.9529941082000732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1920214593410492,
+      "step": 17960
+    },
+    {
+      "epoch": 0.35924,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.030350748697916666,
+      "learning_rate": 0.0001,
+      "loss": 4.4467,
+      "loss/crossentropy": 2.308731436729431,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2238362655043602,
+      "step": 17962
+    },
+    {
+      "epoch": 0.35928,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.04334208170572917,
+      "learning_rate": 0.0001,
+      "loss": 4.1682,
+      "loss/crossentropy": 2.0917986631393433,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2061680108308792,
+      "step": 17964
+    },
+    {
+      "epoch": 0.35932,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 10.265104166666667,
+      "learning_rate": 0.0001,
+      "loss": 4.9501,
+      "loss/crossentropy": 2.16925585269928,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20239810645580292,
+      "step": 17966
+    },
+    {
+      "epoch": 0.35936,
+      "grad_norm": 2.0,
+      "grad_norm_var": 10.219252268473307,
+      "learning_rate": 0.0001,
+      "loss": 4.287,
+      "loss/crossentropy": 2.1585946083068848,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21381714940071106,
+      "step": 17968
+    },
+    {
+      "epoch": 0.3594,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 10.281192016601562,
+      "learning_rate": 0.0001,
+      "loss": 4.0133,
+      "loss/crossentropy": 1.8752552270889282,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19667387753725052,
+      "step": 17970
+    },
+    {
+      "epoch": 0.35944,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 10.291275024414062,
+      "learning_rate": 0.0001,
+      "loss": 3.9797,
+      "loss/crossentropy": 1.8533543944358826,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1844579428434372,
+      "step": 17972
+    },
+    {
+      "epoch": 0.35948,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 10.282754516601562,
+      "learning_rate": 0.0001,
+      "loss": 4.5316,
+      "loss/crossentropy": 2.5859906673431396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23142188042402267,
+      "step": 17974
+    },
+    {
+      "epoch": 0.35952,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 10.236201985677083,
+      "learning_rate": 0.0001,
+      "loss": 4.2607,
+      "loss/crossentropy": 2.2086023092269897,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21764708310365677,
+      "step": 17976
+    },
+    {
+      "epoch": 0.35956,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 10.258858235677083,
+      "learning_rate": 0.0001,
+      "loss": 4.2578,
+      "loss/crossentropy": 2.3860682249069214,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22804389894008636,
+      "step": 17978
+    },
+    {
+      "epoch": 0.3596,
+      "grad_norm": 2.0,
+      "grad_norm_var": 10.314815266927083,
+      "learning_rate": 0.0001,
+      "loss": 4.2055,
+      "loss/crossentropy": 2.1412216424942017,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.200544573366642,
+      "step": 17980
+    },
+    {
+      "epoch": 0.35964,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.022395833333333334,
+      "learning_rate": 0.0001,
+      "loss": 4.0412,
+      "loss/crossentropy": 2.198704957962036,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2043847218155861,
+      "step": 17982
+    },
+    {
+      "epoch": 0.35968,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.010249837239583334,
+      "learning_rate": 0.0001,
+      "loss": 4.037,
+      "loss/crossentropy": 2.0053776502609253,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2072744369506836,
+      "step": 17984
+    },
+    {
+      "epoch": 0.35972,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.04221979777018229,
+      "learning_rate": 0.0001,
+      "loss": 4.127,
+      "loss/crossentropy": 2.5015710592269897,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23083895444869995,
+      "step": 17986
+    },
+    {
+      "epoch": 0.35976,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.04145889282226563,
+      "learning_rate": 0.0001,
+      "loss": 4.0531,
+      "loss/crossentropy": 2.037220776081085,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19487662613391876,
+      "step": 17988
+    },
+    {
+      "epoch": 0.3598,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.03911107381184896,
+      "learning_rate": 0.0001,
+      "loss": 3.9801,
+      "loss/crossentropy": 1.9316660165786743,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20183449983596802,
+      "step": 17990
+    },
+    {
+      "epoch": 0.35984,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.04090067545572917,
+      "learning_rate": 0.0001,
+      "loss": 3.804,
+      "loss/crossentropy": 1.892772138118744,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17082024365663528,
+      "step": 17992
+    },
+    {
+      "epoch": 0.35988,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.04088109334309896,
+      "learning_rate": 0.0001,
+      "loss": 4.0732,
+      "loss/crossentropy": 1.9325169324874878,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1900368332862854,
+      "step": 17994
+    },
+    {
+      "epoch": 0.35992,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.050388336181640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1392,
+      "loss/crossentropy": 1.9746126532554626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21207977831363678,
+      "step": 17996
+    },
+    {
+      "epoch": 0.35996,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.048620351155598956,
+      "learning_rate": 0.0001,
+      "loss": 4.0308,
+      "loss/crossentropy": 2.1249493956565857,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18674498051404953,
+      "step": 17998
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.04625422159830729,
+      "learning_rate": 0.0001,
+      "loss": 4.0854,
+      "loss/crossentropy": 2.1279499530792236,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.214716374874115,
+      "step": 18000
+    },
+    {
+      "epoch": 0.36004,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0177398681640625,
+      "learning_rate": 0.0001,
+      "loss": 3.9238,
+      "loss/crossentropy": 2.301763415336609,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20897136628627777,
+      "step": 18002
+    },
+    {
+      "epoch": 0.36008,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.01788330078125,
+      "learning_rate": 0.0001,
+      "loss": 4.0944,
+      "loss/crossentropy": 1.8079062104225159,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19506293535232544,
+      "step": 18004
+    },
+    {
+      "epoch": 0.36012,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.0205078125,
+      "learning_rate": 0.0001,
+      "loss": 4.0503,
+      "loss/crossentropy": 1.9429230093955994,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19762147217988968,
+      "step": 18006
+    },
+    {
+      "epoch": 0.36016,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.019608561197916666,
+      "learning_rate": 0.0001,
+      "loss": 3.9801,
+      "loss/crossentropy": 2.307368576526642,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21261631697416306,
+      "step": 18008
+    },
+    {
+      "epoch": 0.3602,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.021952311197916668,
+      "learning_rate": 0.0001,
+      "loss": 3.8723,
+      "loss/crossentropy": 2.2834020853042603,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20086131989955902,
+      "step": 18010
+    },
+    {
+      "epoch": 0.36024,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.0117340087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1708,
+      "loss/crossentropy": 2.1730109453201294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20470792055130005,
+      "step": 18012
+    },
+    {
+      "epoch": 0.36028,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012809244791666667,
+      "learning_rate": 0.0001,
+      "loss": 4.4047,
+      "loss/crossentropy": 2.491591691970825,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2361597716808319,
+      "step": 18014
+    },
+    {
+      "epoch": 0.36032,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.013240305582682292,
+      "learning_rate": 0.0001,
+      "loss": 4.167,
+      "loss/crossentropy": 2.024519979953766,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2042919620871544,
+      "step": 18016
+    },
+    {
+      "epoch": 0.36036,
+      "grad_norm": 1.7734375,
+      "grad_norm_var": 0.016123199462890626,
+      "learning_rate": 0.0001,
+      "loss": 3.9521,
+      "loss/crossentropy": 2.055600941181183,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17253537476062775,
+      "step": 18018
+    },
+    {
+      "epoch": 0.3604,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0188629150390625,
+      "learning_rate": 0.0001,
+      "loss": 3.9665,
+      "loss/crossentropy": 1.8796368837356567,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21168731898069382,
+      "step": 18020
+    },
+    {
+      "epoch": 0.36044,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.014825185139973959,
+      "learning_rate": 0.0001,
+      "loss": 4.0291,
+      "loss/crossentropy": 1.9885223507881165,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20676057040691376,
+      "step": 18022
+    },
+    {
+      "epoch": 0.36048,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.010839589436848958,
+      "learning_rate": 0.0001,
+      "loss": 3.8401,
+      "loss/crossentropy": 2.0412577986717224,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18211868405342102,
+      "step": 18024
+    },
+    {
+      "epoch": 0.36052,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.010445149739583333,
+      "learning_rate": 0.0001,
+      "loss": 3.8696,
+      "loss/crossentropy": 2.1391053199768066,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20121745765209198,
+      "step": 18026
+    },
+    {
+      "epoch": 0.36056,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.010087076822916667,
+      "learning_rate": 0.0001,
+      "loss": 3.9683,
+      "loss/crossentropy": 1.8888981938362122,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19372491538524628,
+      "step": 18028
+    },
+    {
+      "epoch": 0.3606,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007087198893229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2467,
+      "loss/crossentropy": 2.002028524875641,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18978165835142136,
+      "step": 18030
+    },
+    {
+      "epoch": 0.36064,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.007828776041666667,
+      "learning_rate": 0.0001,
+      "loss": 4.0703,
+      "loss/crossentropy": 2.3448036909103394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21162079274654388,
+      "step": 18032
+    },
+    {
+      "epoch": 0.36068,
+      "grad_norm": 2.5625,
+      "grad_norm_var": 0.03154474894205729,
+      "learning_rate": 0.0001,
+      "loss": 4.283,
+      "loss/crossentropy": 2.086554765701294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25456031411886215,
+      "step": 18034
+    },
+    {
+      "epoch": 0.36072,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.02948582967122396,
+      "learning_rate": 0.0001,
+      "loss": 3.8122,
+      "loss/crossentropy": 1.9166680574417114,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18382181227207184,
+      "step": 18036
+    },
+    {
+      "epoch": 0.36076,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.029642740885416668,
+      "learning_rate": 0.0001,
+      "loss": 3.9937,
+      "loss/crossentropy": 2.050579786300659,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20126450806856155,
+      "step": 18038
+    },
+    {
+      "epoch": 0.3608,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.030049641927083332,
+      "learning_rate": 0.0001,
+      "loss": 3.9273,
+      "loss/crossentropy": 2.2339882850646973,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20789660513401031,
+      "step": 18040
+    },
+    {
+      "epoch": 0.36084,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.029243977864583333,
+      "learning_rate": 0.0001,
+      "loss": 4.0086,
+      "loss/crossentropy": 2.248544931411743,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20925041288137436,
+      "step": 18042
+    },
+    {
+      "epoch": 0.36088,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.028955078125,
+      "learning_rate": 0.0001,
+      "loss": 4.0448,
+      "loss/crossentropy": 1.8894963264465332,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18232882767915726,
+      "step": 18044
+    },
+    {
+      "epoch": 0.36092,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0289703369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1838,
+      "loss/crossentropy": 2.0220844745635986,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18812591582536697,
+      "step": 18046
+    },
+    {
+      "epoch": 0.36096,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.027913411458333332,
+      "learning_rate": 0.0001,
+      "loss": 4.2088,
+      "loss/crossentropy": 2.1896166801452637,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.205679252743721,
+      "step": 18048
+    },
+    {
+      "epoch": 0.361,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.004686482747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0511,
+      "loss/crossentropy": 2.0670089721679688,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18881987780332565,
+      "step": 18050
+    },
+    {
+      "epoch": 0.36104,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.005500284830729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2426,
+      "loss/crossentropy": 2.155470609664917,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2147108018398285,
+      "step": 18052
+    },
+    {
+      "epoch": 0.36108,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.0076812744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2007,
+      "loss/crossentropy": 1.8182223439216614,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18691811710596085,
+      "step": 18054
+    },
+    {
+      "epoch": 0.36112,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0066314697265625,
+      "learning_rate": 0.0001,
+      "loss": 3.8601,
+      "loss/crossentropy": 1.7366089820861816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18361609429121017,
+      "step": 18056
+    },
+    {
+      "epoch": 0.36116,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.005704498291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.027,
+      "loss/crossentropy": 1.9429153203964233,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1974082887172699,
+      "step": 18058
+    },
+    {
+      "epoch": 0.3612,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.00562744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2751,
+      "loss/crossentropy": 2.230627417564392,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22216492146253586,
+      "step": 18060
+    },
+    {
+      "epoch": 0.36124,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.006371053059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1759,
+      "loss/crossentropy": 2.0506786704063416,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20492341369390488,
+      "step": 18062
+    },
+    {
+      "epoch": 0.36128,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.007116444905598958,
+      "learning_rate": 0.0001,
+      "loss": 3.778,
+      "loss/crossentropy": 2.025477647781372,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20034795254468918,
+      "step": 18064
+    },
+    {
+      "epoch": 0.36132,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.006617991129557291,
+      "learning_rate": 0.0001,
+      "loss": 4.2998,
+      "loss/crossentropy": 1.9828922748565674,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19445167481899261,
+      "step": 18066
+    },
+    {
+      "epoch": 0.36136,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.006485748291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2566,
+      "loss/crossentropy": 2.2869513630867004,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21809721738100052,
+      "step": 18068
+    },
+    {
+      "epoch": 0.3614,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.004129791259765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9636,
+      "loss/crossentropy": 2.1584482192993164,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21123671531677246,
+      "step": 18070
+    },
+    {
+      "epoch": 0.36144,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.0046078999837239586,
+      "learning_rate": 0.0001,
+      "loss": 3.9628,
+      "loss/crossentropy": 2.112669587135315,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20795845240354538,
+      "step": 18072
+    },
+    {
+      "epoch": 0.36148,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0059234619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2516,
+      "loss/crossentropy": 2.3283581733703613,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22724319994449615,
+      "step": 18074
+    },
+    {
+      "epoch": 0.36152,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.006078084309895833,
+      "learning_rate": 0.0001,
+      "loss": 3.8966,
+      "loss/crossentropy": 2.286331057548523,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20937074720859528,
+      "step": 18076
+    },
+    {
+      "epoch": 0.36156,
+      "grad_norm": 1.7578125,
+      "grad_norm_var": 0.009110260009765624,
+      "learning_rate": 0.0001,
+      "loss": 4.1124,
+      "loss/crossentropy": 2.1055954694747925,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1924988478422165,
+      "step": 18078
+    },
+    {
+      "epoch": 0.3616,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.008186594645182291,
+      "learning_rate": 0.0001,
+      "loss": 4.0349,
+      "loss/crossentropy": 2.180675983428955,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21886169910430908,
+      "step": 18080
+    },
+    {
+      "epoch": 0.36164,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.008621978759765624,
+      "learning_rate": 0.0001,
+      "loss": 3.9958,
+      "loss/crossentropy": 1.7595775127410889,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17430071532726288,
+      "step": 18082
+    },
+    {
+      "epoch": 0.36168,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.011472320556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0306,
+      "loss/crossentropy": 2.1191208958625793,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19574576616287231,
+      "step": 18084
+    },
+    {
+      "epoch": 0.36172,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0110595703125,
+      "learning_rate": 0.0001,
+      "loss": 4.1563,
+      "loss/crossentropy": 2.0964609384536743,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19742169976234436,
+      "step": 18086
+    },
+    {
+      "epoch": 0.36176,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0108154296875,
+      "learning_rate": 0.0001,
+      "loss": 4.3584,
+      "loss/crossentropy": 2.3081077337265015,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2177809551358223,
+      "step": 18088
+    },
+    {
+      "epoch": 0.3618,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.009928385416666666,
+      "learning_rate": 0.0001,
+      "loss": 4.0493,
+      "loss/crossentropy": 2.340154528617859,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23811470717191696,
+      "step": 18090
+    },
+    {
+      "epoch": 0.36184,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.009777577718098958,
+      "learning_rate": 0.0001,
+      "loss": 3.828,
+      "loss/crossentropy": 1.9872968196868896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19391798973083496,
+      "step": 18092
+    },
+    {
+      "epoch": 0.36188,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0067779541015625,
+      "learning_rate": 0.0001,
+      "loss": 3.9404,
+      "loss/crossentropy": 1.9162002205848694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.203052818775177,
+      "step": 18094
+    },
+    {
+      "epoch": 0.36192,
+      "grad_norm": 1.7890625,
+      "grad_norm_var": 0.009718577067057291,
+      "learning_rate": 0.0001,
+      "loss": 3.7458,
+      "loss/crossentropy": 2.240887403488159,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19722212105989456,
+      "step": 18096
+    },
+    {
+      "epoch": 0.36196,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009496053059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1473,
+      "loss/crossentropy": 1.8601738214492798,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1938677802681923,
+      "step": 18098
+    },
+    {
+      "epoch": 0.362,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.005729166666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.0131,
+      "loss/crossentropy": 1.9588143229484558,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19559185206890106,
+      "step": 18100
+    },
+    {
+      "epoch": 0.36204,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.0060618082682291664,
+      "learning_rate": 0.0001,
+      "loss": 3.9199,
+      "loss/crossentropy": 1.909518837928772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1974424198269844,
+      "step": 18102
+    },
+    {
+      "epoch": 0.36208,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.005641428629557291,
+      "learning_rate": 0.0001,
+      "loss": 4.15,
+      "loss/crossentropy": 2.0762908458709717,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1981927454471588,
+      "step": 18104
+    },
+    {
+      "epoch": 0.36212,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.006339263916015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3226,
+      "loss/crossentropy": 2.4555106163024902,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22001618146896362,
+      "step": 18106
+    },
+    {
+      "epoch": 0.36216,
+      "grad_norm": 1.7734375,
+      "grad_norm_var": 0.013533528645833333,
+      "learning_rate": 0.0001,
+      "loss": 4.0595,
+      "loss/crossentropy": 1.9983150959014893,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19945750385522842,
+      "step": 18108
+    },
+    {
+      "epoch": 0.3622,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.013181304931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0863,
+      "loss/crossentropy": 2.152313530445099,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21388111263513565,
+      "step": 18110
+    },
+    {
+      "epoch": 0.36224,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.016886393229166668,
+      "learning_rate": 0.0001,
+      "loss": 4.1128,
+      "loss/crossentropy": 1.690669596195221,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1883523240685463,
+      "step": 18112
+    },
+    {
+      "epoch": 0.36228,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.017179107666015624,
+      "learning_rate": 0.0001,
+      "loss": 3.9978,
+      "loss/crossentropy": 1.9404500126838684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18945956230163574,
+      "step": 18114
+    },
+    {
+      "epoch": 0.36232,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.016007486979166666,
+      "learning_rate": 0.0001,
+      "loss": 4.4396,
+      "loss/crossentropy": 2.213471293449402,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22550443559885025,
+      "step": 18116
+    },
+    {
+      "epoch": 0.36236,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.014233144124348958,
+      "learning_rate": 0.0001,
+      "loss": 4.3084,
+      "loss/crossentropy": 2.1008135080337524,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2002386674284935,
+      "step": 18118
+    },
+    {
+      "epoch": 0.3624,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.013866933186848958,
+      "learning_rate": 0.0001,
+      "loss": 3.8664,
+      "loss/crossentropy": 1.9704426527023315,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1806606948375702,
+      "step": 18120
+    },
+    {
+      "epoch": 0.36244,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.01473388671875,
+      "learning_rate": 0.0001,
+      "loss": 3.7456,
+      "loss/crossentropy": 1.8658949732780457,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19861871004104614,
+      "step": 18122
+    },
+    {
+      "epoch": 0.36248,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.011034901936848958,
+      "learning_rate": 0.0001,
+      "loss": 4.0341,
+      "loss/crossentropy": 1.9327979683876038,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19395866990089417,
+      "step": 18124
+    },
+    {
+      "epoch": 0.36252,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.009641265869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.071,
+      "loss/crossentropy": 2.377102255821228,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21928569674491882,
+      "step": 18126
+    },
+    {
+      "epoch": 0.36256,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.006994374593098958,
+      "learning_rate": 0.0001,
+      "loss": 3.9623,
+      "loss/crossentropy": 2.1650161743164062,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2292800396680832,
+      "step": 18128
+    },
+    {
+      "epoch": 0.3626,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.007458241780598959,
+      "learning_rate": 0.0001,
+      "loss": 4.3135,
+      "loss/crossentropy": 2.281827986240387,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2190621793270111,
+      "step": 18130
+    },
+    {
+      "epoch": 0.36264,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.00848388671875,
+      "learning_rate": 0.0001,
+      "loss": 3.8335,
+      "loss/crossentropy": 2.0073219537734985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18984179198741913,
+      "step": 18132
+    },
+    {
+      "epoch": 0.36268,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.01141357421875,
+      "learning_rate": 0.0001,
+      "loss": 3.7551,
+      "loss/crossentropy": 1.7675580978393555,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18771683424711227,
+      "step": 18134
+    },
+    {
+      "epoch": 0.36272,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.011154937744140624,
+      "learning_rate": 0.0001,
+      "loss": 3.8816,
+      "loss/crossentropy": 2.064103126525879,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20835159718990326,
+      "step": 18136
+    },
+    {
+      "epoch": 0.36276,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0126861572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.2122,
+      "loss/crossentropy": 2.1821314096450806,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2391536980867386,
+      "step": 18138
+    },
+    {
+      "epoch": 0.3628,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.010117340087890624,
+      "learning_rate": 0.0001,
+      "loss": 4.2307,
+      "loss/crossentropy": 2.295500636100769,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.215906023979187,
+      "step": 18140
+    },
+    {
+      "epoch": 0.36284,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.010245768229166667,
+      "learning_rate": 0.0001,
+      "loss": 4.1144,
+      "loss/crossentropy": 2.0736570954322815,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18491360545158386,
+      "step": 18142
+    },
+    {
+      "epoch": 0.36288,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.009496053059895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1326,
+      "loss/crossentropy": 2.0513535737991333,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20204297453165054,
+      "step": 18144
+    },
+    {
+      "epoch": 0.36292,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.010601552327473958,
+      "learning_rate": 0.0001,
+      "loss": 4.1648,
+      "loss/crossentropy": 2.1471269130706787,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.204594686627388,
+      "step": 18146
+    },
+    {
+      "epoch": 0.36296,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.011372884114583334,
+      "learning_rate": 0.0001,
+      "loss": 4.2728,
+      "loss/crossentropy": 2.093555986881256,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20173487067222595,
+      "step": 18148
+    },
+    {
+      "epoch": 0.363,
+      "grad_norm": 1.6953125,
+      "grad_norm_var": 0.013627115885416667,
+      "learning_rate": 0.0001,
+      "loss": 3.7183,
+      "loss/crossentropy": 1.6938685178756714,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1703190803527832,
+      "step": 18150
+    },
+    {
+      "epoch": 0.36304,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.013337961832682292,
+      "learning_rate": 0.0001,
+      "loss": 4.116,
+      "loss/crossentropy": 2.180016875267029,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23546195775270462,
+      "step": 18152
+    },
+    {
+      "epoch": 0.36308,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.011277008056640624,
+      "learning_rate": 0.0001,
+      "loss": 4.1826,
+      "loss/crossentropy": 2.145975947380066,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20554804801940918,
+      "step": 18154
+    },
+    {
+      "epoch": 0.36312,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.011131795247395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0113,
+      "loss/crossentropy": 2.068985939025879,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20410801470279694,
+      "step": 18156
+    },
+    {
+      "epoch": 0.36316,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.012572224934895833,
+      "learning_rate": 0.0001,
+      "loss": 3.8273,
+      "loss/crossentropy": 1.8428975343704224,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19165785610675812,
+      "step": 18158
+    },
+    {
+      "epoch": 0.3632,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.0131744384765625,
+      "learning_rate": 0.0001,
+      "loss": 3.7808,
+      "loss/crossentropy": 1.859747588634491,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18524104356765747,
+      "step": 18160
+    },
+    {
+      "epoch": 0.36324,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.011521148681640624,
+      "learning_rate": 0.0001,
+      "loss": 4.1458,
+      "loss/crossentropy": 2.1383343935012817,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21649780869483948,
+      "step": 18162
+    },
+    {
+      "epoch": 0.36328,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010758209228515624,
+      "learning_rate": 0.0001,
+      "loss": 3.9889,
+      "loss/crossentropy": 2.0153123140335083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2131347879767418,
+      "step": 18164
+    },
+    {
+      "epoch": 0.36332,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.007004547119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0827,
+      "loss/crossentropy": 1.8382813930511475,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18116765469312668,
+      "step": 18166
+    },
+    {
+      "epoch": 0.36336,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.007287343343098958,
+      "learning_rate": 0.0001,
+      "loss": 3.8883,
+      "loss/crossentropy": 1.9639039039611816,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2014933079481125,
+      "step": 18168
+    },
+    {
+      "epoch": 0.3634,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.012446848551432292,
+      "learning_rate": 0.0001,
+      "loss": 4.2886,
+      "loss/crossentropy": 2.057366132736206,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2134721800684929,
+      "step": 18170
+    },
+    {
+      "epoch": 0.36344,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.012690989176432292,
+      "learning_rate": 0.0001,
+      "loss": 4.0801,
+      "loss/crossentropy": 2.244979500770569,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.219241164624691,
+      "step": 18172
+    },
+    {
+      "epoch": 0.36348,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.010625966389973958,
+      "learning_rate": 0.0001,
+      "loss": 4.1271,
+      "loss/crossentropy": 2.1159361600875854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19503474980592728,
+      "step": 18174
+    },
+    {
+      "epoch": 0.36352,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0099761962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2031,
+      "loss/crossentropy": 2.240646004676819,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22175125032663345,
+      "step": 18176
+    },
+    {
+      "epoch": 0.36356,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.011327107747395834,
+      "learning_rate": 0.0001,
+      "loss": 3.9508,
+      "loss/crossentropy": 1.7946885228157043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17505639791488647,
+      "step": 18178
+    },
+    {
+      "epoch": 0.3636,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.010420735677083333,
+      "learning_rate": 0.0001,
+      "loss": 4.1138,
+      "loss/crossentropy": 2.2058286666870117,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21548640727996826,
+      "step": 18180
+    },
+    {
+      "epoch": 0.36364,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0101226806640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0831,
+      "loss/crossentropy": 2.168944835662842,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22227579355239868,
+      "step": 18182
+    },
+    {
+      "epoch": 0.36368,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.008392079671223959,
+      "learning_rate": 0.0001,
+      "loss": 4.1482,
+      "loss/crossentropy": 2.019882082939148,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20161078870296478,
+      "step": 18184
+    },
+    {
+      "epoch": 0.36372,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0034993489583333335,
+      "learning_rate": 0.0001,
+      "loss": 4.1305,
+      "loss/crossentropy": 2.000952959060669,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19938969612121582,
+      "step": 18186
+    },
+    {
+      "epoch": 0.36376,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0024920145670572916,
+      "learning_rate": 0.0001,
+      "loss": 3.9158,
+      "loss/crossentropy": 2.0612798929214478,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2022939696907997,
+      "step": 18188
+    },
+    {
+      "epoch": 0.3638,
+      "grad_norm": 1.78125,
+      "grad_norm_var": 0.0042307535807291664,
+      "learning_rate": 0.0001,
+      "loss": 3.8822,
+      "loss/crossentropy": 1.664880096912384,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16950316727161407,
+      "step": 18190
+    },
+    {
+      "epoch": 0.36384,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.003360748291015625,
+      "learning_rate": 0.0001,
+      "loss": 3.9881,
+      "loss/crossentropy": 1.8047854900360107,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1823924407362938,
+      "step": 18192
+    },
+    {
+      "epoch": 0.36388,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.0060808817545572914,
+      "learning_rate": 0.0001,
+      "loss": 3.7857,
+      "loss/crossentropy": 2.167177438735962,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1963154897093773,
+      "step": 18194
+    },
+    {
+      "epoch": 0.36392,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.006151326497395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0372,
+      "loss/crossentropy": 2.396019458770752,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20359576493501663,
+      "step": 18196
+    },
+    {
+      "epoch": 0.36396,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.005866495768229166,
+      "learning_rate": 0.0001,
+      "loss": 3.9349,
+      "loss/crossentropy": 2.0866541862487793,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2082398235797882,
+      "step": 18198
+    },
+    {
+      "epoch": 0.364,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0060791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1046,
+      "loss/crossentropy": 2.170537829399109,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19030719250440598,
+      "step": 18200
+    },
+    {
+      "epoch": 0.36404,
+      "grad_norm": 2.328125,
+      "grad_norm_var": 0.024234771728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.3921,
+      "loss/crossentropy": 1.938852846622467,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19586428999900818,
+      "step": 18202
+    },
+    {
+      "epoch": 0.36408,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.023996734619140626,
+      "learning_rate": 0.0001,
+      "loss": 4.0629,
+      "loss/crossentropy": 1.8909979462623596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17625273764133453,
+      "step": 18204
+    },
+    {
+      "epoch": 0.36412,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.022078450520833334,
+      "learning_rate": 0.0001,
+      "loss": 4.1729,
+      "loss/crossentropy": 1.9127016067504883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1888885200023651,
+      "step": 18206
+    },
+    {
+      "epoch": 0.36416,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.03103612263997396,
+      "learning_rate": 0.0001,
+      "loss": 4.1826,
+      "loss/crossentropy": 2.182482957839966,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21610775589942932,
+      "step": 18208
+    },
+    {
+      "epoch": 0.3642,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.027497355143229166,
+      "learning_rate": 0.0001,
+      "loss": 4.1941,
+      "loss/crossentropy": 2.0900736451148987,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20886409282684326,
+      "step": 18210
+    },
+    {
+      "epoch": 0.36424,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0248199462890625,
+      "learning_rate": 0.0001,
+      "loss": 3.9184,
+      "loss/crossentropy": 2.2132604122161865,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21213480830192566,
+      "step": 18212
+    },
+    {
+      "epoch": 0.36428,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.024825032552083334,
+      "learning_rate": 0.0001,
+      "loss": 4.0002,
+      "loss/crossentropy": 1.7810762524604797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.188262477517128,
+      "step": 18214
+    },
+    {
+      "epoch": 0.36432,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.02490208943684896,
+      "learning_rate": 0.0001,
+      "loss": 3.8339,
+      "loss/crossentropy": 2.072141647338867,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20580865442752838,
+      "step": 18216
+    },
+    {
+      "epoch": 0.36436,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.016778310139973957,
+      "learning_rate": 0.0001,
+      "loss": 4.0777,
+      "loss/crossentropy": 1.9672082662582397,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19521142542362213,
+      "step": 18218
+    },
+    {
+      "epoch": 0.3644,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.017574055989583334,
+      "learning_rate": 0.0001,
+      "loss": 4.0763,
+      "loss/crossentropy": 2.0122682452201843,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21602813154459,
+      "step": 18220
+    },
+    {
+      "epoch": 0.36444,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0169342041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0574,
+      "loss/crossentropy": 2.1709738969802856,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2267712950706482,
+      "step": 18222
+    },
+    {
+      "epoch": 0.36448,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009091949462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2092,
+      "loss/crossentropy": 2.246786117553711,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2128008008003235,
+      "step": 18224
+    },
+    {
+      "epoch": 0.36452,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.0063250223795572914,
+      "learning_rate": 0.0001,
+      "loss": 4.1051,
+      "loss/crossentropy": 1.686498999595642,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18233107775449753,
+      "step": 18226
+    },
+    {
+      "epoch": 0.36456,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.006786855061848959,
+      "learning_rate": 0.0001,
+      "loss": 4.3823,
+      "loss/crossentropy": 2.5849136114120483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23062889277935028,
+      "step": 18228
+    },
+    {
+      "epoch": 0.3646,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.005804189046223958,
+      "learning_rate": 0.0001,
+      "loss": 4.04,
+      "loss/crossentropy": 2.068212151527405,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20198806375265121,
+      "step": 18230
+    },
+    {
+      "epoch": 0.36464,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.00618896484375,
+      "learning_rate": 0.0001,
+      "loss": 3.8729,
+      "loss/crossentropy": 1.727245271205902,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17305339127779007,
+      "step": 18232
+    },
+    {
+      "epoch": 0.36468,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.005574289957682292,
+      "learning_rate": 0.0001,
+      "loss": 4.0496,
+      "loss/crossentropy": 2.0269583463668823,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2060471773147583,
+      "step": 18234
+    },
+    {
+      "epoch": 0.36472,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.004801177978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1707,
+      "loss/crossentropy": 1.9194093346595764,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19781427085399628,
+      "step": 18236
+    },
+    {
+      "epoch": 0.36476,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0055328369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1438,
+      "loss/crossentropy": 1.7539438605308533,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19985094666481018,
+      "step": 18238
+    },
+    {
+      "epoch": 0.3648,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.006068674723307291,
+      "learning_rate": 0.0001,
+      "loss": 4.1287,
+      "loss/crossentropy": 2.1092851161956787,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23365512490272522,
+      "step": 18240
+    },
+    {
+      "epoch": 0.36484,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.00771484375,
+      "learning_rate": 0.0001,
+      "loss": 3.6435,
+      "loss/crossentropy": 1.772037386894226,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18215186148881912,
+      "step": 18242
+    },
+    {
+      "epoch": 0.36488,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.007845052083333333,
+      "learning_rate": 0.0001,
+      "loss": 4.1017,
+      "loss/crossentropy": 2.0108843445777893,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20684240013360977,
+      "step": 18244
+    },
+    {
+      "epoch": 0.36492,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.008348592122395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0077,
+      "loss/crossentropy": 2.0076091289520264,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1842496171593666,
+      "step": 18246
+    },
+    {
+      "epoch": 0.36496,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.008085123697916667,
+      "learning_rate": 0.0001,
+      "loss": 4.1357,
+      "loss/crossentropy": 2.1824593544006348,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19478252530097961,
+      "step": 18248
+    },
+    {
+      "epoch": 0.365,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0068318684895833336,
+      "learning_rate": 0.0001,
+      "loss": 4.2231,
+      "loss/crossentropy": 2.1033846139907837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21361663192510605,
+      "step": 18250
+    },
+    {
+      "epoch": 0.36504,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.009299468994140626,
+      "learning_rate": 0.0001,
+      "loss": 3.7738,
+      "loss/crossentropy": 2.142418146133423,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20618261396884918,
+      "step": 18252
+    },
+    {
+      "epoch": 0.36508,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.008131663004557291,
+      "learning_rate": 0.0001,
+      "loss": 4.1037,
+      "loss/crossentropy": 2.061814546585083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20519836992025375,
+      "step": 18254
+    },
+    {
+      "epoch": 0.36512,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008512115478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0512,
+      "loss/crossentropy": 2.0216987133026123,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1956964135169983,
+      "step": 18256
+    },
+    {
+      "epoch": 0.36516,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.006681315104166667,
+      "learning_rate": 0.0001,
+      "loss": 4.301,
+      "loss/crossentropy": 2.2619231939315796,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21638523787260056,
+      "step": 18258
+    },
+    {
+      "epoch": 0.3652,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.006281534830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.165,
+      "loss/crossentropy": 1.8058243989944458,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19046328961849213,
+      "step": 18260
+    },
+    {
+      "epoch": 0.36524,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.005639394124348958,
+      "learning_rate": 0.0001,
+      "loss": 4.1876,
+      "loss/crossentropy": 2.0969032049179077,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20291081070899963,
+      "step": 18262
+    },
+    {
+      "epoch": 0.36528,
+      "grad_norm": 1.8125,
+      "grad_norm_var": 0.006794993082682292,
+      "learning_rate": 0.0001,
+      "loss": 3.8754,
+      "loss/crossentropy": 1.9567083716392517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19624605774879456,
+      "step": 18264
+    },
+    {
+      "epoch": 0.36532,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.006493123372395834,
+      "learning_rate": 0.0001,
+      "loss": 3.9861,
+      "loss/crossentropy": 1.9015939235687256,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18771862238645554,
+      "step": 18266
+    },
+    {
+      "epoch": 0.36536,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.017575836181640624,
+      "learning_rate": 0.0001,
+      "loss": 4.0288,
+      "loss/crossentropy": 1.9432410597801208,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1785590723156929,
+      "step": 18268
+    },
+    {
+      "epoch": 0.3654,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.017438761393229165,
+      "learning_rate": 0.0001,
+      "loss": 4.3256,
+      "loss/crossentropy": 2.216074585914612,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21463461220264435,
+      "step": 18270
+    },
+    {
+      "epoch": 0.36544,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.020401763916015624,
+      "learning_rate": 0.0001,
+      "loss": 3.7192,
+      "loss/crossentropy": 1.5440006256103516,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1621650904417038,
+      "step": 18272
+    },
+    {
+      "epoch": 0.36548,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.02197850545247396,
+      "learning_rate": 0.0001,
+      "loss": 3.9072,
+      "loss/crossentropy": 1.9478511214256287,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20275350660085678,
+      "step": 18274
+    },
+    {
+      "epoch": 0.36552,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.02217381795247396,
+      "learning_rate": 0.0001,
+      "loss": 4.1986,
+      "loss/crossentropy": 2.16433984041214,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21180611103773117,
+      "step": 18276
+    },
+    {
+      "epoch": 0.36556,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.02182184855143229,
+      "learning_rate": 0.0001,
+      "loss": 4.0345,
+      "loss/crossentropy": 1.9628196954727173,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.198894202709198,
+      "step": 18278
+    },
+    {
+      "epoch": 0.3656,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.020409901936848957,
+      "learning_rate": 0.0001,
+      "loss": 4.0616,
+      "loss/crossentropy": 1.7818017601966858,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19511079788208008,
+      "step": 18280
+    },
+    {
+      "epoch": 0.36564,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.02315241495768229,
+      "learning_rate": 0.0001,
+      "loss": 4.1769,
+      "loss/crossentropy": 2.0149444341659546,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2128366306424141,
+      "step": 18282
+    },
+    {
+      "epoch": 0.36568,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.010351308186848958,
+      "learning_rate": 0.0001,
+      "loss": 4.0475,
+      "loss/crossentropy": 2.1580519676208496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21166586130857468,
+      "step": 18284
+    },
+    {
+      "epoch": 0.36572,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.03432591756184896,
+      "learning_rate": 0.0001,
+      "loss": 3.9196,
+      "loss/crossentropy": 2.308629631996155,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20283856242895126,
+      "step": 18286
+    },
+    {
+      "epoch": 0.36576,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.03199055989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.0587,
+      "loss/crossentropy": 2.1761614084243774,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.212914377450943,
+      "step": 18288
+    },
+    {
+      "epoch": 0.3658,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.029842122395833334,
+      "learning_rate": 0.0001,
+      "loss": 4.191,
+      "loss/crossentropy": 1.9851300120353699,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1898990124464035,
+      "step": 18290
+    },
+    {
+      "epoch": 0.36584,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.031172688802083334,
+      "learning_rate": 0.0001,
+      "loss": 4.028,
+      "loss/crossentropy": 2.1123871207237244,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20982655137777328,
+      "step": 18292
+    },
+    {
+      "epoch": 0.36588,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.03831761678059896,
+      "learning_rate": 0.0001,
+      "loss": 4.3007,
+      "loss/crossentropy": 2.2666051387786865,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23328793793916702,
+      "step": 18294
+    },
+    {
+      "epoch": 0.36592,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.03904393513997396,
+      "learning_rate": 0.0001,
+      "loss": 3.8741,
+      "loss/crossentropy": 1.9444871544837952,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20335105061531067,
+      "step": 18296
+    },
+    {
+      "epoch": 0.36596,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.036717732747395836,
+      "learning_rate": 0.0001,
+      "loss": 4.0799,
+      "loss/crossentropy": 2.136604368686676,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20185644924640656,
+      "step": 18298
+    },
+    {
+      "epoch": 0.366,
+      "grad_norm": 2.59375,
+      "grad_norm_var": 0.059024810791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0605,
+      "loss/crossentropy": 1.9029142260551453,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24649157375097275,
+      "step": 18300
+    },
+    {
+      "epoch": 0.36604,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.0400054931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1685,
+      "loss/crossentropy": 2.0363988876342773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20920708030462265,
+      "step": 18302
+    },
+    {
+      "epoch": 0.36608,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.04248860677083333,
+      "learning_rate": 0.0001,
+      "loss": 3.5556,
+      "loss/crossentropy": 1.8357294797897339,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18201576173305511,
+      "step": 18304
+    },
+    {
+      "epoch": 0.36612,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.04253743489583333,
+      "learning_rate": 0.0001,
+      "loss": 4.3381,
+      "loss/crossentropy": 2.194098114967346,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20245323330163956,
+      "step": 18306
+    },
+    {
+      "epoch": 0.36616,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0409332275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2501,
+      "loss/crossentropy": 2.3840869665145874,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2243075668811798,
+      "step": 18308
+    },
+    {
+      "epoch": 0.3662,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.034398396809895836,
+      "learning_rate": 0.0001,
+      "loss": 4.1895,
+      "loss/crossentropy": 2.3093236684799194,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22531048208475113,
+      "step": 18310
+    },
+    {
+      "epoch": 0.36624,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.03472493489583333,
+      "learning_rate": 0.0001,
+      "loss": 3.7629,
+      "loss/crossentropy": 2.155013680458069,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19654065370559692,
+      "step": 18312
+    },
+    {
+      "epoch": 0.36628,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.03486226399739583,
+      "learning_rate": 0.0001,
+      "loss": 4.0241,
+      "loss/crossentropy": 1.8823603391647339,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18333254009485245,
+      "step": 18314
+    },
+    {
+      "epoch": 0.36632,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0090972900390625,
+      "learning_rate": 0.0001,
+      "loss": 3.8092,
+      "loss/crossentropy": 1.9217600226402283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18095016479492188,
+      "step": 18316
+    },
+    {
+      "epoch": 0.36636,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.009004720052083333,
+      "learning_rate": 0.0001,
+      "loss": 3.8178,
+      "loss/crossentropy": 1.7448294758796692,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18972519785165787,
+      "step": 18318
+    },
+    {
+      "epoch": 0.3664,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.008442942301432292,
+      "learning_rate": 0.0001,
+      "loss": 3.9394,
+      "loss/crossentropy": 1.8730336427688599,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18396812677383423,
+      "step": 18320
+    },
+    {
+      "epoch": 0.36644,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.006955718994140625,
+      "learning_rate": 0.0001,
+      "loss": 3.8763,
+      "loss/crossentropy": 2.075575351715088,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20251936465501785,
+      "step": 18322
+    },
+    {
+      "epoch": 0.36648,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.004131825764973959,
+      "learning_rate": 0.0001,
+      "loss": 4.0955,
+      "loss/crossentropy": 2.083173990249634,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2003224566578865,
+      "step": 18324
+    },
+    {
+      "epoch": 0.36652,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.00396728515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0245,
+      "loss/crossentropy": 2.4401766061782837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21872484683990479,
+      "step": 18326
+    },
+    {
+      "epoch": 0.36656,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0038533528645833333,
+      "learning_rate": 0.0001,
+      "loss": 4.0249,
+      "loss/crossentropy": 2.2402881383895874,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1977379471063614,
+      "step": 18328
+    },
+    {
+      "epoch": 0.3666,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.004650624593098959,
+      "learning_rate": 0.0001,
+      "loss": 4.1047,
+      "loss/crossentropy": 2.2246369123458862,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21132111549377441,
+      "step": 18330
+    },
+    {
+      "epoch": 0.36664,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0043365478515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1607,
+      "loss/crossentropy": 1.8048500418663025,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18298982083797455,
+      "step": 18332
+    },
+    {
+      "epoch": 0.36668,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.006514231363932292,
+      "learning_rate": 0.0001,
+      "loss": 4.0987,
+      "loss/crossentropy": 2.3368008136749268,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21743982285261154,
+      "step": 18334
+    },
+    {
+      "epoch": 0.36672,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.010188547770182292,
+      "learning_rate": 0.0001,
+      "loss": 4.3158,
+      "loss/crossentropy": 2.3520134687423706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22603602707386017,
+      "step": 18336
+    },
+    {
+      "epoch": 0.36676,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.011248524983723958,
+      "learning_rate": 0.0001,
+      "loss": 4.0037,
+      "loss/crossentropy": 2.0071199536323547,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18584266304969788,
+      "step": 18338
+    },
+    {
+      "epoch": 0.3668,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.019260406494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2787,
+      "loss/crossentropy": 1.9329636693000793,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21350353956222534,
+      "step": 18340
+    },
+    {
+      "epoch": 0.36684,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.015990193684895834,
+      "learning_rate": 0.0001,
+      "loss": 4.0387,
+      "loss/crossentropy": 1.8231948018074036,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2064230963587761,
+      "step": 18342
+    },
+    {
+      "epoch": 0.36688,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.01605809529622396,
+      "learning_rate": 0.0001,
+      "loss": 4.0888,
+      "loss/crossentropy": 1.9928399324417114,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1996312066912651,
+      "step": 18344
+    },
+    {
+      "epoch": 0.36692,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.015916951497395835,
+      "learning_rate": 0.0001,
+      "loss": 4.1736,
+      "loss/crossentropy": 2.139783501625061,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20407474786043167,
+      "step": 18346
+    },
+    {
+      "epoch": 0.36696,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.017166900634765624,
+      "learning_rate": 0.0001,
+      "loss": 4.0357,
+      "loss/crossentropy": 1.8519954681396484,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18404167890548706,
+      "step": 18348
+    },
+    {
+      "epoch": 0.367,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.016361236572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1325,
+      "loss/crossentropy": 2.037365674972534,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2384558469057083,
+      "step": 18350
+    },
+    {
+      "epoch": 0.36704,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.015600331624348958,
+      "learning_rate": 0.0001,
+      "loss": 3.887,
+      "loss/crossentropy": 2.0023937821388245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19700156897306442,
+      "step": 18352
+    },
+    {
+      "epoch": 0.36708,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.015282185872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0475,
+      "loss/crossentropy": 1.829107940196991,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19559869915246964,
+      "step": 18354
+    },
+    {
+      "epoch": 0.36712,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.006012980143229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0962,
+      "loss/crossentropy": 2.206624150276184,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2166636437177658,
+      "step": 18356
+    },
+    {
+      "epoch": 0.36716,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.005907185872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1191,
+      "loss/crossentropy": 2.1971875429153442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20924220979213715,
+      "step": 18358
+    },
+    {
+      "epoch": 0.3672,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0069539388020833336,
+      "learning_rate": 0.0001,
+      "loss": 4.4876,
+      "loss/crossentropy": 2.554604172706604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2638286352157593,
+      "step": 18360
+    },
+    {
+      "epoch": 0.36724,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.007981109619140624,
+      "learning_rate": 0.0001,
+      "loss": 3.9076,
+      "loss/crossentropy": 2.088346302509308,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1948780044913292,
+      "step": 18362
+    },
+    {
+      "epoch": 0.36728,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.0078033447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0101,
+      "loss/crossentropy": 2.24162495136261,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21616502106189728,
+      "step": 18364
+    },
+    {
+      "epoch": 0.36732,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.008567047119140626,
+      "learning_rate": 0.0001,
+      "loss": 3.8693,
+      "loss/crossentropy": 2.249878764152527,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2312733307480812,
+      "step": 18366
+    },
+    {
+      "epoch": 0.36736,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.0083404541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0711,
+      "loss/crossentropy": 2.0440531969070435,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18561603128910065,
+      "step": 18368
+    },
+    {
+      "epoch": 0.3674,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.006696573893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2582,
+      "loss/crossentropy": 2.2324228286743164,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20741964876651764,
+      "step": 18370
+    },
+    {
+      "epoch": 0.36744,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.008634440104166667,
+      "learning_rate": 0.0001,
+      "loss": 4.2807,
+      "loss/crossentropy": 2.168972373008728,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.215408056974411,
+      "step": 18372
+    },
+    {
+      "epoch": 0.36748,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.008955637613932291,
+      "learning_rate": 0.0001,
+      "loss": 4.0341,
+      "loss/crossentropy": 2.076392412185669,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2023996263742447,
+      "step": 18374
+    },
+    {
+      "epoch": 0.36752,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.005968983968098958,
+      "learning_rate": 0.0001,
+      "loss": 4.2958,
+      "loss/crossentropy": 2.0497928857803345,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21541842073202133,
+      "step": 18376
+    },
+    {
+      "epoch": 0.36756,
+      "grad_norm": 1.6875,
+      "grad_norm_var": 0.0102294921875,
+      "learning_rate": 0.0001,
+      "loss": 3.6889,
+      "loss/crossentropy": 1.8894451260566711,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18294879794120789,
+      "step": 18378
+    },
+    {
+      "epoch": 0.3676,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.010545857747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0237,
+      "loss/crossentropy": 2.1654014587402344,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18372679501771927,
+      "step": 18380
+    },
+    {
+      "epoch": 0.36764,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.013516998291015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0657,
+      "loss/crossentropy": 2.203396499156952,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22593200951814651,
+      "step": 18382
+    },
+    {
+      "epoch": 0.36768,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.014090728759765626,
+      "learning_rate": 0.0001,
+      "loss": 3.946,
+      "loss/crossentropy": 1.7496543526649475,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1825375333428383,
+      "step": 18384
+    },
+    {
+      "epoch": 0.36772,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.014249420166015625,
+      "learning_rate": 0.0001,
+      "loss": 3.7667,
+      "loss/crossentropy": 1.9488004446029663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18244057893753052,
+      "step": 18386
+    },
+    {
+      "epoch": 0.36776,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.01231689453125,
+      "learning_rate": 0.0001,
+      "loss": 3.9406,
+      "loss/crossentropy": 1.9792875051498413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20204880088567734,
+      "step": 18388
+    },
+    {
+      "epoch": 0.3678,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.012412261962890626,
+      "learning_rate": 0.0001,
+      "loss": 4.0077,
+      "loss/crossentropy": 1.890614092350006,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19333010911941528,
+      "step": 18390
+    },
+    {
+      "epoch": 0.36784,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.011832427978515626,
+      "learning_rate": 0.0001,
+      "loss": 4.0615,
+      "loss/crossentropy": 2.0901471972465515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20534122735261917,
+      "step": 18392
+    },
+    {
+      "epoch": 0.36788,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.00716552734375,
+      "learning_rate": 0.0001,
+      "loss": 4.2502,
+      "loss/crossentropy": 2.2931089401245117,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2130991816520691,
+      "step": 18394
+    },
+    {
+      "epoch": 0.36792,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0077392578125,
+      "learning_rate": 0.0001,
+      "loss": 3.9862,
+      "loss/crossentropy": 2.2144237756729126,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22273198515176773,
+      "step": 18396
+    },
+    {
+      "epoch": 0.36796,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.006624094645182292,
+      "learning_rate": 0.0001,
+      "loss": 3.7365,
+      "loss/crossentropy": 1.9135422110557556,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1983061358332634,
+      "step": 18398
+    },
+    {
+      "epoch": 0.368,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.007456207275390625,
+      "learning_rate": 0.0001,
+      "loss": 3.7556,
+      "loss/crossentropy": 1.692852795124054,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1670849658548832,
+      "step": 18400
+    },
+    {
+      "epoch": 0.36804,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0083160400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0043,
+      "loss/crossentropy": 1.908652126789093,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21787738054990768,
+      "step": 18402
+    },
+    {
+      "epoch": 0.36808,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008432769775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2343,
+      "loss/crossentropy": 2.0613549947738647,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20495950430631638,
+      "step": 18404
+    },
+    {
+      "epoch": 0.36812,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.026759592692057292,
+      "learning_rate": 0.0001,
+      "loss": 4.0549,
+      "loss/crossentropy": 1.9765326976776123,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18859146535396576,
+      "step": 18406
+    },
+    {
+      "epoch": 0.36816,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.027147420247395835,
+      "learning_rate": 0.0001,
+      "loss": 4.0196,
+      "loss/crossentropy": 1.8403696417808533,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18926746398210526,
+      "step": 18408
+    },
+    {
+      "epoch": 0.3682,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.029080963134765624,
+      "learning_rate": 0.0001,
+      "loss": 4.1859,
+      "loss/crossentropy": 2.1792644262313843,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22263357043266296,
+      "step": 18410
+    },
+    {
+      "epoch": 0.36824,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.02910334269205729,
+      "learning_rate": 0.0001,
+      "loss": 4.2387,
+      "loss/crossentropy": 2.079226016998291,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20990607887506485,
+      "step": 18412
+    },
+    {
+      "epoch": 0.36828,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.027457427978515626,
+      "learning_rate": 0.0001,
+      "loss": 4.2093,
+      "loss/crossentropy": 2.1428889632225037,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22140806913375854,
+      "step": 18414
+    },
+    {
+      "epoch": 0.36832,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.024589029947916667,
+      "learning_rate": 0.0001,
+      "loss": 4.3761,
+      "loss/crossentropy": 2.1461609601974487,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20594316720962524,
+      "step": 18416
+    },
+    {
+      "epoch": 0.36836,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.02399470011393229,
+      "learning_rate": 0.0001,
+      "loss": 4.1597,
+      "loss/crossentropy": 1.980036735534668,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.201970636844635,
+      "step": 18418
+    },
+    {
+      "epoch": 0.3684,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.024773915608723957,
+      "learning_rate": 0.0001,
+      "loss": 3.9847,
+      "loss/crossentropy": 1.6273554563522339,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1938486322760582,
+      "step": 18420
+    },
+    {
+      "epoch": 0.36844,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.006251780192057291,
+      "learning_rate": 0.0001,
+      "loss": 4.1784,
+      "loss/crossentropy": 1.9100900292396545,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18063092976808548,
+      "step": 18422
+    },
+    {
+      "epoch": 0.36848,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0063555399576822914,
+      "learning_rate": 0.0001,
+      "loss": 4.1282,
+      "loss/crossentropy": 2.1882822513580322,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21261122822761536,
+      "step": 18424
+    },
+    {
+      "epoch": 0.36852,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.0046384175618489586,
+      "learning_rate": 0.0001,
+      "loss": 3.9556,
+      "loss/crossentropy": 2.210257649421692,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1929008737206459,
+      "step": 18426
+    },
+    {
+      "epoch": 0.36856,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.005041249593098958,
+      "learning_rate": 0.0001,
+      "loss": 4.1691,
+      "loss/crossentropy": 2.0247724056243896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2004971206188202,
+      "step": 18428
+    },
+    {
+      "epoch": 0.3686,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.004133097330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.2092,
+      "loss/crossentropy": 1.9844316244125366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.210312619805336,
+      "step": 18430
+    },
+    {
+      "epoch": 0.36864,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.005000559488932291,
+      "learning_rate": 0.0001,
+      "loss": 4.0368,
+      "loss/crossentropy": 2.2416625022888184,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21065251529216766,
+      "step": 18432
+    },
+    {
+      "epoch": 0.36868,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.020104726155598957,
+      "learning_rate": 0.0001,
+      "loss": 4.4109,
+      "loss/crossentropy": 2.0283551812171936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2150302305817604,
+      "step": 18434
+    },
+    {
+      "epoch": 0.36872,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.01968994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1379,
+      "loss/crossentropy": 2.210233688354492,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21831902861595154,
+      "step": 18436
+    },
+    {
+      "epoch": 0.36876,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0205474853515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9993,
+      "loss/crossentropy": 2.0454147458076477,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2432640790939331,
+      "step": 18438
+    },
+    {
+      "epoch": 0.3688,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.020444488525390624,
+      "learning_rate": 0.0001,
+      "loss": 4.1243,
+      "loss/crossentropy": 2.024592399597168,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20496471971273422,
+      "step": 18440
+    },
+    {
+      "epoch": 0.36884,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.018822987874348957,
+      "learning_rate": 0.0001,
+      "loss": 4.3058,
+      "loss/crossentropy": 2.4205459356307983,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22281523793935776,
+      "step": 18442
+    },
+    {
+      "epoch": 0.36888,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0170806884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0341,
+      "loss/crossentropy": 1.9198943376541138,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1916896402835846,
+      "step": 18444
+    },
+    {
+      "epoch": 0.36892,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0171051025390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0475,
+      "loss/crossentropy": 1.928059160709381,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1899876967072487,
+      "step": 18446
+    },
+    {
+      "epoch": 0.36896,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.020792388916015626,
+      "learning_rate": 0.0001,
+      "loss": 4.3327,
+      "loss/crossentropy": 2.346290349960327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22388508170843124,
+      "step": 18448
+    },
+    {
+      "epoch": 0.369,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.009946441650390625,
+      "learning_rate": 0.0001,
+      "loss": 3.9064,
+      "loss/crossentropy": 2.0108938217163086,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19462434202432632,
+      "step": 18450
+    },
+    {
+      "epoch": 0.36904,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.00986328125,
+      "learning_rate": 0.0001,
+      "loss": 4.2468,
+      "loss/crossentropy": 1.7108886241912842,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18806710839271545,
+      "step": 18452
+    },
+    {
+      "epoch": 0.36908,
+      "grad_norm": 5.1875,
+      "grad_norm_var": 0.6345499674479167,
+      "learning_rate": 0.0001,
+      "loss": 4.6972,
+      "loss/crossentropy": 2.4162802696228027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3217930570244789,
+      "step": 18454
+    },
+    {
+      "epoch": 0.36912,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.639013671875,
+      "learning_rate": 0.0001,
+      "loss": 3.4762,
+      "loss/crossentropy": 1.6459838151931763,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17543485760688782,
+      "step": 18456
+    },
+    {
+      "epoch": 0.36916,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.640679677327474,
+      "learning_rate": 0.0001,
+      "loss": 4.1194,
+      "loss/crossentropy": 2.2111966013908386,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22191359847784042,
+      "step": 18458
+    },
+    {
+      "epoch": 0.3692,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.6325887044270834,
+      "learning_rate": 0.0001,
+      "loss": 4.301,
+      "loss/crossentropy": 2.2514692544937134,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21384654194116592,
+      "step": 18460
+    },
+    {
+      "epoch": 0.36924,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.63665771484375,
+      "learning_rate": 0.0001,
+      "loss": 4.307,
+      "loss/crossentropy": 2.097872793674469,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21747223287820816,
+      "step": 18462
+    },
+    {
+      "epoch": 0.36928,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.6418690999348958,
+      "learning_rate": 0.0001,
+      "loss": 4.1033,
+      "loss/crossentropy": 2.1611807346343994,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2042219042778015,
+      "step": 18464
+    },
+    {
+      "epoch": 0.36932,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.62939453125,
+      "learning_rate": 0.0001,
+      "loss": 3.7871,
+      "loss/crossentropy": 1.8941562175750732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.192266546189785,
+      "step": 18466
+    },
+    {
+      "epoch": 0.36936,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.72021484375,
+      "learning_rate": 0.0001,
+      "loss": 4.0358,
+      "loss/crossentropy": 1.8930317163467407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19436348974704742,
+      "step": 18468
+    },
+    {
+      "epoch": 0.3694,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.15979588826497396,
+      "learning_rate": 0.0001,
+      "loss": 4.1046,
+      "loss/crossentropy": 2.1041005849838257,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21314333379268646,
+      "step": 18470
+    },
+    {
+      "epoch": 0.36944,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.15750732421875,
+      "learning_rate": 0.0001,
+      "loss": 4.2096,
+      "loss/crossentropy": 1.8747637867927551,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19172174483537674,
+      "step": 18472
+    },
+    {
+      "epoch": 0.36948,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.15751113891601562,
+      "learning_rate": 0.0001,
+      "loss": 3.9809,
+      "loss/crossentropy": 2.1883610486984253,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19927512109279633,
+      "step": 18474
+    },
+    {
+      "epoch": 0.36952,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.1606353759765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9303,
+      "loss/crossentropy": 1.972772240638733,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17948149144649506,
+      "step": 18476
+    },
+    {
+      "epoch": 0.36956,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.14570490519205728,
+      "learning_rate": 0.0001,
+      "loss": 4.1246,
+      "loss/crossentropy": 1.8912597298622131,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18716244399547577,
+      "step": 18478
+    },
+    {
+      "epoch": 0.3696,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.1458740234375,
+      "learning_rate": 0.0001,
+      "loss": 4.0816,
+      "loss/crossentropy": 2.044828712940216,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2076854333281517,
+      "step": 18480
+    },
+    {
+      "epoch": 0.36964,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.14629618326822916,
+      "learning_rate": 0.0001,
+      "loss": 4.2007,
+      "loss/crossentropy": 2.0936968326568604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22010911256074905,
+      "step": 18482
+    },
+    {
+      "epoch": 0.36968,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0027414957682291665,
+      "learning_rate": 0.0001,
+      "loss": 4.08,
+      "loss/crossentropy": 2.2678394317626953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2073991820216179,
+      "step": 18484
+    },
+    {
+      "epoch": 0.36972,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0031064351399739585,
+      "learning_rate": 0.0001,
+      "loss": 3.9361,
+      "loss/crossentropy": 2.281362295150757,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2015322744846344,
+      "step": 18486
+    },
+    {
+      "epoch": 0.36976,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0034739176432291665,
+      "learning_rate": 0.0001,
+      "loss": 4.2318,
+      "loss/crossentropy": 2.0029674768447876,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2155751883983612,
+      "step": 18488
+    },
+    {
+      "epoch": 0.3698,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.003525543212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.3761,
+      "loss/crossentropy": 2.370519280433655,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2308923304080963,
+      "step": 18490
+    },
+    {
+      "epoch": 0.36984,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.004109446207682292,
+      "learning_rate": 0.0001,
+      "loss": 4.3137,
+      "loss/crossentropy": 1.9416582584381104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18809420615434647,
+      "step": 18492
+    },
+    {
+      "epoch": 0.36988,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.003885650634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1381,
+      "loss/crossentropy": 2.065304160118103,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19029072672128677,
+      "step": 18494
+    },
+    {
+      "epoch": 0.36992,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.0036516825358072916,
+      "learning_rate": 0.0001,
+      "loss": 4.1366,
+      "loss/crossentropy": 1.9798340201377869,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1716388538479805,
+      "step": 18496
+    },
+    {
+      "epoch": 0.36996,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0035540262858072915,
+      "learning_rate": 0.0001,
+      "loss": 4.2321,
+      "loss/crossentropy": 2.336732864379883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19805839657783508,
+      "step": 18498
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0033444722493489584,
+      "learning_rate": 0.0001,
+      "loss": 4.08,
+      "loss/crossentropy": 1.9556902050971985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19576922804117203,
+      "step": 18500
+    },
+    {
+      "epoch": 0.37004,
+      "grad_norm": 1.796875,
+      "grad_norm_var": 0.006581370035807292,
+      "learning_rate": 0.0001,
+      "loss": 3.7281,
+      "loss/crossentropy": 2.0062427520751953,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19548972696065903,
+      "step": 18502
+    },
+    {
+      "epoch": 0.37008,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0061419169108072914,
+      "learning_rate": 0.0001,
+      "loss": 4.0954,
+      "loss/crossentropy": 2.1030293703079224,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22147215902805328,
+      "step": 18504
+    },
+    {
+      "epoch": 0.37012,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.005163319905598958,
+      "learning_rate": 0.0001,
+      "loss": 3.9545,
+      "loss/crossentropy": 1.7618860006332397,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17934025079011917,
+      "step": 18506
+    },
+    {
+      "epoch": 0.37016,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.005454254150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.239,
+      "loss/crossentropy": 2.4424854516983032,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23131447285413742,
+      "step": 18508
+    },
+    {
+      "epoch": 0.3702,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0076904296875,
+      "learning_rate": 0.0001,
+      "loss": 4.0365,
+      "loss/crossentropy": 2.125267446041107,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19127625226974487,
+      "step": 18510
+    },
+    {
+      "epoch": 0.37024,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.008506011962890626,
+      "learning_rate": 0.0001,
+      "loss": 3.9498,
+      "loss/crossentropy": 2.2222214937210083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.204838365316391,
+      "step": 18512
+    },
+    {
+      "epoch": 0.37028,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.009056599934895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1012,
+      "loss/crossentropy": 2.1930031776428223,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21769095957279205,
+      "step": 18514
+    },
+    {
+      "epoch": 0.37032,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.010814412434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.0068,
+      "loss/crossentropy": 2.078152060508728,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19528701901435852,
+      "step": 18516
+    },
+    {
+      "epoch": 0.37036,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.009016927083333333,
+      "learning_rate": 0.0001,
+      "loss": 4.0135,
+      "loss/crossentropy": 1.9786349534988403,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19193317741155624,
+      "step": 18518
+    },
+    {
+      "epoch": 0.3704,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0119293212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1458,
+      "loss/crossentropy": 2.279360294342041,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20865458250045776,
+      "step": 18520
+    },
+    {
+      "epoch": 0.37044,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.012548828125,
+      "learning_rate": 0.0001,
+      "loss": 4.195,
+      "loss/crossentropy": 2.0668978691101074,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19060996919870377,
+      "step": 18522
+    },
+    {
+      "epoch": 0.37048,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.011116536458333333,
+      "learning_rate": 0.0001,
+      "loss": 4.2661,
+      "loss/crossentropy": 2.4433913230895996,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2263498529791832,
+      "step": 18524
+    },
+    {
+      "epoch": 0.37052,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.009049224853515624,
+      "learning_rate": 0.0001,
+      "loss": 3.7953,
+      "loss/crossentropy": 1.7785582542419434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18417692929506302,
+      "step": 18526
+    },
+    {
+      "epoch": 0.37056,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008499908447265624,
+      "learning_rate": 0.0001,
+      "loss": 4.2958,
+      "loss/crossentropy": 1.9753797054290771,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1898263543844223,
+      "step": 18528
+    },
+    {
+      "epoch": 0.3706,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.007696278889973958,
+      "learning_rate": 0.0001,
+      "loss": 4.1945,
+      "loss/crossentropy": 2.302052319049835,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21289903670549393,
+      "step": 18530
+    },
+    {
+      "epoch": 0.37064,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.007523600260416667,
+      "learning_rate": 0.0001,
+      "loss": 4.1132,
+      "loss/crossentropy": 2.3275071382522583,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2265043556690216,
+      "step": 18532
+    },
+    {
+      "epoch": 0.37068,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.00826416015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2004,
+      "loss/crossentropy": 2.3155715465545654,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20101473480463028,
+      "step": 18534
+    },
+    {
+      "epoch": 0.37072,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.0075681050618489586,
+      "learning_rate": 0.0001,
+      "loss": 3.8546,
+      "loss/crossentropy": 2.0893908739089966,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18266596645116806,
+      "step": 18536
+    },
+    {
+      "epoch": 0.37076,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.006546783447265625,
+      "learning_rate": 0.0001,
+      "loss": 4.297,
+      "loss/crossentropy": 2.075433909893036,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20270193368196487,
+      "step": 18538
+    },
+    {
+      "epoch": 0.3708,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.006605784098307292,
+      "learning_rate": 0.0001,
+      "loss": 4.186,
+      "loss/crossentropy": 2.2795485258102417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21547292172908783,
+      "step": 18540
+    },
+    {
+      "epoch": 0.37084,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.006956990559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0351,
+      "loss/crossentropy": 2.0038467049598694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19927001744508743,
+      "step": 18542
+    },
+    {
+      "epoch": 0.37088,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0056111653645833336,
+      "learning_rate": 0.0001,
+      "loss": 3.979,
+      "loss/crossentropy": 1.8877951502799988,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19283732771873474,
+      "step": 18544
+    },
+    {
+      "epoch": 0.37092,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0065305074055989586,
+      "learning_rate": 0.0001,
+      "loss": 4.0169,
+      "loss/crossentropy": 1.8111292719841003,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20587582886219025,
+      "step": 18546
+    },
+    {
+      "epoch": 0.37096,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.004548136393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0766,
+      "loss/crossentropy": 2.12824147939682,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1996314972639084,
+      "step": 18548
+    },
+    {
+      "epoch": 0.371,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.0035845438639322915,
+      "learning_rate": 0.0001,
+      "loss": 3.9549,
+      "loss/crossentropy": 2.22495698928833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20922152698040009,
+      "step": 18550
+    },
+    {
+      "epoch": 0.37104,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.003110504150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2245,
+      "loss/crossentropy": 2.1434414386749268,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2070355862379074,
+      "step": 18552
+    },
+    {
+      "epoch": 0.37108,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.003979237874348959,
+      "learning_rate": 0.0001,
+      "loss": 4.0359,
+      "loss/crossentropy": 2.248233437538147,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21400754898786545,
+      "step": 18554
+    },
+    {
+      "epoch": 0.37112,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0051513671875,
+      "learning_rate": 0.0001,
+      "loss": 4.2072,
+      "loss/crossentropy": 2.246641755104065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22847777605056763,
+      "step": 18556
+    },
+    {
+      "epoch": 0.37116,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.004654693603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0805,
+      "loss/crossentropy": 2.057798206806183,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23668289184570312,
+      "step": 18558
+    },
+    {
+      "epoch": 0.3712,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.004587554931640625,
+      "learning_rate": 0.0001,
+      "loss": 3.8454,
+      "loss/crossentropy": 1.7570822834968567,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1820850819349289,
+      "step": 18560
+    },
+    {
+      "epoch": 0.37124,
+      "grad_norm": 2.390625,
+      "grad_norm_var": 0.015372467041015626,
+      "learning_rate": 0.0001,
+      "loss": 4.0077,
+      "loss/crossentropy": 1.7747303247451782,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1704205796122551,
+      "step": 18562
+    },
+    {
+      "epoch": 0.37128,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.015632120768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0611,
+      "loss/crossentropy": 2.3060861825942993,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20002726465463638,
+      "step": 18564
+    },
+    {
+      "epoch": 0.37132,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.015242258707682291,
+      "learning_rate": 0.0001,
+      "loss": 4.258,
+      "loss/crossentropy": 2.2742475271224976,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22524481266736984,
+      "step": 18566
+    },
+    {
+      "epoch": 0.37136,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.01647923787434896,
+      "learning_rate": 0.0001,
+      "loss": 3.9816,
+      "loss/crossentropy": 1.7929689288139343,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1819063350558281,
+      "step": 18568
+    },
+    {
+      "epoch": 0.3714,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.015718587239583335,
+      "learning_rate": 0.0001,
+      "loss": 4.2487,
+      "loss/crossentropy": 2.4505950212478638,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22857815772294998,
+      "step": 18570
+    },
+    {
+      "epoch": 0.37144,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.017658487955729166,
+      "learning_rate": 0.0001,
+      "loss": 3.5856,
+      "loss/crossentropy": 1.640372097492218,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16958069056272507,
+      "step": 18572
+    },
+    {
+      "epoch": 0.37148,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.01916681925455729,
+      "learning_rate": 0.0001,
+      "loss": 4.1601,
+      "loss/crossentropy": 2.1556472778320312,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20749886333942413,
+      "step": 18574
+    },
+    {
+      "epoch": 0.37152,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.01846491495768229,
+      "learning_rate": 0.0001,
+      "loss": 4.1579,
+      "loss/crossentropy": 2.1131449937820435,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20596522092819214,
+      "step": 18576
+    },
+    {
+      "epoch": 0.37156,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.008998362223307292,
+      "learning_rate": 0.0001,
+      "loss": 3.8036,
+      "loss/crossentropy": 1.8966050148010254,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19289513677358627,
+      "step": 18578
+    },
+    {
+      "epoch": 0.3716,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.010306803385416667,
+      "learning_rate": 0.0001,
+      "loss": 4.1883,
+      "loss/crossentropy": 2.1016936898231506,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1917622685432434,
+      "step": 18580
+    },
+    {
+      "epoch": 0.37164,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.01512451171875,
+      "learning_rate": 0.0001,
+      "loss": 4.2596,
+      "loss/crossentropy": 2.289618492126465,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2033960521221161,
+      "step": 18582
+    },
+    {
+      "epoch": 0.37168,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.020287068684895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0441,
+      "loss/crossentropy": 2.205715775489807,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20015332102775574,
+      "step": 18584
+    },
+    {
+      "epoch": 0.37172,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.02072118123372396,
+      "learning_rate": 0.0001,
+      "loss": 4.0834,
+      "loss/crossentropy": 2.3037471771240234,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2424507886171341,
+      "step": 18586
+    },
+    {
+      "epoch": 0.37176,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.016686757405598957,
+      "learning_rate": 0.0001,
+      "loss": 4.0718,
+      "loss/crossentropy": 2.1757054328918457,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20777830481529236,
+      "step": 18588
+    },
+    {
+      "epoch": 0.3718,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.015860748291015626,
+      "learning_rate": 0.0001,
+      "loss": 4.2905,
+      "loss/crossentropy": 2.3257133960723877,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22848188877105713,
+      "step": 18590
+    },
+    {
+      "epoch": 0.37184,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.017179107666015624,
+      "learning_rate": 0.0001,
+      "loss": 3.9132,
+      "loss/crossentropy": 2.3375465869903564,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21214719116687775,
+      "step": 18592
+    },
+    {
+      "epoch": 0.37188,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.0162750244140625,
+      "learning_rate": 0.0001,
+      "loss": 4.507,
+      "loss/crossentropy": 2.1896092891693115,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2303280457854271,
+      "step": 18594
+    },
+    {
+      "epoch": 0.37192,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.016287994384765626,
+      "learning_rate": 0.0001,
+      "loss": 3.9078,
+      "loss/crossentropy": 1.8913645148277283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17311514914035797,
+      "step": 18596
+    },
+    {
+      "epoch": 0.37196,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.012442779541015626,
+      "learning_rate": 0.0001,
+      "loss": 4.2212,
+      "loss/crossentropy": 2.5278197526931763,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23568174242973328,
+      "step": 18598
+    },
+    {
+      "epoch": 0.372,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0101959228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1568,
+      "loss/crossentropy": 2.073936700820923,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21885155141353607,
+      "step": 18600
+    },
+    {
+      "epoch": 0.37204,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.012684885660807292,
+      "learning_rate": 0.0001,
+      "loss": 4.388,
+      "loss/crossentropy": 2.162986159324646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2012084275484085,
+      "step": 18602
+    },
+    {
+      "epoch": 0.37208,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.013108062744140624,
+      "learning_rate": 0.0001,
+      "loss": 3.9659,
+      "loss/crossentropy": 1.9332863092422485,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18284077942371368,
+      "step": 18604
+    },
+    {
+      "epoch": 0.37212,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.013809967041015624,
+      "learning_rate": 0.0001,
+      "loss": 4.0279,
+      "loss/crossentropy": 2.083172380924225,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.186289481818676,
+      "step": 18606
+    },
+    {
+      "epoch": 0.37216,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.013588205973307291,
+      "learning_rate": 0.0001,
+      "loss": 3.9932,
+      "loss/crossentropy": 2.100313901901245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1977698802947998,
+      "step": 18608
+    },
+    {
+      "epoch": 0.3722,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.016792805989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.1086,
+      "loss/crossentropy": 2.2336236238479614,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21973469108343124,
+      "step": 18610
+    },
+    {
+      "epoch": 0.37224,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.017048136393229166,
+      "learning_rate": 0.0001,
+      "loss": 3.9422,
+      "loss/crossentropy": 2.0263352394104004,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19232206791639328,
+      "step": 18612
+    },
+    {
+      "epoch": 0.37228,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.01727472941080729,
+      "learning_rate": 0.0001,
+      "loss": 4.2295,
+      "loss/crossentropy": 2.0719032287597656,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22896382957696915,
+      "step": 18614
+    },
+    {
+      "epoch": 0.37232,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.015400950113932292,
+      "learning_rate": 0.0001,
+      "loss": 3.881,
+      "loss/crossentropy": 1.9514707326889038,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1899290755391121,
+      "step": 18616
+    },
+    {
+      "epoch": 0.37236,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.010172526041666666,
+      "learning_rate": 0.0001,
+      "loss": 4.0163,
+      "loss/crossentropy": 2.0810786485671997,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1983262002468109,
+      "step": 18618
+    },
+    {
+      "epoch": 0.3724,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.010302734375,
+      "learning_rate": 0.0001,
+      "loss": 4.1311,
+      "loss/crossentropy": 1.823366403579712,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18820683658123016,
+      "step": 18620
+    },
+    {
+      "epoch": 0.37244,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 0.023933919270833333,
+      "learning_rate": 0.0001,
+      "loss": 4.2361,
+      "loss/crossentropy": 2.2814120054244995,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19500700384378433,
+      "step": 18622
+    },
+    {
+      "epoch": 0.37248,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.023339589436848957,
+      "learning_rate": 0.0001,
+      "loss": 3.9548,
+      "loss/crossentropy": 1.5761349201202393,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17777466028928757,
+      "step": 18624
+    },
+    {
+      "epoch": 0.37252,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.018853505452473957,
+      "learning_rate": 0.0001,
+      "loss": 3.9134,
+      "loss/crossentropy": 1.8996286988258362,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21907050907611847,
+      "step": 18626
+    },
+    {
+      "epoch": 0.37256,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.020869700113932292,
+      "learning_rate": 0.0001,
+      "loss": 4.2361,
+      "loss/crossentropy": 1.6650620698928833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18808475136756897,
+      "step": 18628
+    },
+    {
+      "epoch": 0.3726,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.02156550089518229,
+      "learning_rate": 0.0001,
+      "loss": 4.4004,
+      "loss/crossentropy": 2.2887942790985107,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20416373759508133,
+      "step": 18630
+    },
+    {
+      "epoch": 0.37264,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.018629709879557293,
+      "learning_rate": 0.0001,
+      "loss": 3.9606,
+      "loss/crossentropy": 1.9479430317878723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18489708751440048,
+      "step": 18632
+    },
+    {
+      "epoch": 0.37268,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.016676584879557293,
+      "learning_rate": 0.0001,
+      "loss": 4.3041,
+      "loss/crossentropy": 2.2156901359558105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20997263491153717,
+      "step": 18634
+    },
+    {
+      "epoch": 0.37272,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.015950520833333332,
+      "learning_rate": 0.0001,
+      "loss": 4.1619,
+      "loss/crossentropy": 2.50583279132843,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23347270488739014,
+      "step": 18636
+    },
+    {
+      "epoch": 0.37276,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.008650716145833333,
+      "learning_rate": 0.0001,
+      "loss": 3.8241,
+      "loss/crossentropy": 1.7688243985176086,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18601343035697937,
+      "step": 18638
+    },
+    {
+      "epoch": 0.3728,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009439849853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2241,
+      "loss/crossentropy": 2.3786444664001465,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2287142351269722,
+      "step": 18640
+    },
+    {
+      "epoch": 0.37284,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.009479777018229166,
+      "learning_rate": 0.0001,
+      "loss": 4.243,
+      "loss/crossentropy": 2.2686339616775513,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21086719632148743,
+      "step": 18642
+    },
+    {
+      "epoch": 0.37288,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.007490793863932292,
+      "learning_rate": 0.0001,
+      "loss": 3.8341,
+      "loss/crossentropy": 1.9386130571365356,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19289222359657288,
+      "step": 18644
+    },
+    {
+      "epoch": 0.37292,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.006819407145182292,
+      "learning_rate": 0.0001,
+      "loss": 4.3357,
+      "loss/crossentropy": 2.3655601739883423,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2207036018371582,
+      "step": 18646
+    },
+    {
+      "epoch": 0.37296,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.007100168863932292,
+      "learning_rate": 0.0001,
+      "loss": 3.9039,
+      "loss/crossentropy": 1.9600831270217896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18018172681331635,
+      "step": 18648
+    },
+    {
+      "epoch": 0.373,
+      "grad_norm": 2.515625,
+      "grad_norm_var": 0.024873860677083335,
+      "learning_rate": 0.0001,
+      "loss": 3.9492,
+      "loss/crossentropy": 1.86528480052948,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19763591885566711,
+      "step": 18650
+    },
+    {
+      "epoch": 0.37304,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.024925740559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.07,
+      "loss/crossentropy": 2.0938327312469482,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2051714062690735,
+      "step": 18652
+    },
+    {
+      "epoch": 0.37308,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.022454579671223957,
+      "learning_rate": 0.0001,
+      "loss": 3.878,
+      "loss/crossentropy": 1.964399516582489,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20198997855186462,
+      "step": 18654
+    },
+    {
+      "epoch": 0.37312,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0245849609375,
+      "learning_rate": 0.0001,
+      "loss": 3.9929,
+      "loss/crossentropy": 2.0186336040496826,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19310183823108673,
+      "step": 18656
+    },
+    {
+      "epoch": 0.37316,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.025052642822265624,
+      "learning_rate": 0.0001,
+      "loss": 3.8569,
+      "loss/crossentropy": 1.8805240392684937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20392487198114395,
+      "step": 18658
+    },
+    {
+      "epoch": 0.3732,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.025480143229166665,
+      "learning_rate": 0.0001,
+      "loss": 4.0306,
+      "loss/crossentropy": 2.45102322101593,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2161283940076828,
+      "step": 18660
+    },
+    {
+      "epoch": 0.37324,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.025679270426432293,
+      "learning_rate": 0.0001,
+      "loss": 4.1015,
+      "loss/crossentropy": 2.2971357107162476,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23155176639556885,
+      "step": 18662
+    },
+    {
+      "epoch": 0.37328,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.025614166259765626,
+      "learning_rate": 0.0001,
+      "loss": 4.285,
+      "loss/crossentropy": 2.3736867904663086,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22897879034280777,
+      "step": 18664
+    },
+    {
+      "epoch": 0.37332,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.005500284830729166,
+      "learning_rate": 0.0001,
+      "loss": 3.9884,
+      "loss/crossentropy": 1.5138108134269714,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1526506468653679,
+      "step": 18666
+    },
+    {
+      "epoch": 0.37336,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.0061948140462239586,
+      "learning_rate": 0.0001,
+      "loss": 3.9507,
+      "loss/crossentropy": 2.415435791015625,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21689960360527039,
+      "step": 18668
+    },
+    {
+      "epoch": 0.3734,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.006306966145833333,
+      "learning_rate": 0.0001,
+      "loss": 4.014,
+      "loss/crossentropy": 2.1235941648483276,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20921900868415833,
+      "step": 18670
+    },
+    {
+      "epoch": 0.37344,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.004428863525390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2549,
+      "loss/crossentropy": 2.3612579703330994,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21255160123109818,
+      "step": 18672
+    },
+    {
+      "epoch": 0.37348,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.054351552327473955,
+      "learning_rate": 0.0001,
+      "loss": 4.1722,
+      "loss/crossentropy": 2.23550283908844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18890902400016785,
+      "step": 18674
+    },
+    {
+      "epoch": 0.37352,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.05468317667643229,
+      "learning_rate": 0.0001,
+      "loss": 4.0985,
+      "loss/crossentropy": 1.7998243570327759,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19070486724376678,
+      "step": 18676
+    },
+    {
+      "epoch": 0.37356,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.055214182535807295,
+      "learning_rate": 0.0001,
+      "loss": 4.151,
+      "loss/crossentropy": 1.892760992050171,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19817107915878296,
+      "step": 18678
+    },
+    {
+      "epoch": 0.3736,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.055757395426432294,
+      "learning_rate": 0.0001,
+      "loss": 3.9862,
+      "loss/crossentropy": 1.9619091153144836,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21299303323030472,
+      "step": 18680
+    },
+    {
+      "epoch": 0.37364,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.057889556884765624,
+      "learning_rate": 0.0001,
+      "loss": 3.8419,
+      "loss/crossentropy": 2.0857014656066895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20634424686431885,
+      "step": 18682
+    },
+    {
+      "epoch": 0.37368,
+      "grad_norm": 1.796875,
+      "grad_norm_var": 0.05814793904622396,
+      "learning_rate": 0.0001,
+      "loss": 3.9589,
+      "loss/crossentropy": 2.1093358397483826,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1933949589729309,
+      "step": 18684
+    },
+    {
+      "epoch": 0.37372,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.05987548828125,
+      "learning_rate": 0.0001,
+      "loss": 3.9537,
+      "loss/crossentropy": 2.1040873527526855,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2061748430132866,
+      "step": 18686
+    },
+    {
+      "epoch": 0.37376,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.059081776936848955,
+      "learning_rate": 0.0001,
+      "loss": 4.1676,
+      "loss/crossentropy": 2.4795751571655273,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2075038179755211,
+      "step": 18688
+    },
+    {
+      "epoch": 0.3738,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.006599934895833334,
+      "learning_rate": 0.0001,
+      "loss": 3.9992,
+      "loss/crossentropy": 1.864591360092163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17119919508695602,
+      "step": 18690
+    },
+    {
+      "epoch": 0.37384,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.007063547770182292,
+      "learning_rate": 0.0001,
+      "loss": 3.7565,
+      "loss/crossentropy": 1.6705753207206726,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19292957335710526,
+      "step": 18692
+    },
+    {
+      "epoch": 0.37388,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.004209136962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0448,
+      "loss/crossentropy": 2.16109561920166,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20515639334917068,
+      "step": 18694
+    },
+    {
+      "epoch": 0.37392,
+      "grad_norm": 2.28125,
+      "grad_norm_var": 0.014212799072265626,
+      "learning_rate": 0.0001,
+      "loss": 4.1683,
+      "loss/crossentropy": 1.771731436252594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20304742455482483,
+      "step": 18696
+    },
+    {
+      "epoch": 0.37396,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.014330037434895833,
+      "learning_rate": 0.0001,
+      "loss": 3.9636,
+      "loss/crossentropy": 2.046182096004486,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1976921409368515,
+      "step": 18698
+    },
+    {
+      "epoch": 0.374,
+      "grad_norm": 1.78125,
+      "grad_norm_var": 0.0148345947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0257,
+      "loss/crossentropy": 2.0343292355537415,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19037891179323196,
+      "step": 18700
+    },
+    {
+      "epoch": 0.37404,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.014782460530598958,
+      "learning_rate": 0.0001,
+      "loss": 4.1727,
+      "loss/crossentropy": 2.034367859363556,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20182596892118454,
+      "step": 18702
+    },
+    {
+      "epoch": 0.37408,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.015616861979166667,
+      "learning_rate": 0.0001,
+      "loss": 3.966,
+      "loss/crossentropy": 2.048890709877014,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19922567903995514,
+      "step": 18704
+    },
+    {
+      "epoch": 0.37412,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0163970947265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0864,
+      "loss/crossentropy": 1.8759222626686096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19452380388975143,
+      "step": 18706
+    },
+    {
+      "epoch": 0.37416,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.015681966145833334,
+      "learning_rate": 0.0001,
+      "loss": 3.9964,
+      "loss/crossentropy": 1.693844199180603,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16944261640310287,
+      "step": 18708
+    },
+    {
+      "epoch": 0.3742,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.01727879842122396,
+      "learning_rate": 0.0001,
+      "loss": 4.293,
+      "loss/crossentropy": 2.266264319419861,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2103852555155754,
+      "step": 18710
+    },
+    {
+      "epoch": 0.37424,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.008250935872395834,
+      "learning_rate": 0.0001,
+      "loss": 3.8547,
+      "loss/crossentropy": 2.029367506504059,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18859465420246124,
+      "step": 18712
+    },
+    {
+      "epoch": 0.37428,
+      "grad_norm": 2.640625,
+      "grad_norm_var": 0.039184315999348955,
+      "learning_rate": 0.0001,
+      "loss": 4.1141,
+      "loss/crossentropy": 1.9427701234817505,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1960265040397644,
+      "step": 18714
+    },
+    {
+      "epoch": 0.37432,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.03737360636393229,
+      "learning_rate": 0.0001,
+      "loss": 4.3087,
+      "loss/crossentropy": 2.2656137943267822,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22287357598543167,
+      "step": 18716
+    },
+    {
+      "epoch": 0.37436,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.0365386962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.163,
+      "loss/crossentropy": 2.0447250604629517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20505183935165405,
+      "step": 18718
+    },
+    {
+      "epoch": 0.3744,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.04125137329101562,
+      "learning_rate": 0.0001,
+      "loss": 3.6507,
+      "loss/crossentropy": 1.8176262378692627,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17877614498138428,
+      "step": 18720
+    },
+    {
+      "epoch": 0.37444,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.04017333984375,
+      "learning_rate": 0.0001,
+      "loss": 3.9368,
+      "loss/crossentropy": 1.8362378478050232,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18403278291225433,
+      "step": 18722
+    },
+    {
+      "epoch": 0.37448,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.04104410807291667,
+      "learning_rate": 0.0001,
+      "loss": 4.4322,
+      "loss/crossentropy": 2.43450927734375,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21623602509498596,
+      "step": 18724
+    },
+    {
+      "epoch": 0.37452,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.04133275349934896,
+      "learning_rate": 0.0001,
+      "loss": 4.3905,
+      "loss/crossentropy": 2.227471947669983,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20000503957271576,
+      "step": 18726
+    },
+    {
+      "epoch": 0.37456,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.04317626953125,
+      "learning_rate": 0.0001,
+      "loss": 3.9138,
+      "loss/crossentropy": 1.9392182230949402,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19422952830791473,
+      "step": 18728
+    },
+    {
+      "epoch": 0.3746,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.014469401041666666,
+      "learning_rate": 0.0001,
+      "loss": 4.1284,
+      "loss/crossentropy": 2.0168241262435913,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1774912029504776,
+      "step": 18730
+    },
+    {
+      "epoch": 0.37464,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.012859853108723958,
+      "learning_rate": 0.0001,
+      "loss": 3.8982,
+      "loss/crossentropy": 1.6861125230789185,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16301175951957703,
+      "step": 18732
+    },
+    {
+      "epoch": 0.37468,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.014111328125,
+      "learning_rate": 0.0001,
+      "loss": 4.2366,
+      "loss/crossentropy": 2.2991716861724854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21094633638858795,
+      "step": 18734
+    },
+    {
+      "epoch": 0.37472,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.008115386962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0184,
+      "loss/crossentropy": 1.8746486902236938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19248205423355103,
+      "step": 18736
+    },
+    {
+      "epoch": 0.37476,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.009642537434895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1886,
+      "loss/crossentropy": 2.137068212032318,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1996043175458908,
+      "step": 18738
+    },
+    {
+      "epoch": 0.3748,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.015636952718098958,
+      "learning_rate": 0.0001,
+      "loss": 4.1503,
+      "loss/crossentropy": 1.724601149559021,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23018527776002884,
+      "step": 18740
+    },
+    {
+      "epoch": 0.37484,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.015584309895833334,
+      "learning_rate": 0.0001,
+      "loss": 4.1593,
+      "loss/crossentropy": 2.1407381296157837,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21157334744930267,
+      "step": 18742
+    },
+    {
+      "epoch": 0.37488,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.013090006510416667,
+      "learning_rate": 0.0001,
+      "loss": 4.0371,
+      "loss/crossentropy": 2.2977999448776245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19883693754673004,
+      "step": 18744
+    },
+    {
+      "epoch": 0.37492,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.018155924479166665,
+      "learning_rate": 0.0001,
+      "loss": 3.8912,
+      "loss/crossentropy": 2.096716046333313,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21084215492010117,
+      "step": 18746
+    },
+    {
+      "epoch": 0.37496,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.019254302978515624,
+      "learning_rate": 0.0001,
+      "loss": 3.8508,
+      "loss/crossentropy": 1.7292688488960266,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18280881643295288,
+      "step": 18748
+    },
+    {
+      "epoch": 0.375,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0191314697265625,
+      "learning_rate": 0.0001,
+      "loss": 3.9798,
+      "loss/crossentropy": 1.7562988996505737,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1889612227678299,
+      "step": 18750
+    },
+    {
+      "epoch": 0.37504,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.02008641560872396,
+      "learning_rate": 0.0001,
+      "loss": 4.1498,
+      "loss/crossentropy": 2.1730951070785522,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20177915692329407,
+      "step": 18752
+    },
+    {
+      "epoch": 0.37508,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.018161773681640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0316,
+      "loss/crossentropy": 1.7671055793762207,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17018838971853256,
+      "step": 18754
+    },
+    {
+      "epoch": 0.37512,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010228474934895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1922,
+      "loss/crossentropy": 2.2650365829467773,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22090423107147217,
+      "step": 18756
+    },
+    {
+      "epoch": 0.37516,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.01002197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1985,
+      "loss/crossentropy": 2.252619981765747,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20853855460882187,
+      "step": 18758
+    },
+    {
+      "epoch": 0.3752,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011766560872395833,
+      "learning_rate": 0.0001,
+      "loss": 3.8947,
+      "loss/crossentropy": 2.022092342376709,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19293325394392014,
+      "step": 18760
+    },
+    {
+      "epoch": 0.37524,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010081990559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.0289,
+      "loss/crossentropy": 1.7500890493392944,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18832595646381378,
+      "step": 18762
+    },
+    {
+      "epoch": 0.37528,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.009220123291015625,
+      "learning_rate": 0.0001,
+      "loss": 3.7769,
+      "loss/crossentropy": 1.8201736211776733,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1910964399576187,
+      "step": 18764
+    },
+    {
+      "epoch": 0.37532,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.009989420572916666,
+      "learning_rate": 0.0001,
+      "loss": 4.0508,
+      "loss/crossentropy": 2.1644541025161743,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20023848116397858,
+      "step": 18766
+    },
+    {
+      "epoch": 0.37536,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010436757405598959,
+      "learning_rate": 0.0001,
+      "loss": 4.163,
+      "loss/crossentropy": 2.275505781173706,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23095671832561493,
+      "step": 18768
+    },
+    {
+      "epoch": 0.3754,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010773722330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0557,
+      "loss/crossentropy": 1.976850986480713,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2125934213399887,
+      "step": 18770
+    },
+    {
+      "epoch": 0.37544,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.012035115559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.146,
+      "loss/crossentropy": 2.150836706161499,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19192685186862946,
+      "step": 18772
+    },
+    {
+      "epoch": 0.37548,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.012400054931640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2302,
+      "loss/crossentropy": 2.0790088176727295,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2147749587893486,
+      "step": 18774
+    },
+    {
+      "epoch": 0.37552,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.010697428385416667,
+      "learning_rate": 0.0001,
+      "loss": 3.7918,
+      "loss/crossentropy": 1.8366054892539978,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19662949442863464,
+      "step": 18776
+    },
+    {
+      "epoch": 0.37556,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.008278147379557291,
+      "learning_rate": 0.0001,
+      "loss": 3.9056,
+      "loss/crossentropy": 1.6928801536560059,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18292289972305298,
+      "step": 18778
+    },
+    {
+      "epoch": 0.3756,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.007591756184895834,
+      "learning_rate": 0.0001,
+      "loss": 4.2872,
+      "loss/crossentropy": 2.015128195285797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20940368622541428,
+      "step": 18780
+    },
+    {
+      "epoch": 0.37564,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0061757405598958336,
+      "learning_rate": 0.0001,
+      "loss": 4.3304,
+      "loss/crossentropy": 2.026209592819214,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2227947860956192,
+      "step": 18782
+    },
+    {
+      "epoch": 0.37568,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.004349772135416667,
+      "learning_rate": 0.0001,
+      "loss": 4.0591,
+      "loss/crossentropy": 2.29964280128479,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23132510483264923,
+      "step": 18784
+    },
+    {
+      "epoch": 0.37572,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.005366770426432291,
+      "learning_rate": 0.0001,
+      "loss": 3.9675,
+      "loss/crossentropy": 1.881381332874298,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17967566847801208,
+      "step": 18786
+    },
+    {
+      "epoch": 0.37576,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.004198201497395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1834,
+      "loss/crossentropy": 2.2231001257896423,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21163878589868546,
+      "step": 18788
+    },
+    {
+      "epoch": 0.3758,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.004526519775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0475,
+      "loss/crossentropy": 2.3531078100204468,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21745989471673965,
+      "step": 18790
+    },
+    {
+      "epoch": 0.37584,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.004801432291666667,
+      "learning_rate": 0.0001,
+      "loss": 4.2549,
+      "loss/crossentropy": 2.1178460121154785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23571017384529114,
+      "step": 18792
+    },
+    {
+      "epoch": 0.37588,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.005890909830729167,
+      "learning_rate": 0.0001,
+      "loss": 3.8373,
+      "loss/crossentropy": 2.0659135580062866,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20817245543003082,
+      "step": 18794
+    },
+    {
+      "epoch": 0.37592,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.006318918863932292,
+      "learning_rate": 0.0001,
+      "loss": 4.0103,
+      "loss/crossentropy": 1.579395353794098,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16713083535432816,
+      "step": 18796
+    },
+    {
+      "epoch": 0.37596,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.005940500895182292,
+      "learning_rate": 0.0001,
+      "loss": 4.1272,
+      "loss/crossentropy": 2.102541923522949,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2174636349081993,
+      "step": 18798
+    },
+    {
+      "epoch": 0.376,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0067942301432291664,
+      "learning_rate": 0.0001,
+      "loss": 4.2622,
+      "loss/crossentropy": 1.9342190027236938,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19780533015727997,
+      "step": 18800
+    },
+    {
+      "epoch": 0.37604,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.006121571858723958,
+      "learning_rate": 0.0001,
+      "loss": 4.0659,
+      "loss/crossentropy": 2.140671730041504,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21276966482400894,
+      "step": 18802
+    },
+    {
+      "epoch": 0.37608,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.009590657552083333,
+      "learning_rate": 0.0001,
+      "loss": 4.1929,
+      "loss/crossentropy": 1.9004405736923218,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17683426290750504,
+      "step": 18804
+    },
+    {
+      "epoch": 0.37612,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.010155232747395833,
+      "learning_rate": 0.0001,
+      "loss": 3.7163,
+      "loss/crossentropy": 1.9487649202346802,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19463464617729187,
+      "step": 18806
+    },
+    {
+      "epoch": 0.37616,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.011262766520182292,
+      "learning_rate": 0.0001,
+      "loss": 3.9517,
+      "loss/crossentropy": 2.15252423286438,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20531394332647324,
+      "step": 18808
+    },
+    {
+      "epoch": 0.3762,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.011592356363932292,
+      "learning_rate": 0.0001,
+      "loss": 4.0494,
+      "loss/crossentropy": 2.114433467388153,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20183662325143814,
+      "step": 18810
+    },
+    {
+      "epoch": 0.37624,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.011205037434895834,
+      "learning_rate": 0.0001,
+      "loss": 3.9035,
+      "loss/crossentropy": 1.9229055047035217,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18662934005260468,
+      "step": 18812
+    },
+    {
+      "epoch": 0.37628,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 1.8609944661458333,
+      "learning_rate": 0.0001,
+      "loss": 4.2516,
+      "loss/crossentropy": 1.8242689371109009,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17858239263296127,
+      "step": 18814
+    },
+    {
+      "epoch": 0.37632,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 1.864818318684896,
+      "learning_rate": 0.0001,
+      "loss": 3.8917,
+      "loss/crossentropy": 2.3987890481948853,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2246004194021225,
+      "step": 18816
+    },
+    {
+      "epoch": 0.37636,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 1.8565500895182292,
+      "learning_rate": 0.0001,
+      "loss": 4.3493,
+      "loss/crossentropy": 2.4376614093780518,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24541915208101273,
+      "step": 18818
+    },
+    {
+      "epoch": 0.3764,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 1.865612538655599,
+      "learning_rate": 0.0001,
+      "loss": 4.1344,
+      "loss/crossentropy": 2.2477601766586304,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2178046852350235,
+      "step": 18820
+    },
+    {
+      "epoch": 0.37644,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 1.8636464436848958,
+      "learning_rate": 0.0001,
+      "loss": 3.9685,
+      "loss/crossentropy": 1.9782747626304626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19459142535924911,
+      "step": 18822
+    },
+    {
+      "epoch": 0.37648,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 1.8451738993326823,
+      "learning_rate": 0.0001,
+      "loss": 4.043,
+      "loss/crossentropy": 2.1214100122451782,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2082664743065834,
+      "step": 18824
+    },
+    {
+      "epoch": 0.37652,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 1.8489461263020834,
+      "learning_rate": 0.0001,
+      "loss": 3.8256,
+      "loss/crossentropy": 1.68446546792984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.167055182158947,
+      "step": 18826
+    },
+    {
+      "epoch": 0.37656,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 1.8380022684733073,
+      "learning_rate": 0.0001,
+      "loss": 4.0588,
+      "loss/crossentropy": 1.7720499634742737,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18632127344608307,
+      "step": 18828
+    },
+    {
+      "epoch": 0.3766,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.013492838541666666,
+      "learning_rate": 0.0001,
+      "loss": 4.2288,
+      "loss/crossentropy": 2.283332347869873,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23949767649173737,
+      "step": 18830
+    },
+    {
+      "epoch": 0.37664,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.014644114176432292,
+      "learning_rate": 0.0001,
+      "loss": 3.9134,
+      "loss/crossentropy": 1.9924857020378113,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20181651413440704,
+      "step": 18832
+    },
+    {
+      "epoch": 0.37668,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.011091105143229167,
+      "learning_rate": 0.0001,
+      "loss": 3.8424,
+      "loss/crossentropy": 2.4015761613845825,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2279737964272499,
+      "step": 18834
+    },
+    {
+      "epoch": 0.37672,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.014152018229166667,
+      "learning_rate": 0.0001,
+      "loss": 3.8636,
+      "loss/crossentropy": 2.0717111229896545,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19089852273464203,
+      "step": 18836
+    },
+    {
+      "epoch": 0.37676,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0149566650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2694,
+      "loss/crossentropy": 1.9993728995323181,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19489262998104095,
+      "step": 18838
+    },
+    {
+      "epoch": 0.3768,
+      "grad_norm": 1.8125,
+      "grad_norm_var": 0.015949503580729166,
+      "learning_rate": 0.0001,
+      "loss": 3.7009,
+      "loss/crossentropy": 1.915448248386383,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1923104077577591,
+      "step": 18840
+    },
+    {
+      "epoch": 0.37684,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.016511027018229166,
+      "learning_rate": 0.0001,
+      "loss": 4.0424,
+      "loss/crossentropy": 2.292284607887268,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22159253805875778,
+      "step": 18842
+    },
+    {
+      "epoch": 0.37688,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.012784830729166667,
+      "learning_rate": 0.0001,
+      "loss": 3.9904,
+      "loss/crossentropy": 1.9938938617706299,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19937817752361298,
+      "step": 18844
+    },
+    {
+      "epoch": 0.37692,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.016454060872395832,
+      "learning_rate": 0.0001,
+      "loss": 4.1914,
+      "loss/crossentropy": 2.045258641242981,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2652823179960251,
+      "step": 18846
+    },
+    {
+      "epoch": 0.37696,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.016707356770833334,
+      "learning_rate": 0.0001,
+      "loss": 4.0391,
+      "loss/crossentropy": 2.1330259442329407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19155749678611755,
+      "step": 18848
+    },
+    {
+      "epoch": 0.377,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.017101796468098958,
+      "learning_rate": 0.0001,
+      "loss": 4.1422,
+      "loss/crossentropy": 2.2135089635849,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20062117278575897,
+      "step": 18850
+    },
+    {
+      "epoch": 0.37704,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.013826243082682292,
+      "learning_rate": 0.0001,
+      "loss": 4.1188,
+      "loss/crossentropy": 2.073060691356659,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19443543255329132,
+      "step": 18852
+    },
+    {
+      "epoch": 0.37708,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.014170074462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.089,
+      "loss/crossentropy": 1.930641233921051,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19604724645614624,
+      "step": 18854
+    },
+    {
+      "epoch": 0.37712,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.011252593994140626,
+      "learning_rate": 0.0001,
+      "loss": 4.0238,
+      "loss/crossentropy": 2.0813130140304565,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21019181609153748,
+      "step": 18856
+    },
+    {
+      "epoch": 0.37716,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.010080718994140625,
+      "learning_rate": 0.0001,
+      "loss": 3.9855,
+      "loss/crossentropy": 1.8528355956077576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19945576786994934,
+      "step": 18858
+    },
+    {
+      "epoch": 0.3772,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.011896769205729166,
+      "learning_rate": 0.0001,
+      "loss": 3.9655,
+      "loss/crossentropy": 2.32794725894928,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22688604146242142,
+      "step": 18860
+    },
+    {
+      "epoch": 0.37724,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.007264963785807292,
+      "learning_rate": 0.0001,
+      "loss": 4.25,
+      "loss/crossentropy": 2.061101734638214,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1961553767323494,
+      "step": 18862
+    },
+    {
+      "epoch": 0.37728,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.00604248046875,
+      "learning_rate": 0.0001,
+      "loss": 3.7839,
+      "loss/crossentropy": 1.9127929210662842,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20082338899374008,
+      "step": 18864
+    },
+    {
+      "epoch": 0.37732,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.14662272135416668,
+      "learning_rate": 0.0001,
+      "loss": 4.289,
+      "loss/crossentropy": 2.001932919025421,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1881283074617386,
+      "step": 18866
+    },
+    {
+      "epoch": 0.37736,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.14527180989583333,
+      "learning_rate": 0.0001,
+      "loss": 4.0851,
+      "loss/crossentropy": 1.693075716495514,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18698371946811676,
+      "step": 18868
+    },
+    {
+      "epoch": 0.3774,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.1481402079264323,
+      "learning_rate": 0.0001,
+      "loss": 4.0484,
+      "loss/crossentropy": 1.879252314567566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18412816524505615,
+      "step": 18870
+    },
+    {
+      "epoch": 0.37744,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.1550066630045573,
+      "learning_rate": 0.0001,
+      "loss": 4.3488,
+      "loss/crossentropy": 2.1819299459457397,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20243355631828308,
+      "step": 18872
+    },
+    {
+      "epoch": 0.37748,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.1576568603515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0889,
+      "loss/crossentropy": 2.2487794160842896,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20594948530197144,
+      "step": 18874
+    },
+    {
+      "epoch": 0.37752,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.15466079711914063,
+      "learning_rate": 0.0001,
+      "loss": 4.303,
+      "loss/crossentropy": 1.990889549255371,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19230055809020996,
+      "step": 18876
+    },
+    {
+      "epoch": 0.37756,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.1582763671875,
+      "learning_rate": 0.0001,
+      "loss": 4.0435,
+      "loss/crossentropy": 2.091560959815979,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19504332542419434,
+      "step": 18878
+    },
+    {
+      "epoch": 0.3776,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.1599273681640625,
+      "learning_rate": 0.0001,
+      "loss": 4.0776,
+      "loss/crossentropy": 1.9213955998420715,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1994621828198433,
+      "step": 18880
+    },
+    {
+      "epoch": 0.37764,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.02360814412434896,
+      "learning_rate": 0.0001,
+      "loss": 4.0103,
+      "loss/crossentropy": 2.1728278398513794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21039214730262756,
+      "step": 18882
+    },
+    {
+      "epoch": 0.37768,
+      "grad_norm": 1.78125,
+      "grad_norm_var": 0.026341756184895832,
+      "learning_rate": 0.0001,
+      "loss": 4.052,
+      "loss/crossentropy": 1.889222264289856,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17232363671064377,
+      "step": 18884
+    },
+    {
+      "epoch": 0.37772,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.026228841145833334,
+      "learning_rate": 0.0001,
+      "loss": 4.118,
+      "loss/crossentropy": 2.264981746673584,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.213841512799263,
+      "step": 18886
+    },
+    {
+      "epoch": 0.37776,
+      "grad_norm": 1.78125,
+      "grad_norm_var": 0.013598378499348958,
+      "learning_rate": 0.0001,
+      "loss": 3.8843,
+      "loss/crossentropy": 1.8615361452102661,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1745932251214981,
+      "step": 18888
+    },
+    {
+      "epoch": 0.3778,
+      "grad_norm": 1.765625,
+      "grad_norm_var": 0.009266916910807292,
+      "learning_rate": 0.0001,
+      "loss": 3.5687,
+      "loss/crossentropy": 2.1459723711013794,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21431762725114822,
+      "step": 18890
+    },
+    {
+      "epoch": 0.37784,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010479482014973958,
+      "learning_rate": 0.0001,
+      "loss": 4.2125,
+      "loss/crossentropy": 1.8859283328056335,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20627497136592865,
+      "step": 18892
+    },
+    {
+      "epoch": 0.37788,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.012211100260416666,
+      "learning_rate": 0.0001,
+      "loss": 3.9978,
+      "loss/crossentropy": 1.824287474155426,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17877592146396637,
+      "step": 18894
+    },
+    {
+      "epoch": 0.37792,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.012414296468098959,
+      "learning_rate": 0.0001,
+      "loss": 3.9857,
+      "loss/crossentropy": 2.486180305480957,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2033834606409073,
+      "step": 18896
+    },
+    {
+      "epoch": 0.37796,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.012981923421223958,
+      "learning_rate": 0.0001,
+      "loss": 4.2339,
+      "loss/crossentropy": 2.0677965879440308,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21111362427473068,
+      "step": 18898
+    },
+    {
+      "epoch": 0.378,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.011107381184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2559,
+      "loss/crossentropy": 1.941792368888855,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18762809038162231,
+      "step": 18900
+    },
+    {
+      "epoch": 0.37804,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.012504069010416667,
+      "learning_rate": 0.0001,
+      "loss": 3.8028,
+      "loss/crossentropy": 1.907107174396515,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2050766795873642,
+      "step": 18902
+    },
+    {
+      "epoch": 0.37808,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.009242502848307292,
+      "learning_rate": 0.0001,
+      "loss": 3.9742,
+      "loss/crossentropy": 1.5920222997665405,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17366845160722733,
+      "step": 18904
+    },
+    {
+      "epoch": 0.37812,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0066650390625,
+      "learning_rate": 0.0001,
+      "loss": 3.9581,
+      "loss/crossentropy": 2.054854154586792,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20102889090776443,
+      "step": 18906
+    },
+    {
+      "epoch": 0.37816,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0067789713541666664,
+      "learning_rate": 0.0001,
+      "loss": 4.2582,
+      "loss/crossentropy": 2.42835795879364,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2154095396399498,
+      "step": 18908
+    },
+    {
+      "epoch": 0.3782,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.007169596354166667,
+      "learning_rate": 0.0001,
+      "loss": 4.1466,
+      "loss/crossentropy": 2.1060370206832886,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20842333883047104,
+      "step": 18910
+    },
+    {
+      "epoch": 0.37824,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.008194986979166667,
+      "learning_rate": 0.0001,
+      "loss": 4.1587,
+      "loss/crossentropy": 2.0559674501419067,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2195875570178032,
+      "step": 18912
+    },
+    {
+      "epoch": 0.37828,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008888498942057291,
+      "learning_rate": 0.0001,
+      "loss": 4.2786,
+      "loss/crossentropy": 2.240954041481018,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20908969640731812,
+      "step": 18914
+    },
+    {
+      "epoch": 0.37832,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.008747355143229166,
+      "learning_rate": 0.0001,
+      "loss": 4.0729,
+      "loss/crossentropy": 2.0393139123916626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19933026283979416,
+      "step": 18916
+    },
+    {
+      "epoch": 0.37836,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.007682291666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.0969,
+      "loss/crossentropy": 2.0294516682624817,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20086780190467834,
+      "step": 18918
+    },
+    {
+      "epoch": 0.3784,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.007425944010416667,
+      "learning_rate": 0.0001,
+      "loss": 3.9138,
+      "loss/crossentropy": 1.8509765267372131,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18834587186574936,
+      "step": 18920
+    },
+    {
+      "epoch": 0.37844,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.0068267822265625,
+      "learning_rate": 0.0001,
+      "loss": 3.8669,
+      "loss/crossentropy": 1.8817378878593445,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17679119110107422,
+      "step": 18922
+    },
+    {
+      "epoch": 0.37848,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.005975087483723958,
+      "learning_rate": 0.0001,
+      "loss": 3.9276,
+      "loss/crossentropy": 1.766166627407074,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16948368400335312,
+      "step": 18924
+    },
+    {
+      "epoch": 0.37852,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.006786855061848959,
+      "learning_rate": 0.0001,
+      "loss": 3.8948,
+      "loss/crossentropy": 2.251446485519409,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22272750735282898,
+      "step": 18926
+    },
+    {
+      "epoch": 0.37856,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.0045125325520833336,
+      "learning_rate": 0.0001,
+      "loss": 4.1119,
+      "loss/crossentropy": 2.0006097555160522,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21003134548664093,
+      "step": 18928
+    },
+    {
+      "epoch": 0.3786,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0036936442057291666,
+      "learning_rate": 0.0001,
+      "loss": 3.7731,
+      "loss/crossentropy": 2.2906641960144043,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23650038242340088,
+      "step": 18930
+    },
+    {
+      "epoch": 0.37864,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0039866129557291664,
+      "learning_rate": 0.0001,
+      "loss": 4.1848,
+      "loss/crossentropy": 2.185898005962372,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21387950330972672,
+      "step": 18932
+    },
+    {
+      "epoch": 0.37868,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0028928120930989585,
+      "learning_rate": 0.0001,
+      "loss": 4.1808,
+      "loss/crossentropy": 2.2457324266433716,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21054718643426895,
+      "step": 18934
+    },
+    {
+      "epoch": 0.37872,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0032793680826822915,
+      "learning_rate": 0.0001,
+      "loss": 4.2711,
+      "loss/crossentropy": 2.1835416555404663,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23258651047945023,
+      "step": 18936
+    },
+    {
+      "epoch": 0.37876,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0029436747233072915,
+      "learning_rate": 0.0001,
+      "loss": 4.1555,
+      "loss/crossentropy": 2.1908507347106934,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20261384546756744,
+      "step": 18938
+    },
+    {
+      "epoch": 0.3788,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0035336812337239585,
+      "learning_rate": 0.0001,
+      "loss": 4.1359,
+      "loss/crossentropy": 2.276697278022766,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20843130350112915,
+      "step": 18940
+    },
+    {
+      "epoch": 0.37884,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0023251851399739582,
+      "learning_rate": 0.0001,
+      "loss": 4.1309,
+      "loss/crossentropy": 2.16153222322464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19693513214588165,
+      "step": 18942
+    },
+    {
+      "epoch": 0.37888,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0023590087890625,
+      "learning_rate": 0.0001,
+      "loss": 3.9822,
+      "loss/crossentropy": 2.293270707130432,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21085364371538162,
+      "step": 18944
+    },
+    {
+      "epoch": 0.37892,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.0025042215983072918,
+      "learning_rate": 0.0001,
+      "loss": 4.1291,
+      "loss/crossentropy": 2.0382518768310547,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21567383408546448,
+      "step": 18946
+    },
+    {
+      "epoch": 0.37896,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.00894775390625,
+      "learning_rate": 0.0001,
+      "loss": 3.9993,
+      "loss/crossentropy": 2.038296341896057,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19963379949331284,
+      "step": 18948
+    },
+    {
+      "epoch": 0.379,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.00994873046875,
+      "learning_rate": 0.0001,
+      "loss": 4.2363,
+      "loss/crossentropy": 2.1765074729919434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19551265239715576,
+      "step": 18950
+    },
+    {
+      "epoch": 0.37904,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010591379801432292,
+      "learning_rate": 0.0001,
+      "loss": 3.8641,
+      "loss/crossentropy": 1.870033621788025,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17858022451400757,
+      "step": 18952
+    },
+    {
+      "epoch": 0.37908,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.011083730061848958,
+      "learning_rate": 0.0001,
+      "loss": 4.1877,
+      "loss/crossentropy": 1.8716632723808289,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20534101128578186,
+      "step": 18954
+    },
+    {
+      "epoch": 0.37912,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.0111236572265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1662,
+      "loss/crossentropy": 1.925970435142517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19596435129642487,
+      "step": 18956
+    },
+    {
+      "epoch": 0.37916,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.011800130208333334,
+      "learning_rate": 0.0001,
+      "loss": 4.126,
+      "loss/crossentropy": 2.2280589938163757,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21037188172340393,
+      "step": 18958
+    },
+    {
+      "epoch": 0.3792,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0150299072265625,
+      "learning_rate": 0.0001,
+      "loss": 3.7683,
+      "loss/crossentropy": 1.8350458145141602,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1821446716785431,
+      "step": 18960
+    },
+    {
+      "epoch": 0.37924,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.015018463134765625,
+      "learning_rate": 0.0001,
+      "loss": 4.0376,
+      "loss/crossentropy": 2.0393940210342407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18378175050020218,
+      "step": 18962
+    },
+    {
+      "epoch": 0.37928,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.00897216796875,
+      "learning_rate": 0.0001,
+      "loss": 4.0018,
+      "loss/crossentropy": 1.7470228672027588,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19222351908683777,
+      "step": 18964
+    },
+    {
+      "epoch": 0.37932,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.011742146809895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1791,
+      "loss/crossentropy": 2.3943980932235718,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23214909434318542,
+      "step": 18966
+    },
+    {
+      "epoch": 0.37936,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.012239329020182292,
+      "learning_rate": 0.0001,
+      "loss": 3.7877,
+      "loss/crossentropy": 1.8153263330459595,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19808300584554672,
+      "step": 18968
+    },
+    {
+      "epoch": 0.3794,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.011171213785807292,
+      "learning_rate": 0.0001,
+      "loss": 4.0122,
+      "loss/crossentropy": 2.0037755370140076,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20173701643943787,
+      "step": 18970
+    },
+    {
+      "epoch": 0.37944,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.012168121337890626,
+      "learning_rate": 0.0001,
+      "loss": 4.1069,
+      "loss/crossentropy": 2.144823908805847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21421342343091965,
+      "step": 18972
+    },
+    {
+      "epoch": 0.37948,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.011565907796223959,
+      "learning_rate": 0.0001,
+      "loss": 4.2788,
+      "loss/crossentropy": 2.2151081562042236,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20057211816310883,
+      "step": 18974
+    },
+    {
+      "epoch": 0.37952,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008976236979166666,
+      "learning_rate": 0.0001,
+      "loss": 4.5061,
+      "loss/crossentropy": 2.341770827770233,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21535535156726837,
+      "step": 18976
+    },
+    {
+      "epoch": 0.37956,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.009520467122395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1399,
+      "loss/crossentropy": 1.8952747583389282,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17530933022499084,
+      "step": 18978
+    },
+    {
+      "epoch": 0.3796,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008885701497395834,
+      "learning_rate": 0.0001,
+      "loss": 4.141,
+      "loss/crossentropy": 2.2774226665496826,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20682721585035324,
+      "step": 18980
+    },
+    {
+      "epoch": 0.37964,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.008508046468098959,
+      "learning_rate": 0.0001,
+      "loss": 3.982,
+      "loss/crossentropy": 1.9596800208091736,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18774595111608505,
+      "step": 18982
+    },
+    {
+      "epoch": 0.37968,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.009309895833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.008,
+      "loss/crossentropy": 2.1485220193862915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19221040606498718,
+      "step": 18984
+    },
+    {
+      "epoch": 0.37972,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.009039052327473958,
+      "learning_rate": 0.0001,
+      "loss": 4.142,
+      "loss/crossentropy": 2.2140207290649414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2012096345424652,
+      "step": 18986
+    },
+    {
+      "epoch": 0.37976,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.008063761393229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1088,
+      "loss/crossentropy": 1.9045360684394836,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19290773570537567,
+      "step": 18988
+    },
+    {
+      "epoch": 0.3798,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008072662353515624,
+      "learning_rate": 0.0001,
+      "loss": 4.0349,
+      "loss/crossentropy": 1.8351407051086426,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18892163783311844,
+      "step": 18990
+    },
+    {
+      "epoch": 0.37984,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.007865397135416667,
+      "learning_rate": 0.0001,
+      "loss": 3.9136,
+      "loss/crossentropy": 2.021119713783264,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19214972108602524,
+      "step": 18992
+    },
+    {
+      "epoch": 0.37988,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.006394195556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2961,
+      "loss/crossentropy": 2.0591527223587036,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20295391231775284,
+      "step": 18994
+    },
+    {
+      "epoch": 0.37992,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.006566365559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1135,
+      "loss/crossentropy": 1.9794987440109253,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20964892208576202,
+      "step": 18996
+    },
+    {
+      "epoch": 0.37996,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0042111714680989586,
+      "learning_rate": 0.0001,
+      "loss": 4.0699,
+      "loss/crossentropy": 2.116178512573242,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1962205320596695,
+      "step": 18998
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.003714752197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0683,
+      "loss/crossentropy": 2.2458595037460327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1981743946671486,
+      "step": 19000
+    },
+    {
+      "epoch": 0.38004,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0038266499837239583,
+      "learning_rate": 0.0001,
+      "loss": 4.2085,
+      "loss/crossentropy": 2.0612659454345703,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20526322722434998,
+      "step": 19002
+    },
+    {
+      "epoch": 0.38008,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.00347900390625,
+      "learning_rate": 0.0001,
+      "loss": 4.191,
+      "loss/crossentropy": 2.0810243487358093,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.222077377140522,
+      "step": 19004
+    },
+    {
+      "epoch": 0.38012,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.003940582275390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0957,
+      "loss/crossentropy": 2.2651617527008057,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2106148600578308,
+      "step": 19006
+    },
+    {
+      "epoch": 0.38016,
+      "grad_norm": 1.8125,
+      "grad_norm_var": 0.0056793212890625,
+      "learning_rate": 0.0001,
+      "loss": 3.9671,
+      "loss/crossentropy": 1.9701108932495117,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18778277933597565,
+      "step": 19008
+    },
+    {
+      "epoch": 0.3802,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.007024892171223958,
+      "learning_rate": 0.0001,
+      "loss": 4.1265,
+      "loss/crossentropy": 1.9822896122932434,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25280050933361053,
+      "step": 19010
+    },
+    {
+      "epoch": 0.38024,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0063435872395833336,
+      "learning_rate": 0.0001,
+      "loss": 3.9577,
+      "loss/crossentropy": 2.020545542240143,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19008295983076096,
+      "step": 19012
+    },
+    {
+      "epoch": 0.38028,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.007336171468098959,
+      "learning_rate": 0.0001,
+      "loss": 4.2044,
+      "loss/crossentropy": 2.0922394394874573,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1972956582903862,
+      "step": 19014
+    },
+    {
+      "epoch": 0.38032,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.007067616780598958,
+      "learning_rate": 0.0001,
+      "loss": 3.9726,
+      "loss/crossentropy": 2.028052031993866,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18549586832523346,
+      "step": 19016
+    },
+    {
+      "epoch": 0.38036,
+      "grad_norm": 1.6796875,
+      "grad_norm_var": 0.011395009358723958,
+      "learning_rate": 0.0001,
+      "loss": 4.0118,
+      "loss/crossentropy": 1.9518752098083496,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17747169733047485,
+      "step": 19018
+    },
+    {
+      "epoch": 0.3804,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.013492838541666666,
+      "learning_rate": 0.0001,
+      "loss": 4.1404,
+      "loss/crossentropy": 1.8896766901016235,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20509368181228638,
+      "step": 19020
+    },
+    {
+      "epoch": 0.38044,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.01407470703125,
+      "learning_rate": 0.0001,
+      "loss": 4.1159,
+      "loss/crossentropy": 2.10469388961792,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2058955579996109,
+      "step": 19022
+    },
+    {
+      "epoch": 0.38048,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.012717437744140626,
+      "learning_rate": 0.0001,
+      "loss": 3.8601,
+      "loss/crossentropy": 1.7341394424438477,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19358114898204803,
+      "step": 19024
+    },
+    {
+      "epoch": 0.38052,
+      "grad_norm": 1.7734375,
+      "grad_norm_var": 0.013634999593098959,
+      "learning_rate": 0.0001,
+      "loss": 3.8528,
+      "loss/crossentropy": 2.186724543571472,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1981809437274933,
+      "step": 19026
+    },
+    {
+      "epoch": 0.38056,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.014070638020833333,
+      "learning_rate": 0.0001,
+      "loss": 4.2421,
+      "loss/crossentropy": 2.228869318962097,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20712755620479584,
+      "step": 19028
+    },
+    {
+      "epoch": 0.3806,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0140869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.126,
+      "loss/crossentropy": 2.2783373594284058,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20916558057069778,
+      "step": 19030
+    },
+    {
+      "epoch": 0.38064,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.013270823160807292,
+      "learning_rate": 0.0001,
+      "loss": 4.2744,
+      "loss/crossentropy": 2.2671823501586914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19389723241329193,
+      "step": 19032
+    },
+    {
+      "epoch": 0.38068,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.008540852864583334,
+      "learning_rate": 0.0001,
+      "loss": 4.0267,
+      "loss/crossentropy": 1.948801040649414,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18766260892152786,
+      "step": 19034
+    },
+    {
+      "epoch": 0.38072,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.007818349202473958,
+      "learning_rate": 0.0001,
+      "loss": 4.0856,
+      "loss/crossentropy": 1.895260751247406,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1848379746079445,
+      "step": 19036
+    },
+    {
+      "epoch": 0.38076,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.008504231770833334,
+      "learning_rate": 0.0001,
+      "loss": 3.8189,
+      "loss/crossentropy": 1.8295226097106934,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1816689372062683,
+      "step": 19038
+    },
+    {
+      "epoch": 0.3808,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.007830556233723958,
+      "learning_rate": 0.0001,
+      "loss": 4.0363,
+      "loss/crossentropy": 1.6487661004066467,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.178475059568882,
+      "step": 19040
+    },
+    {
+      "epoch": 0.38084,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.0058095296223958336,
+      "learning_rate": 0.0001,
+      "loss": 3.828,
+      "loss/crossentropy": 1.9264054894447327,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2009180784225464,
+      "step": 19042
+    },
+    {
+      "epoch": 0.38088,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.007989247639973959,
+      "learning_rate": 0.0001,
+      "loss": 4.2636,
+      "loss/crossentropy": 2.3018531799316406,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2039026990532875,
+      "step": 19044
+    },
+    {
+      "epoch": 0.38092,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.009093983968098959,
+      "learning_rate": 0.0001,
+      "loss": 4.0201,
+      "loss/crossentropy": 2.135006010532379,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.172908216714859,
+      "step": 19046
+    },
+    {
+      "epoch": 0.38096,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.009476725260416667,
+      "learning_rate": 0.0001,
+      "loss": 4.2064,
+      "loss/crossentropy": 2.1111900806427,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19583145529031754,
+      "step": 19048
+    },
+    {
+      "epoch": 0.381,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0096099853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1165,
+      "loss/crossentropy": 2.208159327507019,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21107421815395355,
+      "step": 19050
+    },
+    {
+      "epoch": 0.38104,
+      "grad_norm": 2.203125,
+      "grad_norm_var": 0.012894694010416667,
+      "learning_rate": 0.0001,
+      "loss": 4.1722,
+      "loss/crossentropy": 2.144998788833618,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2108323574066162,
+      "step": 19052
+    },
+    {
+      "epoch": 0.38108,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.0116455078125,
+      "learning_rate": 0.0001,
+      "loss": 4.0938,
+      "loss/crossentropy": 2.0046940445899963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20454590767621994,
+      "step": 19054
+    },
+    {
+      "epoch": 0.38112,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.010992177327473958,
+      "learning_rate": 0.0001,
+      "loss": 4.2219,
+      "loss/crossentropy": 2.042281448841095,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21530431509017944,
+      "step": 19056
+    },
+    {
+      "epoch": 0.38116,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0102691650390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2701,
+      "loss/crossentropy": 2.0579177141189575,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20769091695547104,
+      "step": 19058
+    },
+    {
+      "epoch": 0.3812,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.012719472249348959,
+      "learning_rate": 0.0001,
+      "loss": 4.0561,
+      "loss/crossentropy": 1.8340198993682861,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1927536353468895,
+      "step": 19060
+    },
+    {
+      "epoch": 0.38124,
+      "grad_norm": 2.375,
+      "grad_norm_var": 0.016584269205729165,
+      "learning_rate": 0.0001,
+      "loss": 4.0412,
+      "loss/crossentropy": 2.146699070930481,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18840950727462769,
+      "step": 19062
+    },
+    {
+      "epoch": 0.38128,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.017600250244140626,
+      "learning_rate": 0.0001,
+      "loss": 3.7448,
+      "loss/crossentropy": 2.0206886529922485,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20354034006595612,
+      "step": 19064
+    },
+    {
+      "epoch": 0.38132,
+      "grad_norm": 1.78125,
+      "grad_norm_var": 0.023372141520182292,
+      "learning_rate": 0.0001,
+      "loss": 4.0047,
+      "loss/crossentropy": 1.9568690061569214,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.183781698346138,
+      "step": 19066
+    },
+    {
+      "epoch": 0.38136,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.022855631510416665,
+      "learning_rate": 0.0001,
+      "loss": 4.0898,
+      "loss/crossentropy": 2.092573404312134,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20443417131900787,
+      "step": 19068
+    },
+    {
+      "epoch": 0.3814,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.025770823160807293,
+      "learning_rate": 0.0001,
+      "loss": 4.2253,
+      "loss/crossentropy": 2.17366099357605,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2064819559454918,
+      "step": 19070
+    },
+    {
+      "epoch": 0.38144,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0265777587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0398,
+      "loss/crossentropy": 1.7935467958450317,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19256697595119476,
+      "step": 19072
+    },
+    {
+      "epoch": 0.38148,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.026554107666015625,
+      "learning_rate": 0.0001,
+      "loss": 3.9846,
+      "loss/crossentropy": 1.995141625404358,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19474036246538162,
+      "step": 19074
+    },
+    {
+      "epoch": 0.38152,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.023221588134765624,
+      "learning_rate": 0.0001,
+      "loss": 3.9058,
+      "loss/crossentropy": 2.140002131462097,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19610393047332764,
+      "step": 19076
+    },
+    {
+      "epoch": 0.38156,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.013288370768229167,
+      "learning_rate": 0.0001,
+      "loss": 4.1176,
+      "loss/crossentropy": 2.1740564107894897,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20218566060066223,
+      "step": 19078
+    },
+    {
+      "epoch": 0.3816,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.015925089518229168,
+      "learning_rate": 0.0001,
+      "loss": 4.3043,
+      "loss/crossentropy": 2.065530776977539,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21683599054813385,
+      "step": 19080
+    },
+    {
+      "epoch": 0.38164,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.0127349853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2859,
+      "loss/crossentropy": 1.844248354434967,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19479060918092728,
+      "step": 19082
+    },
+    {
+      "epoch": 0.38168,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.011136627197265625,
+      "learning_rate": 0.0001,
+      "loss": 3.886,
+      "loss/crossentropy": 1.8596556186676025,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2059667930006981,
+      "step": 19084
+    },
+    {
+      "epoch": 0.38172,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.014731597900390626,
+      "learning_rate": 0.0001,
+      "loss": 4.0982,
+      "loss/crossentropy": 2.08061683177948,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20623548328876495,
+      "step": 19086
+    },
+    {
+      "epoch": 0.38176,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.014615631103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0013,
+      "loss/crossentropy": 1.888766884803772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18000122904777527,
+      "step": 19088
+    },
+    {
+      "epoch": 0.3818,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.017032877604166666,
+      "learning_rate": 0.0001,
+      "loss": 3.8784,
+      "loss/crossentropy": 2.021374225616455,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21468255668878555,
+      "step": 19090
+    },
+    {
+      "epoch": 0.38184,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.016805013020833332,
+      "learning_rate": 0.0001,
+      "loss": 3.8327,
+      "loss/crossentropy": 2.083697557449341,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21188172698020935,
+      "step": 19092
+    },
+    {
+      "epoch": 0.38188,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.02019017537434896,
+      "learning_rate": 0.0001,
+      "loss": 3.9617,
+      "loss/crossentropy": 2.1484656929969788,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21978207677602768,
+      "step": 19094
+    },
+    {
+      "epoch": 0.38192,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.0218414306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1728,
+      "loss/crossentropy": 2.321221709251404,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22688252478837967,
+      "step": 19096
+    },
+    {
+      "epoch": 0.38196,
+      "grad_norm": 2.703125,
+      "grad_norm_var": 0.0518218994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.157,
+      "loss/crossentropy": 2.0623167753219604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2148035168647766,
+      "step": 19098
+    },
+    {
+      "epoch": 0.382,
+      "grad_norm": 2.234375,
+      "grad_norm_var": 0.05501708984375,
+      "learning_rate": 0.0001,
+      "loss": 4.0255,
+      "loss/crossentropy": 1.8845015168190002,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1880025863647461,
+      "step": 19100
+    },
+    {
+      "epoch": 0.38204,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.05045750935872396,
+      "learning_rate": 0.0001,
+      "loss": 3.8361,
+      "loss/crossentropy": 1.927463173866272,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17446774244308472,
+      "step": 19102
+    },
+    {
+      "epoch": 0.38208,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0536529541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2909,
+      "loss/crossentropy": 2.6780699491500854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22951193153858185,
+      "step": 19104
+    },
+    {
+      "epoch": 0.38212,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.05358250935872396,
+      "learning_rate": 0.0001,
+      "loss": 3.9188,
+      "loss/crossentropy": 2.2530760765075684,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2141779437661171,
+      "step": 19106
+    },
+    {
+      "epoch": 0.38216,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.07016499837239583,
+      "learning_rate": 0.0001,
+      "loss": 4.1647,
+      "loss/crossentropy": 2.0292217135429382,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20607471466064453,
+      "step": 19108
+    },
+    {
+      "epoch": 0.3822,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.07394790649414062,
+      "learning_rate": 0.0001,
+      "loss": 3.8144,
+      "loss/crossentropy": 1.8341345191001892,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18437029421329498,
+      "step": 19110
+    },
+    {
+      "epoch": 0.38224,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.06982421875,
+      "learning_rate": 0.0001,
+      "loss": 4.1782,
+      "loss/crossentropy": 1.9262341260910034,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1894589066505432,
+      "step": 19112
+    },
+    {
+      "epoch": 0.38228,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.04212824503580729,
+      "learning_rate": 0.0001,
+      "loss": 4.2473,
+      "loss/crossentropy": 1.9369717240333557,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17737293988466263,
+      "step": 19114
+    },
+    {
+      "epoch": 0.38232,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.03845621744791667,
+      "learning_rate": 0.0001,
+      "loss": 4.0904,
+      "loss/crossentropy": 1.7547513842582703,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1895698457956314,
+      "step": 19116
+    },
+    {
+      "epoch": 0.38236,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.042789459228515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2814,
+      "loss/crossentropy": 2.2367511987686157,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22208665311336517,
+      "step": 19118
+    },
+    {
+      "epoch": 0.3824,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.040169016520182295,
+      "learning_rate": 0.0001,
+      "loss": 4.1671,
+      "loss/crossentropy": 2.248104691505432,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2132478505373001,
+      "step": 19120
+    },
+    {
+      "epoch": 0.38244,
+      "grad_norm": 1.7890625,
+      "grad_norm_var": 0.040415191650390626,
+      "learning_rate": 0.0001,
+      "loss": 4.0371,
+      "loss/crossentropy": 2.237170696258545,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21920116990804672,
+      "step": 19122
+    },
+    {
+      "epoch": 0.38248,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.018027496337890626,
+      "learning_rate": 0.0001,
+      "loss": 4.1011,
+      "loss/crossentropy": 1.82357919216156,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1723759025335312,
+      "step": 19124
+    },
+    {
+      "epoch": 0.38252,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.018668619791666667,
+      "learning_rate": 0.0001,
+      "loss": 3.8541,
+      "loss/crossentropy": 1.9371765851974487,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19782865047454834,
+      "step": 19126
+    },
+    {
+      "epoch": 0.38256,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.0208404541015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1305,
+      "loss/crossentropy": 1.8295999765396118,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1756351813673973,
+      "step": 19128
+    },
+    {
+      "epoch": 0.3826,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.044077301025390626,
+      "learning_rate": 0.0001,
+      "loss": 4.0657,
+      "loss/crossentropy": 2.0991050601005554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17175115644931793,
+      "step": 19130
+    },
+    {
+      "epoch": 0.38264,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.04390640258789062,
+      "learning_rate": 0.0001,
+      "loss": 4.1359,
+      "loss/crossentropy": 2.2707191705703735,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22085107117891312,
+      "step": 19132
+    },
+    {
+      "epoch": 0.38268,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.03630345662434896,
+      "learning_rate": 0.0001,
+      "loss": 4.352,
+      "loss/crossentropy": 2.237929582595825,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2524951994419098,
+      "step": 19134
+    },
+    {
+      "epoch": 0.38272,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.03629735310872396,
+      "learning_rate": 0.0001,
+      "loss": 4.0293,
+      "loss/crossentropy": 1.9790211915969849,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18156784772872925,
+      "step": 19136
+    },
+    {
+      "epoch": 0.38276,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.034795888264973956,
+      "learning_rate": 0.0001,
+      "loss": 3.9467,
+      "loss/crossentropy": 1.9819305539131165,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18941760063171387,
+      "step": 19138
+    },
+    {
+      "epoch": 0.3828,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.032022857666015626,
+      "learning_rate": 0.0001,
+      "loss": 3.9699,
+      "loss/crossentropy": 1.9541404843330383,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.195252887904644,
+      "step": 19140
+    },
+    {
+      "epoch": 0.38284,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.028824869791666666,
+      "learning_rate": 0.0001,
+      "loss": 3.9563,
+      "loss/crossentropy": 2.1265366673469543,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20113253593444824,
+      "step": 19142
+    },
+    {
+      "epoch": 0.38288,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.0292724609375,
+      "learning_rate": 0.0001,
+      "loss": 4.0248,
+      "loss/crossentropy": 1.767483413219452,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18520043045282364,
+      "step": 19144
+    },
+    {
+      "epoch": 0.38292,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.005352528889973959,
+      "learning_rate": 0.0001,
+      "loss": 4.2516,
+      "loss/crossentropy": 2.1124974489212036,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20909518748521805,
+      "step": 19146
+    },
+    {
+      "epoch": 0.38296,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.005181630452473958,
+      "learning_rate": 0.0001,
+      "loss": 4.3104,
+      "loss/crossentropy": 2.2959529161453247,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2183123677968979,
+      "step": 19148
+    },
+    {
+      "epoch": 0.383,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.005078125,
+      "learning_rate": 0.0001,
+      "loss": 4.3217,
+      "loss/crossentropy": 2.3522391319274902,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22062842547893524,
+      "step": 19150
+    },
+    {
+      "epoch": 0.38304,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.0048095703125,
+      "learning_rate": 0.0001,
+      "loss": 3.7375,
+      "loss/crossentropy": 1.6673399806022644,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16702746599912643,
+      "step": 19152
+    },
+    {
+      "epoch": 0.38308,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.004073079427083333,
+      "learning_rate": 0.0001,
+      "loss": 4.2058,
+      "loss/crossentropy": 2.021436333656311,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19833282381296158,
+      "step": 19154
+    },
+    {
+      "epoch": 0.38312,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.007439931233723958,
+      "learning_rate": 0.0001,
+      "loss": 3.7624,
+      "loss/crossentropy": 1.4123128056526184,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1485099121928215,
+      "step": 19156
+    },
+    {
+      "epoch": 0.38316,
+      "grad_norm": 1.8125,
+      "grad_norm_var": 0.009959920247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.2364,
+      "loss/crossentropy": 1.9472095966339111,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16975026577711105,
+      "step": 19158
+    },
+    {
+      "epoch": 0.3832,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0094390869140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2025,
+      "loss/crossentropy": 2.112083077430725,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2198200300335884,
+      "step": 19160
+    },
+    {
+      "epoch": 0.38324,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0121246337890625,
+      "learning_rate": 0.0001,
+      "loss": 3.9661,
+      "loss/crossentropy": 2.1301331520080566,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2039676010608673,
+      "step": 19162
+    },
+    {
+      "epoch": 0.38328,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.019456990559895835,
+      "learning_rate": 0.0001,
+      "loss": 4.0655,
+      "loss/crossentropy": 1.8461318016052246,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17811527848243713,
+      "step": 19164
+    },
+    {
+      "epoch": 0.38332,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.018973795572916667,
+      "learning_rate": 0.0001,
+      "loss": 4.0408,
+      "loss/crossentropy": 2.102015793323517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21166063100099564,
+      "step": 19166
+    },
+    {
+      "epoch": 0.38336,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.020167795817057292,
+      "learning_rate": 0.0001,
+      "loss": 3.9572,
+      "loss/crossentropy": 1.9496545791625977,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17148377001285553,
+      "step": 19168
+    },
+    {
+      "epoch": 0.3834,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.020643870035807293,
+      "learning_rate": 0.0001,
+      "loss": 4.0675,
+      "loss/crossentropy": 2.059878885746002,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1984856054186821,
+      "step": 19170
+    },
+    {
+      "epoch": 0.38344,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.017923990885416668,
+      "learning_rate": 0.0001,
+      "loss": 3.6476,
+      "loss/crossentropy": 1.7966619729995728,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17823103070259094,
+      "step": 19172
+    },
+    {
+      "epoch": 0.38348,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.01645075480143229,
+      "learning_rate": 0.0001,
+      "loss": 3.6671,
+      "loss/crossentropy": 2.0173062086105347,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19681749492883682,
+      "step": 19174
+    },
+    {
+      "epoch": 0.38352,
+      "grad_norm": 1.7265625,
+      "grad_norm_var": 0.019108072916666666,
+      "learning_rate": 0.0001,
+      "loss": 3.6439,
+      "loss/crossentropy": 1.6330417394638062,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17260564863681793,
+      "step": 19176
+    },
+    {
+      "epoch": 0.38356,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.01646728515625,
+      "learning_rate": 0.0001,
+      "loss": 3.8957,
+      "loss/crossentropy": 1.6575458645820618,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17257437109947205,
+      "step": 19178
+    },
+    {
+      "epoch": 0.3836,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.006266021728515625,
+      "learning_rate": 0.0001,
+      "loss": 3.8633,
+      "loss/crossentropy": 2.094203770160675,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2039681375026703,
+      "step": 19180
+    },
+    {
+      "epoch": 0.38364,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.006306966145833333,
+      "learning_rate": 0.0001,
+      "loss": 3.9579,
+      "loss/crossentropy": 1.9561032056808472,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19183862209320068,
+      "step": 19182
+    },
+    {
+      "epoch": 0.38368,
+      "grad_norm": 2.609375,
+      "grad_norm_var": 0.03765055338541667,
+      "learning_rate": 0.0001,
+      "loss": 4.4288,
+      "loss/crossentropy": 1.950503408908844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.198556549847126,
+      "step": 19184
+    },
+    {
+      "epoch": 0.38372,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0398193359375,
+      "learning_rate": 0.0001,
+      "loss": 4.2354,
+      "loss/crossentropy": 2.421600103378296,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.208856999874115,
+      "step": 19186
+    },
+    {
+      "epoch": 0.38376,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.04036356608072917,
+      "learning_rate": 0.0001,
+      "loss": 3.9318,
+      "loss/crossentropy": 1.7086477279663086,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19327964633703232,
+      "step": 19188
+    },
+    {
+      "epoch": 0.3838,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.041071573893229164,
+      "learning_rate": 0.0001,
+      "loss": 4.1488,
+      "loss/crossentropy": 2.1670665740966797,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19333725422620773,
+      "step": 19190
+    },
+    {
+      "epoch": 0.38384,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.03430582682291667,
+      "learning_rate": 0.0001,
+      "loss": 4.3145,
+      "loss/crossentropy": 2.218670129776001,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20835422724485397,
+      "step": 19192
+    },
+    {
+      "epoch": 0.38388,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.037699127197265626,
+      "learning_rate": 0.0001,
+      "loss": 3.7753,
+      "loss/crossentropy": 1.909380555152893,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18018050491809845,
+      "step": 19194
+    },
+    {
+      "epoch": 0.38392,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.038211822509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1049,
+      "loss/crossentropy": 1.9840248227119446,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19120849668979645,
+      "step": 19196
+    },
+    {
+      "epoch": 0.38396,
+      "grad_norm": 2.265625,
+      "grad_norm_var": 0.042012532552083336,
+      "learning_rate": 0.0001,
+      "loss": 4.4124,
+      "loss/crossentropy": 2.3527809381484985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20009320974349976,
+      "step": 19198
+    },
+    {
+      "epoch": 0.384,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0181793212890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1011,
+      "loss/crossentropy": 2.2111377716064453,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19603972136974335,
+      "step": 19200
+    },
+    {
+      "epoch": 0.38404,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.01685358683268229,
+      "learning_rate": 0.0001,
+      "loss": 3.9995,
+      "loss/crossentropy": 1.6832327842712402,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18738068640232086,
+      "step": 19202
+    },
+    {
+      "epoch": 0.38408,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.015805816650390624,
+      "learning_rate": 0.0001,
+      "loss": 3.9789,
+      "loss/crossentropy": 1.9412715435028076,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1835481896996498,
+      "step": 19204
+    },
+    {
+      "epoch": 0.38412,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.013451131184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.203,
+      "loss/crossentropy": 2.0641706585884094,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20165929198265076,
+      "step": 19206
+    },
+    {
+      "epoch": 0.38416,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.011400349934895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0054,
+      "loss/crossentropy": 2.2053582668304443,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2108134999871254,
+      "step": 19208
+    },
+    {
+      "epoch": 0.3842,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.010015614827473958,
+      "learning_rate": 0.0001,
+      "loss": 3.8549,
+      "loss/crossentropy": 2.0534805059432983,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19088375568389893,
+      "step": 19210
+    },
+    {
+      "epoch": 0.38424,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.010969034830729167,
+      "learning_rate": 0.0001,
+      "loss": 4.06,
+      "loss/crossentropy": 1.8782867789268494,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17865915596485138,
+      "step": 19212
+    },
+    {
+      "epoch": 0.38428,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0059201558430989586,
+      "learning_rate": 0.0001,
+      "loss": 4.0641,
+      "loss/crossentropy": 1.9773831963539124,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20622816681861877,
+      "step": 19214
+    },
+    {
+      "epoch": 0.38432,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0067942301432291664,
+      "learning_rate": 0.0001,
+      "loss": 4.1479,
+      "loss/crossentropy": 2.0890401005744934,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2173013687133789,
+      "step": 19216
+    },
+    {
+      "epoch": 0.38436,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.008676910400390625,
+      "learning_rate": 0.0001,
+      "loss": 3.7221,
+      "loss/crossentropy": 1.858555793762207,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15820113569498062,
+      "step": 19218
+    },
+    {
+      "epoch": 0.3844,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.00911865234375,
+      "learning_rate": 0.0001,
+      "loss": 3.9924,
+      "loss/crossentropy": 2.2149851322174072,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2221473827958107,
+      "step": 19220
+    },
+    {
+      "epoch": 0.38444,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.009490712483723959,
+      "learning_rate": 0.0001,
+      "loss": 3.7057,
+      "loss/crossentropy": 2.086311161518097,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1791597157716751,
+      "step": 19222
+    },
+    {
+      "epoch": 0.38448,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.009112294514973958,
+      "learning_rate": 0.0001,
+      "loss": 3.9117,
+      "loss/crossentropy": 1.810127079486847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2018662691116333,
+      "step": 19224
+    },
+    {
+      "epoch": 0.38452,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.009064737955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.3156,
+      "loss/crossentropy": 2.132863163948059,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22217299789190292,
+      "step": 19226
+    },
+    {
+      "epoch": 0.38456,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.008299763997395833,
+      "learning_rate": 0.0001,
+      "loss": 3.8697,
+      "loss/crossentropy": 2.0617652535438538,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1968286782503128,
+      "step": 19228
+    },
+    {
+      "epoch": 0.3846,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.007250722249348958,
+      "learning_rate": 0.0001,
+      "loss": 3.8566,
+      "loss/crossentropy": 1.9286837577819824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20616496354341507,
+      "step": 19230
+    },
+    {
+      "epoch": 0.38464,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.008150227864583333,
+      "learning_rate": 0.0001,
+      "loss": 4.2854,
+      "loss/crossentropy": 2.009112238883972,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19038759917020798,
+      "step": 19232
+    },
+    {
+      "epoch": 0.38468,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.006818644205729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0879,
+      "loss/crossentropy": 2.1937737464904785,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20164573192596436,
+      "step": 19234
+    },
+    {
+      "epoch": 0.38472,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.00587158203125,
+      "learning_rate": 0.0001,
+      "loss": 4.2356,
+      "loss/crossentropy": 1.981977641582489,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20445218682289124,
+      "step": 19236
+    },
+    {
+      "epoch": 0.38476,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.004648590087890625,
+      "learning_rate": 0.0001,
+      "loss": 3.9515,
+      "loss/crossentropy": 1.6332372426986694,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1629931926727295,
+      "step": 19238
+    },
+    {
+      "epoch": 0.3848,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.003952789306640625,
+      "learning_rate": 0.0001,
+      "loss": 3.9061,
+      "loss/crossentropy": 1.8656854629516602,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18546659499406815,
+      "step": 19240
+    },
+    {
+      "epoch": 0.38484,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.005594635009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1251,
+      "loss/crossentropy": 2.219391703605652,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21327267587184906,
+      "step": 19242
+    },
+    {
+      "epoch": 0.38488,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.007883453369140625,
+      "learning_rate": 0.0001,
+      "loss": 4.448,
+      "loss/crossentropy": 1.9518468976020813,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23419301211833954,
+      "step": 19244
+    },
+    {
+      "epoch": 0.38492,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.007411448160807291,
+      "learning_rate": 0.0001,
+      "loss": 4.2947,
+      "loss/crossentropy": 2.190311014652252,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2218230664730072,
+      "step": 19246
+    },
+    {
+      "epoch": 0.38496,
+      "grad_norm": 1.765625,
+      "grad_norm_var": 0.009633127848307292,
+      "learning_rate": 0.0001,
+      "loss": 3.8475,
+      "loss/crossentropy": 1.8347881436347961,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20036222785711288,
+      "step": 19248
+    },
+    {
+      "epoch": 0.385,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.012157185872395834,
+      "learning_rate": 0.0001,
+      "loss": 3.9732,
+      "loss/crossentropy": 1.949516236782074,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20307819545269012,
+      "step": 19250
+    },
+    {
+      "epoch": 0.38504,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.012892405192057291,
+      "learning_rate": 0.0001,
+      "loss": 4.007,
+      "loss/crossentropy": 1.883664846420288,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1825740560889244,
+      "step": 19252
+    },
+    {
+      "epoch": 0.38508,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.0146881103515625,
+      "learning_rate": 0.0001,
+      "loss": 4.1794,
+      "loss/crossentropy": 2.1245557069778442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20935190469026566,
+      "step": 19254
+    },
+    {
+      "epoch": 0.38512,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.015166982014973959,
+      "learning_rate": 0.0001,
+      "loss": 3.9125,
+      "loss/crossentropy": 1.546354353427887,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15509265661239624,
+      "step": 19256
+    },
+    {
+      "epoch": 0.38516,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.014290110270182291,
+      "learning_rate": 0.0001,
+      "loss": 4.1744,
+      "loss/crossentropy": 1.979533076286316,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20994720607995987,
+      "step": 19258
+    },
+    {
+      "epoch": 0.3852,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.014074452718098958,
+      "learning_rate": 0.0001,
+      "loss": 3.7774,
+      "loss/crossentropy": 2.125569224357605,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21175500005483627,
+      "step": 19260
+    },
+    {
+      "epoch": 0.38524,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.0124267578125,
+      "learning_rate": 0.0001,
+      "loss": 3.911,
+      "loss/crossentropy": 2.027758777141571,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20255477726459503,
+      "step": 19262
+    },
+    {
+      "epoch": 0.38528,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.011058553059895834,
+      "learning_rate": 0.0001,
+      "loss": 3.4728,
+      "loss/crossentropy": 1.9937690496444702,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1836088001728058,
+      "step": 19264
+    },
+    {
+      "epoch": 0.38532,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.00863037109375,
+      "learning_rate": 0.0001,
+      "loss": 4.1797,
+      "loss/crossentropy": 2.0846773386001587,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18216369301080704,
+      "step": 19266
+    },
+    {
+      "epoch": 0.38536,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.008429972330729167,
+      "learning_rate": 0.0001,
+      "loss": 3.8911,
+      "loss/crossentropy": 1.445238471031189,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17925872653722763,
+      "step": 19268
+    },
+    {
+      "epoch": 0.3854,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0061757405598958336,
+      "learning_rate": 0.0001,
+      "loss": 4.0615,
+      "loss/crossentropy": 2.0442845821380615,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2024901956319809,
+      "step": 19270
+    },
+    {
+      "epoch": 0.38544,
+      "grad_norm": 1.7265625,
+      "grad_norm_var": 0.009144846598307292,
+      "learning_rate": 0.0001,
+      "loss": 4.0898,
+      "loss/crossentropy": 1.9336092472076416,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17869817465543747,
+      "step": 19272
+    },
+    {
+      "epoch": 0.38548,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.010884348551432292,
+      "learning_rate": 0.0001,
+      "loss": 4.1939,
+      "loss/crossentropy": 1.815299928188324,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.182998888194561,
+      "step": 19274
+    },
+    {
+      "epoch": 0.38552,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.008886464436848958,
+      "learning_rate": 0.0001,
+      "loss": 4.2121,
+      "loss/crossentropy": 2.091490864753723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19843900948762894,
+      "step": 19276
+    },
+    {
+      "epoch": 0.38556,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009608713785807292,
+      "learning_rate": 0.0001,
+      "loss": 4.098,
+      "loss/crossentropy": 1.9923794865608215,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.186911903321743,
+      "step": 19278
+    },
+    {
+      "epoch": 0.3856,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.009867350260416666,
+      "learning_rate": 0.0001,
+      "loss": 4.0658,
+      "loss/crossentropy": 1.9323501586914062,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1931384950876236,
+      "step": 19280
+    },
+    {
+      "epoch": 0.38564,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.010262044270833333,
+      "learning_rate": 0.0001,
+      "loss": 3.8819,
+      "loss/crossentropy": 1.8469224572181702,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1835423856973648,
+      "step": 19282
+    },
+    {
+      "epoch": 0.38568,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.010453287760416667,
+      "learning_rate": 0.0001,
+      "loss": 3.9027,
+      "loss/crossentropy": 1.9476045370101929,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17014781385660172,
+      "step": 19284
+    },
+    {
+      "epoch": 0.38572,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.010285441080729167,
+      "learning_rate": 0.0001,
+      "loss": 3.9943,
+      "loss/crossentropy": 1.7913609743118286,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18661632388830185,
+      "step": 19286
+    },
+    {
+      "epoch": 0.38576,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.00760498046875,
+      "learning_rate": 0.0001,
+      "loss": 4.0283,
+      "loss/crossentropy": 1.9682837128639221,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1799641102552414,
+      "step": 19288
+    },
+    {
+      "epoch": 0.3858,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.006331125895182292,
+      "learning_rate": 0.0001,
+      "loss": 4.3075,
+      "loss/crossentropy": 2.117633819580078,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.208334282040596,
+      "step": 19290
+    },
+    {
+      "epoch": 0.38584,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.005761464436848958,
+      "learning_rate": 0.0001,
+      "loss": 4.1215,
+      "loss/crossentropy": 2.1208746433258057,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20928698778152466,
+      "step": 19292
+    },
+    {
+      "epoch": 0.38588,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.005549112955729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0572,
+      "loss/crossentropy": 1.9424505829811096,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18668201565742493,
+      "step": 19294
+    },
+    {
+      "epoch": 0.38592,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.005277252197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1575,
+      "loss/crossentropy": 2.251755177974701,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21153685450553894,
+      "step": 19296
+    },
+    {
+      "epoch": 0.38596,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.007114410400390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1885,
+      "loss/crossentropy": 2.1629436016082764,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.206063412129879,
+      "step": 19298
+    },
+    {
+      "epoch": 0.386,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0069732666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1267,
+      "loss/crossentropy": 2.0333832502365112,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1872929260134697,
+      "step": 19300
+    },
+    {
+      "epoch": 0.38604,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.007641347249348959,
+      "learning_rate": 0.0001,
+      "loss": 3.8433,
+      "loss/crossentropy": 1.8364217281341553,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17408161610364914,
+      "step": 19302
+    },
+    {
+      "epoch": 0.38608,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.005973307291666666,
+      "learning_rate": 0.0001,
+      "loss": 3.9175,
+      "loss/crossentropy": 1.975037932395935,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18705998361110687,
+      "step": 19304
+    },
+    {
+      "epoch": 0.38612,
+      "grad_norm": 1.7890625,
+      "grad_norm_var": 0.006477864583333334,
+      "learning_rate": 0.0001,
+      "loss": 3.7868,
+      "loss/crossentropy": 1.6889175176620483,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16312924772500992,
+      "step": 19306
+    },
+    {
+      "epoch": 0.38616,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.006392161051432292,
+      "learning_rate": 0.0001,
+      "loss": 4.1299,
+      "loss/crossentropy": 2.2644211053848267,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18390889465808868,
+      "step": 19308
+    },
+    {
+      "epoch": 0.3862,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.009723917643229166,
+      "learning_rate": 0.0001,
+      "loss": 4.325,
+      "loss/crossentropy": 2.180909037590027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20263531804084778,
+      "step": 19310
+    },
+    {
+      "epoch": 0.38624,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010138956705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.2001,
+      "loss/crossentropy": 2.4007444381713867,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23251044005155563,
+      "step": 19312
+    },
+    {
+      "epoch": 0.38628,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.010534413655598958,
+      "learning_rate": 0.0001,
+      "loss": 3.9862,
+      "loss/crossentropy": 1.8122236728668213,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18613354116678238,
+      "step": 19314
+    },
+    {
+      "epoch": 0.38632,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.010404459635416667,
+      "learning_rate": 0.0001,
+      "loss": 4.1379,
+      "loss/crossentropy": 2.159746825695038,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21711497008800507,
+      "step": 19316
+    },
+    {
+      "epoch": 0.38636,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.009627024332682291,
+      "learning_rate": 0.0001,
+      "loss": 3.9929,
+      "loss/crossentropy": 2.129119336605072,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19623322039842606,
+      "step": 19318
+    },
+    {
+      "epoch": 0.3864,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.009934234619140624,
+      "learning_rate": 0.0001,
+      "loss": 4.1578,
+      "loss/crossentropy": 2.322758913040161,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20978064835071564,
+      "step": 19320
+    },
+    {
+      "epoch": 0.38644,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010322825113932291,
+      "learning_rate": 0.0001,
+      "loss": 4.0266,
+      "loss/crossentropy": 2.2643767595291138,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21338346600532532,
+      "step": 19322
+    },
+    {
+      "epoch": 0.38648,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.010731760660807292,
+      "learning_rate": 0.0001,
+      "loss": 4.0614,
+      "loss/crossentropy": 1.8038535118103027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19042538851499557,
+      "step": 19324
+    },
+    {
+      "epoch": 0.38652,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.0069620768229166664,
+      "learning_rate": 0.0001,
+      "loss": 3.9641,
+      "loss/crossentropy": 2.0847853422164917,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1965423971414566,
+      "step": 19326
+    },
+    {
+      "epoch": 0.38656,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007692209879557292,
+      "learning_rate": 0.0001,
+      "loss": 4.3347,
+      "loss/crossentropy": 2.2121591567993164,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21294504404067993,
+      "step": 19328
+    },
+    {
+      "epoch": 0.3866,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.005985260009765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9869,
+      "loss/crossentropy": 2.100327968597412,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20181559026241302,
+      "step": 19330
+    },
+    {
+      "epoch": 0.38664,
+      "grad_norm": 1.78125,
+      "grad_norm_var": 0.00750732421875,
+      "learning_rate": 0.0001,
+      "loss": 3.642,
+      "loss/crossentropy": 1.7100898623466492,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18045856803655624,
+      "step": 19332
+    },
+    {
+      "epoch": 0.38668,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.008337148030598958,
+      "learning_rate": 0.0001,
+      "loss": 3.712,
+      "loss/crossentropy": 1.6435258388519287,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1821775734424591,
+      "step": 19334
+    },
+    {
+      "epoch": 0.38672,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.008674875895182291,
+      "learning_rate": 0.0001,
+      "loss": 3.8334,
+      "loss/crossentropy": 2.338008165359497,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2040715217590332,
+      "step": 19336
+    },
+    {
+      "epoch": 0.38676,
+      "grad_norm": 1.78125,
+      "grad_norm_var": 0.008454386393229167,
+      "learning_rate": 0.0001,
+      "loss": 3.8911,
+      "loss/crossentropy": 1.8373408913612366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17516936361789703,
+      "step": 19338
+    },
+    {
+      "epoch": 0.3868,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.008422597249348959,
+      "learning_rate": 0.0001,
+      "loss": 3.9902,
+      "loss/crossentropy": 2.0346588492393494,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20979416370391846,
+      "step": 19340
+    },
+    {
+      "epoch": 0.38684,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.009399159749348959,
+      "learning_rate": 0.0001,
+      "loss": 4.1759,
+      "loss/crossentropy": 2.083326816558838,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20702984929084778,
+      "step": 19342
+    },
+    {
+      "epoch": 0.38688,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.012473297119140626,
+      "learning_rate": 0.0001,
+      "loss": 4.611,
+      "loss/crossentropy": 2.4100319147109985,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2617071568965912,
+      "step": 19344
+    },
+    {
+      "epoch": 0.38692,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.012550608317057291,
+      "learning_rate": 0.0001,
+      "loss": 3.8864,
+      "loss/crossentropy": 1.7147992849349976,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.174342080950737,
+      "step": 19346
+    },
+    {
+      "epoch": 0.38696,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.013602701822916667,
+      "learning_rate": 0.0001,
+      "loss": 4.1508,
+      "loss/crossentropy": 2.041933536529541,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2024158239364624,
+      "step": 19348
+    },
+    {
+      "epoch": 0.387,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.012434895833333333,
+      "learning_rate": 0.0001,
+      "loss": 4.0929,
+      "loss/crossentropy": 1.835128128528595,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1861354559659958,
+      "step": 19350
+    },
+    {
+      "epoch": 0.38704,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.013370768229166666,
+      "learning_rate": 0.0001,
+      "loss": 4.164,
+      "loss/crossentropy": 1.9472790360450745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19485513865947723,
+      "step": 19352
+    },
+    {
+      "epoch": 0.38708,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.012428538004557291,
+      "learning_rate": 0.0001,
+      "loss": 4.0217,
+      "loss/crossentropy": 2.0546197295188904,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1883516013622284,
+      "step": 19354
+    },
+    {
+      "epoch": 0.38712,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.013169097900390624,
+      "learning_rate": 0.0001,
+      "loss": 3.6439,
+      "loss/crossentropy": 1.8285245299339294,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1730484738945961,
+      "step": 19356
+    },
+    {
+      "epoch": 0.38716,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.013890584309895834,
+      "learning_rate": 0.0001,
+      "loss": 4.2169,
+      "loss/crossentropy": 2.320949673652649,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21517502516508102,
+      "step": 19358
+    },
+    {
+      "epoch": 0.3872,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011289215087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0196,
+      "loss/crossentropy": 2.2944518327713013,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20343206077814102,
+      "step": 19360
+    },
+    {
+      "epoch": 0.38724,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0112945556640625,
+      "learning_rate": 0.0001,
+      "loss": 4.2113,
+      "loss/crossentropy": 2.134889602661133,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19354654103517532,
+      "step": 19362
+    },
+    {
+      "epoch": 0.38728,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.009639231363932292,
+      "learning_rate": 0.0001,
+      "loss": 4.1244,
+      "loss/crossentropy": 2.206334412097931,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20670340210199356,
+      "step": 19364
+    },
+    {
+      "epoch": 0.38732,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.009769694010416666,
+      "learning_rate": 0.0001,
+      "loss": 3.8194,
+      "loss/crossentropy": 2.1286932229995728,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19814791530370712,
+      "step": 19366
+    },
+    {
+      "epoch": 0.38736,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.0066314697265625,
+      "learning_rate": 0.0001,
+      "loss": 4.1533,
+      "loss/crossentropy": 2.119162678718567,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18044983595609665,
+      "step": 19368
+    },
+    {
+      "epoch": 0.3874,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.005060831705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0061,
+      "loss/crossentropy": 2.4316102266311646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22509240359067917,
+      "step": 19370
+    },
+    {
+      "epoch": 0.38744,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.0060117085774739586,
+      "learning_rate": 0.0001,
+      "loss": 3.9984,
+      "loss/crossentropy": 2.540325403213501,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22844604402780533,
+      "step": 19372
+    },
+    {
+      "epoch": 0.38748,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.005782063802083333,
+      "learning_rate": 0.0001,
+      "loss": 4.1795,
+      "loss/crossentropy": 2.3002058267593384,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21805168688297272,
+      "step": 19374
+    },
+    {
+      "epoch": 0.38752,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0058258056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.274,
+      "loss/crossentropy": 2.2381847500801086,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19565512239933014,
+      "step": 19376
+    },
+    {
+      "epoch": 0.38756,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.0060302734375,
+      "learning_rate": 0.0001,
+      "loss": 4.2683,
+      "loss/crossentropy": 2.101936161518097,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.198813758790493,
+      "step": 19378
+    },
+    {
+      "epoch": 0.3876,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.0060943603515625,
+      "learning_rate": 0.0001,
+      "loss": 3.8679,
+      "loss/crossentropy": 2.1088311672210693,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21237140893936157,
+      "step": 19380
+    },
+    {
+      "epoch": 0.38764,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0060943603515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9588,
+      "loss/crossentropy": 2.0985374450683594,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18979168683290482,
+      "step": 19382
+    },
+    {
+      "epoch": 0.38768,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.005690256754557292,
+      "learning_rate": 0.0001,
+      "loss": 4.0798,
+      "loss/crossentropy": 2.3004449605941772,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2053247168660164,
+      "step": 19384
+    },
+    {
+      "epoch": 0.38772,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.005277252197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0285,
+      "loss/crossentropy": 2.1257707476615906,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22222986072301865,
+      "step": 19386
+    },
+    {
+      "epoch": 0.38776,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.003733062744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.1336,
+      "loss/crossentropy": 2.3034613132476807,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2174137905240059,
+      "step": 19388
+    },
+    {
+      "epoch": 0.3878,
+      "grad_norm": 1.7578125,
+      "grad_norm_var": 0.006245930989583333,
+      "learning_rate": 0.0001,
+      "loss": 3.9854,
+      "loss/crossentropy": 2.1303011178970337,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22152414172887802,
+      "step": 19390
+    },
+    {
+      "epoch": 0.38784,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.010412343343098958,
+      "learning_rate": 0.0001,
+      "loss": 4.1868,
+      "loss/crossentropy": 1.8575092554092407,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18628983944654465,
+      "step": 19392
+    },
+    {
+      "epoch": 0.38788,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011694081624348958,
+      "learning_rate": 0.0001,
+      "loss": 4.1219,
+      "loss/crossentropy": 2.0332913994789124,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2499927133321762,
+      "step": 19394
+    },
+    {
+      "epoch": 0.38792,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.013004302978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.2275,
+      "loss/crossentropy": 2.0926660895347595,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19264871627092361,
+      "step": 19396
+    },
+    {
+      "epoch": 0.38796,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.0156005859375,
+      "learning_rate": 0.0001,
+      "loss": 4.2337,
+      "loss/crossentropy": 2.08814400434494,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21137948334217072,
+      "step": 19398
+    },
+    {
+      "epoch": 0.388,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.019162750244140624,
+      "learning_rate": 0.0001,
+      "loss": 4.0263,
+      "loss/crossentropy": 2.074121594429016,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20524942874908447,
+      "step": 19400
+    },
+    {
+      "epoch": 0.38804,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.020589192708333332,
+      "learning_rate": 0.0001,
+      "loss": 3.9437,
+      "loss/crossentropy": 1.8818755745887756,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1794123500585556,
+      "step": 19402
+    },
+    {
+      "epoch": 0.38808,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.020992024739583334,
+      "learning_rate": 0.0001,
+      "loss": 3.8638,
+      "loss/crossentropy": 2.5050524473190308,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22337081283330917,
+      "step": 19404
+    },
+    {
+      "epoch": 0.38812,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0180572509765625,
+      "learning_rate": 0.0001,
+      "loss": 4.2158,
+      "loss/crossentropy": 2.2035861015319824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20753300189971924,
+      "step": 19406
+    },
+    {
+      "epoch": 0.38816,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.014012654622395834,
+      "learning_rate": 0.0001,
+      "loss": 3.8975,
+      "loss/crossentropy": 2.2270091772079468,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20527157932519913,
+      "step": 19408
+    },
+    {
+      "epoch": 0.3882,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.013504791259765624,
+      "learning_rate": 0.0001,
+      "loss": 3.951,
+      "loss/crossentropy": 1.6546313762664795,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17243140190839767,
+      "step": 19410
+    },
+    {
+      "epoch": 0.38824,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.012756093343098959,
+      "learning_rate": 0.0001,
+      "loss": 4.0653,
+      "loss/crossentropy": 1.8858160376548767,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18621040880680084,
+      "step": 19412
+    },
+    {
+      "epoch": 0.38828,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.011875152587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.4744,
+      "loss/crossentropy": 1.9876565337181091,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30138373374938965,
+      "step": 19414
+    },
+    {
+      "epoch": 0.38832,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.013053131103515626,
+      "learning_rate": 0.0001,
+      "loss": 3.7802,
+      "loss/crossentropy": 2.1887502670288086,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1910400092601776,
+      "step": 19416
+    },
+    {
+      "epoch": 0.38836,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.012511952718098959,
+      "learning_rate": 0.0001,
+      "loss": 3.9398,
+      "loss/crossentropy": 2.246508002281189,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2018478363752365,
+      "step": 19418
+    },
+    {
+      "epoch": 0.3884,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.011510976155598958,
+      "learning_rate": 0.0001,
+      "loss": 4.1703,
+      "loss/crossentropy": 2.041202425956726,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1890542358160019,
+      "step": 19420
+    },
+    {
+      "epoch": 0.38844,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.016904449462890624,
+      "learning_rate": 0.0001,
+      "loss": 4.4154,
+      "loss/crossentropy": 2.0470627546310425,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19648576527833939,
+      "step": 19422
+    },
+    {
+      "epoch": 0.38848,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.0155181884765625,
+      "learning_rate": 0.0001,
+      "loss": 4.1794,
+      "loss/crossentropy": 2.172736167907715,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22087720036506653,
+      "step": 19424
+    },
+    {
+      "epoch": 0.38852,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.015811920166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0106,
+      "loss/crossentropy": 2.0988662242889404,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18553303182125092,
+      "step": 19426
+    },
+    {
+      "epoch": 0.38856,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.017134348551432293,
+      "learning_rate": 0.0001,
+      "loss": 4.1368,
+      "loss/crossentropy": 1.821892261505127,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19670143723487854,
+      "step": 19428
+    },
+    {
+      "epoch": 0.3886,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.016747792561848957,
+      "learning_rate": 0.0001,
+      "loss": 3.7878,
+      "loss/crossentropy": 1.8893300294876099,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18850861489772797,
+      "step": 19430
+    },
+    {
+      "epoch": 0.38864,
+      "grad_norm": 2.984375,
+      "grad_norm_var": 0.07888997395833333,
+      "learning_rate": 0.0001,
+      "loss": 4.243,
+      "loss/crossentropy": 2.210429847240448,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17908993363380432,
+      "step": 19432
+    },
+    {
+      "epoch": 0.38868,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.07731526692708333,
+      "learning_rate": 0.0001,
+      "loss": 3.9939,
+      "loss/crossentropy": 2.164771556854248,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1919529214501381,
+      "step": 19434
+    },
+    {
+      "epoch": 0.38872,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.07656631469726563,
+      "learning_rate": 0.0001,
+      "loss": 4.1721,
+      "loss/crossentropy": 2.1335190534591675,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2263372465968132,
+      "step": 19436
+    },
+    {
+      "epoch": 0.38876,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.073779296875,
+      "learning_rate": 0.0001,
+      "loss": 4.06,
+      "loss/crossentropy": 2.199760317802429,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20560061931610107,
+      "step": 19438
+    },
+    {
+      "epoch": 0.3888,
+      "grad_norm": 1.8125,
+      "grad_norm_var": 0.07665608723958334,
+      "learning_rate": 0.0001,
+      "loss": 3.9367,
+      "loss/crossentropy": 2.2948896884918213,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2016046792268753,
+      "step": 19440
+    },
+    {
+      "epoch": 0.38884,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.07588882446289062,
+      "learning_rate": 0.0001,
+      "loss": 4.1444,
+      "loss/crossentropy": 2.120785415172577,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18029560148715973,
+      "step": 19442
+    },
+    {
+      "epoch": 0.38888,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.07493260701497396,
+      "learning_rate": 0.0001,
+      "loss": 3.8872,
+      "loss/crossentropy": 2.049036145210266,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19056373089551926,
+      "step": 19444
+    },
+    {
+      "epoch": 0.38892,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.07226130167643229,
+      "learning_rate": 0.0001,
+      "loss": 4.2652,
+      "loss/crossentropy": 2.0076504945755005,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20951516926288605,
+      "step": 19446
+    },
+    {
+      "epoch": 0.38896,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.006740061442057291,
+      "learning_rate": 0.0001,
+      "loss": 3.8178,
+      "loss/crossentropy": 1.9398083090782166,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21217594295740128,
+      "step": 19448
+    },
+    {
+      "epoch": 0.389,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.00546875,
+      "learning_rate": 0.0001,
+      "loss": 4.0245,
+      "loss/crossentropy": 2.0976104736328125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19852425903081894,
+      "step": 19450
+    },
+    {
+      "epoch": 0.38904,
+      "grad_norm": 7.625,
+      "grad_norm_var": 2.023509724934896,
+      "learning_rate": 0.0001,
+      "loss": 3.9741,
+      "loss/crossentropy": 1.8561761379241943,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17980807274580002,
+      "step": 19452
+    },
+    {
+      "epoch": 0.38908,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 2.0183570861816404,
+      "learning_rate": 0.0001,
+      "loss": 3.913,
+      "loss/crossentropy": 2.1107255816459656,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20205769687891006,
+      "step": 19454
+    },
+    {
+      "epoch": 0.38912,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 2.009368642171224,
+      "learning_rate": 0.0001,
+      "loss": 3.9216,
+      "loss/crossentropy": 2.018397331237793,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18721628934144974,
+      "step": 19456
+    },
+    {
+      "epoch": 0.38916,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 2.010087076822917,
+      "learning_rate": 0.0001,
+      "loss": 3.8738,
+      "loss/crossentropy": 2.016912341117859,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20780544728040695,
+      "step": 19458
+    },
+    {
+      "epoch": 0.3892,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 2.0083984375,
+      "learning_rate": 0.0001,
+      "loss": 3.9523,
+      "loss/crossentropy": 1.945086121559143,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19544267654418945,
+      "step": 19460
+    },
+    {
+      "epoch": 0.38924,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 2.01300048828125,
+      "learning_rate": 0.0001,
+      "loss": 4.1854,
+      "loss/crossentropy": 2.059622883796692,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21501117944717407,
+      "step": 19462
+    },
+    {
+      "epoch": 0.38928,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 2.024466705322266,
+      "learning_rate": 0.0001,
+      "loss": 3.8336,
+      "loss/crossentropy": 1.8915096521377563,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1852220892906189,
+      "step": 19464
+    },
+    {
+      "epoch": 0.38932,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 2.0162534077962238,
+      "learning_rate": 0.0001,
+      "loss": 4.0612,
+      "loss/crossentropy": 1.5394552946090698,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16873380541801453,
+      "step": 19466
+    },
+    {
+      "epoch": 0.38936,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.03472468058268229,
+      "learning_rate": 0.0001,
+      "loss": 4.2426,
+      "loss/crossentropy": 2.0290380716323853,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19913609325885773,
+      "step": 19468
+    },
+    {
+      "epoch": 0.3894,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.033841705322265624,
+      "learning_rate": 0.0001,
+      "loss": 4.1607,
+      "loss/crossentropy": 2.208653211593628,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.190569207072258,
+      "step": 19470
+    },
+    {
+      "epoch": 0.38944,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0092041015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2646,
+      "loss/crossentropy": 2.2873799800872803,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20589639246463776,
+      "step": 19472
+    },
+    {
+      "epoch": 0.38948,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.008014933268229166,
+      "learning_rate": 0.0001,
+      "loss": 4.1662,
+      "loss/crossentropy": 2.152024030685425,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20272599160671234,
+      "step": 19474
+    },
+    {
+      "epoch": 0.38952,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.009374745686848958,
+      "learning_rate": 0.0001,
+      "loss": 3.9932,
+      "loss/crossentropy": 2.0079659819602966,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19110675901174545,
+      "step": 19476
+    },
+    {
+      "epoch": 0.38956,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.010081990559895834,
+      "learning_rate": 0.0001,
+      "loss": 4.1061,
+      "loss/crossentropy": 2.3006476163864136,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2253967523574829,
+      "step": 19478
+    },
+    {
+      "epoch": 0.3896,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.010389963785807291,
+      "learning_rate": 0.0001,
+      "loss": 4.0678,
+      "loss/crossentropy": 2.1429306864738464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20268720388412476,
+      "step": 19480
+    },
+    {
+      "epoch": 0.38964,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011417643229166666,
+      "learning_rate": 0.0001,
+      "loss": 4.0581,
+      "loss/crossentropy": 2.260764956474304,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2121199518442154,
+      "step": 19482
+    },
+    {
+      "epoch": 0.38968,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010636393229166667,
+      "learning_rate": 0.0001,
+      "loss": 4.1607,
+      "loss/crossentropy": 2.3874053955078125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2229148969054222,
+      "step": 19484
+    },
+    {
+      "epoch": 0.38972,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.009227498372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1456,
+      "loss/crossentropy": 2.2740933895111084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2509172707796097,
+      "step": 19486
+    },
+    {
+      "epoch": 0.38976,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.009679921468098958,
+      "learning_rate": 0.0001,
+      "loss": 4.1831,
+      "loss/crossentropy": 2.1949650049209595,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19400373846292496,
+      "step": 19488
+    },
+    {
+      "epoch": 0.3898,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.009549713134765625,
+      "learning_rate": 0.0001,
+      "loss": 3.9984,
+      "loss/crossentropy": 1.4600969552993774,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15833207219839096,
+      "step": 19490
+    },
+    {
+      "epoch": 0.38984,
+      "grad_norm": 2.453125,
+      "grad_norm_var": 0.019913482666015624,
+      "learning_rate": 0.0001,
+      "loss": 4.1694,
+      "loss/crossentropy": 2.0005985498428345,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20003372430801392,
+      "step": 19492
+    },
+    {
+      "epoch": 0.38988,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.020304107666015626,
+      "learning_rate": 0.0001,
+      "loss": 3.8336,
+      "loss/crossentropy": 1.9609830379486084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19764738529920578,
+      "step": 19494
+    },
+    {
+      "epoch": 0.38992,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.01943359375,
+      "learning_rate": 0.0001,
+      "loss": 3.8708,
+      "loss/crossentropy": 1.7853738069534302,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18979863077402115,
+      "step": 19496
+    },
+    {
+      "epoch": 0.38996,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.018961588541666668,
+      "learning_rate": 0.0001,
+      "loss": 4.0335,
+      "loss/crossentropy": 2.004499912261963,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20765355974435806,
+      "step": 19498
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.019870758056640625,
+      "learning_rate": 0.0001,
+      "loss": 4.1277,
+      "loss/crossentropy": 2.0148558020591736,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18088079243898392,
+      "step": 19500
+    },
+    {
+      "epoch": 0.39004,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.022965240478515624,
+      "learning_rate": 0.0001,
+      "loss": 4.1213,
+      "loss/crossentropy": 2.2602895498275757,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22686412185430527,
+      "step": 19502
+    },
+    {
+      "epoch": 0.39008,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.024326324462890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0541,
+      "loss/crossentropy": 2.1789051294326782,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1989632546901703,
+      "step": 19504
+    },
+    {
+      "epoch": 0.39012,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.024568684895833335,
+      "learning_rate": 0.0001,
+      "loss": 3.9234,
+      "loss/crossentropy": 1.93650484085083,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1976509690284729,
+      "step": 19506
+    },
+    {
+      "epoch": 0.39016,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.008351389567057292,
+      "learning_rate": 0.0001,
+      "loss": 3.9519,
+      "loss/crossentropy": 1.9904287457466125,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17271699011325836,
+      "step": 19508
+    },
+    {
+      "epoch": 0.3902,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008194986979166667,
+      "learning_rate": 0.0001,
+      "loss": 3.7009,
+      "loss/crossentropy": 1.4440776705741882,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16108233481645584,
+      "step": 19510
+    },
+    {
+      "epoch": 0.39024,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.008111317952473959,
+      "learning_rate": 0.0001,
+      "loss": 4.0218,
+      "loss/crossentropy": 2.143462061882019,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2024468332529068,
+      "step": 19512
+    },
+    {
+      "epoch": 0.39028,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.008436838785807291,
+      "learning_rate": 0.0001,
+      "loss": 4.1267,
+      "loss/crossentropy": 2.085771322250366,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22135765105485916,
+      "step": 19514
+    },
+    {
+      "epoch": 0.39032,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.008337148030598958,
+      "learning_rate": 0.0001,
+      "loss": 4.1423,
+      "loss/crossentropy": 2.251617908477783,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2026226669549942,
+      "step": 19516
+    },
+    {
+      "epoch": 0.39036,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.003885650634765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3156,
+      "loss/crossentropy": 2.3808701038360596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2263621687889099,
+      "step": 19518
+    },
+    {
+      "epoch": 0.3904,
+      "grad_norm": 1.8203125,
+      "grad_norm_var": 0.004019927978515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0323,
+      "loss/crossentropy": 2.000342011451721,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20008830726146698,
+      "step": 19520
+    },
+    {
+      "epoch": 0.39044,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.0044830322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.3285,
+      "loss/crossentropy": 2.242555856704712,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2443876415491104,
+      "step": 19522
+    },
+    {
+      "epoch": 0.39048,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.005060831705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.344,
+      "loss/crossentropy": 2.2867462635040283,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21522662043571472,
+      "step": 19524
+    },
+    {
+      "epoch": 0.39052,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0053955078125,
+      "learning_rate": 0.0001,
+      "loss": 4.0117,
+      "loss/crossentropy": 2.1892699003219604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.205935537815094,
+      "step": 19526
+    },
+    {
+      "epoch": 0.39056,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.007972971598307291,
+      "learning_rate": 0.0001,
+      "loss": 3.8673,
+      "loss/crossentropy": 2.182734966278076,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1980958804488182,
+      "step": 19528
+    },
+    {
+      "epoch": 0.3906,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0075927734375,
+      "learning_rate": 0.0001,
+      "loss": 4.0193,
+      "loss/crossentropy": 2.3624355792999268,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22833774983882904,
+      "step": 19530
+    },
+    {
+      "epoch": 0.39064,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.007671864827473959,
+      "learning_rate": 0.0001,
+      "loss": 4.0497,
+      "loss/crossentropy": 2.037912607192993,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2032359093427658,
+      "step": 19532
+    },
+    {
+      "epoch": 0.39068,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.007513173421223958,
+      "learning_rate": 0.0001,
+      "loss": 3.9812,
+      "loss/crossentropy": 2.076040804386139,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21829531341791153,
+      "step": 19534
+    },
+    {
+      "epoch": 0.39072,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0058502197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0407,
+      "loss/crossentropy": 2.2695836424827576,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2125643789768219,
+      "step": 19536
+    },
+    {
+      "epoch": 0.39076,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0061279296875,
+      "learning_rate": 0.0001,
+      "loss": 4.2664,
+      "loss/crossentropy": 1.9343088269233704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2045460194349289,
+      "step": 19538
+    },
+    {
+      "epoch": 0.3908,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.005272420247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.0757,
+      "loss/crossentropy": 1.9968918561935425,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17788998782634735,
+      "step": 19540
+    },
+    {
+      "epoch": 0.39084,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.005467732747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.094,
+      "loss/crossentropy": 2.157910704612732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19516880810260773,
+      "step": 19542
+    },
+    {
+      "epoch": 0.39088,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.004042307535807292,
+      "learning_rate": 0.0001,
+      "loss": 4.1056,
+      "loss/crossentropy": 2.1784998178482056,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20280858874320984,
+      "step": 19544
+    },
+    {
+      "epoch": 0.39092,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.00438232421875,
+      "learning_rate": 0.0001,
+      "loss": 4.0796,
+      "loss/crossentropy": 2.3444844484329224,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2418832629919052,
+      "step": 19546
+    },
+    {
+      "epoch": 0.39096,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0038083394368489585,
+      "learning_rate": 0.0001,
+      "loss": 3.9802,
+      "loss/crossentropy": 1.90557062625885,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2010379433631897,
+      "step": 19548
+    },
+    {
+      "epoch": 0.391,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.004107411702473958,
+      "learning_rate": 0.0001,
+      "loss": 3.7756,
+      "loss/crossentropy": 1.4939787983894348,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1362278200685978,
+      "step": 19550
+    },
+    {
+      "epoch": 0.39104,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0063517252604166664,
+      "learning_rate": 0.0001,
+      "loss": 4.1738,
+      "loss/crossentropy": 2.037553310394287,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2159510999917984,
+      "step": 19552
+    },
+    {
+      "epoch": 0.39108,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.004816691080729167,
+      "learning_rate": 0.0001,
+      "loss": 3.887,
+      "loss/crossentropy": 1.9987242221832275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19919036328792572,
+      "step": 19554
+    },
+    {
+      "epoch": 0.39112,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.0048906962076822914,
+      "learning_rate": 0.0001,
+      "loss": 4.0298,
+      "loss/crossentropy": 2.1593196392059326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21370293200016022,
+      "step": 19556
+    },
+    {
+      "epoch": 0.39116,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.005631510416666667,
+      "learning_rate": 0.0001,
+      "loss": 3.8348,
+      "loss/crossentropy": 1.8323914408683777,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20382094383239746,
+      "step": 19558
+    },
+    {
+      "epoch": 0.3912,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.0053059895833333336,
+      "learning_rate": 0.0001,
+      "loss": 4.1138,
+      "loss/crossentropy": 2.0240999460220337,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19374582171440125,
+      "step": 19560
+    },
+    {
+      "epoch": 0.39124,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.005304972330729167,
+      "learning_rate": 0.0001,
+      "loss": 4.4349,
+      "loss/crossentropy": 2.377955436706543,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2213267982006073,
+      "step": 19562
+    },
+    {
+      "epoch": 0.39128,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.006154123942057292,
+      "learning_rate": 0.0001,
+      "loss": 4.0878,
+      "loss/crossentropy": 1.724283754825592,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20119256526231766,
+      "step": 19564
+    },
+    {
+      "epoch": 0.39132,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.005655670166015625,
+      "learning_rate": 0.0001,
+      "loss": 3.9066,
+      "loss/crossentropy": 2.0352718234062195,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19105417281389236,
+      "step": 19566
+    },
+    {
+      "epoch": 0.39136,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.0037913004557291667,
+      "learning_rate": 0.0001,
+      "loss": 3.7868,
+      "loss/crossentropy": 2.078265905380249,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19992397725582123,
+      "step": 19568
+    },
+    {
+      "epoch": 0.3914,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.004329172770182291,
+      "learning_rate": 0.0001,
+      "loss": 3.8276,
+      "loss/crossentropy": 1.9722577929496765,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17697854340076447,
+      "step": 19570
+    },
+    {
+      "epoch": 0.39144,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.004915364583333333,
+      "learning_rate": 0.0001,
+      "loss": 4.0587,
+      "loss/crossentropy": 2.2238826751708984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20041973888874054,
+      "step": 19572
+    },
+    {
+      "epoch": 0.39148,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0037923177083333333,
+      "learning_rate": 0.0001,
+      "loss": 4.0226,
+      "loss/crossentropy": 1.8374757170677185,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18086174875497818,
+      "step": 19574
+    },
+    {
+      "epoch": 0.39152,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.00635986328125,
+      "learning_rate": 0.0001,
+      "loss": 4.2903,
+      "loss/crossentropy": 2.0642590522766113,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19544780254364014,
+      "step": 19576
+    },
+    {
+      "epoch": 0.39156,
+      "grad_norm": 2.421875,
+      "grad_norm_var": 0.020493316650390624,
+      "learning_rate": 0.0001,
+      "loss": 4.1231,
+      "loss/crossentropy": 2.090702533721924,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19859656691551208,
+      "step": 19578
+    },
+    {
+      "epoch": 0.3916,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.023538970947265626,
+      "learning_rate": 0.0001,
+      "loss": 3.9075,
+      "loss/crossentropy": 1.8087154030799866,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1918642893433571,
+      "step": 19580
+    },
+    {
+      "epoch": 0.39164,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.023527018229166665,
+      "learning_rate": 0.0001,
+      "loss": 3.8869,
+      "loss/crossentropy": 2.0180618166923523,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19190336763858795,
+      "step": 19582
+    },
+    {
+      "epoch": 0.39168,
+      "grad_norm": 2.171875,
+      "grad_norm_var": 0.025886027018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2298,
+      "loss/crossentropy": 1.980249285697937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20591172575950623,
+      "step": 19584
+    },
+    {
+      "epoch": 0.39172,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.026374308268229167,
+      "learning_rate": 0.0001,
+      "loss": 3.9883,
+      "loss/crossentropy": 2.1208410263061523,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20714347064495087,
+      "step": 19586
+    },
+    {
+      "epoch": 0.39176,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.024179840087890626,
+      "learning_rate": 0.0001,
+      "loss": 4.3101,
+      "loss/crossentropy": 2.351140856742859,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26494763791561127,
+      "step": 19588
+    },
+    {
+      "epoch": 0.3918,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.024448394775390625,
+      "learning_rate": 0.0001,
+      "loss": 4.1259,
+      "loss/crossentropy": 1.9821222424507141,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2062402218580246,
+      "step": 19590
+    },
+    {
+      "epoch": 0.39184,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.02462158203125,
+      "learning_rate": 0.0001,
+      "loss": 4.1373,
+      "loss/crossentropy": 2.363589644432068,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22435829043388367,
+      "step": 19592
+    },
+    {
+      "epoch": 0.39188,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.013963826497395833,
+      "learning_rate": 0.0001,
+      "loss": 4.1845,
+      "loss/crossentropy": 1.925455391407013,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17338209599256516,
+      "step": 19594
+    },
+    {
+      "epoch": 0.39192,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.012254842122395833,
+      "learning_rate": 0.0001,
+      "loss": 3.9777,
+      "loss/crossentropy": 2.176175117492676,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21482165157794952,
+      "step": 19596
+    },
+    {
+      "epoch": 0.39196,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.014330037434895833,
+      "learning_rate": 0.0001,
+      "loss": 3.9266,
+      "loss/crossentropy": 2.0013960003852844,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20629072189331055,
+      "step": 19598
+    },
+    {
+      "epoch": 0.392,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.011797841389973958,
+      "learning_rate": 0.0001,
+      "loss": 3.9402,
+      "loss/crossentropy": 2.066355049610138,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22155777364969254,
+      "step": 19600
+    },
+    {
+      "epoch": 0.39204,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.010809071858723958,
+      "learning_rate": 0.0001,
+      "loss": 4.0319,
+      "loss/crossentropy": 2.0513144731521606,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19442218542099,
+      "step": 19602
+    },
+    {
+      "epoch": 0.39208,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.010896809895833333,
+      "learning_rate": 0.0001,
+      "loss": 4.2601,
+      "loss/crossentropy": 1.9635959267616272,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20168906450271606,
+      "step": 19604
+    },
+    {
+      "epoch": 0.39212,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.009025065104166667,
+      "learning_rate": 0.0001,
+      "loss": 3.9406,
+      "loss/crossentropy": 1.9395795464515686,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17990338802337646,
+      "step": 19606
+    },
+    {
+      "epoch": 0.39216,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.011004384358723958,
+      "learning_rate": 0.0001,
+      "loss": 4.1936,
+      "loss/crossentropy": 2.111960232257843,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20662778615951538,
+      "step": 19608
+    },
+    {
+      "epoch": 0.3922,
+      "grad_norm": 1.8125,
+      "grad_norm_var": 0.011466471354166667,
+      "learning_rate": 0.0001,
+      "loss": 3.9238,
+      "loss/crossentropy": 1.855182707309723,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19665290415287018,
+      "step": 19610
+    },
+    {
+      "epoch": 0.39224,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0156494140625,
+      "learning_rate": 0.0001,
+      "loss": 4.262,
+      "loss/crossentropy": 2.1587076783180237,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20080996304750443,
+      "step": 19612
+    },
+    {
+      "epoch": 0.39228,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.01597874959309896,
+      "learning_rate": 0.0001,
+      "loss": 4.1227,
+      "loss/crossentropy": 2.130104422569275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19484283030033112,
+      "step": 19614
+    },
+    {
+      "epoch": 0.39232,
+      "grad_norm": 1.75,
+      "grad_norm_var": 0.017992146809895835,
+      "learning_rate": 0.0001,
+      "loss": 3.9046,
+      "loss/crossentropy": 1.6983461380004883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18040503561496735,
+      "step": 19616
+    },
+    {
+      "epoch": 0.39236,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.017756144205729168,
+      "learning_rate": 0.0001,
+      "loss": 4.0141,
+      "loss/crossentropy": 2.0672999024391174,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20455461740493774,
+      "step": 19618
+    },
+    {
+      "epoch": 0.3924,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.018202463785807293,
+      "learning_rate": 0.0001,
+      "loss": 3.83,
+      "loss/crossentropy": 2.0494508743286133,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18307264149188995,
+      "step": 19620
+    },
+    {
+      "epoch": 0.39244,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.018277740478515624,
+      "learning_rate": 0.0001,
+      "loss": 3.9901,
+      "loss/crossentropy": 1.683717966079712,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16951359808444977,
+      "step": 19622
+    },
+    {
+      "epoch": 0.39248,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.015197499593098959,
+      "learning_rate": 0.0001,
+      "loss": 3.8383,
+      "loss/crossentropy": 2.62760066986084,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22649705410003662,
+      "step": 19624
+    },
+    {
+      "epoch": 0.39252,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 0.024072011311848957,
+      "learning_rate": 0.0001,
+      "loss": 4.19,
+      "loss/crossentropy": 1.857836663722992,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1827384978532791,
+      "step": 19626
+    },
+    {
+      "epoch": 0.39256,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.016108957926432292,
+      "learning_rate": 0.0001,
+      "loss": 3.7635,
+      "loss/crossentropy": 2.1203905940055847,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20033711194992065,
+      "step": 19628
+    },
+    {
+      "epoch": 0.3926,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.01721165974934896,
+      "learning_rate": 0.0001,
+      "loss": 4.4474,
+      "loss/crossentropy": 2.378189444541931,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20892338454723358,
+      "step": 19630
+    },
+    {
+      "epoch": 0.39264,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.015819295247395834,
+      "learning_rate": 0.0001,
+      "loss": 4.0897,
+      "loss/crossentropy": 1.896558940410614,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18873175233602524,
+      "step": 19632
+    },
+    {
+      "epoch": 0.39268,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.020798492431640624,
+      "learning_rate": 0.0001,
+      "loss": 4.3611,
+      "loss/crossentropy": 2.3503568172454834,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23128806054592133,
+      "step": 19634
+    },
+    {
+      "epoch": 0.39272,
+      "grad_norm": 1.7578125,
+      "grad_norm_var": 0.021345011393229165,
+      "learning_rate": 0.0001,
+      "loss": 4.0515,
+      "loss/crossentropy": 2.081954002380371,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20059917122125626,
+      "step": 19636
+    },
+    {
+      "epoch": 0.39276,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.021512858072916665,
+      "learning_rate": 0.0001,
+      "loss": 4.0986,
+      "loss/crossentropy": 1.9573910236358643,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1903771460056305,
+      "step": 19638
+    },
+    {
+      "epoch": 0.3928,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.022025299072265626,
+      "learning_rate": 0.0001,
+      "loss": 3.9333,
+      "loss/crossentropy": 2.098384141921997,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19905343651771545,
+      "step": 19640
+    },
+    {
+      "epoch": 0.39284,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.015892537434895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0854,
+      "loss/crossentropy": 1.8510947227478027,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18955568969249725,
+      "step": 19642
+    },
+    {
+      "epoch": 0.39288,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.016440582275390626,
+      "learning_rate": 0.0001,
+      "loss": 4.2948,
+      "loss/crossentropy": 2.2839618921279907,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23230211436748505,
+      "step": 19644
+    },
+    {
+      "epoch": 0.39292,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.014778391520182291,
+      "learning_rate": 0.0001,
+      "loss": 4.0112,
+      "loss/crossentropy": 1.8055492639541626,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20515839755535126,
+      "step": 19646
+    },
+    {
+      "epoch": 0.39296,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.013348134358723958,
+      "learning_rate": 0.0001,
+      "loss": 4.0556,
+      "loss/crossentropy": 1.6756115555763245,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17216521501541138,
+      "step": 19648
+    },
+    {
+      "epoch": 0.393,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.007287343343098958,
+      "learning_rate": 0.0001,
+      "loss": 4.4001,
+      "loss/crossentropy": 2.3247755765914917,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21122244000434875,
+      "step": 19650
+    },
+    {
+      "epoch": 0.39304,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.004044596354166667,
+      "learning_rate": 0.0001,
+      "loss": 3.895,
+      "loss/crossentropy": 1.9654970169067383,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20018797367811203,
+      "step": 19652
+    },
+    {
+      "epoch": 0.39308,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.0029042561848958335,
+      "learning_rate": 0.0001,
+      "loss": 3.9018,
+      "loss/crossentropy": 1.9368168115615845,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19478464126586914,
+      "step": 19654
+    },
+    {
+      "epoch": 0.39312,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0023671468098958332,
+      "learning_rate": 0.0001,
+      "loss": 3.7869,
+      "loss/crossentropy": 2.099206328392029,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20060646533966064,
+      "step": 19656
+    },
+    {
+      "epoch": 0.39316,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.002561187744140625,
+      "learning_rate": 0.0001,
+      "loss": 4.3438,
+      "loss/crossentropy": 2.1585338711738586,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20852376520633698,
+      "step": 19658
+    },
+    {
+      "epoch": 0.3932,
+      "grad_norm": 1.796875,
+      "grad_norm_var": 0.0044830322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0533,
+      "loss/crossentropy": 2.285220742225647,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1983182057738304,
+      "step": 19660
+    },
+    {
+      "epoch": 0.39324,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.0047190348307291664,
+      "learning_rate": 0.0001,
+      "loss": 4.0046,
+      "loss/crossentropy": 1.955579936504364,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21165720373392105,
+      "step": 19662
+    },
+    {
+      "epoch": 0.39328,
+      "grad_norm": 1.8046875,
+      "grad_norm_var": 0.00645751953125,
+      "learning_rate": 0.0001,
+      "loss": 3.823,
+      "loss/crossentropy": 1.9925233721733093,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18789846450090408,
+      "step": 19664
+    },
+    {
+      "epoch": 0.39332,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.006490071614583333,
+      "learning_rate": 0.0001,
+      "loss": 4.0123,
+      "loss/crossentropy": 1.9267281293869019,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19545376300811768,
+      "step": 19666
+    },
+    {
+      "epoch": 0.39336,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.006493123372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.0401,
+      "loss/crossentropy": 2.179477632045746,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19430068880319595,
+      "step": 19668
+    },
+    {
+      "epoch": 0.3934,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.007228342692057291,
+      "learning_rate": 0.0001,
+      "loss": 4.1508,
+      "loss/crossentropy": 2.1411179900169373,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2276124805212021,
+      "step": 19670
+    },
+    {
+      "epoch": 0.39344,
+      "grad_norm": 2.3125,
+      "grad_norm_var": 0.014438629150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.2689,
+      "loss/crossentropy": 2.085427463054657,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21856296062469482,
+      "step": 19672
+    },
+    {
+      "epoch": 0.39348,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.014495595296223959,
+      "learning_rate": 0.0001,
+      "loss": 4.0302,
+      "loss/crossentropy": 1.7399682402610779,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16918571293354034,
+      "step": 19674
+    },
+    {
+      "epoch": 0.39352,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.011356608072916666,
+      "learning_rate": 0.0001,
+      "loss": 4.0404,
+      "loss/crossentropy": 1.8636209964752197,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17654836922883987,
+      "step": 19676
+    },
+    {
+      "epoch": 0.39356,
+      "grad_norm": 1.7421875,
+      "grad_norm_var": 0.015290323893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0157,
+      "loss/crossentropy": 2.143825590610504,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2125091552734375,
+      "step": 19678
+    },
+    {
+      "epoch": 0.3936,
+      "grad_norm": 1.6953125,
+      "grad_norm_var": 0.01873753865559896,
+      "learning_rate": 0.0001,
+      "loss": 3.7692,
+      "loss/crossentropy": 1.8222747445106506,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17713025212287903,
+      "step": 19680
+    },
+    {
+      "epoch": 0.39364,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.01969172159830729,
+      "learning_rate": 0.0001,
+      "loss": 3.8254,
+      "loss/crossentropy": 1.7044820189476013,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17669418454170227,
+      "step": 19682
+    },
+    {
+      "epoch": 0.39368,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.020411936442057292,
+      "learning_rate": 0.0001,
+      "loss": 4.0987,
+      "loss/crossentropy": 2.144485831260681,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20580272376537323,
+      "step": 19684
+    },
+    {
+      "epoch": 0.39372,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.019774373372395834,
+      "learning_rate": 0.0001,
+      "loss": 4.0994,
+      "loss/crossentropy": 2.17536997795105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21704821288585663,
+      "step": 19686
+    },
+    {
+      "epoch": 0.39376,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.012963612874348959,
+      "learning_rate": 0.0001,
+      "loss": 4.2592,
+      "loss/crossentropy": 2.1875303983688354,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1996757537126541,
+      "step": 19688
+    },
+    {
+      "epoch": 0.3938,
+      "grad_norm": 2.125,
+      "grad_norm_var": 0.014121246337890626,
+      "learning_rate": 0.0001,
+      "loss": 4.1943,
+      "loss/crossentropy": 2.270558476448059,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23406681418418884,
+      "step": 19690
+    },
+    {
+      "epoch": 0.39384,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.015860748291015626,
+      "learning_rate": 0.0001,
+      "loss": 4.2097,
+      "loss/crossentropy": 2.0614060163497925,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21256985515356064,
+      "step": 19692
+    },
+    {
+      "epoch": 0.39388,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.015091705322265624,
+      "learning_rate": 0.0001,
+      "loss": 4.042,
+      "loss/crossentropy": 2.1036725640296936,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19555094093084335,
+      "step": 19694
+    },
+    {
+      "epoch": 0.39392,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.010396321614583334,
+      "learning_rate": 0.0001,
+      "loss": 3.7702,
+      "loss/crossentropy": 1.8713775277137756,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1973343789577484,
+      "step": 19696
+    },
+    {
+      "epoch": 0.39396,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.009411366780598958,
+      "learning_rate": 0.0001,
+      "loss": 3.9581,
+      "loss/crossentropy": 1.9986143708229065,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1925104334950447,
+      "step": 19698
+    },
+    {
+      "epoch": 0.394,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0090484619140625,
+      "learning_rate": 0.0001,
+      "loss": 3.8134,
+      "loss/crossentropy": 1.8336694836616516,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18453969806432724,
+      "step": 19700
+    },
+    {
+      "epoch": 0.39404,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.01065673828125,
+      "learning_rate": 0.0001,
+      "loss": 3.9279,
+      "loss/crossentropy": 1.8948233723640442,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17115242034196854,
+      "step": 19702
+    },
+    {
+      "epoch": 0.39408,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.008958943684895833,
+      "learning_rate": 0.0001,
+      "loss": 4.244,
+      "loss/crossentropy": 2.5506935119628906,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22056522965431213,
+      "step": 19704
+    },
+    {
+      "epoch": 0.39412,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.02890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1832,
+      "loss/crossentropy": 2.127421021461487,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20175430178642273,
+      "step": 19706
+    },
+    {
+      "epoch": 0.39416,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.027197265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0251,
+      "loss/crossentropy": 2.246693968772888,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22178302705287933,
+      "step": 19708
+    },
+    {
+      "epoch": 0.3942,
+      "grad_norm": 1.6953125,
+      "grad_norm_var": 0.03328221638997396,
+      "learning_rate": 0.0001,
+      "loss": 3.9336,
+      "loss/crossentropy": 1.970005750656128,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18967190384864807,
+      "step": 19710
+    },
+    {
+      "epoch": 0.39424,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.032956695556640624,
+      "learning_rate": 0.0001,
+      "loss": 4.1035,
+      "loss/crossentropy": 1.944337785243988,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1952769234776497,
+      "step": 19712
+    },
+    {
+      "epoch": 0.39428,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.0328277587890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2289,
+      "loss/crossentropy": 2.291213870048523,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2077646553516388,
+      "step": 19714
+    },
+    {
+      "epoch": 0.39432,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.03361790974934896,
+      "learning_rate": 0.0001,
+      "loss": 4.0433,
+      "loss/crossentropy": 2.404030203819275,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2200116366147995,
+      "step": 19716
+    },
+    {
+      "epoch": 0.39436,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.032364908854166666,
+      "learning_rate": 0.0001,
+      "loss": 3.8609,
+      "loss/crossentropy": 1.9242625832557678,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19318503141403198,
+      "step": 19718
+    },
+    {
+      "epoch": 0.3944,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.03240966796875,
+      "learning_rate": 0.0001,
+      "loss": 3.9789,
+      "loss/crossentropy": 2.319575071334839,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21668671071529388,
+      "step": 19720
+    },
+    {
+      "epoch": 0.39444,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.012287394205729166,
+      "learning_rate": 0.0001,
+      "loss": 4.2219,
+      "loss/crossentropy": 2.0032835006713867,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19786667823791504,
+      "step": 19722
+    },
+    {
+      "epoch": 0.39448,
+      "grad_norm": 1.7734375,
+      "grad_norm_var": 0.0146636962890625,
+      "learning_rate": 0.0001,
+      "loss": 3.9509,
+      "loss/crossentropy": 1.9524562358856201,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18965402245521545,
+      "step": 19724
+    },
+    {
+      "epoch": 0.39452,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.007521311442057292,
+      "learning_rate": 0.0001,
+      "loss": 3.5677,
+      "loss/crossentropy": 1.5887231826782227,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1512351706624031,
+      "step": 19726
+    },
+    {
+      "epoch": 0.39456,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.008576456705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.1379,
+      "loss/crossentropy": 2.107685923576355,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18645642697811127,
+      "step": 19728
+    },
+    {
+      "epoch": 0.3946,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.009028879801432292,
+      "learning_rate": 0.0001,
+      "loss": 3.9594,
+      "loss/crossentropy": 2.3551766872406006,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2207053080201149,
+      "step": 19730
+    },
+    {
+      "epoch": 0.39464,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.010847727457682291,
+      "learning_rate": 0.0001,
+      "loss": 4.3542,
+      "loss/crossentropy": 2.1983554363250732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23001667857170105,
+      "step": 19732
+    },
+    {
+      "epoch": 0.39468,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009639485677083334,
+      "learning_rate": 0.0001,
+      "loss": 4.3158,
+      "loss/crossentropy": 2.0236815214157104,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19230765849351883,
+      "step": 19734
+    },
+    {
+      "epoch": 0.39472,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.010529581705729167,
+      "learning_rate": 0.0001,
+      "loss": 4.0511,
+      "loss/crossentropy": 2.2027645111083984,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1864710971713066,
+      "step": 19736
+    },
+    {
+      "epoch": 0.39476,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.013152821858723959,
+      "learning_rate": 0.0001,
+      "loss": 4.008,
+      "loss/crossentropy": 1.921549379825592,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21788031607866287,
+      "step": 19738
+    },
+    {
+      "epoch": 0.3948,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.010416666666666666,
+      "learning_rate": 0.0001,
+      "loss": 3.8353,
+      "loss/crossentropy": 2.121790587902069,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21380367130041122,
+      "step": 19740
+    },
+    {
+      "epoch": 0.39484,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.008217112223307291,
+      "learning_rate": 0.0001,
+      "loss": 3.9975,
+      "loss/crossentropy": 1.9595564007759094,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17919503152370453,
+      "step": 19742
+    },
+    {
+      "epoch": 0.39488,
+      "grad_norm": 2.140625,
+      "grad_norm_var": 0.010993448893229167,
+      "learning_rate": 0.0001,
+      "loss": 3.9935,
+      "loss/crossentropy": 2.155984342098236,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2002984657883644,
+      "step": 19744
+    },
+    {
+      "epoch": 0.39492,
+      "grad_norm": 1.7890625,
+      "grad_norm_var": 0.012589518229166667,
+      "learning_rate": 0.0001,
+      "loss": 3.875,
+      "loss/crossentropy": 1.766005277633667,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1717890352010727,
+      "step": 19746
+    },
+    {
+      "epoch": 0.39496,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.010424550374348958,
+      "learning_rate": 0.0001,
+      "loss": 3.9178,
+      "loss/crossentropy": 1.8891428112983704,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1965286061167717,
+      "step": 19748
+    },
+    {
+      "epoch": 0.395,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.010343170166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.3124,
+      "loss/crossentropy": 1.9521282315254211,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21058151125907898,
+      "step": 19750
+    },
+    {
+      "epoch": 0.39504,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.014711252848307292,
+      "learning_rate": 0.0001,
+      "loss": 4.2242,
+      "loss/crossentropy": 2.3142699003219604,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24145027250051498,
+      "step": 19752
+    },
+    {
+      "epoch": 0.39508,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.011777496337890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1059,
+      "loss/crossentropy": 2.0375224351882935,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19370558112859726,
+      "step": 19754
+    },
+    {
+      "epoch": 0.39512,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.011962890625,
+      "learning_rate": 0.0001,
+      "loss": 4.2367,
+      "loss/crossentropy": 2.193490743637085,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2012534961104393,
+      "step": 19756
+    },
+    {
+      "epoch": 0.39516,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.011993153889973959,
+      "learning_rate": 0.0001,
+      "loss": 4.0781,
+      "loss/crossentropy": 2.2397992610931396,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19738413393497467,
+      "step": 19758
+    },
+    {
+      "epoch": 0.3952,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.010223134358723959,
+      "learning_rate": 0.0001,
+      "loss": 3.9153,
+      "loss/crossentropy": 1.7602161169052124,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18599014729261398,
+      "step": 19760
+    },
+    {
+      "epoch": 0.39524,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.008640289306640625,
+      "learning_rate": 0.0001,
+      "loss": 4.182,
+      "loss/crossentropy": 2.536779046058655,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2301289290189743,
+      "step": 19762
+    },
+    {
+      "epoch": 0.39528,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.008965810139973959,
+      "learning_rate": 0.0001,
+      "loss": 3.9999,
+      "loss/crossentropy": 2.3237764835357666,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21476060152053833,
+      "step": 19764
+    },
+    {
+      "epoch": 0.39532,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.011107381184895833,
+      "learning_rate": 0.0001,
+      "loss": 4.2102,
+      "loss/crossentropy": 2.370723605155945,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24277979880571365,
+      "step": 19766
+    },
+    {
+      "epoch": 0.39536,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.006917063395182292,
+      "learning_rate": 0.0001,
+      "loss": 4.0838,
+      "loss/crossentropy": 1.6399320363998413,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1740519106388092,
+      "step": 19768
+    },
+    {
+      "epoch": 0.3954,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.02394383748372396,
+      "learning_rate": 0.0001,
+      "loss": 3.6633,
+      "loss/crossentropy": 1.8633801341056824,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1810051053762436,
+      "step": 19770
+    },
+    {
+      "epoch": 0.39544,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.023583984375,
+      "learning_rate": 0.0001,
+      "loss": 4.1102,
+      "loss/crossentropy": 2.318315625190735,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21214767545461655,
+      "step": 19772
+    },
+    {
+      "epoch": 0.39548,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.026387532552083332,
+      "learning_rate": 0.0001,
+      "loss": 4.1232,
+      "loss/crossentropy": 2.2803520560264587,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21136894822120667,
+      "step": 19774
+    },
+    {
+      "epoch": 0.39552,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.026387532552083332,
+      "learning_rate": 0.0001,
+      "loss": 3.9034,
+      "loss/crossentropy": 2.1529780626296997,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18172463029623032,
+      "step": 19776
+    },
+    {
+      "epoch": 0.39556,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.025585683186848958,
+      "learning_rate": 0.0001,
+      "loss": 4.2851,
+      "loss/crossentropy": 2.1246083974838257,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2181987464427948,
+      "step": 19778
+    },
+    {
+      "epoch": 0.3956,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.029515584309895832,
+      "learning_rate": 0.0001,
+      "loss": 4.0097,
+      "loss/crossentropy": 2.027298629283905,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1877325028181076,
+      "step": 19780
+    },
+    {
+      "epoch": 0.39564,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.0288482666015625,
+      "learning_rate": 0.0001,
+      "loss": 4.2326,
+      "loss/crossentropy": 1.9338072538375854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1746418997645378,
+      "step": 19782
+    },
+    {
+      "epoch": 0.39568,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.03104222615559896,
+      "learning_rate": 0.0001,
+      "loss": 4.2551,
+      "loss/crossentropy": 2.050130307674408,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2011866271495819,
+      "step": 19784
+    },
+    {
+      "epoch": 0.39572,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.014720662434895834,
+      "learning_rate": 0.0001,
+      "loss": 4.3227,
+      "loss/crossentropy": 2.48315691947937,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22591856867074966,
+      "step": 19786
+    },
+    {
+      "epoch": 0.39576,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.021019490559895833,
+      "learning_rate": 0.0001,
+      "loss": 4.3671,
+      "loss/crossentropy": 2.578279137611389,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2216324657201767,
+      "step": 19788
+    },
+    {
+      "epoch": 0.3958,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.018192291259765625,
+      "learning_rate": 0.0001,
+      "loss": 4.259,
+      "loss/crossentropy": 1.7862395644187927,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17009516060352325,
+      "step": 19790
+    },
+    {
+      "epoch": 0.39584,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.017032877604166666,
+      "learning_rate": 0.0001,
+      "loss": 3.8506,
+      "loss/crossentropy": 2.042613208293915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1844482272863388,
+      "step": 19792
+    },
+    {
+      "epoch": 0.39588,
+      "grad_norm": 1.7890625,
+      "grad_norm_var": 0.020926666259765626,
+      "learning_rate": 0.0001,
+      "loss": 4.1919,
+      "loss/crossentropy": 2.317251443862915,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2135745733976364,
+      "step": 19794
+    },
+    {
+      "epoch": 0.39592,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.01964111328125,
+      "learning_rate": 0.0001,
+      "loss": 4.0061,
+      "loss/crossentropy": 1.849199891090393,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20549577474594116,
+      "step": 19796
+    },
+    {
+      "epoch": 0.39596,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.0199859619140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0551,
+      "loss/crossentropy": 2.1876507997512817,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20098386704921722,
+      "step": 19798
+    },
+    {
+      "epoch": 0.396,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.01718317667643229,
+      "learning_rate": 0.0001,
+      "loss": 4.154,
+      "loss/crossentropy": 1.9740530848503113,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19494586437940598,
+      "step": 19800
+    },
+    {
+      "epoch": 0.39604,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.017577107747395834,
+      "learning_rate": 0.0001,
+      "loss": 4.1405,
+      "loss/crossentropy": 2.208943724632263,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22833283245563507,
+      "step": 19802
+    },
+    {
+      "epoch": 0.39608,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.009488677978515625,
+      "learning_rate": 0.0001,
+      "loss": 3.9082,
+      "loss/crossentropy": 2.024399518966675,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19786083698272705,
+      "step": 19804
+    },
+    {
+      "epoch": 0.39612,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.009356435139973958,
+      "learning_rate": 0.0001,
+      "loss": 4.1475,
+      "loss/crossentropy": 2.1283441185951233,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21942409127950668,
+      "step": 19806
+    },
+    {
+      "epoch": 0.39616,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.01080322265625,
+      "learning_rate": 0.0001,
+      "loss": 4.0101,
+      "loss/crossentropy": 2.3675626516342163,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22466859221458435,
+      "step": 19808
+    },
+    {
+      "epoch": 0.3962,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0082672119140625,
+      "learning_rate": 0.0001,
+      "loss": 3.8682,
+      "loss/crossentropy": 1.886322796344757,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1904991939663887,
+      "step": 19810
+    },
+    {
+      "epoch": 0.39624,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.005012003580729166,
+      "learning_rate": 0.0001,
+      "loss": 3.9071,
+      "loss/crossentropy": 2.098154127597809,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2065964937210083,
+      "step": 19812
+    },
+    {
+      "epoch": 0.39628,
+      "grad_norm": 1.8515625,
+      "grad_norm_var": 0.006788889567057292,
+      "learning_rate": 0.0001,
+      "loss": 3.8447,
+      "loss/crossentropy": 2.0527199506759644,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19717100262641907,
+      "step": 19814
+    },
+    {
+      "epoch": 0.39632,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 0.011244455973307291,
+      "learning_rate": 0.0001,
+      "loss": 4.2692,
+      "loss/crossentropy": 2.0650912523269653,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2021411582827568,
+      "step": 19816
+    },
+    {
+      "epoch": 0.39636,
+      "grad_norm": 2.34375,
+      "grad_norm_var": 0.018700154622395833,
+      "learning_rate": 0.0001,
+      "loss": 4.3611,
+      "loss/crossentropy": 2.2239702939987183,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19744005054235458,
+      "step": 19818
+    },
+    {
+      "epoch": 0.3964,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.020369211832682293,
+      "learning_rate": 0.0001,
+      "loss": 3.9029,
+      "loss/crossentropy": 2.123336434364319,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18976306170225143,
+      "step": 19820
+    },
+    {
+      "epoch": 0.39644,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.02154515584309896,
+      "learning_rate": 0.0001,
+      "loss": 3.9966,
+      "loss/crossentropy": 2.121657133102417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2159118503332138,
+      "step": 19822
+    },
+    {
+      "epoch": 0.39648,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.0185455322265625,
+      "learning_rate": 0.0001,
+      "loss": 3.7621,
+      "loss/crossentropy": 1.7323416471481323,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18072299659252167,
+      "step": 19824
+    },
+    {
+      "epoch": 0.39652,
+      "grad_norm": 2.21875,
+      "grad_norm_var": 0.020243072509765626,
+      "learning_rate": 0.0001,
+      "loss": 4.0183,
+      "loss/crossentropy": 2.181841015815735,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21452204138040543,
+      "step": 19826
+    },
+    {
+      "epoch": 0.39656,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.02104670206705729,
+      "learning_rate": 0.0001,
+      "loss": 4.1181,
+      "loss/crossentropy": 1.9022215008735657,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18924792110919952,
+      "step": 19828
+    },
+    {
+      "epoch": 0.3966,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.018317667643229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2505,
+      "loss/crossentropy": 2.1240022778511047,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18618234246969223,
+      "step": 19830
+    },
+    {
+      "epoch": 0.39664,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.015710194905598957,
+      "learning_rate": 0.0001,
+      "loss": 3.8611,
+      "loss/crossentropy": 1.5959683060646057,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17317308485507965,
+      "step": 19832
+    },
+    {
+      "epoch": 0.39668,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.007682291666666666,
+      "learning_rate": 0.0001,
+      "loss": 4.1281,
+      "loss/crossentropy": 2.1812866926193237,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.192174032330513,
+      "step": 19834
+    },
+    {
+      "epoch": 0.39672,
+      "grad_norm": 1.953125,
+      "grad_norm_var": 0.010651652018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.188,
+      "loss/crossentropy": 1.8964568972587585,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20922592282295227,
+      "step": 19836
+    },
+    {
+      "epoch": 0.39676,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.012013498942057292,
+      "learning_rate": 0.0001,
+      "loss": 4.1369,
+      "loss/crossentropy": 2.0822665691375732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18019618093967438,
+      "step": 19838
+    },
+    {
+      "epoch": 0.3968,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.010985310872395833,
+      "learning_rate": 0.0001,
+      "loss": 4.0271,
+      "loss/crossentropy": 2.116270899772644,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21604043990373611,
+      "step": 19840
+    },
+    {
+      "epoch": 0.39684,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.007759602864583334,
+      "learning_rate": 0.0001,
+      "loss": 3.9132,
+      "loss/crossentropy": 1.7761988639831543,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20087965577840805,
+      "step": 19842
+    },
+    {
+      "epoch": 0.39688,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.008649698893229167,
+      "learning_rate": 0.0001,
+      "loss": 4.0187,
+      "loss/crossentropy": 1.6091360449790955,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18282026052474976,
+      "step": 19844
+    },
+    {
+      "epoch": 0.39692,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.009547678629557292,
+      "learning_rate": 0.0001,
+      "loss": 4.4441,
+      "loss/crossentropy": 2.6394678354263306,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20579150319099426,
+      "step": 19846
+    },
+    {
+      "epoch": 0.39696,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.010011545817057292,
+      "learning_rate": 0.0001,
+      "loss": 4.1778,
+      "loss/crossentropy": 2.0911704897880554,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20075388252735138,
+      "step": 19848
+    },
+    {
+      "epoch": 0.397,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.010139719645182291,
+      "learning_rate": 0.0001,
+      "loss": 4.3458,
+      "loss/crossentropy": 2.1299456357955933,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2067631408572197,
+      "step": 19850
+    },
+    {
+      "epoch": 0.39704,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.004937489827473958,
+      "learning_rate": 0.0001,
+      "loss": 3.9037,
+      "loss/crossentropy": 1.9630563855171204,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.187811940908432,
+      "step": 19852
+    },
+    {
+      "epoch": 0.39708,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.004808553059895833,
+      "learning_rate": 0.0001,
+      "loss": 4.1681,
+      "loss/crossentropy": 2.1914453506469727,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2064526304602623,
+      "step": 19854
+    },
+    {
+      "epoch": 0.39712,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.012412261962890626,
+      "learning_rate": 0.0001,
+      "loss": 4.1404,
+      "loss/crossentropy": 2.3003474473953247,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20841488242149353,
+      "step": 19856
+    },
+    {
+      "epoch": 0.39716,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.01682306925455729,
+      "learning_rate": 0.0001,
+      "loss": 3.96,
+      "loss/crossentropy": 2.1625255346298218,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20437289774417877,
+      "step": 19858
+    },
+    {
+      "epoch": 0.3972,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.015868123372395834,
+      "learning_rate": 0.0001,
+      "loss": 3.8594,
+      "loss/crossentropy": 1.6742416620254517,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16787201166152954,
+      "step": 19860
+    },
+    {
+      "epoch": 0.39724,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.01546630859375,
+      "learning_rate": 0.0001,
+      "loss": 3.8889,
+      "loss/crossentropy": 2.0682146549224854,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1715894490480423,
+      "step": 19862
+    },
+    {
+      "epoch": 0.39728,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.017071278889973958,
+      "learning_rate": 0.0001,
+      "loss": 4.0319,
+      "loss/crossentropy": 1.99592924118042,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19570383429527283,
+      "step": 19864
+    },
+    {
+      "epoch": 0.39732,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.01967137654622396,
+      "learning_rate": 0.0001,
+      "loss": 4.0123,
+      "loss/crossentropy": 2.086844265460968,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2123936414718628,
+      "step": 19866
+    },
+    {
+      "epoch": 0.39736,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.017256673177083334,
+      "learning_rate": 0.0001,
+      "loss": 4.2245,
+      "loss/crossentropy": 1.7985658645629883,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18117651343345642,
+      "step": 19868
+    },
+    {
+      "epoch": 0.3974,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.018949381510416665,
+      "learning_rate": 0.0001,
+      "loss": 3.8449,
+      "loss/crossentropy": 1.9804646372795105,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18682067096233368,
+      "step": 19870
+    },
+    {
+      "epoch": 0.39744,
+      "grad_norm": 1.796875,
+      "grad_norm_var": 0.01546630859375,
+      "learning_rate": 0.0001,
+      "loss": 4.0692,
+      "loss/crossentropy": 1.9027678966522217,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18195898830890656,
+      "step": 19872
+    },
+    {
+      "epoch": 0.39748,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.01660944620768229,
+      "learning_rate": 0.0001,
+      "loss": 4.1191,
+      "loss/crossentropy": 2.3370686769485474,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2376203015446663,
+      "step": 19874
+    },
+    {
+      "epoch": 0.39752,
+      "grad_norm": 1.890625,
+      "grad_norm_var": 0.018363189697265626,
+      "learning_rate": 0.0001,
+      "loss": 3.8472,
+      "loss/crossentropy": 1.7463279366493225,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16485755145549774,
+      "step": 19876
+    },
+    {
+      "epoch": 0.39756,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.017967732747395833,
+      "learning_rate": 0.0001,
+      "loss": 4.4039,
+      "loss/crossentropy": 2.1171544194221497,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20631500333547592,
+      "step": 19878
+    },
+    {
+      "epoch": 0.3976,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.016733551025390626,
+      "learning_rate": 0.0001,
+      "loss": 3.7242,
+      "loss/crossentropy": 2.1109176874160767,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18512246757745743,
+      "step": 19880
+    },
+    {
+      "epoch": 0.39764,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.013618977864583333,
+      "learning_rate": 0.0001,
+      "loss": 3.6843,
+      "loss/crossentropy": 1.927691638469696,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19118069112300873,
+      "step": 19882
+    },
+    {
+      "epoch": 0.39768,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.01337890625,
+      "learning_rate": 0.0001,
+      "loss": 3.9539,
+      "loss/crossentropy": 2.041896104812622,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21281379461288452,
+      "step": 19884
+    },
+    {
+      "epoch": 0.39772,
+      "grad_norm": 1.7890625,
+      "grad_norm_var": 0.013272857666015625,
+      "learning_rate": 0.0001,
+      "loss": 3.9146,
+      "loss/crossentropy": 1.8779407739639282,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.180915005505085,
+      "step": 19886
+    },
+    {
+      "epoch": 0.39776,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.019001261393229166,
+      "learning_rate": 0.0001,
+      "loss": 4.2158,
+      "loss/crossentropy": 1.9899646639823914,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19102467596530914,
+      "step": 19888
+    },
+    {
+      "epoch": 0.3978,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.012562815348307292,
+      "learning_rate": 0.0001,
+      "loss": 4.1822,
+      "loss/crossentropy": 2.3294299840927124,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20739557594060898,
+      "step": 19890
+    },
+    {
+      "epoch": 0.39784,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.011864980061848959,
+      "learning_rate": 0.0001,
+      "loss": 4.2829,
+      "loss/crossentropy": 2.2126539945602417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23066286742687225,
+      "step": 19892
+    },
+    {
+      "epoch": 0.39788,
+      "grad_norm": 2.1875,
+      "grad_norm_var": 0.016454060872395832,
+      "learning_rate": 0.0001,
+      "loss": 3.8665,
+      "loss/crossentropy": 1.9949330687522888,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19172564148902893,
+      "step": 19894
+    },
+    {
+      "epoch": 0.39792,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.02158177693684896,
+      "learning_rate": 0.0001,
+      "loss": 3.9567,
+      "loss/crossentropy": 2.0896310210227966,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19692128896713257,
+      "step": 19896
+    },
+    {
+      "epoch": 0.39796,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.020310211181640624,
+      "learning_rate": 0.0001,
+      "loss": 4.1856,
+      "loss/crossentropy": 2.191789746284485,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2073511779308319,
+      "step": 19898
+    },
+    {
+      "epoch": 0.398,
+      "grad_norm": 1.921875,
+      "grad_norm_var": 0.022415924072265624,
+      "learning_rate": 0.0001,
+      "loss": 3.9965,
+      "loss/crossentropy": 1.9925153255462646,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20123805850744247,
+      "step": 19900
+    },
+    {
+      "epoch": 0.39804,
+      "grad_norm": 1.8125,
+      "grad_norm_var": 0.021476236979166667,
+      "learning_rate": 0.0001,
+      "loss": 3.7917,
+      "loss/crossentropy": 1.8903232216835022,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18577788770198822,
+      "step": 19902
+    },
+    {
+      "epoch": 0.39808,
+      "grad_norm": 2.078125,
+      "grad_norm_var": 0.01817804972330729,
+      "learning_rate": 0.0001,
+      "loss": 4.2818,
+      "loss/crossentropy": 2.055288314819336,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19745147973299026,
+      "step": 19904
+    },
+    {
+      "epoch": 0.39812,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.018570709228515624,
+      "learning_rate": 0.0001,
+      "loss": 4.196,
+      "loss/crossentropy": 1.7709746360778809,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17147859930992126,
+      "step": 19906
+    },
+    {
+      "epoch": 0.39816,
+      "grad_norm": 2.0,
+      "grad_norm_var": 0.01843846638997396,
+      "learning_rate": 0.0001,
+      "loss": 4.182,
+      "loss/crossentropy": 2.3295364379882812,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20680927485227585,
+      "step": 19908
+    },
+    {
+      "epoch": 0.3982,
+      "grad_norm": 1.9765625,
+      "grad_norm_var": 0.014414215087890625,
+      "learning_rate": 0.0001,
+      "loss": 4.1633,
+      "loss/crossentropy": 1.9828099608421326,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.202943354845047,
+      "step": 19910
+    },
+    {
+      "epoch": 0.39824,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.009352366129557291,
+      "learning_rate": 0.0001,
+      "loss": 4.2294,
+      "loss/crossentropy": 2.140046715736389,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1967788189649582,
+      "step": 19912
+    },
+    {
+      "epoch": 0.39828,
+      "grad_norm": 1.9453125,
+      "grad_norm_var": 0.008736165364583333,
+      "learning_rate": 0.0001,
+      "loss": 4.1673,
+      "loss/crossentropy": 2.3365899324417114,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2261614426970482,
+      "step": 19914
+    },
+    {
+      "epoch": 0.39832,
+      "grad_norm": 1.9921875,
+      "grad_norm_var": 0.0075642903645833336,
+      "learning_rate": 0.0001,
+      "loss": 3.8716,
+      "loss/crossentropy": 2.101326584815979,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19668899476528168,
+      "step": 19916
+    },
+    {
+      "epoch": 0.39836,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.005037180582682292,
+      "learning_rate": 0.0001,
+      "loss": 4.1572,
+      "loss/crossentropy": 1.9804238080978394,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2117360234260559,
+      "step": 19918
+    },
+    {
+      "epoch": 0.3984,
+      "grad_norm": 1.9140625,
+      "grad_norm_var": 0.0037676493326822915,
+      "learning_rate": 0.0001,
+      "loss": 3.945,
+      "loss/crossentropy": 1.858399510383606,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20013604313135147,
+      "step": 19920
+    },
+    {
+      "epoch": 0.39844,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.005832672119140625,
+      "learning_rate": 0.0001,
+      "loss": 4.0302,
+      "loss/crossentropy": 2.087529957294464,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19001878052949905,
+      "step": 19922
+    },
+    {
+      "epoch": 0.39848,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.0068267822265625,
+      "learning_rate": 0.0001,
+      "loss": 3.9425,
+      "loss/crossentropy": 2.2521530389785767,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20182852447032928,
+      "step": 19924
+    },
+    {
+      "epoch": 0.39852,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.00784912109375,
+      "learning_rate": 0.0001,
+      "loss": 4.3362,
+      "loss/crossentropy": 2.370557188987732,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27234383672475815,
+      "step": 19926
+    },
+    {
+      "epoch": 0.39856,
+      "grad_norm": 13.0625,
+      "grad_norm_var": 7.6871192932128904,
+      "learning_rate": 0.0001,
+      "loss": 4.0383,
+      "loss/crossentropy": 2.18235445022583,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20694056898355484,
+      "step": 19928
+    },
+    {
+      "epoch": 0.3986,
+      "grad_norm": 2.125,
+      "grad_norm_var": 7.659780883789063,
+      "learning_rate": 0.0001,
+      "loss": 3.8946,
+      "loss/crossentropy": 1.9220272898674011,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21540776640176773,
+      "step": 19930
+    },
+    {
+      "epoch": 0.39864,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 7.6551513671875,
+      "learning_rate": 0.0001,
+      "loss": 4.015,
+      "loss/crossentropy": 2.064575970172882,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20172829926013947,
+      "step": 19932
+    },
+    {
+      "epoch": 0.39868,
+      "grad_norm": 2.296875,
+      "grad_norm_var": 7.620402018229167,
+      "learning_rate": 0.0001,
+      "loss": 4.5782,
+      "loss/crossentropy": 2.1824090480804443,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23097260296344757,
+      "step": 19934
+    },
+    {
+      "epoch": 0.39872,
+      "grad_norm": 2.0,
+      "grad_norm_var": 7.594489542643229,
+      "learning_rate": 0.0001,
+      "loss": 3.9227,
+      "loss/crossentropy": 2.211961567401886,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2150636538863182,
+      "step": 19936
+    },
+    {
+      "epoch": 0.39876,
+      "grad_norm": 2.4375,
+      "grad_norm_var": 7.586128743489583,
+      "learning_rate": 0.0001,
+      "loss": 4.1724,
+      "loss/crossentropy": 1.8789254426956177,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18855369836091995,
+      "step": 19938
+    },
+    {
+      "epoch": 0.3988,
+      "grad_norm": 2.03125,
+      "grad_norm_var": 7.576968383789063,
+      "learning_rate": 0.0001,
+      "loss": 4.212,
+      "loss/crossentropy": 2.322842240333557,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19432100653648376,
+      "step": 19940
+    },
+    {
+      "epoch": 0.39884,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 7.590311686197917,
+      "learning_rate": 0.0001,
+      "loss": 4.2201,
+      "loss/crossentropy": 2.1524535417556763,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19931253790855408,
+      "step": 19942
+    },
+    {
+      "epoch": 0.39888,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.0280670166015625,
+      "learning_rate": 0.0001,
+      "loss": 4.0166,
+      "loss/crossentropy": 2.0835599303245544,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18674355000257492,
+      "step": 19944
+    },
+    {
+      "epoch": 0.39892,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.026718902587890624,
+      "learning_rate": 0.0001,
+      "loss": 3.9398,
+      "loss/crossentropy": 1.8204763531684875,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1970055252313614,
+      "step": 19946
+    },
+    {
+      "epoch": 0.39896,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.02769953409830729,
+      "learning_rate": 0.0001,
+      "loss": 3.8403,
+      "loss/crossentropy": 2.1432100534439087,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1960032731294632,
+      "step": 19948
+    },
+    {
+      "epoch": 0.399,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.0227294921875,
+      "learning_rate": 0.0001,
+      "loss": 4.1183,
+      "loss/crossentropy": 1.9959533214569092,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18608924746513367,
+      "step": 19950
+    },
+    {
+      "epoch": 0.39904,
+      "grad_norm": 2.015625,
+      "grad_norm_var": 0.02399266560872396,
+      "learning_rate": 0.0001,
+      "loss": 4.0235,
+      "loss/crossentropy": 1.9874342679977417,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18600600212812424,
+      "step": 19952
+    },
+    {
+      "epoch": 0.39908,
+      "grad_norm": 1.9296875,
+      "grad_norm_var": 0.00849609375,
+      "learning_rate": 0.0001,
+      "loss": 4.2266,
+      "loss/crossentropy": 2.5082361698150635,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22386416047811508,
+      "step": 19954
+    },
+    {
+      "epoch": 0.39912,
+      "grad_norm": 1.90625,
+      "grad_norm_var": 0.008153279622395834,
+      "learning_rate": 0.0001,
+      "loss": 3.9735,
+      "loss/crossentropy": 2.1550523042678833,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2169199213385582,
+      "step": 19956
+    },
+    {
+      "epoch": 0.39916,
+      "grad_norm": 1.9609375,
+      "grad_norm_var": 0.008172353108723959,
+      "learning_rate": 0.0001,
+      "loss": 3.8834,
+      "loss/crossentropy": 2.079386830329895,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19080224633216858,
+      "step": 19958
+    },
+    {
+      "epoch": 0.3992,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.006030019124348958,
+      "learning_rate": 0.0001,
+      "loss": 3.973,
+      "loss/crossentropy": 1.8762348890304565,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1813136711716652,
+      "step": 19960
+    },
+    {
+      "epoch": 0.39924,
+      "grad_norm": 1.84375,
+      "grad_norm_var": 0.006455230712890625,
+      "learning_rate": 0.0001,
+      "loss": 4.0965,
+      "loss/crossentropy": 1.789110004901886,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17166541516780853,
+      "step": 19962
+    },
+    {
+      "epoch": 0.39928,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.009056599934895833,
+      "learning_rate": 0.0001,
+      "loss": 4.0912,
+      "loss/crossentropy": 2.0196239948272705,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20005135238170624,
+      "step": 19964
+    },
+    {
+      "epoch": 0.39932,
+      "grad_norm": 1.8984375,
+      "grad_norm_var": 0.0080718994140625,
+      "learning_rate": 0.0001,
+      "loss": 4.2787,
+      "loss/crossentropy": 2.192206025123596,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1979481726884842,
+      "step": 19966
+    },
+    {
+      "epoch": 0.39936,
+      "grad_norm": 2.0625,
+      "grad_norm_var": 0.007157135009765625,
+      "learning_rate": 0.0001,
+      "loss": 4.3155,
+      "loss/crossentropy": 2.263342499732971,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2187722995877266,
+      "step": 19968
+    },
+    {
+      "epoch": 0.3994,
+      "grad_norm": 2.109375,
+      "grad_norm_var": 0.008161417643229167,
+      "learning_rate": 0.0001,
+      "loss": 4.2727,
+      "loss/crossentropy": 2.3387625217437744,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21375280618667603,
+      "step": 19970
+    },
+    {
+      "epoch": 0.39944,
+      "grad_norm": 1.96875,
+      "grad_norm_var": 0.008733876546223958,
+      "learning_rate": 0.0001,
+      "loss": 4.0442,
+      "loss/crossentropy": 1.5837730765342712,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17303457856178284,
+      "step": 19972
+    },
+    {
+      "epoch": 0.39948,
+      "grad_norm": 1.984375,
+      "grad_norm_var": 0.008678944905598958,
+      "learning_rate": 0.0001,
+      "loss": 3.8398,
+      "loss/crossentropy": 1.9090858697891235,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1838761642575264,
+      "step": 19974
+    },
+    {
+      "epoch": 0.39952,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.008780924479166667,
+      "learning_rate": 0.0001,
+      "loss": 3.9345,
+      "loss/crossentropy": 2.107620596885681,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19132380187511444,
+      "step": 19976
+    },
+    {
+      "epoch": 0.39956,
+      "grad_norm": 1.8671875,
+      "grad_norm_var": 0.009200032552083333,
+      "learning_rate": 0.0001,
+      "loss": 4.0709,
+      "loss/crossentropy": 2.1706109046936035,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19275012612342834,
+      "step": 19978
+    },
+    {
+      "epoch": 0.3996,
+      "grad_norm": 1.8828125,
+      "grad_norm_var": 0.007356516520182292,
+      "learning_rate": 0.0001,
+      "loss": 3.8486,
+      "loss/crossentropy": 2.0092907547950745,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18047627061605453,
+      "step": 19980
+    },
+    {
+      "epoch": 0.39964,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.025349934895833332,
+      "learning_rate": 0.0001,
+      "loss": 4.2226,
+      "loss/crossentropy": 2.1216511726379395,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20591023564338684,
+      "step": 19982
+    },
+    {
+      "epoch": 0.39968,
+      "grad_norm": 2.046875,
+      "grad_norm_var": 0.02671076456705729,
+      "learning_rate": 0.0001,
+      "loss": 3.9895,
+      "loss/crossentropy": 2.3002817630767822,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20734255760908127,
+      "step": 19984
+    },
+    {
+      "epoch": 0.39972,
+      "grad_norm": 2.09375,
+      "grad_norm_var": 0.02664972941080729,
+      "learning_rate": 0.0001,
+      "loss": 4.1494,
+      "loss/crossentropy": 2.3097801208496094,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20194757729768753,
+      "step": 19986
+    },
+    {
+      "epoch": 0.39976,
+      "grad_norm": 2.15625,
+      "grad_norm_var": 0.029130045572916666,
+      "learning_rate": 0.0001,
+      "loss": 3.9706,
+      "loss/crossentropy": 2.2645580768585205,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21866093575954437,
+      "step": 19988
+    },
+    {
+      "epoch": 0.3998,
+      "grad_norm": 1.875,
+      "grad_norm_var": 0.03004150390625,
+      "learning_rate": 0.0001,
+      "loss": 4.0634,
+      "loss/crossentropy": 2.224185347557068,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19011163711547852,
+      "step": 19990
+    },
+    {
+      "epoch": 0.39984,
+      "grad_norm": 1.859375,
+      "grad_norm_var": 0.030248006184895832,
+      "learning_rate": 0.0001,
+      "loss": 3.9552,
+      "loss/crossentropy": 1.958588182926178,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18359722197055817,
+      "step": 19992
+    },
+    {
+      "epoch": 0.39988,
+      "grad_norm": 1.9375,
+      "grad_norm_var": 0.029361724853515625,
+      "learning_rate": 0.0001,
+      "loss": 4.0449,
+      "loss/crossentropy": 2.5081194639205933,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22576630860567093,
+      "step": 19994
+    },
+    {
+      "epoch": 0.39992,
+      "grad_norm": 1.8359375,
+      "grad_norm_var": 0.0306549072265625,
+      "learning_rate": 0.0001,
+      "loss": 3.8833,
+      "loss/crossentropy": 1.9870144724845886,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19262682646512985,
+      "step": 19996
+    },
+    {
+      "epoch": 0.39996,
+      "grad_norm": 1.828125,
+      "grad_norm_var": 0.0118560791015625,
+      "learning_rate": 0.0001,
+      "loss": 4.1321,
+      "loss/crossentropy": 2.227096438407898,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2056111991405487,
+      "step": 19998
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.796875,
+      "grad_norm_var": 0.012837473551432292,
+      "learning_rate": 0.0001,
+      "loss": 4.1659,
+      "loss/crossentropy": 1.8458907008171082,
+      "loss/hidden": 0.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18857256323099136,
+      "step": 20000
+    }
+  ],
+  "logging_steps": 2,
+  "max_steps": 50000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 9223372036854775807,
+  "save_steps": 10000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.08295310655488e+19,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}