diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,12018 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.025,
+  "eval_steps": 500,
+  "global_step": 1000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 2.5e-05,
+      "grad_norm": 1860.177978515625,
+      "learning_rate": 1.0000000000000002e-06,
+      "loss": 346.4354,
+      "loss/crossentropy": 2.979090690612793,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.38270998001098633,
+      "loss/reg": 343.0735778808594,
+      "step": 1
+    },
+    {
+      "epoch": 5e-05,
+      "grad_norm": 19.24078941345215,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 346.2272,
+      "loss/crossentropy": 2.8327953815460205,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.3208012580871582,
+      "loss/reg": 343.0735778808594,
+      "step": 2
+    },
+    {
+      "epoch": 7.5e-05,
+      "grad_norm": 12.568899154663086,
+      "learning_rate": 3e-06,
+      "loss": 346.5434,
+      "loss/crossentropy": 3.0858442783355713,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.38866907358169556,
+      "loss/reg": 343.0688781738281,
+      "step": 3
+    },
+    {
+      "epoch": 0.0001,
+      "grad_norm": 493.76806640625,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 346.1103,
+      "loss/crossentropy": 2.736830711364746,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.3126556873321533,
+      "loss/reg": 343.0608215332031,
+      "step": 4
+    },
+    {
+      "epoch": 0.000125,
+      "grad_norm": 17.433624267578125,
+      "learning_rate": 5e-06,
+      "loss": 346.3604,
+      "loss/crossentropy": 2.9568891525268555,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.3535460829734802,
+      "loss/reg": 343.0499267578125,
+      "step": 5
+    },
+    {
+      "epoch": 0.00015,
+      "grad_norm": 43.95212936401367,
+      "learning_rate": 6e-06,
+      "loss": 346.2806,
+      "loss/crossentropy": 2.8500430583953857,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.3941458761692047,
+      "loss/reg": 343.03643798828125,
+      "step": 6
+    },
+    {
+      "epoch": 0.000175,
+      "grad_norm": 14.617745399475098,
+      "learning_rate": 7.000000000000001e-06,
+      "loss": 346.1611,
+      "loss/crossentropy": 2.815033435821533,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.3255086839199066,
+      "loss/reg": 343.0205993652344,
+      "step": 7
+    },
+    {
+      "epoch": 0.0002,
+      "grad_norm": 94.23649597167969,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 346.2723,
+      "loss/crossentropy": 2.938474178314209,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.33254432678222656,
+      "loss/reg": 343.00128173828125,
+      "step": 8
+    },
+    {
+      "epoch": 0.000225,
+      "grad_norm": 13.357231140136719,
+      "learning_rate": 9e-06,
+      "loss": 345.8072,
+      "loss/crossentropy": 2.565714120864868,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.26191234588623047,
+      "loss/reg": 342.9796142578125,
+      "step": 9
+    },
+    {
+      "epoch": 0.00025,
+      "grad_norm": 7.0878190994262695,
+      "learning_rate": 1e-05,
+      "loss": 346.0762,
+      "loss/crossentropy": 2.8687756061553955,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25318092107772827,
+      "loss/reg": 342.9542236328125,
+      "step": 10
+    },
+    {
+      "epoch": 0.000275,
+      "grad_norm": 33.097476959228516,
+      "learning_rate": 1.1000000000000001e-05,
+      "loss": 345.8165,
+      "loss/crossentropy": 2.6577072143554688,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23302724957466125,
+      "loss/reg": 342.92578125,
+      "step": 11
+    },
+    {
+      "epoch": 0.0003,
+      "grad_norm": 17.795812606811523,
+      "learning_rate": 1.2e-05,
+      "loss": 346.2238,
+      "loss/crossentropy": 3.0676422119140625,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2625121474266052,
+      "loss/reg": 342.8936462402344,
+      "step": 12
+    },
+    {
+      "epoch": 0.000325,
+      "grad_norm": 5.172369003295898,
+      "learning_rate": 1.3000000000000001e-05,
+      "loss": 345.8532,
+      "loss/crossentropy": 2.8076300621032715,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18656937777996063,
+      "loss/reg": 342.8590087890625,
+      "step": 13
+    },
+    {
+      "epoch": 0.00035,
+      "grad_norm": 5.153316497802734,
+      "learning_rate": 1.4000000000000001e-05,
+      "loss": 345.4301,
+      "loss/crossentropy": 2.4606783390045166,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15036556124687195,
+      "loss/reg": 342.8190612792969,
+      "step": 14
+    },
+    {
+      "epoch": 0.000375,
+      "grad_norm": 5.007053852081299,
+      "learning_rate": 1.5e-05,
+      "loss": 345.9839,
+      "loss/crossentropy": 3.0140442848205566,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19039079546928406,
+      "loss/reg": 342.7794189453125,
+      "step": 15
+    },
+    {
+      "epoch": 0.0004,
+      "grad_norm": 9.89411449432373,
+      "grad_norm_var": 218532.61416541884,
+      "learning_rate": 1.6000000000000003e-05,
+      "loss": 345.7775,
+      "loss/crossentropy": 2.900439977645874,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14180609583854675,
+      "loss/reg": 342.7352600097656,
+      "step": 16
+    },
+    {
+      "epoch": 0.000425,
+      "grad_norm": 4.438236713409424,
+      "grad_norm_var": 14521.267629174466,
+      "learning_rate": 1.7000000000000003e-05,
+      "loss": 345.5058,
+      "loss/crossentropy": 2.6860878467559814,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.13272428512573242,
+      "loss/reg": 342.68695068359375,
+      "step": 17
+    },
+    {
+      "epoch": 0.00045,
+      "grad_norm": 6.959152698516846,
+      "grad_norm_var": 14580.739492472594,
+      "learning_rate": 1.8e-05,
+      "loss": 345.8498,
+      "loss/crossentropy": 3.0659737586975098,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14944659173488617,
+      "loss/reg": 342.6343688964844,
+      "step": 18
+    },
+    {
+      "epoch": 0.000475,
+      "grad_norm": 4.815605640411377,
+      "grad_norm_var": 14622.192919380279,
+      "learning_rate": 1.9e-05,
+      "loss": 345.6652,
+      "loss/crossentropy": 2.9661877155303955,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.11781711876392365,
+      "loss/reg": 342.5811462402344,
+      "step": 19
+    },
+    {
+      "epoch": 0.0005,
+      "grad_norm": 20.490373611450195,
+      "grad_norm_var": 526.7206949274603,
+      "learning_rate": 2e-05,
+      "loss": 345.5417,
+      "loss/crossentropy": 2.880610942840576,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1311033070087433,
+      "loss/reg": 342.530029296875,
+      "step": 20
+    },
+    {
+      "epoch": 0.000525,
+      "grad_norm": 8.06376838684082,
+      "grad_norm_var": 534.1263548023838,
+      "learning_rate": 2.1e-05,
+      "loss": 345.2038,
+      "loss/crossentropy": 2.6139538288116455,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1176835298538208,
+      "loss/reg": 342.4721984863281,
+      "step": 21
+    },
+    {
+      "epoch": 0.00055,
+      "grad_norm": 2.9717931747436523,
+      "grad_norm_var": 499.3808875231212,
+      "learning_rate": 2.2000000000000003e-05,
+      "loss": 345.0013,
+      "loss/crossentropy": 2.48826265335083,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.09986373782157898,
+      "loss/reg": 342.4132080078125,
+      "step": 22
+    },
+    {
+      "epoch": 0.000575,
+      "grad_norm": 3.0796430110931396,
+      "grad_norm_var": 509.5546291852752,
+      "learning_rate": 2.3000000000000003e-05,
+      "loss": 345.5402,
+      "loss/crossentropy": 3.083524227142334,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.11023637652397156,
+      "loss/reg": 342.3464050292969,
+      "step": 23
+    },
+    {
+      "epoch": 0.0006,
+      "grad_norm": 4.465290069580078,
+      "grad_norm_var": 66.02530253493347,
+      "learning_rate": 2.4e-05,
+      "loss": 345.118,
+      "loss/crossentropy": 2.7357635498046875,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.10327694565057755,
+      "loss/reg": 342.2790222167969,
+      "step": 24
+    },
+    {
+      "epoch": 0.000625,
+      "grad_norm": 2.7485175132751465,
+      "grad_norm_var": 67.58997383241844,
+      "learning_rate": 2.5e-05,
+      "loss": 345.032,
+      "loss/crossentropy": 2.7314038276672363,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.09067989885807037,
+      "loss/reg": 342.2099609375,
+      "step": 25
+    },
+    {
+      "epoch": 0.00065,
+      "grad_norm": 6.201633930206299,
+      "grad_norm_var": 67.84461638262088,
+      "learning_rate": 2.6000000000000002e-05,
+      "loss": 345.2013,
+      "loss/crossentropy": 2.9543004035949707,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.10873008519411087,
+      "loss/reg": 342.1382141113281,
+      "step": 26
+    },
+    {
+      "epoch": 0.000675,
+      "grad_norm": 2.800253391265869,
+      "grad_norm_var": 26.949349170164666,
+      "learning_rate": 2.7000000000000002e-05,
+      "loss": 345.0995,
+      "loss/crossentropy": 2.9452881813049316,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.09716818481683731,
+      "loss/reg": 342.0570983886719,
+      "step": 27
+    },
+    {
+      "epoch": 0.0007,
+      "grad_norm": 2.699833631515503,
+      "grad_norm_var": 19.218166857921933,
+      "learning_rate": 2.8000000000000003e-05,
+      "loss": 344.7669,
+      "loss/crossentropy": 2.708958864212036,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.08778566122055054,
+      "loss/reg": 341.9701843261719,
+      "step": 28
+    },
+    {
+      "epoch": 0.000725,
+      "grad_norm": 2.547497272491455,
+      "grad_norm_var": 19.91571754218688,
+      "learning_rate": 2.9e-05,
+      "loss": 344.7711,
+      "loss/crossentropy": 2.8091719150543213,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.08377528190612793,
+      "loss/reg": 341.8781433105469,
+      "step": 29
+    },
+    {
+      "epoch": 0.00075,
+      "grad_norm": 2.952777862548828,
+      "grad_norm_var": 20.399598744815588,
+      "learning_rate": 3e-05,
+      "loss": 344.7112,
+      "loss/crossentropy": 2.8418242931365967,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.08888664841651917,
+      "loss/reg": 341.7804260253906,
+      "step": 30
+    },
+    {
+      "epoch": 0.000775,
+      "grad_norm": 2.7115345001220703,
+      "grad_norm_var": 20.92066401501108,
+      "learning_rate": 3.1e-05,
+      "loss": 344.3383,
+      "loss/crossentropy": 2.567899703979492,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.08625812828540802,
+      "loss/reg": 341.68408203125,
+      "step": 31
+    },
+    {
+      "epoch": 0.0008,
+      "grad_norm": 2.8718879222869873,
+      "grad_norm_var": 19.879086013782704,
+      "learning_rate": 3.2000000000000005e-05,
+      "loss": 344.435,
+      "loss/crossentropy": 2.772453546524048,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.08201509714126587,
+      "loss/reg": 341.58056640625,
+      "step": 32
+    },
+    {
+      "epoch": 0.000825,
+      "grad_norm": 3.096867084503174,
+      "grad_norm_var": 20.10115293189847,
+      "learning_rate": 3.3e-05,
+      "loss": 344.5786,
+      "loss/crossentropy": 3.0346808433532715,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.07674358785152435,
+      "loss/reg": 341.4671936035156,
+      "step": 33
+    },
+    {
+      "epoch": 0.00085,
+      "grad_norm": 2.8663108348846436,
+      "grad_norm_var": 20.061121544886294,
+      "learning_rate": 3.4000000000000007e-05,
+      "loss": 344.4023,
+      "loss/crossentropy": 2.9602417945861816,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.09072191268205643,
+      "loss/reg": 341.3513488769531,
+      "step": 34
+    },
+    {
+      "epoch": 0.000875,
+      "grad_norm": 2.5331344604492188,
+      "grad_norm_var": 20.355035956744192,
+      "learning_rate": 3.5e-05,
+      "loss": 343.8108,
+      "loss/crossentropy": 2.4948618412017822,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.07910144329071045,
+      "loss/reg": 341.2367858886719,
+      "step": 35
+    },
+    {
+      "epoch": 0.0009,
+      "grad_norm": 3.133824110031128,
+      "grad_norm_var": 2.3373841825731096,
+      "learning_rate": 3.6e-05,
+      "loss": 344.111,
+      "loss/crossentropy": 2.913165807723999,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.08120056986808777,
+      "loss/reg": 341.11663818359375,
+      "step": 36
+    },
+    {
+      "epoch": 0.000925,
+      "grad_norm": 2.7178568840026855,
+      "grad_norm_var": 0.8591776946022585,
+      "learning_rate": 3.7e-05,
+      "loss": 343.527,
+      "loss/crossentropy": 2.4706835746765137,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.06964104622602463,
+      "loss/reg": 340.9867248535156,
+      "step": 37
+    },
+    {
+      "epoch": 0.00095,
+      "grad_norm": 2.5725350379943848,
+      "grad_norm_var": 0.8786228996593496,
+      "learning_rate": 3.8e-05,
+      "loss": 343.4507,
+      "loss/crossentropy": 2.5263352394104004,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.07618534564971924,
+      "loss/reg": 340.84814453125,
+      "step": 38
+    },
+    {
+      "epoch": 0.000975,
+      "grad_norm": 2.9023842811584473,
+      "grad_norm_var": 0.8816577904134154,
+      "learning_rate": 3.9000000000000006e-05,
+      "loss": 343.5782,
+      "loss/crossentropy": 2.794910192489624,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.07984155416488647,
+      "loss/reg": 340.7034606933594,
+      "step": 39
+    },
+    {
+      "epoch": 0.001,
+      "grad_norm": 2.4993066787719727,
+      "grad_norm_var": 0.7689802883673867,
+      "learning_rate": 4e-05,
+      "loss": 343.2404,
+      "loss/crossentropy": 2.6109530925750732,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.06857939064502716,
+      "loss/reg": 340.5608215332031,
+      "step": 40
+    },
+    {
+      "epoch": 0.001025,
+      "grad_norm": 2.9259226322174072,
+      "grad_norm_var": 0.7652114100620696,
+      "learning_rate": 4.1e-05,
+      "loss": 343.4892,
+      "loss/crossentropy": 2.9959700107574463,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.0883413702249527,
+      "loss/reg": 340.40484619140625,
+      "step": 41
+    },
+    {
+      "epoch": 0.00105,
+      "grad_norm": 2.8494505882263184,
+      "grad_norm_var": 0.03747455037091238,
+      "learning_rate": 4.2e-05,
+      "loss": 342.9757,
+      "loss/crossentropy": 2.6481616497039795,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.07580338418483734,
+      "loss/reg": 340.2517395019531,
+      "step": 42
+    },
+    {
+      "epoch": 0.001075,
+      "grad_norm": 2.8318848609924316,
+      "grad_norm_var": 0.03756942187646525,
+      "learning_rate": 4.3e-05,
+      "loss": 343.0756,
+      "loss/crossentropy": 2.914668083190918,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.07487093657255173,
+      "loss/reg": 340.0860290527344,
+      "step": 43
+    },
+    {
+      "epoch": 0.0011,
+      "grad_norm": 2.5573673248291016,
+      "grad_norm_var": 0.04063739560772698,
+      "learning_rate": 4.4000000000000006e-05,
+      "loss": 342.7438,
+      "loss/crossentropy": 2.758633613586426,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.07187295705080032,
+      "loss/reg": 339.913330078125,
+      "step": 44
+    },
+    {
+      "epoch": 0.001125,
+      "grad_norm": 2.6494765281677246,
+      "grad_norm_var": 0.038049041798770604,
+      "learning_rate": 4.5e-05,
+      "loss": 342.6349,
+      "loss/crossentropy": 2.820467710494995,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.07692387700080872,
+      "loss/reg": 339.7375183105469,
+      "step": 45
+    },
+    {
+      "epoch": 0.00115,
+      "grad_norm": 2.503408432006836,
+      "grad_norm_var": 0.04103864613901654,
+      "learning_rate": 4.600000000000001e-05,
+      "loss": 342.367,
+      "loss/crossentropy": 2.7334773540496826,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.0775354877114296,
+      "loss/reg": 339.5559387207031,
+      "step": 46
+    },
+    {
+      "epoch": 0.001175,
+      "grad_norm": 2.55014967918396,
+      "grad_norm_var": 0.04379427355291886,
+      "learning_rate": 4.7e-05,
+      "loss": 342.1882,
+      "loss/crossentropy": 2.7380266189575195,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.07468418776988983,
+      "loss/reg": 339.37548828125,
+      "step": 47
+    },
+    {
+      "epoch": 0.0012,
+      "grad_norm": 2.682318687438965,
+      "grad_norm_var": 0.04305705435034118,
+      "learning_rate": 4.8e-05,
+      "loss": 341.9997,
+      "loss/crossentropy": 2.7375612258911133,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.06463472545146942,
+      "loss/reg": 339.1974792480469,
+      "step": 48
+    },
+    {
+      "epoch": 0.001225,
+      "grad_norm": 2.775233268737793,
+      "grad_norm_var": 0.03430480419531913,
+      "learning_rate": 4.9e-05,
+      "loss": 342.0445,
+      "loss/crossentropy": 2.9467415809631348,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.07443998754024506,
+      "loss/reg": 339.0233459472656,
+      "step": 49
+    },
+    {
+      "epoch": 0.00125,
+      "grad_norm": 2.5482699871063232,
+      "grad_norm_var": 0.034503307506799766,
+      "learning_rate": 5e-05,
+      "loss": 341.5835,
+      "loss/crossentropy": 2.6657731533050537,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.07320375740528107,
+      "loss/reg": 338.8445129394531,
+      "step": 50
+    },
+    {
+      "epoch": 0.001275,
+      "grad_norm": 2.7958645820617676,
+      "grad_norm_var": 0.032900881109643856,
+      "learning_rate": 5.1000000000000006e-05,
+      "loss": 341.7335,
+      "loss/crossentropy": 3.0148534774780273,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.08110088109970093,
+      "loss/reg": 338.6375427246094,
+      "step": 51
+    },
+    {
+      "epoch": 0.0013,
+      "grad_norm": 2.803079605102539,
+      "grad_norm_var": 0.02142033029363475,
+      "learning_rate": 5.2000000000000004e-05,
+      "loss": 341.2834,
+      "loss/crossentropy": 2.762022018432617,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.07052050530910492,
+      "loss/reg": 338.4508056640625,
+      "step": 52
+    },
+    {
+      "epoch": 0.001325,
+      "grad_norm": 3.264613389968872,
+      "grad_norm_var": 0.041567737457307886,
+      "learning_rate": 5.300000000000001e-05,
+      "loss": 341.1463,
+      "loss/crossentropy": 2.8193376064300537,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.09100329875946045,
+      "loss/reg": 338.2359924316406,
+      "step": 53
+    },
+    {
+      "epoch": 0.00135,
+      "grad_norm": 2.7328436374664307,
+      "grad_norm_var": 0.039766415905754825,
+      "learning_rate": 5.4000000000000005e-05,
+      "loss": 340.6042,
+      "loss/crossentropy": 2.502927780151367,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.07811737805604935,
+      "loss/reg": 338.0231628417969,
+      "step": 54
+    },
+    {
+      "epoch": 0.001375,
+      "grad_norm": 3.4147236347198486,
+      "grad_norm_var": 0.06713011702883306,
+      "learning_rate": 5.500000000000001e-05,
+      "loss": 340.9179,
+      "loss/crossentropy": 3.0260379314422607,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.09565207362174988,
+      "loss/reg": 337.7962341308594,
+      "step": 55
+    },
+    {
+      "epoch": 0.0014,
+      "grad_norm": 2.4889872074127197,
+      "grad_norm_var": 0.0675147239578789,
+      "learning_rate": 5.6000000000000006e-05,
+      "loss": 340.2508,
+      "loss/crossentropy": 2.5990207195281982,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.07676048576831818,
+      "loss/reg": 337.57501220703125,
+      "step": 56
+    },
+    {
+      "epoch": 0.001425,
+      "grad_norm": 2.6865806579589844,
+      "grad_norm_var": 0.06622606037023218,
+      "learning_rate": 5.6999999999999996e-05,
+      "loss": 340.0853,
+      "loss/crossentropy": 2.6511263847351074,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.07812380790710449,
+      "loss/reg": 337.35601806640625,
+      "step": 57
+    },
+    {
+      "epoch": 0.00145,
+      "grad_norm": 2.625458002090454,
+      "grad_norm_var": 0.06664228909772092,
+      "learning_rate": 5.8e-05,
+      "loss": 339.9197,
+      "loss/crossentropy": 2.7347943782806396,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.08070610463619232,
+      "loss/reg": 337.10418701171875,
+      "step": 58
+    },
+    {
+      "epoch": 0.001475,
+      "grad_norm": 3.2201623916625977,
+      "grad_norm_var": 0.08059432957815199,
+      "learning_rate": 5.9e-05,
+      "loss": 339.55,
+      "loss/crossentropy": 2.606354236602783,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.08992001414299011,
+      "loss/reg": 336.853759765625,
+      "step": 59
+    },
+    {
+      "epoch": 0.0015,
+      "grad_norm": 2.9708340167999268,
+      "grad_norm_var": 0.07963074673126963,
+      "learning_rate": 6e-05,
+      "loss": 339.8324,
+      "loss/crossentropy": 3.144277572631836,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.09014974534511566,
+      "loss/reg": 336.59796142578125,
+      "step": 60
+    },
+    {
+      "epoch": 0.001525,
+      "grad_norm": 3.067251443862915,
+      "grad_norm_var": 0.08246093717145656,
+      "learning_rate": 6.1e-05,
+      "loss": 339.0275,
+      "loss/crossentropy": 2.5903847217559814,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.10276832431554794,
+      "loss/reg": 336.3343200683594,
+      "step": 61
+    },
+    {
+      "epoch": 0.00155,
+      "grad_norm": 3.0696287155151367,
+      "grad_norm_var": 0.07855122581927745,
+      "learning_rate": 6.2e-05,
+      "loss": 339.1453,
+      "loss/crossentropy": 2.9826090335845947,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.09507931768894196,
+      "loss/reg": 336.0675964355469,
+      "step": 62
+    },
+    {
+      "epoch": 0.001575,
+      "grad_norm": 4.388347148895264,
+      "grad_norm_var": 0.2147750922195494,
+      "learning_rate": 6.3e-05,
+      "loss": 338.7891,
+      "loss/crossentropy": 2.883762836456299,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.12124631553888321,
+      "loss/reg": 335.7840576171875,
+      "step": 63
+    },
+    {
+      "epoch": 0.0016,
+      "grad_norm": 2.8880841732025146,
+      "grad_norm_var": 0.20950431287563612,
+      "learning_rate": 6.400000000000001e-05,
+      "loss": 338.2764,
+      "loss/crossentropy": 2.6835389137268066,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.09364684671163559,
+      "loss/reg": 335.49920654296875,
+      "step": 64
+    },
+    {
+      "epoch": 0.001625,
+      "grad_norm": 2.813340425491333,
+      "grad_norm_var": 0.2085356207302605,
+      "learning_rate": 6.500000000000001e-05,
+      "loss": 338.2492,
+      "loss/crossentropy": 2.953268527984619,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.09223559498786926,
+      "loss/reg": 335.2037353515625,
+      "step": 65
+    },
+    {
+      "epoch": 0.00165,
+      "grad_norm": 2.741157054901123,
+      "grad_norm_var": 0.19959997265595816,
+      "learning_rate": 6.6e-05,
+      "loss": 337.3503,
+      "loss/crossentropy": 2.3697705268859863,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.07957518845796585,
+      "loss/reg": 334.90087890625,
+      "step": 66
+    },
+    {
+      "epoch": 0.001675,
+      "grad_norm": 3.6085925102233887,
+      "grad_norm_var": 0.21895872310333644,
+      "learning_rate": 6.7e-05,
+      "loss": 337.5594,
+      "loss/crossentropy": 2.8747456073760986,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.09977666288614273,
+      "loss/reg": 334.5848388671875,
+      "step": 67
+    },
+    {
+      "epoch": 0.0017,
+      "grad_norm": 3.3652493953704834,
+      "grad_norm_var": 0.22027918073739935,
+      "learning_rate": 6.800000000000001e-05,
+      "loss": 337.4217,
+      "loss/crossentropy": 3.0556256771087646,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.10333959013223648,
+      "loss/reg": 334.2626953125,
+      "step": 68
+    },
+    {
+      "epoch": 0.001725,
+      "grad_norm": 3.1251964569091797,
+      "grad_norm_var": 0.2181387434263814,
+      "learning_rate": 6.9e-05,
+      "loss": 336.7281,
+      "loss/crossentropy": 2.6668808460235596,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1156100332736969,
+      "loss/reg": 333.9456481933594,
+      "step": 69
+    },
+    {
+      "epoch": 0.00175,
+      "grad_norm": 2.704983711242676,
+      "grad_norm_var": 0.219459742286683,
+      "learning_rate": 7e-05,
+      "loss": 336.4565,
+      "loss/crossentropy": 2.7357680797576904,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.09776544570922852,
+      "loss/reg": 333.6229553222656,
+      "step": 70
+    },
+    {
+      "epoch": 0.001775,
+      "grad_norm": 2.7894554138183594,
+      "grad_norm_var": 0.21546068539459284,
+      "learning_rate": 7.1e-05,
+      "loss": 336.4183,
+      "loss/crossentropy": 3.014235258102417,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.10774749517440796,
+      "loss/reg": 333.2962951660156,
+      "step": 71
+    },
+    {
+      "epoch": 0.0018,
+      "grad_norm": 2.5992395877838135,
+      "grad_norm_var": 0.2081999960008865,
+      "learning_rate": 7.2e-05,
+      "loss": 335.7705,
+      "loss/crossentropy": 2.7298364639282227,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.08281661570072174,
+      "loss/reg": 332.9578552246094,
+      "step": 72
+    },
+    {
+      "epoch": 0.001825,
+      "grad_norm": 2.682684898376465,
+      "grad_norm_var": 0.20838528770162246,
+      "learning_rate": 7.3e-05,
+      "loss": 335.5829,
+      "loss/crossentropy": 2.8684210777282715,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.0963786244392395,
+      "loss/reg": 332.6180725097656,
+      "step": 73
+    },
+    {
+      "epoch": 0.00185,
+      "grad_norm": 2.8167314529418945,
+      "grad_norm_var": 0.20006842089726434,
+      "learning_rate": 7.4e-05,
+      "loss": 335.2414,
+      "loss/crossentropy": 2.8619723320007324,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.09637870639562607,
+      "loss/reg": 332.2830810546875,
+      "step": 74
+    },
+    {
+      "epoch": 0.001875,
+      "grad_norm": 2.792503595352173,
+      "grad_norm_var": 0.20197785150176603,
+      "learning_rate": 7.500000000000001e-05,
+      "loss": 334.7866,
+      "loss/crossentropy": 2.752199411392212,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.10928215831518173,
+      "loss/reg": 331.9251403808594,
+      "step": 75
+    },
+    {
+      "epoch": 0.0019,
+      "grad_norm": 3.629366636276245,
+      "grad_norm_var": 0.22419816294486522,
+      "learning_rate": 7.6e-05,
+      "loss": 334.6898,
+      "loss/crossentropy": 3.008653402328491,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.11774040013551712,
+      "loss/reg": 331.5634460449219,
+      "step": 76
+    },
+    {
+      "epoch": 0.001925,
+      "grad_norm": 2.7170186042785645,
+      "grad_norm_var": 0.2318815003938539,
+      "learning_rate": 7.7e-05,
+      "loss": 334.0274,
+      "loss/crossentropy": 2.7513821125030518,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.0983215868473053,
+      "loss/reg": 331.177734375,
+      "step": 77
+    },
+    {
+      "epoch": 0.00195,
+      "grad_norm": 3.206778049468994,
+      "grad_norm_var": 0.2334942618955277,
+      "learning_rate": 7.800000000000001e-05,
+      "loss": 334.0803,
+      "loss/crossentropy": 3.193337917327881,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.10869012027978897,
+      "loss/reg": 330.7782897949219,
+      "step": 78
+    },
+    {
+      "epoch": 0.001975,
+      "grad_norm": 2.8564422130584717,
+      "grad_norm_var": 0.10767969782056426,
+      "learning_rate": 7.900000000000001e-05,
+      "loss": 333.2658,
+      "loss/crossentropy": 2.764251947402954,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.09988151490688324,
+      "loss/reg": 330.4017028808594,
+      "step": 79
+    },
+    {
+      "epoch": 0.002,
+      "grad_norm": 2.5950961112976074,
+      "grad_norm_var": 0.11579763493383836,
+      "learning_rate": 8e-05,
+      "loss": 332.7263,
+      "loss/crossentropy": 2.626741409301758,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.088912233710289,
+      "loss/reg": 330.01068115234375,
+      "step": 80
+    },
+    {
+      "epoch": 0.002025,
+      "grad_norm": 2.9864420890808105,
+      "grad_norm_var": 0.11474153182901219,
+      "learning_rate": 8.1e-05,
+      "loss": 332.4809,
+      "loss/crossentropy": 2.768440008163452,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.11446887254714966,
+      "loss/reg": 329.5980224609375,
+      "step": 81
+    },
+    {
+      "epoch": 0.00205,
+      "grad_norm": 2.9913902282714844,
+      "grad_norm_var": 0.11165182755541589,
+      "learning_rate": 8.2e-05,
+      "loss": 332.2192,
+      "loss/crossentropy": 2.921024799346924,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.10505437850952148,
+      "loss/reg": 329.193115234375,
+      "step": 82
+    },
+    {
+      "epoch": 0.002075,
+      "grad_norm": 3.1324057579040527,
+      "grad_norm_var": 0.08506906493201592,
+      "learning_rate": 8.3e-05,
+      "loss": 331.8305,
+      "loss/crossentropy": 2.9233036041259766,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.11960877478122711,
+      "loss/reg": 328.78759765625,
+      "step": 83
+    },
+    {
+      "epoch": 0.0021,
+      "grad_norm": 4.698559761047363,
+      "grad_norm_var": 0.27231954898200667,
+      "learning_rate": 8.4e-05,
+      "loss": 331.6368,
+      "loss/crossentropy": 3.0924150943756104,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17615610361099243,
+      "loss/reg": 328.3682861328125,
+      "step": 84
+    },
+    {
+      "epoch": 0.002125,
+      "grad_norm": 3.177739143371582,
+      "grad_norm_var": 0.27322718837700205,
+      "learning_rate": 8.5e-05,
+      "loss": 330.8532,
+      "loss/crossentropy": 2.7933664321899414,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.11796407401561737,
+      "loss/reg": 327.9419250488281,
+      "step": 85
+    },
+    {
+      "epoch": 0.00215,
+      "grad_norm": 3.030458688735962,
+      "grad_norm_var": 0.26602324580689635,
+      "learning_rate": 8.6e-05,
+      "loss": 330.4252,
+      "loss/crossentropy": 2.8065192699432373,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.11260520666837692,
+      "loss/reg": 327.506103515625,
+      "step": 86
+    },
+    {
+      "epoch": 0.002175,
+      "grad_norm": 2.590994119644165,
+      "grad_norm_var": 0.2752177678969853,
+      "learning_rate": 8.7e-05,
+      "loss": 329.8946,
+      "loss/crossentropy": 2.706664562225342,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.11704559624195099,
+      "loss/reg": 327.0708923339844,
+      "step": 87
+    },
+    {
+      "epoch": 0.0022,
+      "grad_norm": 2.971489429473877,
+      "grad_norm_var": 0.2624243358406903,
+      "learning_rate": 8.800000000000001e-05,
+      "loss": 329.6451,
+      "loss/crossentropy": 2.8904268741607666,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.11559218168258667,
+      "loss/reg": 326.6390380859375,
+      "step": 88
+    },
+    {
+      "epoch": 0.002225,
+      "grad_norm": 2.968080759048462,
+      "grad_norm_var": 0.25335665900247384,
+      "learning_rate": 8.900000000000001e-05,
+      "loss": 329.3453,
+      "loss/crossentropy": 3.0395331382751465,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.12317082285881042,
+      "loss/reg": 326.1826477050781,
+      "step": 89
+    },
+    {
+      "epoch": 0.00225,
+      "grad_norm": 2.7764782905578613,
+      "grad_norm_var": 0.2548311632382782,
+      "learning_rate": 9e-05,
+      "loss": 328.5304,
+      "loss/crossentropy": 2.6738674640655518,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1210193932056427,
+      "loss/reg": 325.73553466796875,
+      "step": 90
+    },
+    {
+      "epoch": 0.002275,
+      "grad_norm": 2.9657976627349854,
+      "grad_norm_var": 0.25029449720438685,
+      "learning_rate": 9.1e-05,
+      "loss": 327.8257,
+      "loss/crossentropy": 2.4202983379364014,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.13400676846504211,
+      "loss/reg": 325.27142333984375,
+      "step": 91
+    },
+    {
+      "epoch": 0.0023,
+      "grad_norm": 2.650662422180176,
+      "grad_norm_var": 0.23859044240333346,
+      "learning_rate": 9.200000000000001e-05,
+      "loss": 327.4639,
+      "loss/crossentropy": 2.5483720302581787,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.11590175330638885,
+      "loss/reg": 324.7995910644531,
+      "step": 92
+    },
+    {
+      "epoch": 0.002325,
+      "grad_norm": 3.443924903869629,
+      "grad_norm_var": 0.24227501888755848,
+      "learning_rate": 9.300000000000001e-05,
+      "loss": 327.2633,
+      "loss/crossentropy": 2.815500497817993,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1212223544716835,
+      "loss/reg": 324.32659912109375,
+      "step": 93
+    },
+    {
+      "epoch": 0.00235,
+      "grad_norm": 3.134320020675659,
+      "grad_norm_var": 0.24123508076126352,
+      "learning_rate": 9.4e-05,
+      "loss": 326.7527,
+      "loss/crossentropy": 2.785496473312378,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.13180789351463318,
+      "loss/reg": 323.8353576660156,
+      "step": 94
+    },
+    {
+      "epoch": 0.002375,
+      "grad_norm": 2.610710382461548,
+      "grad_norm_var": 0.2516995603676444,
+      "learning_rate": 9.5e-05,
+      "loss": 326.0419,
+      "loss/crossentropy": 2.5883114337921143,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.11013612151145935,
+      "loss/reg": 323.34344482421875,
+      "step": 95
+    },
+    {
+      "epoch": 0.0024,
+      "grad_norm": 3.3038907051086426,
+      "grad_norm_var": 0.24055345506962927,
+      "learning_rate": 9.6e-05,
+      "loss": 326.3564,
+      "loss/crossentropy": 3.360260009765625,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.12902778387069702,
+      "loss/reg": 322.8670959472656,
+      "step": 96
+    },
+    {
+      "epoch": 0.002425,
+      "grad_norm": 3.0808234214782715,
+      "grad_norm_var": 0.23981223839680202,
+      "learning_rate": 9.7e-05,
+      "loss": 325.4279,
+      "loss/crossentropy": 2.9151549339294434,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1561610996723175,
+      "loss/reg": 322.3565979003906,
+      "step": 97
+    },
+    {
+      "epoch": 0.00245,
+      "grad_norm": 3.7542147636413574,
+      "grad_norm_var": 0.2655938131607021,
+      "learning_rate": 9.8e-05,
+      "loss": 324.8318,
+      "loss/crossentropy": 2.83591628074646,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.13835349678993225,
+      "loss/reg": 321.8575134277344,
+      "step": 98
+    },
+    {
+      "epoch": 0.002475,
+      "grad_norm": 4.176578044891357,
+      "grad_norm_var": 0.3322401459220373,
+      "learning_rate": 9.900000000000001e-05,
+      "loss": 324.9787,
+      "loss/crossentropy": 3.4443886280059814,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20227839052677155,
+      "loss/reg": 321.3320007324219,
+      "step": 99
+    },
+    {
+      "epoch": 0.0025,
+      "grad_norm": 3.2133636474609375,
+      "grad_norm_var": 0.17501650801164412,
+      "learning_rate": 0.0001,
+      "loss": 323.56,
+      "loss/crossentropy": 2.631444215774536,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.13826847076416016,
+      "loss/reg": 320.7903137207031,
+      "step": 100
+    },
+    {
+      "epoch": 0.002525,
+      "grad_norm": 3.3536131381988525,
+      "grad_norm_var": 0.17840699933572804,
+      "learning_rate": 0.0001,
+      "loss": 323.2537,
+      "loss/crossentropy": 2.823065757751465,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19433599710464478,
+      "loss/reg": 320.23626708984375,
+      "step": 101
+    },
+    {
+      "epoch": 0.00255,
+      "grad_norm": 2.846142053604126,
+      "grad_norm_var": 0.18289270736203184,
+      "learning_rate": 0.0001,
+      "loss": 322.8064,
+      "loss/crossentropy": 2.977403163909912,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14205776154994965,
+      "loss/reg": 319.68695068359375,
+      "step": 102
+    },
+    {
+      "epoch": 0.002575,
+      "grad_norm": 2.762763500213623,
+      "grad_norm_var": 0.1727341123234955,
+      "learning_rate": 0.0001,
+      "loss": 322.0067,
+      "loss/crossentropy": 2.7616305351257324,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.13712918758392334,
+      "loss/reg": 319.10791015625,
+      "step": 103
+    },
+    {
+      "epoch": 0.0026,
+      "grad_norm": 2.5470879077911377,
+      "grad_norm_var": 0.192723543133589,
+      "learning_rate": 0.0001,
+      "loss": 321.2095,
+      "loss/crossentropy": 2.548597812652588,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.12305562198162079,
+      "loss/reg": 318.5378723144531,
+      "step": 104
+    },
+    {
+      "epoch": 0.002625,
+      "grad_norm": 4.06820821762085,
+      "grad_norm_var": 0.2491215448449329,
+      "learning_rate": 0.0001,
+      "loss": 320.8897,
+      "loss/crossentropy": 2.7753727436065674,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16259068250656128,
+      "loss/reg": 317.9517517089844,
+      "step": 105
+    },
+    {
+      "epoch": 0.00265,
+      "grad_norm": 2.7462382316589355,
+      "grad_norm_var": 0.25075746320380765,
+      "learning_rate": 0.0001,
+      "loss": 320.5133,
+      "loss/crossentropy": 3.033684492111206,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1321936696767807,
+      "loss/reg": 317.347412109375,
+      "step": 106
+    },
+    {
+      "epoch": 0.002675,
+      "grad_norm": 2.9147019386291504,
+      "grad_norm_var": 0.2522855635872415,
+      "learning_rate": 0.0001,
+      "loss": 319.5785,
+      "loss/crossentropy": 2.6685800552368164,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14506930112838745,
+      "loss/reg": 316.7648620605469,
+      "step": 107
+    },
+    {
+      "epoch": 0.0027,
+      "grad_norm": 3.1959245204925537,
+      "grad_norm_var": 0.2336231557989919,
+      "learning_rate": 0.0001,
+      "loss": 319.3165,
+      "loss/crossentropy": 2.9722442626953125,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14754945039749146,
+      "loss/reg": 316.1966857910156,
+      "step": 108
+    },
+    {
+      "epoch": 0.002725,
+      "grad_norm": 3.798722267150879,
+      "grad_norm_var": 0.2531703418887588,
+      "learning_rate": 0.0001,
+      "loss": 319.04,
+      "loss/crossentropy": 3.2570812702178955,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16630737483501434,
+      "loss/reg": 315.6166076660156,
+      "step": 109
+    },
+    {
+      "epoch": 0.00275,
+      "grad_norm": 2.6525461673736572,
+      "grad_norm_var": 0.27312977627090806,
+      "learning_rate": 0.0001,
+      "loss": 317.8362,
+      "loss/crossentropy": 2.6261277198791504,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14820465445518494,
+      "loss/reg": 315.0619201660156,
+      "step": 110
+    },
+    {
+      "epoch": 0.002775,
+      "grad_norm": 3.1293742656707764,
+      "grad_norm_var": 0.2499447068577022,
+      "learning_rate": 0.0001,
+      "loss": 317.2627,
+      "loss/crossentropy": 2.623718023300171,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14719776809215546,
+      "loss/reg": 314.4918212890625,
+      "step": 111
+    },
+    {
+      "epoch": 0.0028,
+      "grad_norm": 15.958823204040527,
+      "grad_norm_var": 10.398151501641397,
+      "learning_rate": 0.0001,
+      "loss": 316.8877,
+      "loss/crossentropy": 2.8262336254119873,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.129408597946167,
+      "loss/reg": 313.9320373535156,
+      "step": 112
+    },
+    {
+      "epoch": 0.002825,
+      "grad_norm": 3.9413416385650635,
+      "grad_norm_var": 10.337541876862074,
+      "learning_rate": 0.0001,
+      "loss": 316.3959,
+      "loss/crossentropy": 2.8791425228118896,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15744474530220032,
+      "loss/reg": 313.3592834472656,
+      "step": 113
+    },
+    {
+      "epoch": 0.00285,
+      "grad_norm": 2.8014914989471436,
+      "grad_norm_var": 10.433906911606512,
+      "learning_rate": 0.0001,
+      "loss": 315.6562,
+      "loss/crossentropy": 2.7462589740753174,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14004458487033844,
+      "loss/reg": 312.7699279785156,
+      "step": 114
+    },
+    {
+      "epoch": 0.002875,
+      "grad_norm": 2.7303318977355957,
+      "grad_norm_var": 10.531872222449612,
+      "learning_rate": 0.0001,
+      "loss": 315.0281,
+      "loss/crossentropy": 2.6918511390686035,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.13197588920593262,
+      "loss/reg": 312.2042541503906,
+      "step": 115
+    },
+    {
+      "epoch": 0.0029,
+      "grad_norm": 3.1385786533355713,
+      "grad_norm_var": 10.5392309058265,
+      "learning_rate": 0.0001,
+      "loss": 314.5278,
+      "loss/crossentropy": 2.748079299926758,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15540769696235657,
+      "loss/reg": 311.6242980957031,
+      "step": 116
+    },
+    {
+      "epoch": 0.002925,
+      "grad_norm": 2.856105327606201,
+      "grad_norm_var": 10.591715440353928,
+      "learning_rate": 0.0001,
+      "loss": 313.9369,
+      "loss/crossentropy": 2.706995725631714,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1622162163257599,
+      "loss/reg": 311.0676574707031,
+      "step": 117
+    },
+    {
+      "epoch": 0.00295,
+      "grad_norm": 3.0335209369659424,
+      "grad_norm_var": 10.568067027910804,
+      "learning_rate": 0.0001,
+      "loss": 313.4418,
+      "loss/crossentropy": 2.7954583168029785,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14605656266212463,
+      "loss/reg": 310.50030517578125,
+      "step": 118
+    },
+    {
+      "epoch": 0.002975,
+      "grad_norm": 17.778278350830078,
+      "grad_norm_var": 22.39839291836262,
+      "learning_rate": 0.0001,
+      "loss": 313.1016,
+      "loss/crossentropy": 3.015498399734497,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14635121822357178,
+      "loss/reg": 309.9397277832031,
+      "step": 119
+    },
+    {
+      "epoch": 0.003,
+      "grad_norm": 2.5111374855041504,
+      "grad_norm_var": 22.409419960433475,
+      "learning_rate": 0.0001,
+      "loss": 312.4484,
+      "loss/crossentropy": 2.9234232902526855,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15030893683433533,
+      "loss/reg": 309.3746337890625,
+      "step": 120
+    },
+    {
+      "epoch": 0.003025,
+      "grad_norm": 3.0075595378875732,
+      "grad_norm_var": 22.58724529715737,
+      "learning_rate": 0.0001,
+      "loss": 311.6718,
+      "loss/crossentropy": 2.7128963470458984,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15696708858013153,
+      "loss/reg": 308.8019714355469,
+      "step": 121
+    },
+    {
+      "epoch": 0.00305,
+      "grad_norm": 2.8516271114349365,
+      "grad_norm_var": 22.55961193976418,
+      "learning_rate": 0.0001,
+      "loss": 311.5919,
+      "loss/crossentropy": 3.201800584793091,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1552673727273941,
+      "loss/reg": 308.2347717285156,
+      "step": 122
+    },
+    {
+      "epoch": 0.003075,
+      "grad_norm": 2.7313201427459717,
+      "grad_norm_var": 22.60704699907139,
+      "learning_rate": 0.0001,
+      "loss": 310.4214,
+      "loss/crossentropy": 2.6282145977020264,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.13782621920108795,
+      "loss/reg": 307.6553955078125,
+      "step": 123
+    },
+    {
+      "epoch": 0.0031,
+      "grad_norm": 3.00699520111084,
+      "grad_norm_var": 22.64860965541018,
+      "learning_rate": 0.0001,
+      "loss": 309.9948,
+      "loss/crossentropy": 2.773193597793579,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15227621793746948,
+      "loss/reg": 307.0693359375,
+      "step": 124
+    },
+    {
+      "epoch": 0.003125,
+      "grad_norm": 2.9907143115997314,
+      "grad_norm_var": 22.791413325771067,
+      "learning_rate": 0.0001,
+      "loss": 309.6375,
+      "loss/crossentropy": 3.001366138458252,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14761751890182495,
+      "loss/reg": 306.488525390625,
+      "step": 125
+    },
+    {
+      "epoch": 0.00315,
+      "grad_norm": 2.7048888206481934,
+      "grad_norm_var": 22.777330344225966,
+      "learning_rate": 0.0001,
+      "loss": 308.8778,
+      "loss/crossentropy": 2.8160345554351807,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.144578754901886,
+      "loss/reg": 305.9171447753906,
+      "step": 126
+    },
+    {
+      "epoch": 0.003175,
+      "grad_norm": 3.0401437282562256,
+      "grad_norm_var": 22.796493590486268,
+      "learning_rate": 0.0001,
+      "loss": 308.0832,
+      "loss/crossentropy": 2.607736825942993,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.13729776442050934,
+      "loss/reg": 305.3381652832031,
+      "step": 127
+    },
+    {
+      "epoch": 0.0032,
+      "grad_norm": 2.584563732147217,
+      "grad_norm_var": 13.885740820076437,
+      "learning_rate": 0.0001,
+      "loss": 307.6585,
+      "loss/crossentropy": 2.750492572784424,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.13798421621322632,
+      "loss/reg": 304.77001953125,
+      "step": 128
+    },
+    {
+      "epoch": 0.003225,
+      "grad_norm": 2.501339912414551,
+      "grad_norm_var": 13.999106697222624,
+      "learning_rate": 0.0001,
+      "loss": 306.8389,
+      "loss/crossentropy": 2.499746084213257,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14136828482151031,
+      "loss/reg": 304.19781494140625,
+      "step": 129
+    },
+    {
+      "epoch": 0.00325,
+      "grad_norm": 3.1528725624084473,
+      "grad_norm_var": 13.961598599715407,
+      "learning_rate": 0.0001,
+      "loss": 306.6842,
+      "loss/crossentropy": 2.9020321369171143,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15830577909946442,
+      "loss/reg": 303.62384033203125,
+      "step": 130
+    },
+    {
+      "epoch": 0.003275,
+      "grad_norm": 2.893075704574585,
+      "grad_norm_var": 13.94028717778272,
+      "learning_rate": 0.0001,
+      "loss": 305.9274,
+      "loss/crossentropy": 2.743729591369629,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14728468656539917,
+      "loss/reg": 303.0364074707031,
+      "step": 131
+    },
+    {
+      "epoch": 0.0033,
+      "grad_norm": 3.0060131549835205,
+      "grad_norm_var": 13.953057327464618,
+      "learning_rate": 0.0001,
+      "loss": 305.4517,
+      "loss/crossentropy": 2.834550142288208,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16237227618694305,
+      "loss/reg": 302.4547424316406,
+      "step": 132
+    },
+    {
+      "epoch": 0.003325,
+      "grad_norm": 3.0694620609283447,
+      "grad_norm_var": 13.929317309938057,
+      "learning_rate": 0.0001,
+      "loss": 304.5747,
+      "loss/crossentropy": 2.5429139137268066,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1559884250164032,
+      "loss/reg": 301.8758544921875,
+      "step": 133
+    },
+    {
+      "epoch": 0.00335,
+      "grad_norm": 2.8518848419189453,
+      "grad_norm_var": 13.950038126308876,
+      "learning_rate": 0.0001,
+      "loss": 304.1976,
+      "loss/crossentropy": 2.765056610107422,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14854753017425537,
+      "loss/reg": 301.28399658203125,
+      "step": 134
+    },
+    {
+      "epoch": 0.003375,
+      "grad_norm": 3.1176578998565674,
+      "grad_norm_var": 0.044937653308535865,
+      "learning_rate": 0.0001,
+      "loss": 303.7528,
+      "loss/crossentropy": 2.9260571002960205,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1468181163072586,
+      "loss/reg": 300.6799011230469,
+      "step": 135
+    },
+    {
+      "epoch": 0.0034,
+      "grad_norm": 2.880009889602661,
+      "grad_norm_var": 0.035480645828153946,
+      "learning_rate": 0.0001,
+      "loss": 302.9741,
+      "loss/crossentropy": 2.736276149749756,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1469912827014923,
+      "loss/reg": 300.0908508300781,
+      "step": 136
+    },
+    {
+      "epoch": 0.003425,
+      "grad_norm": 3.2396042346954346,
+      "grad_norm_var": 0.04219284656746065,
+      "learning_rate": 0.0001,
+      "loss": 302.0787,
+      "loss/crossentropy": 2.436084747314453,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15764698386192322,
+      "loss/reg": 299.4849548339844,
+      "step": 137
+    },
+    {
+      "epoch": 0.00345,
+      "grad_norm": 2.818772315979004,
+      "grad_norm_var": 0.04253304441969187,
+      "learning_rate": 0.0001,
+      "loss": 301.9017,
+      "loss/crossentropy": 2.8690006732940674,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.13555486500263214,
+      "loss/reg": 298.8971252441406,
+      "step": 138
+    },
+    {
+      "epoch": 0.003475,
+      "grad_norm": 3.1128132343292236,
+      "grad_norm_var": 0.042447214695410435,
+      "learning_rate": 0.0001,
+      "loss": 301.0966,
+      "loss/crossentropy": 2.6258418560028076,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17426033318042755,
+      "loss/reg": 298.2965393066406,
+      "step": 139
+    },
+    {
+      "epoch": 0.0035,
+      "grad_norm": 3.2976646423339844,
+      "grad_norm_var": 0.05049181223860704,
+      "learning_rate": 0.0001,
+      "loss": 301.2447,
+      "loss/crossentropy": 3.3375208377838135,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19012662768363953,
+      "loss/reg": 297.7170104980469,
+      "step": 140
+    },
+    {
+      "epoch": 0.003525,
+      "grad_norm": 2.7214531898498535,
+      "grad_norm_var": 0.05369940885392074,
+      "learning_rate": 0.0001,
+      "loss": 299.7301,
+      "loss/crossentropy": 2.44716477394104,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14688129723072052,
+      "loss/reg": 297.13604736328125,
+      "step": 141
+    },
+    {
+      "epoch": 0.00355,
+      "grad_norm": 3.3967669010162354,
+      "grad_norm_var": 0.0622042440975612,
+      "learning_rate": 0.0001,
+      "loss": 299.2944,
+      "loss/crossentropy": 2.5750997066497803,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17346450686454773,
+      "loss/reg": 296.5458068847656,
+      "step": 142
+    },
+    {
+      "epoch": 0.003575,
+      "grad_norm": 2.718705654144287,
+      "grad_norm_var": 0.06609520888261215,
+      "learning_rate": 0.0001,
+      "loss": 298.7035,
+      "loss/crossentropy": 2.576488733291626,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1555417776107788,
+      "loss/reg": 295.9714660644531,
+      "step": 143
+    },
+    {
+      "epoch": 0.0036,
+      "grad_norm": 3.459041118621826,
+      "grad_norm_var": 0.07009550701137926,
+      "learning_rate": 0.0001,
+      "loss": 298.3419,
+      "loss/crossentropy": 2.7831904888153076,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16926178336143494,
+      "loss/reg": 295.3894958496094,
+      "step": 144
+    },
+    {
+      "epoch": 0.003625,
+      "grad_norm": 3.599822998046875,
+      "grad_norm_var": 0.07030535777098426,
+      "learning_rate": 0.0001,
+      "loss": 298.2544,
+      "loss/crossentropy": 3.2596793174743652,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18854370713233948,
+      "loss/reg": 294.80621337890625,
+      "step": 145
+    },
+    {
+      "epoch": 0.00365,
+      "grad_norm": 3.6367578506469727,
+      "grad_norm_var": 0.08941673085661629,
+      "learning_rate": 0.0001,
+      "loss": 297.6298,
+      "loss/crossentropy": 3.228911876678467,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15578624606132507,
+      "loss/reg": 294.24505615234375,
+      "step": 146
+    },
+    {
+      "epoch": 0.003675,
+      "grad_norm": 3.3731987476348877,
+      "grad_norm_var": 0.08969931689454427,
+      "learning_rate": 0.0001,
+      "loss": 296.5468,
+      "loss/crossentropy": 2.6783909797668457,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18989479541778564,
+      "loss/reg": 293.678466796875,
+      "step": 147
+    },
+    {
+      "epoch": 0.0037,
+      "grad_norm": 3.004645347595215,
+      "grad_norm_var": 0.08972454925701427,
+      "learning_rate": 0.0001,
+      "loss": 296.139,
+      "loss/crossentropy": 2.887582302093506,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15772318840026855,
+      "loss/reg": 293.09368896484375,
+      "step": 148
+    },
+    {
+      "epoch": 0.003725,
+      "grad_norm": 3.8851213455200195,
+      "grad_norm_var": 0.1232384713331688,
+      "learning_rate": 0.0001,
+      "loss": 295.1292,
+      "loss/crossentropy": 2.43943452835083,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17688119411468506,
+      "loss/reg": 292.5129089355469,
+      "step": 149
+    },
+    {
+      "epoch": 0.00375,
+      "grad_norm": 2.969780206680298,
+      "grad_norm_var": 0.11871959357580385,
+      "learning_rate": 0.0001,
+      "loss": 294.82,
+      "loss/crossentropy": 2.720794439315796,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14547483623027802,
+      "loss/reg": 291.9537658691406,
+      "step": 150
+    },
+    {
+      "epoch": 0.003775,
+      "grad_norm": 3.3458027839660645,
+      "grad_norm_var": 0.1194074455570662,
+      "learning_rate": 0.0001,
+      "loss": 294.4663,
+      "loss/crossentropy": 2.914644718170166,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19076432287693024,
+      "loss/reg": 291.3608703613281,
+      "step": 151
+    },
+    {
+      "epoch": 0.0038,
+      "grad_norm": 2.9391496181488037,
+      "grad_norm_var": 0.11697470608086556,
+      "learning_rate": 0.0001,
+      "loss": 294.1329,
+      "loss/crossentropy": 3.177144765853882,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17206624150276184,
+      "loss/reg": 290.7837219238281,
+      "step": 152
+    },
+    {
+      "epoch": 0.003825,
+      "grad_norm": 3.448251485824585,
+      "grad_norm_var": 0.12024250794295975,
+      "learning_rate": 0.0001,
+      "loss": 293.2957,
+      "loss/crossentropy": 2.916529893875122,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18904531002044678,
+      "loss/reg": 290.1900939941406,
+      "step": 153
+    },
+    {
+      "epoch": 0.00385,
+      "grad_norm": 2.7693159580230713,
+      "grad_norm_var": 0.12312676691166922,
+      "learning_rate": 0.0001,
+      "loss": 292.1465,
+      "loss/crossentropy": 2.3761188983917236,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16432620584964752,
+      "loss/reg": 289.6059875488281,
+      "step": 154
+    },
+    {
+      "epoch": 0.003875,
+      "grad_norm": 2.8227667808532715,
+      "grad_norm_var": 0.13291251787620695,
+      "learning_rate": 0.0001,
+      "loss": 291.6444,
+      "loss/crossentropy": 2.4790537357330322,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15044492483139038,
+      "loss/reg": 289.0148620605469,
+      "step": 155
+    },
+    {
+      "epoch": 0.0039,
+      "grad_norm": 2.769251823425293,
+      "grad_norm_var": 0.14431173200531475,
+      "learning_rate": 0.0001,
+      "loss": 291.3051,
+      "loss/crossentropy": 2.724376916885376,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15646930038928986,
+      "loss/reg": 288.42431640625,
+      "step": 156
+    },
+    {
+      "epoch": 0.003925,
+      "grad_norm": 2.9450161457061768,
+      "grad_norm_var": 0.1338045365052719,
+      "learning_rate": 0.0001,
+      "loss": 290.7092,
+      "loss/crossentropy": 2.7232112884521484,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17326687276363373,
+      "loss/reg": 287.81268310546875,
+      "step": 157
+    },
+    {
+      "epoch": 0.00395,
+      "grad_norm": 2.7544944286346436,
+      "grad_norm_var": 0.1421121590872642,
+      "learning_rate": 0.0001,
+      "loss": 289.9852,
+      "loss/crossentropy": 2.628100633621216,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15836934745311737,
+      "loss/reg": 287.1986999511719,
+      "step": 158
+    },
+    {
+      "epoch": 0.003975,
+      "grad_norm": 3.282729148864746,
+      "grad_norm_var": 0.12936684677523522,
+      "learning_rate": 0.0001,
+      "loss": 289.6616,
+      "loss/crossentropy": 2.853585720062256,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20318502187728882,
+      "loss/reg": 286.6048583984375,
+      "step": 159
+    },
+    {
+      "epoch": 0.004,
+      "grad_norm": 2.7608420848846436,
+      "grad_norm_var": 0.1345857719286831,
+      "learning_rate": 0.0001,
+      "loss": 288.7971,
+      "loss/crossentropy": 2.639310836791992,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15059445798397064,
+      "loss/reg": 286.0071716308594,
+      "step": 160
+    },
+    {
+      "epoch": 0.004025,
+      "grad_norm": 2.615666151046753,
+      "grad_norm_var": 0.1353317229456477,
+      "learning_rate": 0.0001,
+      "loss": 288.2175,
+      "loss/crossentropy": 2.6622371673583984,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1381928026676178,
+      "loss/reg": 285.41705322265625,
+      "step": 161
+    },
+    {
+      "epoch": 0.00405,
+      "grad_norm": 2.699389934539795,
+      "grad_norm_var": 0.12099720896805645,
+      "learning_rate": 0.0001,
+      "loss": 287.8594,
+      "loss/crossentropy": 2.9092233180999756,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14686957001686096,
+      "loss/reg": 284.8033142089844,
+      "step": 162
+    },
+    {
+      "epoch": 0.004075,
+      "grad_norm": 2.52980375289917,
+      "grad_norm_var": 0.12619606783641663,
+      "learning_rate": 0.0001,
+      "loss": 287.099,
+      "loss/crossentropy": 2.7334139347076416,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1602194458246231,
+      "loss/reg": 284.2053527832031,
+      "step": 163
+    },
+    {
+      "epoch": 0.0041,
+      "grad_norm": 2.9028260707855225,
+      "grad_norm_var": 0.1263923635326608,
+      "learning_rate": 0.0001,
+      "loss": 286.6346,
+      "loss/crossentropy": 2.8802831172943115,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1618650257587433,
+      "loss/reg": 283.59246826171875,
+      "step": 164
+    },
+    {
+      "epoch": 0.004125,
+      "grad_norm": 3.010094404220581,
+      "grad_norm_var": 0.06689759190688846,
+      "learning_rate": 0.0001,
+      "loss": 286.137,
+      "loss/crossentropy": 2.9807565212249756,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17984265089035034,
+      "loss/reg": 282.9764404296875,
+      "step": 165
+    },
+    {
+      "epoch": 0.00415,
+      "grad_norm": 2.747976779937744,
+      "grad_norm_var": 0.06821403807519436,
+      "learning_rate": 0.0001,
+      "loss": 285.1411,
+      "loss/crossentropy": 2.6191065311431885,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17148274183273315,
+      "loss/reg": 282.3504943847656,
+      "step": 166
+    },
+    {
+      "epoch": 0.004175,
+      "grad_norm": 2.72310471534729,
+      "grad_norm_var": 0.05514136557764762,
+      "learning_rate": 0.0001,
+      "loss": 284.5961,
+      "loss/crossentropy": 2.7177231311798096,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15355822443962097,
+      "loss/reg": 281.72479248046875,
+      "step": 167
+    },
+    {
+      "epoch": 0.0042,
+      "grad_norm": 2.5135159492492676,
+      "grad_norm_var": 0.06183281601049633,
+      "learning_rate": 0.0001,
+      "loss": 283.7144,
+      "loss/crossentropy": 2.4525184631347656,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.13518790900707245,
+      "loss/reg": 281.1266784667969,
+      "step": 168
+    },
+    {
+      "epoch": 0.004225,
+      "grad_norm": 2.6755948066711426,
+      "grad_norm_var": 0.03554926963080618,
+      "learning_rate": 0.0001,
+      "loss": 283.5094,
+      "loss/crossentropy": 2.8291640281677246,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17008362710475922,
+      "loss/reg": 280.5101318359375,
+      "step": 169
+    },
+    {
+      "epoch": 0.00425,
+      "grad_norm": 3.0268986225128174,
+      "grad_norm_var": 0.03923814612245735,
+      "learning_rate": 0.0001,
+      "loss": 283.1964,
+      "loss/crossentropy": 3.105539083480835,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18000483512878418,
+      "loss/reg": 279.9109191894531,
+      "step": 170
+    },
+    {
+      "epoch": 0.004275,
+      "grad_norm": 2.5846283435821533,
+      "grad_norm_var": 0.04201988364738796,
+      "learning_rate": 0.0001,
+      "loss": 282.1588,
+      "loss/crossentropy": 2.6842048168182373,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16073930263519287,
+      "loss/reg": 279.3138732910156,
+      "step": 171
+    },
+    {
+      "epoch": 0.0043,
+      "grad_norm": 2.8193604946136475,
+      "grad_norm_var": 0.0420791835209684,
+      "learning_rate": 0.0001,
+      "loss": 281.6269,
+      "loss/crossentropy": 2.7535877227783203,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17149388790130615,
+      "loss/reg": 278.7017822265625,
+      "step": 172
+    },
+    {
+      "epoch": 0.004325,
+      "grad_norm": 2.695707321166992,
+      "grad_norm_var": 0.04071110069445254,
+      "learning_rate": 0.0001,
+      "loss": 280.9163,
+      "loss/crossentropy": 2.694293260574341,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15487328171730042,
+      "loss/reg": 278.0671081542969,
+      "step": 173
+    },
+    {
+      "epoch": 0.00435,
+      "grad_norm": 3.0142641067504883,
+      "grad_norm_var": 0.04434257349895461,
+      "learning_rate": 0.0001,
+      "loss": 280.5464,
+      "loss/crossentropy": 2.912869691848755,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1794903725385666,
+      "loss/reg": 277.45404052734375,
+      "step": 174
+    },
+    {
+      "epoch": 0.004375,
+      "grad_norm": 3.4833621978759766,
+      "grad_norm_var": 0.060102318830361885,
+      "learning_rate": 0.0001,
+      "loss": 280.0757,
+      "loss/crossentropy": 3.031020164489746,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1958397924900055,
+      "loss/reg": 276.84881591796875,
+      "step": 175
+    },
+    {
+      "epoch": 0.0044,
+      "grad_norm": 2.6899683475494385,
+      "grad_norm_var": 0.060788090400682414,
+      "learning_rate": 0.0001,
+      "loss": 279.0604,
+      "loss/crossentropy": 2.689755439758301,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1408371478319168,
+      "loss/reg": 276.22979736328125,
+      "step": 176
+    },
+    {
+      "epoch": 0.004425,
+      "grad_norm": 2.686345338821411,
+      "grad_norm_var": 0.059403126018499584,
+      "learning_rate": 0.0001,
+      "loss": 278.8184,
+      "loss/crossentropy": 3.0344579219818115,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1603156328201294,
+      "loss/reg": 275.6236267089844,
+      "step": 177
+    },
+    {
+      "epoch": 0.00445,
+      "grad_norm": 2.9537172317504883,
+      "grad_norm_var": 0.060028034669986144,
+      "learning_rate": 0.0001,
+      "loss": 277.8445,
+      "loss/crossentropy": 2.6497297286987305,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1772124469280243,
+      "loss/reg": 275.01751708984375,
+      "step": 178
+    },
+    {
+      "epoch": 0.004475,
+      "grad_norm": 2.6830756664276123,
+      "grad_norm_var": 0.055646030147755474,
+      "learning_rate": 0.0001,
+      "loss": 277.3781,
+      "loss/crossentropy": 2.7995009422302246,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15789446234703064,
+      "loss/reg": 274.420654296875,
+      "step": 179
+    },
+    {
+      "epoch": 0.0045,
+      "grad_norm": 2.8642189502716064,
+      "grad_norm_var": 0.05534192722217289,
+      "learning_rate": 0.0001,
+      "loss": 276.8572,
+      "loss/crossentropy": 2.8247008323669434,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19499436020851135,
+      "loss/reg": 273.8375244140625,
+      "step": 180
+    },
+    {
+      "epoch": 0.004525,
+      "grad_norm": 2.3742892742156982,
+      "grad_norm_var": 0.06476700330631478,
+      "learning_rate": 0.0001,
+      "loss": 276.0537,
+      "loss/crossentropy": 2.6782188415527344,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14065401256084442,
+      "loss/reg": 273.2348327636719,
+      "step": 181
+    },
+    {
+      "epoch": 0.00455,
+      "grad_norm": 2.8511719703674316,
+      "grad_norm_var": 0.06494378033136528,
+      "learning_rate": 0.0001,
+      "loss": 275.8217,
+      "loss/crossentropy": 3.0019702911376953,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17122197151184082,
+      "loss/reg": 272.6484375,
+      "step": 182
+    },
+    {
+      "epoch": 0.004575,
+      "grad_norm": 2.4934184551239014,
+      "grad_norm_var": 0.07028818692612214,
+      "learning_rate": 0.0001,
+      "loss": 275.005,
+      "loss/crossentropy": 2.765932559967041,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1687726229429245,
+      "loss/reg": 272.0703125,
+      "step": 183
+    },
+    {
+      "epoch": 0.0046,
+      "grad_norm": 3.0428080558776855,
+      "grad_norm_var": 0.06930197860402411,
+      "learning_rate": 0.0001,
+      "loss": 274.319,
+      "loss/crossentropy": 2.6140480041503906,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20197075605392456,
+      "loss/reg": 271.50299072265625,
+      "step": 184
+    },
+    {
+      "epoch": 0.004625,
+      "grad_norm": 2.7042367458343506,
+      "grad_norm_var": 0.06884502087402401,
+      "learning_rate": 0.0001,
+      "loss": 273.7751,
+      "loss/crossentropy": 2.7085049152374268,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16187983751296997,
+      "loss/reg": 270.90478515625,
+      "step": 185
+    },
+    {
+      "epoch": 0.00465,
+      "grad_norm": 2.4192299842834473,
+      "grad_norm_var": 0.07438801189002385,
+      "learning_rate": 0.0001,
+      "loss": 272.9952,
+      "loss/crossentropy": 2.509551525115967,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15606564283370972,
+      "loss/reg": 270.3295593261719,
+      "step": 186
+    },
+    {
+      "epoch": 0.004675,
+      "grad_norm": 2.6057181358337402,
+      "grad_norm_var": 0.07388755541776296,
+      "learning_rate": 0.0001,
+      "loss": 272.7285,
+      "loss/crossentropy": 2.832514762878418,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1625734269618988,
+      "loss/reg": 269.7333984375,
+      "step": 187
+    },
+    {
+      "epoch": 0.0047,
+      "grad_norm": 2.5072972774505615,
+      "grad_norm_var": 0.07807856050180723,
+      "learning_rate": 0.0001,
+      "loss": 272.1894,
+      "loss/crossentropy": 2.8929169178009033,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15297307074069977,
+      "loss/reg": 269.1435241699219,
+      "step": 188
+    },
+    {
+      "epoch": 0.004725,
+      "grad_norm": 2.668285369873047,
+      "grad_norm_var": 0.07833979493371525,
+      "learning_rate": 0.0001,
+      "loss": 271.5161,
+      "loss/crossentropy": 2.798143148422241,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1715821623802185,
+      "loss/reg": 268.5463562011719,
+      "step": 189
+    },
+    {
+      "epoch": 0.00475,
+      "grad_norm": 2.869302272796631,
+      "grad_norm_var": 0.07459542490979629,
+      "learning_rate": 0.0001,
+      "loss": 271.0226,
+      "loss/crossentropy": 2.913928270339966,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16282354295253754,
+      "loss/reg": 267.94586181640625,
+      "step": 190
+    },
+    {
+      "epoch": 0.004775,
+      "grad_norm": 2.4999892711639404,
+      "grad_norm_var": 0.03802990774269475,
+      "learning_rate": 0.0001,
+      "loss": 270.1609,
+      "loss/crossentropy": 2.6663661003112793,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1597089171409607,
+      "loss/reg": 267.33489990234375,
+      "step": 191
+    },
+    {
+      "epoch": 0.0048,
+      "grad_norm": 2.7010915279388428,
+      "grad_norm_var": 0.03804935894705916,
+      "learning_rate": 0.0001,
+      "loss": 269.6252,
+      "loss/crossentropy": 2.7287211418151855,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1760062277317047,
+      "loss/reg": 266.720458984375,
+      "step": 192
+    },
+    {
+      "epoch": 0.004825,
+      "grad_norm": 2.635450839996338,
+      "grad_norm_var": 0.03818693477124479,
+      "learning_rate": 0.0001,
+      "loss": 269.0576,
+      "loss/crossentropy": 2.7622997760772705,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1753700077533722,
+      "loss/reg": 266.11993408203125,
+      "step": 193
+    },
+    {
+      "epoch": 0.00485,
+      "grad_norm": 2.54504656791687,
+      "grad_norm_var": 0.03368765909132942,
+      "learning_rate": 0.0001,
+      "loss": 268.4639,
+      "loss/crossentropy": 2.774796962738037,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16008035838603973,
+      "loss/reg": 265.5290222167969,
+      "step": 194
+    },
+    {
+      "epoch": 0.004875,
+      "grad_norm": 2.8884477615356445,
+      "grad_norm_var": 0.03711886375720705,
+      "learning_rate": 0.0001,
+      "loss": 267.7791,
+      "loss/crossentropy": 2.678165912628174,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16350409388542175,
+      "loss/reg": 264.9374694824219,
+      "step": 195
+    },
+    {
+      "epoch": 0.0049,
+      "grad_norm": 3.0534660816192627,
+      "grad_norm_var": 0.044336834869120406,
+      "learning_rate": 0.0001,
+      "loss": 267.5437,
+      "loss/crossentropy": 2.990952491760254,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20335815846920013,
+      "loss/reg": 264.3493957519531,
+      "step": 196
+    },
+    {
+      "epoch": 0.004925,
+      "grad_norm": 2.8830363750457764,
+      "grad_norm_var": 0.03986402384874263,
+      "learning_rate": 0.0001,
+      "loss": 266.6284,
+      "loss/crossentropy": 2.674790859222412,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18334966897964478,
+      "loss/reg": 263.77020263671875,
+      "step": 197
+    },
+    {
+      "epoch": 0.00495,
+      "grad_norm": 3.1203083992004395,
+      "grad_norm_var": 0.04943917591481958,
+      "learning_rate": 0.0001,
+      "loss": 265.9243,
+      "loss/crossentropy": 2.5693366527557373,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16054046154022217,
+      "loss/reg": 263.1944580078125,
+      "step": 198
+    },
+    {
+      "epoch": 0.004975,
+      "grad_norm": 2.7858941555023193,
+      "grad_norm_var": 0.04566411871986986,
+      "learning_rate": 0.0001,
+      "loss": 265.842,
+      "loss/crossentropy": 3.063640832901001,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1918787658214569,
+      "loss/reg": 262.5865478515625,
+      "step": 199
+    },
+    {
+      "epoch": 0.005,
+      "grad_norm": 4.905181407928467,
+      "grad_norm_var": 0.3362427866457504,
+      "learning_rate": 0.0001,
+      "loss": 265.0334,
+      "loss/crossentropy": 2.75864839553833,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.28906920552253723,
+      "loss/reg": 261.9857177734375,
+      "step": 200
+    },
+    {
+      "epoch": 0.005025,
+      "grad_norm": 2.969449281692505,
+      "grad_norm_var": 0.33506015925643545,
+      "learning_rate": 0.0001,
+      "loss": 264.667,
+      "loss/crossentropy": 3.093539237976074,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19913670420646667,
+      "loss/reg": 261.37432861328125,
+      "step": 201
+    },
+    {
+      "epoch": 0.00505,
+      "grad_norm": 3.2088537216186523,
+      "grad_norm_var": 0.32566799990582446,
+      "learning_rate": 0.0001,
+      "loss": 264.2178,
+      "loss/crossentropy": 3.255889415740967,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20012566447257996,
+      "loss/reg": 260.76177978515625,
+      "step": 202
+    },
+    {
+      "epoch": 0.005075,
+      "grad_norm": 3.11022686958313,
+      "grad_norm_var": 0.3199018318121316,
+      "learning_rate": 0.0001,
+      "loss": 263.1096,
+      "loss/crossentropy": 2.753197193145752,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19567176699638367,
+      "loss/reg": 260.16070556640625,
+      "step": 203
+    },
+    {
+      "epoch": 0.0051,
+      "grad_norm": 2.5052154064178467,
+      "grad_norm_var": 0.32002761472599534,
+      "learning_rate": 0.0001,
+      "loss": 262.7169,
+      "loss/crossentropy": 2.981990098953247,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17743879556655884,
+      "loss/reg": 259.55743408203125,
+      "step": 204
+    },
+    {
+      "epoch": 0.005125,
+      "grad_norm": 3.437074899673462,
+      "grad_norm_var": 0.3271340622124995,
+      "learning_rate": 0.0001,
+      "loss": 262.1845,
+      "loss/crossentropy": 3.0462844371795654,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17453166842460632,
+      "loss/reg": 258.96368408203125,
+      "step": 205
+    },
+    {
+      "epoch": 0.00515,
+      "grad_norm": 3.3540172576904297,
+      "grad_norm_var": 0.33289475403727997,
+      "learning_rate": 0.0001,
+      "loss": 261.2446,
+      "loss/crossentropy": 2.684351682662964,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18615120649337769,
+      "loss/reg": 258.37408447265625,
+      "step": 206
+    },
+    {
+      "epoch": 0.005175,
+      "grad_norm": 2.6485707759857178,
+      "grad_norm_var": 0.32362257450873433,
+      "learning_rate": 0.0001,
+      "loss": 260.6506,
+      "loss/crossentropy": 2.712137222290039,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16978460550308228,
+      "loss/reg": 257.7686767578125,
+      "step": 207
+    },
+    {
+      "epoch": 0.0052,
+      "grad_norm": 2.638559103012085,
+      "grad_norm_var": 0.326750686147299,
+      "learning_rate": 0.0001,
+      "loss": 259.9714,
+      "loss/crossentropy": 2.6442058086395264,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16394683718681335,
+      "loss/reg": 257.1632385253906,
+      "step": 208
+    },
+    {
+      "epoch": 0.005225,
+      "grad_norm": 2.62022066116333,
+      "grad_norm_var": 0.32759289115149387,
+      "learning_rate": 0.0001,
+      "loss": 259.6152,
+      "loss/crossentropy": 2.86051607131958,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1837988793849945,
+      "loss/reg": 256.5708923339844,
+      "step": 209
+    },
+    {
+      "epoch": 0.00525,
+      "grad_norm": 3.0118000507354736,
+      "grad_norm_var": 0.31027566591164846,
+      "learning_rate": 0.0001,
+      "loss": 258.8864,
+      "loss/crossentropy": 2.7158284187316895,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18835628032684326,
+      "loss/reg": 255.98226928710938,
+      "step": 210
+    },
+    {
+      "epoch": 0.005275,
+      "grad_norm": 2.738693952560425,
+      "grad_norm_var": 0.31532774991742346,
+      "learning_rate": 0.0001,
+      "loss": 258.2583,
+      "loss/crossentropy": 2.6809372901916504,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1898835152387619,
+      "loss/reg": 255.387451171875,
+      "step": 211
+    },
+    {
+      "epoch": 0.0053,
+      "grad_norm": 2.739337205886841,
+      "grad_norm_var": 0.32184874512720374,
+      "learning_rate": 0.0001,
+      "loss": 257.8907,
+      "loss/crossentropy": 2.906891107559204,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1842600256204605,
+      "loss/reg": 254.79959106445312,
+      "step": 212
+    },
+    {
+      "epoch": 0.005325,
+      "grad_norm": 5.883702754974365,
+      "grad_norm_var": 0.820888078604249,
+      "learning_rate": 0.0001,
+      "loss": 257.1401,
+      "loss/crossentropy": 2.7662065029144287,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17315037548542023,
+      "loss/reg": 254.2007293701172,
+      "step": 213
+    },
+    {
+      "epoch": 0.00535,
+      "grad_norm": 3.2972004413604736,
+      "grad_norm_var": 0.8202608766175215,
+      "learning_rate": 0.0001,
+      "loss": 256.5279,
+      "loss/crossentropy": 2.737523317337036,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1854221224784851,
+      "loss/reg": 253.6049346923828,
+      "step": 214
+    },
+    {
+      "epoch": 0.005375,
+      "grad_norm": 3.02815580368042,
+      "grad_norm_var": 0.8092324619971265,
+      "learning_rate": 0.0001,
+      "loss": 256.0627,
+      "loss/crossentropy": 2.8625571727752686,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19949959218502045,
+      "loss/reg": 253.0006866455078,
+      "step": 215
+    },
+    {
+      "epoch": 0.0054,
+      "grad_norm": 3.2634494304656982,
+      "grad_norm_var": 0.6166894091876105,
+      "learning_rate": 0.0001,
+      "loss": 255.4964,
+      "loss/crossentropy": 2.924158811569214,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16927270591259003,
+      "loss/reg": 252.40298461914062,
+      "step": 216
+    },
+    {
+      "epoch": 0.005425,
+      "grad_norm": 3.190610647201538,
+      "grad_norm_var": 0.614321823565594,
+      "learning_rate": 0.0001,
+      "loss": 254.9168,
+      "loss/crossentropy": 2.959676742553711,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1694745123386383,
+      "loss/reg": 251.78762817382812,
+      "step": 217
+    },
+    {
+      "epoch": 0.00545,
+      "grad_norm": 2.910163402557373,
+      "grad_norm_var": 0.6182401597326245,
+      "learning_rate": 0.0001,
+      "loss": 254.3778,
+      "loss/crossentropy": 3.0126121044158936,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19637390971183777,
+      "loss/reg": 251.16880798339844,
+      "step": 218
+    },
+    {
+      "epoch": 0.005475,
+      "grad_norm": 2.5100033283233643,
+      "grad_norm_var": 0.6438249148085596,
+      "learning_rate": 0.0001,
+      "loss": 253.619,
+      "loss/crossentropy": 2.8760910034179688,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17724955081939697,
+      "loss/reg": 250.565673828125,
+      "step": 219
+    },
+    {
+      "epoch": 0.0055,
+      "grad_norm": 2.6412065029144287,
+      "grad_norm_var": 0.6339957102239564,
+      "learning_rate": 0.0001,
+      "loss": 253.0932,
+      "loss/crossentropy": 2.9451711177825928,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1847916692495346,
+      "loss/reg": 249.96322631835938,
+      "step": 220
+    },
+    {
+      "epoch": 0.005525,
+      "grad_norm": 2.4829912185668945,
+      "grad_norm_var": 0.6504949610281291,
+      "learning_rate": 0.0001,
+      "loss": 251.9582,
+      "loss/crossentropy": 2.4317898750305176,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15323440730571747,
+      "loss/reg": 249.37318420410156,
+      "step": 221
+    },
+    {
+      "epoch": 0.00555,
+      "grad_norm": 2.709721565246582,
+      "grad_norm_var": 0.6511748224192303,
+      "learning_rate": 0.0001,
+      "loss": 251.9418,
+      "loss/crossentropy": 2.9580609798431396,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19112282991409302,
+      "loss/reg": 248.7926788330078,
+      "step": 222
+    },
+    {
+      "epoch": 0.005575,
+      "grad_norm": 2.7896413803100586,
+      "grad_norm_var": 0.6454388623794479,
+      "learning_rate": 0.0001,
+      "loss": 251.1011,
+      "loss/crossentropy": 2.6989221572875977,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18577811121940613,
+      "loss/reg": 248.21641540527344,
+      "step": 223
+    },
+    {
+      "epoch": 0.0056,
+      "grad_norm": 2.763256549835205,
+      "grad_norm_var": 0.6399279824277821,
+      "learning_rate": 0.0001,
+      "loss": 250.6324,
+      "loss/crossentropy": 2.8038721084594727,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19551604986190796,
+      "loss/reg": 247.633056640625,
+      "step": 224
+    },
+    {
+      "epoch": 0.005625,
+      "grad_norm": 2.6225526332855225,
+      "grad_norm_var": 0.6397989634558596,
+      "learning_rate": 0.0001,
+      "loss": 250.0981,
+      "loss/crossentropy": 2.873964548110962,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18002215027809143,
+      "loss/reg": 247.0440673828125,
+      "step": 225
+    },
+    {
+      "epoch": 0.00565,
+      "grad_norm": 2.7640974521636963,
+      "grad_norm_var": 0.6444463916989489,
+      "learning_rate": 0.0001,
+      "loss": 249.6578,
+      "loss/crossentropy": 3.0052733421325684,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19472980499267578,
+      "loss/reg": 246.45779418945312,
+      "step": 226
+    },
+    {
+      "epoch": 0.005675,
+      "grad_norm": 4.11354398727417,
+      "grad_norm_var": 0.7108481451659288,
+      "learning_rate": 0.0001,
+      "loss": 249.1703,
+      "loss/crossentropy": 3.090904951095581,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20938721299171448,
+      "loss/reg": 245.87001037597656,
+      "step": 227
+    },
+    {
+      "epoch": 0.0057,
+      "grad_norm": 2.749560594558716,
+      "grad_norm_var": 0.7103537111133307,
+      "learning_rate": 0.0001,
+      "loss": 247.8932,
+      "loss/crossentropy": 2.4375710487365723,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16894645988941193,
+      "loss/reg": 245.28668212890625,
+      "step": 228
+    },
+    {
+      "epoch": 0.005725,
+      "grad_norm": 2.942262649536133,
+      "grad_norm_var": 0.16230004133669051,
+      "learning_rate": 0.0001,
+      "loss": 247.8172,
+      "loss/crossentropy": 2.9423258304595947,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1861574649810791,
+      "loss/reg": 244.6887664794922,
+      "step": 229
+    },
+    {
+      "epoch": 0.00575,
+      "grad_norm": 2.7158396244049072,
+      "grad_norm_var": 0.1544682228743985,
+      "learning_rate": 0.0001,
+      "loss": 246.9261,
+      "loss/crossentropy": 2.6655476093292236,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1642124503850937,
+      "loss/reg": 244.0963592529297,
+      "step": 230
+    },
+    {
+      "epoch": 0.005775,
+      "grad_norm": 3.031425714492798,
+      "grad_norm_var": 0.15453029560278514,
+      "learning_rate": 0.0001,
+      "loss": 246.3331,
+      "loss/crossentropy": 2.641087532043457,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1917201280593872,
+      "loss/reg": 243.5002899169922,
+      "step": 231
+    },
+    {
+      "epoch": 0.0058,
+      "grad_norm": 2.781378984451294,
+      "grad_norm_var": 0.1448915546965355,
+      "learning_rate": 0.0001,
+      "loss": 245.8157,
+      "loss/crossentropy": 2.7308170795440674,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.178788423538208,
+      "loss/reg": 242.90609741210938,
+      "step": 232
+    },
+    {
+      "epoch": 0.005825,
+      "grad_norm": 2.6632707118988037,
+      "grad_norm_var": 0.13884268173051903,
+      "learning_rate": 0.0001,
+      "loss": 245.2104,
+      "loss/crossentropy": 2.7034757137298584,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19913603365421295,
+      "loss/reg": 242.30780029296875,
+      "step": 233
+    },
+    {
+      "epoch": 0.00585,
+      "grad_norm": 2.6279611587524414,
+      "grad_norm_var": 0.14059426093737837,
+      "learning_rate": 0.0001,
+      "loss": 244.448,
+      "loss/crossentropy": 2.5541391372680664,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1769060492515564,
+      "loss/reg": 241.71693420410156,
+      "step": 234
+    },
+    {
+      "epoch": 0.005875,
+      "grad_norm": 2.669217586517334,
+      "grad_norm_var": 0.13587813046198485,
+      "learning_rate": 0.0001,
+      "loss": 243.9826,
+      "loss/crossentropy": 2.677557945251465,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1696961522102356,
+      "loss/reg": 241.13534545898438,
+      "step": 235
+    },
+    {
+      "epoch": 0.0059,
+      "grad_norm": 4.654603958129883,
+      "grad_norm_var": 0.34211473789542096,
+      "learning_rate": 0.0001,
+      "loss": 243.4455,
+      "loss/crossentropy": 2.6129984855651855,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24794022738933563,
+      "loss/reg": 240.58453369140625,
+      "step": 236
+    },
+    {
+      "epoch": 0.005925,
+      "grad_norm": 2.7966091632843018,
+      "grad_norm_var": 0.32904384319109853,
+      "learning_rate": 0.0001,
+      "loss": 243.0811,
+      "loss/crossentropy": 2.8610379695892334,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18956679105758667,
+      "loss/reg": 240.03045654296875,
+      "step": 237
+    },
+    {
+      "epoch": 0.00595,
+      "grad_norm": 2.4518496990203857,
+      "grad_norm_var": 0.3418803558489948,
+      "learning_rate": 0.0001,
+      "loss": 242.0064,
+      "loss/crossentropy": 2.412545680999756,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.14585809409618378,
+      "loss/reg": 239.44796752929688,
+      "step": 238
+    },
+    {
+      "epoch": 0.005975,
+      "grad_norm": 14.31179141998291,
+      "grad_norm_var": 8.399062121815462,
+      "learning_rate": 0.0001,
+      "loss": 241.8318,
+      "loss/crossentropy": 2.7919762134552,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17732976377010345,
+      "loss/reg": 238.86245727539062,
+      "step": 239
+    },
+    {
+      "epoch": 0.006,
+      "grad_norm": 2.6747586727142334,
+      "grad_norm_var": 8.410206107231419,
+      "learning_rate": 0.0001,
+      "loss": 241.3024,
+      "loss/crossentropy": 2.8322153091430664,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19286081194877625,
+      "loss/reg": 238.27737426757812,
+      "step": 240
+    },
+    {
+      "epoch": 0.006025,
+      "grad_norm": 2.5404036045074463,
+      "grad_norm_var": 8.421998600034392,
+      "learning_rate": 0.0001,
+      "loss": 240.354,
+      "loss/crossentropy": 2.509608507156372,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15739041566848755,
+      "loss/reg": 237.68695068359375,
+      "step": 241
+    },
+    {
+      "epoch": 0.00605,
+      "grad_norm": 2.88310170173645,
+      "grad_norm_var": 8.408739063367712,
+      "learning_rate": 0.0001,
+      "loss": 239.9893,
+      "loss/crossentropy": 2.714906692504883,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17041458189487457,
+      "loss/reg": 237.10398864746094,
+      "step": 242
+    },
+    {
+      "epoch": 0.006075,
+      "grad_norm": 2.5657036304473877,
+      "grad_norm_var": 8.465489057845808,
+      "learning_rate": 0.0001,
+      "loss": 239.4222,
+      "loss/crossentropy": 2.726501226425171,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17326998710632324,
+      "loss/reg": 236.5224151611328,
+      "step": 243
+    },
+    {
+      "epoch": 0.0061,
+      "grad_norm": 3.74009370803833,
+      "grad_norm_var": 8.41895240057901,
+      "learning_rate": 0.0001,
+      "loss": 238.9056,
+      "loss/crossentropy": 2.748196840286255,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21651685237884521,
+      "loss/reg": 235.94091796875,
+      "step": 244
+    },
+    {
+      "epoch": 0.006125,
+      "grad_norm": 2.7053475379943848,
+      "grad_norm_var": 8.444126473060866,
+      "learning_rate": 0.0001,
+      "loss": 238.032,
+      "loss/crossentropy": 2.4857230186462402,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1798083782196045,
+      "loss/reg": 235.36642456054688,
+      "step": 245
+    },
+    {
+      "epoch": 0.00615,
+      "grad_norm": 3.132035255432129,
+      "grad_norm_var": 8.405148171874606,
+      "learning_rate": 0.0001,
+      "loss": 237.9786,
+      "loss/crossentropy": 2.9837818145751953,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2059348225593567,
+      "loss/reg": 234.78887939453125,
+      "step": 246
+    },
+    {
+      "epoch": 0.006175,
+      "grad_norm": 2.4898736476898193,
+      "grad_norm_var": 8.467374226379958,
+      "learning_rate": 0.0001,
+      "loss": 237.2757,
+      "loss/crossentropy": 2.880187749862671,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17742741107940674,
+      "loss/reg": 234.21804809570312,
+      "step": 247
+    },
+    {
+      "epoch": 0.0062,
+      "grad_norm": 2.655679702758789,
+      "grad_norm_var": 8.482173935471064,
+      "learning_rate": 0.0001,
+      "loss": 236.6548,
+      "loss/crossentropy": 2.8338496685028076,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1855936199426651,
+      "loss/reg": 233.63536071777344,
+      "step": 248
+    },
+    {
+      "epoch": 0.006225,
+      "grad_norm": 2.869889259338379,
+      "grad_norm_var": 8.459100961664873,
+      "learning_rate": 0.0001,
+      "loss": 236.2722,
+      "loss/crossentropy": 3.00722074508667,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19864147901535034,
+      "loss/reg": 233.06637573242188,
+      "step": 249
+    },
+    {
+      "epoch": 0.00625,
+      "grad_norm": 2.8619699478149414,
+      "grad_norm_var": 8.431865311995418,
+      "learning_rate": 0.0001,
+      "loss": 235.1938,
+      "loss/crossentropy": 2.527130365371704,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18343941867351532,
+      "loss/reg": 232.48324584960938,
+      "step": 250
+    },
+    {
+      "epoch": 0.006275,
+      "grad_norm": 2.570348024368286,
+      "grad_norm_var": 8.445078379379984,
+      "learning_rate": 0.0001,
+      "loss": 234.9153,
+      "loss/crossentropy": 2.8627750873565674,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16326285898685455,
+      "loss/reg": 231.88925170898438,
+      "step": 251
+    },
+    {
+      "epoch": 0.0063,
+      "grad_norm": 3.70983624458313,
+      "grad_norm_var": 8.370411445912024,
+      "learning_rate": 0.0001,
+      "loss": 234.453,
+      "loss/crossentropy": 2.9494543075561523,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1932690143585205,
+      "loss/reg": 231.31028747558594,
+      "step": 252
+    },
+    {
+      "epoch": 0.006325,
+      "grad_norm": 2.636204957962036,
+      "grad_norm_var": 8.38834540620384,
+      "learning_rate": 0.0001,
+      "loss": 233.79,
+      "loss/crossentropy": 2.870368003845215,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18195605278015137,
+      "loss/reg": 230.73770141601562,
+      "step": 253
+    },
+    {
+      "epoch": 0.00635,
+      "grad_norm": 2.7411065101623535,
+      "grad_norm_var": 8.35122443905512,
+      "learning_rate": 0.0001,
+      "loss": 233.4192,
+      "loss/crossentropy": 3.0693013668060303,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17932993173599243,
+      "loss/reg": 230.17056274414062,
+      "step": 254
+    },
+    {
+      "epoch": 0.006375,
+      "grad_norm": 2.631800889968872,
+      "grad_norm_var": 0.1459736729753594,
+      "learning_rate": 0.0001,
+      "loss": 232.6018,
+      "loss/crossentropy": 2.812204360961914,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1809658408164978,
+      "loss/reg": 229.60862731933594,
+      "step": 255
+    },
+    {
+      "epoch": 0.0064,
+      "grad_norm": 2.6473164558410645,
+      "grad_norm_var": 0.14661806909008704,
+      "learning_rate": 0.0001,
+      "loss": 232.0037,
+      "loss/crossentropy": 2.8154137134552,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16016216576099396,
+      "loss/reg": 229.02810668945312,
+      "step": 256
+    },
+    {
+      "epoch": 0.006425,
+      "grad_norm": 2.8096730709075928,
+      "grad_norm_var": 0.14052644786456767,
+      "learning_rate": 0.0001,
+      "loss": 231.4434,
+      "loss/crossentropy": 2.8410181999206543,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.15286508202552795,
+      "loss/reg": 228.4495391845703,
+      "step": 257
+    },
+    {
+      "epoch": 0.00645,
+      "grad_norm": 2.902038335800171,
+      "grad_norm_var": 0.14062455090234588,
+      "learning_rate": 0.0001,
+      "loss": 230.797,
+      "loss/crossentropy": 2.71943998336792,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1998460292816162,
+      "loss/reg": 227.87771606445312,
+      "step": 258
+    },
+    {
+      "epoch": 0.006475,
+      "grad_norm": 2.605620861053467,
+      "grad_norm_var": 0.1391881027932695,
+      "learning_rate": 0.0001,
+      "loss": 230.5228,
+      "loss/crossentropy": 3.017008066177368,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18362455070018768,
+      "loss/reg": 227.32220458984375,
+      "step": 259
+    },
+    {
+      "epoch": 0.0065,
+      "grad_norm": 3.0760796070098877,
+      "grad_norm_var": 0.08854286659723035,
+      "learning_rate": 0.0001,
+      "loss": 229.5804,
+      "loss/crossentropy": 2.6169934272766113,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20932170748710632,
+      "loss/reg": 226.75404357910156,
+      "step": 260
+    },
+    {
+      "epoch": 0.006525,
+      "grad_norm": 2.7260305881500244,
+      "grad_norm_var": 0.08826637957723141,
+      "learning_rate": 0.0001,
+      "loss": 229.1744,
+      "loss/crossentropy": 2.800166368484497,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16581743955612183,
+      "loss/reg": 226.20838928222656,
+      "step": 261
+    },
+    {
+      "epoch": 0.00655,
+      "grad_norm": 2.520145893096924,
+      "grad_norm_var": 0.08593156013952026,
+      "learning_rate": 0.0001,
+      "loss": 228.4811,
+      "loss/crossentropy": 2.66050124168396,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17131495475769043,
+      "loss/reg": 225.64930725097656,
+      "step": 262
+    },
+    {
+      "epoch": 0.006575,
+      "grad_norm": 2.666048288345337,
+      "grad_norm_var": 0.08109508789745874,
+      "learning_rate": 0.0001,
+      "loss": 228.21,
+      "loss/crossentropy": 2.948183298110962,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1878964900970459,
+      "loss/reg": 225.0739288330078,
+      "step": 263
+    },
+    {
+      "epoch": 0.0066,
+      "grad_norm": 2.762319326400757,
+      "grad_norm_var": 0.07990506415366098,
+      "learning_rate": 0.0001,
+      "loss": 227.5642,
+      "loss/crossentropy": 2.8666961193084717,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19062572717666626,
+      "loss/reg": 224.5068817138672,
+      "step": 264
+    },
+    {
+      "epoch": 0.006625,
+      "grad_norm": 3.026460647583008,
+      "grad_norm_var": 0.08297919370627369,
+      "learning_rate": 0.0001,
+      "loss": 227.1483,
+      "loss/crossentropy": 3.004094362258911,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1900184154510498,
+      "loss/reg": 223.9541473388672,
+      "step": 265
+    },
+    {
+      "epoch": 0.00665,
+      "grad_norm": 2.7408447265625,
+      "grad_norm_var": 0.08298920636462223,
+      "learning_rate": 0.0001,
+      "loss": 226.3396,
+      "loss/crossentropy": 2.745393753051758,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20416490733623505,
+      "loss/reg": 223.39002990722656,
+      "step": 266
+    },
+    {
+      "epoch": 0.006675,
+      "grad_norm": 2.686131238937378,
+      "grad_norm_var": 0.0803088906266936,
+      "learning_rate": 0.0001,
+      "loss": 225.7382,
+      "loss/crossentropy": 2.730952262878418,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19333088397979736,
+      "loss/reg": 222.8138885498047,
+      "step": 267
+    },
+    {
+      "epoch": 0.0067,
+      "grad_norm": 3.0181736946105957,
+      "grad_norm_var": 0.026807333288596878,
+      "learning_rate": 0.0001,
+      "loss": 225.3613,
+      "loss/crossentropy": 2.9188590049743652,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2012787163257599,
+      "loss/reg": 222.24114990234375,
+      "step": 268
+    },
+    {
+      "epoch": 0.006725,
+      "grad_norm": 2.6647708415985107,
+      "grad_norm_var": 0.026378256663085863,
+      "learning_rate": 0.0001,
+      "loss": 224.5769,
+      "loss/crossentropy": 2.7387688159942627,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18105106055736542,
+      "loss/reg": 221.6570587158203,
+      "step": 269
+    },
+    {
+      "epoch": 0.00675,
+      "grad_norm": 3.005382776260376,
+      "grad_norm_var": 0.029935448640787853,
+      "learning_rate": 0.0001,
+      "loss": 223.7069,
+      "loss/crossentropy": 2.4513492584228516,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17486891150474548,
+      "loss/reg": 221.08065795898438,
+      "step": 270
+    },
+    {
+      "epoch": 0.006775,
+      "grad_norm": 2.534144163131714,
+      "grad_norm_var": 0.03246837804460086,
+      "learning_rate": 0.0001,
+      "loss": 223.3926,
+      "loss/crossentropy": 2.698563575744629,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1854040026664734,
+      "loss/reg": 220.50863647460938,
+      "step": 271
+    },
+    {
+      "epoch": 0.0068,
+      "grad_norm": 2.7375473976135254,
+      "grad_norm_var": 0.031447726867909864,
+      "learning_rate": 0.0001,
+      "loss": 223.1195,
+      "loss/crossentropy": 2.990828037261963,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19145527482032776,
+      "loss/reg": 219.937255859375,
+      "step": 272
+    },
+    {
+      "epoch": 0.006825,
+      "grad_norm": 2.594477415084839,
+      "grad_norm_var": 0.03349317904539077,
+      "learning_rate": 0.0001,
+      "loss": 222.1966,
+      "loss/crossentropy": 2.6549155712127686,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18063394725322723,
+      "loss/reg": 219.36102294921875,
+      "step": 273
+    },
+    {
+      "epoch": 0.00685,
+      "grad_norm": 3.0771312713623047,
+      "grad_norm_var": 0.03857028263729087,
+      "learning_rate": 0.0001,
+      "loss": 221.9057,
+      "loss/crossentropy": 2.904212713241577,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19918082654476166,
+      "loss/reg": 218.80227661132812,
+      "step": 274
+    },
+    {
+      "epoch": 0.006875,
+      "grad_norm": 2.548748731613159,
+      "grad_norm_var": 0.040076406355323495,
+      "learning_rate": 0.0001,
+      "loss": 221.1609,
+      "loss/crossentropy": 2.7445976734161377,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17896625399589539,
+      "loss/reg": 218.23736572265625,
+      "step": 275
+    },
+    {
+      "epoch": 0.0069,
+      "grad_norm": 3.4485270977020264,
+      "grad_norm_var": 0.0637460442930499,
+      "learning_rate": 0.0001,
+      "loss": 220.8704,
+      "loss/crossentropy": 2.9816956520080566,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19691282510757446,
+      "loss/reg": 217.6918182373047,
+      "step": 276
+    },
+    {
+      "epoch": 0.006925,
+      "grad_norm": 2.622490644454956,
+      "grad_norm_var": 0.06540004680390439,
+      "learning_rate": 0.0001,
+      "loss": 219.9821,
+      "loss/crossentropy": 2.6642203330993652,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1839863508939743,
+      "loss/reg": 217.13385009765625,
+      "step": 277
+    },
+    {
+      "epoch": 0.00695,
+      "grad_norm": 2.6010499000549316,
+      "grad_norm_var": 0.06288917084829108,
+      "learning_rate": 0.0001,
+      "loss": 219.4899,
+      "loss/crossentropy": 2.710230827331543,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19618386030197144,
+      "loss/reg": 216.58343505859375,
+      "step": 278
+    },
+    {
+      "epoch": 0.006975,
+      "grad_norm": 2.880260705947876,
+      "grad_norm_var": 0.06204859700921664,
+      "learning_rate": 0.0001,
+      "loss": 219.1407,
+      "loss/crossentropy": 2.898496150970459,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2143867015838623,
+      "loss/reg": 216.02783203125,
+      "step": 279
+    },
+    {
+      "epoch": 0.007,
+      "grad_norm": 3.233513593673706,
+      "grad_norm_var": 0.07297482892120115,
+      "learning_rate": 0.0001,
+      "loss": 218.5671,
+      "loss/crossentropy": 2.876899003982544,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20586279034614563,
+      "loss/reg": 215.48439025878906,
+      "step": 280
+    },
+    {
+      "epoch": 0.007025,
+      "grad_norm": 2.415728807449341,
+      "grad_norm_var": 0.08099970934078797,
+      "learning_rate": 0.0001,
+      "loss": 217.7508,
+      "loss/crossentropy": 2.639965057373047,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1753409504890442,
+      "loss/reg": 214.93548583984375,
+      "step": 281
+    },
+    {
+      "epoch": 0.00705,
+      "grad_norm": 2.7094812393188477,
+      "grad_norm_var": 0.08131089617161227,
+      "learning_rate": 0.0001,
+      "loss": 217.2597,
+      "loss/crossentropy": 2.687687873840332,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18249084055423737,
+      "loss/reg": 214.38949584960938,
+      "step": 282
+    },
+    {
+      "epoch": 0.007075,
+      "grad_norm": 2.616089105606079,
+      "grad_norm_var": 0.08266783158203429,
+      "learning_rate": 0.0001,
+      "loss": 216.6144,
+      "loss/crossentropy": 2.6135380268096924,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1692454218864441,
+      "loss/reg": 213.83157348632812,
+      "step": 283
+    },
+    {
+      "epoch": 0.0071,
+      "grad_norm": 2.598839521408081,
+      "grad_norm_var": 0.08113636926467128,
+      "learning_rate": 0.0001,
+      "loss": 216.4407,
+      "loss/crossentropy": 2.969308853149414,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18686597049236298,
+      "loss/reg": 213.28448486328125,
+      "step": 284
+    },
+    {
+      "epoch": 0.007125,
+      "grad_norm": 2.68222713470459,
+      "grad_norm_var": 0.0809151212890202,
+      "learning_rate": 0.0001,
+      "loss": 215.8076,
+      "loss/crossentropy": 2.8860280513763428,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17992877960205078,
+      "loss/reg": 212.74166870117188,
+      "step": 285
+    },
+    {
+      "epoch": 0.00715,
+      "grad_norm": 2.798988103866577,
+      "grad_norm_var": 0.07707527762625711,
+      "learning_rate": 0.0001,
+      "loss": 215.3757,
+      "loss/crossentropy": 2.9692792892456055,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21567609906196594,
+      "loss/reg": 212.19068908691406,
+      "step": 286
+    },
+    {
+      "epoch": 0.007175,
+      "grad_norm": 2.5780839920043945,
+      "grad_norm_var": 0.07589498443210232,
+      "learning_rate": 0.0001,
+      "loss": 214.3753,
+      "loss/crossentropy": 2.5596561431884766,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.177793949842453,
+      "loss/reg": 211.63784790039062,
+      "step": 287
+    },
+    {
+      "epoch": 0.0072,
+      "grad_norm": 2.6318469047546387,
+      "grad_norm_var": 0.07689489347470727,
+      "learning_rate": 0.0001,
+      "loss": 214.0618,
+      "loss/crossentropy": 2.7902228832244873,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18190298974514008,
+      "loss/reg": 211.0896759033203,
+      "step": 288
+    },
+    {
+      "epoch": 0.007225,
+      "grad_norm": 3.0802104473114014,
+      "grad_norm_var": 0.08141687457795115,
+      "learning_rate": 0.0001,
+      "loss": 213.4452,
+      "loss/crossentropy": 2.7214579582214355,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18961231410503387,
+      "loss/reg": 210.5341796875,
+      "step": 289
+    },
+    {
+      "epoch": 0.00725,
+      "grad_norm": 3.087554693222046,
+      "grad_norm_var": 0.08183286107987228,
+      "learning_rate": 0.0001,
+      "loss": 213.064,
+      "loss/crossentropy": 2.886679172515869,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17951728403568268,
+      "loss/reg": 209.9978485107422,
+      "step": 290
+    },
+    {
+      "epoch": 0.007275,
+      "grad_norm": 2.7192165851593018,
+      "grad_norm_var": 0.07831674565225223,
+      "learning_rate": 0.0001,
+      "loss": 212.4935,
+      "loss/crossentropy": 2.835261821746826,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19229570031166077,
+      "loss/reg": 209.4658966064453,
+      "step": 291
+    },
+    {
+      "epoch": 0.0073,
+      "grad_norm": 2.9835152626037598,
+      "grad_norm_var": 0.051250203523905684,
+      "learning_rate": 0.0001,
+      "loss": 211.8015,
+      "loss/crossentropy": 2.691941499710083,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18396204710006714,
+      "loss/reg": 208.92559814453125,
+      "step": 292
+    },
+    {
+      "epoch": 0.007325,
+      "grad_norm": 3.2173898220062256,
+      "grad_norm_var": 0.062069919928694615,
+      "learning_rate": 0.0001,
+      "loss": 211.2997,
+      "loss/crossentropy": 2.6862618923187256,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22537732124328613,
+      "loss/reg": 208.3881072998047,
+      "step": 293
+    },
+    {
+      "epoch": 0.00735,
+      "grad_norm": 3.5398621559143066,
+      "grad_norm_var": 0.09198591178203515,
+      "learning_rate": 0.0001,
+      "loss": 210.6651,
+      "loss/crossentropy": 2.612825393676758,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19290409982204437,
+      "loss/reg": 207.85940551757812,
+      "step": 294
+    },
+    {
+      "epoch": 0.007375,
+      "grad_norm": 2.561976671218872,
+      "grad_norm_var": 0.09749160768811103,
+      "learning_rate": 0.0001,
+      "loss": 210.1769,
+      "loss/crossentropy": 2.6648244857788086,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20216211676597595,
+      "loss/reg": 207.30992126464844,
+      "step": 295
+    },
+    {
+      "epoch": 0.0074,
+      "grad_norm": 7.271378040313721,
+      "grad_norm_var": 1.3278853272864029,
+      "learning_rate": 0.0001,
+      "loss": 209.9314,
+      "loss/crossentropy": 2.9801483154296875,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19208793342113495,
+      "loss/reg": 206.75917053222656,
+      "step": 296
+    },
+    {
+      "epoch": 0.007425,
+      "grad_norm": 3.091381788253784,
+      "grad_norm_var": 1.295378929230882,
+      "learning_rate": 0.0001,
+      "loss": 209.1877,
+      "loss/crossentropy": 2.7932381629943848,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18018564581871033,
+      "loss/reg": 206.21424865722656,
+      "step": 297
+    },
+    {
+      "epoch": 0.00745,
+      "grad_norm": 2.8066790103912354,
+      "grad_norm_var": 1.2904482820319183,
+      "learning_rate": 0.0001,
+      "loss": 208.8806,
+      "loss/crossentropy": 3.017073392868042,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19379809498786926,
+      "loss/reg": 205.66973876953125,
+      "step": 298
+    },
+    {
+      "epoch": 0.007475,
+      "grad_norm": 2.856220006942749,
+      "grad_norm_var": 1.2772274114279487,
+      "learning_rate": 0.0001,
+      "loss": 208.1113,
+      "loss/crossentropy": 2.81599497795105,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18530045449733734,
+      "loss/reg": 205.11001586914062,
+      "step": 299
+    },
+    {
+      "epoch": 0.0075,
+      "grad_norm": 2.9903173446655273,
+      "grad_norm_var": 1.2576931439437788,
+      "learning_rate": 0.0001,
+      "loss": 207.5814,
+      "loss/crossentropy": 2.826772451400757,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19662591814994812,
+      "loss/reg": 204.55799865722656,
+      "step": 300
+    },
+    {
+      "epoch": 0.007525,
+      "grad_norm": 2.5510308742523193,
+      "grad_norm_var": 1.2674948009951321,
+      "learning_rate": 0.0001,
+      "loss": 206.9224,
+      "loss/crossentropy": 2.712320327758789,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19106364250183105,
+      "loss/reg": 204.01902770996094,
+      "step": 301
+    },
+    {
+      "epoch": 0.00755,
+      "grad_norm": 2.508366107940674,
+      "grad_norm_var": 1.2872607464713248,
+      "learning_rate": 0.0001,
+      "loss": 206.1158,
+      "loss/crossentropy": 2.4741883277893066,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17461839318275452,
+      "loss/reg": 203.46702575683594,
+      "step": 302
+    },
+    {
+      "epoch": 0.007575,
+      "grad_norm": 2.976346969604492,
+      "grad_norm_var": 1.266555341337554,
+      "learning_rate": 0.0001,
+      "loss": 205.8845,
+      "loss/crossentropy": 2.732536554336548,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2265537679195404,
+      "loss/reg": 202.92544555664062,
+      "step": 303
+    },
+    {
+      "epoch": 0.0076,
+      "grad_norm": 2.8182311058044434,
+      "grad_norm_var": 1.2551146741564971,
+      "learning_rate": 0.0001,
+      "loss": 205.3437,
+      "loss/crossentropy": 2.774559736251831,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19560466706752777,
+      "loss/reg": 202.37356567382812,
+      "step": 304
+    },
+    {
+      "epoch": 0.007625,
+      "grad_norm": 2.6641106605529785,
+      "grad_norm_var": 1.2720952145177449,
+      "learning_rate": 0.0001,
+      "loss": 204.813,
+      "loss/crossentropy": 2.7902915477752686,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18885600566864014,
+      "loss/reg": 201.83387756347656,
+      "step": 305
+    },
+    {
+      "epoch": 0.00765,
+      "grad_norm": 2.7573277950286865,
+      "grad_norm_var": 1.2823306075296226,
+      "learning_rate": 0.0001,
+      "loss": 204.3929,
+      "loss/crossentropy": 2.89617919921875,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20617598295211792,
+      "loss/reg": 201.29051208496094,
+      "step": 306
+    },
+    {
+      "epoch": 0.007675,
+      "grad_norm": 2.9883832931518555,
+      "grad_norm_var": 1.2715927971824363,
+      "learning_rate": 0.0001,
+      "loss": 203.6661,
+      "loss/crossentropy": 2.736328363418579,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17194727063179016,
+      "loss/reg": 200.75784301757812,
+      "step": 307
+    },
+    {
+      "epoch": 0.0077,
+      "grad_norm": 2.671340227127075,
+      "grad_norm_var": 1.2850880861748617,
+      "learning_rate": 0.0001,
+      "loss": 203.2853,
+      "loss/crossentropy": 2.837782859802246,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20544278621673584,
+      "loss/reg": 200.2420654296875,
+      "step": 308
+    },
+    {
+      "epoch": 0.007725,
+      "grad_norm": 2.5738258361816406,
+      "grad_norm_var": 1.304496016414119,
+      "learning_rate": 0.0001,
+      "loss": 202.6491,
+      "loss/crossentropy": 2.764209270477295,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17526961863040924,
+      "loss/reg": 199.7096710205078,
+      "step": 309
+    },
+    {
+      "epoch": 0.00775,
+      "grad_norm": 2.727646827697754,
+      "grad_norm_var": 1.2982730821691781,
+      "learning_rate": 0.0001,
+      "loss": 201.8776,
+      "loss/crossentropy": 2.507113456726074,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18204720318317413,
+      "loss/reg": 199.1884765625,
+      "step": 310
+    },
+    {
+      "epoch": 0.007775,
+      "grad_norm": 3.019986629486084,
+      "grad_norm_var": 1.2815257147172248,
+      "learning_rate": 0.0001,
+      "loss": 201.4175,
+      "loss/crossentropy": 2.580369234085083,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18372410535812378,
+      "loss/reg": 198.6533660888672,
+      "step": 311
+    },
+    {
+      "epoch": 0.0078,
+      "grad_norm": 4.663088321685791,
+      "grad_norm_var": 0.24891895975728176,
+      "learning_rate": 0.0001,
+      "loss": 201.1794,
+      "loss/crossentropy": 2.814059019088745,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23769119381904602,
+      "loss/reg": 198.1276092529297,
+      "step": 312
+    },
+    {
+      "epoch": 0.007825,
+      "grad_norm": 2.9102060794830322,
+      "grad_norm_var": 0.24674634897498038,
+      "learning_rate": 0.0001,
+      "loss": 200.4867,
+      "loss/crossentropy": 2.6932103633880615,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1972815990447998,
+      "loss/reg": 197.5962371826172,
+      "step": 313
+    },
+    {
+      "epoch": 0.00785,
+      "grad_norm": 2.906273603439331,
+      "grad_norm_var": 0.24605808105580232,
+      "learning_rate": 0.0001,
+      "loss": 200.3487,
+      "loss/crossentropy": 3.0405361652374268,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22804442048072815,
+      "loss/reg": 197.0801544189453,
+      "step": 314
+    },
+    {
+      "epoch": 0.007875,
+      "grad_norm": 2.5689239501953125,
+      "grad_norm_var": 0.2533312249281851,
+      "learning_rate": 0.0001,
+      "loss": 199.4335,
+      "loss/crossentropy": 2.7003679275512695,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1923004686832428,
+      "loss/reg": 196.54086303710938,
+      "step": 315
+    },
+    {
+      "epoch": 0.0079,
+      "grad_norm": 2.8070003986358643,
+      "grad_norm_var": 0.2530642011976994,
+      "learning_rate": 0.0001,
+      "loss": 199.2196,
+      "loss/crossentropy": 3.0146713256835938,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1969759613275528,
+      "loss/reg": 196.0079803466797,
+      "step": 316
+    },
+    {
+      "epoch": 0.007925,
+      "grad_norm": 2.6455047130584717,
+      "grad_norm_var": 0.2494529065085686,
+      "learning_rate": 0.0001,
+      "loss": 198.6955,
+      "loss/crossentropy": 3.0136356353759766,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20378613471984863,
+      "loss/reg": 195.4780731201172,
+      "step": 317
+    },
+    {
+      "epoch": 0.00795,
+      "grad_norm": 2.6797735691070557,
+      "grad_norm_var": 0.2426149646107582,
+      "learning_rate": 0.0001,
+      "loss": 197.5625,
+      "loss/crossentropy": 2.4390504360198975,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17309829592704773,
+      "loss/reg": 194.95030212402344,
+      "step": 318
+    },
+    {
+      "epoch": 0.007975,
+      "grad_norm": 2.668614149093628,
+      "grad_norm_var": 0.2453445922218293,
+      "learning_rate": 0.0001,
+      "loss": 197.1961,
+      "loss/crossentropy": 2.5966238975524902,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1794152557849884,
+      "loss/reg": 194.42005920410156,
+      "step": 319
+    },
+    {
+      "epoch": 0.008,
+      "grad_norm": 2.552947521209717,
+      "grad_norm_var": 0.2519063072997559,
+      "learning_rate": 0.0001,
+      "loss": 196.5726,
+      "loss/crossentropy": 2.5113775730133057,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16936884820461273,
+      "loss/reg": 193.891845703125,
+      "step": 320
+    },
+    {
+      "epoch": 0.008025,
+      "grad_norm": 2.836556911468506,
+      "grad_norm_var": 0.2491962625539448,
+      "learning_rate": 0.0001,
+      "loss": 196.1308,
+      "loss/crossentropy": 2.5704989433288574,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19436271488666534,
+      "loss/reg": 193.3658905029297,
+      "step": 321
+    },
+    {
+      "epoch": 0.00805,
+      "grad_norm": 2.6601991653442383,
+      "grad_norm_var": 0.25129150502635655,
+      "learning_rate": 0.0001,
+      "loss": 196.0067,
+      "loss/crossentropy": 2.9537734985351562,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21296489238739014,
+      "loss/reg": 192.83993530273438,
+      "step": 322
+    },
+    {
+      "epoch": 0.008075,
+      "grad_norm": 2.7744781970977783,
+      "grad_norm_var": 0.25070402667003294,
+      "learning_rate": 0.0001,
+      "loss": 195.5274,
+      "loss/crossentropy": 3.016360282897949,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19973087310791016,
+      "loss/reg": 192.3113250732422,
+      "step": 323
+    },
+    {
+      "epoch": 0.0081,
+      "grad_norm": 3.1214191913604736,
+      "grad_norm_var": 0.252394334847159,
+      "learning_rate": 0.0001,
+      "loss": 195.3513,
+      "loss/crossentropy": 3.329318046569824,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23599132895469666,
+      "loss/reg": 191.78594970703125,
+      "step": 324
+    },
+    {
+      "epoch": 0.008125,
+      "grad_norm": 2.6547012329101562,
+      "grad_norm_var": 0.2494769798717158,
+      "learning_rate": 0.0001,
+      "loss": 194.3078,
+      "loss/crossentropy": 2.844395399093628,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20515291392803192,
+      "loss/reg": 191.2582550048828,
+      "step": 325
+    },
+    {
+      "epoch": 0.00815,
+      "grad_norm": 2.800093650817871,
+      "grad_norm_var": 0.24826251752535028,
+      "learning_rate": 0.0001,
+      "loss": 193.7962,
+      "loss/crossentropy": 2.856586217880249,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1940813660621643,
+      "loss/reg": 190.74557495117188,
+      "step": 326
+    },
+    {
+      "epoch": 0.008175,
+      "grad_norm": 2.558241844177246,
+      "grad_norm_var": 0.2536998205740133,
+      "learning_rate": 0.0001,
+      "loss": 192.9891,
+      "loss/crossentropy": 2.562021493911743,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18647362291812897,
+      "loss/reg": 190.24061584472656,
+      "step": 327
+    },
+    {
+      "epoch": 0.0082,
+      "grad_norm": 2.952324390411377,
+      "grad_norm_var": 0.026016228267095206,
+      "learning_rate": 0.0001,
+      "loss": 192.9934,
+      "loss/crossentropy": 3.0561769008636475,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20818649232387543,
+      "loss/reg": 189.72903442382812,
+      "step": 328
+    },
+    {
+      "epoch": 0.008225,
+      "grad_norm": 3.3000900745391846,
+      "grad_norm_var": 0.043529065715877856,
+      "learning_rate": 0.0001,
+      "loss": 192.3072,
+      "loss/crossentropy": 2.8769357204437256,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21247068047523499,
+      "loss/reg": 189.21775817871094,
+      "step": 329
+    },
+    {
+      "epoch": 0.00825,
+      "grad_norm": 2.693880081176758,
+      "grad_norm_var": 0.04278518629817896,
+      "learning_rate": 0.0001,
+      "loss": 191.9053,
+      "loss/crossentropy": 2.990717649459839,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20985952019691467,
+      "loss/reg": 188.7047119140625,
+      "step": 330
+    },
+    {
+      "epoch": 0.008275,
+      "grad_norm": 2.718574285507202,
+      "grad_norm_var": 0.04022917313372621,
+      "learning_rate": 0.0001,
+      "loss": 191.1251,
+      "loss/crossentropy": 2.72544527053833,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18823279440402985,
+      "loss/reg": 188.21141052246094,
+      "step": 331
+    },
+    {
+      "epoch": 0.0083,
+      "grad_norm": 2.4027438163757324,
+      "grad_norm_var": 0.04880048181735764,
+      "learning_rate": 0.0001,
+      "loss": 190.5635,
+      "loss/crossentropy": 2.6736438274383545,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16881534457206726,
+      "loss/reg": 187.72108459472656,
+      "step": 332
+    },
+    {
+      "epoch": 0.008325,
+      "grad_norm": 4.033091068267822,
+      "grad_norm_var": 0.14957197834728883,
+      "learning_rate": 0.0001,
+      "loss": 190.4425,
+      "loss/crossentropy": 2.9758944511413574,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23931945860385895,
+      "loss/reg": 187.22726440429688,
+      "step": 333
+    },
+    {
+      "epoch": 0.00835,
+      "grad_norm": 2.87990140914917,
+      "grad_norm_var": 0.14785355856326585,
+      "learning_rate": 0.0001,
+      "loss": 189.6929,
+      "loss/crossentropy": 2.762582302093506,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1953504979610443,
+      "loss/reg": 186.7349853515625,
+      "step": 334
+    },
+    {
+      "epoch": 0.008375,
+      "grad_norm": 2.569755792617798,
+      "grad_norm_var": 0.15086170983481276,
+      "learning_rate": 0.0001,
+      "loss": 189.2558,
+      "loss/crossentropy": 2.8176991939544678,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2055705040693283,
+      "loss/reg": 186.23252868652344,
+      "step": 335
+    },
+    {
+      "epoch": 0.0084,
+      "grad_norm": 2.7810733318328857,
+      "grad_norm_var": 0.14525191609682603,
+      "learning_rate": 0.0001,
+      "loss": 188.7248,
+      "loss/crossentropy": 2.7931947708129883,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20690809190273285,
+      "loss/reg": 185.72470092773438,
+      "step": 336
+    },
+    {
+      "epoch": 0.008425,
+      "grad_norm": 2.921773910522461,
+      "grad_norm_var": 0.1454556654203382,
+      "learning_rate": 0.0001,
+      "loss": 188.15,
+      "loss/crossentropy": 2.723609447479248,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2181951254606247,
+      "loss/reg": 185.20816040039062,
+      "step": 337
+    },
+    {
+      "epoch": 0.00845,
+      "grad_norm": 3.092437505722046,
+      "grad_norm_var": 0.14539310884848244,
+      "learning_rate": 0.0001,
+      "loss": 187.8704,
+      "loss/crossentropy": 2.962968111038208,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2084237039089203,
+      "loss/reg": 184.69903564453125,
+      "step": 338
+    },
+    {
+      "epoch": 0.008475,
+      "grad_norm": 3.1183600425720215,
+      "grad_norm_var": 0.14744546795223623,
+      "learning_rate": 0.0001,
+      "loss": 187.327,
+      "loss/crossentropy": 2.9327218532562256,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21510222554206848,
+      "loss/reg": 184.17919921875,
+      "step": 339
+    },
+    {
+      "epoch": 0.0085,
+      "grad_norm": 3.118776559829712,
+      "grad_norm_var": 0.14737225768415765,
+      "learning_rate": 0.0001,
+      "loss": 186.937,
+      "loss/crossentropy": 3.054023265838623,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21871572732925415,
+      "loss/reg": 183.66429138183594,
+      "step": 340
+    },
+    {
+      "epoch": 0.008525,
+      "grad_norm": 3.009570360183716,
+      "grad_norm_var": 0.14305740155455335,
+      "learning_rate": 0.0001,
+      "loss": 186.0073,
+      "loss/crossentropy": 2.6760642528533936,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1764220893383026,
+      "loss/reg": 183.15478515625,
+      "step": 341
+    },
+    {
+      "epoch": 0.00855,
+      "grad_norm": 2.6750733852386475,
+      "grad_norm_var": 0.14627338296657608,
+      "learning_rate": 0.0001,
+      "loss": 185.687,
+      "loss/crossentropy": 2.8624348640441895,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1818752884864807,
+      "loss/reg": 182.64266967773438,
+      "step": 342
+    },
+    {
+      "epoch": 0.008575,
+      "grad_norm": 2.536036252975464,
+      "grad_norm_var": 0.14739482829090184,
+      "learning_rate": 0.0001,
+      "loss": 185.0605,
+      "loss/crossentropy": 2.755174398422241,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1642666757106781,
+      "loss/reg": 182.1410369873047,
+      "step": 343
+    },
+    {
+      "epoch": 0.0086,
+      "grad_norm": 2.7478532791137695,
+      "grad_norm_var": 0.14926881515059046,
+      "learning_rate": 0.0001,
+      "loss": 184.4167,
+      "loss/crossentropy": 2.5819690227508545,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19197826087474823,
+      "loss/reg": 181.64279174804688,
+      "step": 344
+    },
+    {
+      "epoch": 0.008625,
+      "grad_norm": 2.914630651473999,
+      "grad_norm_var": 0.1386317271370025,
+      "learning_rate": 0.0001,
+      "loss": 184.1337,
+      "loss/crossentropy": 2.7838339805603027,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19958782196044922,
+      "loss/reg": 181.15025329589844,
+      "step": 345
+    },
+    {
+      "epoch": 0.00865,
+      "grad_norm": 2.6812732219696045,
+      "grad_norm_var": 0.13896854058942837,
+      "learning_rate": 0.0001,
+      "loss": 183.8571,
+      "loss/crossentropy": 3.0270869731903076,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17617946863174438,
+      "loss/reg": 180.65380859375,
+      "step": 346
+    },
+    {
+      "epoch": 0.008675,
+      "grad_norm": 2.7997968196868896,
+      "grad_norm_var": 0.13755082338921434,
+      "learning_rate": 0.0001,
+      "loss": 183.1842,
+      "loss/crossentropy": 2.830026865005493,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2012362778186798,
+      "loss/reg": 180.15289306640625,
+      "step": 347
+    },
+    {
+      "epoch": 0.0087,
+      "grad_norm": 2.632431745529175,
+      "grad_norm_var": 0.1258451860116639,
+      "learning_rate": 0.0001,
+      "loss": 182.56,
+      "loss/crossentropy": 2.704882860183716,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1936817169189453,
+      "loss/reg": 179.66139221191406,
+      "step": 348
+    },
+    {
+      "epoch": 0.008725,
+      "grad_norm": 2.971341371536255,
+      "grad_norm_var": 0.036883841878242646,
+      "learning_rate": 0.0001,
+      "loss": 182.6145,
+      "loss/crossentropy": 3.2539191246032715,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19264021515846252,
+      "loss/reg": 179.1678924560547,
+      "step": 349
+    },
+    {
+      "epoch": 0.00875,
+      "grad_norm": 3.221050500869751,
+      "grad_norm_var": 0.04594406550980518,
+      "learning_rate": 0.0001,
+      "loss": 181.849,
+      "loss/crossentropy": 2.9532337188720703,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2272038757801056,
+      "loss/reg": 178.6685333251953,
+      "step": 350
+    },
+    {
+      "epoch": 0.008775,
+      "grad_norm": 2.788480520248413,
+      "grad_norm_var": 0.04041268740233453,
+      "learning_rate": 0.0001,
+      "loss": 181.0548,
+      "loss/crossentropy": 2.6735174655914307,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20720481872558594,
+      "loss/reg": 178.17410278320312,
+      "step": 351
+    },
+    {
+      "epoch": 0.0088,
+      "grad_norm": 2.8164358139038086,
+      "grad_norm_var": 0.040045045030925056,
+      "learning_rate": 0.0001,
+      "loss": 180.8875,
+      "loss/crossentropy": 3.0074617862701416,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20497429370880127,
+      "loss/reg": 177.67507934570312,
+      "step": 352
+    },
+    {
+      "epoch": 0.008825,
+      "grad_norm": 2.459709882736206,
+      "grad_norm_var": 0.05068183434945368,
+      "learning_rate": 0.0001,
+      "loss": 180.1328,
+      "loss/crossentropy": 2.7839858531951904,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16963033378124237,
+      "loss/reg": 177.17913818359375,
+      "step": 353
+    },
+    {
+      "epoch": 0.00885,
+      "grad_norm": 2.582552909851074,
+      "grad_norm_var": 0.05037757045170774,
+      "learning_rate": 0.0001,
+      "loss": 179.5129,
+      "loss/crossentropy": 2.63332200050354,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18901947140693665,
+      "loss/reg": 176.6905975341797,
+      "step": 354
+    },
+    {
+      "epoch": 0.008875,
+      "grad_norm": 2.6625685691833496,
+      "grad_norm_var": 0.04505259166354089,
+      "learning_rate": 0.0001,
+      "loss": 179.0017,
+      "loss/crossentropy": 2.6259818077087402,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1885865181684494,
+      "loss/reg": 176.18716430664062,
+      "step": 355
+    },
+    {
+      "epoch": 0.0089,
+      "grad_norm": 2.9031405448913574,
+      "grad_norm_var": 0.03846567871036252,
+      "learning_rate": 0.0001,
+      "loss": 178.6655,
+      "loss/crossentropy": 2.7855353355407715,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19966638088226318,
+      "loss/reg": 175.6802520751953,
+      "step": 356
+    },
+    {
+      "epoch": 0.008925,
+      "grad_norm": 2.6948978900909424,
+      "grad_norm_var": 0.03481774262957226,
+      "learning_rate": 0.0001,
+      "loss": 178.1122,
+      "loss/crossentropy": 2.7389872074127197,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20075306296348572,
+      "loss/reg": 175.1724090576172,
+      "step": 357
+    },
+    {
+      "epoch": 0.00895,
+      "grad_norm": 2.755830764770508,
+      "grad_norm_var": 0.03435983560266701,
+      "learning_rate": 0.0001,
+      "loss": 177.4502,
+      "loss/crossentropy": 2.594728946685791,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19329878687858582,
+      "loss/reg": 174.6621551513672,
+      "step": 358
+    },
+    {
+      "epoch": 0.008975,
+      "grad_norm": 2.4237000942230225,
+      "grad_norm_var": 0.038510630346210446,
+      "learning_rate": 0.0001,
+      "loss": 176.9384,
+      "loss/crossentropy": 2.599057197570801,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1731238067150116,
+      "loss/reg": 174.16619873046875,
+      "step": 359
+    },
+    {
+      "epoch": 0.009,
+      "grad_norm": 2.7944130897521973,
+      "grad_norm_var": 0.03861118264588394,
+      "learning_rate": 0.0001,
+      "loss": 176.9084,
+      "loss/crossentropy": 2.9929091930389404,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24037709832191467,
+      "loss/reg": 173.67515563964844,
+      "step": 360
+    },
+    {
+      "epoch": 0.009025,
+      "grad_norm": 2.8927605152130127,
+      "grad_norm_var": 0.03817964658272786,
+      "learning_rate": 0.0001,
+      "loss": 176.0866,
+      "loss/crossentropy": 2.695925712585449,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2011643946170807,
+      "loss/reg": 173.1894989013672,
+      "step": 361
+    },
+    {
+      "epoch": 0.00905,
+      "grad_norm": 2.654205799102783,
+      "grad_norm_var": 0.038491602775841474,
+      "learning_rate": 0.0001,
+      "loss": 175.8716,
+      "loss/crossentropy": 2.9696269035339355,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19813314080238342,
+      "loss/reg": 172.70388793945312,
+      "step": 362
+    },
+    {
+      "epoch": 0.009075,
+      "grad_norm": 2.500924587249756,
+      "grad_norm_var": 0.0422227970984134,
+      "learning_rate": 0.0001,
+      "loss": 175.0925,
+      "loss/crossentropy": 2.694688558578491,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19237878918647766,
+      "loss/reg": 172.20538330078125,
+      "step": 363
+    },
+    {
+      "epoch": 0.0091,
+      "grad_norm": 2.856943130493164,
+      "grad_norm_var": 0.04231316052960944,
+      "learning_rate": 0.0001,
+      "loss": 174.687,
+      "loss/crossentropy": 2.7701804637908936,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19237452745437622,
+      "loss/reg": 171.72442626953125,
+      "step": 364
+    },
+    {
+      "epoch": 0.009125,
+      "grad_norm": 3.0055325031280518,
+      "grad_norm_var": 0.043401276039503264,
+      "learning_rate": 0.0001,
+      "loss": 174.6459,
+      "loss/crossentropy": 3.1644747257232666,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24399538338184357,
+      "loss/reg": 171.23745727539062,
+      "step": 365
+    },
+    {
+      "epoch": 0.00915,
+      "grad_norm": 2.485426664352417,
+      "grad_norm_var": 0.031101142054486056,
+      "learning_rate": 0.0001,
+      "loss": 173.739,
+      "loss/crossentropy": 2.804879903793335,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1790430247783661,
+      "loss/reg": 170.75502014160156,
+      "step": 366
+    },
+    {
+      "epoch": 0.009175,
+      "grad_norm": 2.922252655029297,
+      "grad_norm_var": 0.033711321846642286,
+      "learning_rate": 0.0001,
+      "loss": 173.451,
+      "loss/crossentropy": 2.9688096046447754,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20513179898262024,
+      "loss/reg": 170.2770538330078,
+      "step": 367
+    },
+    {
+      "epoch": 0.0092,
+      "grad_norm": 2.730825662612915,
+      "grad_norm_var": 0.032991054055673645,
+      "learning_rate": 0.0001,
+      "loss": 172.7964,
+      "loss/crossentropy": 2.7933456897735596,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20646029710769653,
+      "loss/reg": 169.796630859375,
+      "step": 368
+    },
+    {
+      "epoch": 0.009225,
+      "grad_norm": 2.482426881790161,
+      "grad_norm_var": 0.03227169195139723,
+      "learning_rate": 0.0001,
+      "loss": 172.0632,
+      "loss/crossentropy": 2.5588743686676025,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18388806283473969,
+      "loss/reg": 169.3204803466797,
+      "step": 369
+    },
+    {
+      "epoch": 0.00925,
+      "grad_norm": 2.739718198776245,
+      "grad_norm_var": 0.03115998847033552,
+      "learning_rate": 0.0001,
+      "loss": 171.5819,
+      "loss/crossentropy": 2.54241681098938,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19010567665100098,
+      "loss/reg": 168.84933471679688,
+      "step": 370
+    },
+    {
+      "epoch": 0.009275,
+      "grad_norm": 2.8765339851379395,
+      "grad_norm_var": 0.03240860179928783,
+      "learning_rate": 0.0001,
+      "loss": 171.0873,
+      "loss/crossentropy": 2.5227410793304443,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18850305676460266,
+      "loss/reg": 168.37608337402344,
+      "step": 371
+    },
+    {
+      "epoch": 0.0093,
+      "grad_norm": 2.648451089859009,
+      "grad_norm_var": 0.030667067483867782,
+      "learning_rate": 0.0001,
+      "loss": 170.7962,
+      "loss/crossentropy": 2.701418161392212,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18235236406326294,
+      "loss/reg": 167.9124755859375,
+      "step": 372
+    },
+    {
+      "epoch": 0.009325,
+      "grad_norm": 2.3994297981262207,
+      "grad_norm_var": 0.03697651271872549,
+      "learning_rate": 0.0001,
+      "loss": 170.5056,
+      "loss/crossentropy": 2.8883211612701416,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17240813374519348,
+      "loss/reg": 167.44482421875,
+      "step": 373
+    },
+    {
+      "epoch": 0.00935,
+      "grad_norm": 2.760050058364868,
+      "grad_norm_var": 0.03701011100701498,
+      "learning_rate": 0.0001,
+      "loss": 170.0168,
+      "loss/crossentropy": 2.828566551208496,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20667554438114166,
+      "loss/reg": 166.98158264160156,
+      "step": 374
+    },
+    {
+      "epoch": 0.009375,
+      "grad_norm": 2.598747968673706,
+      "grad_norm_var": 0.032514977652635696,
+      "learning_rate": 0.0001,
+      "loss": 169.915,
+      "loss/crossentropy": 3.2055323123931885,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18947294354438782,
+      "loss/reg": 166.51995849609375,
+      "step": 375
+    },
+    {
+      "epoch": 0.0094,
+      "grad_norm": 2.5448288917541504,
+      "grad_norm_var": 0.03357553294952117,
+      "learning_rate": 0.0001,
+      "loss": 169.237,
+      "loss/crossentropy": 2.9864261150360107,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19935306906700134,
+      "loss/reg": 166.05117797851562,
+      "step": 376
+    },
+    {
+      "epoch": 0.009425,
+      "grad_norm": 2.9218602180480957,
+      "grad_norm_var": 0.034400838745598135,
+      "learning_rate": 0.0001,
+      "loss": 169.3156,
+      "loss/crossentropy": 3.513453722000122,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21727558970451355,
+      "loss/reg": 165.5848846435547,
+      "step": 377
+    },
+    {
+      "epoch": 0.00945,
+      "grad_norm": 2.6745686531066895,
+      "grad_norm_var": 0.03431461157821616,
+      "learning_rate": 0.0001,
+      "loss": 167.9062,
+      "loss/crossentropy": 2.607114791870117,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17556855082511902,
+      "loss/reg": 165.1234893798828,
+      "step": 378
+    },
+    {
+      "epoch": 0.009475,
+      "grad_norm": 2.8298020362854004,
+      "grad_norm_var": 0.03248619908528383,
+      "learning_rate": 0.0001,
+      "loss": 167.8138,
+      "loss/crossentropy": 2.9435441493988037,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20177403092384338,
+      "loss/reg": 164.66851806640625,
+      "step": 379
+    },
+    {
+      "epoch": 0.0095,
+      "grad_norm": 2.4895458221435547,
+      "grad_norm_var": 0.03408372867624981,
+      "learning_rate": 0.0001,
+      "loss": 167.1126,
+      "loss/crossentropy": 2.7143633365631104,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1809697151184082,
+      "loss/reg": 164.2172393798828,
+      "step": 380
+    },
+    {
+      "epoch": 0.009525,
+      "grad_norm": 2.7371509075164795,
+      "grad_norm_var": 0.027450997371564274,
+      "learning_rate": 0.0001,
+      "loss": 166.7423,
+      "loss/crossentropy": 2.7732856273651123,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1965845823287964,
+      "loss/reg": 163.7724609375,
+      "step": 381
+    },
+    {
+      "epoch": 0.00955,
+      "grad_norm": 2.873455047607422,
+      "grad_norm_var": 0.026918816484265782,
+      "learning_rate": 0.0001,
+      "loss": 166.429,
+      "loss/crossentropy": 2.9108941555023193,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20937275886535645,
+      "loss/reg": 163.30874633789062,
+      "step": 382
+    },
+    {
+      "epoch": 0.009575,
+      "grad_norm": 2.4213316440582275,
+      "grad_norm_var": 0.02788105642846309,
+      "learning_rate": 0.0001,
+      "loss": 165.6412,
+      "loss/crossentropy": 2.609408140182495,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18198969960212708,
+      "loss/reg": 162.84983825683594,
+      "step": 383
+    },
+    {
+      "epoch": 0.0096,
+      "grad_norm": 2.3272271156311035,
+      "grad_norm_var": 0.03481792449697399,
+      "learning_rate": 0.0001,
+      "loss": 165.1343,
+      "loss/crossentropy": 2.555964469909668,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19197088479995728,
+      "loss/reg": 162.3863983154297,
+      "step": 384
+    },
+    {
+      "epoch": 0.009625,
+      "grad_norm": 2.934056282043457,
+      "grad_norm_var": 0.03775698672306949,
+      "learning_rate": 0.0001,
+      "loss": 165.2302,
+      "loss/crossentropy": 3.0558810234069824,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2526497542858124,
+      "loss/reg": 161.92169189453125,
+      "step": 385
+    },
+    {
+      "epoch": 0.00965,
+      "grad_norm": 2.6826000213623047,
+      "grad_norm_var": 0.03745695106176162,
+      "learning_rate": 0.0001,
+      "loss": 163.9906,
+      "loss/crossentropy": 2.3432304859161377,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18046601116657257,
+      "loss/reg": 161.46690368652344,
+      "step": 386
+    },
+    {
+      "epoch": 0.009675,
+      "grad_norm": 2.6484644412994385,
+      "grad_norm_var": 0.03442670004298994,
+      "learning_rate": 0.0001,
+      "loss": 163.9132,
+      "loss/crossentropy": 2.702923536300659,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2004016935825348,
+      "loss/reg": 161.00990295410156,
+      "step": 387
+    },
+    {
+      "epoch": 0.0097,
+      "grad_norm": 2.6372387409210205,
+      "grad_norm_var": 0.03444542888671454,
+      "learning_rate": 0.0001,
+      "loss": 163.1478,
+      "loss/crossentropy": 2.4251551628112793,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.16877132654190063,
+      "loss/reg": 160.5538787841797,
+      "step": 388
+    },
+    {
+      "epoch": 0.009725,
+      "grad_norm": 2.702305793762207,
+      "grad_norm_var": 0.029857082413892993,
+      "learning_rate": 0.0001,
+      "loss": 163.1074,
+      "loss/crossentropy": 2.8232805728912354,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.185012549161911,
+      "loss/reg": 160.09906005859375,
+      "step": 389
+    },
+    {
+      "epoch": 0.00975,
+      "grad_norm": 2.8600754737854004,
+      "grad_norm_var": 0.031630664651837746,
+      "learning_rate": 0.0001,
+      "loss": 162.6523,
+      "loss/crossentropy": 2.8138694763183594,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19447121024131775,
+      "loss/reg": 159.64395141601562,
+      "step": 390
+    },
+    {
+      "epoch": 0.009775,
+      "grad_norm": 3.1134562492370605,
+      "grad_norm_var": 0.0425983283858803,
+      "learning_rate": 0.0001,
+      "loss": 162.2286,
+      "loss/crossentropy": 2.8298885822296143,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19982922077178955,
+      "loss/reg": 159.1988983154297,
+      "step": 391
+    },
+    {
+      "epoch": 0.0098,
+      "grad_norm": 2.7824184894561768,
+      "grad_norm_var": 0.04081881578391986,
+      "learning_rate": 0.0001,
+      "loss": 161.8162,
+      "loss/crossentropy": 2.8324615955352783,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23998713493347168,
+      "loss/reg": 158.74378967285156,
+      "step": 392
+    },
+    {
+      "epoch": 0.009825,
+      "grad_norm": 2.5551321506500244,
+      "grad_norm_var": 0.039707183410192214,
+      "learning_rate": 0.0001,
+      "loss": 161.2205,
+      "loss/crossentropy": 2.7324817180633545,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20197457075119019,
+      "loss/reg": 158.28604125976562,
+      "step": 393
+    },
+    {
+      "epoch": 0.00985,
+      "grad_norm": 2.9523658752441406,
+      "grad_norm_var": 0.04342908454165884,
+      "learning_rate": 0.0001,
+      "loss": 161.0285,
+      "loss/crossentropy": 2.9625911712646484,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23311267793178558,
+      "loss/reg": 157.83279418945312,
+      "step": 394
+    },
+    {
+      "epoch": 0.009875,
+      "grad_norm": 2.851621627807617,
+      "grad_norm_var": 0.043773443776307396,
+      "learning_rate": 0.0001,
+      "loss": 160.4359,
+      "loss/crossentropy": 2.845726728439331,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20467609167099,
+      "loss/reg": 157.38543701171875,
+      "step": 395
+    },
+    {
+      "epoch": 0.0099,
+      "grad_norm": 2.7198143005371094,
+      "grad_norm_var": 0.03991894337190445,
+      "learning_rate": 0.0001,
+      "loss": 159.8738,
+      "loss/crossentropy": 2.746978998184204,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19763833284378052,
+      "loss/reg": 156.92922973632812,
+      "step": 396
+    },
+    {
+      "epoch": 0.009925,
+      "grad_norm": 3.205587148666382,
+      "grad_norm_var": 0.05361669114409532,
+      "learning_rate": 0.0001,
+      "loss": 159.7989,
+      "loss/crossentropy": 3.081794261932373,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2381635308265686,
+      "loss/reg": 156.47898864746094,
+      "step": 397
+    },
+    {
+      "epoch": 0.00995,
+      "grad_norm": 2.616609573364258,
+      "grad_norm_var": 0.05408374472743371,
+      "learning_rate": 0.0001,
+      "loss": 159.1617,
+      "loss/crossentropy": 2.927096128463745,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21107865869998932,
+      "loss/reg": 156.02352905273438,
+      "step": 398
+    },
+    {
+      "epoch": 0.009975,
+      "grad_norm": 2.7694790363311768,
+      "grad_norm_var": 0.04637258989533374,
+      "learning_rate": 0.0001,
+      "loss": 158.5532,
+      "loss/crossentropy": 2.7599570751190186,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22352541983127594,
+      "loss/reg": 155.56968688964844,
+      "step": 399
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.700110912322998,
+      "grad_norm_var": 0.032929538641367155,
+      "learning_rate": 0.0001,
+      "loss": 157.9857,
+      "loss/crossentropy": 2.6631603240966797,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20680838823318481,
+      "loss/reg": 155.11572265625,
+      "step": 400
+    },
+    {
+      "epoch": 0.010025,
+      "grad_norm": 2.723098039627075,
+      "grad_norm_var": 0.031819586107117694,
+      "learning_rate": 0.0001,
+      "loss": 157.8913,
+      "loss/crossentropy": 3.0035312175750732,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21532365679740906,
+      "loss/reg": 154.67247009277344,
+      "step": 401
+    },
+    {
+      "epoch": 0.01005,
+      "grad_norm": 3.2790751457214355,
+      "grad_norm_var": 0.046109071751097255,
+      "learning_rate": 0.0001,
+      "loss": 157.6402,
+      "loss/crossentropy": 3.1786797046661377,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2247261106967926,
+      "loss/reg": 154.2367706298828,
+      "step": 402
+    },
+    {
+      "epoch": 0.010075,
+      "grad_norm": 3.035478353500366,
+      "grad_norm_var": 0.04662890368927582,
+      "learning_rate": 0.0001,
+      "loss": 156.9016,
+      "loss/crossentropy": 2.859160900115967,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23791690170764923,
+      "loss/reg": 153.80447387695312,
+      "step": 403
+    },
+    {
+      "epoch": 0.0101,
+      "grad_norm": 3.228116750717163,
+      "grad_norm_var": 0.05216118625241369,
+      "learning_rate": 0.0001,
+      "loss": 156.2435,
+      "loss/crossentropy": 2.6365485191345215,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23830467462539673,
+      "loss/reg": 153.36862182617188,
+      "step": 404
+    },
+    {
+      "epoch": 0.010125,
+      "grad_norm": 2.873171091079712,
+      "grad_norm_var": 0.049916639205365716,
+      "learning_rate": 0.0001,
+      "loss": 156.137,
+      "loss/crossentropy": 2.9804184436798096,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2179238498210907,
+      "loss/reg": 152.9386444091797,
+      "step": 405
+    },
+    {
+      "epoch": 0.01015,
+      "grad_norm": 2.98555326461792,
+      "grad_norm_var": 0.05037325371620132,
+      "learning_rate": 0.0001,
+      "loss": 155.7607,
+      "loss/crossentropy": 3.0471410751342773,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20413793623447418,
+      "loss/reg": 152.50946044921875,
+      "step": 406
+    },
+    {
+      "epoch": 0.010175,
+      "grad_norm": 7.3229899406433105,
+      "grad_norm_var": 1.2780035865068755,
+      "learning_rate": 0.0001,
+      "loss": 155.6208,
+      "loss/crossentropy": 3.179004192352295,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.3593263626098633,
+      "loss/reg": 152.0824432373047,
+      "step": 407
+    },
+    {
+      "epoch": 0.0102,
+      "grad_norm": 2.5819976329803467,
+      "grad_norm_var": 1.2906719922528245,
+      "learning_rate": 0.0001,
+      "loss": 154.8975,
+      "loss/crossentropy": 3.030137538909912,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20869307219982147,
+      "loss/reg": 151.65867614746094,
+      "step": 408
+    },
+    {
+      "epoch": 0.010225,
+      "grad_norm": 2.881765604019165,
+      "grad_norm_var": 1.2714323685464743,
+      "learning_rate": 0.0001,
+      "loss": 154.2088,
+      "loss/crossentropy": 2.7575833797454834,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2238922119140625,
+      "loss/reg": 151.2273712158203,
+      "step": 409
+    },
+    {
+      "epoch": 0.01025,
+      "grad_norm": 3.032489061355591,
+      "grad_norm_var": 1.269504032877872,
+      "learning_rate": 0.0001,
+      "loss": 154.2042,
+      "loss/crossentropy": 3.1616742610931396,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24033766984939575,
+      "loss/reg": 150.8022003173828,
+      "step": 410
+    },
+    {
+      "epoch": 0.010275,
+      "grad_norm": 2.8271679878234863,
+      "grad_norm_var": 1.270597194896881,
+      "learning_rate": 0.0001,
+      "loss": 153.2748,
+      "loss/crossentropy": 2.699521780014038,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20507997274398804,
+      "loss/reg": 150.3701934814453,
+      "step": 411
+    },
+    {
+      "epoch": 0.0103,
+      "grad_norm": 2.8864877223968506,
+      "grad_norm_var": 1.2622421001984312,
+      "learning_rate": 0.0001,
+      "loss": 152.7086,
+      "loss/crossentropy": 2.557340383529663,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20991156995296478,
+      "loss/reg": 149.94137573242188,
+      "step": 412
+    },
+    {
+      "epoch": 0.010325,
+      "grad_norm": 2.665037155151367,
+      "grad_norm_var": 1.278971707345721,
+      "learning_rate": 0.0001,
+      "loss": 152.6803,
+      "loss/crossentropy": 2.9646856784820557,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19997093081474304,
+      "loss/reg": 149.51559448242188,
+      "step": 413
+    },
+    {
+      "epoch": 0.01035,
+      "grad_norm": 2.9376747608184814,
+      "grad_norm_var": 1.2625575568111116,
+      "learning_rate": 0.0001,
+      "loss": 151.8695,
+      "loss/crossentropy": 2.5759167671203613,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2004423588514328,
+      "loss/reg": 149.09315490722656,
+      "step": 414
+    },
+    {
+      "epoch": 0.010375,
+      "grad_norm": 2.8804266452789307,
+      "grad_norm_var": 1.2573930188301232,
+      "learning_rate": 0.0001,
+      "loss": 151.6834,
+      "loss/crossentropy": 2.805604934692383,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19938749074935913,
+      "loss/reg": 148.67840576171875,
+      "step": 415
+    },
+    {
+      "epoch": 0.0104,
+      "grad_norm": 2.844179153442383,
+      "grad_norm_var": 1.2495192648568405,
+      "learning_rate": 0.0001,
+      "loss": 151.2533,
+      "loss/crossentropy": 2.804511547088623,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19970285892486572,
+      "loss/reg": 148.2490692138672,
+      "step": 416
+    },
+    {
+      "epoch": 0.010425,
+      "grad_norm": 2.5592336654663086,
+      "grad_norm_var": 1.2613231291949638,
+      "learning_rate": 0.0001,
+      "loss": 150.7822,
+      "loss/crossentropy": 2.7462940216064453,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20548762381076813,
+      "loss/reg": 147.8304443359375,
+      "step": 417
+    },
+    {
+      "epoch": 0.01045,
+      "grad_norm": 2.6821486949920654,
+      "grad_norm_var": 1.2754135282406718,
+      "learning_rate": 0.0001,
+      "loss": 150.2964,
+      "loss/crossentropy": 2.685582399368286,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20678475499153137,
+      "loss/reg": 147.40402221679688,
+      "step": 418
+    },
+    {
+      "epoch": 0.010475,
+      "grad_norm": 2.9765594005584717,
+      "grad_norm_var": 1.2764437045171377,
+      "learning_rate": 0.0001,
+      "loss": 150.4573,
+      "loss/crossentropy": 3.2364401817321777,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2370205670595169,
+      "loss/reg": 146.98388671875,
+      "step": 419
+    },
+    {
+      "epoch": 0.0105,
+      "grad_norm": 2.849670886993408,
+      "grad_norm_var": 1.2807121780616888,
+      "learning_rate": 0.0001,
+      "loss": 149.4418,
+      "loss/crossentropy": 2.6446595191955566,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23380397260189056,
+      "loss/reg": 146.56333923339844,
+      "step": 420
+    },
+    {
+      "epoch": 0.010525,
+      "grad_norm": 2.668278694152832,
+      "grad_norm_var": 1.289851246662234,
+      "learning_rate": 0.0001,
+      "loss": 149.259,
+      "loss/crossentropy": 2.9065048694610596,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21474584937095642,
+      "loss/reg": 146.1377716064453,
+      "step": 421
+    },
+    {
+      "epoch": 0.01055,
+      "grad_norm": 2.594972848892212,
+      "grad_norm_var": 1.3052862072893956,
+      "learning_rate": 0.0001,
+      "loss": 148.7854,
+      "loss/crossentropy": 2.8732926845550537,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19787928462028503,
+      "loss/reg": 145.71426391601562,
+      "step": 422
+    },
+    {
+      "epoch": 0.010575,
+      "grad_norm": 3.493597984313965,
+      "grad_norm_var": 0.05255425549001392,
+      "learning_rate": 0.0001,
+      "loss": 148.5448,
+      "loss/crossentropy": 2.979468584060669,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2681131958961487,
+      "loss/reg": 145.29721069335938,
+      "step": 423
+    },
+    {
+      "epoch": 0.0106,
+      "grad_norm": 2.545316219329834,
+      "grad_norm_var": 0.05387626475652117,
+      "learning_rate": 0.0001,
+      "loss": 147.7063,
+      "loss/crossentropy": 2.627589225769043,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1998068243265152,
+      "loss/reg": 144.87893676757812,
+      "step": 424
+    },
+    {
+      "epoch": 0.010625,
+      "grad_norm": 2.801800489425659,
+      "grad_norm_var": 0.05375398155694313,
+      "learning_rate": 0.0001,
+      "loss": 147.7385,
+      "loss/crossentropy": 3.051680088043213,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23187664151191711,
+      "loss/reg": 144.4549560546875,
+      "step": 425
+    },
+    {
+      "epoch": 0.01065,
+      "grad_norm": 2.5229389667510986,
+      "grad_norm_var": 0.056076010247692425,
+      "learning_rate": 0.0001,
+      "loss": 146.9956,
+      "loss/crossentropy": 2.7526652812957764,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20328941941261292,
+      "loss/reg": 144.0396270751953,
+      "step": 426
+    },
+    {
+      "epoch": 0.010675,
+      "grad_norm": 2.885693311691284,
+      "grad_norm_var": 0.05653354974819133,
+      "learning_rate": 0.0001,
+      "loss": 146.6748,
+      "loss/crossentropy": 2.816437244415283,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22479213774204254,
+      "loss/reg": 143.633544921875,
+      "step": 427
+    },
+    {
+      "epoch": 0.0107,
+      "grad_norm": 2.888580083847046,
+      "grad_norm_var": 0.05655805617718836,
+      "learning_rate": 0.0001,
+      "loss": 146.5772,
+      "loss/crossentropy": 3.1261234283447266,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2150314301252365,
+      "loss/reg": 143.23602294921875,
+      "step": 428
+    },
+    {
+      "epoch": 0.010725,
+      "grad_norm": 2.679644823074341,
+      "grad_norm_var": 0.05630900067725785,
+      "learning_rate": 0.0001,
+      "loss": 145.6467,
+      "loss/crossentropy": 2.6079299449920654,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21127401292324066,
+      "loss/reg": 142.82745361328125,
+      "step": 429
+    },
+    {
+      "epoch": 0.01075,
+      "grad_norm": 2.5422425270080566,
+      "grad_norm_var": 0.05885842547493757,
+      "learning_rate": 0.0001,
+      "loss": 145.2591,
+      "loss/crossentropy": 2.6437766551971436,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19190073013305664,
+      "loss/reg": 142.42344665527344,
+      "step": 430
+    },
+    {
+      "epoch": 0.010775,
+      "grad_norm": 2.6776621341705322,
+      "grad_norm_var": 0.058603604392913886,
+      "learning_rate": 0.0001,
+      "loss": 145.2115,
+      "loss/crossentropy": 2.9833149909973145,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2141813039779663,
+      "loss/reg": 142.0140380859375,
+      "step": 431
+    },
+    {
+      "epoch": 0.0108,
+      "grad_norm": 2.8127005100250244,
+      "grad_norm_var": 0.058326000336826195,
+      "learning_rate": 0.0001,
+      "loss": 145.2914,
+      "loss/crossentropy": 3.4850382804870605,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1998693197965622,
+      "loss/reg": 141.60650634765625,
+      "step": 432
+    },
+    {
+      "epoch": 0.010825,
+      "grad_norm": 2.7787222862243652,
+      "grad_norm_var": 0.05542301102962029,
+      "learning_rate": 0.0001,
+      "loss": 144.1324,
+      "loss/crossentropy": 2.7224390506744385,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21212972700595856,
+      "loss/reg": 141.1977996826172,
+      "step": 433
+    },
+    {
+      "epoch": 0.01085,
+      "grad_norm": 2.582594156265259,
+      "grad_norm_var": 0.057275397262241276,
+      "learning_rate": 0.0001,
+      "loss": 143.8046,
+      "loss/crossentropy": 2.824321985244751,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19551266729831696,
+      "loss/reg": 140.7847900390625,
+      "step": 434
+    },
+    {
+      "epoch": 0.010875,
+      "grad_norm": 2.746096134185791,
+      "grad_norm_var": 0.05421119495302354,
+      "learning_rate": 0.0001,
+      "loss": 143.3764,
+      "loss/crossentropy": 2.780687093734741,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21173939108848572,
+      "loss/reg": 140.38397216796875,
+      "step": 435
+    },
+    {
+      "epoch": 0.0109,
+      "grad_norm": 3.0453367233276367,
+      "grad_norm_var": 0.05908933324654957,
+      "learning_rate": 0.0001,
+      "loss": 143.2668,
+      "loss/crossentropy": 3.0895285606384277,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2058122158050537,
+      "loss/reg": 139.97145080566406,
+      "step": 436
+    },
+    {
+      "epoch": 0.010925,
+      "grad_norm": 2.8102662563323975,
+      "grad_norm_var": 0.058487291701452664,
+      "learning_rate": 0.0001,
+      "loss": 142.5939,
+      "loss/crossentropy": 2.8098881244659424,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21639777719974518,
+      "loss/reg": 139.56761169433594,
+      "step": 437
+    },
+    {
+      "epoch": 0.01095,
+      "grad_norm": 2.786220073699951,
+      "grad_norm_var": 0.05616962127030523,
+      "learning_rate": 0.0001,
+      "loss": 142.2782,
+      "loss/crossentropy": 2.889941453933716,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22128039598464966,
+      "loss/reg": 139.16693115234375,
+      "step": 438
+    },
+    {
+      "epoch": 0.010975,
+      "grad_norm": 2.496029853820801,
+      "grad_norm_var": 0.024443678618327407,
+      "learning_rate": 0.0001,
+      "loss": 141.7307,
+      "loss/crossentropy": 2.751573085784912,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19620397686958313,
+      "loss/reg": 138.782958984375,
+      "step": 439
+    },
+    {
+      "epoch": 0.011,
+      "grad_norm": 4.073153495788574,
+      "grad_norm_var": 0.13370943824087528,
+      "learning_rate": 0.0001,
+      "loss": 141.1084,
+      "loss/crossentropy": 2.547888994216919,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1862032264471054,
+      "loss/reg": 138.37429809570312,
+      "step": 440
+    },
+    {
+      "epoch": 0.011025,
+      "grad_norm": 2.8439884185791016,
+      "grad_norm_var": 0.13371489998645056,
+      "learning_rate": 0.0001,
+      "loss": 141.1169,
+      "loss/crossentropy": 2.9181034564971924,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22650295495986938,
+      "loss/reg": 137.97227478027344,
+      "step": 441
+    },
+    {
+      "epoch": 0.01105,
+      "grad_norm": 3.0714433193206787,
+      "grad_norm_var": 0.13055613818799353,
+      "learning_rate": 0.0001,
+      "loss": 140.6507,
+      "loss/crossentropy": 2.857541084289551,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21894115209579468,
+      "loss/reg": 137.57418823242188,
+      "step": 442
+    },
+    {
+      "epoch": 0.011075,
+      "grad_norm": 2.4183578491210938,
+      "grad_norm_var": 0.14245098271452789,
+      "learning_rate": 0.0001,
+      "loss": 140.0377,
+      "loss/crossentropy": 2.6772847175598145,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18452061712741852,
+      "loss/reg": 137.17588806152344,
+      "step": 443
+    },
+    {
+      "epoch": 0.0111,
+      "grad_norm": 3.0140979290008545,
+      "grad_norm_var": 0.14444423385113864,
+      "learning_rate": 0.0001,
+      "loss": 139.7851,
+      "loss/crossentropy": 2.8040971755981445,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19091713428497314,
+      "loss/reg": 136.79010009765625,
+      "step": 444
+    },
+    {
+      "epoch": 0.011125,
+      "grad_norm": 2.9585142135620117,
+      "grad_norm_var": 0.14348511163545558,
+      "learning_rate": 0.0001,
+      "loss": 139.3333,
+      "loss/crossentropy": 2.7211837768554688,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2089160978794098,
+      "loss/reg": 136.40313720703125,
+      "step": 445
+    },
+    {
+      "epoch": 0.01115,
+      "grad_norm": 2.8655238151550293,
+      "grad_norm_var": 0.1365966991771605,
+      "learning_rate": 0.0001,
+      "loss": 139.1563,
+      "loss/crossentropy": 2.9270730018615723,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21874740719795227,
+      "loss/reg": 136.01043701171875,
+      "step": 446
+    },
+    {
+      "epoch": 0.011175,
+      "grad_norm": 3.5568222999572754,
+      "grad_norm_var": 0.16191349512506553,
+      "learning_rate": 0.0001,
+      "loss": 138.9605,
+      "loss/crossentropy": 3.0990545749664307,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2374783754348755,
+      "loss/reg": 135.62399291992188,
+      "step": 447
+    },
+    {
+      "epoch": 0.0112,
+      "grad_norm": 3.069288730621338,
+      "grad_norm_var": 0.16205836586185096,
+      "learning_rate": 0.0001,
+      "loss": 138.2294,
+      "loss/crossentropy": 2.8115599155426025,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19097256660461426,
+      "loss/reg": 135.22686767578125,
+      "step": 448
+    },
+    {
+      "epoch": 0.011225,
+      "grad_norm": 2.829019784927368,
+      "grad_norm_var": 0.16110285265393878,
+      "learning_rate": 0.0001,
+      "loss": 137.3381,
+      "loss/crossentropy": 2.2983205318450928,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20224307477474213,
+      "loss/reg": 134.8375701904297,
+      "step": 449
+    },
+    {
+      "epoch": 0.01125,
+      "grad_norm": 2.959508180618286,
+      "grad_norm_var": 0.1516222356653503,
+      "learning_rate": 0.0001,
+      "loss": 137.696,
+      "loss/crossentropy": 3.0192501544952393,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22333860397338867,
+      "loss/reg": 134.4534149169922,
+      "step": 450
+    },
+    {
+      "epoch": 0.011275,
+      "grad_norm": 2.742422342300415,
+      "grad_norm_var": 0.15173348060539713,
+      "learning_rate": 0.0001,
+      "loss": 137.0164,
+      "loss/crossentropy": 2.738880157470703,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2160961925983429,
+      "loss/reg": 134.0614776611328,
+      "step": 451
+    },
+    {
+      "epoch": 0.0113,
+      "grad_norm": 2.8377208709716797,
+      "grad_norm_var": 0.1523766132789395,
+      "learning_rate": 0.0001,
+      "loss": 136.7287,
+      "loss/crossentropy": 2.8547892570495605,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21014469861984253,
+      "loss/reg": 133.6637420654297,
+      "step": 452
+    },
+    {
+      "epoch": 0.011325,
+      "grad_norm": 2.859851121902466,
+      "grad_norm_var": 0.15155175629658876,
+      "learning_rate": 0.0001,
+      "loss": 136.1562,
+      "loss/crossentropy": 2.66414213180542,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22573256492614746,
+      "loss/reg": 133.26629638671875,
+      "step": 453
+    },
+    {
+      "epoch": 0.01135,
+      "grad_norm": 2.6915457248687744,
+      "grad_norm_var": 0.15432295238523253,
+      "learning_rate": 0.0001,
+      "loss": 135.9197,
+      "loss/crossentropy": 2.8435401916503906,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19813670217990875,
+      "loss/reg": 132.87806701660156,
+      "step": 454
+    },
+    {
+      "epoch": 0.011375,
+      "grad_norm": 2.7335429191589355,
+      "grad_norm_var": 0.14329945186020698,
+      "learning_rate": 0.0001,
+      "loss": 135.273,
+      "loss/crossentropy": 2.5880157947540283,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19232623279094696,
+      "loss/reg": 132.49264526367188,
+      "step": 455
+    },
+    {
+      "epoch": 0.0114,
+      "grad_norm": 2.695430040359497,
+      "grad_norm_var": 0.059341799627403206,
+      "learning_rate": 0.0001,
+      "loss": 135.0541,
+      "loss/crossentropy": 2.746819257736206,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20626962184906006,
+      "loss/reg": 132.10101318359375,
+      "step": 456
+    },
+    {
+      "epoch": 0.011425,
+      "grad_norm": 2.8293468952178955,
+      "grad_norm_var": 0.05943368425061877,
+      "learning_rate": 0.0001,
+      "loss": 134.5696,
+      "loss/crossentropy": 2.6393778324127197,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21043707430362701,
+      "loss/reg": 131.71983337402344,
+      "step": 457
+    },
+    {
+      "epoch": 0.01145,
+      "grad_norm": 2.8794851303100586,
+      "grad_norm_var": 0.05692067856874985,
+      "learning_rate": 0.0001,
+      "loss": 134.1734,
+      "loss/crossentropy": 2.650709390640259,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19179099798202515,
+      "loss/reg": 131.3309326171875,
+      "step": 458
+    },
+    {
+      "epoch": 0.011475,
+      "grad_norm": 2.8426058292388916,
+      "grad_norm_var": 0.042549658611572026,
+      "learning_rate": 0.0001,
+      "loss": 133.6375,
+      "loss/crossentropy": 2.513119697570801,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17655596137046814,
+      "loss/reg": 130.9477996826172,
+      "step": 459
+    },
+    {
+      "epoch": 0.0115,
+      "grad_norm": 3.155630588531494,
+      "grad_norm_var": 0.045996375141740486,
+      "learning_rate": 0.0001,
+      "loss": 133.6152,
+      "loss/crossentropy": 2.833134651184082,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21260111033916473,
+      "loss/reg": 130.5695037841797,
+      "step": 460
+    },
+    {
+      "epoch": 0.011525,
+      "grad_norm": 2.494323492050171,
+      "grad_norm_var": 0.05625290696184659,
+      "learning_rate": 0.0001,
+      "loss": 133.0751,
+      "loss/crossentropy": 2.6999471187591553,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18340814113616943,
+      "loss/reg": 130.19174194335938,
+      "step": 461
+    },
+    {
+      "epoch": 0.01155,
+      "grad_norm": 2.6319777965545654,
+      "grad_norm_var": 0.06003884724692152,
+      "learning_rate": 0.0001,
+      "loss": 132.8793,
+      "loss/crossentropy": 2.860808849334717,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20245122909545898,
+      "loss/reg": 129.8160400390625,
+      "step": 462
+    },
+    {
+      "epoch": 0.011575,
+      "grad_norm": 2.514202117919922,
+      "grad_norm_var": 0.03153201551004561,
+      "learning_rate": 0.0001,
+      "loss": 132.3611,
+      "loss/crossentropy": 2.730762481689453,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18745794892311096,
+      "loss/reg": 129.44293212890625,
+      "step": 463
+    },
+    {
+      "epoch": 0.0116,
+      "grad_norm": 2.5370423793792725,
+      "grad_norm_var": 0.029975769359208235,
+      "learning_rate": 0.0001,
+      "loss": 131.8961,
+      "loss/crossentropy": 2.640779972076416,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18950456380844116,
+      "loss/reg": 129.0658416748047,
+      "step": 464
+    },
+    {
+      "epoch": 0.011625,
+      "grad_norm": 4.233380317687988,
+      "grad_norm_var": 0.16530188527350068,
+      "learning_rate": 0.0001,
+      "loss": 131.9097,
+      "loss/crossentropy": 2.9330646991729736,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2857138216495514,
+      "loss/reg": 128.69088745117188,
+      "step": 465
+    },
+    {
+      "epoch": 0.01165,
+      "grad_norm": 2.779022216796875,
+      "grad_norm_var": 0.16475971985575838,
+      "learning_rate": 0.0001,
+      "loss": 131.55,
+      "loss/crossentropy": 3.005692481994629,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22528377175331116,
+      "loss/reg": 128.3190460205078,
+      "step": 466
+    },
+    {
+      "epoch": 0.011675,
+      "grad_norm": 2.610145092010498,
+      "grad_norm_var": 0.16759359645252517,
+      "learning_rate": 0.0001,
+      "loss": 130.7001,
+      "loss/crossentropy": 2.568135976791382,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17949801683425903,
+      "loss/reg": 127.9524917602539,
+      "step": 467
+    },
+    {
+      "epoch": 0.0117,
+      "grad_norm": 2.7065391540527344,
+      "grad_norm_var": 0.1685835608909353,
+      "learning_rate": 0.0001,
+      "loss": 130.8896,
+      "loss/crossentropy": 3.0850789546966553,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21300601959228516,
+      "loss/reg": 127.591552734375,
+      "step": 468
+    },
+    {
+      "epoch": 0.011725,
+      "grad_norm": 2.577502965927124,
+      "grad_norm_var": 0.17224012017982868,
+      "learning_rate": 0.0001,
+      "loss": 130.1112,
+      "loss/crossentropy": 2.673410177230835,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21455711126327515,
+      "loss/reg": 127.2232666015625,
+      "step": 469
+    },
+    {
+      "epoch": 0.01175,
+      "grad_norm": 2.620901346206665,
+      "grad_norm_var": 0.17363936391112228,
+      "learning_rate": 0.0001,
+      "loss": 130.2284,
+      "loss/crossentropy": 3.125420093536377,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23448872566223145,
+      "loss/reg": 126.86849975585938,
+      "step": 470
+    },
+    {
+      "epoch": 0.011775,
+      "grad_norm": 2.6767570972442627,
+      "grad_norm_var": 0.1743635181200638,
+      "learning_rate": 0.0001,
+      "loss": 129.5788,
+      "loss/crossentropy": 2.8606624603271484,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.214532732963562,
+      "loss/reg": 126.50355529785156,
+      "step": 471
+    },
+    {
+      "epoch": 0.0118,
+      "grad_norm": 2.6148667335510254,
+      "grad_norm_var": 0.17588189249079184,
+      "learning_rate": 0.0001,
+      "loss": 129.3386,
+      "loss/crossentropy": 2.9896390438079834,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20493671298027039,
+      "loss/reg": 126.14399719238281,
+      "step": 472
+    },
+    {
+      "epoch": 0.011825,
+      "grad_norm": 2.431541681289673,
+      "grad_norm_var": 0.1838967324892555,
+      "learning_rate": 0.0001,
+      "loss": 128.7304,
+      "loss/crossentropy": 2.748108386993408,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20910008251667023,
+      "loss/reg": 125.7732162475586,
+      "step": 473
+    },
+    {
+      "epoch": 0.01185,
+      "grad_norm": 2.770341157913208,
+      "grad_norm_var": 0.18303516965960565,
+      "learning_rate": 0.0001,
+      "loss": 128.8167,
+      "loss/crossentropy": 3.200181245803833,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20696839690208435,
+      "loss/reg": 125.40956115722656,
+      "step": 474
+    },
+    {
+      "epoch": 0.011875,
+      "grad_norm": 2.664780378341675,
+      "grad_norm_var": 0.1831074521196359,
+      "learning_rate": 0.0001,
+      "loss": 128.242,
+      "loss/crossentropy": 2.969964027404785,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22718313336372375,
+      "loss/reg": 125.04483032226562,
+      "step": 475
+    },
+    {
+      "epoch": 0.0119,
+      "grad_norm": 2.510708808898926,
+      "grad_norm_var": 0.17432457651303418,
+      "learning_rate": 0.0001,
+      "loss": 127.4987,
+      "loss/crossentropy": 2.6007182598114014,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21277430653572083,
+      "loss/reg": 124.68523406982422,
+      "step": 476
+    },
+    {
+      "epoch": 0.011925,
+      "grad_norm": 2.9702818393707275,
+      "grad_norm_var": 0.17474036873515666,
+      "learning_rate": 0.0001,
+      "loss": 127.41,
+      "loss/crossentropy": 2.86385178565979,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2170620560646057,
+      "loss/reg": 124.32904052734375,
+      "step": 477
+    },
+    {
+      "epoch": 0.01195,
+      "grad_norm": 2.6219518184661865,
+      "grad_norm_var": 0.17489188976909203,
+      "learning_rate": 0.0001,
+      "loss": 126.9219,
+      "loss/crossentropy": 2.7511582374572754,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2034904807806015,
+      "loss/reg": 123.96720886230469,
+      "step": 478
+    },
+    {
+      "epoch": 0.011975,
+      "grad_norm": 2.599919080734253,
+      "grad_norm_var": 0.17277049923038704,
+      "learning_rate": 0.0001,
+      "loss": 126.5413,
+      "loss/crossentropy": 2.7556185722351074,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1808776557445526,
+      "loss/reg": 123.60482025146484,
+      "step": 479
+    },
+    {
+      "epoch": 0.012,
+      "grad_norm": 2.5657715797424316,
+      "grad_norm_var": 0.17202413016778048,
+      "learning_rate": 0.0001,
+      "loss": 126.311,
+      "loss/crossentropy": 2.8689675331115723,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20293821394443512,
+      "loss/reg": 123.23907470703125,
+      "step": 480
+    },
+    {
+      "epoch": 0.012025,
+      "grad_norm": 3.2952754497528076,
+      "grad_norm_var": 0.04112811192426212,
+      "learning_rate": 0.0001,
+      "loss": 125.9246,
+      "loss/crossentropy": 2.818702220916748,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2267230898141861,
+      "loss/reg": 122.87913513183594,
+      "step": 481
+    },
+    {
+      "epoch": 0.01205,
+      "grad_norm": 2.615551233291626,
+      "grad_norm_var": 0.040825667865743515,
+      "learning_rate": 0.0001,
+      "loss": 125.7597,
+      "loss/crossentropy": 3.017942428588867,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21883933246135712,
+      "loss/reg": 122.52295684814453,
+      "step": 482
+    },
+    {
+      "epoch": 0.012075,
+      "grad_norm": 3.4193038940429688,
+      "grad_norm_var": 0.07439346615546445,
+      "learning_rate": 0.0001,
+      "loss": 125.1558,
+      "loss/crossentropy": 2.7724950313568115,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21510450541973114,
+      "loss/reg": 122.16816711425781,
+      "step": 483
+    },
+    {
+      "epoch": 0.0121,
+      "grad_norm": 2.505560874938965,
+      "grad_norm_var": 0.07751650924940326,
+      "learning_rate": 0.0001,
+      "loss": 124.927,
+      "loss/crossentropy": 2.9188194274902344,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19483289122581482,
+      "loss/reg": 121.81333923339844,
+      "step": 484
+    },
+    {
+      "epoch": 0.012125,
+      "grad_norm": 2.7361960411071777,
+      "grad_norm_var": 0.07615337485008998,
+      "learning_rate": 0.0001,
+      "loss": 124.5378,
+      "loss/crossentropy": 2.8456363677978516,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22739849984645844,
+      "loss/reg": 121.46476745605469,
+      "step": 485
+    },
+    {
+      "epoch": 0.01215,
+      "grad_norm": 2.5417439937591553,
+      "grad_norm_var": 0.07765668354199894,
+      "learning_rate": 0.0001,
+      "loss": 124.0698,
+      "loss/crossentropy": 2.7495946884155273,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20217089354991913,
+      "loss/reg": 121.11808013916016,
+      "step": 486
+    },
+    {
+      "epoch": 0.012175,
+      "grad_norm": 2.67401385307312,
+      "grad_norm_var": 0.07767344047614415,
+      "learning_rate": 0.0001,
+      "loss": 123.7152,
+      "loss/crossentropy": 2.7279324531555176,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22579890489578247,
+      "loss/reg": 120.76143646240234,
+      "step": 487
+    },
+    {
+      "epoch": 0.0122,
+      "grad_norm": 2.488034248352051,
+      "grad_norm_var": 0.08047557627683753,
+      "learning_rate": 0.0001,
+      "loss": 123.451,
+      "loss/crossentropy": 2.8445310592651367,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19405901432037354,
+      "loss/reg": 120.41239929199219,
+      "step": 488
+    },
+    {
+      "epoch": 0.012225,
+      "grad_norm": 2.5098259449005127,
+      "grad_norm_var": 0.07791882719567696,
+      "learning_rate": 0.0001,
+      "loss": 123.1306,
+      "loss/crossentropy": 2.8643126487731934,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19319729506969452,
+      "loss/reg": 120.0730972290039,
+      "step": 489
+    },
+    {
+      "epoch": 0.01225,
+      "grad_norm": 2.768183469772339,
+      "grad_norm_var": 0.07790408271164324,
+      "learning_rate": 0.0001,
+      "loss": 123.1131,
+      "loss/crossentropy": 3.1591904163360596,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23094454407691956,
+      "loss/reg": 119.72293853759766,
+      "step": 490
+    },
+    {
+      "epoch": 0.012275,
+      "grad_norm": 2.5662379264831543,
+      "grad_norm_var": 0.0792095113967413,
+      "learning_rate": 0.0001,
+      "loss": 122.4756,
+      "loss/crossentropy": 2.9084203243255615,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19523289799690247,
+      "loss/reg": 119.37194061279297,
+      "step": 491
+    },
+    {
+      "epoch": 0.0123,
+      "grad_norm": 2.7122464179992676,
+      "grad_norm_var": 0.07634484398728673,
+      "learning_rate": 0.0001,
+      "loss": 122.0971,
+      "loss/crossentropy": 2.8523452281951904,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22050230205059052,
+      "loss/reg": 119.02421569824219,
+      "step": 492
+    },
+    {
+      "epoch": 0.012325,
+      "grad_norm": 2.6066133975982666,
+      "grad_norm_var": 0.0726872533289639,
+      "learning_rate": 0.0001,
+      "loss": 121.826,
+      "loss/crossentropy": 2.943951368331909,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2006131410598755,
+      "loss/reg": 118.68144989013672,
+      "step": 493
+    },
+    {
+      "epoch": 0.01235,
+      "grad_norm": 2.752058506011963,
+      "grad_norm_var": 0.07236263717393664,
+      "learning_rate": 0.0001,
+      "loss": 121.3462,
+      "loss/crossentropy": 2.8118808269500732,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19854308664798737,
+      "loss/reg": 118.33576965332031,
+      "step": 494
+    },
+    {
+      "epoch": 0.012375,
+      "grad_norm": 2.920879364013672,
+      "grad_norm_var": 0.07409949102501111,
+      "learning_rate": 0.0001,
+      "loss": 120.8552,
+      "loss/crossentropy": 2.6617374420166016,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19966919720172882,
+      "loss/reg": 117.99381256103516,
+      "step": 495
+    },
+    {
+      "epoch": 0.0124,
+      "grad_norm": 2.7351551055908203,
+      "grad_norm_var": 0.07218718704733244,
+      "learning_rate": 0.0001,
+      "loss": 121.014,
+      "loss/crossentropy": 3.1586992740631104,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20681920647621155,
+      "loss/reg": 117.6484603881836,
+      "step": 496
+    },
+    {
+      "epoch": 0.012425,
+      "grad_norm": 2.35911226272583,
+      "grad_norm_var": 0.05770549480844901,
+      "learning_rate": 0.0001,
+      "loss": 120.1416,
+      "loss/crossentropy": 2.6408603191375732,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19257068634033203,
+      "loss/reg": 117.30819702148438,
+      "step": 497
+    },
+    {
+      "epoch": 0.01245,
+      "grad_norm": 2.738931655883789,
+      "grad_norm_var": 0.057565104717087694,
+      "learning_rate": 0.0001,
+      "loss": 119.9577,
+      "loss/crossentropy": 2.8233439922332764,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.17061106860637665,
+      "loss/reg": 116.96371459960938,
+      "step": 498
+    },
+    {
+      "epoch": 0.012475,
+      "grad_norm": 3.1137208938598633,
+      "grad_norm_var": 0.03367133349540812,
+      "learning_rate": 0.0001,
+      "loss": 119.923,
+      "loss/crossentropy": 3.0651209354400635,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23511168360710144,
+      "loss/reg": 116.62273406982422,
+      "step": 499
+    },
+    {
+      "epoch": 0.0125,
+      "grad_norm": 2.931697130203247,
+      "grad_norm_var": 0.03564747630760602,
+      "learning_rate": 0.0001,
+      "loss": 119.9291,
+      "loss/crossentropy": 3.406992197036743,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23726436495780945,
+      "loss/reg": 116.28487396240234,
+      "step": 500
+    },
+    {
+      "epoch": 0.012525,
+      "grad_norm": 2.700242280960083,
+      "grad_norm_var": 0.035541163062665034,
+      "learning_rate": 0.0001,
+      "loss": 119.0308,
+      "loss/crossentropy": 2.8376407623291016,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24376104772090912,
+      "loss/reg": 115.94940948486328,
+      "step": 501
+    },
+    {
+      "epoch": 0.01255,
+      "grad_norm": 2.5305213928222656,
+      "grad_norm_var": 0.03577823695906375,
+      "learning_rate": 0.0001,
+      "loss": 118.5299,
+      "loss/crossentropy": 2.7097437381744385,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21053358912467957,
+      "loss/reg": 115.60960388183594,
+      "step": 502
+    },
+    {
+      "epoch": 0.012575,
+      "grad_norm": 2.849966526031494,
+      "grad_norm_var": 0.03723922016397386,
+      "learning_rate": 0.0001,
+      "loss": 118.188,
+      "loss/crossentropy": 2.718074083328247,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20146432518959045,
+      "loss/reg": 115.26842498779297,
+      "step": 503
+    },
+    {
+      "epoch": 0.0126,
+      "grad_norm": 6.95852518081665,
+      "grad_norm_var": 1.1568663516812001,
+      "learning_rate": 0.0001,
+      "loss": 118.1133,
+      "loss/crossentropy": 2.935373306274414,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23765507340431213,
+      "loss/reg": 114.94031524658203,
+      "step": 504
+    },
+    {
+      "epoch": 0.012625,
+      "grad_norm": 2.5411369800567627,
+      "grad_norm_var": 1.1549454537059127,
+      "learning_rate": 0.0001,
+      "loss": 117.2537,
+      "loss/crossentropy": 2.4439382553100586,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19584742188453674,
+      "loss/reg": 114.61389923095703,
+      "step": 505
+    },
+    {
+      "epoch": 0.01265,
+      "grad_norm": 2.5562095642089844,
+      "grad_norm_var": 1.1639262533207444,
+      "learning_rate": 0.0001,
+      "loss": 117.1436,
+      "loss/crossentropy": 2.656878709793091,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20227234065532684,
+      "loss/reg": 114.28450012207031,
+      "step": 506
+    },
+    {
+      "epoch": 0.012675,
+      "grad_norm": 2.452930212020874,
+      "grad_norm_var": 1.1708788671982782,
+      "learning_rate": 0.0001,
+      "loss": 116.7552,
+      "loss/crossentropy": 2.6194908618927,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18559934198856354,
+      "loss/reg": 113.95010375976562,
+      "step": 507
+    },
+    {
+      "epoch": 0.0127,
+      "grad_norm": 2.600403308868408,
+      "grad_norm_var": 1.1754484294589225,
+      "learning_rate": 0.0001,
+      "loss": 116.878,
+      "loss/crossentropy": 3.047492504119873,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21343189477920532,
+      "loss/reg": 113.61707305908203,
+      "step": 508
+    },
+    {
+      "epoch": 0.012725,
+      "grad_norm": 2.8134710788726807,
+      "grad_norm_var": 1.1683965532079448,
+      "learning_rate": 0.0001,
+      "loss": 116.1436,
+      "loss/crossentropy": 2.6284308433532715,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22599661350250244,
+      "loss/reg": 113.28915405273438,
+      "step": 509
+    },
+    {
+      "epoch": 0.01275,
+      "grad_norm": 3.0693750381469727,
+      "grad_norm_var": 1.1653763573131297,
+      "learning_rate": 0.0001,
+      "loss": 116.0373,
+      "loss/crossentropy": 2.8555266857147217,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21752798557281494,
+      "loss/reg": 112.96427154541016,
+      "step": 510
+    },
+    {
+      "epoch": 0.012775,
+      "grad_norm": 3.6787264347076416,
+      "grad_norm_var": 1.1940838877816609,
+      "learning_rate": 0.0001,
+      "loss": 115.4654,
+      "loss/crossentropy": 2.5687825679779053,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2617979049682617,
+      "loss/reg": 112.63484191894531,
+      "step": 511
+    },
+    {
+      "epoch": 0.0128,
+      "grad_norm": 2.7873494625091553,
+      "grad_norm_var": 1.192136957506784,
+      "learning_rate": 0.0001,
+      "loss": 115.1653,
+      "loss/crossentropy": 2.6631510257720947,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.191485196352005,
+      "loss/reg": 112.31067657470703,
+      "step": 512
+    },
+    {
+      "epoch": 0.012825,
+      "grad_norm": 2.6256515979766846,
+      "grad_norm_var": 1.1722853783887095,
+      "learning_rate": 0.0001,
+      "loss": 114.9084,
+      "loss/crossentropy": 2.7325098514556885,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19808584451675415,
+      "loss/reg": 111.97781372070312,
+      "step": 513
+    },
+    {
+      "epoch": 0.01285,
+      "grad_norm": 2.967942237854004,
+      "grad_norm_var": 1.1657807662503947,
+      "learning_rate": 0.0001,
+      "loss": 114.5064,
+      "loss/crossentropy": 2.6529791355133057,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20329859852790833,
+      "loss/reg": 111.650146484375,
+      "step": 514
+    },
+    {
+      "epoch": 0.012875,
+      "grad_norm": 2.934296131134033,
+      "grad_norm_var": 1.166833422533542,
+      "learning_rate": 0.0001,
+      "loss": 114.5007,
+      "loss/crossentropy": 2.951188087463379,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22600007057189941,
+      "loss/reg": 111.32347869873047,
+      "step": 515
+    },
+    {
+      "epoch": 0.0129,
+      "grad_norm": 2.6087300777435303,
+      "grad_norm_var": 1.178981137512823,
+      "learning_rate": 0.0001,
+      "loss": 114.303,
+      "loss/crossentropy": 3.0892579555511475,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.219014972448349,
+      "loss/reg": 110.99468994140625,
+      "step": 516
+    },
+    {
+      "epoch": 0.012925,
+      "grad_norm": 2.6592049598693848,
+      "grad_norm_var": 1.1809575567663138,
+      "learning_rate": 0.0001,
+      "loss": 113.656,
+      "loss/crossentropy": 2.759434461593628,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23085039854049683,
+      "loss/reg": 110.66567993164062,
+      "step": 517
+    },
+    {
+      "epoch": 0.01295,
+      "grad_norm": 2.7460856437683105,
+      "grad_norm_var": 1.169228407645687,
+      "learning_rate": 0.0001,
+      "loss": 113.58,
+      "loss/crossentropy": 3.019660472869873,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22078245878219604,
+      "loss/reg": 110.3395767211914,
+      "step": 518
+    },
+    {
+      "epoch": 0.012975,
+      "grad_norm": 2.9937329292297363,
+      "grad_norm_var": 1.1666258859232923,
+      "learning_rate": 0.0001,
+      "loss": 112.7048,
+      "loss/crossentropy": 2.477316379547119,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21349506080150604,
+      "loss/reg": 110.01399993896484,
+      "step": 519
+    },
+    {
+      "epoch": 0.013,
+      "grad_norm": 2.8256266117095947,
+      "grad_norm_var": 0.0870475905341967,
+      "learning_rate": 0.0001,
+      "loss": 112.8148,
+      "loss/crossentropy": 2.904346466064453,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22682815790176392,
+      "loss/reg": 109.68363952636719,
+      "step": 520
+    },
+    {
+      "epoch": 0.013025,
+      "grad_norm": 2.6826443672180176,
+      "grad_norm_var": 0.08334319224761823,
+      "learning_rate": 0.0001,
+      "loss": 112.4453,
+      "loss/crossentropy": 2.8740198612213135,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21089796721935272,
+      "loss/reg": 109.36034393310547,
+      "step": 521
+    },
+    {
+      "epoch": 0.01305,
+      "grad_norm": 2.876612424850464,
+      "grad_norm_var": 0.07880413790800654,
+      "learning_rate": 0.0001,
+      "loss": 112.014,
+      "loss/crossentropy": 2.752030372619629,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22804582118988037,
+      "loss/reg": 109.03395080566406,
+      "step": 522
+    },
+    {
+      "epoch": 0.013075,
+      "grad_norm": 2.6571719646453857,
+      "grad_norm_var": 0.07107003720365886,
+      "learning_rate": 0.0001,
+      "loss": 112.0514,
+      "loss/crossentropy": 3.134460926055908,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20280694961547852,
+      "loss/reg": 108.7141342163086,
+      "step": 523
+    },
+    {
+      "epoch": 0.0131,
+      "grad_norm": 2.6310062408447266,
+      "grad_norm_var": 0.07012872943871476,
+      "learning_rate": 0.0001,
+      "loss": 111.5231,
+      "loss/crossentropy": 2.899681329727173,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2234225869178772,
+      "loss/reg": 108.39997863769531,
+      "step": 524
+    },
+    {
+      "epoch": 0.013125,
+      "grad_norm": 2.8861429691314697,
+      "grad_norm_var": 0.07013051549444356,
+      "learning_rate": 0.0001,
+      "loss": 111.3517,
+      "loss/crossentropy": 3.016164779663086,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2486555576324463,
+      "loss/reg": 108.08689880371094,
+      "step": 525
+    },
+    {
+      "epoch": 0.01315,
+      "grad_norm": 11.108626365661621,
+      "grad_norm_var": 4.342596426812534,
+      "learning_rate": 0.0001,
+      "loss": 111.424,
+      "loss/crossentropy": 3.3841826915740967,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.27035826444625854,
+      "loss/reg": 107.76946258544922,
+      "step": 526
+    },
+    {
+      "epoch": 0.013175,
+      "grad_norm": 2.765962600708008,
+      "grad_norm_var": 4.355189952794281,
+      "learning_rate": 0.0001,
+      "loss": 110.3478,
+      "loss/crossentropy": 2.66182279586792,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2297273576259613,
+      "loss/reg": 107.45628356933594,
+      "step": 527
+    },
+    {
+      "epoch": 0.0132,
+      "grad_norm": 3.090799570083618,
+      "grad_norm_var": 4.340312503643294,
+      "learning_rate": 0.0001,
+      "loss": 110.4536,
+      "loss/crossentropy": 3.0606637001037598,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2523239552974701,
+      "loss/reg": 107.14060974121094,
+      "step": 528
+    },
+    {
+      "epoch": 0.013225,
+      "grad_norm": 2.63191819190979,
+      "grad_norm_var": 4.339737919036126,
+      "learning_rate": 0.0001,
+      "loss": 109.6669,
+      "loss/crossentropy": 2.6498641967773438,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1968306005001068,
+      "loss/reg": 106.8201904296875,
+      "step": 529
+    },
+    {
+      "epoch": 0.01325,
+      "grad_norm": 2.5872347354888916,
+      "grad_norm_var": 4.366497639190141,
+      "learning_rate": 0.0001,
+      "loss": 109.3358,
+      "loss/crossentropy": 2.615196466445923,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21458838880062103,
+      "loss/reg": 106.50599670410156,
+      "step": 530
+    },
+    {
+      "epoch": 0.013275,
+      "grad_norm": 3.557237148284912,
+      "grad_norm_var": 4.360969037365885,
+      "learning_rate": 0.0001,
+      "loss": 109.2467,
+      "loss/crossentropy": 2.8124053478240967,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2458515763282776,
+      "loss/reg": 106.18840789794922,
+      "step": 531
+    },
+    {
+      "epoch": 0.0133,
+      "grad_norm": 9.031005859375,
+      "grad_norm_var": 6.319656798143334,
+      "learning_rate": 0.0001,
+      "loss": 108.9254,
+      "loss/crossentropy": 2.764629602432251,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2858869433403015,
+      "loss/reg": 105.87490844726562,
+      "step": 532
+    },
+    {
+      "epoch": 0.013325,
+      "grad_norm": 3.397695302963257,
+      "grad_norm_var": 6.247992121947124,
+      "learning_rate": 0.0001,
+      "loss": 108.913,
+      "loss/crossentropy": 3.053879737854004,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2985405921936035,
+      "loss/reg": 105.5605697631836,
+      "step": 533
+    },
+    {
+      "epoch": 0.01335,
+      "grad_norm": 5.6299591064453125,
+      "grad_norm_var": 6.370482684906529,
+      "learning_rate": 0.0001,
+      "loss": 108.2695,
+      "loss/crossentropy": 2.793619155883789,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2356250137090683,
+      "loss/reg": 105.24028778076172,
+      "step": 534
+    },
+    {
+      "epoch": 0.013375,
+      "grad_norm": 3.0469837188720703,
+      "grad_norm_var": 6.363802254153521,
+      "learning_rate": 0.0001,
+      "loss": 108.5512,
+      "loss/crossentropy": 3.3861045837402344,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23634299635887146,
+      "loss/reg": 104.92872619628906,
+      "step": 535
+    },
+    {
+      "epoch": 0.0134,
+      "grad_norm": 2.819078207015991,
+      "grad_norm_var": 6.364797923503401,
+      "learning_rate": 0.0001,
+      "loss": 107.8459,
+      "loss/crossentropy": 2.994354724884033,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23442190885543823,
+      "loss/reg": 104.6171646118164,
+      "step": 536
+    },
+    {
+      "epoch": 0.013425,
+      "grad_norm": 5.84004545211792,
+      "grad_norm_var": 6.449067359728785,
+      "learning_rate": 0.0001,
+      "loss": 107.6525,
+      "loss/crossentropy": 3.0567657947540283,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2871782183647156,
+      "loss/reg": 104.30860137939453,
+      "step": 537
+    },
+    {
+      "epoch": 0.01345,
+      "grad_norm": 2.5338470935821533,
+      "grad_norm_var": 6.515056601417896,
+      "learning_rate": 0.0001,
+      "loss": 106.7596,
+      "loss/crossentropy": 2.5571489334106445,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20038628578186035,
+      "loss/reg": 104.00208282470703,
+      "step": 538
+    },
+    {
+      "epoch": 0.013475,
+      "grad_norm": 2.7171146869659424,
+      "grad_norm_var": 6.503442502818018,
+      "learning_rate": 0.0001,
+      "loss": 106.7376,
+      "loss/crossentropy": 2.8135974407196045,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22917550802230835,
+      "loss/reg": 103.69479370117188,
+      "step": 539
+    },
+    {
+      "epoch": 0.0135,
+      "grad_norm": 6.479078769683838,
+      "grad_norm_var": 6.653581035332929,
+      "learning_rate": 0.0001,
+      "loss": 106.5785,
+      "loss/crossentropy": 2.8556628227233887,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.33066555857658386,
+      "loss/reg": 103.3922119140625,
+      "step": 540
+    },
+    {
+      "epoch": 0.013525,
+      "grad_norm": 2.6559464931488037,
+      "grad_norm_var": 6.702825655017004,
+      "learning_rate": 0.0001,
+      "loss": 106.0303,
+      "loss/crossentropy": 2.7272911071777344,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21066486835479736,
+      "loss/reg": 103.09235382080078,
+      "step": 541
+    },
+    {
+      "epoch": 0.01355,
+      "grad_norm": 2.68630313873291,
+      "grad_norm_var": 3.5670498293655744,
+      "learning_rate": 0.0001,
+      "loss": 105.6865,
+      "loss/crossentropy": 2.6875898838043213,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20091715455055237,
+      "loss/reg": 102.79798126220703,
+      "step": 542
+    },
+    {
+      "epoch": 0.013575,
+      "grad_norm": 2.6095407009124756,
+      "grad_norm_var": 3.59101884290791,
+      "learning_rate": 0.0001,
+      "loss": 105.3207,
+      "loss/crossentropy": 2.607323408126831,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20690517127513885,
+      "loss/reg": 102.50646209716797,
+      "step": 543
+    },
+    {
+      "epoch": 0.0136,
+      "grad_norm": 2.6883742809295654,
+      "grad_norm_var": 3.6409168446953246,
+      "learning_rate": 0.0001,
+      "loss": 105.2911,
+      "loss/crossentropy": 2.8654534816741943,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21081852912902832,
+      "loss/reg": 102.21478271484375,
+      "step": 544
+    },
+    {
+      "epoch": 0.013625,
+      "grad_norm": 2.5874383449554443,
+      "grad_norm_var": 3.648009256619466,
+      "learning_rate": 0.0001,
+      "loss": 104.9956,
+      "loss/crossentropy": 2.858520984649658,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21754832565784454,
+      "loss/reg": 101.91951751708984,
+      "step": 545
+    },
+    {
+      "epoch": 0.01365,
+      "grad_norm": 2.54441499710083,
+      "grad_norm_var": 3.6550717570432996,
+      "learning_rate": 0.0001,
+      "loss": 104.3815,
+      "loss/crossentropy": 2.5437026023864746,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2128550112247467,
+      "loss/reg": 101.62492370605469,
+      "step": 546
+    },
+    {
+      "epoch": 0.013675,
+      "grad_norm": 2.422410488128662,
+      "grad_norm_var": 3.7725212936238335,
+      "learning_rate": 0.0001,
+      "loss": 104.2026,
+      "loss/crossentropy": 2.678537607192993,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19898445904254913,
+      "loss/reg": 101.32511138916016,
+      "step": 547
+    },
+    {
+      "epoch": 0.0137,
+      "grad_norm": 2.816225051879883,
+      "grad_norm_var": 1.7943565080708708,
+      "learning_rate": 0.0001,
+      "loss": 104.3567,
+      "loss/crossentropy": 3.10886549949646,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22006894648075104,
+      "loss/reg": 101.02771759033203,
+      "step": 548
+    },
+    {
+      "epoch": 0.013725,
+      "grad_norm": 2.9122800827026367,
+      "grad_norm_var": 1.805488475198342,
+      "learning_rate": 0.0001,
+      "loss": 103.6283,
+      "loss/crossentropy": 2.6595304012298584,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23327180743217468,
+      "loss/reg": 100.73548126220703,
+      "step": 549
+    },
+    {
+      "epoch": 0.01375,
+      "grad_norm": 2.715149402618408,
+      "grad_norm_var": 1.4355691908428239,
+      "learning_rate": 0.0001,
+      "loss": 103.5837,
+      "loss/crossentropy": 2.9111499786376953,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2255546748638153,
+      "loss/reg": 100.4470443725586,
+      "step": 550
+    },
+    {
+      "epoch": 0.013775,
+      "grad_norm": 2.6600630283355713,
+      "grad_norm_var": 1.4491900778787985,
+      "learning_rate": 0.0001,
+      "loss": 102.9596,
+      "loss/crossentropy": 2.591811180114746,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2164371758699417,
+      "loss/reg": 100.15138244628906,
+      "step": 551
+    },
+    {
+      "epoch": 0.0138,
+      "grad_norm": 2.811176061630249,
+      "grad_norm_var": 1.4494957147530347,
+      "learning_rate": 0.0001,
+      "loss": 102.7969,
+      "loss/crossentropy": 2.7154479026794434,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22633354365825653,
+      "loss/reg": 99.85511779785156,
+      "step": 552
+    },
+    {
+      "epoch": 0.013825,
+      "grad_norm": 2.5404551029205322,
+      "grad_norm_var": 0.9266648578686081,
+      "learning_rate": 0.0001,
+      "loss": 102.5948,
+      "loss/crossentropy": 2.83707857131958,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19505436718463898,
+      "loss/reg": 99.56267547607422,
+      "step": 553
+    },
+    {
+      "epoch": 0.01385,
+      "grad_norm": 3.074483871459961,
+      "grad_norm_var": 0.9186296960512579,
+      "learning_rate": 0.0001,
+      "loss": 102.4509,
+      "loss/crossentropy": 2.912346363067627,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2622039020061493,
+      "loss/reg": 99.27632904052734,
+      "step": 554
+    },
+    {
+      "epoch": 0.013875,
+      "grad_norm": 2.686178207397461,
+      "grad_norm_var": 0.9195780649456933,
+      "learning_rate": 0.0001,
+      "loss": 102.0401,
+      "loss/crossentropy": 2.8341047763824463,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21929171681404114,
+      "loss/reg": 98.98668670654297,
+      "step": 555
+    },
+    {
+      "epoch": 0.0139,
+      "grad_norm": 2.656536340713501,
+      "grad_norm_var": 0.024253446700591517,
+      "learning_rate": 0.0001,
+      "loss": 101.5869,
+      "loss/crossentropy": 2.6577670574188232,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23103003203868866,
+      "loss/reg": 98.69813537597656,
+      "step": 556
+    },
+    {
+      "epoch": 0.013925,
+      "grad_norm": 2.5510590076446533,
+      "grad_norm_var": 0.025440849818042465,
+      "learning_rate": 0.0001,
+      "loss": 100.9329,
+      "loss/crossentropy": 2.326843738555908,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18946564197540283,
+      "loss/reg": 98.41661071777344,
+      "step": 557
+    },
+    {
+      "epoch": 0.01395,
+      "grad_norm": 2.6039011478424072,
+      "grad_norm_var": 0.02585234669650897,
+      "learning_rate": 0.0001,
+      "loss": 101.0391,
+      "loss/crossentropy": 2.6925387382507324,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2089564949274063,
+      "loss/reg": 98.13763427734375,
+      "step": 558
+    },
+    {
+      "epoch": 0.013975,
+      "grad_norm": 2.4973230361938477,
+      "grad_norm_var": 0.027693340503882762,
+      "learning_rate": 0.0001,
+      "loss": 100.7989,
+      "loss/crossentropy": 2.747386932373047,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2009141445159912,
+      "loss/reg": 97.85063934326172,
+      "step": 559
+    },
+    {
+      "epoch": 0.014,
+      "grad_norm": 2.5563371181488037,
+      "grad_norm_var": 0.028511705384668556,
+      "learning_rate": 0.0001,
+      "loss": 100.641,
+      "loss/crossentropy": 2.8731026649475098,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20369890332221985,
+      "loss/reg": 97.56423950195312,
+      "step": 560
+    },
+    {
+      "epoch": 0.014025,
+      "grad_norm": 2.81292724609375,
+      "grad_norm_var": 0.029366212464935634,
+      "learning_rate": 0.0001,
+      "loss": 100.1638,
+      "loss/crossentropy": 2.6780025959014893,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20065699517726898,
+      "loss/reg": 97.28515625,
+      "step": 561
+    },
+    {
+      "epoch": 0.01405,
+      "grad_norm": 2.867063522338867,
+      "grad_norm_var": 0.03009105233082489,
+      "learning_rate": 0.0001,
+      "loss": 99.9774,
+      "loss/crossentropy": 2.7445759773254395,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2291974425315857,
+      "loss/reg": 97.00367736816406,
+      "step": 562
+    },
+    {
+      "epoch": 0.014075,
+      "grad_norm": 2.4345862865448,
+      "grad_norm_var": 0.02965133530149539,
+      "learning_rate": 0.0001,
+      "loss": 99.6903,
+      "loss/crossentropy": 2.768172264099121,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19444233179092407,
+      "loss/reg": 96.72771453857422,
+      "step": 563
+    },
+    {
+      "epoch": 0.0141,
+      "grad_norm": 2.8722984790802,
+      "grad_norm_var": 0.0307187897240811,
+      "learning_rate": 0.0001,
+      "loss": 99.6226,
+      "loss/crossentropy": 2.9429454803466797,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2194785326719284,
+      "loss/reg": 96.46016693115234,
+      "step": 564
+    },
+    {
+      "epoch": 0.014125,
+      "grad_norm": 2.9882256984710693,
+      "grad_norm_var": 0.03319604425916699,
+      "learning_rate": 0.0001,
+      "loss": 99.6137,
+      "loss/crossentropy": 3.175783395767212,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24568451941013336,
+      "loss/reg": 96.19220733642578,
+      "step": 565
+    },
+    {
+      "epoch": 0.01415,
+      "grad_norm": 2.822582721710205,
+      "grad_norm_var": 0.03402003702614896,
+      "learning_rate": 0.0001,
+      "loss": 99.1781,
+      "loss/crossentropy": 3.020627021789551,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2289256751537323,
+      "loss/reg": 95.92849731445312,
+      "step": 566
+    },
+    {
+      "epoch": 0.014175,
+      "grad_norm": 2.7644307613372803,
+      "grad_norm_var": 0.03394051714071698,
+      "learning_rate": 0.0001,
+      "loss": 98.6117,
+      "loss/crossentropy": 2.714923858642578,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2449963539838791,
+      "loss/reg": 95.65177154541016,
+      "step": 567
+    },
+    {
+      "epoch": 0.0142,
+      "grad_norm": 2.699716091156006,
+      "grad_norm_var": 0.03338014972604976,
+      "learning_rate": 0.0001,
+      "loss": 98.4869,
+      "loss/crossentropy": 2.897766351699829,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2104054093360901,
+      "loss/reg": 95.37874603271484,
+      "step": 568
+    },
+    {
+      "epoch": 0.014225,
+      "grad_norm": 2.8579864501953125,
+      "grad_norm_var": 0.03232346391338676,
+      "learning_rate": 0.0001,
+      "loss": 98.2656,
+      "loss/crossentropy": 2.9202752113342285,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23916882276535034,
+      "loss/reg": 95.10614013671875,
+      "step": 569
+    },
+    {
+      "epoch": 0.01425,
+      "grad_norm": 2.753019332885742,
+      "grad_norm_var": 0.024192763356738093,
+      "learning_rate": 0.0001,
+      "loss": 98.0069,
+      "loss/crossentropy": 2.944240093231201,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22491341829299927,
+      "loss/reg": 94.8377914428711,
+      "step": 570
+    },
+    {
+      "epoch": 0.014275,
+      "grad_norm": 2.60459566116333,
+      "grad_norm_var": 0.02491149826441017,
+      "learning_rate": 0.0001,
+      "loss": 97.608,
+      "loss/crossentropy": 2.819348096847534,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22620517015457153,
+      "loss/reg": 94.56243133544922,
+      "step": 571
+    },
+    {
+      "epoch": 0.0143,
+      "grad_norm": 2.870256185531616,
+      "grad_norm_var": 0.02627376883378929,
+      "learning_rate": 0.0001,
+      "loss": 97.2411,
+      "loss/crossentropy": 2.742431879043579,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21404124796390533,
+      "loss/reg": 94.28459167480469,
+      "step": 572
+    },
+    {
+      "epoch": 0.014325,
+      "grad_norm": 2.3389241695404053,
+      "grad_norm_var": 0.033928965438431644,
+      "learning_rate": 0.0001,
+      "loss": 96.7127,
+      "loss/crossentropy": 2.517883777618408,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19041433930397034,
+      "loss/reg": 94.0043716430664,
+      "step": 573
+    },
+    {
+      "epoch": 0.01435,
+      "grad_norm": 2.59843111038208,
+      "grad_norm_var": 0.034007496068778426,
+      "learning_rate": 0.0001,
+      "loss": 96.8607,
+      "loss/crossentropy": 2.9151978492736816,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21700671315193176,
+      "loss/reg": 93.72846221923828,
+      "step": 574
+    },
+    {
+      "epoch": 0.014375,
+      "grad_norm": 3.3285093307495117,
+      "grad_norm_var": 0.05376453050990312,
+      "learning_rate": 0.0001,
+      "loss": 96.8123,
+      "loss/crossentropy": 3.1045849323272705,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2494984120130539,
+      "loss/reg": 93.45817565917969,
+      "step": 575
+    },
+    {
+      "epoch": 0.0144,
+      "grad_norm": 2.6628782749176025,
+      "grad_norm_var": 0.0515720576900262,
+      "learning_rate": 0.0001,
+      "loss": 96.3792,
+      "loss/crossentropy": 2.955598831176758,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2415456622838974,
+      "loss/reg": 93.18202209472656,
+      "step": 576
+    },
+    {
+      "epoch": 0.014425,
+      "grad_norm": 4.186471939086914,
+      "grad_norm_var": 0.1778464831949935,
+      "learning_rate": 0.0001,
+      "loss": 96.081,
+      "loss/crossentropy": 2.8772761821746826,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2912898063659668,
+      "loss/reg": 92.9124526977539,
+      "step": 577
+    },
+    {
+      "epoch": 0.01445,
+      "grad_norm": 2.668438196182251,
+      "grad_norm_var": 0.17994305561740684,
+      "learning_rate": 0.0001,
+      "loss": 95.7919,
+      "loss/crossentropy": 2.9419538974761963,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21084845066070557,
+      "loss/reg": 92.63912200927734,
+      "step": 578
+    },
+    {
+      "epoch": 0.014475,
+      "grad_norm": 2.549457311630249,
+      "grad_norm_var": 0.17454752775228427,
+      "learning_rate": 0.0001,
+      "loss": 95.1398,
+      "loss/crossentropy": 2.5472424030303955,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2210979163646698,
+      "loss/reg": 92.37150573730469,
+      "step": 579
+    },
+    {
+      "epoch": 0.0145,
+      "grad_norm": 3.0102109909057617,
+      "grad_norm_var": 0.17618512136317932,
+      "learning_rate": 0.0001,
+      "loss": 95.1484,
+      "loss/crossentropy": 2.8119232654571533,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2312673032283783,
+      "loss/reg": 92.10520935058594,
+      "step": 580
+    },
+    {
+      "epoch": 0.014525,
+      "grad_norm": 2.673635721206665,
+      "grad_norm_var": 0.17684562367797388,
+      "learning_rate": 0.0001,
+      "loss": 94.916,
+      "loss/crossentropy": 2.840714693069458,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24130380153656006,
+      "loss/reg": 91.83395385742188,
+      "step": 581
+    },
+    {
+      "epoch": 0.01455,
+      "grad_norm": 2.600407361984253,
+      "grad_norm_var": 0.18035328363555816,
+      "learning_rate": 0.0001,
+      "loss": 94.4715,
+      "loss/crossentropy": 2.7048165798187256,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.202452152967453,
+      "loss/reg": 91.56422424316406,
+      "step": 582
+    },
+    {
+      "epoch": 0.014575,
+      "grad_norm": 3.064331531524658,
+      "grad_norm_var": 0.18363414575108336,
+      "learning_rate": 0.0001,
+      "loss": 94.3897,
+      "loss/crossentropy": 2.871046304702759,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21794039011001587,
+      "loss/reg": 91.30074310302734,
+      "step": 583
+    },
+    {
+      "epoch": 0.0146,
+      "grad_norm": 2.4271910190582275,
+      "grad_norm_var": 0.19343539696492276,
+      "learning_rate": 0.0001,
+      "loss": 94.3164,
+      "loss/crossentropy": 3.068648099899292,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20914003252983093,
+      "loss/reg": 91.03858947753906,
+      "step": 584
+    },
+    {
+      "epoch": 0.014625,
+      "grad_norm": 2.773268699645996,
+      "grad_norm_var": 0.1935076502725669,
+      "learning_rate": 0.0001,
+      "loss": 93.7491,
+      "loss/crossentropy": 2.722926378250122,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24941931664943695,
+      "loss/reg": 90.77677154541016,
+      "step": 585
+    },
+    {
+      "epoch": 0.01465,
+      "grad_norm": 2.473400592803955,
+      "grad_norm_var": 0.2008682828648034,
+      "learning_rate": 0.0001,
+      "loss": 93.6203,
+      "loss/crossentropy": 2.8903560638427734,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21103140711784363,
+      "loss/reg": 90.51893615722656,
+      "step": 586
+    },
+    {
+      "epoch": 0.014675,
+      "grad_norm": 2.6214702129364014,
+      "grad_norm_var": 0.20044215566161913,
+      "learning_rate": 0.0001,
+      "loss": 93.2864,
+      "loss/crossentropy": 2.8173437118530273,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21141797304153442,
+      "loss/reg": 90.25767517089844,
+      "step": 587
+    },
+    {
+      "epoch": 0.0147,
+      "grad_norm": 2.776019811630249,
+      "grad_norm_var": 0.20015155933538128,
+      "learning_rate": 0.0001,
+      "loss": 92.9381,
+      "loss/crossentropy": 2.7188355922698975,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21627959609031677,
+      "loss/reg": 90.00298309326172,
+      "step": 588
+    },
+    {
+      "epoch": 0.014725,
+      "grad_norm": 2.63840389251709,
+      "grad_norm_var": 0.18746319834137826,
+      "learning_rate": 0.0001,
+      "loss": 92.7113,
+      "loss/crossentropy": 2.7562100887298584,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2082541435956955,
+      "loss/reg": 89.74687957763672,
+      "step": 589
+    },
+    {
+      "epoch": 0.01475,
+      "grad_norm": 2.835721015930176,
+      "grad_norm_var": 0.1841056372587597,
+      "learning_rate": 0.0001,
+      "loss": 92.324,
+      "loss/crossentropy": 2.6325843334198,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20466585457324982,
+      "loss/reg": 89.4867935180664,
+      "step": 590
+    },
+    {
+      "epoch": 0.014775,
+      "grad_norm": 2.687760829925537,
+      "grad_norm_var": 0.16722875087179906,
+      "learning_rate": 0.0001,
+      "loss": 92.2685,
+      "loss/crossentropy": 2.82861328125,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2116677463054657,
+      "loss/reg": 89.22821044921875,
+      "step": 591
+    },
+    {
+      "epoch": 0.0148,
+      "grad_norm": 3.145275354385376,
+      "grad_norm_var": 0.17356006417378372,
+      "learning_rate": 0.0001,
+      "loss": 91.9296,
+      "loss/crossentropy": 2.7166497707366943,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24187356233596802,
+      "loss/reg": 88.9710693359375,
+      "step": 592
+    },
+    {
+      "epoch": 0.014825,
+      "grad_norm": 2.6644723415374756,
+      "grad_norm_var": 0.04118301322724444,
+      "learning_rate": 0.0001,
+      "loss": 91.8205,
+      "loss/crossentropy": 2.891160249710083,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2127176821231842,
+      "loss/reg": 88.71666717529297,
+      "step": 593
+    },
+    {
+      "epoch": 0.01485,
+      "grad_norm": 2.5222256183624268,
+      "grad_norm_var": 0.043633350924598586,
+      "learning_rate": 0.0001,
+      "loss": 91.2693,
+      "loss/crossentropy": 2.588534355163574,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2142462432384491,
+      "loss/reg": 88.46653747558594,
+      "step": 594
+    },
+    {
+      "epoch": 0.014875,
+      "grad_norm": 2.6689634323120117,
+      "grad_norm_var": 0.04186501943967471,
+      "learning_rate": 0.0001,
+      "loss": 91.1498,
+      "loss/crossentropy": 2.731534957885742,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20442616939544678,
+      "loss/reg": 88.2138442993164,
+      "step": 595
+    },
+    {
+      "epoch": 0.0149,
+      "grad_norm": 2.733809471130371,
+      "grad_norm_var": 0.03608913512671942,
+      "learning_rate": 0.0001,
+      "loss": 91.2427,
+      "loss/crossentropy": 3.0743908882141113,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21326132118701935,
+      "loss/reg": 87.95504760742188,
+      "step": 596
+    },
+    {
+      "epoch": 0.014925,
+      "grad_norm": 2.6387839317321777,
+      "grad_norm_var": 0.03631845228885571,
+      "learning_rate": 0.0001,
+      "loss": 90.743,
+      "loss/crossentropy": 2.8121533393859863,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22556063532829285,
+      "loss/reg": 87.70523834228516,
+      "step": 597
+    },
+    {
+      "epoch": 0.01495,
+      "grad_norm": 2.6532163619995117,
+      "grad_norm_var": 0.035760032396463734,
+      "learning_rate": 0.0001,
+      "loss": 90.4519,
+      "loss/crossentropy": 2.7932353019714355,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20535191893577576,
+      "loss/reg": 87.45335388183594,
+      "step": 598
+    },
+    {
+      "epoch": 0.014975,
+      "grad_norm": 2.998858690261841,
+      "grad_norm_var": 0.03291526795530686,
+      "learning_rate": 0.0001,
+      "loss": 90.3378,
+      "loss/crossentropy": 2.908860206604004,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22954756021499634,
+      "loss/reg": 87.19943237304688,
+      "step": 599
+    },
+    {
+      "epoch": 0.015,
+      "grad_norm": 13.023781776428223,
+      "grad_norm_var": 6.660256756747305,
+      "learning_rate": 0.0001,
+      "loss": 90.0128,
+      "loss/crossentropy": 2.8209877014160156,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2434820979833603,
+      "loss/reg": 86.94837188720703,
+      "step": 600
+    },
+    {
+      "epoch": 0.015025,
+      "grad_norm": 3.2308173179626465,
+      "grad_norm_var": 6.637182891814268,
+      "learning_rate": 0.0001,
+      "loss": 89.5445,
+      "loss/crossentropy": 2.5709166526794434,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.27399659156799316,
+      "loss/reg": 86.69955444335938,
+      "step": 601
+    },
+    {
+      "epoch": 0.01505,
+      "grad_norm": 2.7746541500091553,
+      "grad_norm_var": 6.605854606820831,
+      "learning_rate": 0.0001,
+      "loss": 89.6479,
+      "loss/crossentropy": 2.955522060394287,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2408815622329712,
+      "loss/reg": 86.45153045654297,
+      "step": 602
+    },
+    {
+      "epoch": 0.015075,
+      "grad_norm": 2.534477472305298,
+      "grad_norm_var": 6.615513089198629,
+      "learning_rate": 0.0001,
+      "loss": 89.5647,
+      "loss/crossentropy": 3.1383554935455322,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22149553894996643,
+      "loss/reg": 86.2048110961914,
+      "step": 603
+    },
+    {
+      "epoch": 0.0151,
+      "grad_norm": 2.6757757663726807,
+      "grad_norm_var": 6.624587476581763,
+      "learning_rate": 0.0001,
+      "loss": 88.9872,
+      "loss/crossentropy": 2.8011903762817383,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22577175498008728,
+      "loss/reg": 85.960205078125,
+      "step": 604
+    },
+    {
+      "epoch": 0.015125,
+      "grad_norm": 2.5399699211120605,
+      "grad_norm_var": 6.635210790627498,
+      "learning_rate": 0.0001,
+      "loss": 88.6121,
+      "loss/crossentropy": 2.686098337173462,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21132361888885498,
+      "loss/reg": 85.71470642089844,
+      "step": 605
+    },
+    {
+      "epoch": 0.01515,
+      "grad_norm": 3.094851016998291,
+      "grad_norm_var": 6.620065609075882,
+      "learning_rate": 0.0001,
+      "loss": 88.5017,
+      "loss/crossentropy": 2.7985479831695557,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23451220989227295,
+      "loss/reg": 85.46868896484375,
+      "step": 606
+    },
+    {
+      "epoch": 0.015175,
+      "grad_norm": 2.8335752487182617,
+      "grad_norm_var": 6.607319105523461,
+      "learning_rate": 0.0001,
+      "loss": 88.5627,
+      "loss/crossentropy": 3.1135289669036865,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22703945636749268,
+      "loss/reg": 85.22212219238281,
+      "step": 607
+    },
+    {
+      "epoch": 0.0152,
+      "grad_norm": 2.805544137954712,
+      "grad_norm_var": 6.627015267189299,
+      "learning_rate": 0.0001,
+      "loss": 88.0827,
+      "loss/crossentropy": 2.8752803802490234,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22529396414756775,
+      "loss/reg": 84.98210144042969,
+      "step": 608
+    },
+    {
+      "epoch": 0.015225,
+      "grad_norm": 2.7001230716705322,
+      "grad_norm_var": 6.623600272248156,
+      "learning_rate": 0.0001,
+      "loss": 87.6396,
+      "loss/crossentropy": 2.6898162364959717,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21284155547618866,
+      "loss/reg": 84.73692321777344,
+      "step": 609
+    },
+    {
+      "epoch": 0.01525,
+      "grad_norm": 3.2781596183776855,
+      "grad_norm_var": 6.570657725923031,
+      "learning_rate": 0.0001,
+      "loss": 87.7335,
+      "loss/crossentropy": 2.9826276302337646,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25785699486732483,
+      "loss/reg": 84.4930191040039,
+      "step": 610
+    },
+    {
+      "epoch": 0.015275,
+      "grad_norm": 2.449453830718994,
+      "grad_norm_var": 6.596501814133948,
+      "learning_rate": 0.0001,
+      "loss": 87.1477,
+      "loss/crossentropy": 2.6783015727996826,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22048676013946533,
+      "loss/reg": 84.2489242553711,
+      "step": 611
+    },
+    {
+      "epoch": 0.0153,
+      "grad_norm": 2.459174394607544,
+      "grad_norm_var": 6.62690543519524,
+      "learning_rate": 0.0001,
+      "loss": 86.776,
+      "loss/crossentropy": 2.574934244155884,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19860099256038666,
+      "loss/reg": 84.0024185180664,
+      "step": 612
+    },
+    {
+      "epoch": 0.015325,
+      "grad_norm": 2.941655158996582,
+      "grad_norm_var": 6.601163552477004,
+      "learning_rate": 0.0001,
+      "loss": 86.9776,
+      "loss/crossentropy": 2.986813545227051,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23044732213020325,
+      "loss/reg": 83.76029968261719,
+      "step": 613
+    },
+    {
+      "epoch": 0.01535,
+      "grad_norm": 2.6428897380828857,
+      "grad_norm_var": 6.602249575617103,
+      "learning_rate": 0.0001,
+      "loss": 86.4506,
+      "loss/crossentropy": 2.6974997520446777,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2369295060634613,
+      "loss/reg": 83.51612091064453,
+      "step": 614
+    },
+    {
+      "epoch": 0.015375,
+      "grad_norm": 2.6571388244628906,
+      "grad_norm_var": 6.629487272361301,
+      "learning_rate": 0.0001,
+      "loss": 86.3883,
+      "loss/crossentropy": 2.886151075363159,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22736769914627075,
+      "loss/reg": 83.27473449707031,
+      "step": 615
+    },
+    {
+      "epoch": 0.0154,
+      "grad_norm": 2.609266996383667,
+      "grad_norm_var": 0.06577351547558692,
+      "learning_rate": 0.0001,
+      "loss": 86.0519,
+      "loss/crossentropy": 2.794356346130371,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22279015183448792,
+      "loss/reg": 83.0347900390625,
+      "step": 616
+    },
+    {
+      "epoch": 0.015425,
+      "grad_norm": 3.1104328632354736,
+      "grad_norm_var": 0.05918982306040803,
+      "learning_rate": 0.0001,
+      "loss": 85.9367,
+      "loss/crossentropy": 2.882596015930176,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25845298171043396,
+      "loss/reg": 82.79560852050781,
+      "step": 617
+    },
+    {
+      "epoch": 0.01545,
+      "grad_norm": 2.470853090286255,
+      "grad_norm_var": 0.06423085419138465,
+      "learning_rate": 0.0001,
+      "loss": 85.5665,
+      "loss/crossentropy": 2.796099901199341,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.214504212141037,
+      "loss/reg": 82.55593872070312,
+      "step": 618
+    },
+    {
+      "epoch": 0.015475,
+      "grad_norm": 2.714553117752075,
+      "grad_norm_var": 0.06137795437797967,
+      "learning_rate": 0.0001,
+      "loss": 85.3188,
+      "loss/crossentropy": 2.7812206745147705,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2175397127866745,
+      "loss/reg": 82.32003021240234,
+      "step": 619
+    },
+    {
+      "epoch": 0.0155,
+      "grad_norm": 2.775290012359619,
+      "grad_norm_var": 0.061025800256583315,
+      "learning_rate": 0.0001,
+      "loss": 85.1917,
+      "loss/crossentropy": 2.8803231716156006,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22989241778850555,
+      "loss/reg": 82.0815200805664,
+      "step": 620
+    },
+    {
+      "epoch": 0.015525,
+      "grad_norm": 3.064432382583618,
+      "grad_norm_var": 0.06316760074340261,
+      "learning_rate": 0.0001,
+      "loss": 84.9975,
+      "loss/crossentropy": 2.9176337718963623,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2374771535396576,
+      "loss/reg": 81.8424072265625,
+      "step": 621
+    },
+    {
+      "epoch": 0.01555,
+      "grad_norm": 2.5047502517700195,
+      "grad_norm_var": 0.0607852310360857,
+      "learning_rate": 0.0001,
+      "loss": 84.4023,
+      "loss/crossentropy": 2.5925066471099854,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20711900293827057,
+      "loss/reg": 81.6026611328125,
+      "step": 622
+    },
+    {
+      "epoch": 0.015575,
+      "grad_norm": 2.9647669792175293,
+      "grad_norm_var": 0.06330394741956387,
+      "learning_rate": 0.0001,
+      "loss": 84.5351,
+      "loss/crossentropy": 2.934492826461792,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23662039637565613,
+      "loss/reg": 81.36399841308594,
+      "step": 623
+    },
+    {
+      "epoch": 0.0156,
+      "grad_norm": 2.6432931423187256,
+      "grad_norm_var": 0.06394843640099997,
+      "learning_rate": 0.0001,
+      "loss": 84.4146,
+      "loss/crossentropy": 3.0514731407165527,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22952324151992798,
+      "loss/reg": 81.13361358642578,
+      "step": 624
+    },
+    {
+      "epoch": 0.015625,
+      "grad_norm": 2.54736065864563,
+      "grad_norm_var": 0.06640534283560531,
+      "learning_rate": 0.0001,
+      "loss": 83.7896,
+      "loss/crossentropy": 2.670376777648926,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21750324964523315,
+      "loss/reg": 80.90174865722656,
+      "step": 625
+    },
+    {
+      "epoch": 0.01565,
+      "grad_norm": 2.8746328353881836,
+      "grad_norm_var": 0.047605595082000934,
+      "learning_rate": 0.0001,
+      "loss": 83.6016,
+      "loss/crossentropy": 2.705821990966797,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22585441172122955,
+      "loss/reg": 80.66989135742188,
+      "step": 626
+    },
+    {
+      "epoch": 0.015675,
+      "grad_norm": 4.1663737297058105,
+      "grad_norm_var": 0.17119830661165428,
+      "learning_rate": 0.0001,
+      "loss": 83.7693,
+      "loss/crossentropy": 3.0719547271728516,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25689268112182617,
+      "loss/reg": 80.44044494628906,
+      "step": 627
+    },
+    {
+      "epoch": 0.0157,
+      "grad_norm": 2.9214797019958496,
+      "grad_norm_var": 0.16221115285843457,
+      "learning_rate": 0.0001,
+      "loss": 83.4368,
+      "loss/crossentropy": 2.9782841205596924,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24059516191482544,
+      "loss/reg": 80.21793365478516,
+      "step": 628
+    },
+    {
+      "epoch": 0.015725,
+      "grad_norm": 2.3547921180725098,
+      "grad_norm_var": 0.17660964070513122,
+      "learning_rate": 0.0001,
+      "loss": 82.9563,
+      "loss/crossentropy": 2.759288787841797,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2049524188041687,
+      "loss/reg": 79.99205780029297,
+      "step": 629
+    },
+    {
+      "epoch": 0.01575,
+      "grad_norm": 2.5864202976226807,
+      "grad_norm_var": 0.17809647704259438,
+      "learning_rate": 0.0001,
+      "loss": 82.856,
+      "loss/crossentropy": 2.891166925430298,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19927763938903809,
+      "loss/reg": 79.76554870605469,
+      "step": 630
+    },
+    {
+      "epoch": 0.015775,
+      "grad_norm": 2.579041004180908,
+      "grad_norm_var": 0.1800732301463995,
+      "learning_rate": 0.0001,
+      "loss": 82.3371,
+      "loss/crossentropy": 2.578899621963501,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21475175023078918,
+      "loss/reg": 79.54348754882812,
+      "step": 631
+    },
+    {
+      "epoch": 0.0158,
+      "grad_norm": 2.8605313301086426,
+      "grad_norm_var": 0.17744545594942238,
+      "learning_rate": 0.0001,
+      "loss": 82.2226,
+      "loss/crossentropy": 2.6955201625823975,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20546796917915344,
+      "loss/reg": 79.32160186767578,
+      "step": 632
+    },
+    {
+      "epoch": 0.015825,
+      "grad_norm": 2.458681344985962,
+      "grad_norm_var": 0.1788587470198704,
+      "learning_rate": 0.0001,
+      "loss": 81.9029,
+      "loss/crossentropy": 2.6130025386810303,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19218957424163818,
+      "loss/reg": 79.09767150878906,
+      "step": 633
+    },
+    {
+      "epoch": 0.01585,
+      "grad_norm": 2.6389007568359375,
+      "grad_norm_var": 0.1736867369098557,
+      "learning_rate": 0.0001,
+      "loss": 82.1122,
+      "loss/crossentropy": 2.994948148727417,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24412985146045685,
+      "loss/reg": 78.87307739257812,
+      "step": 634
+    },
+    {
+      "epoch": 0.015875,
+      "grad_norm": 2.7426064014434814,
+      "grad_norm_var": 0.17345014249303006,
+      "learning_rate": 0.0001,
+      "loss": 81.7675,
+      "loss/crossentropy": 2.9031403064727783,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2145020067691803,
+      "loss/reg": 78.64982604980469,
+      "step": 635
+    },
+    {
+      "epoch": 0.0159,
+      "grad_norm": 2.5872960090637207,
+      "grad_norm_var": 0.176095637618936,
+      "learning_rate": 0.0001,
+      "loss": 81.2096,
+      "loss/crossentropy": 2.5721232891082764,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20824509859085083,
+      "loss/reg": 78.42925262451172,
+      "step": 636
+    },
+    {
+      "epoch": 0.015925,
+      "grad_norm": 2.639803886413574,
+      "grad_norm_var": 0.17131557533067104,
+      "learning_rate": 0.0001,
+      "loss": 81.7399,
+      "loss/crossentropy": 3.311671018600464,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2240554839372635,
+      "loss/reg": 78.20418548583984,
+      "step": 637
+    },
+    {
+      "epoch": 0.01595,
+      "grad_norm": 2.7064106464385986,
+      "grad_norm_var": 0.1671441066181302,
+      "learning_rate": 0.0001,
+      "loss": 81.0741,
+      "loss/crossentropy": 2.8790698051452637,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21729126572608948,
+      "loss/reg": 77.97772979736328,
+      "step": 638
+    },
+    {
+      "epoch": 0.015975,
+      "grad_norm": 2.6912827491760254,
+      "grad_norm_var": 0.16460811219505161,
+      "learning_rate": 0.0001,
+      "loss": 81.0138,
+      "loss/crossentropy": 3.024972915649414,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2377013862133026,
+      "loss/reg": 77.75112915039062,
+      "step": 639
+    },
+    {
+      "epoch": 0.016,
+      "grad_norm": 2.8207814693450928,
+      "grad_norm_var": 0.16405338147699144,
+      "learning_rate": 0.0001,
+      "loss": 80.6592,
+      "loss/crossentropy": 2.9033145904541016,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23379957675933838,
+      "loss/reg": 77.52207946777344,
+      "step": 640
+    },
+    {
+      "epoch": 0.016025,
+      "grad_norm": 2.738354444503784,
+      "grad_norm_var": 0.16089216214350396,
+      "learning_rate": 0.0001,
+      "loss": 80.3146,
+      "loss/crossentropy": 2.778174877166748,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24535414576530457,
+      "loss/reg": 77.29109954833984,
+      "step": 641
+    },
+    {
+      "epoch": 0.01605,
+      "grad_norm": 3.2551162242889404,
+      "grad_norm_var": 0.17509802330359298,
+      "learning_rate": 0.0001,
+      "loss": 80.2895,
+      "loss/crossentropy": 2.9776339530944824,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24247264862060547,
+      "loss/reg": 77.0693588256836,
+      "step": 642
+    },
+    {
+      "epoch": 0.016075,
+      "grad_norm": 2.881504535675049,
+      "grad_norm_var": 0.043638895164599255,
+      "learning_rate": 0.0001,
+      "loss": 80.1407,
+      "loss/crossentropy": 3.0473055839538574,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24725341796875,
+      "loss/reg": 76.84616088867188,
+      "step": 643
+    },
+    {
+      "epoch": 0.0161,
+      "grad_norm": 2.992424249649048,
+      "grad_norm_var": 0.045893014160188275,
+      "learning_rate": 0.0001,
+      "loss": 79.5319,
+      "loss/crossentropy": 2.674531936645508,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23270609974861145,
+      "loss/reg": 76.62467956542969,
+      "step": 644
+    },
+    {
+      "epoch": 0.016125,
+      "grad_norm": 2.4216911792755127,
+      "grad_norm_var": 0.04290734773771661,
+      "learning_rate": 0.0001,
+      "loss": 79.4997,
+      "loss/crossentropy": 2.8802490234375,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21543535590171814,
+      "loss/reg": 76.40399932861328,
+      "step": 645
+    },
+    {
+      "epoch": 0.01615,
+      "grad_norm": 2.5457189083099365,
+      "grad_norm_var": 0.043763224077495264,
+      "learning_rate": 0.0001,
+      "loss": 79.235,
+      "loss/crossentropy": 2.8406364917755127,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20543472468852997,
+      "loss/reg": 76.18891143798828,
+      "step": 646
+    },
+    {
+      "epoch": 0.016175,
+      "grad_norm": 2.762531280517578,
+      "grad_norm_var": 0.04235751730274515,
+      "learning_rate": 0.0001,
+      "loss": 78.9388,
+      "loss/crossentropy": 2.7533273696899414,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21493667364120483,
+      "loss/reg": 75.97050476074219,
+      "step": 647
+    },
+    {
+      "epoch": 0.0162,
+      "grad_norm": 2.390070676803589,
+      "grad_norm_var": 0.048252346296710394,
+      "learning_rate": 0.0001,
+      "loss": 78.7404,
+      "loss/crossentropy": 2.7817790508270264,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21161305904388428,
+      "loss/reg": 75.74702453613281,
+      "step": 648
+    },
+    {
+      "epoch": 0.016225,
+      "grad_norm": 2.534242868423462,
+      "grad_norm_var": 0.046131862120249896,
+      "learning_rate": 0.0001,
+      "loss": 78.2552,
+      "loss/crossentropy": 2.5109705924987793,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20796047151088715,
+      "loss/reg": 75.53626251220703,
+      "step": 649
+    },
+    {
+      "epoch": 0.01625,
+      "grad_norm": 2.9576361179351807,
+      "grad_norm_var": 0.04948971532782949,
+      "learning_rate": 0.0001,
+      "loss": 78.448,
+      "loss/crossentropy": 2.9173483848571777,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20767123997211456,
+      "loss/reg": 75.32296752929688,
+      "step": 650
+    },
+    {
+      "epoch": 0.016275,
+      "grad_norm": 3.188835620880127,
+      "grad_norm_var": 0.06273138119426373,
+      "learning_rate": 0.0001,
+      "loss": 78.3685,
+      "loss/crossentropy": 3.024512529373169,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23480483889579773,
+      "loss/reg": 75.10919189453125,
+      "step": 651
+    },
+    {
+      "epoch": 0.0163,
+      "grad_norm": 2.9349732398986816,
+      "grad_norm_var": 0.06241445749091478,
+      "learning_rate": 0.0001,
+      "loss": 78.1327,
+      "loss/crossentropy": 3.004549264907837,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2360680103302002,
+      "loss/reg": 74.8920669555664,
+      "step": 652
+    },
+    {
+      "epoch": 0.016325,
+      "grad_norm": 2.927232265472412,
+      "grad_norm_var": 0.06224965786214606,
+      "learning_rate": 0.0001,
+      "loss": 77.718,
+      "loss/crossentropy": 2.8078725337982178,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22627294063568115,
+      "loss/reg": 74.68383026123047,
+      "step": 653
+    },
+    {
+      "epoch": 0.01635,
+      "grad_norm": 2.6294076442718506,
+      "grad_norm_var": 0.06354828695962549,
+      "learning_rate": 0.0001,
+      "loss": 77.6217,
+      "loss/crossentropy": 2.906186580657959,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24207134544849396,
+      "loss/reg": 74.47340393066406,
+      "step": 654
+    },
+    {
+      "epoch": 0.016375,
+      "grad_norm": 2.4185891151428223,
+      "grad_norm_var": 0.07185744774492757,
+      "learning_rate": 0.0001,
+      "loss": 77.1087,
+      "loss/crossentropy": 2.6219069957733154,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21973133087158203,
+      "loss/reg": 74.26707458496094,
+      "step": 655
+    },
+    {
+      "epoch": 0.0164,
+      "grad_norm": 3.0168731212615967,
+      "grad_norm_var": 0.07545913020925733,
+      "learning_rate": 0.0001,
+      "loss": 77.1465,
+      "loss/crossentropy": 2.838118076324463,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24065163731575012,
+      "loss/reg": 74.0677719116211,
+      "step": 656
+    },
+    {
+      "epoch": 0.016425,
+      "grad_norm": 2.7818074226379395,
+      "grad_norm_var": 0.07529414177001831,
+      "learning_rate": 0.0001,
+      "loss": 76.9287,
+      "loss/crossentropy": 2.8338921070098877,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22610336542129517,
+      "loss/reg": 73.86873626708984,
+      "step": 657
+    },
+    {
+      "epoch": 0.01645,
+      "grad_norm": 2.854801654815674,
+      "grad_norm_var": 0.06047968099176065,
+      "learning_rate": 0.0001,
+      "loss": 76.6613,
+      "loss/crossentropy": 2.7867751121520996,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.209943026304245,
+      "loss/reg": 73.6645736694336,
+      "step": 658
+    },
+    {
+      "epoch": 0.016475,
+      "grad_norm": 2.9934234619140625,
+      "grad_norm_var": 0.063002636345387,
+      "learning_rate": 0.0001,
+      "loss": 76.8956,
+      "loss/crossentropy": 3.177863359451294,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24757607281208038,
+      "loss/reg": 73.47013854980469,
+      "step": 659
+    },
+    {
+      "epoch": 0.0165,
+      "grad_norm": 2.6304805278778076,
+      "grad_norm_var": 0.06054759846060639,
+      "learning_rate": 0.0001,
+      "loss": 76.1319,
+      "loss/crossentropy": 2.6409659385681152,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21888282895088196,
+      "loss/reg": 73.27206420898438,
+      "step": 660
+    },
+    {
+      "epoch": 0.016525,
+      "grad_norm": 2.7408792972564697,
+      "grad_norm_var": 0.05297394175892407,
+      "learning_rate": 0.0001,
+      "loss": 76.4855,
+      "loss/crossentropy": 3.1915743350982666,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2194949984550476,
+      "loss/reg": 73.07437896728516,
+      "step": 661
+    },
+    {
+      "epoch": 0.01655,
+      "grad_norm": 2.7700512409210205,
+      "grad_norm_var": 0.04943414917226316,
+      "learning_rate": 0.0001,
+      "loss": 75.886,
+      "loss/crossentropy": 2.7812137603759766,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2344488501548767,
+      "loss/reg": 72.8703384399414,
+      "step": 662
+    },
+    {
+      "epoch": 0.016575,
+      "grad_norm": 2.6418073177337646,
+      "grad_norm_var": 0.050678375391551594,
+      "learning_rate": 0.0001,
+      "loss": 75.6395,
+      "loss/crossentropy": 2.7417471408843994,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2233429104089737,
+      "loss/reg": 72.67445373535156,
+      "step": 663
+    },
+    {
+      "epoch": 0.0166,
+      "grad_norm": 3.158125162124634,
+      "grad_norm_var": 0.04805692783096613,
+      "learning_rate": 0.0001,
+      "loss": 75.6306,
+      "loss/crossentropy": 2.9241199493408203,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22319327294826508,
+      "loss/reg": 72.48330688476562,
+      "step": 664
+    },
+    {
+      "epoch": 0.016625,
+      "grad_norm": 2.736354351043701,
+      "grad_norm_var": 0.042809702674100926,
+      "learning_rate": 0.0001,
+      "loss": 75.3328,
+      "loss/crossentropy": 2.8228261470794678,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22843888401985168,
+      "loss/reg": 72.28150939941406,
+      "step": 665
+    },
+    {
+      "epoch": 0.01665,
+      "grad_norm": 2.7677245140075684,
+      "grad_norm_var": 0.04199219130631846,
+      "learning_rate": 0.0001,
+      "loss": 74.9261,
+      "loss/crossentropy": 2.627197027206421,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22012290358543396,
+      "loss/reg": 72.07875061035156,
+      "step": 666
+    },
+    {
+      "epoch": 0.016675,
+      "grad_norm": 2.5004570484161377,
+      "grad_norm_var": 0.03816500903700728,
+      "learning_rate": 0.0001,
+      "loss": 74.777,
+      "loss/crossentropy": 2.672111988067627,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22588777542114258,
+      "loss/reg": 71.87902069091797,
+      "step": 667
+    },
+    {
+      "epoch": 0.0167,
+      "grad_norm": 2.466562509536743,
+      "grad_norm_var": 0.042288959656074356,
+      "learning_rate": 0.0001,
+      "loss": 74.7091,
+      "loss/crossentropy": 2.8123865127563477,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21958181262016296,
+      "loss/reg": 71.6771469116211,
+      "step": 668
+    },
+    {
+      "epoch": 0.016725,
+      "grad_norm": 2.751133918762207,
+      "grad_norm_var": 0.04011649012775607,
+      "learning_rate": 0.0001,
+      "loss": 74.4675,
+      "loss/crossentropy": 2.754987955093384,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23784713447093964,
+      "loss/reg": 71.4747085571289,
+      "step": 669
+    },
+    {
+      "epoch": 0.01675,
+      "grad_norm": 2.6867825984954834,
+      "grad_norm_var": 0.03946736718336652,
+      "learning_rate": 0.0001,
+      "loss": 74.5895,
+      "loss/crossentropy": 3.101982593536377,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21628157794475555,
+      "loss/reg": 71.27123260498047,
+      "step": 670
+    },
+    {
+      "epoch": 0.016775,
+      "grad_norm": 2.500910997390747,
+      "grad_norm_var": 0.03631099988891246,
+      "learning_rate": 0.0001,
+      "loss": 73.8414,
+      "loss/crossentropy": 2.5617518424987793,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20989301800727844,
+      "loss/reg": 71.06979370117188,
+      "step": 671
+    },
+    {
+      "epoch": 0.0168,
+      "grad_norm": 2.473396062850952,
+      "grad_norm_var": 0.03542460303709258,
+      "learning_rate": 0.0001,
+      "loss": 73.9086,
+      "loss/crossentropy": 2.824228286743164,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2157234251499176,
+      "loss/reg": 70.86860656738281,
+      "step": 672
+    },
+    {
+      "epoch": 0.016825,
+      "grad_norm": 2.8463189601898193,
+      "grad_norm_var": 0.036251456664882414,
+      "learning_rate": 0.0001,
+      "loss": 73.6464,
+      "loss/crossentropy": 2.7426350116729736,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23804283142089844,
+      "loss/reg": 70.66572570800781,
+      "step": 673
+    },
+    {
+      "epoch": 0.01685,
+      "grad_norm": 2.4676475524902344,
+      "grad_norm_var": 0.03865839021441365,
+      "learning_rate": 0.0001,
+      "loss": 73.4992,
+      "loss/crossentropy": 2.813382863998413,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21776646375656128,
+      "loss/reg": 70.46805572509766,
+      "step": 674
+    },
+    {
+      "epoch": 0.016875,
+      "grad_norm": 3.2696657180786133,
+      "grad_norm_var": 0.054391622405608804,
+      "learning_rate": 0.0001,
+      "loss": 73.4994,
+      "loss/crossentropy": 2.9931130409240723,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2343236804008484,
+      "loss/reg": 70.27191925048828,
+      "step": 675
+    },
+    {
+      "epoch": 0.0169,
+      "grad_norm": 2.960261344909668,
+      "grad_norm_var": 0.05755957057574411,
+      "learning_rate": 0.0001,
+      "loss": 73.1637,
+      "loss/crossentropy": 2.845167398452759,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24311169981956482,
+      "loss/reg": 70.0754165649414,
+      "step": 676
+    },
+    {
+      "epoch": 0.016925,
+      "grad_norm": 2.8058366775512695,
+      "grad_norm_var": 0.057886073712972795,
+      "learning_rate": 0.0001,
+      "loss": 72.9262,
+      "loss/crossentropy": 2.7900030612945557,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2522013783454895,
+      "loss/reg": 69.8840103149414,
+      "step": 677
+    },
+    {
+      "epoch": 0.01695,
+      "grad_norm": 2.6820642948150635,
+      "grad_norm_var": 0.05799027827774973,
+      "learning_rate": 0.0001,
+      "loss": 72.9552,
+      "loss/crossentropy": 3.0160844326019287,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25032639503479004,
+      "loss/reg": 69.68882751464844,
+      "step": 678
+    },
+    {
+      "epoch": 0.016975,
+      "grad_norm": 2.756159782409668,
+      "grad_norm_var": 0.05742948572952381,
+      "learning_rate": 0.0001,
+      "loss": 72.4603,
+      "loss/crossentropy": 2.7341785430908203,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23339924216270447,
+      "loss/reg": 69.49270629882812,
+      "step": 679
+    },
+    {
+      "epoch": 0.017,
+      "grad_norm": 2.85196852684021,
+      "grad_norm_var": 0.04619244950055569,
+      "learning_rate": 0.0001,
+      "loss": 72.4025,
+      "loss/crossentropy": 2.8541834354400635,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2513972520828247,
+      "loss/reg": 69.29695129394531,
+      "step": 680
+    },
+    {
+      "epoch": 0.017025,
+      "grad_norm": 2.745920419692993,
+      "grad_norm_var": 0.046218769763096884,
+      "learning_rate": 0.0001,
+      "loss": 72.2013,
+      "loss/crossentropy": 2.8387668132781982,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2571982443332672,
+      "loss/reg": 69.1053237915039,
+      "step": 681
+    },
+    {
+      "epoch": 0.01705,
+      "grad_norm": 2.7895922660827637,
+      "grad_norm_var": 0.046385473001735275,
+      "learning_rate": 0.0001,
+      "loss": 72.1027,
+      "loss/crossentropy": 2.9553239345550537,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23462611436843872,
+      "loss/reg": 68.9127426147461,
+      "step": 682
+    },
+    {
+      "epoch": 0.017075,
+      "grad_norm": 3.0583336353302,
+      "grad_norm_var": 0.049345512699435545,
+      "learning_rate": 0.0001,
+      "loss": 71.6065,
+      "loss/crossentropy": 2.6561527252197266,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22944068908691406,
+      "loss/reg": 68.72093963623047,
+      "step": 683
+    },
+    {
+      "epoch": 0.0171,
+      "grad_norm": 10.506353378295898,
+      "grad_norm_var": 3.7778572455504134,
+      "learning_rate": 0.0001,
+      "loss": 71.4093,
+      "loss/crossentropy": 2.677732467651367,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20256610214710236,
+      "loss/reg": 68.52899169921875,
+      "step": 684
+    },
+    {
+      "epoch": 0.017125,
+      "grad_norm": 2.7688822746276855,
+      "grad_norm_var": 3.77667386049867,
+      "learning_rate": 0.0001,
+      "loss": 71.4833,
+      "loss/crossentropy": 2.910447359085083,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23493064939975739,
+      "loss/reg": 68.33790588378906,
+      "step": 685
+    },
+    {
+      "epoch": 0.01715,
+      "grad_norm": 2.6812407970428467,
+      "grad_norm_var": 3.77709980042232,
+      "learning_rate": 0.0001,
+      "loss": 70.9918,
+      "loss/crossentropy": 2.609196424484253,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23388031125068665,
+      "loss/reg": 68.14875030517578,
+      "step": 686
+    },
+    {
+      "epoch": 0.017175,
+      "grad_norm": 2.8212389945983887,
+      "grad_norm_var": 3.7510797794332498,
+      "learning_rate": 0.0001,
+      "loss": 71.0687,
+      "loss/crossentropy": 2.8833634853363037,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22528675198554993,
+      "loss/reg": 67.9600830078125,
+      "step": 687
+    },
+    {
+      "epoch": 0.0172,
+      "grad_norm": 2.6118838787078857,
+      "grad_norm_var": 3.737378850831258,
+      "learning_rate": 0.0001,
+      "loss": 70.4259,
+      "loss/crossentropy": 2.438934564590454,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21688072383403778,
+      "loss/reg": 67.7700424194336,
+      "step": 688
+    },
+    {
+      "epoch": 0.017225,
+      "grad_norm": 2.8803842067718506,
+      "grad_norm_var": 3.7354408858260792,
+      "learning_rate": 0.0001,
+      "loss": 70.6865,
+      "loss/crossentropy": 2.8623788356781006,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23739641904830933,
+      "loss/reg": 67.58670043945312,
+      "step": 689
+    },
+    {
+      "epoch": 0.01725,
+      "grad_norm": 2.9390385150909424,
+      "grad_norm_var": 3.697573889963882,
+      "learning_rate": 0.0001,
+      "loss": 70.1731,
+      "loss/crossentropy": 2.5417747497558594,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2275683581829071,
+      "loss/reg": 67.40377044677734,
+      "step": 690
+    },
+    {
+      "epoch": 0.017275,
+      "grad_norm": 3.4491701126098633,
+      "grad_norm_var": 3.6983698569325107,
+      "learning_rate": 0.0001,
+      "loss": 70.0581,
+      "loss/crossentropy": 2.6056199073791504,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23559321463108063,
+      "loss/reg": 67.2168960571289,
+      "step": 691
+    },
+    {
+      "epoch": 0.0173,
+      "grad_norm": 3.0338120460510254,
+      "grad_norm_var": 3.695064661679089,
+      "learning_rate": 0.0001,
+      "loss": 70.0615,
+      "loss/crossentropy": 2.7806742191314697,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24799373745918274,
+      "loss/reg": 67.03279876708984,
+      "step": 692
+    },
+    {
+      "epoch": 0.017325,
+      "grad_norm": 2.7937796115875244,
+      "grad_norm_var": 3.6959266334784027,
+      "learning_rate": 0.0001,
+      "loss": 69.9194,
+      "loss/crossentropy": 2.838820219039917,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22827139496803284,
+      "loss/reg": 66.85226440429688,
+      "step": 693
+    },
+    {
+      "epoch": 0.01735,
+      "grad_norm": 2.4988961219787598,
+      "grad_norm_var": 3.713984810158103,
+      "learning_rate": 0.0001,
+      "loss": 69.5952,
+      "loss/crossentropy": 2.718376398086548,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20215165615081787,
+      "loss/reg": 66.67467498779297,
+      "step": 694
+    },
+    {
+      "epoch": 0.017375,
+      "grad_norm": 2.665208339691162,
+      "grad_norm_var": 3.721389950709211,
+      "learning_rate": 0.0001,
+      "loss": 69.5769,
+      "loss/crossentropy": 2.8547093868255615,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2223222255706787,
+      "loss/reg": 66.49986267089844,
+      "step": 695
+    },
+    {
+      "epoch": 0.0174,
+      "grad_norm": 2.7634010314941406,
+      "grad_norm_var": 3.7273892640509603,
+      "learning_rate": 0.0001,
+      "loss": 69.3502,
+      "loss/crossentropy": 2.8282363414764404,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2077331691980362,
+      "loss/reg": 66.31417846679688,
+      "step": 696
+    },
+    {
+      "epoch": 0.017425,
+      "grad_norm": 3.8694913387298584,
+      "grad_norm_var": 3.7213441994990677,
+      "learning_rate": 0.0001,
+      "loss": 70.1241,
+      "loss/crossentropy": 3.720386266708374,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2679758071899414,
+      "loss/reg": 66.13578796386719,
+      "step": 697
+    },
+    {
+      "epoch": 0.01745,
+      "grad_norm": 5.2169013023376465,
+      "grad_norm_var": 3.897477580961625,
+      "learning_rate": 0.0001,
+      "loss": 69.5964,
+      "loss/crossentropy": 3.217229127883911,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.42148905992507935,
+      "loss/reg": 65.95770263671875,
+      "step": 698
+    },
+    {
+      "epoch": 0.017475,
+      "grad_norm": 3.6213738918304443,
+      "grad_norm_var": 3.8815159738632965,
+      "learning_rate": 0.0001,
+      "loss": 68.5239,
+      "loss/crossentropy": 2.502047061920166,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24625252187252045,
+      "loss/reg": 65.7756118774414,
+      "step": 699
+    },
+    {
+      "epoch": 0.0175,
+      "grad_norm": 2.716905355453491,
+      "grad_norm_var": 0.46975474422527747,
+      "learning_rate": 0.0001,
+      "loss": 68.5619,
+      "loss/crossentropy": 2.764882802963257,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20099103450775146,
+      "loss/reg": 65.5960693359375,
+      "step": 700
+    },
+    {
+      "epoch": 0.017525,
+      "grad_norm": 2.6961023807525635,
+      "grad_norm_var": 0.47313618338585167,
+      "learning_rate": 0.0001,
+      "loss": 68.6203,
+      "loss/crossentropy": 2.978670358657837,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2190450131893158,
+      "loss/reg": 65.42259216308594,
+      "step": 701
+    },
+    {
+      "epoch": 0.01755,
+      "grad_norm": 2.7581896781921387,
+      "grad_norm_var": 0.46942862049069445,
+      "learning_rate": 0.0001,
+      "loss": 68.2458,
+      "loss/crossentropy": 2.773253917694092,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2290889024734497,
+      "loss/reg": 65.24346160888672,
+      "step": 702
+    },
+    {
+      "epoch": 0.017575,
+      "grad_norm": 2.856942653656006,
+      "grad_norm_var": 0.468259868516004,
+      "learning_rate": 0.0001,
+      "loss": 68.2141,
+      "loss/crossentropy": 2.897153854370117,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24900856614112854,
+      "loss/reg": 65.06795501708984,
+      "step": 703
+    },
+    {
+      "epoch": 0.0176,
+      "grad_norm": 2.5926945209503174,
+      "grad_norm_var": 0.469495224772921,
+      "learning_rate": 0.0001,
+      "loss": 67.9397,
+      "loss/crossentropy": 2.8245160579681396,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2201787531375885,
+      "loss/reg": 64.8949966430664,
+      "step": 704
+    },
+    {
+      "epoch": 0.017625,
+      "grad_norm": 2.5850887298583984,
+      "grad_norm_var": 0.48298250086362465,
+      "learning_rate": 0.0001,
+      "loss": 67.705,
+      "loss/crossentropy": 2.749037265777588,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2325897514820099,
+      "loss/reg": 64.72339630126953,
+      "step": 705
+    },
+    {
+      "epoch": 0.01765,
+      "grad_norm": 2.7953805923461914,
+      "grad_norm_var": 0.486705412463463,
+      "learning_rate": 0.0001,
+      "loss": 68.0719,
+      "loss/crossentropy": 3.281177282333374,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23989194631576538,
+      "loss/reg": 64.55084991455078,
+      "step": 706
+    },
+    {
+      "epoch": 0.017675,
+      "grad_norm": 2.8562774658203125,
+      "grad_norm_var": 0.4776801572940296,
+      "learning_rate": 0.0001,
+      "loss": 67.7388,
+      "loss/crossentropy": 3.113478899002075,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2437664419412613,
+      "loss/reg": 64.38153839111328,
+      "step": 707
+    },
+    {
+      "epoch": 0.0177,
+      "grad_norm": 2.6234002113342285,
+      "grad_norm_var": 0.4874532296205454,
+      "learning_rate": 0.0001,
+      "loss": 67.1132,
+      "loss/crossentropy": 2.668200731277466,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2298266887664795,
+      "loss/reg": 64.21515655517578,
+      "step": 708
+    },
+    {
+      "epoch": 0.017725,
+      "grad_norm": 3.07820725440979,
+      "grad_norm_var": 0.48490202715237274,
+      "learning_rate": 0.0001,
+      "loss": 67.0545,
+      "loss/crossentropy": 2.7720115184783936,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24164924025535583,
+      "loss/reg": 64.04081726074219,
+      "step": 709
+    },
+    {
+      "epoch": 0.01775,
+      "grad_norm": 3.3483076095581055,
+      "grad_norm_var": 0.4718668398271717,
+      "learning_rate": 0.0001,
+      "loss": 67.0299,
+      "loss/crossentropy": 2.9127249717712402,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24779383838176727,
+      "loss/reg": 63.869354248046875,
+      "step": 710
+    },
+    {
+      "epoch": 0.017775,
+      "grad_norm": 2.7884206771850586,
+      "grad_norm_var": 0.46624379181974973,
+      "learning_rate": 0.0001,
+      "loss": 66.733,
+      "loss/crossentropy": 2.7825210094451904,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2500211000442505,
+      "loss/reg": 63.70044708251953,
+      "step": 711
+    },
+    {
+      "epoch": 0.0178,
+      "grad_norm": 2.7660884857177734,
+      "grad_norm_var": 0.4661333259783419,
+      "learning_rate": 0.0001,
+      "loss": 66.8389,
+      "loss/crossentropy": 3.069040298461914,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24203762412071228,
+      "loss/reg": 63.52777862548828,
+      "step": 712
+    },
+    {
+      "epoch": 0.017825,
+      "grad_norm": 2.770434617996216,
+      "grad_norm_var": 0.42492635693953673,
+      "learning_rate": 0.0001,
+      "loss": 66.559,
+      "loss/crossentropy": 2.9704928398132324,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23337113857269287,
+      "loss/reg": 63.3551139831543,
+      "step": 713
+    },
+    {
+      "epoch": 0.01785,
+      "grad_norm": 2.4780144691467285,
+      "grad_norm_var": 0.08580528270122821,
+      "learning_rate": 0.0001,
+      "loss": 66.2605,
+      "loss/crossentropy": 2.8575258255004883,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2185363471508026,
+      "loss/reg": 63.184391021728516,
+      "step": 714
+    },
+    {
+      "epoch": 0.017875,
+      "grad_norm": 2.620495080947876,
+      "grad_norm_var": 0.043238218869502514,
+      "learning_rate": 0.0001,
+      "loss": 65.9609,
+      "loss/crossentropy": 2.7202653884887695,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22694343328475952,
+      "loss/reg": 63.01369094848633,
+      "step": 715
+    },
+    {
+      "epoch": 0.0179,
+      "grad_norm": 2.542977809906006,
+      "grad_norm_var": 0.04637604671543893,
+      "learning_rate": 0.0001,
+      "loss": 65.9499,
+      "loss/crossentropy": 2.8846218585968018,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21847984194755554,
+      "loss/reg": 62.84674835205078,
+      "step": 716
+    },
+    {
+      "epoch": 0.017925,
+      "grad_norm": 2.3846898078918457,
+      "grad_norm_var": 0.05508256728689957,
+      "learning_rate": 0.0001,
+      "loss": 65.4993,
+      "loss/crossentropy": 2.6143085956573486,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20310330390930176,
+      "loss/reg": 62.681915283203125,
+      "step": 717
+    },
+    {
+      "epoch": 0.01795,
+      "grad_norm": 2.73012375831604,
+      "grad_norm_var": 0.05506504227467772,
+      "learning_rate": 0.0001,
+      "loss": 65.5976,
+      "loss/crossentropy": 2.8506314754486084,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22879712283611298,
+      "loss/reg": 62.518165588378906,
+      "step": 718
+    },
+    {
+      "epoch": 0.017975,
+      "grad_norm": 2.4760758876800537,
+      "grad_norm_var": 0.05812137756337279,
+      "learning_rate": 0.0001,
+      "loss": 65.5302,
+      "loss/crossentropy": 2.9573020935058594,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2235342562198639,
+      "loss/reg": 62.34933090209961,
+      "step": 719
+    },
+    {
+      "epoch": 0.018,
+      "grad_norm": 2.9984333515167236,
+      "grad_norm_var": 0.06180506886074681,
+      "learning_rate": 0.0001,
+      "loss": 65.2478,
+      "loss/crossentropy": 2.825148344039917,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24277126789093018,
+      "loss/reg": 62.17986297607422,
+      "step": 720
+    },
+    {
+      "epoch": 0.018025,
+      "grad_norm": 2.449723243713379,
+      "grad_norm_var": 0.06574898435085288,
+      "learning_rate": 0.0001,
+      "loss": 64.8758,
+      "loss/crossentropy": 2.6417102813720703,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22177964448928833,
+      "loss/reg": 62.012306213378906,
+      "step": 721
+    },
+    {
+      "epoch": 0.01805,
+      "grad_norm": 2.6764326095581055,
+      "grad_norm_var": 0.06562316783263214,
+      "learning_rate": 0.0001,
+      "loss": 64.9029,
+      "loss/crossentropy": 2.828677177429199,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23033303022384644,
+      "loss/reg": 61.84387969970703,
+      "step": 722
+    },
+    {
+      "epoch": 0.018075,
+      "grad_norm": 2.8159947395324707,
+      "grad_norm_var": 0.06501549731222649,
+      "learning_rate": 0.0001,
+      "loss": 65.0769,
+      "loss/crossentropy": 3.174980878829956,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22877538204193115,
+      "loss/reg": 61.67316818237305,
+      "step": 723
+    },
+    {
+      "epoch": 0.0181,
+      "grad_norm": 3.2005398273468018,
+      "grad_norm_var": 0.07826629049727458,
+      "learning_rate": 0.0001,
+      "loss": 64.9549,
+      "loss/crossentropy": 3.203920364379883,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24600070714950562,
+      "loss/reg": 61.50501251220703,
+      "step": 724
+    },
+    {
+      "epoch": 0.018125,
+      "grad_norm": 2.9479753971099854,
+      "grad_norm_var": 0.07376285343432466,
+      "learning_rate": 0.0001,
+      "loss": 64.4933,
+      "loss/crossentropy": 2.9181010723114014,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23694497346878052,
+      "loss/reg": 61.338253021240234,
+      "step": 725
+    },
+    {
+      "epoch": 0.01815,
+      "grad_norm": 3.087327718734741,
+      "grad_norm_var": 0.05718879220165339,
+      "learning_rate": 0.0001,
+      "loss": 64.0707,
+      "loss/crossentropy": 2.6616029739379883,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2374861240386963,
+      "loss/reg": 61.17159652709961,
+      "step": 726
+    },
+    {
+      "epoch": 0.018175,
+      "grad_norm": 2.491910934448242,
+      "grad_norm_var": 0.06050683436232778,
+      "learning_rate": 0.0001,
+      "loss": 63.8745,
+      "loss/crossentropy": 2.6548359394073486,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2142171561717987,
+      "loss/reg": 61.00545120239258,
+      "step": 727
+    },
+    {
+      "epoch": 0.0182,
+      "grad_norm": 2.7890079021453857,
+      "grad_norm_var": 0.060696315605900726,
+      "learning_rate": 0.0001,
+      "loss": 63.9746,
+      "loss/crossentropy": 2.92938232421875,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20511876046657562,
+      "loss/reg": 60.84012985229492,
+      "step": 728
+    },
+    {
+      "epoch": 0.018225,
+      "grad_norm": 2.7196593284606934,
+      "grad_norm_var": 0.06049068327400467,
+      "learning_rate": 0.0001,
+      "loss": 63.5493,
+      "loss/crossentropy": 2.639850378036499,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23349660634994507,
+      "loss/reg": 60.675933837890625,
+      "step": 729
+    },
+    {
+      "epoch": 0.01825,
+      "grad_norm": 2.650775671005249,
+      "grad_norm_var": 0.05694124485670666,
+      "learning_rate": 0.0001,
+      "loss": 63.5801,
+      "loss/crossentropy": 2.8353641033172607,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23342850804328918,
+      "loss/reg": 60.51133346557617,
+      "step": 730
+    },
+    {
+      "epoch": 0.018275,
+      "grad_norm": 2.598869800567627,
+      "grad_norm_var": 0.05726858156747217,
+      "learning_rate": 0.0001,
+      "loss": 63.436,
+      "loss/crossentropy": 2.8569741249084473,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23304705321788788,
+      "loss/reg": 60.345970153808594,
+      "step": 731
+    },
+    {
+      "epoch": 0.0183,
+      "grad_norm": 2.6495063304901123,
+      "grad_norm_var": 0.055427494487368514,
+      "learning_rate": 0.0001,
+      "loss": 63.132,
+      "loss/crossentropy": 2.7324588298797607,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21595758199691772,
+      "loss/reg": 60.183570861816406,
+      "step": 732
+    },
+    {
+      "epoch": 0.018325,
+      "grad_norm": 2.603133201599121,
+      "grad_norm_var": 0.04837598895656067,
+      "learning_rate": 0.0001,
+      "loss": 62.695,
+      "loss/crossentropy": 2.4560225009918213,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21853625774383545,
+      "loss/reg": 60.02042770385742,
+      "step": 733
+    },
+    {
+      "epoch": 0.01835,
+      "grad_norm": 3.1586227416992188,
+      "grad_norm_var": 0.05912500309994281,
+      "learning_rate": 0.0001,
+      "loss": 63.1673,
+      "loss/crossentropy": 3.0193772315979004,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.28715232014656067,
+      "loss/reg": 59.86080551147461,
+      "step": 734
+    },
+    {
+      "epoch": 0.018375,
+      "grad_norm": 2.6718556880950928,
+      "grad_norm_var": 0.053857831483746094,
+      "learning_rate": 0.0001,
+      "loss": 62.6891,
+      "loss/crossentropy": 2.7639520168304443,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22497007250785828,
+      "loss/reg": 59.70021057128906,
+      "step": 735
+    },
+    {
+      "epoch": 0.0184,
+      "grad_norm": 2.7182719707489014,
+      "grad_norm_var": 0.05067343602402825,
+      "learning_rate": 0.0001,
+      "loss": 62.8571,
+      "loss/crossentropy": 3.092790365219116,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22454728186130524,
+      "loss/reg": 59.53980255126953,
+      "step": 736
+    },
+    {
+      "epoch": 0.018425,
+      "grad_norm": 2.63033127784729,
+      "grad_norm_var": 0.04513557987957005,
+      "learning_rate": 0.0001,
+      "loss": 62.4324,
+      "loss/crossentropy": 2.814718008041382,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23878329992294312,
+      "loss/reg": 59.37885284423828,
+      "step": 737
+    },
+    {
+      "epoch": 0.01845,
+      "grad_norm": 2.9615981578826904,
+      "grad_norm_var": 0.046446030177343306,
+      "learning_rate": 0.0001,
+      "loss": 62.5695,
+      "loss/crossentropy": 3.1102991104125977,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2397473156452179,
+      "loss/reg": 59.219425201416016,
+      "step": 738
+    },
+    {
+      "epoch": 0.018475,
+      "grad_norm": 2.7155027389526367,
+      "grad_norm_var": 0.04677527116378618,
+      "learning_rate": 0.0001,
+      "loss": 62.0607,
+      "loss/crossentropy": 2.7697606086730957,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23005826771259308,
+      "loss/reg": 59.06086349487305,
+      "step": 739
+    },
+    {
+      "epoch": 0.0185,
+      "grad_norm": 2.794189214706421,
+      "grad_norm_var": 0.03469948264410713,
+      "learning_rate": 0.0001,
+      "loss": 61.9033,
+      "loss/crossentropy": 2.7485909461975098,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24854815006256104,
+      "loss/reg": 58.9061279296875,
+      "step": 740
+    },
+    {
+      "epoch": 0.018525,
+      "grad_norm": 2.6141436100006104,
+      "grad_norm_var": 0.03337714746116092,
+      "learning_rate": 0.0001,
+      "loss": 61.7766,
+      "loss/crossentropy": 2.809553384780884,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21490439772605896,
+      "loss/reg": 58.752098083496094,
+      "step": 741
+    },
+    {
+      "epoch": 0.01855,
+      "grad_norm": 2.6113595962524414,
+      "grad_norm_var": 0.025552325556027947,
+      "learning_rate": 0.0001,
+      "loss": 61.7345,
+      "loss/crossentropy": 2.905996322631836,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22810769081115723,
+      "loss/reg": 58.60040283203125,
+      "step": 742
+    },
+    {
+      "epoch": 0.018575,
+      "grad_norm": 2.6710562705993652,
+      "grad_norm_var": 0.022320882287599632,
+      "learning_rate": 0.0001,
+      "loss": 61.7047,
+      "loss/crossentropy": 3.0256869792938232,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23153835535049438,
+      "loss/reg": 58.4475212097168,
+      "step": 743
+    },
+    {
+      "epoch": 0.0186,
+      "grad_norm": 2.6590538024902344,
+      "grad_norm_var": 0.022221697868613433,
+      "learning_rate": 0.0001,
+      "loss": 61.3096,
+      "loss/crossentropy": 2.7802655696868896,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23257814347743988,
+      "loss/reg": 58.29676818847656,
+      "step": 744
+    },
+    {
+      "epoch": 0.018625,
+      "grad_norm": 2.6750831604003906,
+      "grad_norm_var": 0.022313711031313233,
+      "learning_rate": 0.0001,
+      "loss": 60.9886,
+      "loss/crossentropy": 2.6292662620544434,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21325910091400146,
+      "loss/reg": 58.146026611328125,
+      "step": 745
+    },
+    {
+      "epoch": 0.01865,
+      "grad_norm": 2.910602569580078,
+      "grad_norm_var": 0.02443077895978618,
+      "learning_rate": 0.0001,
+      "loss": 61.2949,
+      "loss/crossentropy": 3.0462257862091064,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25037214159965515,
+      "loss/reg": 57.99826431274414,
+      "step": 746
+    },
+    {
+      "epoch": 0.018675,
+      "grad_norm": 2.7559032440185547,
+      "grad_norm_var": 0.023274603878721933,
+      "learning_rate": 0.0001,
+      "loss": 61.0211,
+      "loss/crossentropy": 2.9264190196990967,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24024531245231628,
+      "loss/reg": 57.85441970825195,
+      "step": 747
+    },
+    {
+      "epoch": 0.0187,
+      "grad_norm": 11.09821891784668,
+      "grad_norm_var": 4.385431661130277,
+      "learning_rate": 0.0001,
+      "loss": 60.7765,
+      "loss/crossentropy": 2.7616488933563232,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.3043861985206604,
+      "loss/reg": 57.71049118041992,
+      "step": 748
+    },
+    {
+      "epoch": 0.018725,
+      "grad_norm": 3.4629063606262207,
+      "grad_norm_var": 4.355694283900243,
+      "learning_rate": 0.0001,
+      "loss": 60.7808,
+      "loss/crossentropy": 2.9968600273132324,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22214075922966003,
+      "loss/reg": 57.56179428100586,
+      "step": 749
+    },
+    {
+      "epoch": 0.01875,
+      "grad_norm": 2.9865288734436035,
+      "grad_norm_var": 4.361232034357083,
+      "learning_rate": 0.0001,
+      "loss": 60.3824,
+      "loss/crossentropy": 2.712460994720459,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25824302434921265,
+      "loss/reg": 57.4117431640625,
+      "step": 750
+    },
+    {
+      "epoch": 0.018775,
+      "grad_norm": 2.966188430786133,
+      "grad_norm_var": 4.341660332338729,
+      "learning_rate": 0.0001,
+      "loss": 60.2275,
+      "loss/crossentropy": 2.749971389770508,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2149219810962677,
+      "loss/reg": 57.26259231567383,
+      "step": 751
+    },
+    {
+      "epoch": 0.0188,
+      "grad_norm": 3.0538909435272217,
+      "grad_norm_var": 4.32146321783135,
+      "learning_rate": 0.0001,
+      "loss": 60.1386,
+      "loss/crossentropy": 2.7705466747283936,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25299209356307983,
+      "loss/reg": 57.11503219604492,
+      "step": 752
+    },
+    {
+      "epoch": 0.018825,
+      "grad_norm": 3.7716422080993652,
+      "grad_norm_var": 4.293677767872144,
+      "learning_rate": 0.0001,
+      "loss": 60.374,
+      "loss/crossentropy": 3.1197762489318848,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2896609902381897,
+      "loss/reg": 56.964542388916016,
+      "step": 753
+    },
+    {
+      "epoch": 0.01885,
+      "grad_norm": 2.8715505599975586,
+      "grad_norm_var": 4.29967918105209,
+      "learning_rate": 0.0001,
+      "loss": 59.6617,
+      "loss/crossentropy": 2.6158196926116943,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23370970785617828,
+      "loss/reg": 56.81219482421875,
+      "step": 754
+    },
+    {
+      "epoch": 0.018875,
+      "grad_norm": 2.87619686126709,
+      "grad_norm_var": 4.286335448113463,
+      "learning_rate": 0.0001,
+      "loss": 59.5822,
+      "loss/crossentropy": 2.6575210094451904,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.26069819927215576,
+      "loss/reg": 56.66400146484375,
+      "step": 755
+    },
+    {
+      "epoch": 0.0189,
+      "grad_norm": 2.656797409057617,
+      "grad_norm_var": 4.299046394966134,
+      "learning_rate": 0.0001,
+      "loss": 59.3342,
+      "loss/crossentropy": 2.594022035598755,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22433573007583618,
+      "loss/reg": 56.515830993652344,
+      "step": 756
+    },
+    {
+      "epoch": 0.018925,
+      "grad_norm": 2.692230224609375,
+      "grad_norm_var": 4.291088604047604,
+      "learning_rate": 0.0001,
+      "loss": 59.2204,
+      "loss/crossentropy": 2.6446311473846436,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2099241316318512,
+      "loss/reg": 56.36582565307617,
+      "step": 757
+    },
+    {
+      "epoch": 0.01895,
+      "grad_norm": 2.687993288040161,
+      "grad_norm_var": 4.283193607489185,
+      "learning_rate": 0.0001,
+      "loss": 59.2018,
+      "loss/crossentropy": 2.7599661350250244,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22358274459838867,
+      "loss/reg": 56.21822738647461,
+      "step": 758
+    },
+    {
+      "epoch": 0.018975,
+      "grad_norm": 2.611598253250122,
+      "grad_norm_var": 4.289389567894268,
+      "learning_rate": 0.0001,
+      "loss": 59.0686,
+      "loss/crossentropy": 2.787243604660034,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21356835961341858,
+      "loss/reg": 56.06779098510742,
+      "step": 759
+    },
+    {
+      "epoch": 0.019,
+      "grad_norm": 2.4653584957122803,
+      "grad_norm_var": 4.31141311016122,
+      "learning_rate": 0.0001,
+      "loss": 59.0029,
+      "loss/crossentropy": 2.8415911197662354,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24345867335796356,
+      "loss/reg": 55.9178466796875,
+      "step": 760
+    },
+    {
+      "epoch": 0.019025,
+      "grad_norm": 2.8678863048553467,
+      "grad_norm_var": 4.2948716677726795,
+      "learning_rate": 0.0001,
+      "loss": 58.6391,
+      "loss/crossentropy": 2.6471433639526367,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21930553019046783,
+      "loss/reg": 55.772621154785156,
+      "step": 761
+    },
+    {
+      "epoch": 0.01905,
+      "grad_norm": 2.790867805480957,
+      "grad_norm_var": 4.303915496486924,
+      "learning_rate": 0.0001,
+      "loss": 58.5621,
+      "loss/crossentropy": 2.691446542739868,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2425747513771057,
+      "loss/reg": 55.628082275390625,
+      "step": 762
+    },
+    {
+      "epoch": 0.019075,
+      "grad_norm": 2.7855591773986816,
+      "grad_norm_var": 4.301370303985943,
+      "learning_rate": 0.0001,
+      "loss": 58.4557,
+      "loss/crossentropy": 2.732412815093994,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23795194923877716,
+      "loss/reg": 55.48536682128906,
+      "step": 763
+    },
+    {
+      "epoch": 0.0191,
+      "grad_norm": 2.5391337871551514,
+      "grad_norm_var": 0.11220097224740945,
+      "learning_rate": 0.0001,
+      "loss": 58.1866,
+      "loss/crossentropy": 2.628573179244995,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21366839110851288,
+      "loss/reg": 55.344329833984375,
+      "step": 764
+    },
+    {
+      "epoch": 0.019125,
+      "grad_norm": 2.4486775398254395,
+      "grad_norm_var": 0.09771899643028424,
+      "learning_rate": 0.0001,
+      "loss": 58.3236,
+      "loss/crossentropy": 2.8930563926696777,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22997558116912842,
+      "loss/reg": 55.20055389404297,
+      "step": 765
+    },
+    {
+      "epoch": 0.01915,
+      "grad_norm": 2.766298532485962,
+      "grad_norm_var": 0.09577246439944981,
+      "learning_rate": 0.0001,
+      "loss": 58.3324,
+      "loss/crossentropy": 3.025320291519165,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24967965483665466,
+      "loss/reg": 55.057376861572266,
+      "step": 766
+    },
+    {
+      "epoch": 0.019175,
+      "grad_norm": 3.201991558074951,
+      "grad_norm_var": 0.1043707670856025,
+      "learning_rate": 0.0001,
+      "loss": 58.3737,
+      "loss/crossentropy": 3.2072536945343018,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24961410462856293,
+      "loss/reg": 54.916839599609375,
+      "step": 767
+    },
+    {
+      "epoch": 0.0192,
+      "grad_norm": 2.3997886180877686,
+      "grad_norm_var": 0.11053669150879818,
+      "learning_rate": 0.0001,
+      "loss": 57.8128,
+      "loss/crossentropy": 2.814892053604126,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.221253901720047,
+      "loss/reg": 54.776668548583984,
+      "step": 768
+    },
+    {
+      "epoch": 0.019225,
+      "grad_norm": 2.467320680618286,
+      "grad_norm_var": 0.04390441270999475,
+      "learning_rate": 0.0001,
+      "loss": 57.848,
+      "loss/crossentropy": 2.9831323623657227,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22602063417434692,
+      "loss/reg": 54.63887405395508,
+      "step": 769
+    },
+    {
+      "epoch": 0.01925,
+      "grad_norm": 2.8920133113861084,
+      "grad_norm_var": 0.04441070048479005,
+      "learning_rate": 0.0001,
+      "loss": 57.6435,
+      "loss/crossentropy": 2.9186885356903076,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2257421761751175,
+      "loss/reg": 54.49909210205078,
+      "step": 770
+    },
+    {
+      "epoch": 0.019275,
+      "grad_norm": 2.7764594554901123,
+      "grad_norm_var": 0.04264750323779935,
+      "learning_rate": 0.0001,
+      "loss": 56.9818,
+      "loss/crossentropy": 2.4131717681884766,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20915502309799194,
+      "loss/reg": 54.35948181152344,
+      "step": 771
+    },
+    {
+      "epoch": 0.0193,
+      "grad_norm": 2.9066452980041504,
+      "grad_norm_var": 0.04542215413656076,
+      "learning_rate": 0.0001,
+      "loss": 57.4967,
+      "loss/crossentropy": 3.0247690677642822,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2509494423866272,
+      "loss/reg": 54.22095489501953,
+      "step": 772
+    },
+    {
+      "epoch": 0.019325,
+      "grad_norm": 2.6702942848205566,
+      "grad_norm_var": 0.04549320067607141,
+      "learning_rate": 0.0001,
+      "loss": 57.237,
+      "loss/crossentropy": 2.915168523788452,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23676593601703644,
+      "loss/reg": 54.08510208129883,
+      "step": 773
+    },
+    {
+      "epoch": 0.01935,
+      "grad_norm": 2.5919129848480225,
+      "grad_norm_var": 0.046286340421070805,
+      "learning_rate": 0.0001,
+      "loss": 56.9803,
+      "loss/crossentropy": 2.8107874393463135,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2236163318157196,
+      "loss/reg": 53.94586181640625,
+      "step": 774
+    },
+    {
+      "epoch": 0.019375,
+      "grad_norm": 2.6954429149627686,
+      "grad_norm_var": 0.045750154457365015,
+      "learning_rate": 0.0001,
+      "loss": 57.0791,
+      "loss/crossentropy": 3.017504930496216,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2500165104866028,
+      "loss/reg": 53.81157684326172,
+      "step": 775
+    },
+    {
+      "epoch": 0.0194,
+      "grad_norm": 2.492471218109131,
+      "grad_norm_var": 0.044933029105976394,
+      "learning_rate": 0.0001,
+      "loss": 56.6907,
+      "loss/crossentropy": 2.7895073890686035,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2253413051366806,
+      "loss/reg": 53.67582702636719,
+      "step": 776
+    },
+    {
+      "epoch": 0.019425,
+      "grad_norm": 2.369546413421631,
+      "grad_norm_var": 0.04968441666320113,
+      "learning_rate": 0.0001,
+      "loss": 56.6367,
+      "loss/crossentropy": 2.9010658264160156,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.1958373337984085,
+      "loss/reg": 53.53976058959961,
+      "step": 777
+    },
+    {
+      "epoch": 0.01945,
+      "grad_norm": 3.021898031234741,
+      "grad_norm_var": 0.05660028336942536,
+      "learning_rate": 0.0001,
+      "loss": 56.5493,
+      "loss/crossentropy": 2.893462896347046,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2516905665397644,
+      "loss/reg": 53.404178619384766,
+      "step": 778
+    },
+    {
+      "epoch": 0.019475,
+      "grad_norm": 2.691014528274536,
+      "grad_norm_var": 0.055942876476975316,
+      "learning_rate": 0.0001,
+      "loss": 55.9807,
+      "loss/crossentropy": 2.465963840484619,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24134577810764313,
+      "loss/reg": 53.27339553833008,
+      "step": 779
+    },
+    {
+      "epoch": 0.0195,
+      "grad_norm": 2.3977248668670654,
+      "grad_norm_var": 0.05990861359730009,
+      "learning_rate": 0.0001,
+      "loss": 56.0824,
+      "loss/crossentropy": 2.7291431427001953,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21200142800807953,
+      "loss/reg": 53.141212463378906,
+      "step": 780
+    },
+    {
+      "epoch": 0.019525,
+      "grad_norm": 2.8265068531036377,
+      "grad_norm_var": 0.057462358496513606,
+      "learning_rate": 0.0001,
+      "loss": 56.3862,
+      "loss/crossentropy": 3.13495135307312,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24010971188545227,
+      "loss/reg": 53.011112213134766,
+      "step": 781
+    },
+    {
+      "epoch": 0.01955,
+      "grad_norm": 2.549685001373291,
+      "grad_norm_var": 0.05842115228572311,
+      "learning_rate": 0.0001,
+      "loss": 55.8148,
+      "loss/crossentropy": 2.680000066757202,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25199833512306213,
+      "loss/reg": 52.88276672363281,
+      "step": 782
+    },
+    {
+      "epoch": 0.019575,
+      "grad_norm": 2.601576328277588,
+      "grad_norm_var": 0.03951790591236583,
+      "learning_rate": 0.0001,
+      "loss": 55.6334,
+      "loss/crossentropy": 2.6754848957061768,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2111392617225647,
+      "loss/reg": 52.746726989746094,
+      "step": 783
+    },
+    {
+      "epoch": 0.0196,
+      "grad_norm": 8.212234497070312,
+      "grad_norm_var": 1.9595461171198734,
+      "learning_rate": 0.0001,
+      "loss": 55.6696,
+      "loss/crossentropy": 2.8070499897003174,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25141945481300354,
+      "loss/reg": 52.61111068725586,
+      "step": 784
+    },
+    {
+      "epoch": 0.019625,
+      "grad_norm": 2.7455339431762695,
+      "grad_norm_var": 1.9442466683120376,
+      "learning_rate": 0.0001,
+      "loss": 55.6931,
+      "loss/crossentropy": 2.996835947036743,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21819031238555908,
+      "loss/reg": 52.47810363769531,
+      "step": 785
+    },
+    {
+      "epoch": 0.01965,
+      "grad_norm": 2.630183696746826,
+      "grad_norm_var": 1.9532633581534922,
+      "learning_rate": 0.0001,
+      "loss": 55.4071,
+      "loss/crossentropy": 2.804309844970703,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25376999378204346,
+      "loss/reg": 52.3490104675293,
+      "step": 786
+    },
+    {
+      "epoch": 0.019675,
+      "grad_norm": 3.049198865890503,
+      "grad_norm_var": 1.9493762909636376,
+      "learning_rate": 0.0001,
+      "loss": 55.0084,
+      "loss/crossentropy": 2.542832374572754,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24326905608177185,
+      "loss/reg": 52.222328186035156,
+      "step": 787
+    },
+    {
+      "epoch": 0.0197,
+      "grad_norm": 4.536539554595947,
+      "grad_norm_var": 2.0889857251666357,
+      "learning_rate": 0.0001,
+      "loss": 55.1162,
+      "loss/crossentropy": 2.8076493740081787,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21975521743297577,
+      "loss/reg": 52.08875274658203,
+      "step": 788
+    },
+    {
+      "epoch": 0.019725,
+      "grad_norm": 4.922837734222412,
+      "grad_norm_var": 2.2680069995824983,
+      "learning_rate": 0.0001,
+      "loss": 55.0792,
+      "loss/crossentropy": 2.878237724304199,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2420842945575714,
+      "loss/reg": 51.95884323120117,
+      "step": 789
+    },
+    {
+      "epoch": 0.01975,
+      "grad_norm": 2.30218243598938,
+      "grad_norm_var": 2.2994830337610286,
+      "learning_rate": 0.0001,
+      "loss": 54.8028,
+      "loss/crossentropy": 2.758139133453369,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2122175395488739,
+      "loss/reg": 51.83245086669922,
+      "step": 790
+    },
+    {
+      "epoch": 0.019775,
+      "grad_norm": 2.6652355194091797,
+      "grad_norm_var": 2.3017848488088544,
+      "learning_rate": 0.0001,
+      "loss": 54.6171,
+      "loss/crossentropy": 2.6950416564941406,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2227291464805603,
+      "loss/reg": 51.699317932128906,
+      "step": 791
+    },
+    {
+      "epoch": 0.0198,
+      "grad_norm": 2.5915586948394775,
+      "grad_norm_var": 2.29237841360286,
+      "learning_rate": 0.0001,
+      "loss": 54.58,
+      "loss/crossentropy": 2.798182487487793,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21353884041309357,
+      "loss/reg": 51.568267822265625,
+      "step": 792
+    },
+    {
+      "epoch": 0.019825,
+      "grad_norm": 2.6132395267486572,
+      "grad_norm_var": 2.2672515903509027,
+      "learning_rate": 0.0001,
+      "loss": 54.4863,
+      "loss/crossentropy": 2.8284451961517334,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22251488268375397,
+      "loss/reg": 51.43537902832031,
+      "step": 793
+    },
+    {
+      "epoch": 0.01985,
+      "grad_norm": 2.9160501956939697,
+      "grad_norm_var": 2.2714860685983003,
+      "learning_rate": 0.0001,
+      "loss": 54.4318,
+      "loss/crossentropy": 2.8874566555023193,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24243222177028656,
+      "loss/reg": 51.301944732666016,
+      "step": 794
+    },
+    {
+      "epoch": 0.019875,
+      "grad_norm": 2.767122983932495,
+      "grad_norm_var": 2.266016244984087,
+      "learning_rate": 0.0001,
+      "loss": 53.9738,
+      "loss/crossentropy": 2.5667829513549805,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2378443330526352,
+      "loss/reg": 51.16920852661133,
+      "step": 795
+    },
+    {
+      "epoch": 0.0199,
+      "grad_norm": 2.9074504375457764,
+      "grad_norm_var": 2.222940734299857,
+      "learning_rate": 0.0001,
+      "loss": 53.9377,
+      "loss/crossentropy": 2.659839391708374,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23761096596717834,
+      "loss/reg": 51.040260314941406,
+      "step": 796
+    },
+    {
+      "epoch": 0.019925,
+      "grad_norm": 2.6332762241363525,
+      "grad_norm_var": 2.237533280064099,
+      "learning_rate": 0.0001,
+      "loss": 53.8857,
+      "loss/crossentropy": 2.7650234699249268,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21053263545036316,
+      "loss/reg": 50.91017532348633,
+      "step": 797
+    },
+    {
+      "epoch": 0.01995,
+      "grad_norm": 2.6881556510925293,
+      "grad_norm_var": 2.225058902631474,
+      "learning_rate": 0.0001,
+      "loss": 53.6563,
+      "loss/crossentropy": 2.6306800842285156,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24274012446403503,
+      "loss/reg": 50.78291320800781,
+      "step": 798
+    },
+    {
+      "epoch": 0.019975,
+      "grad_norm": 2.7099571228027344,
+      "grad_norm_var": 2.215716208024157,
+      "learning_rate": 0.0001,
+      "loss": 53.4381,
+      "loss/crossentropy": 2.52323842048645,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.259515643119812,
+      "loss/reg": 50.65534591674805,
+      "step": 799
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.960604429244995,
+      "grad_norm_var": 0.5037824266901583,
+      "learning_rate": 0.0001,
+      "loss": 53.5623,
+      "loss/crossentropy": 2.7882955074310303,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24377551674842834,
+      "loss/reg": 50.53020095825195,
+      "step": 800
+    },
+    {
+      "epoch": 0.020025,
+      "grad_norm": 2.7040579319000244,
+      "grad_norm_var": 0.505172444748995,
+      "learning_rate": 0.0001,
+      "loss": 53.4338,
+      "loss/crossentropy": 2.8009700775146484,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2242647111415863,
+      "loss/reg": 50.408573150634766,
+      "step": 801
+    },
+    {
+      "epoch": 0.02005,
+      "grad_norm": 2.7675962448120117,
+      "grad_norm_var": 0.5000376610377392,
+      "learning_rate": 0.0001,
+      "loss": 53.1763,
+      "loss/crossentropy": 2.6563568115234375,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23715972900390625,
+      "loss/reg": 50.28279113769531,
+      "step": 802
+    },
+    {
+      "epoch": 0.020075,
+      "grad_norm": 2.9913876056671143,
+      "grad_norm_var": 0.49973967585988155,
+      "learning_rate": 0.0001,
+      "loss": 53.2435,
+      "loss/crossentropy": 2.856863498687744,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22863171994686127,
+      "loss/reg": 50.158050537109375,
+      "step": 803
+    },
+    {
+      "epoch": 0.0201,
+      "grad_norm": 2.579055070877075,
+      "grad_norm_var": 0.3329253447165852,
+      "learning_rate": 0.0001,
+      "loss": 53.0455,
+      "loss/crossentropy": 2.7912509441375732,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21997642517089844,
+      "loss/reg": 50.0342903137207,
+      "step": 804
+    },
+    {
+      "epoch": 0.020125,
+      "grad_norm": 2.906846046447754,
+      "grad_norm_var": 0.03177485529399533,
+      "learning_rate": 0.0001,
+      "loss": 52.9991,
+      "loss/crossentropy": 2.8479995727539062,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24032440781593323,
+      "loss/reg": 49.91073989868164,
+      "step": 805
+    },
+    {
+      "epoch": 0.02015,
+      "grad_norm": 3.0859761238098145,
+      "grad_norm_var": 0.025305915418744006,
+      "learning_rate": 0.0001,
+      "loss": 52.9709,
+      "loss/crossentropy": 2.920821189880371,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.26482486724853516,
+      "loss/reg": 49.78524398803711,
+      "step": 806
+    },
+    {
+      "epoch": 0.020175,
+      "grad_norm": 2.7434208393096924,
+      "grad_norm_var": 0.02448665601768667,
+      "learning_rate": 0.0001,
+      "loss": 52.9071,
+      "loss/crossentropy": 3.0034008026123047,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24188174307346344,
+      "loss/reg": 49.66181945800781,
+      "step": 807
+    },
+    {
+      "epoch": 0.0202,
+      "grad_norm": 2.88177227973938,
+      "grad_norm_var": 0.02225149356486682,
+      "learning_rate": 0.0001,
+      "loss": 52.4653,
+      "loss/crossentropy": 2.6840085983276367,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.245322123169899,
+      "loss/reg": 49.53594207763672,
+      "step": 808
+    },
+    {
+      "epoch": 0.020225,
+      "grad_norm": 3.1109049320220947,
+      "grad_norm_var": 0.02510624438714686,
+      "learning_rate": 0.0001,
+      "loss": 52.547,
+      "loss/crossentropy": 2.8825902938842773,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2539767920970917,
+      "loss/reg": 49.41044616699219,
+      "step": 809
+    },
+    {
+      "epoch": 0.02025,
+      "grad_norm": 2.4495651721954346,
+      "grad_norm_var": 0.033640854815441185,
+      "learning_rate": 0.0001,
+      "loss": 51.7535,
+      "loss/crossentropy": 2.28265380859375,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.18538084626197815,
+      "loss/reg": 49.28546905517578,
+      "step": 810
+    },
+    {
+      "epoch": 0.020275,
+      "grad_norm": 2.864192247390747,
+      "grad_norm_var": 0.03373374858250576,
+      "learning_rate": 0.0001,
+      "loss": 52.0966,
+      "loss/crossentropy": 2.682021379470825,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2524881362915039,
+      "loss/reg": 49.162086486816406,
+      "step": 811
+    },
+    {
+      "epoch": 0.0203,
+      "grad_norm": 2.7251250743865967,
+      "grad_norm_var": 0.03347917919778235,
+      "learning_rate": 0.0001,
+      "loss": 52.1571,
+      "loss/crossentropy": 2.893673896789551,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2223658412694931,
+      "loss/reg": 49.04104232788086,
+      "step": 812
+    },
+    {
+      "epoch": 0.020325,
+      "grad_norm": 2.496249198913574,
+      "grad_norm_var": 0.037700954552336914,
+      "learning_rate": 0.0001,
+      "loss": 51.914,
+      "loss/crossentropy": 2.755688190460205,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23637332022190094,
+      "loss/reg": 48.92195510864258,
+      "step": 813
+    },
+    {
+      "epoch": 0.02035,
+      "grad_norm": 2.5698087215423584,
+      "grad_norm_var": 0.04020791484434175,
+      "learning_rate": 0.0001,
+      "loss": 51.7827,
+      "loss/crossentropy": 2.7431833744049072,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23644611239433289,
+      "loss/reg": 48.80311584472656,
+      "step": 814
+    },
+    {
+      "epoch": 0.020375,
+      "grad_norm": 2.952713966369629,
+      "grad_norm_var": 0.04148941052159193,
+      "learning_rate": 0.0001,
+      "loss": 51.7364,
+      "loss/crossentropy": 2.824826717376709,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23083820939064026,
+      "loss/reg": 48.68075942993164,
+      "step": 815
+    },
+    {
+      "epoch": 0.0204,
+      "grad_norm": 2.635850191116333,
+      "grad_norm_var": 0.0410977076632508,
+      "learning_rate": 0.0001,
+      "loss": 51.7439,
+      "loss/crossentropy": 2.93487548828125,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24603211879730225,
+      "loss/reg": 48.56298828125,
+      "step": 816
+    },
+    {
+      "epoch": 0.020425,
+      "grad_norm": 2.40620493888855,
+      "grad_norm_var": 0.049620007024852225,
+      "learning_rate": 0.0001,
+      "loss": 51.304,
+      "loss/crossentropy": 2.656850814819336,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2041395902633667,
+      "loss/reg": 48.44300079345703,
+      "step": 817
+    },
+    {
+      "epoch": 0.02045,
+      "grad_norm": 2.646141290664673,
+      "grad_norm_var": 0.05042569960910441,
+      "learning_rate": 0.0001,
+      "loss": 51.235,
+      "loss/crossentropy": 2.664440155029297,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2458214908838272,
+      "loss/reg": 48.3246955871582,
+      "step": 818
+    },
+    {
+      "epoch": 0.020475,
+      "grad_norm": 2.8259634971618652,
+      "grad_norm_var": 0.04687417195843082,
+      "learning_rate": 0.0001,
+      "loss": 51.6837,
+      "loss/crossentropy": 3.2239608764648438,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2503930926322937,
+      "loss/reg": 48.2093505859375,
+      "step": 819
+    },
+    {
+      "epoch": 0.0205,
+      "grad_norm": 2.993224620819092,
+      "grad_norm_var": 0.04857006914218402,
+      "learning_rate": 0.0001,
+      "loss": 51.27,
+      "loss/crossentropy": 2.8996071815490723,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.27634817361831665,
+      "loss/reg": 48.094078063964844,
+      "step": 820
+    },
+    {
+      "epoch": 0.020525,
+      "grad_norm": 3.131995916366577,
+      "grad_norm_var": 0.05589532321223274,
+      "learning_rate": 0.0001,
+      "loss": 51.3641,
+      "loss/crossentropy": 3.1158416271209717,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.272818386554718,
+      "loss/reg": 47.975460052490234,
+      "step": 821
+    },
+    {
+      "epoch": 0.02055,
+      "grad_norm": 2.6322062015533447,
+      "grad_norm_var": 0.05040003879316449,
+      "learning_rate": 0.0001,
+      "loss": 50.9655,
+      "loss/crossentropy": 2.874999761581421,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2292400300502777,
+      "loss/reg": 47.86125564575195,
+      "step": 822
+    },
+    {
+      "epoch": 0.020575,
+      "grad_norm": 2.631922960281372,
+      "grad_norm_var": 0.05133554293392611,
+      "learning_rate": 0.0001,
+      "loss": 50.457,
+      "loss/crossentropy": 2.5026562213897705,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20776653289794922,
+      "loss/reg": 47.74656677246094,
+      "step": 823
+    },
+    {
+      "epoch": 0.0206,
+      "grad_norm": 2.881382942199707,
+      "grad_norm_var": 0.051328562127300144,
+      "learning_rate": 0.0001,
+      "loss": 50.542,
+      "loss/crossentropy": 2.688274621963501,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22304531931877136,
+      "loss/reg": 47.63066482543945,
+      "step": 824
+    },
+    {
+      "epoch": 0.020625,
+      "grad_norm": 2.683779239654541,
+      "grad_norm_var": 0.04201158273686285,
+      "learning_rate": 0.0001,
+      "loss": 50.5567,
+      "loss/crossentropy": 2.7942123413085938,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2437441498041153,
+      "loss/reg": 47.518707275390625,
+      "step": 825
+    },
+    {
+      "epoch": 0.02065,
+      "grad_norm": 2.7014639377593994,
+      "grad_norm_var": 0.03688115494732453,
+      "learning_rate": 0.0001,
+      "loss": 50.5527,
+      "loss/crossentropy": 2.8857181072235107,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.26020658016204834,
+      "loss/reg": 47.40679168701172,
+      "step": 826
+    },
+    {
+      "epoch": 0.020675,
+      "grad_norm": 13.778948783874512,
+      "grad_norm_var": 7.668981462187241,
+      "learning_rate": 0.0001,
+      "loss": 50.3418,
+      "loss/crossentropy": 2.836272716522217,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.213333398103714,
+      "loss/reg": 47.29219436645508,
+      "step": 827
+    },
+    {
+      "epoch": 0.0207,
+      "grad_norm": 2.6730599403381348,
+      "grad_norm_var": 7.673962997287033,
+      "learning_rate": 0.0001,
+      "loss": 50.3743,
+      "loss/crossentropy": 2.9492764472961426,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24545426666736603,
+      "loss/reg": 47.17961120605469,
+      "step": 828
+    },
+    {
+      "epoch": 0.020725,
+      "grad_norm": 2.44629168510437,
+      "grad_norm_var": 7.680239164125849,
+      "learning_rate": 0.0001,
+      "loss": 49.7865,
+      "loss/crossentropy": 2.5040016174316406,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21603502333164215,
+      "loss/reg": 47.06645965576172,
+      "step": 829
+    },
+    {
+      "epoch": 0.02075,
+      "grad_norm": 2.927917003631592,
+      "grad_norm_var": 7.6480446113119305,
+      "learning_rate": 0.0001,
+      "loss": 50.4814,
+      "loss/crossentropy": 3.270429849624634,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25596797466278076,
+      "loss/reg": 46.9549674987793,
+      "step": 830
+    },
+    {
+      "epoch": 0.020775,
+      "grad_norm": 3.077221155166626,
+      "grad_norm_var": 7.641018421905691,
+      "learning_rate": 0.0001,
+      "loss": 50.196,
+      "loss/crossentropy": 3.097292423248291,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2553516924381256,
+      "loss/reg": 46.843326568603516,
+      "step": 831
+    },
+    {
+      "epoch": 0.0208,
+      "grad_norm": 3.5145583152770996,
+      "grad_norm_var": 7.5948155070728856,
+      "learning_rate": 0.0001,
+      "loss": 49.5672,
+      "loss/crossentropy": 2.601334571838379,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2364489734172821,
+      "loss/reg": 46.729393005371094,
+      "step": 832
+    },
+    {
+      "epoch": 0.020825,
+      "grad_norm": 3.11055326461792,
+      "grad_norm_var": 7.523380552917458,
+      "learning_rate": 0.0001,
+      "loss": 49.711,
+      "loss/crossentropy": 2.8619778156280518,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2326866090297699,
+      "loss/reg": 46.61635971069336,
+      "step": 833
+    },
+    {
+      "epoch": 0.02085,
+      "grad_norm": 3.700695276260376,
+      "grad_norm_var": 7.467056690865655,
+      "learning_rate": 0.0001,
+      "loss": 49.5927,
+      "loss/crossentropy": 2.8446598052978516,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2441275715827942,
+      "loss/reg": 46.50389099121094,
+      "step": 834
+    },
+    {
+      "epoch": 0.020875,
+      "grad_norm": 2.789363145828247,
+      "grad_norm_var": 7.470951661037448,
+      "learning_rate": 0.0001,
+      "loss": 49.4515,
+      "loss/crossentropy": 2.822880268096924,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23577329516410828,
+      "loss/reg": 46.39282989501953,
+      "step": 835
+    },
+    {
+      "epoch": 0.0209,
+      "grad_norm": 2.612974166870117,
+      "grad_norm_var": 7.510988449534897,
+      "learning_rate": 0.0001,
+      "loss": 49.3286,
+      "loss/crossentropy": 2.8413946628570557,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.20617251098155975,
+      "loss/reg": 46.281070709228516,
+      "step": 836
+    },
+    {
+      "epoch": 0.020925,
+      "grad_norm": 2.69990611076355,
+      "grad_norm_var": 7.548519312088183,
+      "learning_rate": 0.0001,
+      "loss": 49.2338,
+      "loss/crossentropy": 2.829338550567627,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23464342951774597,
+      "loss/reg": 46.16982650756836,
+      "step": 837
+    },
+    {
+      "epoch": 0.02095,
+      "grad_norm": 3.1259517669677734,
+      "grad_norm_var": 7.503078866049228,
+      "learning_rate": 0.0001,
+      "loss": 48.9702,
+      "loss/crossentropy": 2.642885208129883,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2709121108055115,
+      "loss/reg": 46.05638885498047,
+      "step": 838
+    },
+    {
+      "epoch": 0.020975,
+      "grad_norm": 2.6453773975372314,
+      "grad_norm_var": 7.50138088085608,
+      "learning_rate": 0.0001,
+      "loss": 49.101,
+      "loss/crossentropy": 2.90779972076416,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2505376935005188,
+      "loss/reg": 45.94264602661133,
+      "step": 839
+    },
+    {
+      "epoch": 0.021,
+      "grad_norm": 2.7109532356262207,
+      "grad_norm_var": 7.519198653159886,
+      "learning_rate": 0.0001,
+      "loss": 48.8333,
+      "loss/crossentropy": 2.7746942043304443,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23057350516319275,
+      "loss/reg": 45.8280029296875,
+      "step": 840
+    },
+    {
+      "epoch": 0.021025,
+      "grad_norm": 2.9908559322357178,
+      "grad_norm_var": 7.488604931370973,
+      "learning_rate": 0.0001,
+      "loss": 48.7675,
+      "loss/crossentropy": 2.7737483978271484,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.27381837368011475,
+      "loss/reg": 45.71989822387695,
+      "step": 841
+    },
+    {
+      "epoch": 0.02105,
+      "grad_norm": 2.8714518547058105,
+      "grad_norm_var": 7.47017858293949,
+      "learning_rate": 0.0001,
+      "loss": 48.9529,
+      "loss/crossentropy": 3.073143243789673,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.26654019951820374,
+      "loss/reg": 45.61322021484375,
+      "step": 842
+    },
+    {
+      "epoch": 0.021075,
+      "grad_norm": 3.0701894760131836,
+      "grad_norm_var": 0.11045821533903463,
+      "learning_rate": 0.0001,
+      "loss": 48.3711,
+      "loss/crossentropy": 2.606783390045166,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.26009494066238403,
+      "loss/reg": 45.504241943359375,
+      "step": 843
+    },
+    {
+      "epoch": 0.0211,
+      "grad_norm": 2.5917296409606934,
+      "grad_norm_var": 0.11371707836486636,
+      "learning_rate": 0.0001,
+      "loss": 48.4286,
+      "loss/crossentropy": 2.8071954250335693,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2240188866853714,
+      "loss/reg": 45.39739227294922,
+      "step": 844
+    },
+    {
+      "epoch": 0.021125,
+      "grad_norm": 2.990241289138794,
+      "grad_norm_var": 0.09710077586091354,
+      "learning_rate": 0.0001,
+      "loss": 48.4181,
+      "loss/crossentropy": 2.8722949028015137,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25466662645339966,
+      "loss/reg": 45.29114532470703,
+      "step": 845
+    },
+    {
+      "epoch": 0.02115,
+      "grad_norm": 2.7038068771362305,
+      "grad_norm_var": 0.10132916045419596,
+      "learning_rate": 0.0001,
+      "loss": 48.5483,
+      "loss/crossentropy": 3.1085779666900635,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25524720549583435,
+      "loss/reg": 45.18445587158203,
+      "step": 846
+    },
+    {
+      "epoch": 0.021175,
+      "grad_norm": 2.6508195400238037,
+      "grad_norm_var": 0.10548054452361626,
+      "learning_rate": 0.0001,
+      "loss": 48.4558,
+      "loss/crossentropy": 3.1368627548217773,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23976629972457886,
+      "loss/reg": 45.079139709472656,
+      "step": 847
+    },
+    {
+      "epoch": 0.0212,
+      "grad_norm": 2.988497734069824,
+      "grad_norm_var": 0.0813341385370144,
+      "learning_rate": 0.0001,
+      "loss": 48.1072,
+      "loss/crossentropy": 2.8759114742279053,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25515347719192505,
+      "loss/reg": 44.97610855102539,
+      "step": 848
+    },
+    {
+      "epoch": 0.021225,
+      "grad_norm": 2.769603729248047,
+      "grad_norm_var": 0.0786111905047143,
+      "learning_rate": 0.0001,
+      "loss": 48.2049,
+      "loss/crossentropy": 3.070866346359253,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.258023738861084,
+      "loss/reg": 44.8759765625,
+      "step": 849
+    },
+    {
+      "epoch": 0.02125,
+      "grad_norm": 2.505613327026367,
+      "grad_norm_var": 0.03543295310148456,
+      "learning_rate": 0.0001,
+      "loss": 47.7345,
+      "loss/crossentropy": 2.7244465351104736,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2349836379289627,
+      "loss/reg": 44.77507019042969,
+      "step": 850
+    },
+    {
+      "epoch": 0.021275,
+      "grad_norm": 2.5850486755371094,
+      "grad_norm_var": 0.038190999955643436,
+      "learning_rate": 0.0001,
+      "loss": 47.6796,
+      "loss/crossentropy": 2.797024726867676,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21167223155498505,
+      "loss/reg": 44.670875549316406,
+      "step": 851
+    },
+    {
+      "epoch": 0.0213,
+      "grad_norm": 3.048123359680176,
+      "grad_norm_var": 0.04021511405729043,
+      "learning_rate": 0.0001,
+      "loss": 47.6982,
+      "loss/crossentropy": 2.882145881652832,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25044912099838257,
+      "loss/reg": 44.56555938720703,
+      "step": 852
+    },
+    {
+      "epoch": 0.021325,
+      "grad_norm": 2.6726913452148438,
+      "grad_norm_var": 0.04065821192390615,
+      "learning_rate": 0.0001,
+      "loss": 47.6222,
+      "loss/crossentropy": 2.9055726528167725,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2560499608516693,
+      "loss/reg": 44.460567474365234,
+      "step": 853
+    },
+    {
+      "epoch": 0.02135,
+      "grad_norm": 2.616454601287842,
+      "grad_norm_var": 0.03525310786582168,
+      "learning_rate": 0.0001,
+      "loss": 47.3723,
+      "loss/crossentropy": 2.7777538299560547,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23695862293243408,
+      "loss/reg": 44.357627868652344,
+      "step": 854
+    },
+    {
+      "epoch": 0.021375,
+      "grad_norm": 2.6965930461883545,
+      "grad_norm_var": 0.03452699702972097,
+      "learning_rate": 0.0001,
+      "loss": 47.1636,
+      "loss/crossentropy": 2.6849050521850586,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22069215774536133,
+      "loss/reg": 44.257965087890625,
+      "step": 855
+    },
+    {
+      "epoch": 0.0214,
+      "grad_norm": 2.5102341175079346,
+      "grad_norm_var": 0.0388638936666986,
+      "learning_rate": 0.0001,
+      "loss": 47.262,
+      "loss/crossentropy": 2.860992908477783,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24046388268470764,
+      "loss/reg": 44.160499572753906,
+      "step": 856
+    },
+    {
+      "epoch": 0.021425,
+      "grad_norm": 2.9143855571746826,
+      "grad_norm_var": 0.036940528281403574,
+      "learning_rate": 0.0001,
+      "loss": 47.0549,
+      "loss/crossentropy": 2.7531962394714355,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23869255185127258,
+      "loss/reg": 44.063053131103516,
+      "step": 857
+    },
+    {
+      "epoch": 0.02145,
+      "grad_norm": 2.7312750816345215,
+      "grad_norm_var": 0.03611533132415896,
+      "learning_rate": 0.0001,
+      "loss": 47.2825,
+      "loss/crossentropy": 3.063075304031372,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2570626139640808,
+      "loss/reg": 43.962337493896484,
+      "step": 858
+    },
+    {
+      "epoch": 0.021475,
+      "grad_norm": 2.6700384616851807,
+      "grad_norm_var": 0.029190745995408444,
+      "learning_rate": 0.0001,
+      "loss": 46.8626,
+      "loss/crossentropy": 2.765472173690796,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23290973901748657,
+      "loss/reg": 43.86418533325195,
+      "step": 859
+    },
+    {
+      "epoch": 0.0215,
+      "grad_norm": 2.5979349613189697,
+      "grad_norm_var": 0.029080552835625658,
+      "learning_rate": 0.0001,
+      "loss": 46.8266,
+      "loss/crossentropy": 2.843446969985962,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22045856714248657,
+      "loss/reg": 43.76265335083008,
+      "step": 860
+    },
+    {
+      "epoch": 0.021525,
+      "grad_norm": 2.8044228553771973,
+      "grad_norm_var": 0.0247465536201607,
+      "learning_rate": 0.0001,
+      "loss": 46.8942,
+      "loss/crossentropy": 2.9794631004333496,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25055602192878723,
+      "loss/reg": 43.664146423339844,
+      "step": 861
+    },
+    {
+      "epoch": 0.02155,
+      "grad_norm": 3.0274248123168945,
+      "grad_norm_var": 0.03074023090764418,
+      "learning_rate": 0.0001,
+      "loss": 46.4618,
+      "loss/crossentropy": 2.6372880935668945,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.26163622736930847,
+      "loss/reg": 43.56292724609375,
+      "step": 862
+    },
+    {
+      "epoch": 0.021575,
+      "grad_norm": 2.771188974380493,
+      "grad_norm_var": 0.030265496058056365,
+      "learning_rate": 0.0001,
+      "loss": 46.687,
+      "loss/crossentropy": 2.977721691131592,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24795976281166077,
+      "loss/reg": 43.46126937866211,
+      "step": 863
+    },
+    {
+      "epoch": 0.0216,
+      "grad_norm": 3.0198278427124023,
+      "grad_norm_var": 0.031346752653106096,
+      "learning_rate": 0.0001,
+      "loss": 46.4047,
+      "loss/crossentropy": 2.805306911468506,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23715007305145264,
+      "loss/reg": 43.36224365234375,
+      "step": 864
+    },
+    {
+      "epoch": 0.021625,
+      "grad_norm": 2.8970999717712402,
+      "grad_norm_var": 0.032758795573165383,
+      "learning_rate": 0.0001,
+      "loss": 46.3235,
+      "loss/crossentropy": 2.81503963470459,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2449081391096115,
+      "loss/reg": 43.263526916503906,
+      "step": 865
+    },
+    {
+      "epoch": 0.02165,
+      "grad_norm": 2.6958978176116943,
+      "grad_norm_var": 0.028713014180188404,
+      "learning_rate": 0.0001,
+      "loss": 46.4609,
+      "loss/crossentropy": 3.0649611949920654,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23137107491493225,
+      "loss/reg": 43.164615631103516,
+      "step": 866
+    },
+    {
+      "epoch": 0.021675,
+      "grad_norm": 2.738786458969116,
+      "grad_norm_var": 0.026477629275204133,
+      "learning_rate": 0.0001,
+      "loss": 46.2192,
+      "loss/crossentropy": 2.900941848754883,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25012895464897156,
+      "loss/reg": 43.06816482543945,
+      "step": 867
+    },
+    {
+      "epoch": 0.0217,
+      "grad_norm": 2.6869115829467773,
+      "grad_norm_var": 0.021515463936426234,
+      "learning_rate": 0.0001,
+      "loss": 46.0673,
+      "loss/crossentropy": 2.8564064502716064,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23792710900306702,
+      "loss/reg": 42.972923278808594,
+      "step": 868
+    },
+    {
+      "epoch": 0.021725,
+      "grad_norm": 2.827066421508789,
+      "grad_norm_var": 0.02134784747631405,
+      "learning_rate": 0.0001,
+      "loss": 45.8876,
+      "loss/crossentropy": 2.7518844604492188,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2576528787612915,
+      "loss/reg": 42.87803268432617,
+      "step": 869
+    },
+    {
+      "epoch": 0.02175,
+      "grad_norm": 2.626676321029663,
+      "grad_norm_var": 0.02115486089260775,
+      "learning_rate": 0.0001,
+      "loss": 45.7523,
+      "loss/crossentropy": 2.7327208518981934,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23792991042137146,
+      "loss/reg": 42.78163528442383,
+      "step": 870
+    },
+    {
+      "epoch": 0.021775,
+      "grad_norm": 2.7007803916931152,
+      "grad_norm_var": 0.021118609979900307,
+      "learning_rate": 0.0001,
+      "loss": 45.7201,
+      "loss/crossentropy": 2.7678470611572266,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.26777875423431396,
+      "loss/reg": 42.68449020385742,
+      "step": 871
+    },
+    {
+      "epoch": 0.0218,
+      "grad_norm": 2.867694616317749,
+      "grad_norm_var": 0.01702195773885459,
+      "learning_rate": 0.0001,
+      "loss": 45.5534,
+      "loss/crossentropy": 2.7220120429992676,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24349060654640198,
+      "loss/reg": 42.58792495727539,
+      "step": 872
+    },
+    {
+      "epoch": 0.021825,
+      "grad_norm": 2.7054524421691895,
+      "grad_norm_var": 0.016176199233639704,
+      "learning_rate": 0.0001,
+      "loss": 45.5576,
+      "loss/crossentropy": 2.821178913116455,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24668864905834198,
+      "loss/reg": 42.489776611328125,
+      "step": 873
+    },
+    {
+      "epoch": 0.02185,
+      "grad_norm": 2.7255043983459473,
+      "grad_norm_var": 0.01621040773991519,
+      "learning_rate": 0.0001,
+      "loss": 45.2197,
+      "loss/crossentropy": 2.5901880264282227,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24032941460609436,
+      "loss/reg": 42.389225006103516,
+      "step": 874
+    },
+    {
+      "epoch": 0.021875,
+      "grad_norm": 2.680649757385254,
+      "grad_norm_var": 0.016072239099798935,
+      "learning_rate": 0.0001,
+      "loss": 45.4495,
+      "loss/crossentropy": 2.920884132385254,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24124376475811005,
+      "loss/reg": 42.28740692138672,
+      "step": 875
+    },
+    {
+      "epoch": 0.0219,
+      "grad_norm": 2.7948641777038574,
+      "grad_norm_var": 0.013890606167317543,
+      "learning_rate": 0.0001,
+      "loss": 45.2185,
+      "loss/crossentropy": 2.7859625816345215,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24146077036857605,
+      "loss/reg": 42.191123962402344,
+      "step": 876
+    },
+    {
+      "epoch": 0.021925,
+      "grad_norm": 2.503937005996704,
+      "grad_norm_var": 0.018781331446677995,
+      "learning_rate": 0.0001,
+      "loss": 45.2798,
+      "loss/crossentropy": 2.9733173847198486,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21078920364379883,
+      "loss/reg": 42.09564971923828,
+      "step": 877
+    },
+    {
+      "epoch": 0.02195,
+      "grad_norm": 2.723689556121826,
+      "grad_norm_var": 0.013994920468044824,
+      "learning_rate": 0.0001,
+      "loss": 45.1466,
+      "loss/crossentropy": 2.9063451290130615,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24084623157978058,
+      "loss/reg": 41.999412536621094,
+      "step": 878
+    },
+    {
+      "epoch": 0.021975,
+      "grad_norm": 2.604905366897583,
+      "grad_norm_var": 0.015206201216361136,
+      "learning_rate": 0.0001,
+      "loss": 44.9635,
+      "loss/crossentropy": 2.8358519077301025,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22244781255722046,
+      "loss/reg": 41.905216217041016,
+      "step": 879
+    },
+    {
+      "epoch": 0.022,
+      "grad_norm": 2.6830010414123535,
+      "grad_norm_var": 0.009616840170906471,
+      "learning_rate": 0.0001,
+      "loss": 44.8298,
+      "loss/crossentropy": 2.7782702445983887,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2416587471961975,
+      "loss/reg": 41.80987548828125,
+      "step": 880
+    },
+    {
+      "epoch": 0.022025,
+      "grad_norm": 2.701080322265625,
+      "grad_norm_var": 0.007296400643871455,
+      "learning_rate": 0.0001,
+      "loss": 44.458,
+      "loss/crossentropy": 2.5311830043792725,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21052095293998718,
+      "loss/reg": 41.71625900268555,
+      "step": 881
+    },
+    {
+      "epoch": 0.02205,
+      "grad_norm": 2.5038342475891113,
+      "grad_norm_var": 0.009814048940694183,
+      "learning_rate": 0.0001,
+      "loss": 44.6839,
+      "loss/crossentropy": 2.8363115787506104,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22428202629089355,
+      "loss/reg": 41.623291015625,
+      "step": 882
+    },
+    {
+      "epoch": 0.022075,
+      "grad_norm": 2.757178783416748,
+      "grad_norm_var": 0.009949491806351792,
+      "learning_rate": 0.0001,
+      "loss": 44.6845,
+      "loss/crossentropy": 2.905855417251587,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2473723292350769,
+      "loss/reg": 41.531280517578125,
+      "step": 883
+    },
+    {
+      "epoch": 0.0221,
+      "grad_norm": 2.660141706466675,
+      "grad_norm_var": 0.0100171783636083,
+      "learning_rate": 0.0001,
+      "loss": 44.3141,
+      "loss/crossentropy": 2.6371467113494873,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23851189017295837,
+      "loss/reg": 41.43846893310547,
+      "step": 884
+    },
+    {
+      "epoch": 0.022125,
+      "grad_norm": 4.251846790313721,
+      "grad_norm_var": 0.1626166050996685,
+      "learning_rate": 0.0001,
+      "loss": 44.3927,
+      "loss/crossentropy": 2.787137746810913,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25860458612442017,
+      "loss/reg": 41.346946716308594,
+      "step": 885
+    },
+    {
+      "epoch": 0.02215,
+      "grad_norm": 2.9530351161956787,
+      "grad_norm_var": 0.16257111931878943,
+      "learning_rate": 0.0001,
+      "loss": 44.171,
+      "loss/crossentropy": 2.694733142852783,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22275252640247345,
+      "loss/reg": 41.25349426269531,
+      "step": 886
+    },
+    {
+      "epoch": 0.022175,
+      "grad_norm": 2.991832971572876,
+      "grad_norm_var": 0.1639725008568424,
+      "learning_rate": 0.0001,
+      "loss": 44.5867,
+      "loss/crossentropy": 3.156198501586914,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2699030637741089,
+      "loss/reg": 41.16058349609375,
+      "step": 887
+    },
+    {
+      "epoch": 0.0222,
+      "grad_norm": 3.1930086612701416,
+      "grad_norm_var": 0.1726863652995225,
+      "learning_rate": 0.0001,
+      "loss": 43.9933,
+      "loss/crossentropy": 2.6939613819122314,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22894969582557678,
+      "loss/reg": 41.07038879394531,
+      "step": 888
+    },
+    {
+      "epoch": 0.022225,
+      "grad_norm": 2.8903820514678955,
+      "grad_norm_var": 0.17151552786031227,
+      "learning_rate": 0.0001,
+      "loss": 44.0888,
+      "loss/crossentropy": 2.862319231033325,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24869811534881592,
+      "loss/reg": 40.97781753540039,
+      "step": 889
+    },
+    {
+      "epoch": 0.02225,
+      "grad_norm": 2.881136655807495,
+      "grad_norm_var": 0.17042145948309156,
+      "learning_rate": 0.0001,
+      "loss": 44.3592,
+      "loss/crossentropy": 3.2091801166534424,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2633001506328583,
+      "loss/reg": 40.88669967651367,
+      "step": 890
+    },
+    {
+      "epoch": 0.022275,
+      "grad_norm": 2.62827467918396,
+      "grad_norm_var": 0.17185170990455442,
+      "learning_rate": 0.0001,
+      "loss": 43.7441,
+      "loss/crossentropy": 2.69647479057312,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25092294812202454,
+      "loss/reg": 40.79666519165039,
+      "step": 891
+    },
+    {
+      "epoch": 0.0223,
+      "grad_norm": 2.436519145965576,
+      "grad_norm_var": 0.18287652337390886,
+      "learning_rate": 0.0001,
+      "loss": 43.7102,
+      "loss/crossentropy": 2.7679152488708496,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23737739026546478,
+      "loss/reg": 40.704925537109375,
+      "step": 892
+    },
+    {
+      "epoch": 0.022325,
+      "grad_norm": 2.700011968612671,
+      "grad_norm_var": 0.17661805600996155,
+      "learning_rate": 0.0001,
+      "loss": 43.6811,
+      "loss/crossentropy": 2.8262124061584473,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23837611079216003,
+      "loss/reg": 40.61653137207031,
+      "step": 893
+    },
+    {
+      "epoch": 0.02235,
+      "grad_norm": 2.5235769748687744,
+      "grad_norm_var": 0.18242413999566054,
+      "learning_rate": 0.0001,
+      "loss": 43.4445,
+      "loss/crossentropy": 2.6917617321014404,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22554537653923035,
+      "loss/reg": 40.527217864990234,
+      "step": 894
+    },
+    {
+      "epoch": 0.022375,
+      "grad_norm": 2.7341766357421875,
+      "grad_norm_var": 0.17950288283988294,
+      "learning_rate": 0.0001,
+      "loss": 43.7484,
+      "loss/crossentropy": 3.029411792755127,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.28348681330680847,
+      "loss/reg": 40.43548583984375,
+      "step": 895
+    },
+    {
+      "epoch": 0.0224,
+      "grad_norm": 2.659266233444214,
+      "grad_norm_var": 0.18004463619057087,
+      "learning_rate": 0.0001,
+      "loss": 43.1614,
+      "loss/crossentropy": 2.5794694423675537,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23672150075435638,
+      "loss/reg": 40.34519577026367,
+      "step": 896
+    },
+    {
+      "epoch": 0.022425,
+      "grad_norm": 3.1245830059051514,
+      "grad_norm_var": 0.1833206141640287,
+      "learning_rate": 0.0001,
+      "loss": 43.5047,
+      "loss/crossentropy": 3.009214162826538,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24004200100898743,
+      "loss/reg": 40.25539779663086,
+      "step": 897
+    },
+    {
+      "epoch": 0.02245,
+      "grad_norm": 3.283797264099121,
+      "grad_norm_var": 0.1834653295253453,
+      "learning_rate": 0.0001,
+      "loss": 42.9276,
+      "loss/crossentropy": 2.5301198959350586,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23050135374069214,
+      "loss/reg": 40.16700744628906,
+      "step": 898
+    },
+    {
+      "epoch": 0.022475,
+      "grad_norm": 2.4279603958129883,
+      "grad_norm_var": 0.19724598877930527,
+      "learning_rate": 0.0001,
+      "loss": 43.0253,
+      "loss/crossentropy": 2.7136290073394775,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23277553915977478,
+      "loss/reg": 40.078914642333984,
+      "step": 899
+    },
+    {
+      "epoch": 0.0225,
+      "grad_norm": 2.5088298320770264,
+      "grad_norm_var": 0.2034398420188303,
+      "learning_rate": 0.0001,
+      "loss": 43.1582,
+      "loss/crossentropy": 2.948103189468384,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21640148758888245,
+      "loss/reg": 39.99374008178711,
+      "step": 900
+    },
+    {
+      "epoch": 0.022525,
+      "grad_norm": 2.6329402923583984,
+      "grad_norm_var": 0.07258482335469481,
+      "learning_rate": 0.0001,
+      "loss": 43.0828,
+      "loss/crossentropy": 2.9187188148498535,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25746428966522217,
+      "loss/reg": 39.90663528442383,
+      "step": 901
+    },
+    {
+      "epoch": 0.02255,
+      "grad_norm": 2.9308879375457764,
+      "grad_norm_var": 0.07212100124237182,
+      "learning_rate": 0.0001,
+      "loss": 42.8767,
+      "loss/crossentropy": 2.779430866241455,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2763053774833679,
+      "loss/reg": 39.82095718383789,
+      "step": 902
+    },
+    {
+      "epoch": 0.022575,
+      "grad_norm": 2.6100316047668457,
+      "grad_norm_var": 0.07066177999980709,
+      "learning_rate": 0.0001,
+      "loss": 42.7168,
+      "loss/crossentropy": 2.7670066356658936,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21381208300590515,
+      "loss/reg": 39.73594665527344,
+      "step": 903
+    },
+    {
+      "epoch": 0.0226,
+      "grad_norm": 2.6517529487609863,
+      "grad_norm_var": 0.057746798972484,
+      "learning_rate": 0.0001,
+      "loss": 42.6053,
+      "loss/crossentropy": 2.7067525386810303,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2483087182044983,
+      "loss/reg": 39.65021896362305,
+      "step": 904
+    },
+    {
+      "epoch": 0.022625,
+      "grad_norm": 2.896390914916992,
+      "grad_norm_var": 0.057880348550103614,
+      "learning_rate": 0.0001,
+      "loss": 42.4123,
+      "loss/crossentropy": 2.6049654483795166,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24248483777046204,
+      "loss/reg": 39.56484603881836,
+      "step": 905
+    },
+    {
+      "epoch": 0.02265,
+      "grad_norm": 2.7035367488861084,
+      "grad_norm_var": 0.05619899439359794,
+      "learning_rate": 0.0001,
+      "loss": 42.5853,
+      "loss/crossentropy": 2.8769829273223877,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2273760437965393,
+      "loss/reg": 39.48089599609375,
+      "step": 906
+    },
+    {
+      "epoch": 0.022675,
+      "grad_norm": 2.7665393352508545,
+      "grad_norm_var": 0.05578056314836128,
+      "learning_rate": 0.0001,
+      "loss": 42.4535,
+      "loss/crossentropy": 2.8200128078460693,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2363729327917099,
+      "loss/reg": 39.39712905883789,
+      "step": 907
+    },
+    {
+      "epoch": 0.0227,
+      "grad_norm": 2.480107545852661,
+      "grad_norm_var": 0.05422606208670284,
+      "learning_rate": 0.0001,
+      "loss": 42.4059,
+      "loss/crossentropy": 2.8648123741149902,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22969950735569,
+      "loss/reg": 39.31142044067383,
+      "step": 908
+    },
+    {
+      "epoch": 0.022725,
+      "grad_norm": 2.7900424003601074,
+      "grad_norm_var": 0.05440689578418396,
+      "learning_rate": 0.0001,
+      "loss": 42.4378,
+      "loss/crossentropy": 2.9586613178253174,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25124895572662354,
+      "loss/reg": 39.2279052734375,
+      "step": 909
+    },
+    {
+      "epoch": 0.02275,
+      "grad_norm": 2.8045692443847656,
+      "grad_norm_var": 0.051503902709208425,
+      "learning_rate": 0.0001,
+      "loss": 42.0578,
+      "loss/crossentropy": 2.6795575618743896,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23502734303474426,
+      "loss/reg": 39.14320373535156,
+      "step": 910
+    },
+    {
+      "epoch": 0.022775,
+      "grad_norm": 3.0532169342041016,
+      "grad_norm_var": 0.05717807714263259,
+      "learning_rate": 0.0001,
+      "loss": 42.4633,
+      "loss/crossentropy": 3.152160882949829,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25327080488204956,
+      "loss/reg": 39.05790710449219,
+      "step": 911
+    },
+    {
+      "epoch": 0.0228,
+      "grad_norm": 2.6680328845977783,
+      "grad_norm_var": 0.05705311999868907,
+      "learning_rate": 0.0001,
+      "loss": 42.2034,
+      "loss/crossentropy": 2.99041748046875,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23903341591358185,
+      "loss/reg": 38.97396469116211,
+      "step": 912
+    },
+    {
+      "epoch": 0.022825,
+      "grad_norm": 3.3302087783813477,
+      "grad_norm_var": 0.0693946111033938,
+      "learning_rate": 0.0001,
+      "loss": 42.0845,
+      "loss/crossentropy": 2.930839776992798,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2614016532897949,
+      "loss/reg": 38.892234802246094,
+      "step": 913
+    },
+    {
+      "epoch": 0.02285,
+      "grad_norm": 2.6165523529052734,
+      "grad_norm_var": 0.05272697596004718,
+      "learning_rate": 0.0001,
+      "loss": 41.7748,
+      "loss/crossentropy": 2.7452762126922607,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2193608283996582,
+      "loss/reg": 38.81016540527344,
+      "step": 914
+    },
+    {
+      "epoch": 0.022875,
+      "grad_norm": 2.933302402496338,
+      "grad_norm_var": 0.04752966494528721,
+      "learning_rate": 0.0001,
+      "loss": 41.9394,
+      "loss/crossentropy": 2.9566924571990967,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25487080216407776,
+      "loss/reg": 38.72779846191406,
+      "step": 915
+    },
+    {
+      "epoch": 0.0229,
+      "grad_norm": 2.6887619495391846,
+      "grad_norm_var": 0.043202036673411236,
+      "learning_rate": 0.0001,
+      "loss": 41.6663,
+      "loss/crossentropy": 2.798774242401123,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2221754789352417,
+      "loss/reg": 38.645347595214844,
+      "step": 916
+    },
+    {
+      "epoch": 0.022925,
+      "grad_norm": 2.8121073246002197,
+      "grad_norm_var": 0.04158046028161924,
+      "learning_rate": 0.0001,
+      "loss": 41.463,
+      "loss/crossentropy": 2.6423110961914062,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2565896213054657,
+      "loss/reg": 38.56413269042969,
+      "step": 917
+    },
+    {
+      "epoch": 0.02295,
+      "grad_norm": 2.509282350540161,
+      "grad_norm_var": 0.045107458768999464,
+      "learning_rate": 0.0001,
+      "loss": 41.6478,
+      "loss/crossentropy": 2.930504322052002,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2339283674955368,
+      "loss/reg": 38.48335647583008,
+      "step": 918
+    },
+    {
+      "epoch": 0.022975,
+      "grad_norm": 2.6414058208465576,
+      "grad_norm_var": 0.04450125049984308,
+      "learning_rate": 0.0001,
+      "loss": 41.2222,
+      "loss/crossentropy": 2.609210252761841,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2125830054283142,
+      "loss/reg": 38.40039825439453,
+      "step": 919
+    },
+    {
+      "epoch": 0.023,
+      "grad_norm": 2.715888261795044,
+      "grad_norm_var": 0.0437333648592715,
+      "learning_rate": 0.0001,
+      "loss": 41.2045,
+      "loss/crossentropy": 2.648813009262085,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2370688021183014,
+      "loss/reg": 38.318572998046875,
+      "step": 920
+    },
+    {
+      "epoch": 0.023025,
+      "grad_norm": 2.711751937866211,
+      "grad_norm_var": 0.04289092327507144,
+      "learning_rate": 0.0001,
+      "loss": 41.3706,
+      "loss/crossentropy": 2.885514736175537,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24639633297920227,
+      "loss/reg": 38.238643646240234,
+      "step": 921
+    },
+    {
+      "epoch": 0.02305,
+      "grad_norm": 2.6456990242004395,
+      "grad_norm_var": 0.04356690227298922,
+      "learning_rate": 0.0001,
+      "loss": 41.3587,
+      "loss/crossentropy": 2.941075563430786,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2592896819114685,
+      "loss/reg": 38.15837478637695,
+      "step": 922
+    },
+    {
+      "epoch": 0.023075,
+      "grad_norm": 2.8968873023986816,
+      "grad_norm_var": 0.044734353597355184,
+      "learning_rate": 0.0001,
+      "loss": 41.2417,
+      "loss/crossentropy": 2.9061999320983887,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25584694743156433,
+      "loss/reg": 38.07965850830078,
+      "step": 923
+    },
+    {
+      "epoch": 0.0231,
+      "grad_norm": 2.6788389682769775,
+      "grad_norm_var": 0.039558045732947515,
+      "learning_rate": 0.0001,
+      "loss": 40.9876,
+      "loss/crossentropy": 2.746546506881714,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23900818824768066,
+      "loss/reg": 38.002079010009766,
+      "step": 924
+    },
+    {
+      "epoch": 0.023125,
+      "grad_norm": 3.128474712371826,
+      "grad_norm_var": 0.047123059326641466,
+      "learning_rate": 0.0001,
+      "loss": 41.2668,
+      "loss/crossentropy": 3.087491035461426,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25750723481178284,
+      "loss/reg": 37.921783447265625,
+      "step": 925
+    },
+    {
+      "epoch": 0.02315,
+      "grad_norm": 2.8882362842559814,
+      "grad_norm_var": 0.04758715374512785,
+      "learning_rate": 0.0001,
+      "loss": 40.9121,
+      "loss/crossentropy": 2.817408800125122,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2543855309486389,
+      "loss/reg": 37.84029006958008,
+      "step": 926
+    },
+    {
+      "epoch": 0.023175,
+      "grad_norm": 2.8361799716949463,
+      "grad_norm_var": 0.043418151431792985,
+      "learning_rate": 0.0001,
+      "loss": 40.832,
+      "loss/crossentropy": 2.8375983238220215,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23513871431350708,
+      "loss/reg": 37.75931167602539,
+      "step": 927
+    },
+    {
+      "epoch": 0.0232,
+      "grad_norm": 2.909255266189575,
+      "grad_norm_var": 0.043008241613055276,
+      "learning_rate": 0.0001,
+      "loss": 41.1667,
+      "loss/crossentropy": 3.246528387069702,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24166589975357056,
+      "loss/reg": 37.67852783203125,
+      "step": 928
+    },
+    {
+      "epoch": 0.023225,
+      "grad_norm": 2.552696704864502,
+      "grad_norm_var": 0.026750676712286037,
+      "learning_rate": 0.0001,
+      "loss": 40.818,
+      "loss/crossentropy": 2.9691615104675293,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24937693774700165,
+      "loss/reg": 37.59941101074219,
+      "step": 929
+    },
+    {
+      "epoch": 0.02325,
+      "grad_norm": 2.887279748916626,
+      "grad_norm_var": 0.026141477988341915,
+      "learning_rate": 0.0001,
+      "loss": 40.7307,
+      "loss/crossentropy": 2.956307888031006,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2544440031051636,
+      "loss/reg": 37.51991653442383,
+      "step": 930
+    },
+    {
+      "epoch": 0.023275,
+      "grad_norm": 2.615774154663086,
+      "grad_norm_var": 0.02583631071739016,
+      "learning_rate": 0.0001,
+      "loss": 40.504,
+      "loss/crossentropy": 2.836076259613037,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22688478231430054,
+      "loss/reg": 37.4410285949707,
+      "step": 931
+    },
+    {
+      "epoch": 0.0233,
+      "grad_norm": 3.346874475479126,
+      "grad_norm_var": 0.046882285076200486,
+      "learning_rate": 0.0001,
+      "loss": 40.2933,
+      "loss/crossentropy": 2.698227643966675,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23271670937538147,
+      "loss/reg": 37.36236572265625,
+      "step": 932
+    },
+    {
+      "epoch": 0.023325,
+      "grad_norm": 3.105316162109375,
+      "grad_norm_var": 0.052785925698018585,
+      "learning_rate": 0.0001,
+      "loss": 40.2366,
+      "loss/crossentropy": 2.690444231033325,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2598923146724701,
+      "loss/reg": 37.28624725341797,
+      "step": 933
+    },
+    {
+      "epoch": 0.02335,
+      "grad_norm": 2.8327033519744873,
+      "grad_norm_var": 0.046059668983682654,
+      "learning_rate": 0.0001,
+      "loss": 40.4552,
+      "loss/crossentropy": 3.0131001472473145,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23372459411621094,
+      "loss/reg": 37.20838165283203,
+      "step": 934
+    },
+    {
+      "epoch": 0.023375,
+      "grad_norm": 2.720252513885498,
+      "grad_norm_var": 0.04439112918428994,
+      "learning_rate": 0.0001,
+      "loss": 40.1157,
+      "loss/crossentropy": 2.748929977416992,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23494702577590942,
+      "loss/reg": 37.13186264038086,
+      "step": 935
+    },
+    {
+      "epoch": 0.0234,
+      "grad_norm": 2.510647773742676,
+      "grad_norm_var": 0.05047514191979848,
+      "learning_rate": 0.0001,
+      "loss": 39.8904,
+      "loss/crossentropy": 2.602600336074829,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23265476524829865,
+      "loss/reg": 37.05512237548828,
+      "step": 936
+    },
+    {
+      "epoch": 0.023425,
+      "grad_norm": 2.9770853519439697,
+      "grad_norm_var": 0.05072093631072979,
+      "learning_rate": 0.0001,
+      "loss": 40.3788,
+      "loss/crossentropy": 3.1354868412017822,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2637995779514313,
+      "loss/reg": 36.97951889038086,
+      "step": 937
+    },
+    {
+      "epoch": 0.02345,
+      "grad_norm": 2.5489563941955566,
+      "grad_norm_var": 0.053886506006241085,
+      "learning_rate": 0.0001,
+      "loss": 39.9812,
+      "loss/crossentropy": 2.8318233489990234,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24620817601680756,
+      "loss/reg": 36.90315246582031,
+      "step": 938
+    },
+    {
+      "epoch": 0.023475,
+      "grad_norm": 2.9009687900543213,
+      "grad_norm_var": 0.05391865958705443,
+      "learning_rate": 0.0001,
+      "loss": 39.9177,
+      "loss/crossentropy": 2.8244364261627197,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2657071053981781,
+      "loss/reg": 36.82755661010742,
+      "step": 939
+    },
+    {
+      "epoch": 0.0235,
+      "grad_norm": 2.9793155193328857,
+      "grad_norm_var": 0.05310601324464391,
+      "learning_rate": 0.0001,
+      "loss": 40.0234,
+      "loss/crossentropy": 3.0105204582214355,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2618894875049591,
+      "loss/reg": 36.750953674316406,
+      "step": 940
+    },
+    {
+      "epoch": 0.023525,
+      "grad_norm": 2.8539981842041016,
+      "grad_norm_var": 0.0479435574644467,
+      "learning_rate": 0.0001,
+      "loss": 39.8117,
+      "loss/crossentropy": 2.8921923637390137,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2426648885011673,
+      "loss/reg": 36.67680358886719,
+      "step": 941
+    },
+    {
+      "epoch": 0.02355,
+      "grad_norm": 2.6014649868011475,
+      "grad_norm_var": 0.05130008365882167,
+      "learning_rate": 0.0001,
+      "loss": 39.7408,
+      "loss/crossentropy": 2.896888017654419,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2410784810781479,
+      "loss/reg": 36.60279083251953,
+      "step": 942
+    },
+    {
+      "epoch": 0.023575,
+      "grad_norm": 2.8801045417785645,
+      "grad_norm_var": 0.05149391710301548,
+      "learning_rate": 0.0001,
+      "loss": 39.7228,
+      "loss/crossentropy": 2.9397897720336914,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25510135293006897,
+      "loss/reg": 36.52793502807617,
+      "step": 943
+    },
+    {
+      "epoch": 0.0236,
+      "grad_norm": 2.8186585903167725,
+      "grad_norm_var": 0.05100626896681509,
+      "learning_rate": 0.0001,
+      "loss": 39.47,
+      "loss/crossentropy": 2.75411057472229,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.26445087790489197,
+      "loss/reg": 36.451393127441406,
+      "step": 944
+    },
+    {
+      "epoch": 0.023625,
+      "grad_norm": 2.9633240699768066,
+      "grad_norm_var": 0.04686836082955779,
+      "learning_rate": 0.0001,
+      "loss": 39.3906,
+      "loss/crossentropy": 2.7534892559051514,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25982236862182617,
+      "loss/reg": 36.377254486083984,
+      "step": 945
+    },
+    {
+      "epoch": 0.02365,
+      "grad_norm": 2.929901123046875,
+      "grad_norm_var": 0.04721409535759804,
+      "learning_rate": 0.0001,
+      "loss": 39.2373,
+      "loss/crossentropy": 2.6932575702667236,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2382897436618805,
+      "loss/reg": 36.305755615234375,
+      "step": 946
+    },
+    {
+      "epoch": 0.023675,
+      "grad_norm": 2.675022602081299,
+      "grad_norm_var": 0.04559039355538038,
+      "learning_rate": 0.0001,
+      "loss": 39.2789,
+      "loss/crossentropy": 2.8028876781463623,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24303217232227325,
+      "loss/reg": 36.23302459716797,
+      "step": 947
+    },
+    {
+      "epoch": 0.0237,
+      "grad_norm": 2.8433735370635986,
+      "grad_norm_var": 0.0282651774333182,
+      "learning_rate": 0.0001,
+      "loss": 39.181,
+      "loss/crossentropy": 2.774716854095459,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24557200074195862,
+      "loss/reg": 36.16071319580078,
+      "step": 948
+    },
+    {
+      "epoch": 0.023725,
+      "grad_norm": 2.809319257736206,
+      "grad_norm_var": 0.022532732866163013,
+      "learning_rate": 0.0001,
+      "loss": 39.33,
+      "loss/crossentropy": 3.011596918106079,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.228348970413208,
+      "loss/reg": 36.09001159667969,
+      "step": 949
+    },
+    {
+      "epoch": 0.02375,
+      "grad_norm": 2.9273681640625,
+      "grad_norm_var": 0.023470027420536846,
+      "learning_rate": 0.0001,
+      "loss": 38.9413,
+      "loss/crossentropy": 2.6855454444885254,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23887701332569122,
+      "loss/reg": 36.01683044433594,
+      "step": 950
+    },
+    {
+      "epoch": 0.023775,
+      "grad_norm": 3.389801263809204,
+      "grad_norm_var": 0.043589378411585736,
+      "learning_rate": 0.0001,
+      "loss": 39.4398,
+      "loss/crossentropy": 3.211602210998535,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.283491313457489,
+      "loss/reg": 35.944705963134766,
+      "step": 951
+    },
+    {
+      "epoch": 0.0238,
+      "grad_norm": 2.8046481609344482,
+      "grad_norm_var": 0.03566620795445464,
+      "learning_rate": 0.0001,
+      "loss": 38.8264,
+      "loss/crossentropy": 2.689462661743164,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2632759213447571,
+      "loss/reg": 35.87367248535156,
+      "step": 952
+    },
+    {
+      "epoch": 0.023825,
+      "grad_norm": 2.9345734119415283,
+      "grad_norm_var": 0.03516626203340228,
+      "learning_rate": 0.0001,
+      "loss": 39.0027,
+      "loss/crossentropy": 2.9450294971466064,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2565668821334839,
+      "loss/reg": 35.8011474609375,
+      "step": 953
+    },
+    {
+      "epoch": 0.02385,
+      "grad_norm": 2.4629881381988525,
+      "grad_norm_var": 0.03926570002250382,
+      "learning_rate": 0.0001,
+      "loss": 38.6895,
+      "loss/crossentropy": 2.7172951698303223,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24316415190696716,
+      "loss/reg": 35.72907257080078,
+      "step": 954
+    },
+    {
+      "epoch": 0.023875,
+      "grad_norm": 3.0658795833587646,
+      "grad_norm_var": 0.041845868526290055,
+      "learning_rate": 0.0001,
+      "loss": 38.8817,
+      "loss/crossentropy": 2.9409422874450684,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2831678092479706,
+      "loss/reg": 35.65754699707031,
+      "step": 955
+    },
+    {
+      "epoch": 0.0239,
+      "grad_norm": 2.8264553546905518,
+      "grad_norm_var": 0.04110340640186886,
+      "learning_rate": 0.0001,
+      "loss": 38.6798,
+      "loss/crossentropy": 2.841045618057251,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25310787558555603,
+      "loss/reg": 35.58567428588867,
+      "step": 956
+    },
+    {
+      "epoch": 0.023925,
+      "grad_norm": 2.8380401134490967,
+      "grad_norm_var": 0.041135667710252004,
+      "learning_rate": 0.0001,
+      "loss": 38.459,
+      "loss/crossentropy": 2.6821844577789307,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2615047097206116,
+      "loss/reg": 35.515350341796875,
+      "step": 957
+    },
+    {
+      "epoch": 0.02395,
+      "grad_norm": 3.5808513164520264,
+      "grad_norm_var": 0.06723561445601618,
+      "learning_rate": 0.0001,
+      "loss": 38.8643,
+      "loss/crossentropy": 3.114391803741455,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.30714309215545654,
+      "loss/reg": 35.44272994995117,
+      "step": 958
+    },
+    {
+      "epoch": 0.023975,
+      "grad_norm": 2.6595239639282227,
+      "grad_norm_var": 0.07150567007109672,
+      "learning_rate": 0.0001,
+      "loss": 38.0653,
+      "loss/crossentropy": 2.4589178562164307,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23348893225193024,
+      "loss/reg": 35.37288284301758,
+      "step": 959
+    },
+    {
+      "epoch": 0.024,
+      "grad_norm": 2.541332483291626,
+      "grad_norm_var": 0.07962008638647618,
+      "learning_rate": 0.0001,
+      "loss": 38.2545,
+      "loss/crossentropy": 2.7399823665618896,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2123335599899292,
+      "loss/reg": 35.30220413208008,
+      "step": 960
+    },
+    {
+      "epoch": 0.024025,
+      "grad_norm": 3.438565254211426,
+      "grad_norm_var": 0.09833307021827574,
+      "learning_rate": 0.0001,
+      "loss": 38.371,
+      "loss/crossentropy": 2.8811817169189453,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25691962242126465,
+      "loss/reg": 35.232913970947266,
+      "step": 961
+    },
+    {
+      "epoch": 0.02405,
+      "grad_norm": 2.7037241458892822,
+      "grad_norm_var": 0.10124614126430061,
+      "learning_rate": 0.0001,
+      "loss": 38.2645,
+      "loss/crossentropy": 2.8557090759277344,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24377232789993286,
+      "loss/reg": 35.1650505065918,
+      "step": 962
+    },
+    {
+      "epoch": 0.024075,
+      "grad_norm": 2.69218111038208,
+      "grad_norm_var": 0.10073533014069654,
+      "learning_rate": 0.0001,
+      "loss": 38.0522,
+      "loss/crossentropy": 2.7217583656311035,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2358003854751587,
+      "loss/reg": 35.09463882446289,
+      "step": 963
+    },
+    {
+      "epoch": 0.0241,
+      "grad_norm": 3.008634090423584,
+      "grad_norm_var": 0.10103115408336763,
+      "learning_rate": 0.0001,
+      "loss": 38.3344,
+      "loss/crossentropy": 3.036750316619873,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2726588249206543,
+      "loss/reg": 35.02499771118164,
+      "step": 964
+    },
+    {
+      "epoch": 0.024125,
+      "grad_norm": 2.662174940109253,
+      "grad_norm_var": 0.10451155871548182,
+      "learning_rate": 0.0001,
+      "loss": 38.3726,
+      "loss/crossentropy": 3.1680073738098145,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24728718400001526,
+      "loss/reg": 34.95726013183594,
+      "step": 965
+    },
+    {
+      "epoch": 0.02415,
+      "grad_norm": 2.8360965251922607,
+      "grad_norm_var": 0.10480316259746587,
+      "learning_rate": 0.0001,
+      "loss": 37.8496,
+      "loss/crossentropy": 2.714099884033203,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2466110736131668,
+      "loss/reg": 34.888893127441406,
+      "step": 966
+    },
+    {
+      "epoch": 0.024175,
+      "grad_norm": 2.738298177719116,
+      "grad_norm_var": 0.08903093948484392,
+      "learning_rate": 0.0001,
+      "loss": 38.2404,
+      "loss/crossentropy": 3.152971029281616,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.268291711807251,
+      "loss/reg": 34.819091796875,
+      "step": 967
+    },
+    {
+      "epoch": 0.0242,
+      "grad_norm": 2.773857831954956,
+      "grad_norm_var": 0.08932614783542002,
+      "learning_rate": 0.0001,
+      "loss": 38.0543,
+      "loss/crossentropy": 3.0643906593322754,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2395814061164856,
+      "loss/reg": 34.7503547668457,
+      "step": 968
+    },
+    {
+      "epoch": 0.024225,
+      "grad_norm": 2.555189609527588,
+      "grad_norm_var": 0.09455968532326603,
+      "learning_rate": 0.0001,
+      "loss": 37.7541,
+      "loss/crossentropy": 2.834400177001953,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23769161105155945,
+      "loss/reg": 34.68204879760742,
+      "step": 969
+    },
+    {
+      "epoch": 0.02425,
+      "grad_norm": 3.086888551712036,
+      "grad_norm_var": 0.08781776896609182,
+      "learning_rate": 0.0001,
+      "loss": 37.8627,
+      "loss/crossentropy": 2.9810502529144287,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.26743876934051514,
+      "loss/reg": 34.61421585083008,
+      "step": 970
+    },
+    {
+      "epoch": 0.024275,
+      "grad_norm": 4.088360786437988,
+      "grad_norm_var": 0.17911672859183908,
+      "learning_rate": 0.0001,
+      "loss": 37.7253,
+      "loss/crossentropy": 2.8682823181152344,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.30868783593177795,
+      "loss/reg": 34.54836654663086,
+      "step": 971
+    },
+    {
+      "epoch": 0.0243,
+      "grad_norm": 3.1919515132904053,
+      "grad_norm_var": 0.1819625200877444,
+      "learning_rate": 0.0001,
+      "loss": 37.4812,
+      "loss/crossentropy": 2.7390143871307373,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.26100480556488037,
+      "loss/reg": 34.481136322021484,
+      "step": 972
+    },
+    {
+      "epoch": 0.024325,
+      "grad_norm": 2.575105667114258,
+      "grad_norm_var": 0.19063724665019186,
+      "learning_rate": 0.0001,
+      "loss": 37.4356,
+      "loss/crossentropy": 2.8003249168395996,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2204829901456833,
+      "loss/reg": 34.41484069824219,
+      "step": 973
+    },
+    {
+      "epoch": 0.02435,
+      "grad_norm": 2.9630026817321777,
+      "grad_norm_var": 0.16218006358210366,
+      "learning_rate": 0.0001,
+      "loss": 37.519,
+      "loss/crossentropy": 2.894831657409668,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2747398912906647,
+      "loss/reg": 34.34943771362305,
+      "step": 974
+    },
+    {
+      "epoch": 0.024375,
+      "grad_norm": 2.6025750637054443,
+      "grad_norm_var": 0.16426326415777884,
+      "learning_rate": 0.0001,
+      "loss": 37.0734,
+      "loss/crossentropy": 2.554583787918091,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23361030220985413,
+      "loss/reg": 34.28517532348633,
+      "step": 975
+    },
+    {
+      "epoch": 0.0244,
+      "grad_norm": 2.6642050743103027,
+      "grad_norm_var": 0.15927148910242578,
+      "learning_rate": 0.0001,
+      "loss": 37.4822,
+      "loss/crossentropy": 3.0164380073547363,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24384766817092896,
+      "loss/reg": 34.22189712524414,
+      "step": 976
+    },
+    {
+      "epoch": 0.024425,
+      "grad_norm": 3.38037371635437,
+      "grad_norm_var": 0.15539215192371383,
+      "learning_rate": 0.0001,
+      "loss": 37.2464,
+      "loss/crossentropy": 2.8206450939178467,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2702368497848511,
+      "loss/reg": 34.155494689941406,
+      "step": 977
+    },
+    {
+      "epoch": 0.02445,
+      "grad_norm": 4.222375392913818,
+      "grad_norm_var": 0.2582409245142865,
+      "learning_rate": 0.0001,
+      "loss": 37.0559,
+      "loss/crossentropy": 2.7178001403808594,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24727877974510193,
+      "loss/reg": 34.0908317565918,
+      "step": 978
+    },
+    {
+      "epoch": 0.024475,
+      "grad_norm": 2.7761597633361816,
+      "grad_norm_var": 0.2552061228003131,
+      "learning_rate": 0.0001,
+      "loss": 37.1297,
+      "loss/crossentropy": 2.860049247741699,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24505718052387238,
+      "loss/reg": 34.02457046508789,
+      "step": 979
+    },
+    {
+      "epoch": 0.0245,
+      "grad_norm": 3.6322226524353027,
+      "grad_norm_var": 0.2795770565181319,
+      "learning_rate": 0.0001,
+      "loss": 36.8498,
+      "loss/crossentropy": 2.6204307079315186,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2698652446269989,
+      "loss/reg": 33.95948791503906,
+      "step": 980
+    },
+    {
+      "epoch": 0.024525,
+      "grad_norm": 2.981858491897583,
+      "grad_norm_var": 0.2695698630538421,
+      "learning_rate": 0.0001,
+      "loss": 36.9264,
+      "loss/crossentropy": 2.737006187438965,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.29314208030700684,
+      "loss/reg": 33.89624786376953,
+      "step": 981
+    },
+    {
+      "epoch": 0.02455,
+      "grad_norm": 2.74503755569458,
+      "grad_norm_var": 0.2728889013063165,
+      "learning_rate": 0.0001,
+      "loss": 36.943,
+      "loss/crossentropy": 2.8646838665008545,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24609914422035217,
+      "loss/reg": 33.8322639465332,
+      "step": 982
+    },
+    {
+      "epoch": 0.024575,
+      "grad_norm": 3.004685640335083,
+      "grad_norm_var": 0.26585896787462554,
+      "learning_rate": 0.0001,
+      "loss": 36.8484,
+      "loss/crossentropy": 2.8109850883483887,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.27065545320510864,
+      "loss/reg": 33.76673889160156,
+      "step": 983
+    },
+    {
+      "epoch": 0.0246,
+      "grad_norm": 2.5903549194335938,
+      "grad_norm_var": 0.2753986673068833,
+      "learning_rate": 0.0001,
+      "loss": 36.6906,
+      "loss/crossentropy": 2.7470083236694336,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23907317221164703,
+      "loss/reg": 33.70456314086914,
+      "step": 984
+    },
+    {
+      "epoch": 0.024625,
+      "grad_norm": 3.030545234680176,
+      "grad_norm_var": 0.2571285872129806,
+      "learning_rate": 0.0001,
+      "loss": 36.79,
+      "loss/crossentropy": 2.900418996810913,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.24839098751544952,
+      "loss/reg": 33.64120101928711,
+      "step": 985
+    },
+    {
+      "epoch": 0.02465,
+      "grad_norm": 4.879204750061035,
+      "grad_norm_var": 0.45573044942737867,
+      "learning_rate": 0.0001,
+      "loss": 36.6408,
+      "loss/crossentropy": 2.826655626296997,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2366310954093933,
+      "loss/reg": 33.57756042480469,
+      "step": 986
+    },
+    {
+      "epoch": 0.024675,
+      "grad_norm": 2.7205452919006348,
+      "grad_norm_var": 0.4121069666646612,
+      "learning_rate": 0.0001,
+      "loss": 36.6661,
+      "loss/crossentropy": 2.88576078414917,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2657344341278076,
+      "loss/reg": 33.51458740234375,
+      "step": 987
+    },
+    {
+      "epoch": 0.0247,
+      "grad_norm": 2.815150737762451,
+      "grad_norm_var": 0.4174920306889968,
+      "learning_rate": 0.0001,
+      "loss": 36.4976,
+      "loss/crossentropy": 2.8254778385162354,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22087544202804565,
+      "loss/reg": 33.45124435424805,
+      "step": 988
+    },
+    {
+      "epoch": 0.024725,
+      "grad_norm": 2.422433614730835,
+      "grad_norm_var": 0.4296126043076429,
+      "learning_rate": 0.0001,
+      "loss": 36.3345,
+      "loss/crossentropy": 2.748582124710083,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.19592465460300446,
+      "loss/reg": 33.390045166015625,
+      "step": 989
+    },
+    {
+      "epoch": 0.02475,
+      "grad_norm": 3.05722713470459,
+      "grad_norm_var": 0.42857927278229774,
+      "learning_rate": 0.0001,
+      "loss": 36.6306,
+      "loss/crossentropy": 3.035149574279785,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2665707468986511,
+      "loss/reg": 33.32892990112305,
+      "step": 990
+    },
+    {
+      "epoch": 0.024775,
+      "grad_norm": 2.711097240447998,
+      "grad_norm_var": 0.42218565465504604,
+      "learning_rate": 0.0001,
+      "loss": 36.1321,
+      "loss/crossentropy": 2.623349189758301,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.23991812765598297,
+      "loss/reg": 33.26885986328125,
+      "step": 991
+    },
+    {
+      "epoch": 0.0248,
+      "grad_norm": 2.7607216835021973,
+      "grad_norm_var": 0.4171327516174946,
+      "learning_rate": 0.0001,
+      "loss": 36.2218,
+      "loss/crossentropy": 2.785637140274048,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.22857247292995453,
+      "loss/reg": 33.20759201049805,
+      "step": 992
+    },
+    {
+      "epoch": 0.024825,
+      "grad_norm": 2.80138897895813,
+      "grad_norm_var": 0.4170671328116583,
+      "learning_rate": 0.0001,
+      "loss": 36.5573,
+      "loss/crossentropy": 3.1539735794067383,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2574368119239807,
+      "loss/reg": 33.14588165283203,
+      "step": 993
+    },
+    {
+      "epoch": 0.02485,
+      "grad_norm": 2.9705312252044678,
+      "grad_norm_var": 0.3229893124537945,
+      "learning_rate": 0.0001,
+      "loss": 36.1044,
+      "loss/crossentropy": 2.742678165435791,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.27794939279556274,
+      "loss/reg": 33.08378219604492,
+      "step": 994
+    },
+    {
+      "epoch": 0.024875,
+      "grad_norm": 3.338322401046753,
+      "grad_norm_var": 0.3264354213345731,
+      "learning_rate": 0.0001,
+      "loss": 36.47,
+      "loss/crossentropy": 3.1671180725097656,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2787696421146393,
+      "loss/reg": 33.024147033691406,
+      "step": 995
+    },
+    {
+      "epoch": 0.0249,
+      "grad_norm": 2.7873196601867676,
+      "grad_norm_var": 0.3030776384523582,
+      "learning_rate": 0.0001,
+      "loss": 36.0396,
+      "loss/crossentropy": 2.798398733139038,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2761346995830536,
+      "loss/reg": 32.96507263183594,
+      "step": 996
+    },
+    {
+      "epoch": 0.024925,
+      "grad_norm": 2.831047534942627,
+      "grad_norm_var": 0.3043818652657305,
+      "learning_rate": 0.0001,
+      "loss": 35.9753,
+      "loss/crossentropy": 2.8229103088378906,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2463880479335785,
+      "loss/reg": 32.90603256225586,
+      "step": 997
+    },
+    {
+      "epoch": 0.02495,
+      "grad_norm": 2.631510019302368,
+      "grad_norm_var": 0.3085412012750756,
+      "learning_rate": 0.0001,
+      "loss": 35.9105,
+      "loss/crossentropy": 2.8491134643554688,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.21344730257987976,
+      "loss/reg": 32.84790802001953,
+      "step": 998
+    },
+    {
+      "epoch": 0.024975,
+      "grad_norm": 2.7426092624664307,
+      "grad_norm_var": 0.311255195789613,
+      "learning_rate": 0.0001,
+      "loss": 35.8303,
+      "loss/crossentropy": 2.779572010040283,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.2633829116821289,
+      "loss/reg": 32.78731155395508,
+      "step": 999
+    },
+    {
+      "epoch": 0.025,
+      "grad_norm": 2.9469587802886963,
+      "grad_norm_var": 0.30242983856450084,
+      "learning_rate": 0.0001,
+      "loss": 36.044,
+      "loss/crossentropy": 3.054746150970459,
+      "loss/hidden": 0.0,
+      "loss/logits": 0.25968533754348755,
+      "loss/reg": 32.72958755493164,
+      "step": 1000
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 40000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 9223372036854775807,
+  "save_steps": 1000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": true,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 6.455688167424e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}