diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,24019 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.3333333333333333,
+  "eval_steps": 2000,
+  "global_step": 2000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.00016666666666666666,
+      "grad_norm": 43.25,
+      "learning_rate": 0.0001,
+      "loss": 7.897,
+      "loss/crossentropy": 1.8981975317001343,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2279614694416523,
+      "step": 1
+    },
+    {
+      "epoch": 0.0003333333333333333,
+      "grad_norm": 42.25,
+      "learning_rate": 9.999999314610822e-05,
+      "loss": 7.7785,
+      "loss/crossentropy": 1.6111224591732025,
+      "loss/hidden": 3.9296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.332294587045908,
+      "step": 2
+    },
+    {
+      "epoch": 0.0005,
+      "grad_norm": 36.0,
+      "learning_rate": 9.999997258443473e-05,
+      "loss": 7.8681,
+      "loss/crossentropy": 1.038797065615654,
+      "loss/hidden": 4.09375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20423871837556362,
+      "step": 3
+    },
+    {
+      "epoch": 0.0006666666666666666,
+      "grad_norm": 35.0,
+      "learning_rate": 9.999993831498517e-05,
+      "loss": 7.5829,
+      "loss/crossentropy": 1.2436729669570923,
+      "loss/hidden": 4.01171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19205103814601898,
+      "step": 4
+    },
+    {
+      "epoch": 0.0008333333333333334,
+      "grad_norm": 36.0,
+      "learning_rate": 9.999989033776898e-05,
+      "loss": 7.6967,
+      "loss/crossentropy": 1.8320814073085785,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22012443095445633,
+      "step": 5
+    },
+    {
+      "epoch": 0.001,
+      "grad_norm": 34.5,
+      "learning_rate": 9.999982865279924e-05,
+      "loss": 7.6994,
+      "loss/crossentropy": 2.6133211851119995,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18267794325947762,
+      "step": 6
+    },
+    {
+      "epoch": 0.0011666666666666668,
+      "grad_norm": 40.5,
+      "learning_rate": 9.999975326009292e-05,
+      "loss": 8.0194,
+      "loss/crossentropy": 1.7613287270069122,
+      "loss/hidden": 4.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28519386798143387,
+      "step": 7
+    },
+    {
+      "epoch": 0.0013333333333333333,
+      "grad_norm": 35.25,
+      "learning_rate": 9.999966415967066e-05,
+      "loss": 7.8227,
+      "loss/crossentropy": 2.1208671629428864,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20836012810468674,
+      "step": 8
+    },
+    {
+      "epoch": 0.0015,
+      "grad_norm": 34.25,
+      "learning_rate": 9.999956135155687e-05,
+      "loss": 7.8838,
+      "loss/crossentropy": 1.7976483404636383,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18757568299770355,
+      "step": 9
+    },
+    {
+      "epoch": 0.0016666666666666668,
+      "grad_norm": 34.75,
+      "learning_rate": 9.999944483577981e-05,
+      "loss": 7.9165,
+      "loss/crossentropy": 1.3925501853227615,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1962764672935009,
+      "step": 10
+    },
+    {
+      "epoch": 0.0018333333333333333,
+      "grad_norm": 32.5,
+      "learning_rate": 9.999931461237134e-05,
+      "loss": 7.9337,
+      "loss/crossentropy": 1.7288123071193695,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22744527459144592,
+      "step": 11
+    },
+    {
+      "epoch": 0.002,
+      "grad_norm": 33.5,
+      "learning_rate": 9.999917068136722e-05,
+      "loss": 7.5202,
+      "loss/crossentropy": 1.494637444615364,
+      "loss/hidden": 4.015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2517388202250004,
+      "step": 12
+    },
+    {
+      "epoch": 0.0021666666666666666,
+      "grad_norm": 34.5,
+      "learning_rate": 9.999901304280685e-05,
+      "loss": 7.4411,
+      "loss/crossentropy": 1.8212255239486694,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15588481724262238,
+      "step": 13
+    },
+    {
+      "epoch": 0.0023333333333333335,
+      "grad_norm": 33.25,
+      "learning_rate": 9.999884169673351e-05,
+      "loss": 7.4569,
+      "loss/crossentropy": 1.6355410516262054,
+      "loss/hidden": 4.08984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22944416105747223,
+      "step": 14
+    },
+    {
+      "epoch": 0.0025,
+      "grad_norm": 34.75,
+      "learning_rate": 9.999865664319414e-05,
+      "loss": 7.9158,
+      "loss/crossentropy": 1.8614172041416168,
+      "loss/hidden": 3.7890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25939401239156723,
+      "step": 15
+    },
+    {
+      "epoch": 0.0026666666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 10.78515625,
+      "learning_rate": 9.999845788223949e-05,
+      "loss": 7.7363,
+      "loss/crossentropy": 2.09798863530159,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22896145656704903,
+      "step": 16
+    },
+    {
+      "epoch": 0.0028333333333333335,
+      "grad_norm": 32.5,
+      "grad_norm_var": 7.325,
+      "learning_rate": 9.999824541392405e-05,
+      "loss": 7.6403,
+      "loss/crossentropy": 1.6914259791374207,
+      "loss/hidden": 4.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33019889891147614,
+      "step": 17
+    },
+    {
+      "epoch": 0.003,
+      "grad_norm": 31.0,
+      "grad_norm_var": 4.54765625,
+      "learning_rate": 9.999801923830603e-05,
+      "loss": 7.5091,
+      "loss/crossentropy": 1.7090625166893005,
+      "loss/hidden": 3.94140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25124871358275414,
+      "step": 18
+    },
+    {
+      "epoch": 0.0031666666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 5.603580729166667,
+      "learning_rate": 9.99977793554475e-05,
+      "loss": 7.5787,
+      "loss/crossentropy": 1.4764716178178787,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20044904574751854,
+      "step": 19
+    },
+    {
+      "epoch": 0.0033333333333333335,
+      "grad_norm": 36.25,
+      "grad_norm_var": 5.861393229166667,
+      "learning_rate": 9.999752576541418e-05,
+      "loss": 7.7661,
+      "loss/crossentropy": 1.8880396485328674,
+      "loss/hidden": 3.7265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2577027827501297,
+      "step": 20
+    },
+    {
+      "epoch": 0.0035,
+      "grad_norm": 35.25,
+      "grad_norm_var": 5.708268229166666,
+      "learning_rate": 9.999725846827562e-05,
+      "loss": 7.8717,
+      "loss/crossentropy": 1.2173301130533218,
+      "loss/hidden": 3.921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23028925992548466,
+      "step": 21
+    },
+    {
+      "epoch": 0.0036666666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.9556640625,
+      "learning_rate": 9.999697746410508e-05,
+      "loss": 7.3192,
+      "loss/crossentropy": 1.9408041834831238,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17967402562499046,
+      "step": 22
+    },
+    {
+      "epoch": 0.003833333333333333,
+      "grad_norm": 37.0,
+      "grad_norm_var": 3.6478515625,
+      "learning_rate": 9.99966827529796e-05,
+      "loss": 7.7772,
+      "loss/crossentropy": 1.696288287639618,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2100219465792179,
+      "step": 23
+    },
+    {
+      "epoch": 0.004,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.4863932291666666,
+      "learning_rate": 9.999637433497999e-05,
+      "loss": 7.7023,
+      "loss/crossentropy": 1.5910453498363495,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17898119986057281,
+      "step": 24
+    },
+    {
+      "epoch": 0.004166666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 3.695572916666667,
+      "learning_rate": 9.999605221019081e-05,
+      "loss": 7.3402,
+      "loss/crossentropy": 1.7376164197921753,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18538454920053482,
+      "step": 25
+    },
+    {
+      "epoch": 0.004333333333333333,
+      "grad_norm": 34.5,
+      "grad_norm_var": 3.65625,
+      "learning_rate": 9.999571637870036e-05,
+      "loss": 7.5857,
+      "loss/crossentropy": 1.3886851519346237,
+      "loss/hidden": 3.7890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1782984510064125,
+      "step": 26
+    },
+    {
+      "epoch": 0.0045,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.59375,
+      "learning_rate": 9.99953668406007e-05,
+      "loss": 7.6528,
+      "loss/crossentropy": 2.2225378453731537,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19338875636458397,
+      "step": 27
+    },
+    {
+      "epoch": 0.004666666666666667,
+      "grad_norm": 34.5,
+      "grad_norm_var": 3.65625,
+      "learning_rate": 9.999500359598768e-05,
+      "loss": 7.7959,
+      "loss/crossentropy": 1.7327675968408585,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21430451981723309,
+      "step": 28
+    },
+    {
+      "epoch": 0.004833333333333334,
+      "grad_norm": 33.75,
+      "grad_norm_var": 3.59765625,
+      "learning_rate": 9.999462664496088e-05,
+      "loss": 7.6499,
+      "loss/crossentropy": 1.8107008635997772,
+      "loss/hidden": 4.0859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28763537108898163,
+      "step": 29
+    },
+    {
+      "epoch": 0.005,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.6104166666666666,
+      "learning_rate": 9.999423598762363e-05,
+      "loss": 7.6058,
+      "loss/crossentropy": 1.9658692479133606,
+      "loss/hidden": 3.75,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2879307046532631,
+      "step": 30
+    },
+    {
+      "epoch": 0.005166666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.542708333333333,
+      "learning_rate": 9.999383162408304e-05,
+      "loss": 7.4496,
+      "loss/crossentropy": 1.8217417299747467,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19359543919563293,
+      "step": 31
+    },
+    {
+      "epoch": 0.005333333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.542708333333333,
+      "learning_rate": 9.999341355444995e-05,
+      "loss": 7.6384,
+      "loss/crossentropy": 2.139420449733734,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1821003034710884,
+      "step": 32
+    },
+    {
+      "epoch": 0.0055,
+      "grad_norm": 34.5,
+      "grad_norm_var": 3.534375,
+      "learning_rate": 9.999298177883903e-05,
+      "loss": 7.8442,
+      "loss/crossentropy": 1.346135452389717,
+      "loss/hidden": 3.921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2800942696630955,
+      "step": 33
+    },
+    {
+      "epoch": 0.005666666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.0927083333333334,
+      "learning_rate": 9.99925362973686e-05,
+      "loss": 7.5856,
+      "loss/crossentropy": 1.7728485763072968,
+      "loss/hidden": 4.1171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27092037349939346,
+      "step": 34
+    },
+    {
+      "epoch": 0.005833333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.074934895833333,
+      "learning_rate": 9.999207711016081e-05,
+      "loss": 8.1591,
+      "loss/crossentropy": 2.1373378336429596,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28596626967191696,
+      "step": 35
+    },
+    {
+      "epoch": 0.006,
+      "grad_norm": 36.5,
+      "grad_norm_var": 2.1561848958333334,
+      "learning_rate": 9.999160421734155e-05,
+      "loss": 7.711,
+      "loss/crossentropy": 1.7681190073490143,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18538705632090569,
+      "step": 36
+    },
+    {
+      "epoch": 0.006166666666666667,
+      "grad_norm": 35.25,
+      "grad_norm_var": 2.1561848958333334,
+      "learning_rate": 9.999111761904046e-05,
+      "loss": 7.6903,
+      "loss/crossentropy": 1.7745196521282196,
+      "loss/hidden": 4.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4391864165663719,
+      "step": 37
+    },
+    {
+      "epoch": 0.006333333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.0952473958333333,
+      "learning_rate": 9.999061731539094e-05,
+      "loss": 7.5037,
+      "loss/crossentropy": 1.6296418905258179,
+      "loss/hidden": 3.875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22025315463542938,
+      "step": 38
+    },
+    {
+      "epoch": 0.0065,
+      "grad_norm": 34.75,
+      "grad_norm_var": 1.4999348958333334,
+      "learning_rate": 9.999010330653018e-05,
+      "loss": 7.8698,
+      "loss/crossentropy": 1.5856382250785828,
+      "loss/hidden": 3.91015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2322024367749691,
+      "step": 39
+    },
+    {
+      "epoch": 0.006666666666666667,
+      "grad_norm": 35.25,
+      "grad_norm_var": 1.6275390625,
+      "learning_rate": 9.998957559259906e-05,
+      "loss": 7.7027,
+      "loss/crossentropy": 1.5911406129598618,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2131226249039173,
+      "step": 40
+    },
+    {
+      "epoch": 0.006833333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.6275390625,
+      "learning_rate": 9.998903417374228e-05,
+      "loss": 7.8013,
+      "loss/crossentropy": 2.2657040655612946,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17953365668654442,
+      "step": 41
+    },
+    {
+      "epoch": 0.007,
+      "grad_norm": 34.25,
+      "grad_norm_var": 1.6113932291666666,
+      "learning_rate": 9.998847905010826e-05,
+      "loss": 7.4123,
+      "loss/crossentropy": 1.5879982709884644,
+      "loss/hidden": 3.796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19292457774281502,
+      "step": 42
+    },
+    {
+      "epoch": 0.007166666666666667,
+      "grad_norm": 35.0,
+      "grad_norm_var": 1.6754557291666667,
+      "learning_rate": 9.998791022184922e-05,
+      "loss": 7.5851,
+      "loss/crossentropy": 1.9873057901859283,
+      "loss/hidden": 3.87890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2376515343785286,
+      "step": 43
+    },
+    {
+      "epoch": 0.007333333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.7113932291666667,
+      "learning_rate": 9.998732768912104e-05,
+      "loss": 7.2571,
+      "loss/crossentropy": 1.7832402884960175,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17459676414728165,
+      "step": 44
+    },
+    {
+      "epoch": 0.0075,
+      "grad_norm": 33.5,
+      "grad_norm_var": 1.7197265625,
+      "learning_rate": 9.99867314520835e-05,
+      "loss": 7.3922,
+      "loss/crossentropy": 2.2741356790065765,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1955919899046421,
+      "step": 45
+    },
+    {
+      "epoch": 0.007666666666666666,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.15625,
+      "learning_rate": 9.998612151090003e-05,
+      "loss": 7.2748,
+      "loss/crossentropy": 1.3483582437038422,
+      "loss/hidden": 3.9609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2215445153415203,
+      "step": 46
+    },
+    {
+      "epoch": 0.007833333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.8363932291666667,
+      "learning_rate": 9.998549786573785e-05,
+      "loss": 7.3114,
+      "loss/crossentropy": 1.8974584341049194,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1681991945952177,
+      "step": 47
+    },
+    {
+      "epoch": 0.008,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.770768229166667,
+      "learning_rate": 9.998486051676792e-05,
+      "loss": 7.3338,
+      "loss/crossentropy": 1.4651325196027756,
+      "loss/hidden": 3.91015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21610706113278866,
+      "step": 48
+    },
+    {
+      "epoch": 0.008166666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.768684895833333,
+      "learning_rate": 9.9984209464165e-05,
+      "loss": 7.5725,
+      "loss/crossentropy": 1.9949549734592438,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15607954934239388,
+      "step": 49
+    },
+    {
+      "epoch": 0.008333333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.9379557291666667,
+      "learning_rate": 9.998354470810757e-05,
+      "loss": 7.2875,
+      "loss/crossentropy": 1.897917628288269,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19233481213450432,
+      "step": 50
+    },
+    {
+      "epoch": 0.0085,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.9379557291666667,
+      "learning_rate": 9.998286624877786e-05,
+      "loss": 7.9821,
+      "loss/crossentropy": 2.4633734226226807,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.201544351875782,
+      "step": 51
+    },
+    {
+      "epoch": 0.008666666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.4567057291666665,
+      "learning_rate": 9.99821740863619e-05,
+      "loss": 7.7769,
+      "loss/crossentropy": 1.9304746389389038,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2369798980653286,
+      "step": 52
+    },
+    {
+      "epoch": 0.008833333333333334,
+      "grad_norm": 34.0,
+      "grad_norm_var": 2.1832682291666665,
+      "learning_rate": 9.998146822104943e-05,
+      "loss": 7.3161,
+      "loss/crossentropy": 1.5423674881458282,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24242010712623596,
+      "step": 53
+    },
+    {
+      "epoch": 0.009,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.184830729166667,
+      "learning_rate": 9.998074865303399e-05,
+      "loss": 7.7667,
+      "loss/crossentropy": 1.9618560820817947,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15662527084350586,
+      "step": 54
+    },
+    {
+      "epoch": 0.009166666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.126822916666667,
+      "learning_rate": 9.998001538251282e-05,
+      "loss": 7.3556,
+      "loss/crossentropy": 2.1333842277526855,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19949736073613167,
+      "step": 55
+    },
+    {
+      "epoch": 0.009333333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.154166666666667,
+      "learning_rate": 9.997926840968699e-05,
+      "loss": 7.3285,
+      "loss/crossentropy": 1.6652204990386963,
+      "loss/hidden": 3.66015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18719187378883362,
+      "step": 56
+    },
+    {
+      "epoch": 0.0095,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.155143229166667,
+      "learning_rate": 9.997850773476126e-05,
+      "loss": 7.1769,
+      "loss/crossentropy": 1.5638276934623718,
+      "loss/hidden": 3.9609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20844870433211327,
+      "step": 57
+    },
+    {
+      "epoch": 0.009666666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 2.104622395833333,
+      "learning_rate": 9.997773335794416e-05,
+      "loss": 7.5886,
+      "loss/crossentropy": 1.6454956978559494,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2049052193760872,
+      "step": 58
+    },
+    {
+      "epoch": 0.009833333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.7666666666666666,
+      "learning_rate": 9.997694527944803e-05,
+      "loss": 7.3413,
+      "loss/crossentropy": 1.7107558101415634,
+      "loss/hidden": 4.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1903497278690338,
+      "step": 59
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.79140625,
+      "learning_rate": 9.99761434994889e-05,
+      "loss": 7.5478,
+      "loss/crossentropy": 1.7333263158798218,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15633369609713554,
+      "step": 60
+    },
+    {
+      "epoch": 0.010166666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.71015625,
+      "learning_rate": 9.997532801828658e-05,
+      "loss": 7.8502,
+      "loss/crossentropy": 1.5546621978282928,
+      "loss/hidden": 4.078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29293418303132057,
+      "step": 61
+    },
+    {
+      "epoch": 0.010333333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.6603515625,
+      "learning_rate": 9.997449883606466e-05,
+      "loss": 7.699,
+      "loss/crossentropy": 1.9417144358158112,
+      "loss/hidden": 4.078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33094572275877,
+      "step": 62
+    },
+    {
+      "epoch": 0.0105,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.6270833333333334,
+      "learning_rate": 9.997365595305044e-05,
+      "loss": 7.655,
+      "loss/crossentropy": 1.8994901180267334,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19589544087648392,
+      "step": 63
+    },
+    {
+      "epoch": 0.010666666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.5885416666666667,
+      "learning_rate": 9.997279936947502e-05,
+      "loss": 7.9396,
+      "loss/crossentropy": 1.4546025916934013,
+      "loss/hidden": 3.95703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2566604632884264,
+      "step": 64
+    },
+    {
+      "epoch": 0.010833333333333334,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.6051432291666667,
+      "learning_rate": 9.997192908557323e-05,
+      "loss": 7.5829,
+      "loss/crossentropy": 1.7165934145450592,
+      "loss/hidden": 3.80078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3070928454399109,
+      "step": 65
+    },
+    {
+      "epoch": 0.011,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.5785807291666667,
+      "learning_rate": 9.997104510158365e-05,
+      "loss": 7.832,
+      "loss/crossentropy": 1.4161300361156464,
+      "loss/hidden": 4.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2050262913107872,
+      "step": 66
+    },
+    {
+      "epoch": 0.011166666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.5296223958333333,
+      "learning_rate": 9.997014741774866e-05,
+      "loss": 7.3665,
+      "loss/crossentropy": 1.6561641842126846,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13321572355926037,
+      "step": 67
+    },
+    {
+      "epoch": 0.011333333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.4546223958333333,
+      "learning_rate": 9.996923603431433e-05,
+      "loss": 7.3003,
+      "loss/crossentropy": 1.888094425201416,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15798387117683887,
+      "step": 68
+    },
+    {
+      "epoch": 0.0115,
+      "grad_norm": 33.5,
+      "grad_norm_var": 1.3634765625,
+      "learning_rate": 9.996831095153055e-05,
+      "loss": 7.5084,
+      "loss/crossentropy": 1.8267690539360046,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13853692263364792,
+      "step": 69
+    },
+    {
+      "epoch": 0.011666666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.2650390625,
+      "learning_rate": 9.996737216965092e-05,
+      "loss": 7.5366,
+      "loss/crossentropy": 1.8325877487659454,
+      "loss/hidden": 3.91015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24080873653292656,
+      "step": 70
+    },
+    {
+      "epoch": 0.011833333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 1.39765625,
+      "learning_rate": 9.996641968893282e-05,
+      "loss": 7.563,
+      "loss/crossentropy": 2.386824905872345,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20230185240507126,
+      "step": 71
+    },
+    {
+      "epoch": 0.012,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.01015625,
+      "learning_rate": 9.996545350963738e-05,
+      "loss": 7.9256,
+      "loss/crossentropy": 1.899581179022789,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17962408438324928,
+      "step": 72
+    },
+    {
+      "epoch": 0.012166666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 0.946875,
+      "learning_rate": 9.996447363202946e-05,
+      "loss": 7.4534,
+      "loss/crossentropy": 1.597171574831009,
+      "loss/hidden": 3.7421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22687402367591858,
+      "step": 73
+    },
+    {
+      "epoch": 0.012333333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 0.8997395833333334,
+      "learning_rate": 9.996348005637775e-05,
+      "loss": 7.5359,
+      "loss/crossentropy": 1.8461322486400604,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18764278665184975,
+      "step": 74
+    },
+    {
+      "epoch": 0.0125,
+      "grad_norm": 32.5,
+      "grad_norm_var": 0.8171223958333333,
+      "learning_rate": 9.996247278295458e-05,
+      "loss": 7.8896,
+      "loss/crossentropy": 1.44264367967844,
+      "loss/hidden": 4.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.44364645332098007,
+      "step": 75
+    },
+    {
+      "epoch": 0.012666666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 0.8247395833333333,
+      "learning_rate": 9.996145181203615e-05,
+      "loss": 7.2413,
+      "loss/crossentropy": 1.9038055539131165,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2209215946495533,
+      "step": 76
+    },
+    {
+      "epoch": 0.012833333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.0202473958333333,
+      "learning_rate": 9.996041714390235e-05,
+      "loss": 7.6161,
+      "loss/crossentropy": 1.6209959387779236,
+      "loss/hidden": 4.03125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21705806627869606,
+      "step": 77
+    },
+    {
+      "epoch": 0.013,
+      "grad_norm": 31.875,
+      "grad_norm_var": 0.9708333333333333,
+      "learning_rate": 9.995936877883682e-05,
+      "loss": 7.5658,
+      "loss/crossentropy": 2.0096256136894226,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1687867697328329,
+      "step": 78
+    },
+    {
+      "epoch": 0.013166666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 0.75390625,
+      "learning_rate": 9.9958306717127e-05,
+      "loss": 7.5281,
+      "loss/crossentropy": 1.5593473613262177,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18466703221201897,
+      "step": 79
+    },
+    {
+      "epoch": 0.013333333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 0.7447916666666666,
+      "learning_rate": 9.995723095906407e-05,
+      "loss": 7.3577,
+      "loss/crossentropy": 1.8190798908472061,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17776387929916382,
+      "step": 80
+    },
+    {
+      "epoch": 0.0135,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.6676432291666667,
+      "learning_rate": 9.995614150494293e-05,
+      "loss": 7.3202,
+      "loss/crossentropy": 1.827375441789627,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16521281003952026,
+      "step": 81
+    },
+    {
+      "epoch": 0.013666666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.6593098958333334,
+      "learning_rate": 9.995503835506226e-05,
+      "loss": 7.5337,
+      "loss/crossentropy": 1.511421576142311,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18143922463059425,
+      "step": 82
+    },
+    {
+      "epoch": 0.013833333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.7077473958333333,
+      "learning_rate": 9.995392150972451e-05,
+      "loss": 7.6254,
+      "loss/crossentropy": 1.9109731316566467,
+      "loss/hidden": 4.05078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21341488882899284,
+      "step": 83
+    },
+    {
+      "epoch": 0.014,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.7247395833333334,
+      "learning_rate": 9.995279096923585e-05,
+      "loss": 7.4453,
+      "loss/crossentropy": 2.3015710711479187,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21305473893880844,
+      "step": 84
+    },
+    {
+      "epoch": 0.014166666666666666,
+      "grad_norm": 33.75,
+      "grad_norm_var": 1.7739583333333333,
+      "learning_rate": 9.995164673390625e-05,
+      "loss": 7.5234,
+      "loss/crossentropy": 2.2502674758434296,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2330060638487339,
+      "step": 85
+    },
+    {
+      "epoch": 0.014333333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.9239583333333334,
+      "learning_rate": 9.995048880404938e-05,
+      "loss": 7.4712,
+      "loss/crossentropy": 1.7803797721862793,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21768440306186676,
+      "step": 86
+    },
+    {
+      "epoch": 0.0145,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.7322916666666666,
+      "learning_rate": 9.994931717998272e-05,
+      "loss": 7.5616,
+      "loss/crossentropy": 1.902416467666626,
+      "loss/hidden": 3.74609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2145845927298069,
+      "step": 87
+    },
+    {
+      "epoch": 0.014666666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.73515625,
+      "learning_rate": 9.994813186202747e-05,
+      "loss": 7.5587,
+      "loss/crossentropy": 1.828644409775734,
+      "loss/hidden": 3.6484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22059463895857334,
+      "step": 88
+    },
+    {
+      "epoch": 0.014833333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.79765625,
+      "learning_rate": 9.994693285050857e-05,
+      "loss": 7.1863,
+      "loss/crossentropy": 1.3599056899547577,
+      "loss/hidden": 3.85546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15615848824381828,
+      "step": 89
+    },
+    {
+      "epoch": 0.015,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.69140625,
+      "learning_rate": 9.994572014575476e-05,
+      "loss": 8.0386,
+      "loss/crossentropy": 2.1659975349903107,
+      "loss/hidden": 3.6875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2879357635974884,
+      "step": 90
+    },
+    {
+      "epoch": 0.015166666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 1.95390625,
+      "learning_rate": 9.994449374809851e-05,
+      "loss": 7.5539,
+      "loss/crossentropy": 1.7670844793319702,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21495118737220764,
+      "step": 91
+    },
+    {
+      "epoch": 0.015333333333333332,
+      "grad_norm": 35.0,
+      "grad_norm_var": 2.5041015625,
+      "learning_rate": 9.994325365787602e-05,
+      "loss": 7.8947,
+      "loss/crossentropy": 1.9785922467708588,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19296985119581223,
+      "step": 92
+    },
+    {
+      "epoch": 0.0155,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.34765625,
+      "learning_rate": 9.99419998754273e-05,
+      "loss": 7.5811,
+      "loss/crossentropy": 2.220028728246689,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1642632707953453,
+      "step": 93
+    },
+    {
+      "epoch": 0.015666666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.3421223958333335,
+      "learning_rate": 9.994073240109606e-05,
+      "loss": 7.2548,
+      "loss/crossentropy": 1.7792484760284424,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19576582685112953,
+      "step": 94
+    },
+    {
+      "epoch": 0.015833333333333335,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.401497395833333,
+      "learning_rate": 9.993945123522978e-05,
+      "loss": 7.9,
+      "loss/crossentropy": 1.4793440699577332,
+      "loss/hidden": 3.98046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24769844487309456,
+      "step": 95
+    },
+    {
+      "epoch": 0.016,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.4202473958333335,
+      "learning_rate": 9.993815637817974e-05,
+      "loss": 7.381,
+      "loss/crossentropy": 1.5669107139110565,
+      "loss/hidden": 3.97265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2881624363362789,
+      "step": 96
+    },
+    {
+      "epoch": 0.016166666666666666,
+      "grad_norm": 35.25,
+      "grad_norm_var": 1.8577473958333333,
+      "learning_rate": 9.993684783030088e-05,
+      "loss": 7.6191,
+      "loss/crossentropy": 1.7162407040596008,
+      "loss/hidden": 3.9140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23535295203328133,
+      "step": 97
+    },
+    {
+      "epoch": 0.01633333333333333,
+      "grad_norm": 35.5,
+      "grad_norm_var": 2.3181640625,
+      "learning_rate": 9.993552559195197e-05,
+      "loss": 7.6683,
+      "loss/crossentropy": 1.6846329420804977,
+      "loss/hidden": 3.9453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30070426501333714,
+      "step": 98
+    },
+    {
+      "epoch": 0.0165,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.3181640625,
+      "learning_rate": 9.993418966349552e-05,
+      "loss": 7.5142,
+      "loss/crossentropy": 1.3872482031583786,
+      "loss/hidden": 4.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20113259553909302,
+      "step": 99
+    },
+    {
+      "epoch": 0.016666666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.8103515625,
+      "learning_rate": 9.993284004529775e-05,
+      "loss": 7.3211,
+      "loss/crossentropy": 1.6917027533054352,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15302151814103127,
+      "step": 100
+    },
+    {
+      "epoch": 0.016833333333333332,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.9947265625,
+      "learning_rate": 9.99314767377287e-05,
+      "loss": 7.4487,
+      "loss/crossentropy": 2.0142119228839874,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2277221530675888,
+      "step": 101
+    },
+    {
+      "epoch": 0.017,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.7119140625,
+      "learning_rate": 9.993009974116211e-05,
+      "loss": 7.2728,
+      "loss/crossentropy": 2.0172336399555206,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15139513090252876,
+      "step": 102
+    },
+    {
+      "epoch": 0.017166666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.7119140625,
+      "learning_rate": 9.992870905597548e-05,
+      "loss": 7.473,
+      "loss/crossentropy": 1.7352718710899353,
+      "loss/hidden": 3.83203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2756512500345707,
+      "step": 103
+    },
+    {
+      "epoch": 0.017333333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.7931640625,
+      "learning_rate": 9.992730468255011e-05,
+      "loss": 7.3698,
+      "loss/crossentropy": 1.532135546207428,
+      "loss/hidden": 3.72265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1990617997944355,
+      "step": 104
+    },
+    {
+      "epoch": 0.0175,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.7775390625,
+      "learning_rate": 9.9925886621271e-05,
+      "loss": 7.4661,
+      "loss/crossentropy": 2.0795831382274628,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19490709155797958,
+      "step": 105
+    },
+    {
+      "epoch": 0.017666666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.69375,
+      "learning_rate": 9.992445487252691e-05,
+      "loss": 7.528,
+      "loss/crossentropy": 1.4112937152385712,
+      "loss/hidden": 4.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22824537754058838,
+      "step": 106
+    },
+    {
+      "epoch": 0.017833333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.6177083333333333,
+      "learning_rate": 9.992300943671036e-05,
+      "loss": 7.7207,
+      "loss/crossentropy": 2.027792990207672,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22932051867246628,
+      "step": 107
+    },
+    {
+      "epoch": 0.018,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.334375,
+      "learning_rate": 9.992155031421764e-05,
+      "loss": 7.5793,
+      "loss/crossentropy": 1.7205888107419014,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1724705696105957,
+      "step": 108
+    },
+    {
+      "epoch": 0.018166666666666668,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.3872395833333333,
+      "learning_rate": 9.992007750544876e-05,
+      "loss": 8.0001,
+      "loss/crossentropy": 1.776763528585434,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17769636027514935,
+      "step": 109
+    },
+    {
+      "epoch": 0.018333333333333333,
+      "grad_norm": 35.0,
+      "grad_norm_var": 2.7684895833333334,
+      "learning_rate": 9.991859101080751e-05,
+      "loss": 7.4431,
+      "loss/crossentropy": 1.863470584154129,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21127909794449806,
+      "step": 110
+    },
+    {
+      "epoch": 0.0185,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.9280598958333335,
+      "learning_rate": 9.991709083070143e-05,
+      "loss": 7.3494,
+      "loss/crossentropy": 1.776065930724144,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17032187432050705,
+      "step": 111
+    },
+    {
+      "epoch": 0.018666666666666668,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.9072265625,
+      "learning_rate": 9.991557696554177e-05,
+      "loss": 7.6787,
+      "loss/crossentropy": 1.9118266999721527,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19897421821951866,
+      "step": 112
+    },
+    {
+      "epoch": 0.018833333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.4009765625,
+      "learning_rate": 9.991404941574361e-05,
+      "loss": 7.5492,
+      "loss/crossentropy": 2.13837730884552,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24482352659106255,
+      "step": 113
+    },
+    {
+      "epoch": 0.019,
+      "grad_norm": 34.0,
+      "grad_norm_var": 1.9150390625,
+      "learning_rate": 9.99125081817257e-05,
+      "loss": 7.4473,
+      "loss/crossentropy": 1.9174355566501617,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25030214712023735,
+      "step": 114
+    },
+    {
+      "epoch": 0.019166666666666665,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.0134765625,
+      "learning_rate": 9.99109532639106e-05,
+      "loss": 7.5955,
+      "loss/crossentropy": 1.432920753955841,
+      "loss/hidden": 4.02734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1886458545923233,
+      "step": 115
+    },
+    {
+      "epoch": 0.019333333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.6893229166666666,
+      "learning_rate": 9.990938466272459e-05,
+      "loss": 7.4896,
+      "loss/crossentropy": 1.4262737929821014,
+      "loss/hidden": 3.96484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22360584139823914,
+      "step": 116
+    },
+    {
+      "epoch": 0.0195,
+      "grad_norm": 34.75,
+      "grad_norm_var": 1.89765625,
+      "learning_rate": 9.990780237859769e-05,
+      "loss": 8.213,
+      "loss/crossentropy": 1.6001249551773071,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23247257247567177,
+      "step": 117
+    },
+    {
+      "epoch": 0.019666666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.9145833333333333,
+      "learning_rate": 9.990620641196374e-05,
+      "loss": 7.7368,
+      "loss/crossentropy": 1.6282226294279099,
+      "loss/hidden": 3.90625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24213666282594204,
+      "step": 118
+    },
+    {
+      "epoch": 0.019833333333333335,
+      "grad_norm": 39.5,
+      "grad_norm_var": 4.934375,
+      "learning_rate": 9.990459676326024e-05,
+      "loss": 7.5643,
+      "loss/crossentropy": 2.126531273126602,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16531133279204369,
+      "step": 119
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 31.625,
+      "grad_norm_var": 4.910872395833334,
+      "learning_rate": 9.990297343292851e-05,
+      "loss": 7.5654,
+      "loss/crossentropy": 1.669142171740532,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2322467900812626,
+      "step": 120
+    },
+    {
+      "epoch": 0.020166666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.890559895833333,
+      "learning_rate": 9.990133642141359e-05,
+      "loss": 7.6424,
+      "loss/crossentropy": 1.842205137014389,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2020024061203003,
+      "step": 121
+    },
+    {
+      "epoch": 0.02033333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.805989583333333,
+      "learning_rate": 9.989968572916426e-05,
+      "loss": 7.4694,
+      "loss/crossentropy": 1.8283292353153229,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1664537452161312,
+      "step": 122
+    },
+    {
+      "epoch": 0.0205,
+      "grad_norm": 31.75,
+      "grad_norm_var": 4.884375,
+      "learning_rate": 9.989802135663308e-05,
+      "loss": 7.1245,
+      "loss/crossentropy": 1.530500888824463,
+      "loss/hidden": 3.94140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17333029955625534,
+      "step": 123
+    },
+    {
+      "epoch": 0.020666666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.626041666666667,
+      "learning_rate": 9.989634330427636e-05,
+      "loss": 7.4663,
+      "loss/crossentropy": 1.8538174033164978,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27487967163324356,
+      "step": 124
+    },
+    {
+      "epoch": 0.020833333333333332,
+      "grad_norm": 35.75,
+      "grad_norm_var": 5.083072916666667,
+      "learning_rate": 9.989465157255412e-05,
+      "loss": 7.7927,
+      "loss/crossentropy": 1.7388005554676056,
+      "loss/hidden": 3.99609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2771507576107979,
+      "step": 125
+    },
+    {
+      "epoch": 0.021,
+      "grad_norm": 37.0,
+      "grad_norm_var": 5.820572916666666,
+      "learning_rate": 9.989294616193017e-05,
+      "loss": 7.6527,
+      "loss/crossentropy": 1.881695032119751,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19622422382235527,
+      "step": 126
+    },
+    {
+      "epoch": 0.021166666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 5.950455729166666,
+      "learning_rate": 9.989122707287208e-05,
+      "loss": 7.5189,
+      "loss/crossentropy": 2.330846667289734,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2320748008787632,
+      "step": 127
+    },
+    {
+      "epoch": 0.021333333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.881184895833333,
+      "learning_rate": 9.988949430585111e-05,
+      "loss": 7.9233,
+      "loss/crossentropy": 1.7099304348230362,
+      "loss/hidden": 3.9296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2644583098590374,
+      "step": 128
+    },
+    {
+      "epoch": 0.0215,
+      "grad_norm": 29.375,
+      "grad_norm_var": 6.849739583333333,
+      "learning_rate": 9.988774786134234e-05,
+      "loss": 7.3151,
+      "loss/crossentropy": 1.8030900359153748,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22125136107206345,
+      "step": 129
+    },
+    {
+      "epoch": 0.021666666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 6.79375,
+      "learning_rate": 9.988598773982454e-05,
+      "loss": 7.9058,
+      "loss/crossentropy": 2.1423310041427612,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18011631816625595,
+      "step": 130
+    },
+    {
+      "epoch": 0.021833333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 6.5791015625,
+      "learning_rate": 9.988421394178027e-05,
+      "loss": 7.3133,
+      "loss/crossentropy": 1.898929089307785,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2054540552198887,
+      "step": 131
+    },
+    {
+      "epoch": 0.022,
+      "grad_norm": 33.75,
+      "grad_norm_var": 6.4291015625,
+      "learning_rate": 9.988242646769584e-05,
+      "loss": 7.1352,
+      "loss/crossentropy": 1.5720456838607788,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16385111585259438,
+      "step": 132
+    },
+    {
+      "epoch": 0.022166666666666668,
+      "grad_norm": 32.5,
+      "grad_norm_var": 6.2791015625,
+      "learning_rate": 9.988062531806126e-05,
+      "loss": 7.5522,
+      "loss/crossentropy": 1.3341676890850067,
+      "loss/hidden": 4.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23463992774486542,
+      "step": 133
+    },
+    {
+      "epoch": 0.022333333333333334,
+      "grad_norm": 35.25,
+      "grad_norm_var": 6.581184895833333,
+      "learning_rate": 9.987881049337037e-05,
+      "loss": 7.9398,
+      "loss/crossentropy": 1.5141830444335938,
+      "loss/hidden": 4.03125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30013222992420197,
+      "step": 134
+    },
+    {
+      "epoch": 0.0225,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.8369140625,
+      "learning_rate": 9.98769819941207e-05,
+      "loss": 7.3277,
+      "loss/crossentropy": 1.5133470296859741,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20712239667773247,
+      "step": 135
+    },
+    {
+      "epoch": 0.02266666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.75390625,
+      "learning_rate": 9.987513982081351e-05,
+      "loss": 7.6822,
+      "loss/crossentropy": 1.5610093474388123,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17235243692994118,
+      "step": 136
+    },
+    {
+      "epoch": 0.022833333333333334,
+      "grad_norm": 36.25,
+      "grad_norm_var": 4.40625,
+      "learning_rate": 9.987328397395387e-05,
+      "loss": 7.2063,
+      "loss/crossentropy": 1.771727591753006,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18995914980769157,
+      "step": 137
+    },
+    {
+      "epoch": 0.023,
+      "grad_norm": 31.875,
+      "grad_norm_var": 4.3306640625,
+      "learning_rate": 9.98714144540506e-05,
+      "loss": 7.7502,
+      "loss/crossentropy": 1.9235819280147552,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17999661713838577,
+      "step": 138
+    },
+    {
+      "epoch": 0.023166666666666665,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.1791015625,
+      "learning_rate": 9.986953126161619e-05,
+      "loss": 7.6574,
+      "loss/crossentropy": 1.362782821059227,
+      "loss/hidden": 3.85546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25264550372958183,
+      "step": 139
+    },
+    {
+      "epoch": 0.023333333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.54765625,
+      "learning_rate": 9.986763439716696e-05,
+      "loss": 7.56,
+      "loss/crossentropy": 1.556288480758667,
+      "loss/hidden": 3.78515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2722315713763237,
+      "step": 140
+    },
+    {
+      "epoch": 0.0235,
+      "grad_norm": 31.25,
+      "grad_norm_var": 4.26640625,
+      "learning_rate": 9.986572386122291e-05,
+      "loss": 7.703,
+      "loss/crossentropy": 2.008838027715683,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2090667188167572,
+      "step": 141
+    },
+    {
+      "epoch": 0.023666666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.066666666666667,
+      "learning_rate": 9.986379965430786e-05,
+      "loss": 7.5106,
+      "loss/crossentropy": 2.2713093757629395,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22611282020807266,
+      "step": 142
+    },
+    {
+      "epoch": 0.023833333333333335,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.845833333333333,
+      "learning_rate": 9.986186177694933e-05,
+      "loss": 7.4256,
+      "loss/crossentropy": 1.7764946222305298,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21423206105828285,
+      "step": 143
+    },
+    {
+      "epoch": 0.024,
+      "grad_norm": 36.75,
+      "grad_norm_var": 3.8684895833333335,
+      "learning_rate": 9.98599102296786e-05,
+      "loss": 7.7863,
+      "loss/crossentropy": 1.7998189330101013,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19024800322949886,
+      "step": 144
+    },
+    {
+      "epoch": 0.024166666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.1936848958333335,
+      "learning_rate": 9.98579450130307e-05,
+      "loss": 7.3844,
+      "loss/crossentropy": 2.0455740094184875,
+      "loss/hidden": 3.71875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23831216618418694,
+      "step": 145
+    },
+    {
+      "epoch": 0.024333333333333332,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.270247395833333,
+      "learning_rate": 9.985596612754439e-05,
+      "loss": 7.4585,
+      "loss/crossentropy": 1.2537191361188889,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14199252426624298,
+      "step": 146
+    },
+    {
+      "epoch": 0.0245,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.2455729166666667,
+      "learning_rate": 9.985397357376222e-05,
+      "loss": 7.57,
+      "loss/crossentropy": 1.9460046589374542,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17490505427122116,
+      "step": 147
+    },
+    {
+      "epoch": 0.024666666666666667,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.3207682291666667,
+      "learning_rate": 9.985196735223045e-05,
+      "loss": 7.6974,
+      "loss/crossentropy": 1.9865233302116394,
+      "loss/hidden": 3.83203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22223564982414246,
+      "step": 148
+    },
+    {
+      "epoch": 0.024833333333333332,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.6059895833333333,
+      "learning_rate": 9.98499474634991e-05,
+      "loss": 7.5172,
+      "loss/crossentropy": 1.265583649277687,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19561972841620445,
+      "step": 149
+    },
+    {
+      "epoch": 0.025,
+      "grad_norm": 34.75,
+      "grad_norm_var": 3.468489583333333,
+      "learning_rate": 9.98479139081219e-05,
+      "loss": 7.5424,
+      "loss/crossentropy": 1.336432695388794,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17582042515277863,
+      "step": 150
+    },
+    {
+      "epoch": 0.025166666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.386458333333333,
+      "learning_rate": 9.98458666866564e-05,
+      "loss": 7.8285,
+      "loss/crossentropy": 1.2626894861459732,
+      "loss/hidden": 4.13671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26004448160529137,
+      "step": 151
+    },
+    {
+      "epoch": 0.025333333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.4458333333333333,
+      "learning_rate": 9.984380579966385e-05,
+      "loss": 7.9453,
+      "loss/crossentropy": 1.5037457644939423,
+      "loss/hidden": 4.0234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3038301654160023,
+      "step": 152
+    },
+    {
+      "epoch": 0.0255,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.8059895833333335,
+      "learning_rate": 9.984173124770923e-05,
+      "loss": 7.4471,
+      "loss/crossentropy": 2.0935803651809692,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20670584589242935,
+      "step": 153
+    },
+    {
+      "epoch": 0.025666666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.23515625,
+      "learning_rate": 9.983964303136133e-05,
+      "loss": 7.7246,
+      "loss/crossentropy": 1.491058737039566,
+      "loss/hidden": 3.8046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.175271263346076,
+      "step": 154
+    },
+    {
+      "epoch": 0.025833333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.21015625,
+      "learning_rate": 9.983754115119261e-05,
+      "loss": 7.4375,
+      "loss/crossentropy": 2.090191066265106,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.160873519256711,
+      "step": 155
+    },
+    {
+      "epoch": 0.026,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.1462890625,
+      "learning_rate": 9.983542560777935e-05,
+      "loss": 7.4173,
+      "loss/crossentropy": 2.091566741466522,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24305898323655128,
+      "step": 156
+    },
+    {
+      "epoch": 0.026166666666666668,
+      "grad_norm": 34.5,
+      "grad_norm_var": 3.3291015625,
+      "learning_rate": 9.983329640170149e-05,
+      "loss": 7.6379,
+      "loss/crossentropy": 1.7479139864444733,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24493517726659775,
+      "step": 157
+    },
+    {
+      "epoch": 0.026333333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.5,
+      "learning_rate": 9.983115353354281e-05,
+      "loss": 7.2407,
+      "loss/crossentropy": 1.7053819745779037,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2257254458963871,
+      "step": 158
+    },
+    {
+      "epoch": 0.0265,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.7900390625,
+      "learning_rate": 9.982899700389076e-05,
+      "loss": 7.4953,
+      "loss/crossentropy": 1.3609639704227448,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15046744793653488,
+      "step": 159
+    },
+    {
+      "epoch": 0.02666666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.4337890625,
+      "learning_rate": 9.982682681333658e-05,
+      "loss": 7.5805,
+      "loss/crossentropy": 1.8961246311664581,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21236862987279892,
+      "step": 160
+    },
+    {
+      "epoch": 0.026833333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.4197265625,
+      "learning_rate": 9.982464296247522e-05,
+      "loss": 7.5454,
+      "loss/crossentropy": 1.612160935997963,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18199052661657333,
+      "step": 161
+    },
+    {
+      "epoch": 0.027,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.2416015625,
+      "learning_rate": 9.982244545190542e-05,
+      "loss": 7.783,
+      "loss/crossentropy": 1.516614407300949,
+      "loss/hidden": 3.87890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23189150914549828,
+      "step": 162
+    },
+    {
+      "epoch": 0.027166666666666665,
+      "grad_norm": 34.75,
+      "grad_norm_var": 2.6759765625,
+      "learning_rate": 9.982023428222962e-05,
+      "loss": 7.5819,
+      "loss/crossentropy": 1.9319280982017517,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2132977582514286,
+      "step": 163
+    },
+    {
+      "epoch": 0.027333333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.6791666666666667,
+      "learning_rate": 9.981800945405403e-05,
+      "loss": 7.4188,
+      "loss/crossentropy": 1.921055793762207,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22658692672848701,
+      "step": 164
+    },
+    {
+      "epoch": 0.0275,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.6447265625,
+      "learning_rate": 9.981577096798863e-05,
+      "loss": 7.628,
+      "loss/crossentropy": 2.343845933675766,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18943049386143684,
+      "step": 165
+    },
+    {
+      "epoch": 0.027666666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.2958333333333334,
+      "learning_rate": 9.981351882464706e-05,
+      "loss": 7.7791,
+      "loss/crossentropy": 1.7194806337356567,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22096674144268036,
+      "step": 166
+    },
+    {
+      "epoch": 0.027833333333333335,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.2926432291666665,
+      "learning_rate": 9.98112530246468e-05,
+      "loss": 7.9327,
+      "loss/crossentropy": 1.6965374052524567,
+      "loss/hidden": 3.8671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.314368050545454,
+      "step": 167
+    },
+    {
+      "epoch": 0.028,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.1770833333333335,
+      "learning_rate": 9.980897356860901e-05,
+      "loss": 7.1302,
+      "loss/crossentropy": 1.3775997012853622,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17563216760754585,
+      "step": 168
+    },
+    {
+      "epoch": 0.028166666666666666,
+      "grad_norm": 34.0,
+      "grad_norm_var": 2.314583333333333,
+      "learning_rate": 9.980668045715864e-05,
+      "loss": 7.7548,
+      "loss/crossentropy": 1.6328150779008865,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19686350598931313,
+      "step": 169
+    },
+    {
+      "epoch": 0.028333333333333332,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.01640625,
+      "learning_rate": 9.980437369092431e-05,
+      "loss": 7.6132,
+      "loss/crossentropy": 1.4576155543327332,
+      "loss/hidden": 4.0703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2253819815814495,
+      "step": 170
+    },
+    {
+      "epoch": 0.0285,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.04140625,
+      "learning_rate": 9.980205327053848e-05,
+      "loss": 7.8547,
+      "loss/crossentropy": 1.8768835365772247,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.173042681068182,
+      "step": 171
+    },
+    {
+      "epoch": 0.028666666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.97890625,
+      "learning_rate": 9.97997191966373e-05,
+      "loss": 7.4031,
+      "loss/crossentropy": 1.8408287167549133,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19031177461147308,
+      "step": 172
+    },
+    {
+      "epoch": 0.028833333333333332,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.709375,
+      "learning_rate": 9.979737146986064e-05,
+      "loss": 7.4604,
+      "loss/crossentropy": 1.340224727988243,
+      "loss/hidden": 3.7578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2081661932170391,
+      "step": 173
+    },
+    {
+      "epoch": 0.029,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.5462890625,
+      "learning_rate": 9.979501009085219e-05,
+      "loss": 7.415,
+      "loss/crossentropy": 1.5763408243656158,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17047973722219467,
+      "step": 174
+    },
+    {
+      "epoch": 0.029166666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.1393229166666667,
+      "learning_rate": 9.979263506025929e-05,
+      "loss": 7.5373,
+      "loss/crossentropy": 1.7883712649345398,
+      "loss/hidden": 3.74609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2959150858223438,
+      "step": 175
+    },
+    {
+      "epoch": 0.029333333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 1.1822916666666667,
+      "learning_rate": 9.97902463787331e-05,
+      "loss": 7.4311,
+      "loss/crossentropy": 2.3578516244888306,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19385182484984398,
+      "step": 176
+    },
+    {
+      "epoch": 0.0295,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.4254557291666667,
+      "learning_rate": 9.978784404692847e-05,
+      "loss": 7.4492,
+      "loss/crossentropy": 2.283793807029724,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21292581409215927,
+      "step": 177
+    },
+    {
+      "epoch": 0.029666666666666668,
+      "grad_norm": 34.0,
+      "grad_norm_var": 1.4926432291666667,
+      "learning_rate": 9.978542806550402e-05,
+      "loss": 7.7152,
+      "loss/crossentropy": 1.6890181303024292,
+      "loss/hidden": 4.0234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3450893461704254,
+      "step": 178
+    },
+    {
+      "epoch": 0.029833333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.2455729166666667,
+      "learning_rate": 9.97829984351221e-05,
+      "loss": 7.3753,
+      "loss/crossentropy": 1.7534202337265015,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19514819234609604,
+      "step": 179
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.5104166666666667,
+      "learning_rate": 9.978055515644882e-05,
+      "loss": 7.2476,
+      "loss/crossentropy": 1.5852151215076447,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15638437122106552,
+      "step": 180
+    },
+    {
+      "epoch": 0.030166666666666668,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.4817057291666667,
+      "learning_rate": 9.977809823015401e-05,
+      "loss": 7.1402,
+      "loss/crossentropy": 1.579344391822815,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1618119813501835,
+      "step": 181
+    },
+    {
+      "epoch": 0.030333333333333334,
+      "grad_norm": 34.5,
+      "grad_norm_var": 1.7729166666666667,
+      "learning_rate": 9.977562765691124e-05,
+      "loss": 7.4241,
+      "loss/crossentropy": 1.1309615969657898,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15628819353878498,
+      "step": 182
+    },
+    {
+      "epoch": 0.0305,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.7645182291666666,
+      "learning_rate": 9.977314343739786e-05,
+      "loss": 7.5023,
+      "loss/crossentropy": 2.053345203399658,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2063504196703434,
+      "step": 183
+    },
+    {
+      "epoch": 0.030666666666666665,
+      "grad_norm": 34.25,
+      "grad_norm_var": 1.7697916666666667,
+      "learning_rate": 9.977064557229492e-05,
+      "loss": 7.5849,
+      "loss/crossentropy": 1.5982548594474792,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3230434563010931,
+      "step": 184
+    },
+    {
+      "epoch": 0.030833333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.653125,
+      "learning_rate": 9.97681340622872e-05,
+      "loss": 7.6597,
+      "loss/crossentropy": 1.831273764371872,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21502476185560226,
+      "step": 185
+    },
+    {
+      "epoch": 0.031,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.5708333333333333,
+      "learning_rate": 9.976560890806328e-05,
+      "loss": 7.2985,
+      "loss/crossentropy": 1.863772302865982,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2501535527408123,
+      "step": 186
+    },
+    {
+      "epoch": 0.031166666666666665,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.5643229166666666,
+      "learning_rate": 9.976307011031542e-05,
+      "loss": 7.632,
+      "loss/crossentropy": 1.398108333349228,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17524024844169617,
+      "step": 187
+    },
+    {
+      "epoch": 0.03133333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.6942057291666666,
+      "learning_rate": 9.976051766973966e-05,
+      "loss": 7.4031,
+      "loss/crossentropy": 1.7360979616641998,
+      "loss/hidden": 3.92578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2732916437089443,
+      "step": 188
+    },
+    {
+      "epoch": 0.0315,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.7145182291666667,
+      "learning_rate": 9.975795158703576e-05,
+      "loss": 7.7187,
+      "loss/crossentropy": 1.8443644046783447,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19181860238313675,
+      "step": 189
+    },
+    {
+      "epoch": 0.03166666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.155143229166667,
+      "learning_rate": 9.975537186290724e-05,
+      "loss": 7.2779,
+      "loss/crossentropy": 2.078747868537903,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27637293189764023,
+      "step": 190
+    },
+    {
+      "epoch": 0.03183333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.2499348958333334,
+      "learning_rate": 9.975277849806133e-05,
+      "loss": 7.3447,
+      "loss/crossentropy": 1.783536285161972,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20468712225556374,
+      "step": 191
+    },
+    {
+      "epoch": 0.032,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.9874348958333334,
+      "learning_rate": 9.9750171493209e-05,
+      "loss": 7.4426,
+      "loss/crossentropy": 2.0766963958740234,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23221758753061295,
+      "step": 192
+    },
+    {
+      "epoch": 0.03216666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.8921223958333333,
+      "learning_rate": 9.974755084906502e-05,
+      "loss": 7.433,
+      "loss/crossentropy": 1.5522357821464539,
+      "loss/hidden": 3.76171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23938192054629326,
+      "step": 193
+    },
+    {
+      "epoch": 0.03233333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.671875,
+      "learning_rate": 9.974491656634782e-05,
+      "loss": 7.0971,
+      "loss/crossentropy": 2.2775827646255493,
+      "loss/hidden": 3.83203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2913665287196636,
+      "step": 194
+    },
+    {
+      "epoch": 0.0325,
+      "grad_norm": 34.0,
+      "grad_norm_var": 1.8921223958333333,
+      "learning_rate": 9.974226864577961e-05,
+      "loss": 7.4177,
+      "loss/crossentropy": 1.771312177181244,
+      "loss/hidden": 3.71875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22810643911361694,
+      "step": 195
+    },
+    {
+      "epoch": 0.03266666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.7072265625,
+      "learning_rate": 9.973960708808633e-05,
+      "loss": 7.633,
+      "loss/crossentropy": 1.507501482963562,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1433581579476595,
+      "step": 196
+    },
+    {
+      "epoch": 0.03283333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 1.7416666666666667,
+      "learning_rate": 9.973693189399766e-05,
+      "loss": 7.7088,
+      "loss/crossentropy": 1.6331115067005157,
+      "loss/hidden": 3.8671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3173863925039768,
+      "step": 197
+    },
+    {
+      "epoch": 0.033,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.5291666666666666,
+      "learning_rate": 9.973424306424705e-05,
+      "loss": 7.5782,
+      "loss/crossentropy": 1.5179617553949356,
+      "loss/hidden": 3.8984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.281230915337801,
+      "step": 198
+    },
+    {
+      "epoch": 0.033166666666666664,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.5205729166666666,
+      "learning_rate": 9.973154059957162e-05,
+      "loss": 8.0153,
+      "loss/crossentropy": 1.9138246178627014,
+      "loss/hidden": 3.8359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20401380956172943,
+      "step": 199
+    },
+    {
+      "epoch": 0.03333333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.3114583333333334,
+      "learning_rate": 9.972882450071228e-05,
+      "loss": 7.0994,
+      "loss/crossentropy": 1.0574043989181519,
+      "loss/hidden": 3.921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23022147826850414,
+      "step": 200
+    },
+    {
+      "epoch": 0.0335,
+      "grad_norm": 34.0,
+      "grad_norm_var": 1.478125,
+      "learning_rate": 9.972609476841367e-05,
+      "loss": 7.7352,
+      "loss/crossentropy": 1.9633951485157013,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26096950471401215,
+      "step": 201
+    },
+    {
+      "epoch": 0.033666666666666664,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.46015625,
+      "learning_rate": 9.972335140342415e-05,
+      "loss": 7.4408,
+      "loss/crossentropy": 1.3950132727622986,
+      "loss/hidden": 3.8203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20604024082422256,
+      "step": 202
+    },
+    {
+      "epoch": 0.03383333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.5622395833333333,
+      "learning_rate": 9.972059440649584e-05,
+      "loss": 7.1746,
+      "loss/crossentropy": 1.7117474675178528,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18884649127721786,
+      "step": 203
+    },
+    {
+      "epoch": 0.034,
+      "grad_norm": 33.75,
+      "grad_norm_var": 1.5577473958333334,
+      "learning_rate": 9.971782377838457e-05,
+      "loss": 7.9798,
+      "loss/crossentropy": 1.7780987322330475,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20581259950995445,
+      "step": 204
+    },
+    {
+      "epoch": 0.034166666666666665,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.9639973958333334,
+      "learning_rate": 9.971503951984995e-05,
+      "loss": 7.2605,
+      "loss/crossentropy": 1.6917162835597992,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1562095284461975,
+      "step": 205
+    },
+    {
+      "epoch": 0.034333333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.5421223958333334,
+      "learning_rate": 9.971224163165527e-05,
+      "loss": 7.2398,
+      "loss/crossentropy": 1.9893342852592468,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18143022432923317,
+      "step": 206
+    },
+    {
+      "epoch": 0.0345,
+      "grad_norm": 34.0,
+      "grad_norm_var": 1.6311848958333333,
+      "learning_rate": 9.970943011456761e-05,
+      "loss": 8.0924,
+      "loss/crossentropy": 1.8004242777824402,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21368973329663277,
+      "step": 207
+    },
+    {
+      "epoch": 0.034666666666666665,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.8004557291666667,
+      "learning_rate": 9.970660496935776e-05,
+      "loss": 7.2989,
+      "loss/crossentropy": 1.4333279579877853,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17135126143693924,
+      "step": 208
+    },
+    {
+      "epoch": 0.034833333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.7705729166666666,
+      "learning_rate": 9.970376619680024e-05,
+      "loss": 7.344,
+      "loss/crossentropy": 1.4728360325098038,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17651018127799034,
+      "step": 209
+    },
+    {
+      "epoch": 0.035,
+      "grad_norm": 34.0,
+      "grad_norm_var": 1.8978515625,
+      "learning_rate": 9.970091379767331e-05,
+      "loss": 7.598,
+      "loss/crossentropy": 1.7544775307178497,
+      "loss/hidden": 3.7109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23908454179763794,
+      "step": 210
+    },
+    {
+      "epoch": 0.035166666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.0541015625,
+      "learning_rate": 9.9698047772759e-05,
+      "loss": 7.623,
+      "loss/crossentropy": 2.131743371486664,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19084632396697998,
+      "step": 211
+    },
+    {
+      "epoch": 0.035333333333333335,
+      "grad_norm": 35.0,
+      "grad_norm_var": 2.4853515625,
+      "learning_rate": 9.969516812284301e-05,
+      "loss": 7.3634,
+      "loss/crossentropy": 2.212506026029587,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2108316533267498,
+      "step": 212
+    },
+    {
+      "epoch": 0.0355,
+      "grad_norm": 36.25,
+      "grad_norm_var": 3.2681640625,
+      "learning_rate": 9.969227484871484e-05,
+      "loss": 7.0909,
+      "loss/crossentropy": 1.9730757176876068,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.220124252140522,
+      "step": 213
+    },
+    {
+      "epoch": 0.035666666666666666,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.2968098958333334,
+      "learning_rate": 9.968936795116768e-05,
+      "loss": 7.7264,
+      "loss/crossentropy": 1.7381523847579956,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20243105106055737,
+      "step": 214
+    },
+    {
+      "epoch": 0.035833333333333335,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.4764973958333334,
+      "learning_rate": 9.968644743099848e-05,
+      "loss": 7.3717,
+      "loss/crossentropy": 1.4052337929606438,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12791427364572883,
+      "step": 215
+    },
+    {
+      "epoch": 0.036,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.3843098958333333,
+      "learning_rate": 9.968351328900794e-05,
+      "loss": 7.4818,
+      "loss/crossentropy": 1.563568338751793,
+      "loss/hidden": 3.89453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22113186493515968,
+      "step": 216
+    },
+    {
+      "epoch": 0.036166666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.5702473958333334,
+      "learning_rate": 9.968056552600043e-05,
+      "loss": 7.3022,
+      "loss/crossentropy": 1.722176730632782,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1667504534125328,
+      "step": 217
+    },
+    {
+      "epoch": 0.036333333333333336,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.566666666666667,
+      "learning_rate": 9.967760414278411e-05,
+      "loss": 7.4332,
+      "loss/crossentropy": 2.4729353487491608,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22087901085615158,
+      "step": 218
+    },
+    {
+      "epoch": 0.0365,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.3822916666666667,
+      "learning_rate": 9.967462914017088e-05,
+      "loss": 7.337,
+      "loss/crossentropy": 1.458601787686348,
+      "loss/hidden": 3.82421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18009812012314796,
+      "step": 219
+    },
+    {
+      "epoch": 0.03666666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.351822916666667,
+      "learning_rate": 9.967164051897633e-05,
+      "loss": 7.9017,
+      "loss/crossentropy": 1.590276151895523,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2577647417783737,
+      "step": 220
+    },
+    {
+      "epoch": 0.036833333333333336,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.8916666666666666,
+      "learning_rate": 9.966863828001982e-05,
+      "loss": 7.2061,
+      "loss/crossentropy": 1.7242890745401382,
+      "loss/hidden": 4.04296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3013566806912422,
+      "step": 221
+    },
+    {
+      "epoch": 0.037,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.890625,
+      "learning_rate": 9.966562242412442e-05,
+      "loss": 7.6513,
+      "loss/crossentropy": 1.62159264087677,
+      "loss/hidden": 3.71875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20868892595171928,
+      "step": 222
+    },
+    {
+      "epoch": 0.03716666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.9822265625,
+      "learning_rate": 9.966259295211697e-05,
+      "loss": 7.7656,
+      "loss/crossentropy": 1.6929957270622253,
+      "loss/hidden": 3.8515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.314834401011467,
+      "step": 223
+    },
+    {
+      "epoch": 0.037333333333333336,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.856705729166667,
+      "learning_rate": 9.965954986482799e-05,
+      "loss": 7.5551,
+      "loss/crossentropy": 1.4845804572105408,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14850697666406631,
+      "step": 224
+    },
+    {
+      "epoch": 0.0375,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.8556640625,
+      "learning_rate": 9.965649316309178e-05,
+      "loss": 7.677,
+      "loss/crossentropy": 1.7364672124385834,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18876685202121735,
+      "step": 225
+    },
+    {
+      "epoch": 0.03766666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.9559895833333334,
+      "learning_rate": 9.965342284774632e-05,
+      "loss": 6.9854,
+      "loss/crossentropy": 1.9357840418815613,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18362392112612724,
+      "step": 226
+    },
+    {
+      "epoch": 0.03783333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.517708333333333,
+      "learning_rate": 9.965033891963338e-05,
+      "loss": 7.6782,
+      "loss/crossentropy": 1.895157277584076,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26038768514990807,
+      "step": 227
+    },
+    {
+      "epoch": 0.038,
+      "grad_norm": 35.75,
+      "grad_norm_var": 2.7622395833333333,
+      "learning_rate": 9.964724137959843e-05,
+      "loss": 7.9638,
+      "loss/crossentropy": 1.6238293796777725,
+      "loss/hidden": 3.78125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16440559178590775,
+      "step": 228
+    },
+    {
+      "epoch": 0.03816666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.9893229166666666,
+      "learning_rate": 9.964413022849068e-05,
+      "loss": 7.364,
+      "loss/crossentropy": 2.079767644405365,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1978134848177433,
+      "step": 229
+    },
+    {
+      "epoch": 0.03833333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.9770182291666667,
+      "learning_rate": 9.964100546716309e-05,
+      "loss": 7.2248,
+      "loss/crossentropy": 2.071979194879532,
+      "loss/hidden": 3.85546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2368118055164814,
+      "step": 230
+    },
+    {
+      "epoch": 0.0385,
+      "grad_norm": 33.75,
+      "grad_norm_var": 1.8738932291666666,
+      "learning_rate": 9.963786709647228e-05,
+      "loss": 7.5364,
+      "loss/crossentropy": 1.8584957420825958,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19067588821053505,
+      "step": 231
+    },
+    {
+      "epoch": 0.03866666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.8832682291666667,
+      "learning_rate": 9.963471511727868e-05,
+      "loss": 7.5332,
+      "loss/crossentropy": 1.5630061328411102,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16000641509890556,
+      "step": 232
+    },
+    {
+      "epoch": 0.03883333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.5421223958333334,
+      "learning_rate": 9.963154953044645e-05,
+      "loss": 7.7968,
+      "loss/crossentropy": 2.40740567445755,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1722939983010292,
+      "step": 233
+    },
+    {
+      "epoch": 0.039,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.386458333333333,
+      "learning_rate": 9.962837033684343e-05,
+      "loss": 6.9222,
+      "loss/crossentropy": 1.3479905128479004,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1708462443202734,
+      "step": 234
+    },
+    {
+      "epoch": 0.03916666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.3872395833333333,
+      "learning_rate": 9.96251775373412e-05,
+      "loss": 7.5628,
+      "loss/crossentropy": 1.9919890463352203,
+      "loss/hidden": 3.71484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19337913393974304,
+      "step": 235
+    },
+    {
+      "epoch": 0.03933333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.3997395833333335,
+      "learning_rate": 9.962197113281509e-05,
+      "loss": 7.6578,
+      "loss/crossentropy": 1.9954728782176971,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21360453963279724,
+      "step": 236
+    },
+    {
+      "epoch": 0.0395,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.1184895833333335,
+      "learning_rate": 9.961875112414416e-05,
+      "loss": 7.7701,
+      "loss/crossentropy": 1.5776352882385254,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2165641412138939,
+      "step": 237
+    },
+    {
+      "epoch": 0.03966666666666667,
+      "grad_norm": 37.5,
+      "grad_norm_var": 3.769791666666667,
+      "learning_rate": 9.961551751221121e-05,
+      "loss": 7.4688,
+      "loss/crossentropy": 1.602569431066513,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16928716376423836,
+      "step": 238
+    },
+    {
+      "epoch": 0.03983333333333333,
+      "grad_norm": 35.75,
+      "grad_norm_var": 4.255143229166666,
+      "learning_rate": 9.961227029790272e-05,
+      "loss": 7.3368,
+      "loss/crossentropy": 1.9378673136234283,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1705906204879284,
+      "step": 239
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 31.0,
+      "grad_norm_var": 4.312955729166666,
+      "learning_rate": 9.960900948210896e-05,
+      "loss": 7.3995,
+      "loss/crossentropy": 1.6474340558052063,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22635122761130333,
+      "step": 240
+    },
+    {
+      "epoch": 0.04016666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.333268229166666,
+      "learning_rate": 9.96057350657239e-05,
+      "loss": 7.7658,
+      "loss/crossentropy": 2.135969579219818,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1689089611172676,
+      "step": 241
+    },
+    {
+      "epoch": 0.04033333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 4.2369140625,
+      "learning_rate": 9.960244704964521e-05,
+      "loss": 7.3876,
+      "loss/crossentropy": 1.2477717697620392,
+      "loss/hidden": 3.8671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1663598008453846,
+      "step": 242
+    },
+    {
+      "epoch": 0.0405,
+      "grad_norm": 33.75,
+      "grad_norm_var": 4.2884765625,
+      "learning_rate": 9.959914543477435e-05,
+      "loss": 7.5481,
+      "loss/crossentropy": 1.8062715828418732,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21589022129774094,
+      "step": 243
+    },
+    {
+      "epoch": 0.04066666666666666,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.8108723958333335,
+      "learning_rate": 9.959583022201647e-05,
+      "loss": 7.2943,
+      "loss/crossentropy": 2.5664525628089905,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.182389248162508,
+      "step": 244
+    },
+    {
+      "epoch": 0.04083333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.848372395833333,
+      "learning_rate": 9.959250141228045e-05,
+      "loss": 7.3098,
+      "loss/crossentropy": 1.983911782503128,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23118621855974197,
+      "step": 245
+    },
+    {
+      "epoch": 0.041,
+      "grad_norm": 29.75,
+      "grad_norm_var": 4.30625,
+      "learning_rate": 9.95891590064789e-05,
+      "loss": 7.5478,
+      "loss/crossentropy": 1.5972959399223328,
+      "loss/hidden": 3.796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.208140280097723,
+      "step": 246
+    },
+    {
+      "epoch": 0.041166666666666664,
+      "grad_norm": 37.75,
+      "grad_norm_var": 5.90625,
+      "learning_rate": 9.958580300552815e-05,
+      "loss": 7.2526,
+      "loss/crossentropy": 1.255698412656784,
+      "loss/hidden": 4.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24902774393558502,
+      "step": 247
+    },
+    {
+      "epoch": 0.04133333333333333,
+      "grad_norm": 60.75,
+      "grad_norm_var": 54.296875,
+      "learning_rate": 9.958243341034827e-05,
+      "loss": 8.0128,
+      "loss/crossentropy": 1.4112011045217514,
+      "loss/hidden": 4.0234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3044033832848072,
+      "step": 248
+    },
+    {
+      "epoch": 0.0415,
+      "grad_norm": 34.75,
+      "grad_norm_var": 54.10182291666667,
+      "learning_rate": 9.957905022186309e-05,
+      "loss": 7.643,
+      "loss/crossentropy": 2.386468082666397,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2023131437599659,
+      "step": 249
+    },
+    {
+      "epoch": 0.041666666666666664,
+      "grad_norm": 31.75,
+      "grad_norm_var": 52.558268229166664,
+      "learning_rate": 9.957565344100009e-05,
+      "loss": 7.2652,
+      "loss/crossentropy": 1.6484432220458984,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1775803379714489,
+      "step": 250
+    },
+    {
+      "epoch": 0.041833333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 54.62291666666667,
+      "learning_rate": 9.957224306869053e-05,
+      "loss": 7.2139,
+      "loss/crossentropy": 2.3155510425567627,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.200841274112463,
+      "step": 251
+    },
+    {
+      "epoch": 0.042,
+      "grad_norm": 30.5,
+      "grad_norm_var": 55.110416666666666,
+      "learning_rate": 9.956881910586937e-05,
+      "loss": 7.2056,
+      "loss/crossentropy": 1.4926374852657318,
+      "loss/hidden": 3.8125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19533399865031242,
+      "step": 252
+    },
+    {
+      "epoch": 0.042166666666666665,
+      "grad_norm": 31.25,
+      "grad_norm_var": 55.40807291666667,
+      "learning_rate": 9.956538155347534e-05,
+      "loss": 7.3433,
+      "loss/crossentropy": 2.0166503190994263,
+      "loss/hidden": 3.796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26178446784615517,
+      "step": 253
+    },
+    {
+      "epoch": 0.042333333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 54.967708333333334,
+      "learning_rate": 9.956193041245084e-05,
+      "loss": 7.5805,
+      "loss/crossentropy": 1.2202660590410233,
+      "loss/hidden": 3.875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1833195984363556,
+      "step": 254
+    },
+    {
+      "epoch": 0.0425,
+      "grad_norm": 31.0,
+      "grad_norm_var": 55.44765625,
+      "learning_rate": 9.955846568374201e-05,
+      "loss": 7.2663,
+      "loss/crossentropy": 1.2276012897491455,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13461654260754585,
+      "step": 255
+    },
+    {
+      "epoch": 0.042666666666666665,
+      "grad_norm": 32.5,
+      "grad_norm_var": 54.99140625,
+      "learning_rate": 9.955498736829875e-05,
+      "loss": 7.3589,
+      "loss/crossentropy": 1.6177268624305725,
+      "loss/hidden": 3.7421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20657369121909142,
+      "step": 256
+    },
+    {
+      "epoch": 0.042833333333333334,
+      "grad_norm": 39.0,
+      "grad_norm_var": 56.19375,
+      "learning_rate": 9.955149546707465e-05,
+      "loss": 7.3266,
+      "loss/crossentropy": 1.5220600366592407,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1910797841846943,
+      "step": 257
+    },
+    {
+      "epoch": 0.043,
+      "grad_norm": 32.75,
+      "grad_norm_var": 55.8416015625,
+      "learning_rate": 9.954798998102702e-05,
+      "loss": 7.8094,
+      "loss/crossentropy": 1.7598982155323029,
+      "loss/hidden": 3.91796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28901227191090584,
+      "step": 258
+    },
+    {
+      "epoch": 0.043166666666666666,
+      "grad_norm": 64.5,
+      "grad_norm_var": 111.5759765625,
+      "learning_rate": 9.954447091111694e-05,
+      "loss": 7.4358,
+      "loss/crossentropy": 2.1494410634040833,
+      "loss/hidden": 4.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33057959005236626,
+      "step": 259
+    },
+    {
+      "epoch": 0.043333333333333335,
+      "grad_norm": 35.75,
+      "grad_norm_var": 111.18587239583333,
+      "learning_rate": 9.954093825830917e-05,
+      "loss": 7.3195,
+      "loss/crossentropy": 1.4215206280350685,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12461109273135662,
+      "step": 260
+    },
+    {
+      "epoch": 0.0435,
+      "grad_norm": 32.75,
+      "grad_norm_var": 110.76087239583333,
+      "learning_rate": 9.953739202357218e-05,
+      "loss": 7.3571,
+      "loss/crossentropy": 1.2692449390888214,
+      "loss/hidden": 3.96875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19402378797531128,
+      "step": 261
+    },
+    {
+      "epoch": 0.043666666666666666,
+      "grad_norm": 36.5,
+      "grad_norm_var": 107.39993489583334,
+      "learning_rate": 9.953383220787824e-05,
+      "loss": 8.1526,
+      "loss/crossentropy": 2.4642735719680786,
+      "loss/hidden": 3.09765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19126908108592033,
+      "step": 262
+    },
+    {
+      "epoch": 0.043833333333333335,
+      "grad_norm": 31.5,
+      "grad_norm_var": 109.27493489583334,
+      "learning_rate": 9.953025881220325e-05,
+      "loss": 7.6018,
+      "loss/crossentropy": 1.7057749032974243,
+      "loss/hidden": 3.87890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23253849148750305,
+      "step": 263
+    },
+    {
+      "epoch": 0.044,
+      "grad_norm": 31.25,
+      "grad_norm_var": 68.98899739583334,
+      "learning_rate": 9.952667183752689e-05,
+      "loss": 7.2282,
+      "loss/crossentropy": 1.576460361480713,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20516490936279297,
+      "step": 264
+    },
+    {
+      "epoch": 0.04416666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 70.1666015625,
+      "learning_rate": 9.952307128483256e-05,
+      "loss": 7.4597,
+      "loss/crossentropy": 1.3988469243049622,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24469899386167526,
+      "step": 265
+    },
+    {
+      "epoch": 0.044333333333333336,
+      "grad_norm": 32.75,
+      "grad_norm_var": 69.85305989583334,
+      "learning_rate": 9.951945715510738e-05,
+      "loss": 7.5851,
+      "loss/crossentropy": 1.6784016638994217,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26239439100027084,
+      "step": 266
+    },
+    {
+      "epoch": 0.0445,
+      "grad_norm": 30.375,
+      "grad_norm_var": 68.84212239583333,
+      "learning_rate": 9.951582944934215e-05,
+      "loss": 7.3346,
+      "loss/crossentropy": 1.2082529813051224,
+      "loss/hidden": 3.76171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16079665906727314,
+      "step": 267
+    },
+    {
+      "epoch": 0.04466666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 68.91354166666666,
+      "learning_rate": 9.951218816853145e-05,
+      "loss": 7.2935,
+      "loss/crossentropy": 1.6022576689720154,
+      "loss/hidden": 3.77734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19950025528669357,
+      "step": 268
+    },
+    {
+      "epoch": 0.044833333333333336,
+      "grad_norm": 96.5,
+      "grad_norm_var": 304.83307291666665,
+      "learning_rate": 9.950853331367356e-05,
+      "loss": 7.6964,
+      "loss/crossentropy": 1.8837226927280426,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.241776242852211,
+      "step": 269
+    },
+    {
+      "epoch": 0.045,
+      "grad_norm": 39.0,
+      "grad_norm_var": 302.23541666666665,
+      "learning_rate": 9.950486488577045e-05,
+      "loss": 7.5045,
+      "loss/crossentropy": 1.924521803855896,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1783529743552208,
+      "step": 270
+    },
+    {
+      "epoch": 0.04516666666666667,
+      "grad_norm": 36.75,
+      "grad_norm_var": 298.0247395833333,
+      "learning_rate": 9.950118288582788e-05,
+      "loss": 7.2832,
+      "loss/crossentropy": 1.426510602235794,
+      "loss/hidden": 3.73046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20130225643515587,
+      "step": 271
+    },
+    {
+      "epoch": 0.04533333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 297.57057291666666,
+      "learning_rate": 9.949748731485527e-05,
+      "loss": 7.7463,
+      "loss/crossentropy": 1.5387413799762726,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13837396539747715,
+      "step": 272
+    },
+    {
+      "epoch": 0.0455,
+      "grad_norm": 34.5,
+      "grad_norm_var": 299.1830729166667,
+      "learning_rate": 9.949377817386579e-05,
+      "loss": 7.748,
+      "loss/crossentropy": 2.084898829460144,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17804009094834328,
+      "step": 273
+    },
+    {
+      "epoch": 0.04566666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 299.40520833333335,
+      "learning_rate": 9.949005546387631e-05,
+      "loss": 7.6664,
+      "loss/crossentropy": 1.4455165565013885,
+      "loss/hidden": 3.734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17068586684763432,
+      "step": 274
+    },
+    {
+      "epoch": 0.04583333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 258.03098958333334,
+      "learning_rate": 9.948631918590746e-05,
+      "loss": 7.3021,
+      "loss/crossentropy": 1.790118157863617,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20695554837584496,
+      "step": 275
+    },
+    {
+      "epoch": 0.046,
+      "grad_norm": 31.375,
+      "grad_norm_var": 260.02024739583334,
+      "learning_rate": 9.948256934098352e-05,
+      "loss": 7.4683,
+      "loss/crossentropy": 1.9009050130844116,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14622126519680023,
+      "step": 276
+    },
+    {
+      "epoch": 0.04616666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 261.7728515625,
+      "learning_rate": 9.947880593013255e-05,
+      "loss": 6.8859,
+      "loss/crossentropy": 1.3700038492679596,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1501402109861374,
+      "step": 277
+    },
+    {
+      "epoch": 0.04633333333333333,
+      "grad_norm": 35.0,
+      "grad_norm_var": 261.9494140625,
+      "learning_rate": 9.947502895438631e-05,
+      "loss": 7.5624,
+      "loss/crossentropy": 1.6421467065811157,
+      "loss/hidden": 3.97265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2923738732933998,
+      "step": 278
+    },
+    {
+      "epoch": 0.0465,
+      "grad_norm": 31.25,
+      "grad_norm_var": 262.1228515625,
+      "learning_rate": 9.94712384147803e-05,
+      "loss": 7.6512,
+      "loss/crossentropy": 2.2700000405311584,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1824687384068966,
+      "step": 279
+    },
+    {
+      "epoch": 0.04666666666666667,
+      "grad_norm": 77.5,
+      "grad_norm_var": 363.0056640625,
+      "learning_rate": 9.94674343123537e-05,
+      "loss": 7.6463,
+      "loss/crossentropy": 1.9362052083015442,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22555846720933914,
+      "step": 280
+    },
+    {
+      "epoch": 0.04683333333333333,
+      "grad_norm": 60.75,
+      "grad_norm_var": 384.05462239583335,
+      "learning_rate": 9.946361664814943e-05,
+      "loss": 7.5631,
+      "loss/crossentropy": 1.9529215693473816,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21802689135074615,
+      "step": 281
+    },
+    {
+      "epoch": 0.047,
+      "grad_norm": 39.75,
+      "grad_norm_var": 379.08899739583336,
+      "learning_rate": 9.945978542321411e-05,
+      "loss": 7.4859,
+      "loss/crossentropy": 1.9232031404972076,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19063835218548775,
+      "step": 282
+    },
+    {
+      "epoch": 0.04716666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 374.39348958333335,
+      "learning_rate": 9.945594063859809e-05,
+      "loss": 7.6737,
+      "loss/crossentropy": 1.237590491771698,
+      "loss/hidden": 3.94921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17227556183934212,
+      "step": 283
+    },
+    {
+      "epoch": 0.04733333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 373.63307291666666,
+      "learning_rate": 9.945208229535548e-05,
+      "loss": 7.5977,
+      "loss/crossentropy": 1.904694378376007,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2495473362505436,
+      "step": 284
+    },
+    {
+      "epoch": 0.0475,
+      "grad_norm": 52.75,
+      "grad_norm_var": 175.61875,
+      "learning_rate": 9.944821039454402e-05,
+      "loss": 7.4746,
+      "loss/crossentropy": 1.3100051432847977,
+      "loss/hidden": 3.83203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20047403126955032,
+      "step": 285
+    },
+    {
+      "epoch": 0.04766666666666667,
+      "grad_norm": 44.0,
+      "grad_norm_var": 176.97291666666666,
+      "learning_rate": 9.944432493722524e-05,
+      "loss": 7.3571,
+      "loss/crossentropy": 1.7345851063728333,
+      "loss/hidden": 3.87109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.209664486348629,
+      "step": 286
+    },
+    {
+      "epoch": 0.04783333333333333,
+      "grad_norm": 34.5,
+      "grad_norm_var": 178.15182291666667,
+      "learning_rate": 9.944042592446434e-05,
+      "loss": 7.9188,
+      "loss/crossentropy": 1.5532788634300232,
+      "loss/hidden": 4.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21716440096497536,
+      "step": 287
+    },
+    {
+      "epoch": 0.048,
+      "grad_norm": 31.375,
+      "grad_norm_var": 179.72180989583333,
+      "learning_rate": 9.943651335733028e-05,
+      "loss": 7.3091,
+      "loss/crossentropy": 1.5013292729854584,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17087434977293015,
+      "step": 288
+    },
+    {
+      "epoch": 0.04816666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 181.50305989583333,
+      "learning_rate": 9.94325872368957e-05,
+      "loss": 7.3496,
+      "loss/crossentropy": 1.9843324422836304,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15040899068117142,
+      "step": 289
+    },
+    {
+      "epoch": 0.04833333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 181.06920572916667,
+      "learning_rate": 9.942864756423697e-05,
+      "loss": 7.6047,
+      "loss/crossentropy": 1.3669559061527252,
+      "loss/hidden": 3.875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3323919512331486,
+      "step": 290
+    },
+    {
+      "epoch": 0.0485,
+      "grad_norm": 36.25,
+      "grad_norm_var": 175.45618489583333,
+      "learning_rate": 9.942469434043418e-05,
+      "loss": 7.8764,
+      "loss/crossentropy": 2.0171509087085724,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17283064499497414,
+      "step": 291
+    },
+    {
+      "epoch": 0.048666666666666664,
+      "grad_norm": 32.75,
+      "grad_norm_var": 174.05182291666668,
+      "learning_rate": 9.942072756657112e-05,
+      "loss": 7.3889,
+      "loss/crossentropy": 1.4584332257509232,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1842716448009014,
+      "step": 292
+    },
+    {
+      "epoch": 0.04883333333333333,
+      "grad_norm": 56.5,
+      "grad_norm_var": 183.81354166666668,
+      "learning_rate": 9.941674724373531e-05,
+      "loss": 7.5858,
+      "loss/crossentropy": 1.7785181254148483,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1826663427054882,
+      "step": 293
+    },
+    {
+      "epoch": 0.049,
+      "grad_norm": 34.5,
+      "grad_norm_var": 184.25625,
+      "learning_rate": 9.941275337301796e-05,
+      "loss": 7.5304,
+      "loss/crossentropy": 1.922767847776413,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18793915212154388,
+      "step": 294
+    },
+    {
+      "epoch": 0.049166666666666664,
+      "grad_norm": 29.875,
+      "grad_norm_var": 186.2306640625,
+      "learning_rate": 9.940874595551404e-05,
+      "loss": 7.4593,
+      "loss/crossentropy": 1.6790326088666916,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.226290762424469,
+      "step": 295
+    },
+    {
+      "epoch": 0.04933333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 94.9650390625,
+      "learning_rate": 9.940472499232217e-05,
+      "loss": 7.3456,
+      "loss/crossentropy": 1.5425264686346054,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20014165341854095,
+      "step": 296
+    },
+    {
+      "epoch": 0.0495,
+      "grad_norm": 32.0,
+      "grad_norm_var": 61.42337239583333,
+      "learning_rate": 9.940069048454476e-05,
+      "loss": 7.2312,
+      "loss/crossentropy": 1.7203574776649475,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1594822872430086,
+      "step": 297
+    },
+    {
+      "epoch": 0.049666666666666665,
+      "grad_norm": 30.75,
+      "grad_norm_var": 62.857747395833336,
+      "learning_rate": 9.939664243328788e-05,
+      "loss": 7.2087,
+      "loss/crossentropy": 1.5136828869581223,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17818651348352432,
+      "step": 298
+    },
+    {
+      "epoch": 0.049833333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 63.895572916666666,
+      "learning_rate": 9.939258083966131e-05,
+      "loss": 7.4739,
+      "loss/crossentropy": 1.2135891020298004,
+      "loss/hidden": 4.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1756036840379238,
+      "step": 299
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 35.0,
+      "grad_norm_var": 62.13170572916667,
+      "learning_rate": 9.938850570477858e-05,
+      "loss": 7.8539,
+      "loss/crossentropy": 2.2176790833473206,
+      "loss/hidden": 3.66015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21778493002057076,
+      "step": 300
+    },
+    {
+      "epoch": 0.050166666666666665,
+      "grad_norm": 32.25,
+      "grad_norm_var": 43.361393229166666,
+      "learning_rate": 9.938441702975689e-05,
+      "loss": 7.7701,
+      "loss/crossentropy": 1.50228750705719,
+      "loss/hidden": 3.984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21811051666736603,
+      "step": 301
+    },
+    {
+      "epoch": 0.050333333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 37.8150390625,
+      "learning_rate": 9.93803148157172e-05,
+      "loss": 7.7161,
+      "loss/crossentropy": 1.6196162104606628,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2511756159365177,
+      "step": 302
+    },
+    {
+      "epoch": 0.0505,
+      "grad_norm": 31.875,
+      "grad_norm_var": 38.16640625,
+      "learning_rate": 9.937619906378413e-05,
+      "loss": 7.6818,
+      "loss/crossentropy": 1.9243209064006805,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1884879358112812,
+      "step": 303
+    },
+    {
+      "epoch": 0.050666666666666665,
+      "grad_norm": 36.75,
+      "grad_norm_var": 38.01243489583333,
+      "learning_rate": 9.937206977508604e-05,
+      "loss": 7.7187,
+      "loss/crossentropy": 2.00638946890831,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20743847452104092,
+      "step": 304
+    },
+    {
+      "epoch": 0.050833333333333335,
+      "grad_norm": 35.25,
+      "grad_norm_var": 37.69680989583333,
+      "learning_rate": 9.936792695075502e-05,
+      "loss": 7.5495,
+      "loss/crossentropy": 2.144765794277191,
+      "loss/hidden": 4.06640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34854231029748917,
+      "step": 305
+    },
+    {
+      "epoch": 0.051,
+      "grad_norm": 32.5,
+      "grad_norm_var": 37.8212890625,
+      "learning_rate": 9.936377059192683e-05,
+      "loss": 8.0608,
+      "loss/crossentropy": 1.8376803696155548,
+      "loss/hidden": 4.09765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3084591254591942,
+      "step": 306
+    },
+    {
+      "epoch": 0.051166666666666666,
+      "grad_norm": 29.125,
+      "grad_norm_var": 39.428125,
+      "learning_rate": 9.935960069974096e-05,
+      "loss": 7.4009,
+      "loss/crossentropy": 1.757315680384636,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1837361603975296,
+      "step": 307
+    },
+    {
+      "epoch": 0.051333333333333335,
+      "grad_norm": 33.75,
+      "grad_norm_var": 39.303125,
+      "learning_rate": 9.935541727534062e-05,
+      "loss": 7.2565,
+      "loss/crossentropy": 1.2428034543991089,
+      "loss/hidden": 4.09765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2133203148841858,
+      "step": 308
+    },
+    {
+      "epoch": 0.0515,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.004166666666666,
+      "learning_rate": 9.93512203198727e-05,
+      "loss": 7.4448,
+      "loss/crossentropy": 1.4536133259534836,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1890176311135292,
+      "step": 309
+    },
+    {
+      "epoch": 0.051666666666666666,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.9497395833333333,
+      "learning_rate": 9.934700983448785e-05,
+      "loss": 7.3572,
+      "loss/crossentropy": 1.5229679942131042,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16795078665018082,
+      "step": 310
+    },
+    {
+      "epoch": 0.051833333333333335,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.5184895833333334,
+      "learning_rate": 9.934278582034037e-05,
+      "loss": 8.021,
+      "loss/crossentropy": 2.2002266943454742,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30353260040283203,
+      "step": 311
+    },
+    {
+      "epoch": 0.052,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.737434895833333,
+      "learning_rate": 9.93385482785883e-05,
+      "loss": 7.6316,
+      "loss/crossentropy": 1.4069420397281647,
+      "loss/hidden": 3.796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1764775663614273,
+      "step": 312
+    },
+    {
+      "epoch": 0.05216666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.7983723958333333,
+      "learning_rate": 9.93342972103934e-05,
+      "loss": 7.0989,
+      "loss/crossentropy": 1.8300898224115372,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20034699887037277,
+      "step": 313
+    },
+    {
+      "epoch": 0.052333333333333336,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.5468098958333334,
+      "learning_rate": 9.933003261692113e-05,
+      "loss": 7.3283,
+      "loss/crossentropy": 1.3944992870092392,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12819642573595047,
+      "step": 314
+    },
+    {
+      "epoch": 0.0525,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.476822916666667,
+      "learning_rate": 9.932575449934062e-05,
+      "loss": 7.7229,
+      "loss/crossentropy": 2.354605346918106,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17047492787241936,
+      "step": 315
+    },
+    {
+      "epoch": 0.05266666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.1768229166666666,
+      "learning_rate": 9.932146285882477e-05,
+      "loss": 7.3177,
+      "loss/crossentropy": 1.7730609476566315,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17645995318889618,
+      "step": 316
+    },
+    {
+      "epoch": 0.052833333333333336,
+      "grad_norm": 35.0,
+      "grad_norm_var": 3.4833333333333334,
+      "learning_rate": 9.931715769655015e-05,
+      "loss": 7.7961,
+      "loss/crossentropy": 1.6061375439167023,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2202853336930275,
+      "step": 317
+    },
+    {
+      "epoch": 0.053,
+      "grad_norm": 31.625,
+      "grad_norm_var": 3.574934895833333,
+      "learning_rate": 9.931283901369706e-05,
+      "loss": 7.5045,
+      "loss/crossentropy": 1.0404724478721619,
+      "loss/hidden": 3.9140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22050953656435013,
+      "step": 318
+    },
+    {
+      "epoch": 0.05316666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.7330729166666665,
+      "learning_rate": 9.930850681144945e-05,
+      "loss": 7.6846,
+      "loss/crossentropy": 1.7995603382587433,
+      "loss/hidden": 3.91015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.261033620685339,
+      "step": 319
+    },
+    {
+      "epoch": 0.05333333333333334,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.8541666666666665,
+      "learning_rate": 9.930416109099505e-05,
+      "loss": 7.693,
+      "loss/crossentropy": 1.8140327036380768,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21637173742055893,
+      "step": 320
+    },
+    {
+      "epoch": 0.0535,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.419791666666667,
+      "learning_rate": 9.929980185352526e-05,
+      "loss": 7.2824,
+      "loss/crossentropy": 1.6707937568426132,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1599106565117836,
+      "step": 321
+    },
+    {
+      "epoch": 0.05366666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.45,
+      "learning_rate": 9.929542910023517e-05,
+      "loss": 7.4991,
+      "loss/crossentropy": 2.0937286615371704,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20110001415014267,
+      "step": 322
+    },
+    {
+      "epoch": 0.05383333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.7447265625,
+      "learning_rate": 9.929104283232362e-05,
+      "loss": 7.2991,
+      "loss/crossentropy": 1.5034100413322449,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1632218137383461,
+      "step": 323
+    },
+    {
+      "epoch": 0.054,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.6593098958333334,
+      "learning_rate": 9.928664305099314e-05,
+      "loss": 7.4129,
+      "loss/crossentropy": 1.6899998188018799,
+      "loss/hidden": 3.734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24160628765821457,
+      "step": 324
+    },
+    {
+      "epoch": 0.05416666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.8684895833333333,
+      "learning_rate": 9.928222975744991e-05,
+      "loss": 7.5294,
+      "loss/crossentropy": 1.9032530784606934,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20945952832698822,
+      "step": 325
+    },
+    {
+      "epoch": 0.05433333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.7494140625,
+      "learning_rate": 9.927780295290389e-05,
+      "loss": 7.4902,
+      "loss/crossentropy": 2.1650781631469727,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21734672039747238,
+      "step": 326
+    },
+    {
+      "epoch": 0.0545,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.9447265625,
+      "learning_rate": 9.927336263856872e-05,
+      "loss": 7.2849,
+      "loss/crossentropy": 1.3737293183803558,
+      "loss/hidden": 3.859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19318902119994164,
+      "step": 327
+    },
+    {
+      "epoch": 0.05466666666666667,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.875,
+      "learning_rate": 9.926890881566171e-05,
+      "loss": 7.4656,
+      "loss/crossentropy": 1.3970333486795425,
+      "loss/hidden": 3.84765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25607638619840145,
+      "step": 328
+    },
+    {
+      "epoch": 0.05483333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.928125,
+      "learning_rate": 9.926444148540393e-05,
+      "loss": 7.5143,
+      "loss/crossentropy": 2.113563746213913,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19357622042298317,
+      "step": 329
+    },
+    {
+      "epoch": 0.055,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.9125,
+      "learning_rate": 9.925996064902011e-05,
+      "loss": 7.6879,
+      "loss/crossentropy": 1.4198494106531143,
+      "loss/hidden": 3.92578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19689097069203854,
+      "step": 330
+    },
+    {
+      "epoch": 0.05516666666666667,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.9114583333333333,
+      "learning_rate": 9.92554663077387e-05,
+      "loss": 7.3615,
+      "loss/crossentropy": 1.9942026734352112,
+      "loss/hidden": 3.7734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2034350484609604,
+      "step": 331
+    },
+    {
+      "epoch": 0.05533333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.8643229166666666,
+      "learning_rate": 9.925095846279184e-05,
+      "loss": 7.5943,
+      "loss/crossentropy": 1.7958940863609314,
+      "loss/hidden": 3.8515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2254759781062603,
+      "step": 332
+    },
+    {
+      "epoch": 0.0555,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.3447916666666666,
+      "learning_rate": 9.924643711541539e-05,
+      "loss": 7.6929,
+      "loss/crossentropy": 2.2087718546390533,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.217594176530838,
+      "step": 333
+    },
+    {
+      "epoch": 0.05566666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.5572265625,
+      "learning_rate": 9.92419022668489e-05,
+      "loss": 6.992,
+      "loss/crossentropy": 2.0006081461906433,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1748015247285366,
+      "step": 334
+    },
+    {
+      "epoch": 0.05583333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.603125,
+      "learning_rate": 9.923735391833564e-05,
+      "loss": 7.1254,
+      "loss/crossentropy": 1.3763177990913391,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16850856319069862,
+      "step": 335
+    },
+    {
+      "epoch": 0.056,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.5259765625,
+      "learning_rate": 9.923279207112255e-05,
+      "loss": 7.5483,
+      "loss/crossentropy": 2.0636999011039734,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19349276646971703,
+      "step": 336
+    },
+    {
+      "epoch": 0.05616666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.27890625,
+      "learning_rate": 9.922821672646027e-05,
+      "loss": 7.5106,
+      "loss/crossentropy": 2.108673393726349,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1966811791062355,
+      "step": 337
+    },
+    {
+      "epoch": 0.05633333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.1582682291666666,
+      "learning_rate": 9.922362788560319e-05,
+      "loss": 7.4267,
+      "loss/crossentropy": 1.1151272654533386,
+      "loss/hidden": 3.9375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22225746139883995,
+      "step": 338
+    },
+    {
+      "epoch": 0.0565,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.1747395833333334,
+      "learning_rate": 9.921902554980934e-05,
+      "loss": 7.04,
+      "loss/crossentropy": 1.8475845456123352,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21776658669114113,
+      "step": 339
+    },
+    {
+      "epoch": 0.056666666666666664,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.2489583333333334,
+      "learning_rate": 9.921440972034049e-05,
+      "loss": 7.5363,
+      "loss/crossentropy": 1.732244223356247,
+      "loss/hidden": 3.73046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18700235709547997,
+      "step": 340
+    },
+    {
+      "epoch": 0.05683333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.1958333333333333,
+      "learning_rate": 9.92097803984621e-05,
+      "loss": 7.2846,
+      "loss/crossentropy": 1.6501949727535248,
+      "loss/hidden": 3.765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21733957901597023,
+      "step": 341
+    },
+    {
+      "epoch": 0.057,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.1306640625,
+      "learning_rate": 9.920513758544332e-05,
+      "loss": 7.4429,
+      "loss/crossentropy": 2.1270740926265717,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16204610094428062,
+      "step": 342
+    },
+    {
+      "epoch": 0.057166666666666664,
+      "grad_norm": 34.75,
+      "grad_norm_var": 1.5955729166666666,
+      "learning_rate": 9.920048128255699e-05,
+      "loss": 7.7165,
+      "loss/crossentropy": 1.5288592278957367,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17006384581327438,
+      "step": 343
+    },
+    {
+      "epoch": 0.05733333333333333,
+      "grad_norm": 34.75,
+      "grad_norm_var": 2.126822916666667,
+      "learning_rate": 9.919581149107968e-05,
+      "loss": 7.4922,
+      "loss/crossentropy": 1.3599306643009186,
+      "loss/hidden": 4.02734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30936116725206375,
+      "step": 344
+    },
+    {
+      "epoch": 0.0575,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.100455729166667,
+      "learning_rate": 9.919112821229163e-05,
+      "loss": 7.6191,
+      "loss/crossentropy": 1.263101488351822,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21637150272727013,
+      "step": 345
+    },
+    {
+      "epoch": 0.057666666666666665,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.1009765625,
+      "learning_rate": 9.918643144747681e-05,
+      "loss": 7.8287,
+      "loss/crossentropy": 1.9307979047298431,
+      "loss/hidden": 3.92578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3323311507701874,
+      "step": 346
+    },
+    {
+      "epoch": 0.057833333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.1077473958333335,
+      "learning_rate": 9.918172119792282e-05,
+      "loss": 7.665,
+      "loss/crossentropy": 2.110259026288986,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2538214847445488,
+      "step": 347
+    },
+    {
+      "epoch": 0.058,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.0530598958333335,
+      "learning_rate": 9.917699746492104e-05,
+      "loss": 7.3206,
+      "loss/crossentropy": 1.6958958506584167,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17854224145412445,
+      "step": 348
+    },
+    {
+      "epoch": 0.058166666666666665,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.106705729166667,
+      "learning_rate": 9.917226024976649e-05,
+      "loss": 7.5838,
+      "loss/crossentropy": 1.89556485414505,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.252014197409153,
+      "step": 349
+    },
+    {
+      "epoch": 0.058333333333333334,
+      "grad_norm": 34.25,
+      "grad_norm_var": 2.2306640625,
+      "learning_rate": 9.91675095537579e-05,
+      "loss": 7.6167,
+      "loss/crossentropy": 2.5545510947704315,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20364287123084068,
+      "step": 350
+    },
+    {
+      "epoch": 0.0585,
+      "grad_norm": 35.5,
+      "grad_norm_var": 2.796875,
+      "learning_rate": 9.916274537819775e-05,
+      "loss": 8.0692,
+      "loss/crossentropy": 1.9260480403900146,
+      "loss/hidden": 3.734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2535390965640545,
+      "step": 351
+    },
+    {
+      "epoch": 0.058666666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.0587890625,
+      "learning_rate": 9.915796772439207e-05,
+      "loss": 6.9244,
+      "loss/crossentropy": 1.3378092050552368,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1299455240368843,
+      "step": 352
+    },
+    {
+      "epoch": 0.058833333333333335,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.943489583333333,
+      "learning_rate": 9.915317659365077e-05,
+      "loss": 7.1952,
+      "loss/crossentropy": 2.034640520811081,
+      "loss/hidden": 3.8203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21266617625951767,
+      "step": 353
+    },
+    {
+      "epoch": 0.059,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.088997395833333,
+      "learning_rate": 9.914837198728733e-05,
+      "loss": 7.1558,
+      "loss/crossentropy": 2.037965625524521,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1865476667881012,
+      "step": 354
+    },
+    {
+      "epoch": 0.059166666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.8997395833333335,
+      "learning_rate": 9.914355390661896e-05,
+      "loss": 7.4129,
+      "loss/crossentropy": 1.7422327101230621,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17895380780100822,
+      "step": 355
+    },
+    {
+      "epoch": 0.059333333333333335,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.90625,
+      "learning_rate": 9.913872235296657e-05,
+      "loss": 7.8146,
+      "loss/crossentropy": 1.9063669443130493,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20077048987150192,
+      "step": 356
+    },
+    {
+      "epoch": 0.0595,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.0608723958333335,
+      "learning_rate": 9.913387732765475e-05,
+      "loss": 7.0098,
+      "loss/crossentropy": 1.535146176815033,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23291584476828575,
+      "step": 357
+    },
+    {
+      "epoch": 0.059666666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.135872395833333,
+      "learning_rate": 9.91290188320118e-05,
+      "loss": 6.9395,
+      "loss/crossentropy": 1.5794707238674164,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20209918171167374,
+      "step": 358
+    },
+    {
+      "epoch": 0.059833333333333336,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.7822265625,
+      "learning_rate": 9.91241468673697e-05,
+      "loss": 7.1428,
+      "loss/crossentropy": 1.551845133304596,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16993633657693863,
+      "step": 359
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 28.375,
+      "grad_norm_var": 3.09765625,
+      "learning_rate": 9.911926143506412e-05,
+      "loss": 7.1726,
+      "loss/crossentropy": 2.04102224111557,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19895951822400093,
+      "step": 360
+    },
+    {
+      "epoch": 0.06016666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.2681640625,
+      "learning_rate": 9.911436253643445e-05,
+      "loss": 7.4218,
+      "loss/crossentropy": 1.2187197357416153,
+      "loss/hidden": 3.890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22673745080828667,
+      "step": 361
+    },
+    {
+      "epoch": 0.060333333333333336,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.3119140625,
+      "learning_rate": 9.910945017282372e-05,
+      "loss": 7.0264,
+      "loss/crossentropy": 1.0910635814070702,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1586693823337555,
+      "step": 362
+    },
+    {
+      "epoch": 0.0605,
+      "grad_norm": 31.625,
+      "grad_norm_var": 3.29140625,
+      "learning_rate": 9.91045243455787e-05,
+      "loss": 7.4063,
+      "loss/crossentropy": 1.703506886959076,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1660308763384819,
+      "step": 363
+    },
+    {
+      "epoch": 0.06066666666666667,
+      "grad_norm": 34.5,
+      "grad_norm_var": 3.709375,
+      "learning_rate": 9.909958505604984e-05,
+      "loss": 7.2245,
+      "loss/crossentropy": 1.409069374203682,
+      "loss/hidden": 3.7109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22200162708759308,
+      "step": 364
+    },
+    {
+      "epoch": 0.060833333333333336,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.66015625,
+      "learning_rate": 9.909463230559127e-05,
+      "loss": 7.655,
+      "loss/crossentropy": 1.4122879952192307,
+      "loss/hidden": 3.83984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18040809780359268,
+      "step": 365
+    },
+    {
+      "epoch": 0.061,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.2874348958333335,
+      "learning_rate": 9.908966609556079e-05,
+      "loss": 8.0707,
+      "loss/crossentropy": 2.1443492472171783,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1719832681119442,
+      "step": 366
+    },
+    {
+      "epoch": 0.06116666666666667,
+      "grad_norm": 35.25,
+      "grad_norm_var": 3.1645182291666667,
+      "learning_rate": 9.908468642731995e-05,
+      "loss": 7.4423,
+      "loss/crossentropy": 1.669471025466919,
+      "loss/hidden": 3.84765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.194247554987669,
+      "step": 367
+    },
+    {
+      "epoch": 0.06133333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.096809895833333,
+      "learning_rate": 9.907969330223395e-05,
+      "loss": 7.497,
+      "loss/crossentropy": 1.650326743721962,
+      "loss/hidden": 3.71875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18958208337426186,
+      "step": 368
+    },
+    {
+      "epoch": 0.0615,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.1416015625,
+      "learning_rate": 9.907468672167165e-05,
+      "loss": 7.1178,
+      "loss/crossentropy": 1.2656744867563248,
+      "loss/hidden": 3.71484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1717012356966734,
+      "step": 369
+    },
+    {
+      "epoch": 0.06166666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.9494140625,
+      "learning_rate": 9.906966668700567e-05,
+      "loss": 7.068,
+      "loss/crossentropy": 1.3486099988222122,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1758405789732933,
+      "step": 370
+    },
+    {
+      "epoch": 0.06183333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.919791666666667,
+      "learning_rate": 9.906463319961225e-05,
+      "loss": 7.6119,
+      "loss/crossentropy": 2.083003431558609,
+      "loss/hidden": 3.81640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21161380410194397,
+      "step": 371
+    },
+    {
+      "epoch": 0.062,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.94765625,
+      "learning_rate": 9.90595862608714e-05,
+      "loss": 7.1607,
+      "loss/crossentropy": 1.420634001493454,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16716723702847958,
+      "step": 372
+    },
+    {
+      "epoch": 0.06216666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.8666015625,
+      "learning_rate": 9.90545258721667e-05,
+      "loss": 7.2752,
+      "loss/crossentropy": 2.1014665961265564,
+      "loss/hidden": 3.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15661929920315742,
+      "step": 373
+    },
+    {
+      "epoch": 0.06233333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.1009765625,
+      "learning_rate": 9.904945203488554e-05,
+      "loss": 7.1848,
+      "loss/crossentropy": 1.617295429110527,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20422854647040367,
+      "step": 374
+    },
+    {
+      "epoch": 0.0625,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.0306640625,
+      "learning_rate": 9.904436475041891e-05,
+      "loss": 7.4797,
+      "loss/crossentropy": 2.26700422167778,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18778881430625916,
+      "step": 375
+    },
+    {
+      "epoch": 0.06266666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.9375,
+      "learning_rate": 9.903926402016153e-05,
+      "loss": 7.4606,
+      "loss/crossentropy": 1.9848062098026276,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26049788668751717,
+      "step": 376
+    },
+    {
+      "epoch": 0.06283333333333334,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.7978515625,
+      "learning_rate": 9.903414984551179e-05,
+      "loss": 7.249,
+      "loss/crossentropy": 1.9200113117694855,
+      "loss/hidden": 3.6484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23478441685438156,
+      "step": 377
+    },
+    {
+      "epoch": 0.063,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.8384765625,
+      "learning_rate": 9.902902222787175e-05,
+      "loss": 7.4366,
+      "loss/crossentropy": 1.7129599452018738,
+      "loss/hidden": 3.80078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18362978473305702,
+      "step": 378
+    },
+    {
+      "epoch": 0.06316666666666666,
+      "grad_norm": 34.0,
+      "grad_norm_var": 2.91640625,
+      "learning_rate": 9.902388116864722e-05,
+      "loss": 7.4767,
+      "loss/crossentropy": 1.3501964658498764,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1189290564507246,
+      "step": 379
+    },
+    {
+      "epoch": 0.06333333333333334,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.7309895833333333,
+      "learning_rate": 9.901872666924764e-05,
+      "loss": 7.3649,
+      "loss/crossentropy": 2.340759515762329,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20152597874403,
+      "step": 380
+    },
+    {
+      "epoch": 0.0635,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.716666666666667,
+      "learning_rate": 9.901355873108609e-05,
+      "loss": 7.2519,
+      "loss/crossentropy": 1.875748872756958,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19418711960315704,
+      "step": 381
+    },
+    {
+      "epoch": 0.06366666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.5895182291666665,
+      "learning_rate": 9.900837735557947e-05,
+      "loss": 7.3722,
+      "loss/crossentropy": 1.8264379799365997,
+      "loss/hidden": 3.72265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21041027829051018,
+      "step": 382
+    },
+    {
+      "epoch": 0.06383333333333334,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.1556640625,
+      "learning_rate": 9.900318254414821e-05,
+      "loss": 7.4797,
+      "loss/crossentropy": 2.100499212741852,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21592680737376213,
+      "step": 383
+    },
+    {
+      "epoch": 0.064,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.5125,
+      "learning_rate": 9.899797429821656e-05,
+      "loss": 7.0055,
+      "loss/crossentropy": 1.8394896984100342,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21314410492777824,
+      "step": 384
+    },
+    {
+      "epoch": 0.06416666666666666,
+      "grad_norm": 37.0,
+      "grad_norm_var": 3.9247395833333334,
+      "learning_rate": 9.899275261921234e-05,
+      "loss": 7.7596,
+      "loss/crossentropy": 1.4905634820461273,
+      "loss/hidden": 3.8828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21177972108125687,
+      "step": 385
+    },
+    {
+      "epoch": 0.06433333333333334,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.98515625,
+      "learning_rate": 9.898751750856713e-05,
+      "loss": 7.8125,
+      "loss/crossentropy": 1.5256913900375366,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17715035378932953,
+      "step": 386
+    },
+    {
+      "epoch": 0.0645,
+      "grad_norm": 30.25,
+      "grad_norm_var": 4.3025390625,
+      "learning_rate": 9.898226896771619e-05,
+      "loss": 7.4127,
+      "loss/crossentropy": 1.7017391622066498,
+      "loss/hidden": 3.9765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36126620322465897,
+      "step": 387
+    },
+    {
+      "epoch": 0.06466666666666666,
+      "grad_norm": 29.75,
+      "grad_norm_var": 4.735872395833334,
+      "learning_rate": 9.897700699809837e-05,
+      "loss": 7.3459,
+      "loss/crossentropy": 2.117743670940399,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20285234227776527,
+      "step": 388
+    },
+    {
+      "epoch": 0.06483333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.59140625,
+      "learning_rate": 9.897173160115632e-05,
+      "loss": 7.3591,
+      "loss/crossentropy": 1.51130610704422,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15151387825608253,
+      "step": 389
+    },
+    {
+      "epoch": 0.065,
+      "grad_norm": 33.75,
+      "grad_norm_var": 4.47890625,
+      "learning_rate": 9.896644277833631e-05,
+      "loss": 7.651,
+      "loss/crossentropy": 1.2360347509384155,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14697962626814842,
+      "step": 390
+    },
+    {
+      "epoch": 0.06516666666666666,
+      "grad_norm": 34.75,
+      "grad_norm_var": 4.85625,
+      "learning_rate": 9.896114053108829e-05,
+      "loss": 7.6821,
+      "loss/crossentropy": 1.6834103763103485,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30382155627012253,
+      "step": 391
+    },
+    {
+      "epoch": 0.06533333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.845572916666667,
+      "learning_rate": 9.895582486086592e-05,
+      "loss": 7.3562,
+      "loss/crossentropy": 1.5958560854196548,
+      "loss/hidden": 3.78515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1785712093114853,
+      "step": 392
+    },
+    {
+      "epoch": 0.0655,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.612434895833333,
+      "learning_rate": 9.89504957691265e-05,
+      "loss": 7.5119,
+      "loss/crossentropy": 1.34417524933815,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18383424542844296,
+      "step": 393
+    },
+    {
+      "epoch": 0.06566666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.5994140625,
+      "learning_rate": 9.894515325733103e-05,
+      "loss": 7.6725,
+      "loss/crossentropy": 1.6958922445774078,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.174136221408844,
+      "step": 394
+    },
+    {
+      "epoch": 0.06583333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.60625,
+      "learning_rate": 9.893979732694421e-05,
+      "loss": 7.1315,
+      "loss/crossentropy": 1.6653735637664795,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16760963946580887,
+      "step": 395
+    },
+    {
+      "epoch": 0.066,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.1884765625,
+      "learning_rate": 9.89344279794344e-05,
+      "loss": 7.1095,
+      "loss/crossentropy": 1.5993010103702545,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17182908952236176,
+      "step": 396
+    },
+    {
+      "epoch": 0.06616666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.340559895833334,
+      "learning_rate": 9.892904521627361e-05,
+      "loss": 7.0819,
+      "loss/crossentropy": 1.4599365592002869,
+      "loss/hidden": 3.8125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18065180256962776,
+      "step": 397
+    },
+    {
+      "epoch": 0.06633333333333333,
+      "grad_norm": 64.0,
+      "grad_norm_var": 66.31920572916667,
+      "learning_rate": 9.892364903893759e-05,
+      "loss": 7.313,
+      "loss/crossentropy": 2.08347424864769,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21691031754016876,
+      "step": 398
+    },
+    {
+      "epoch": 0.0665,
+      "grad_norm": 33.0,
+      "grad_norm_var": 66.39993489583334,
+      "learning_rate": 9.891823944890568e-05,
+      "loss": 7.1682,
+      "loss/crossentropy": 1.3314939439296722,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19921602681279182,
+      "step": 399
+    },
+    {
+      "epoch": 0.06666666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 65.73587239583334,
+      "learning_rate": 9.8912816447661e-05,
+      "loss": 7.2361,
+      "loss/crossentropy": 1.031214565038681,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15684598125517368,
+      "step": 400
+    },
+    {
+      "epoch": 0.06683333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 65.76666666666667,
+      "learning_rate": 9.890738003669029e-05,
+      "loss": 7.2621,
+      "loss/crossentropy": 2.0488553047180176,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18182619661092758,
+      "step": 401
+    },
+    {
+      "epoch": 0.067,
+      "grad_norm": 31.125,
+      "grad_norm_var": 66.3369140625,
+      "learning_rate": 9.890193021748395e-05,
+      "loss": 7.341,
+      "loss/crossentropy": 1.5651976019144058,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1589721255004406,
+      "step": 402
+    },
+    {
+      "epoch": 0.06716666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 65.9931640625,
+      "learning_rate": 9.88964669915361e-05,
+      "loss": 7.3422,
+      "loss/crossentropy": 1.7897996976971626,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14813226275146008,
+      "step": 403
+    },
+    {
+      "epoch": 0.06733333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 65.2666015625,
+      "learning_rate": 9.889099036034451e-05,
+      "loss": 7.2486,
+      "loss/crossentropy": 2.028013914823532,
+      "loss/hidden": 4.015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2523065656423569,
+      "step": 404
+    },
+    {
+      "epoch": 0.0675,
+      "grad_norm": 34.25,
+      "grad_norm_var": 64.75670572916667,
+      "learning_rate": 9.888550032541059e-05,
+      "loss": 7.6157,
+      "loss/crossentropy": 1.7820807695388794,
+      "loss/hidden": 3.8359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1764237843453884,
+      "step": 405
+    },
+    {
+      "epoch": 0.06766666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 65.39791666666666,
+      "learning_rate": 9.887999688823955e-05,
+      "loss": 7.3084,
+      "loss/crossentropy": 1.5304799228906631,
+      "loss/hidden": 3.9453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26905737072229385,
+      "step": 406
+    },
+    {
+      "epoch": 0.06783333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 65.54557291666667,
+      "learning_rate": 9.88744800503401e-05,
+      "loss": 7.4814,
+      "loss/crossentropy": 1.909523993730545,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22289742529392242,
+      "step": 407
+    },
+    {
+      "epoch": 0.068,
+      "grad_norm": 35.5,
+      "grad_norm_var": 65.42708333333333,
+      "learning_rate": 9.886894981322476e-05,
+      "loss": 7.5579,
+      "loss/crossentropy": 1.9321822822093964,
+      "loss/hidden": 4.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21899055689573288,
+      "step": 408
+    },
+    {
+      "epoch": 0.06816666666666667,
+      "grad_norm": 35.0,
+      "grad_norm_var": 65.45625,
+      "learning_rate": 9.886340617840968e-05,
+      "loss": 7.4951,
+      "loss/crossentropy": 1.745114415884018,
+      "loss/hidden": 3.74609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19060330465435982,
+      "step": 409
+    },
+    {
+      "epoch": 0.06833333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 66.01223958333334,
+      "learning_rate": 9.885784914741465e-05,
+      "loss": 7.514,
+      "loss/crossentropy": 2.1204441487789154,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19498489424586296,
+      "step": 410
+    },
+    {
+      "epoch": 0.0685,
+      "grad_norm": 30.5,
+      "grad_norm_var": 66.3900390625,
+      "learning_rate": 9.88522787217632e-05,
+      "loss": 7.2348,
+      "loss/crossentropy": 1.9602341055870056,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16496751084923744,
+      "step": 411
+    },
+    {
+      "epoch": 0.06866666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 65.15,
+      "learning_rate": 9.884669490298244e-05,
+      "loss": 7.1868,
+      "loss/crossentropy": 1.9193522334098816,
+      "loss/hidden": 3.7734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22780518978834152,
+      "step": 412
+    },
+    {
+      "epoch": 0.06883333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 65.78639322916666,
+      "learning_rate": 9.884109769260325e-05,
+      "loss": 7.2402,
+      "loss/crossentropy": 1.510194793343544,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18128687515854836,
+      "step": 413
+    },
+    {
+      "epoch": 0.069,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.3613932291666666,
+      "learning_rate": 9.883548709216013e-05,
+      "loss": 7.3283,
+      "loss/crossentropy": 1.604835420846939,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1667223460972309,
+      "step": 414
+    },
+    {
+      "epoch": 0.06916666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.60390625,
+      "learning_rate": 9.882986310319124e-05,
+      "loss": 7.3877,
+      "loss/crossentropy": 1.790344089269638,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17915276437997818,
+      "step": 415
+    },
+    {
+      "epoch": 0.06933333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.67265625,
+      "learning_rate": 9.882422572723844e-05,
+      "loss": 7.0171,
+      "loss/crossentropy": 2.0589410960674286,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21364706382155418,
+      "step": 416
+    },
+    {
+      "epoch": 0.0695,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.814583333333333,
+      "learning_rate": 9.881857496584726e-05,
+      "loss": 7.111,
+      "loss/crossentropy": 1.3649069666862488,
+      "loss/hidden": 3.74609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23485032096505165,
+      "step": 417
+    },
+    {
+      "epoch": 0.06966666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.8962890625,
+      "learning_rate": 9.881291082056685e-05,
+      "loss": 7.2745,
+      "loss/crossentropy": 1.6852681040763855,
+      "loss/hidden": 3.73046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23358748480677605,
+      "step": 418
+    },
+    {
+      "epoch": 0.06983333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.8541015625,
+      "learning_rate": 9.880723329295012e-05,
+      "loss": 7.4246,
+      "loss/crossentropy": 1.7184284329414368,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20467643439769745,
+      "step": 419
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.968489583333333,
+      "learning_rate": 9.880154238455356e-05,
+      "loss": 7.1977,
+      "loss/crossentropy": 1.639914184808731,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27305275201797485,
+      "step": 420
+    },
+    {
+      "epoch": 0.07016666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.6655598958333333,
+      "learning_rate": 9.879583809693738e-05,
+      "loss": 7.1936,
+      "loss/crossentropy": 1.7328845262527466,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14637303911149502,
+      "step": 421
+    },
+    {
+      "epoch": 0.07033333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.035416666666667,
+      "learning_rate": 9.879012043166542e-05,
+      "loss": 7.6863,
+      "loss/crossentropy": 1.6795161664485931,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17538448050618172,
+      "step": 422
+    },
+    {
+      "epoch": 0.0705,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.9895833333333335,
+      "learning_rate": 9.878438939030526e-05,
+      "loss": 7.4762,
+      "loss/crossentropy": 1.9363873898983002,
+      "loss/hidden": 3.85546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.42479467391967773,
+      "step": 423
+    },
+    {
+      "epoch": 0.07066666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.9259765625,
+      "learning_rate": 9.877864497442804e-05,
+      "loss": 7.6811,
+      "loss/crossentropy": 1.9779807329177856,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22114330902695656,
+      "step": 424
+    },
+    {
+      "epoch": 0.07083333333333333,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.3530598958333333,
+      "learning_rate": 9.877288718560866e-05,
+      "loss": 7.1135,
+      "loss/crossentropy": 1.2891452014446259,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12475473806262016,
+      "step": 425
+    },
+    {
+      "epoch": 0.071,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.6009765625,
+      "learning_rate": 9.876711602542563e-05,
+      "loss": 7.2409,
+      "loss/crossentropy": 1.786192610859871,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18037456832826138,
+      "step": 426
+    },
+    {
+      "epoch": 0.07116666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.8072265625,
+      "learning_rate": 9.876133149546118e-05,
+      "loss": 7.4427,
+      "loss/crossentropy": 1.9531354904174805,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1618359051644802,
+      "step": 427
+    },
+    {
+      "epoch": 0.07133333333333333,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.528580729166667,
+      "learning_rate": 9.875553359730114e-05,
+      "loss": 7.2901,
+      "loss/crossentropy": 1.4478426724672318,
+      "loss/hidden": 3.93359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1801770105957985,
+      "step": 428
+    },
+    {
+      "epoch": 0.0715,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.601822916666667,
+      "learning_rate": 9.874972233253504e-05,
+      "loss": 7.0818,
+      "loss/crossentropy": 1.631743460893631,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19842953979969025,
+      "step": 429
+    },
+    {
+      "epoch": 0.07166666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.595572916666667,
+      "learning_rate": 9.874389770275607e-05,
+      "loss": 7.5555,
+      "loss/crossentropy": 1.5108791291713715,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16849511303007603,
+      "step": 430
+    },
+    {
+      "epoch": 0.07183333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.6655598958333333,
+      "learning_rate": 9.87380597095611e-05,
+      "loss": 7.6512,
+      "loss/crossentropy": 1.9580343961715698,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21497854217886925,
+      "step": 431
+    },
+    {
+      "epoch": 0.072,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.7375,
+      "learning_rate": 9.873220835455064e-05,
+      "loss": 7.5562,
+      "loss/crossentropy": 1.9114446938037872,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15941833890974522,
+      "step": 432
+    },
+    {
+      "epoch": 0.07216666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.77890625,
+      "learning_rate": 9.872634363932887e-05,
+      "loss": 7.3277,
+      "loss/crossentropy": 1.5630664974451065,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15427661500871181,
+      "step": 433
+    },
+    {
+      "epoch": 0.07233333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.7166015625,
+      "learning_rate": 9.872046556550363e-05,
+      "loss": 7.2233,
+      "loss/crossentropy": 1.781863272190094,
+      "loss/hidden": 4.01171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.256728108972311,
+      "step": 434
+    },
+    {
+      "epoch": 0.0725,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.7264973958333334,
+      "learning_rate": 9.871457413468644e-05,
+      "loss": 7.226,
+      "loss/crossentropy": 1.3000462800264359,
+      "loss/hidden": 3.83984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21796053275465965,
+      "step": 435
+    },
+    {
+      "epoch": 0.07266666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.629166666666667,
+      "learning_rate": 9.870866934849248e-05,
+      "loss": 7.351,
+      "loss/crossentropy": 1.6976856887340546,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1842471994459629,
+      "step": 436
+    },
+    {
+      "epoch": 0.07283333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.508072916666667,
+      "learning_rate": 9.870275120854054e-05,
+      "loss": 7.5839,
+      "loss/crossentropy": 2.3903128504753113,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18865225464105606,
+      "step": 437
+    },
+    {
+      "epoch": 0.073,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.1864583333333334,
+      "learning_rate": 9.869681971645315e-05,
+      "loss": 7.4518,
+      "loss/crossentropy": 1.8442813605070114,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1843016818165779,
+      "step": 438
+    },
+    {
+      "epoch": 0.07316666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.218489583333333,
+      "learning_rate": 9.869087487385644e-05,
+      "loss": 7.6065,
+      "loss/crossentropy": 1.7373953759670258,
+      "loss/hidden": 3.7265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17827446945011616,
+      "step": 439
+    },
+    {
+      "epoch": 0.07333333333333333,
+      "grad_norm": 36.0,
+      "grad_norm_var": 3.442122395833333,
+      "learning_rate": 9.868491668238025e-05,
+      "loss": 7.2492,
+      "loss/crossentropy": 1.6353029012680054,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1661812923848629,
+      "step": 440
+    },
+    {
+      "epoch": 0.0735,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.787955729166667,
+      "learning_rate": 9.867894514365802e-05,
+      "loss": 7.4579,
+      "loss/crossentropy": 1.672650307416916,
+      "loss/hidden": 3.875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2703779861330986,
+      "step": 441
+    },
+    {
+      "epoch": 0.07366666666666667,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.1738932291666666,
+      "learning_rate": 9.867296025932688e-05,
+      "loss": 7.2466,
+      "loss/crossentropy": 2.3615481853485107,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19613812491297722,
+      "step": 442
+    },
+    {
+      "epoch": 0.07383333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.1738932291666666,
+      "learning_rate": 9.866696203102766e-05,
+      "loss": 7.2187,
+      "loss/crossentropy": 1.9904030561447144,
+      "loss/hidden": 3.890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2621547132730484,
+      "step": 443
+    },
+    {
+      "epoch": 0.074,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.8697916666666667,
+      "learning_rate": 9.866095046040478e-05,
+      "loss": 7.6522,
+      "loss/crossentropy": 1.8616594970226288,
+      "loss/hidden": 3.13671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16203163377940655,
+      "step": 444
+    },
+    {
+      "epoch": 0.07416666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.8809895833333334,
+      "learning_rate": 9.865492554910633e-05,
+      "loss": 7.303,
+      "loss/crossentropy": 1.496483102440834,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24139206856489182,
+      "step": 445
+    },
+    {
+      "epoch": 0.07433333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.7625,
+      "learning_rate": 9.86488872987841e-05,
+      "loss": 7.356,
+      "loss/crossentropy": 1.637431263923645,
+      "loss/hidden": 3.98046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2300253063440323,
+      "step": 446
+    },
+    {
+      "epoch": 0.0745,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.7634765625,
+      "learning_rate": 9.864283571109352e-05,
+      "loss": 7.8614,
+      "loss/crossentropy": 1.583919882774353,
+      "loss/hidden": 3.80078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2739395461976528,
+      "step": 447
+    },
+    {
+      "epoch": 0.07466666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.7244140625,
+      "learning_rate": 9.863677078769362e-05,
+      "loss": 7.5423,
+      "loss/crossentropy": 1.9376616775989532,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2414015680551529,
+      "step": 448
+    },
+    {
+      "epoch": 0.07483333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.5125,
+      "learning_rate": 9.863069253024719e-05,
+      "loss": 7.9243,
+      "loss/crossentropy": 1.9691549688577652,
+      "loss/hidden": 3.83203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21448491513729095,
+      "step": 449
+    },
+    {
+      "epoch": 0.075,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.5087890625,
+      "learning_rate": 9.862460094042056e-05,
+      "loss": 7.2924,
+      "loss/crossentropy": 1.596924401819706,
+      "loss/hidden": 3.9609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16464821249246597,
+      "step": 450
+    },
+    {
+      "epoch": 0.07516666666666667,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.5176432291666666,
+      "learning_rate": 9.861849601988383e-05,
+      "loss": 7.5016,
+      "loss/crossentropy": 1.297846108675003,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18855424225330353,
+      "step": 451
+    },
+    {
+      "epoch": 0.07533333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.9525390625,
+      "learning_rate": 9.861237777031068e-05,
+      "loss": 7.2361,
+      "loss/crossentropy": 1.8280795812606812,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18522650003433228,
+      "step": 452
+    },
+    {
+      "epoch": 0.0755,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.87890625,
+      "learning_rate": 9.860624619337844e-05,
+      "loss": 7.3614,
+      "loss/crossentropy": 1.5700929462909698,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16174762323498726,
+      "step": 453
+    },
+    {
+      "epoch": 0.07566666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.7587890625,
+      "learning_rate": 9.860010129076813e-05,
+      "loss": 6.8208,
+      "loss/crossentropy": 1.4433760195970535,
+      "loss/hidden": 3.7421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16576368734240532,
+      "step": 454
+    },
+    {
+      "epoch": 0.07583333333333334,
+      "grad_norm": 34.25,
+      "grad_norm_var": 2.035872395833333,
+      "learning_rate": 9.859394306416444e-05,
+      "loss": 7.6301,
+      "loss/crossentropy": 2.117895543575287,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21888913959264755,
+      "step": 455
+    },
+    {
+      "epoch": 0.076,
+      "grad_norm": 33.5,
+      "grad_norm_var": 1.1843098958333333,
+      "learning_rate": 9.858777151525564e-05,
+      "loss": 7.0118,
+      "loss/crossentropy": 1.6520485877990723,
+      "loss/hidden": 3.84765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1626514047384262,
+      "step": 456
+    },
+    {
+      "epoch": 0.07616666666666666,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.2122395833333333,
+      "learning_rate": 9.85815866457337e-05,
+      "loss": 7.1072,
+      "loss/crossentropy": 2.178823322057724,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18832968920469284,
+      "step": 457
+    },
+    {
+      "epoch": 0.07633333333333334,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.2122395833333333,
+      "learning_rate": 9.857538845729426e-05,
+      "loss": 7.4747,
+      "loss/crossentropy": 1.6742831617593765,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1979827582836151,
+      "step": 458
+    },
+    {
+      "epoch": 0.0765,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.1739583333333334,
+      "learning_rate": 9.856917695163658e-05,
+      "loss": 7.2882,
+      "loss/crossentropy": 1.699859082698822,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1798916831612587,
+      "step": 459
+    },
+    {
+      "epoch": 0.07666666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.1497395833333333,
+      "learning_rate": 9.856295213046357e-05,
+      "loss": 7.5482,
+      "loss/crossentropy": 2.0418995320796967,
+      "loss/hidden": 3.75,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2375405840575695,
+      "step": 460
+    },
+    {
+      "epoch": 0.07683333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.2364583333333334,
+      "learning_rate": 9.855671399548181e-05,
+      "loss": 7.3054,
+      "loss/crossentropy": 1.9323958605527878,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22189056873321533,
+      "step": 461
+    },
+    {
+      "epoch": 0.077,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.32890625,
+      "learning_rate": 9.855046254840151e-05,
+      "loss": 7.5128,
+      "loss/crossentropy": 1.9981287121772766,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1936916969716549,
+      "step": 462
+    },
+    {
+      "epoch": 0.07716666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.3,
+      "learning_rate": 9.854419779093655e-05,
+      "loss": 7.1306,
+      "loss/crossentropy": 1.2438909262418747,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17436053417623043,
+      "step": 463
+    },
+    {
+      "epoch": 0.07733333333333334,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.6176432291666667,
+      "learning_rate": 9.853791972480445e-05,
+      "loss": 7.3704,
+      "loss/crossentropy": 1.6113817691802979,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15345285832881927,
+      "step": 464
+    },
+    {
+      "epoch": 0.0775,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.7830729166666666,
+      "learning_rate": 9.853162835172637e-05,
+      "loss": 7.4493,
+      "loss/crossentropy": 2.1560588479042053,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16945412755012512,
+      "step": 465
+    },
+    {
+      "epoch": 0.07766666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.7760416666666667,
+      "learning_rate": 9.852532367342713e-05,
+      "loss": 7.2615,
+      "loss/crossentropy": 1.7896616458892822,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18198765441775322,
+      "step": 466
+    },
+    {
+      "epoch": 0.07783333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.8080729166666667,
+      "learning_rate": 9.851900569163519e-05,
+      "loss": 7.4714,
+      "loss/crossentropy": 1.9437064826488495,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1927110031247139,
+      "step": 467
+    },
+    {
+      "epoch": 0.078,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.4393229166666666,
+      "learning_rate": 9.851267440808265e-05,
+      "loss": 7.3594,
+      "loss/crossentropy": 1.3540785908699036,
+      "loss/hidden": 3.98046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20039800181984901,
+      "step": 468
+    },
+    {
+      "epoch": 0.07816666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.4830729166666667,
+      "learning_rate": 9.85063298245053e-05,
+      "loss": 7.1811,
+      "loss/crossentropy": 1.8484355509281158,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1995554454624653,
+      "step": 469
+    },
+    {
+      "epoch": 0.07833333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.5926432291666666,
+      "learning_rate": 9.84999719426425e-05,
+      "loss": 7.7469,
+      "loss/crossentropy": 1.7779522836208344,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1738000623881817,
+      "step": 470
+    },
+    {
+      "epoch": 0.0785,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.3119140625,
+      "learning_rate": 9.849360076423734e-05,
+      "loss": 7.5347,
+      "loss/crossentropy": 1.979307770729065,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16858231276273727,
+      "step": 471
+    },
+    {
+      "epoch": 0.07866666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.14375,
+      "learning_rate": 9.84872162910365e-05,
+      "loss": 7.2582,
+      "loss/crossentropy": 1.9429263174533844,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1769695095717907,
+      "step": 472
+    },
+    {
+      "epoch": 0.07883333333333334,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.61015625,
+      "learning_rate": 9.84808185247903e-05,
+      "loss": 7.573,
+      "loss/crossentropy": 2.1421655118465424,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17848249524831772,
+      "step": 473
+    },
+    {
+      "epoch": 0.079,
+      "grad_norm": 36.75,
+      "grad_norm_var": 3.32890625,
+      "learning_rate": 9.847440746725275e-05,
+      "loss": 7.7908,
+      "loss/crossentropy": 1.3447946310043335,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18741309270262718,
+      "step": 474
+    },
+    {
+      "epoch": 0.07916666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.338997395833333,
+      "learning_rate": 9.846798312018146e-05,
+      "loss": 7.5913,
+      "loss/crossentropy": 1.8275827169418335,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24438031017780304,
+      "step": 475
+    },
+    {
+      "epoch": 0.07933333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.32890625,
+      "learning_rate": 9.846154548533773e-05,
+      "loss": 7.4547,
+      "loss/crossentropy": 1.9462362825870514,
+      "loss/hidden": 3.71875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21810081601142883,
+      "step": 476
+    },
+    {
+      "epoch": 0.0795,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.4309895833333335,
+      "learning_rate": 9.845509456448643e-05,
+      "loss": 7.3036,
+      "loss/crossentropy": 1.7459705471992493,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16330504044890404,
+      "step": 477
+    },
+    {
+      "epoch": 0.07966666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.4895833333333335,
+      "learning_rate": 9.844863035939615e-05,
+      "loss": 7.5502,
+      "loss/crossentropy": 1.9055275321006775,
+      "loss/hidden": 3.796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28089214861392975,
+      "step": 478
+    },
+    {
+      "epoch": 0.07983333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.5082682291666667,
+      "learning_rate": 9.844215287183909e-05,
+      "loss": 7.1068,
+      "loss/crossentropy": 1.7861053347587585,
+      "loss/hidden": 3.79296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25769438222050667,
+      "step": 479
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 34.5,
+      "grad_norm_var": 3.732291666666667,
+      "learning_rate": 9.843566210359106e-05,
+      "loss": 7.0123,
+      "loss/crossentropy": 1.8309744149446487,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15748615190386772,
+      "step": 480
+    },
+    {
+      "epoch": 0.08016666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.542122395833333,
+      "learning_rate": 9.842915805643155e-05,
+      "loss": 7.2267,
+      "loss/crossentropy": 1.8069412112236023,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1999414637684822,
+      "step": 481
+    },
+    {
+      "epoch": 0.08033333333333334,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.6910807291666665,
+      "learning_rate": 9.842264073214371e-05,
+      "loss": 7.073,
+      "loss/crossentropy": 1.5245321840047836,
+      "loss/hidden": 3.87890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22637157142162323,
+      "step": 482
+    },
+    {
+      "epoch": 0.0805,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.7676432291666666,
+      "learning_rate": 9.841611013251429e-05,
+      "loss": 7.5032,
+      "loss/crossentropy": 1.4558211117982864,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18064381927251816,
+      "step": 483
+    },
+    {
+      "epoch": 0.08066666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.761458333333333,
+      "learning_rate": 9.840956625933367e-05,
+      "loss": 7.281,
+      "loss/crossentropy": 1.9358763992786407,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18417445197701454,
+      "step": 484
+    },
+    {
+      "epoch": 0.08083333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.9369140625,
+      "learning_rate": 9.840300911439591e-05,
+      "loss": 7.3674,
+      "loss/crossentropy": 1.7015015482902527,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20439771935343742,
+      "step": 485
+    },
+    {
+      "epoch": 0.081,
+      "grad_norm": 44.25,
+      "grad_norm_var": 13.275455729166667,
+      "learning_rate": 9.839643869949866e-05,
+      "loss": 7.7514,
+      "loss/crossentropy": 2.1199212074279785,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22786704450845718,
+      "step": 486
+    },
+    {
+      "epoch": 0.08116666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 13.0056640625,
+      "learning_rate": 9.838985501644328e-05,
+      "loss": 7.1685,
+      "loss/crossentropy": 1.7645389437675476,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18691401928663254,
+      "step": 487
+    },
+    {
+      "epoch": 0.08133333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 13.152018229166666,
+      "learning_rate": 9.83832580670347e-05,
+      "loss": 7.4557,
+      "loss/crossentropy": 1.8709838688373566,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20340480655431747,
+      "step": 488
+    },
+    {
+      "epoch": 0.0815,
+      "grad_norm": 30.125,
+      "grad_norm_var": 12.5947265625,
+      "learning_rate": 9.837664785308149e-05,
+      "loss": 7.2281,
+      "loss/crossentropy": 2.070108652114868,
+      "loss/hidden": 3.71484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24872201681137085,
+      "step": 489
+    },
+    {
+      "epoch": 0.08166666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 11.5400390625,
+      "learning_rate": 9.837002437639593e-05,
+      "loss": 7.7274,
+      "loss/crossentropy": 2.13846293091774,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18616901524364948,
+      "step": 490
+    },
+    {
+      "epoch": 0.08183333333333333,
+      "grad_norm": 36.75,
+      "grad_norm_var": 12.466666666666667,
+      "learning_rate": 9.836338763879385e-05,
+      "loss": 7.4742,
+      "loss/crossentropy": 1.9132429659366608,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15966995060443878,
+      "step": 491
+    },
+    {
+      "epoch": 0.082,
+      "grad_norm": 38.0,
+      "grad_norm_var": 13.966666666666667,
+      "learning_rate": 9.835673764209474e-05,
+      "loss": 7.6467,
+      "loss/crossentropy": 2.3943788409233093,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1816554293036461,
+      "step": 492
+    },
+    {
+      "epoch": 0.08216666666666667,
+      "grad_norm": 40.25,
+      "grad_norm_var": 15.869791666666666,
+      "learning_rate": 9.835007438812177e-05,
+      "loss": 7.6718,
+      "loss/crossentropy": 1.3702804148197174,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1656254678964615,
+      "step": 493
+    },
+    {
+      "epoch": 0.08233333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 16.619791666666668,
+      "learning_rate": 9.834339787870166e-05,
+      "loss": 7.3474,
+      "loss/crossentropy": 2.013328567147255,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17504287511110306,
+      "step": 494
+    },
+    {
+      "epoch": 0.0825,
+      "grad_norm": 33.0,
+      "grad_norm_var": 16.501822916666665,
+      "learning_rate": 9.833670811566485e-05,
+      "loss": 7.8286,
+      "loss/crossentropy": 1.9959434270858765,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1881503090262413,
+      "step": 495
+    },
+    {
+      "epoch": 0.08266666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 16.998372395833332,
+      "learning_rate": 9.833000510084537e-05,
+      "loss": 7.7687,
+      "loss/crossentropy": 2.2734366953372955,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16209129244089127,
+      "step": 496
+    },
+    {
+      "epoch": 0.08283333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 17.259309895833333,
+      "learning_rate": 9.832328883608088e-05,
+      "loss": 7.016,
+      "loss/crossentropy": 1.9175287038087845,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17320701107382774,
+      "step": 497
+    },
+    {
+      "epoch": 0.083,
+      "grad_norm": 29.875,
+      "grad_norm_var": 17.534375,
+      "learning_rate": 9.83165593232127e-05,
+      "loss": 7.2644,
+      "loss/crossentropy": 2.0322545766830444,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17185277864336967,
+      "step": 498
+    },
+    {
+      "epoch": 0.08316666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 17.59140625,
+      "learning_rate": 9.830981656408574e-05,
+      "loss": 7.6535,
+      "loss/crossentropy": 1.8783454298973083,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2332802191376686,
+      "step": 499
+    },
+    {
+      "epoch": 0.08333333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 17.86015625,
+      "learning_rate": 9.830306056054858e-05,
+      "loss": 7.3534,
+      "loss/crossentropy": 1.6666748374700546,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23043451458215714,
+      "step": 500
+    },
+    {
+      "epoch": 0.0835,
+      "grad_norm": 32.0,
+      "grad_norm_var": 17.2712890625,
+      "learning_rate": 9.829629131445342e-05,
+      "loss": 7.964,
+      "loss/crossentropy": 1.8988508880138397,
+      "loss/hidden": 3.859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32320621982216835,
+      "step": 501
+    },
+    {
+      "epoch": 0.08366666666666667,
+      "grad_norm": 36.0,
+      "grad_norm_var": 9.6744140625,
+      "learning_rate": 9.828950882765608e-05,
+      "loss": 7.7202,
+      "loss/crossentropy": 1.911725103855133,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27852168306708336,
+      "step": 502
+    },
+    {
+      "epoch": 0.08383333333333333,
+      "grad_norm": 38.5,
+      "grad_norm_var": 11.5556640625,
+      "learning_rate": 9.828271310201601e-05,
+      "loss": 7.521,
+      "loss/crossentropy": 1.8464531302452087,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19315709546208382,
+      "step": 503
+    },
+    {
+      "epoch": 0.084,
+      "grad_norm": 34.25,
+      "grad_norm_var": 10.964322916666667,
+      "learning_rate": 9.827590413939632e-05,
+      "loss": 7.3371,
+      "loss/crossentropy": 1.4560853987932205,
+      "loss/hidden": 3.87890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2105232998728752,
+      "step": 504
+    },
+    {
+      "epoch": 0.08416666666666667,
+      "grad_norm": 40.0,
+      "grad_norm_var": 12.512434895833334,
+      "learning_rate": 9.82690819416637e-05,
+      "loss": 7.3674,
+      "loss/crossentropy": 2.154931664466858,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2120799981057644,
+      "step": 505
+    },
+    {
+      "epoch": 0.08433333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 12.581184895833333,
+      "learning_rate": 9.826224651068852e-05,
+      "loss": 7.4003,
+      "loss/crossentropy": 2.2104691863059998,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20807649940252304,
+      "step": 506
+    },
+    {
+      "epoch": 0.0845,
+      "grad_norm": 32.0,
+      "grad_norm_var": 12.3634765625,
+      "learning_rate": 9.825539784834472e-05,
+      "loss": 8.0154,
+      "loss/crossentropy": 1.8775488883256912,
+      "loss/hidden": 3.85546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29413507506251335,
+      "step": 507
+    },
+    {
+      "epoch": 0.08466666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 11.7619140625,
+      "learning_rate": 9.824853595650991e-05,
+      "loss": 7.2414,
+      "loss/crossentropy": 1.8616193979978561,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22998263500630856,
+      "step": 508
+    },
+    {
+      "epoch": 0.08483333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 8.452018229166667,
+      "learning_rate": 9.824166083706534e-05,
+      "loss": 7.461,
+      "loss/crossentropy": 2.28107687830925,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1916147843003273,
+      "step": 509
+    },
+    {
+      "epoch": 0.085,
+      "grad_norm": 30.875,
+      "grad_norm_var": 8.339322916666667,
+      "learning_rate": 9.823477249189586e-05,
+      "loss": 7.3315,
+      "loss/crossentropy": 2.21771964430809,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17464283481240273,
+      "step": 510
+    },
+    {
+      "epoch": 0.08516666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 8.677018229166666,
+      "learning_rate": 9.822787092288991e-05,
+      "loss": 7.526,
+      "loss/crossentropy": 1.6273227035999298,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14881786331534386,
+      "step": 511
+    },
+    {
+      "epoch": 0.08533333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 9.379166666666666,
+      "learning_rate": 9.822095613193962e-05,
+      "loss": 7.4387,
+      "loss/crossentropy": 1.1738679260015488,
+      "loss/hidden": 3.9296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23782077431678772,
+      "step": 512
+    },
+    {
+      "epoch": 0.0855,
+      "grad_norm": 29.875,
+      "grad_norm_var": 9.801497395833334,
+      "learning_rate": 9.821402812094073e-05,
+      "loss": 7.2145,
+      "loss/crossentropy": 1.9264242947101593,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16293403506278992,
+      "step": 513
+    },
+    {
+      "epoch": 0.08566666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 9.315625,
+      "learning_rate": 9.820708689179259e-05,
+      "loss": 7.5502,
+      "loss/crossentropy": 1.9040753841400146,
+      "loss/hidden": 3.81640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22584915533661842,
+      "step": 514
+    },
+    {
+      "epoch": 0.08583333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 9.315625,
+      "learning_rate": 9.820013244639816e-05,
+      "loss": 7.3902,
+      "loss/crossentropy": 1.9383987933397293,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17753108590841293,
+      "step": 515
+    },
+    {
+      "epoch": 0.086,
+      "grad_norm": 37.25,
+      "grad_norm_var": 10.2884765625,
+      "learning_rate": 9.819316478666405e-05,
+      "loss": 7.2979,
+      "loss/crossentropy": 1.6980541944503784,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2852817215025425,
+      "step": 516
+    },
+    {
+      "epoch": 0.08616666666666667,
+      "grad_norm": 77.0,
+      "grad_norm_var": 130.1478515625,
+      "learning_rate": 9.81861839145005e-05,
+      "loss": 7.2376,
+      "loss/crossentropy": 2.135787755250931,
+      "loss/hidden": 3.84765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31826721876859665,
+      "step": 517
+    },
+    {
+      "epoch": 0.08633333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 131.303125,
+      "learning_rate": 9.817918983182132e-05,
+      "loss": 7.5889,
+      "loss/crossentropy": 1.9717459678649902,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1891697198152542,
+      "step": 518
+    },
+    {
+      "epoch": 0.0865,
+      "grad_norm": 33.5,
+      "grad_norm_var": 130.96979166666668,
+      "learning_rate": 9.8172182540544e-05,
+      "loss": 7.5686,
+      "loss/crossentropy": 1.85391965508461,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21009477972984314,
+      "step": 519
+    },
+    {
+      "epoch": 0.08666666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 131.11432291666668,
+      "learning_rate": 9.816516204258963e-05,
+      "loss": 7.6624,
+      "loss/crossentropy": 1.716525286436081,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20092326402664185,
+      "step": 520
+    },
+    {
+      "epoch": 0.08683333333333333,
+      "grad_norm": 37.0,
+      "grad_norm_var": 129.79557291666666,
+      "learning_rate": 9.815812833988291e-05,
+      "loss": 7.5984,
+      "loss/crossentropy": 1.9357898831367493,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20662782341241837,
+      "step": 521
+    },
+    {
+      "epoch": 0.087,
+      "grad_norm": 31.625,
+      "grad_norm_var": 129.95983072916667,
+      "learning_rate": 9.815108143435218e-05,
+      "loss": 7.5694,
+      "loss/crossentropy": 1.6256851255893707,
+      "loss/hidden": 3.94140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22203047946095467,
+      "step": 522
+    },
+    {
+      "epoch": 0.08716666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 131.28958333333333,
+      "learning_rate": 9.814402132792939e-05,
+      "loss": 7.1982,
+      "loss/crossentropy": 1.38237664103508,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18348556011915207,
+      "step": 523
+    },
+    {
+      "epoch": 0.08733333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 130.13515625,
+      "learning_rate": 9.81369480225501e-05,
+      "loss": 7.2048,
+      "loss/crossentropy": 1.1930311918258667,
+      "loss/hidden": 3.71875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1644559781998396,
+      "step": 524
+    },
+    {
+      "epoch": 0.0875,
+      "grad_norm": 32.0,
+      "grad_norm_var": 130.32473958333333,
+      "learning_rate": 9.812986152015348e-05,
+      "loss": 7.0943,
+      "loss/crossentropy": 1.6827171742916107,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1376042291522026,
+      "step": 525
+    },
+    {
+      "epoch": 0.08766666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 129.41608072916668,
+      "learning_rate": 9.812276182268236e-05,
+      "loss": 7.2546,
+      "loss/crossentropy": 1.7442258596420288,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1855621635913849,
+      "step": 526
+    },
+    {
+      "epoch": 0.08783333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 128.06432291666667,
+      "learning_rate": 9.811564893208318e-05,
+      "loss": 7.2576,
+      "loss/crossentropy": 1.7366014122962952,
+      "loss/hidden": 3.66015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1919305920600891,
+      "step": 527
+    },
+    {
+      "epoch": 0.088,
+      "grad_norm": 30.0,
+      "grad_norm_var": 127.27057291666667,
+      "learning_rate": 9.810852285030593e-05,
+      "loss": 7.2357,
+      "loss/crossentropy": 1.479769915342331,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13667869940400124,
+      "step": 528
+    },
+    {
+      "epoch": 0.08816666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 127.0875,
+      "learning_rate": 9.81013835793043e-05,
+      "loss": 7.0018,
+      "loss/crossentropy": 1.913949340581894,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20627281442284584,
+      "step": 529
+    },
+    {
+      "epoch": 0.08833333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 127.33645833333334,
+      "learning_rate": 9.809423112103554e-05,
+      "loss": 7.578,
+      "loss/crossentropy": 1.991028755903244,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2587154358625412,
+      "step": 530
+    },
+    {
+      "epoch": 0.0885,
+      "grad_norm": 32.5,
+      "grad_norm_var": 127.33645833333334,
+      "learning_rate": 9.808706547746057e-05,
+      "loss": 7.4202,
+      "loss/crossentropy": 1.8278934955596924,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1403353549540043,
+      "step": 531
+    },
+    {
+      "epoch": 0.08866666666666667,
+      "grad_norm": 31.25,
+      "grad_norm_var": 128.16145833333334,
+      "learning_rate": 9.807988665054386e-05,
+      "loss": 7.0638,
+      "loss/crossentropy": 1.9256342053413391,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19789009541273117,
+      "step": 532
+    },
+    {
+      "epoch": 0.08883333333333333,
+      "grad_norm": 37.25,
+      "grad_norm_var": 4.812239583333334,
+      "learning_rate": 9.807269464225355e-05,
+      "loss": 7.3253,
+      "loss/crossentropy": 1.6666708588600159,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19620881974697113,
+      "step": 533
+    },
+    {
+      "epoch": 0.089,
+      "grad_norm": 31.0,
+      "grad_norm_var": 4.918684895833334,
+      "learning_rate": 9.806548945456134e-05,
+      "loss": 7.0865,
+      "loss/crossentropy": 1.6663388460874557,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18542758375406265,
+      "step": 534
+    },
+    {
+      "epoch": 0.08916666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 4.918684895833334,
+      "learning_rate": 9.80582710894426e-05,
+      "loss": 7.663,
+      "loss/crossentropy": 1.5742804110050201,
+      "loss/hidden": 3.86328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2117019146680832,
+      "step": 535
+    },
+    {
+      "epoch": 0.08933333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 4.882291666666666,
+      "learning_rate": 9.805103954887627e-05,
+      "loss": 7.3142,
+      "loss/crossentropy": 1.4173726439476013,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1896781139075756,
+      "step": 536
+    },
+    {
+      "epoch": 0.0895,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.595833333333333,
+      "learning_rate": 9.804379483484494e-05,
+      "loss": 7.037,
+      "loss/crossentropy": 1.822264939546585,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26772022992372513,
+      "step": 537
+    },
+    {
+      "epoch": 0.08966666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.823958333333333,
+      "learning_rate": 9.803653694933476e-05,
+      "loss": 7.1956,
+      "loss/crossentropy": 1.9131834506988525,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18607193604111671,
+      "step": 538
+    },
+    {
+      "epoch": 0.08983333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.434309895833333,
+      "learning_rate": 9.802926589433553e-05,
+      "loss": 7.3716,
+      "loss/crossentropy": 1.3513479083776474,
+      "loss/hidden": 3.87109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18404807522892952,
+      "step": 539
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.468489583333333,
+      "learning_rate": 9.802198167184067e-05,
+      "loss": 7.3129,
+      "loss/crossentropy": 2.4426297545433044,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2037721537053585,
+      "step": 540
+    },
+    {
+      "epoch": 0.09016666666666667,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.4686848958333334,
+      "learning_rate": 9.801468428384716e-05,
+      "loss": 7.6509,
+      "loss/crossentropy": 1.5746603906154633,
+      "loss/hidden": 3.9921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32794898748397827,
+      "step": 541
+    },
+    {
+      "epoch": 0.09033333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.5546223958333334,
+      "learning_rate": 9.800737373235565e-05,
+      "loss": 7.4872,
+      "loss/crossentropy": 1.8053328394889832,
+      "loss/hidden": 3.8046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20955198258161545,
+      "step": 542
+    },
+    {
+      "epoch": 0.0905,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.2177083333333334,
+      "learning_rate": 9.800005001937034e-05,
+      "loss": 7.3155,
+      "loss/crossentropy": 1.6847141683101654,
+      "loss/hidden": 3.73828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24826514720916748,
+      "step": 543
+    },
+    {
+      "epoch": 0.09066666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.0770833333333334,
+      "learning_rate": 9.799271314689908e-05,
+      "loss": 7.4947,
+      "loss/crossentropy": 1.4387599974870682,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17265227809548378,
+      "step": 544
+    },
+    {
+      "epoch": 0.09083333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.9389973958333333,
+      "learning_rate": 9.798536311695334e-05,
+      "loss": 7.5155,
+      "loss/crossentropy": 1.610468253493309,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25361471623182297,
+      "step": 545
+    },
+    {
+      "epoch": 0.091,
+      "grad_norm": 39.0,
+      "grad_norm_var": 6.040559895833334,
+      "learning_rate": 9.797799993154814e-05,
+      "loss": 7.6191,
+      "loss/crossentropy": 1.585631400346756,
+      "loss/hidden": 3.88671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23871760815382004,
+      "step": 546
+    },
+    {
+      "epoch": 0.09116666666666666,
+      "grad_norm": 34.25,
+      "grad_norm_var": 6.259309895833334,
+      "learning_rate": 9.797062359270215e-05,
+      "loss": 7.075,
+      "loss/crossentropy": 1.7494496703147888,
+      "loss/hidden": 3.75,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20255015417933464,
+      "step": 547
+    },
+    {
+      "epoch": 0.09133333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 6.156184895833333,
+      "learning_rate": 9.796323410243763e-05,
+      "loss": 7.2868,
+      "loss/crossentropy": 1.8040901720523834,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2106856070458889,
+      "step": 548
+    },
+    {
+      "epoch": 0.0915,
+      "grad_norm": 33.5,
+      "grad_norm_var": 4.687434895833333,
+      "learning_rate": 9.795583146278046e-05,
+      "loss": 7.5655,
+      "loss/crossentropy": 1.967070758342743,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1635872982442379,
+      "step": 549
+    },
+    {
+      "epoch": 0.09166666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 4.687434895833333,
+      "learning_rate": 9.794841567576011e-05,
+      "loss": 7.183,
+      "loss/crossentropy": 1.89350163936615,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2356087677180767,
+      "step": 550
+    },
+    {
+      "epoch": 0.09183333333333334,
+      "grad_norm": 28.75,
+      "grad_norm_var": 5.350455729166667,
+      "learning_rate": 9.794098674340965e-05,
+      "loss": 7.5694,
+      "loss/crossentropy": 2.145504057407379,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2829388454556465,
+      "step": 551
+    },
+    {
+      "epoch": 0.092,
+      "grad_norm": 30.5,
+      "grad_norm_var": 5.495833333333334,
+      "learning_rate": 9.793354466776579e-05,
+      "loss": 7.3258,
+      "loss/crossentropy": 1.336853101849556,
+      "loss/hidden": 3.78515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22842231392860413,
+      "step": 552
+    },
+    {
+      "epoch": 0.09216666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 5.576497395833333,
+      "learning_rate": 9.79260894508688e-05,
+      "loss": 6.9005,
+      "loss/crossentropy": 1.2748893350362778,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14185206405818462,
+      "step": 553
+    },
+    {
+      "epoch": 0.09233333333333334,
+      "grad_norm": 35.0,
+      "grad_norm_var": 5.898958333333334,
+      "learning_rate": 9.791862109476257e-05,
+      "loss": 7.2931,
+      "loss/crossentropy": 1.4451977908611298,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1574489250779152,
+      "step": 554
+    },
+    {
+      "epoch": 0.0925,
+      "grad_norm": 35.5,
+      "grad_norm_var": 6.5625,
+      "learning_rate": 9.791113960149458e-05,
+      "loss": 7.5447,
+      "loss/crossentropy": 1.8290373086929321,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21042313054203987,
+      "step": 555
+    },
+    {
+      "epoch": 0.09266666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 6.383333333333334,
+      "learning_rate": 9.790364497311597e-05,
+      "loss": 7.0703,
+      "loss/crossentropy": 1.9699373841285706,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19719154760241508,
+      "step": 556
+    },
+    {
+      "epoch": 0.09283333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 6.440625,
+      "learning_rate": 9.789613721168139e-05,
+      "loss": 7.7407,
+      "loss/crossentropy": 1.7264699935913086,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2011517956852913,
+      "step": 557
+    },
+    {
+      "epoch": 0.093,
+      "grad_norm": 29.5,
+      "grad_norm_var": 6.69765625,
+      "learning_rate": 9.788861631924913e-05,
+      "loss": 7.3241,
+      "loss/crossentropy": 1.5741243362426758,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18323720432817936,
+      "step": 558
+    },
+    {
+      "epoch": 0.09316666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 6.639518229166667,
+      "learning_rate": 9.788108229788111e-05,
+      "loss": 7.5273,
+      "loss/crossentropy": 1.6940201073884964,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25649141147732735,
+      "step": 559
+    },
+    {
+      "epoch": 0.09333333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 6.770572916666667,
+      "learning_rate": 9.787353514964284e-05,
+      "loss": 7.481,
+      "loss/crossentropy": 1.7400060892105103,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1604272499680519,
+      "step": 560
+    },
+    {
+      "epoch": 0.0935,
+      "grad_norm": 28.75,
+      "grad_norm_var": 7.62890625,
+      "learning_rate": 9.786597487660337e-05,
+      "loss": 7.2236,
+      "loss/crossentropy": 1.8776799738407135,
+      "loss/hidden": 3.71484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2707614079117775,
+      "step": 561
+    },
+    {
+      "epoch": 0.09366666666666666,
+      "grad_norm": 35.75,
+      "grad_norm_var": 5.357291666666667,
+      "learning_rate": 9.785840148083543e-05,
+      "loss": 7.4374,
+      "loss/crossentropy": 1.3171101808547974,
+      "loss/hidden": 3.7578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17544611543416977,
+      "step": 562
+    },
+    {
+      "epoch": 0.09383333333333334,
+      "grad_norm": 35.25,
+      "grad_norm_var": 5.715625,
+      "learning_rate": 9.785081496441527e-05,
+      "loss": 7.4671,
+      "loss/crossentropy": 1.6518716365098953,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1516691166907549,
+      "step": 563
+    },
+    {
+      "epoch": 0.094,
+      "grad_norm": 34.0,
+      "grad_norm_var": 5.943489583333333,
+      "learning_rate": 9.784321532942282e-05,
+      "loss": 7.7993,
+      "loss/crossentropy": 2.034118413925171,
+      "loss/hidden": 3.796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2103406973183155,
+      "step": 564
+    },
+    {
+      "epoch": 0.09416666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 5.838997395833333,
+      "learning_rate": 9.783560257794154e-05,
+      "loss": 7.3867,
+      "loss/crossentropy": 1.5135559290647507,
+      "loss/hidden": 3.6484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25094762071967125,
+      "step": 565
+    },
+    {
+      "epoch": 0.09433333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.762434895833334,
+      "learning_rate": 9.78279767120585e-05,
+      "loss": 7.5445,
+      "loss/crossentropy": 2.080939292907715,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28340235352516174,
+      "step": 566
+    },
+    {
+      "epoch": 0.0945,
+      "grad_norm": 31.0,
+      "grad_norm_var": 5.049934895833333,
+      "learning_rate": 9.782033773386439e-05,
+      "loss": 7.5806,
+      "loss/crossentropy": 2.0637872517108917,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1680529285222292,
+      "step": 567
+    },
+    {
+      "epoch": 0.09466666666666666,
+      "grad_norm": 28.75,
+      "grad_norm_var": 5.666080729166667,
+      "learning_rate": 9.781268564545348e-05,
+      "loss": 7.2124,
+      "loss/crossentropy": 1.145897313952446,
+      "loss/hidden": 3.90234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2204378955066204,
+      "step": 568
+    },
+    {
+      "epoch": 0.09483333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 5.450455729166666,
+      "learning_rate": 9.780502044892362e-05,
+      "loss": 7.5132,
+      "loss/crossentropy": 1.9616839289665222,
+      "loss/hidden": 3.76953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20735982805490494,
+      "step": 569
+    },
+    {
+      "epoch": 0.095,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.9322265625,
+      "learning_rate": 9.779734214637628e-05,
+      "loss": 7.4043,
+      "loss/crossentropy": 1.7016425430774689,
+      "loss/hidden": 3.77734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18523864448070526,
+      "step": 570
+    },
+    {
+      "epoch": 0.09516666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 4.280143229166667,
+      "learning_rate": 9.778965073991651e-05,
+      "loss": 7.8276,
+      "loss/crossentropy": 1.8775156438350677,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16149216145277023,
+      "step": 571
+    },
+    {
+      "epoch": 0.09533333333333334,
+      "grad_norm": 35.5,
+      "grad_norm_var": 5.02890625,
+      "learning_rate": 9.778194623165296e-05,
+      "loss": 7.4528,
+      "loss/crossentropy": 1.6817852705717087,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18132995441555977,
+      "step": 572
+    },
+    {
+      "epoch": 0.0955,
+      "grad_norm": 33.75,
+      "grad_norm_var": 5.109309895833333,
+      "learning_rate": 9.777422862369783e-05,
+      "loss": 7.2692,
+      "loss/crossentropy": 2.2099765241146088,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20043417066335678,
+      "step": 573
+    },
+    {
+      "epoch": 0.09566666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 4.820833333333334,
+      "learning_rate": 9.776649791816698e-05,
+      "loss": 7.3347,
+      "loss/crossentropy": 1.9577098190784454,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18493201583623886,
+      "step": 574
+    },
+    {
+      "epoch": 0.09583333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 5.168489583333334,
+      "learning_rate": 9.77587541171798e-05,
+      "loss": 7.2916,
+      "loss/crossentropy": 1.350791186094284,
+      "loss/hidden": 3.8984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19409921392798424,
+      "step": 575
+    },
+    {
+      "epoch": 0.096,
+      "grad_norm": 30.375,
+      "grad_norm_var": 5.311458333333333,
+      "learning_rate": 9.775099722285935e-05,
+      "loss": 7.3015,
+      "loss/crossentropy": 1.5742547512054443,
+      "loss/hidden": 3.8203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21891336888074875,
+      "step": 576
+    },
+    {
+      "epoch": 0.09616666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 4.605989583333334,
+      "learning_rate": 9.774322723733216e-05,
+      "loss": 7.1642,
+      "loss/crossentropy": 1.5461835861206055,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26148809865117073,
+      "step": 577
+    },
+    {
+      "epoch": 0.09633333333333334,
+      "grad_norm": 29.875,
+      "grad_norm_var": 4.058268229166667,
+      "learning_rate": 9.773544416272845e-05,
+      "loss": 7.2441,
+      "loss/crossentropy": 2.068322002887726,
+      "loss/hidden": 3.83203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1912810057401657,
+      "step": 578
+    },
+    {
+      "epoch": 0.0965,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.3421223958333335,
+      "learning_rate": 9.772764800118199e-05,
+      "loss": 7.6185,
+      "loss/crossentropy": 2.2380817532539368,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23637336120009422,
+      "step": 579
+    },
+    {
+      "epoch": 0.09666666666666666,
+      "grad_norm": 36.0,
+      "grad_norm_var": 4.185872395833333,
+      "learning_rate": 9.771983875483013e-05,
+      "loss": 7.4395,
+      "loss/crossentropy": 1.504803255200386,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18147898092865944,
+      "step": 580
+    },
+    {
+      "epoch": 0.09683333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 4.208333333333333,
+      "learning_rate": 9.771201642581385e-05,
+      "loss": 7.4978,
+      "loss/crossentropy": 1.624559223651886,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3103417083621025,
+      "step": 581
+    },
+    {
+      "epoch": 0.097,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.182291666666667,
+      "learning_rate": 9.770418101627765e-05,
+      "loss": 7.4196,
+      "loss/crossentropy": 1.2693295776844025,
+      "loss/hidden": 3.83984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16165144927799702,
+      "step": 582
+    },
+    {
+      "epoch": 0.09716666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 4.182291666666667,
+      "learning_rate": 9.769633252836969e-05,
+      "loss": 7.2533,
+      "loss/crossentropy": 1.8926941752433777,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1859600469470024,
+      "step": 583
+    },
+    {
+      "epoch": 0.09733333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 3.512955729166667,
+      "learning_rate": 9.768847096424164e-05,
+      "loss": 7.1366,
+      "loss/crossentropy": 0.7712745144963264,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1119179055094719,
+      "step": 584
+    },
+    {
+      "epoch": 0.0975,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.87265625,
+      "learning_rate": 9.76805963260488e-05,
+      "loss": 7.4932,
+      "loss/crossentropy": 1.6978729963302612,
+      "loss/hidden": 3.73046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20137057453393936,
+      "step": 585
+    },
+    {
+      "epoch": 0.09766666666666667,
+      "grad_norm": 29.125,
+      "grad_norm_var": 4.3244140625,
+      "learning_rate": 9.767270861595005e-05,
+      "loss": 6.989,
+      "loss/crossentropy": 1.3800832033157349,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1577189937233925,
+      "step": 586
+    },
+    {
+      "epoch": 0.09783333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 4.1712890625,
+      "learning_rate": 9.766480783610788e-05,
+      "loss": 7.3945,
+      "loss/crossentropy": 2.3877436816692352,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17484752088785172,
+      "step": 587
+    },
+    {
+      "epoch": 0.098,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.145768229166667,
+      "learning_rate": 9.765689398868831e-05,
+      "loss": 7.2735,
+      "loss/crossentropy": 1.9268846809864044,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14904212579131126,
+      "step": 588
+    },
+    {
+      "epoch": 0.09816666666666667,
+      "grad_norm": 35.25,
+      "grad_norm_var": 3.8004557291666665,
+      "learning_rate": 9.764896707586096e-05,
+      "loss": 7.1183,
+      "loss/crossentropy": 1.3810617476701736,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1700935810804367,
+      "step": 589
+    },
+    {
+      "epoch": 0.09833333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.7447916666666665,
+      "learning_rate": 9.764102709979902e-05,
+      "loss": 7.2089,
+      "loss/crossentropy": 1.7945798188447952,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17525190860033035,
+      "step": 590
+    },
+    {
+      "epoch": 0.0985,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.566666666666667,
+      "learning_rate": 9.763307406267932e-05,
+      "loss": 7.3247,
+      "loss/crossentropy": 1.9337197840213776,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1758846901357174,
+      "step": 591
+    },
+    {
+      "epoch": 0.09866666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.7080729166666666,
+      "learning_rate": 9.76251079666822e-05,
+      "loss": 7.5682,
+      "loss/crossentropy": 1.7730904519557953,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1683446653187275,
+      "step": 592
+    },
+    {
+      "epoch": 0.09883333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.7025390625,
+      "learning_rate": 9.761712881399164e-05,
+      "loss": 7.4159,
+      "loss/crossentropy": 1.4231205731630325,
+      "loss/hidden": 3.82421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24406572431325912,
+      "step": 593
+    },
+    {
+      "epoch": 0.099,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.5462890625,
+      "learning_rate": 9.760913660679515e-05,
+      "loss": 7.5347,
+      "loss/crossentropy": 1.8549748659133911,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25112149119377136,
+      "step": 594
+    },
+    {
+      "epoch": 0.09916666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.856184895833333,
+      "learning_rate": 9.760113134728384e-05,
+      "loss": 7.5034,
+      "loss/crossentropy": 2.0482011437416077,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23262568935751915,
+      "step": 595
+    },
+    {
+      "epoch": 0.09933333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.4518229166666665,
+      "learning_rate": 9.75931130376524e-05,
+      "loss": 6.9944,
+      "loss/crossentropy": 1.2663083150982857,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10222736466675997,
+      "step": 596
+    },
+    {
+      "epoch": 0.0995,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.8372395833333335,
+      "learning_rate": 9.75850816800991e-05,
+      "loss": 7.3032,
+      "loss/crossentropy": 1.7903995215892792,
+      "loss/hidden": 3.89453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2412303052842617,
+      "step": 597
+    },
+    {
+      "epoch": 0.09966666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.093489583333333,
+      "learning_rate": 9.757703727682574e-05,
+      "loss": 7.4625,
+      "loss/crossentropy": 2.283373177051544,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26123256236314774,
+      "step": 598
+    },
+    {
+      "epoch": 0.09983333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.1802083333333333,
+      "learning_rate": 9.756897983003781e-05,
+      "loss": 7.1114,
+      "loss/crossentropy": 1.7814842015504837,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2132834978401661,
+      "step": 599
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.178580729166667,
+      "learning_rate": 9.756090934194427e-05,
+      "loss": 7.3947,
+      "loss/crossentropy": 1.76243194937706,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20218634605407715,
+      "step": 600
+    },
+    {
+      "epoch": 0.10016666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.937239583333333,
+      "learning_rate": 9.755282581475769e-05,
+      "loss": 7.3161,
+      "loss/crossentropy": 1.6957703530788422,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16431233659386635,
+      "step": 601
+    },
+    {
+      "epoch": 0.10033333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.5228515625,
+      "learning_rate": 9.75447292506942e-05,
+      "loss": 7.0909,
+      "loss/crossentropy": 1.6883865594863892,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16238689422607422,
+      "step": 602
+    },
+    {
+      "epoch": 0.1005,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.471875,
+      "learning_rate": 9.753661965197354e-05,
+      "loss": 7.434,
+      "loss/crossentropy": 2.007025808095932,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25700215622782707,
+      "step": 603
+    },
+    {
+      "epoch": 0.10066666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.372916666666667,
+      "learning_rate": 9.752849702081901e-05,
+      "loss": 7.2278,
+      "loss/crossentropy": 2.1384987235069275,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18107693642377853,
+      "step": 604
+    },
+    {
+      "epoch": 0.10083333333333333,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.0580729166666667,
+      "learning_rate": 9.752036135945744e-05,
+      "loss": 7.4111,
+      "loss/crossentropy": 1.9143229275941849,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16873291693627834,
+      "step": 605
+    },
+    {
+      "epoch": 0.101,
+      "grad_norm": 34.0,
+      "grad_norm_var": 2.380989583333333,
+      "learning_rate": 9.751221267011929e-05,
+      "loss": 7.335,
+      "loss/crossentropy": 2.115232676267624,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24951379001140594,
+      "step": 606
+    },
+    {
+      "epoch": 0.10116666666666667,
+      "grad_norm": 51.75,
+      "grad_norm_var": 26.980989583333333,
+      "learning_rate": 9.750405095503859e-05,
+      "loss": 7.6198,
+      "loss/crossentropy": 1.5640252828598022,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16927756741642952,
+      "step": 607
+    },
+    {
+      "epoch": 0.10133333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 26.103059895833333,
+      "learning_rate": 9.749587621645288e-05,
+      "loss": 7.3387,
+      "loss/crossentropy": 1.8963924050331116,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.251237440854311,
+      "step": 608
+    },
+    {
+      "epoch": 0.1015,
+      "grad_norm": 28.875,
+      "grad_norm_var": 27.092122395833332,
+      "learning_rate": 9.748768845660334e-05,
+      "loss": 7.0424,
+      "loss/crossentropy": 1.470510482788086,
+      "loss/hidden": 3.80859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2107672020792961,
+      "step": 609
+    },
+    {
+      "epoch": 0.10166666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 27.773893229166667,
+      "learning_rate": 9.74794876777347e-05,
+      "loss": 7.4808,
+      "loss/crossentropy": 1.5921064913272858,
+      "loss/hidden": 4.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2697357200086117,
+      "step": 610
+    },
+    {
+      "epoch": 0.10183333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 29.089322916666667,
+      "learning_rate": 9.74712738820952e-05,
+      "loss": 6.9466,
+      "loss/crossentropy": 1.3399248868227005,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15232133492827415,
+      "step": 611
+    },
+    {
+      "epoch": 0.102,
+      "grad_norm": 28.25,
+      "grad_norm_var": 30.2791015625,
+      "learning_rate": 9.746304707193675e-05,
+      "loss": 7.1313,
+      "loss/crossentropy": 1.8080547451972961,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18052439764142036,
+      "step": 612
+    },
+    {
+      "epoch": 0.10216666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 30.258072916666666,
+      "learning_rate": 9.745480724951473e-05,
+      "loss": 7.5941,
+      "loss/crossentropy": 1.3583181947469711,
+      "loss/hidden": 3.76171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17031322419643402,
+      "step": 613
+    },
+    {
+      "epoch": 0.10233333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 30.208268229166666,
+      "learning_rate": 9.744655441708818e-05,
+      "loss": 7.4358,
+      "loss/crossentropy": 1.250537633895874,
+      "loss/hidden": 4.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2280135713517666,
+      "step": 614
+    },
+    {
+      "epoch": 0.1025,
+      "grad_norm": 32.25,
+      "grad_norm_var": 29.9103515625,
+      "learning_rate": 9.743828857691963e-05,
+      "loss": 7.3615,
+      "loss/crossentropy": 2.1062381863594055,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17050448432564735,
+      "step": 615
+    },
+    {
+      "epoch": 0.10266666666666667,
+      "grad_norm": 142.0,
+      "grad_norm_var": 779.3535807291667,
+      "learning_rate": 9.743000973127523e-05,
+      "loss": 8.0423,
+      "loss/crossentropy": 1.8291618078947067,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22066161036491394,
+      "step": 616
+    },
+    {
+      "epoch": 0.10283333333333333,
+      "grad_norm": 47.25,
+      "grad_norm_var": 777.6393229166666,
+      "learning_rate": 9.742171788242466e-05,
+      "loss": 7.6019,
+      "loss/crossentropy": 1.6715676486492157,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28256499022245407,
+      "step": 617
+    },
+    {
+      "epoch": 0.103,
+      "grad_norm": 35.5,
+      "grad_norm_var": 773.9143229166667,
+      "learning_rate": 9.741341303264118e-05,
+      "loss": 7.0277,
+      "loss/crossentropy": 1.7895082533359528,
+      "loss/hidden": 3.91015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20992443338036537,
+      "step": 618
+    },
+    {
+      "epoch": 0.10316666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 773.59375,
+      "learning_rate": 9.74050951842016e-05,
+      "loss": 7.3626,
+      "loss/crossentropy": 2.2063084542751312,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2172108329832554,
+      "step": 619
+    },
+    {
+      "epoch": 0.10333333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 774.8228515625,
+      "learning_rate": 9.739676433938633e-05,
+      "loss": 7.0412,
+      "loss/crossentropy": 1.7916174978017807,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22405482083559036,
+      "step": 620
+    },
+    {
+      "epoch": 0.1035,
+      "grad_norm": 31.5,
+      "grad_norm_var": 777.8134765625,
+      "learning_rate": 9.73884205004793e-05,
+      "loss": 7.8157,
+      "loss/crossentropy": 2.1516139805316925,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20110300928354263,
+      "step": 621
+    },
+    {
+      "epoch": 0.10366666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 780.9291015625,
+      "learning_rate": 9.7380063669768e-05,
+      "loss": 7.1226,
+      "loss/crossentropy": 1.5412001609802246,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18404862098395824,
+      "step": 622
+    },
+    {
+      "epoch": 0.10383333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 774.8806640625,
+      "learning_rate": 9.737169384954355e-05,
+      "loss": 7.4503,
+      "loss/crossentropy": 1.2898205816745758,
+      "loss/hidden": 3.8515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25114247761666775,
+      "step": 623
+    },
+    {
+      "epoch": 0.104,
+      "grad_norm": 31.375,
+      "grad_norm_var": 776.528125,
+      "learning_rate": 9.736331104210056e-05,
+      "loss": 7.4958,
+      "loss/crossentropy": 2.409509092569351,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2068478837609291,
+      "step": 624
+    },
+    {
+      "epoch": 0.10416666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 773.3455729166667,
+      "learning_rate": 9.735491524973722e-05,
+      "loss": 7.438,
+      "loss/crossentropy": 1.5180833488702774,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1766592413187027,
+      "step": 625
+    },
+    {
+      "epoch": 0.10433333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 770.8962890625,
+      "learning_rate": 9.73465064747553e-05,
+      "loss": 7.0654,
+      "loss/crossentropy": 1.275202915072441,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13855969533324242,
+      "step": 626
+    },
+    {
+      "epoch": 0.1045,
+      "grad_norm": 30.125,
+      "grad_norm_var": 768.5738932291666,
+      "learning_rate": 9.73380847194601e-05,
+      "loss": 7.1306,
+      "loss/crossentropy": 1.3508297204971313,
+      "loss/hidden": 3.89453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19833296723663807,
+      "step": 627
+    },
+    {
+      "epoch": 0.10466666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 762.7978515625,
+      "learning_rate": 9.732964998616046e-05,
+      "loss": 7.5178,
+      "loss/crossentropy": 1.835921436548233,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15121160633862019,
+      "step": 628
+    },
+    {
+      "epoch": 0.10483333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 761.8895833333333,
+      "learning_rate": 9.732120227716888e-05,
+      "loss": 7.2138,
+      "loss/crossentropy": 1.433204710483551,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13274148851633072,
+      "step": 629
+    },
+    {
+      "epoch": 0.105,
+      "grad_norm": 29.5,
+      "grad_norm_var": 764.4384765625,
+      "learning_rate": 9.73127415948013e-05,
+      "loss": 7.4102,
+      "loss/crossentropy": 2.046613782644272,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2395108975470066,
+      "step": 630
+    },
+    {
+      "epoch": 0.10516666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 768.5837890625,
+      "learning_rate": 9.730426794137727e-05,
+      "loss": 7.2557,
+      "loss/crossentropy": 1.1972267031669617,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11847973614931107,
+      "step": 631
+    },
+    {
+      "epoch": 0.10533333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 18.2759765625,
+      "learning_rate": 9.72957813192199e-05,
+      "loss": 7.1137,
+      "loss/crossentropy": 1.4940354526042938,
+      "loss/hidden": 3.87109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2576959580183029,
+      "step": 632
+    },
+    {
+      "epoch": 0.1055,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.0072265625,
+      "learning_rate": 9.728728173065585e-05,
+      "loss": 7.5056,
+      "loss/crossentropy": 2.48933482170105,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18084804341197014,
+      "step": 633
+    },
+    {
+      "epoch": 0.10566666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 2.3744140625,
+      "learning_rate": 9.72787691780153e-05,
+      "loss": 7.2325,
+      "loss/crossentropy": 1.2645312547683716,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29484754242002964,
+      "step": 634
+    },
+    {
+      "epoch": 0.10583333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.4176432291666665,
+      "learning_rate": 9.727024366363206e-05,
+      "loss": 7.8715,
+      "loss/crossentropy": 1.912324070930481,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4577797334641218,
+      "step": 635
+    },
+    {
+      "epoch": 0.106,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.3833333333333333,
+      "learning_rate": 9.726170518984341e-05,
+      "loss": 7.8196,
+      "loss/crossentropy": 1.9435686469078064,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2176922857761383,
+      "step": 636
+    },
+    {
+      "epoch": 0.10616666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.3833333333333333,
+      "learning_rate": 9.725315375899024e-05,
+      "loss": 7.2167,
+      "loss/crossentropy": 1.6355883032083511,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19487426057457924,
+      "step": 637
+    },
+    {
+      "epoch": 0.10633333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.3559895833333333,
+      "learning_rate": 9.724458937341698e-05,
+      "loss": 7.6199,
+      "loss/crossentropy": 1.9535970985889435,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2554296776652336,
+      "step": 638
+    },
+    {
+      "epoch": 0.1065,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.80390625,
+      "learning_rate": 9.723601203547158e-05,
+      "loss": 7.4487,
+      "loss/crossentropy": 1.7397211343050003,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16598396748304367,
+      "step": 639
+    },
+    {
+      "epoch": 0.10666666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.7916015625,
+      "learning_rate": 9.722742174750558e-05,
+      "loss": 7.1779,
+      "loss/crossentropy": 1.8731798231601715,
+      "loss/hidden": 3.9921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2941042631864548,
+      "step": 640
+    },
+    {
+      "epoch": 0.10683333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.0497395833333334,
+      "learning_rate": 9.721881851187406e-05,
+      "loss": 7.215,
+      "loss/crossentropy": 1.6555385440587997,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18280649557709694,
+      "step": 641
+    },
+    {
+      "epoch": 0.107,
+      "grad_norm": 35.5,
+      "grad_norm_var": 3.8080729166666667,
+      "learning_rate": 9.721020233093563e-05,
+      "loss": 7.3108,
+      "loss/crossentropy": 1.8831823766231537,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18430525064468384,
+      "step": 642
+    },
+    {
+      "epoch": 0.10716666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.7427083333333333,
+      "learning_rate": 9.72015732070525e-05,
+      "loss": 7.0786,
+      "loss/crossentropy": 1.8577512204647064,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1396968588232994,
+      "step": 643
+    },
+    {
+      "epoch": 0.10733333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.6770833333333335,
+      "learning_rate": 9.719293114259033e-05,
+      "loss": 6.9481,
+      "loss/crossentropy": 1.5243144929409027,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17269080504775047,
+      "step": 644
+    },
+    {
+      "epoch": 0.1075,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.9900390625,
+      "learning_rate": 9.718427613991848e-05,
+      "loss": 7.1693,
+      "loss/crossentropy": 1.1140879094600677,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16539104841649532,
+      "step": 645
+    },
+    {
+      "epoch": 0.10766666666666666,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.911393229166667,
+      "learning_rate": 9.717560820140969e-05,
+      "loss": 7.503,
+      "loss/crossentropy": 1.3911971896886826,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1835576556622982,
+      "step": 646
+    },
+    {
+      "epoch": 0.10783333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 3.7108723958333334,
+      "learning_rate": 9.716692732944035e-05,
+      "loss": 7.1375,
+      "loss/crossentropy": 1.6554150804877281,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1653379462659359,
+      "step": 647
+    },
+    {
+      "epoch": 0.108,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.5431640625,
+      "learning_rate": 9.715823352639037e-05,
+      "loss": 7.4142,
+      "loss/crossentropy": 2.2835015952587128,
+      "loss/hidden": 3.05078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2020583711564541,
+      "step": 648
+    },
+    {
+      "epoch": 0.10816666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.440625,
+      "learning_rate": 9.714952679464323e-05,
+      "loss": 7.3722,
+      "loss/crossentropy": 1.7637509107589722,
+      "loss/hidden": 3.10546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12249446101486683,
+      "step": 649
+    },
+    {
+      "epoch": 0.10833333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.115625,
+      "learning_rate": 9.71408071365859e-05,
+      "loss": 7.4532,
+      "loss/crossentropy": 1.142805352807045,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1244389396160841,
+      "step": 650
+    },
+    {
+      "epoch": 0.1085,
+      "grad_norm": 29.25,
+      "grad_norm_var": 3.434309895833333,
+      "learning_rate": 9.713207455460894e-05,
+      "loss": 7.6449,
+      "loss/crossentropy": 1.4701527208089828,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15961984731256962,
+      "step": 651
+    },
+    {
+      "epoch": 0.10866666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.2014973958333335,
+      "learning_rate": 9.71233290511064e-05,
+      "loss": 7.1868,
+      "loss/crossentropy": 1.9166098833084106,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16668128222227097,
+      "step": 652
+    },
+    {
+      "epoch": 0.10883333333333334,
+      "grad_norm": 41.25,
+      "grad_norm_var": 9.092122395833334,
+      "learning_rate": 9.711457062847595e-05,
+      "loss": 7.2294,
+      "loss/crossentropy": 1.8596365749835968,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15339374542236328,
+      "step": 653
+    },
+    {
+      "epoch": 0.109,
+      "grad_norm": 33.25,
+      "grad_norm_var": 9.1681640625,
+      "learning_rate": 9.710579928911876e-05,
+      "loss": 7.4448,
+      "loss/crossentropy": 1.525905817747116,
+      "loss/hidden": 3.9765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1737709790468216,
+      "step": 654
+    },
+    {
+      "epoch": 0.10916666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 8.883072916666666,
+      "learning_rate": 9.709701503543954e-05,
+      "loss": 7.4211,
+      "loss/crossentropy": 1.5346709489822388,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19532465934753418,
+      "step": 655
+    },
+    {
+      "epoch": 0.10933333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 8.973893229166666,
+      "learning_rate": 9.708821786984652e-05,
+      "loss": 7.2277,
+      "loss/crossentropy": 1.3570376336574554,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1451853420585394,
+      "step": 656
+    },
+    {
+      "epoch": 0.1095,
+      "grad_norm": 36.0,
+      "grad_norm_var": 9.730143229166666,
+      "learning_rate": 9.707940779475151e-05,
+      "loss": 7.7429,
+      "loss/crossentropy": 1.5309327840805054,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18355575948953629,
+      "step": 657
+    },
+    {
+      "epoch": 0.10966666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 8.978580729166667,
+      "learning_rate": 9.707058481256985e-05,
+      "loss": 7.267,
+      "loss/crossentropy": 1.7733177542686462,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14523290283977985,
+      "step": 658
+    },
+    {
+      "epoch": 0.10983333333333334,
+      "grad_norm": 28.25,
+      "grad_norm_var": 9.732291666666667,
+      "learning_rate": 9.706174892572039e-05,
+      "loss": 7.0048,
+      "loss/crossentropy": 0.9952821731567383,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15045959129929543,
+      "step": 659
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 28.875,
+      "grad_norm_var": 10.3087890625,
+      "learning_rate": 9.705290013662556e-05,
+      "loss": 7.3332,
+      "loss/crossentropy": 1.9855782389640808,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20127024874091148,
+      "step": 660
+    },
+    {
+      "epoch": 0.11016666666666666,
+      "grad_norm": 31.75,
+      "grad_norm_var": 10.065625,
+      "learning_rate": 9.704403844771128e-05,
+      "loss": 7.2978,
+      "loss/crossentropy": 1.6638555526733398,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18440712057054043,
+      "step": 661
+    },
+    {
+      "epoch": 0.11033333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 9.869205729166667,
+      "learning_rate": 9.703516386140705e-05,
+      "loss": 7.3387,
+      "loss/crossentropy": 1.9379086792469025,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1722083631902933,
+      "step": 662
+    },
+    {
+      "epoch": 0.1105,
+      "grad_norm": 29.625,
+      "grad_norm_var": 9.745572916666667,
+      "learning_rate": 9.70262763801459e-05,
+      "loss": 7.0218,
+      "loss/crossentropy": 1.532268926501274,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1991143524646759,
+      "step": 663
+    },
+    {
+      "epoch": 0.11066666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 9.813541666666667,
+      "learning_rate": 9.701737600636436e-05,
+      "loss": 7.2071,
+      "loss/crossentropy": 1.3929783999919891,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18860838934779167,
+      "step": 664
+    },
+    {
+      "epoch": 0.11083333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 9.831705729166666,
+      "learning_rate": 9.700846274250251e-05,
+      "loss": 7.6409,
+      "loss/crossentropy": 1.9433818459510803,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20732827112078667,
+      "step": 665
+    },
+    {
+      "epoch": 0.111,
+      "grad_norm": 30.5,
+      "grad_norm_var": 9.942643229166666,
+      "learning_rate": 9.699953659100401e-05,
+      "loss": 7.0308,
+      "loss/crossentropy": 1.2917715013027191,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19213541224598885,
+      "step": 666
+    },
+    {
+      "epoch": 0.11116666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 9.514322916666666,
+      "learning_rate": 9.699059755431598e-05,
+      "loss": 7.2249,
+      "loss/crossentropy": 1.8711410611867905,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18579615280032158,
+      "step": 667
+    },
+    {
+      "epoch": 0.11133333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 9.514322916666666,
+      "learning_rate": 9.698164563488914e-05,
+      "loss": 7.1074,
+      "loss/crossentropy": 1.856014922261238,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16111868992447853,
+      "step": 668
+    },
+    {
+      "epoch": 0.1115,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.530989583333333,
+      "learning_rate": 9.697268083517767e-05,
+      "loss": 7.3351,
+      "loss/crossentropy": 2.1788209080696106,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27165431156754494,
+      "step": 669
+    },
+    {
+      "epoch": 0.11166666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.3666015625,
+      "learning_rate": 9.696370315763936e-05,
+      "loss": 7.4897,
+      "loss/crossentropy": 1.7634121179580688,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25094933807849884,
+      "step": 670
+    },
+    {
+      "epoch": 0.11183333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.3643229166666666,
+      "learning_rate": 9.695471260473545e-05,
+      "loss": 7.36,
+      "loss/crossentropy": 1.8961014151573181,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15150394663214684,
+      "step": 671
+    },
+    {
+      "epoch": 0.112,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.3754557291666667,
+      "learning_rate": 9.69457091789308e-05,
+      "loss": 7.5605,
+      "loss/crossentropy": 2.0419953167438507,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2538929544389248,
+      "step": 672
+    },
+    {
+      "epoch": 0.11216666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.7655598958333334,
+      "learning_rate": 9.693669288269372e-05,
+      "loss": 7.4923,
+      "loss/crossentropy": 2.064723700284958,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18939067423343658,
+      "step": 673
+    },
+    {
+      "epoch": 0.11233333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.634375,
+      "learning_rate": 9.692766371849606e-05,
+      "loss": 7.2448,
+      "loss/crossentropy": 1.8495427966117859,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1740853115916252,
+      "step": 674
+    },
+    {
+      "epoch": 0.1125,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.2072265625,
+      "learning_rate": 9.691862168881325e-05,
+      "loss": 6.9931,
+      "loss/crossentropy": 1.6820146292448044,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15422948263585567,
+      "step": 675
+    },
+    {
+      "epoch": 0.11266666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 0.9135416666666667,
+      "learning_rate": 9.690956679612421e-05,
+      "loss": 7.205,
+      "loss/crossentropy": 1.7395752966403961,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20282979309558868,
+      "step": 676
+    },
+    {
+      "epoch": 0.11283333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 0.9018229166666667,
+      "learning_rate": 9.690049904291139e-05,
+      "loss": 8.064,
+      "loss/crossentropy": 1.8543930053710938,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18171074986457825,
+      "step": 677
+    },
+    {
+      "epoch": 0.113,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.3780598958333334,
+      "learning_rate": 9.689141843166074e-05,
+      "loss": 7.3536,
+      "loss/crossentropy": 1.9578366875648499,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20767145231366158,
+      "step": 678
+    },
+    {
+      "epoch": 0.11316666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.3405598958333333,
+      "learning_rate": 9.688232496486178e-05,
+      "loss": 7.2639,
+      "loss/crossentropy": 2.4087424874305725,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19602132216095924,
+      "step": 679
+    },
+    {
+      "epoch": 0.11333333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.3317057291666667,
+      "learning_rate": 9.687321864500755e-05,
+      "loss": 7.4667,
+      "loss/crossentropy": 1.4875228255987167,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15017814375460148,
+      "step": 680
+    },
+    {
+      "epoch": 0.1135,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.3317057291666667,
+      "learning_rate": 9.686409947459458e-05,
+      "loss": 7.6094,
+      "loss/crossentropy": 2.1256344318389893,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23931212723255157,
+      "step": 681
+    },
+    {
+      "epoch": 0.11366666666666667,
+      "grad_norm": 36.75,
+      "grad_norm_var": 3.4410807291666665,
+      "learning_rate": 9.685496745612295e-05,
+      "loss": 7.2968,
+      "loss/crossentropy": 1.9394277036190033,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17435170337557793,
+      "step": 682
+    },
+    {
+      "epoch": 0.11383333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.6822916666666665,
+      "learning_rate": 9.684582259209624e-05,
+      "loss": 7.4703,
+      "loss/crossentropy": 2.1526367366313934,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18739807233214378,
+      "step": 683
+    },
+    {
+      "epoch": 0.114,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.7739583333333333,
+      "learning_rate": 9.683666488502158e-05,
+      "loss": 7.0513,
+      "loss/crossentropy": 1.5884509682655334,
+      "loss/hidden": 3.640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1915692798793316,
+      "step": 684
+    },
+    {
+      "epoch": 0.11416666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.5083333333333333,
+      "learning_rate": 9.682749433740962e-05,
+      "loss": 7.1726,
+      "loss/crossentropy": 1.665308564901352,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17285103723406792,
+      "step": 685
+    },
+    {
+      "epoch": 0.11433333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.702018229166667,
+      "learning_rate": 9.68183109517745e-05,
+      "loss": 7.5115,
+      "loss/crossentropy": 2.2219296991825104,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17908718809485435,
+      "step": 686
+    },
+    {
+      "epoch": 0.1145,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.787239583333333,
+      "learning_rate": 9.68091147306339e-05,
+      "loss": 7.5444,
+      "loss/crossentropy": 1.3817338794469833,
+      "loss/hidden": 3.875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24499402940273285,
+      "step": 687
+    },
+    {
+      "epoch": 0.11466666666666667,
+      "grad_norm": 42.25,
+      "grad_norm_var": 11.265625,
+      "learning_rate": 9.6799905676509e-05,
+      "loss": 7.4661,
+      "loss/crossentropy": 1.3822373747825623,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24674061685800552,
+      "step": 688
+    },
+    {
+      "epoch": 0.11483333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 11.220572916666667,
+      "learning_rate": 9.679068379192456e-05,
+      "loss": 7.2038,
+      "loss/crossentropy": 1.2653769254684448,
+      "loss/hidden": 3.7265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23560648038983345,
+      "step": 689
+    },
+    {
+      "epoch": 0.115,
+      "grad_norm": 30.125,
+      "grad_norm_var": 11.326822916666666,
+      "learning_rate": 9.678144907940876e-05,
+      "loss": 7.1295,
+      "loss/crossentropy": 1.3056302964687347,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16892738826572895,
+      "step": 690
+    },
+    {
+      "epoch": 0.11516666666666667,
+      "grad_norm": 29.125,
+      "grad_norm_var": 11.826822916666666,
+      "learning_rate": 9.677220154149337e-05,
+      "loss": 7.369,
+      "loss/crossentropy": 2.204236626625061,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2106693685054779,
+      "step": 691
+    },
+    {
+      "epoch": 0.11533333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 11.92890625,
+      "learning_rate": 9.676294118071367e-05,
+      "loss": 7.444,
+      "loss/crossentropy": 1.4087094068527222,
+      "loss/hidden": 3.796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25383753329515457,
+      "step": 692
+    },
+    {
+      "epoch": 0.1155,
+      "grad_norm": 29.125,
+      "grad_norm_var": 12.2791015625,
+      "learning_rate": 9.675366799960841e-05,
+      "loss": 7.1626,
+      "loss/crossentropy": 1.372077763080597,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16764922067523003,
+      "step": 693
+    },
+    {
+      "epoch": 0.11566666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 11.4869140625,
+      "learning_rate": 9.674438200071991e-05,
+      "loss": 7.2944,
+      "loss/crossentropy": 1.6388816386461258,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2240111120045185,
+      "step": 694
+    },
+    {
+      "epoch": 0.11583333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 11.39140625,
+      "learning_rate": 9.6735083186594e-05,
+      "loss": 7.6349,
+      "loss/crossentropy": 1.794825941324234,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1945985034108162,
+      "step": 695
+    },
+    {
+      "epoch": 0.116,
+      "grad_norm": 32.5,
+      "grad_norm_var": 11.294791666666667,
+      "learning_rate": 9.672577155977993e-05,
+      "loss": 7.3889,
+      "loss/crossentropy": 1.771435633301735,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18000315874814987,
+      "step": 696
+    },
+    {
+      "epoch": 0.11616666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 11.489322916666667,
+      "learning_rate": 9.671644712283061e-05,
+      "loss": 6.8488,
+      "loss/crossentropy": 1.3064327538013458,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17428554967045784,
+      "step": 697
+    },
+    {
+      "epoch": 0.11633333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 10.105143229166666,
+      "learning_rate": 9.670710987830233e-05,
+      "loss": 7.0415,
+      "loss/crossentropy": 2.0627946108579636,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21828969940543175,
+      "step": 698
+    },
+    {
+      "epoch": 0.1165,
+      "grad_norm": 37.25,
+      "grad_norm_var": 12.034309895833333,
+      "learning_rate": 9.669775982875501e-05,
+      "loss": 7.2626,
+      "loss/crossentropy": 1.859133243560791,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1761200949549675,
+      "step": 699
+    },
+    {
+      "epoch": 0.11666666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 11.979622395833333,
+      "learning_rate": 9.668839697675196e-05,
+      "loss": 7.4687,
+      "loss/crossentropy": 1.504253938794136,
+      "loss/hidden": 3.71875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20222976431250572,
+      "step": 700
+    },
+    {
+      "epoch": 0.11683333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 11.928059895833334,
+      "learning_rate": 9.667902132486009e-05,
+      "loss": 7.4905,
+      "loss/crossentropy": 1.32832533121109,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18155817314982414,
+      "step": 701
+    },
+    {
+      "epoch": 0.117,
+      "grad_norm": 29.875,
+      "grad_norm_var": 12.130208333333334,
+      "learning_rate": 9.666963287564979e-05,
+      "loss": 7.6311,
+      "loss/crossentropy": 1.641435131430626,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20181964337825775,
+      "step": 702
+    },
+    {
+      "epoch": 0.11716666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 12.160872395833334,
+      "learning_rate": 9.666023163169493e-05,
+      "loss": 7.0104,
+      "loss/crossentropy": 1.3292317688465118,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1403409019112587,
+      "step": 703
+    },
+    {
+      "epoch": 0.11733333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.547916666666667,
+      "learning_rate": 9.665081759557295e-05,
+      "loss": 8.0035,
+      "loss/crossentropy": 2.21759694814682,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.271518062800169,
+      "step": 704
+    },
+    {
+      "epoch": 0.1175,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.5416015625,
+      "learning_rate": 9.664139076986473e-05,
+      "loss": 7.3329,
+      "loss/crossentropy": 1.9011651873588562,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.184487696737051,
+      "step": 705
+    },
+    {
+      "epoch": 0.11766666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 4.460872395833333,
+      "learning_rate": 9.663195115715471e-05,
+      "loss": 7.7688,
+      "loss/crossentropy": 2.1999647319316864,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23816713690757751,
+      "step": 706
+    },
+    {
+      "epoch": 0.11783333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 4.634375,
+      "learning_rate": 9.66224987600308e-05,
+      "loss": 7.6553,
+      "loss/crossentropy": 1.5069388449192047,
+      "loss/hidden": 4.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32512833178043365,
+      "step": 707
+    },
+    {
+      "epoch": 0.118,
+      "grad_norm": 29.625,
+      "grad_norm_var": 4.7228515625,
+      "learning_rate": 9.661303358108445e-05,
+      "loss": 6.9435,
+      "loss/crossentropy": 1.475489467382431,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16300288774073124,
+      "step": 708
+    },
+    {
+      "epoch": 0.11816666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.302083333333333,
+      "learning_rate": 9.660355562291055e-05,
+      "loss": 6.8902,
+      "loss/crossentropy": 1.5365915447473526,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14909449964761734,
+      "step": 709
+    },
+    {
+      "epoch": 0.11833333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.27265625,
+      "learning_rate": 9.659406488810759e-05,
+      "loss": 7.1034,
+      "loss/crossentropy": 1.4892998337745667,
+      "loss/hidden": 3.6484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22402291372418404,
+      "step": 710
+    },
+    {
+      "epoch": 0.1185,
+      "grad_norm": 31.75,
+      "grad_norm_var": 4.11640625,
+      "learning_rate": 9.658456137927745e-05,
+      "loss": 7.6042,
+      "loss/crossentropy": 1.8311169147491455,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20424065738916397,
+      "step": 711
+    },
+    {
+      "epoch": 0.11866666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.151041666666667,
+      "learning_rate": 9.657504509902562e-05,
+      "loss": 7.231,
+      "loss/crossentropy": 1.7308376729488373,
+      "loss/hidden": 3.66015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2392408549785614,
+      "step": 712
+    },
+    {
+      "epoch": 0.11883333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.06875,
+      "learning_rate": 9.656551604996102e-05,
+      "loss": 7.6568,
+      "loss/crossentropy": 1.7638636529445648,
+      "loss/hidden": 4.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2292976714670658,
+      "step": 713
+    },
+    {
+      "epoch": 0.119,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.784375,
+      "learning_rate": 9.655597423469609e-05,
+      "loss": 7.1736,
+      "loss/crossentropy": 1.7928911745548248,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18390469625592232,
+      "step": 714
+    },
+    {
+      "epoch": 0.11916666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 2.184375,
+      "learning_rate": 9.654641965584678e-05,
+      "loss": 7.4753,
+      "loss/crossentropy": 1.7224461138248444,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19840865582227707,
+      "step": 715
+    },
+    {
+      "epoch": 0.11933333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.026497395833333,
+      "learning_rate": 9.653685231603256e-05,
+      "loss": 7.4207,
+      "loss/crossentropy": 2.0060774981975555,
+      "loss/hidden": 3.73046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2470613420009613,
+      "step": 716
+    },
+    {
+      "epoch": 0.1195,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.135872395833333,
+      "learning_rate": 9.652727221787631e-05,
+      "loss": 7.0611,
+      "loss/crossentropy": 1.905442088842392,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17968276515603065,
+      "step": 717
+    },
+    {
+      "epoch": 0.11966666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.060416666666667,
+      "learning_rate": 9.65176793640045e-05,
+      "loss": 7.6799,
+      "loss/crossentropy": 1.8074587881565094,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18332719430327415,
+      "step": 718
+    },
+    {
+      "epoch": 0.11983333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.0843098958333335,
+      "learning_rate": 9.650807375704708e-05,
+      "loss": 7.265,
+      "loss/crossentropy": 2.222355842590332,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20805594697594643,
+      "step": 719
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.292708333333333,
+      "learning_rate": 9.649845539963747e-05,
+      "loss": 7.3168,
+      "loss/crossentropy": 1.362179696559906,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1665158737450838,
+      "step": 720
+    },
+    {
+      "epoch": 0.12016666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 2.8218098958333333,
+      "learning_rate": 9.648882429441257e-05,
+      "loss": 7.4782,
+      "loss/crossentropy": 1.5363773852586746,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1421268694102764,
+      "step": 721
+    },
+    {
+      "epoch": 0.12033333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.9184895833333333,
+      "learning_rate": 9.647918044401285e-05,
+      "loss": 7.3579,
+      "loss/crossentropy": 2.0128808319568634,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19340229779481888,
+      "step": 722
+    },
+    {
+      "epoch": 0.1205,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.60390625,
+      "learning_rate": 9.646952385108218e-05,
+      "loss": 7.5683,
+      "loss/crossentropy": 1.2376703917980194,
+      "loss/hidden": 4.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24080761522054672,
+      "step": 723
+    },
+    {
+      "epoch": 0.12066666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.50390625,
+      "learning_rate": 9.645985451826803e-05,
+      "loss": 7.3693,
+      "loss/crossentropy": 2.0515187084674835,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17883748933672905,
+      "step": 724
+    },
+    {
+      "epoch": 0.12083333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.8375,
+      "learning_rate": 9.645017244822123e-05,
+      "loss": 7.7341,
+      "loss/crossentropy": 1.279893919825554,
+      "loss/hidden": 3.9921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2662294916808605,
+      "step": 725
+    },
+    {
+      "epoch": 0.121,
+      "grad_norm": 35.75,
+      "grad_norm_var": 3.9309895833333335,
+      "learning_rate": 9.644047764359622e-05,
+      "loss": 7.4369,
+      "loss/crossentropy": 1.72224622964859,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16291150823235512,
+      "step": 726
+    },
+    {
+      "epoch": 0.12116666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.933333333333333,
+      "learning_rate": 9.643077010705087e-05,
+      "loss": 7.0033,
+      "loss/crossentropy": 1.627916619181633,
+      "loss/hidden": 3.75,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19500724971294403,
+      "step": 727
+    },
+    {
+      "epoch": 0.12133333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 4.187239583333334,
+      "learning_rate": 9.642104984124656e-05,
+      "loss": 7.1655,
+      "loss/crossentropy": 1.731247067451477,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16741443425416946,
+      "step": 728
+    },
+    {
+      "epoch": 0.1215,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.178125,
+      "learning_rate": 9.641131684884817e-05,
+      "loss": 7.227,
+      "loss/crossentropy": 2.0340212881565094,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1663869433104992,
+      "step": 729
+    },
+    {
+      "epoch": 0.12166666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 4.4009765625,
+      "learning_rate": 9.640157113252403e-05,
+      "loss": 7.5722,
+      "loss/crossentropy": 1.8398717641830444,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15602401085197926,
+      "step": 730
+    },
+    {
+      "epoch": 0.12183333333333334,
+      "grad_norm": 30.5,
+      "grad_norm_var": 4.0962890625,
+      "learning_rate": 9.6391812694946e-05,
+      "loss": 7.4194,
+      "loss/crossentropy": 2.0755681097507477,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18884403258562088,
+      "step": 731
+    },
+    {
+      "epoch": 0.122,
+      "grad_norm": 31.625,
+      "grad_norm_var": 3.981705729166667,
+      "learning_rate": 9.63820415387894e-05,
+      "loss": 6.9202,
+      "loss/crossentropy": 2.044442892074585,
+      "loss/hidden": 3.796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1960710808634758,
+      "step": 732
+    },
+    {
+      "epoch": 0.12216666666666667,
+      "grad_norm": 39.75,
+      "grad_norm_var": 7.405143229166667,
+      "learning_rate": 9.637225766673307e-05,
+      "loss": 7.4404,
+      "loss/crossentropy": 1.885232001543045,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18598414212465286,
+      "step": 733
+    },
+    {
+      "epoch": 0.12233333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 7.453580729166666,
+      "learning_rate": 9.636246108145929e-05,
+      "loss": 7.172,
+      "loss/crossentropy": 1.3374506533145905,
+      "loss/hidden": 4.05859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1551193678751588,
+      "step": 734
+    },
+    {
+      "epoch": 0.1225,
+      "grad_norm": 30.625,
+      "grad_norm_var": 7.242643229166666,
+      "learning_rate": 9.635265178565385e-05,
+      "loss": 7.0339,
+      "loss/crossentropy": 1.5045725405216217,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19045464508235455,
+      "step": 735
+    },
+    {
+      "epoch": 0.12266666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 6.741666666666666,
+      "learning_rate": 9.634282978200604e-05,
+      "loss": 7.8247,
+      "loss/crossentropy": 1.854149729013443,
+      "loss/hidden": 3.72265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25182773917913437,
+      "step": 736
+    },
+    {
+      "epoch": 0.12283333333333334,
+      "grad_norm": 29.875,
+      "grad_norm_var": 6.917122395833333,
+      "learning_rate": 9.63329950732086e-05,
+      "loss": 7.5027,
+      "loss/crossentropy": 1.5925196409225464,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.197333712130785,
+      "step": 737
+    },
+    {
+      "epoch": 0.123,
+      "grad_norm": 29.375,
+      "grad_norm_var": 7.195572916666666,
+      "learning_rate": 9.632314766195781e-05,
+      "loss": 7.0563,
+      "loss/crossentropy": 1.354059174656868,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21647107601165771,
+      "step": 738
+    },
+    {
+      "epoch": 0.12316666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 7.145768229166666,
+      "learning_rate": 9.631328755095333e-05,
+      "loss": 7.8115,
+      "loss/crossentropy": 1.5845647156238556,
+      "loss/hidden": 3.7734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21428315714001656,
+      "step": 739
+    },
+    {
+      "epoch": 0.12333333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 6.900455729166667,
+      "learning_rate": 9.630341474289842e-05,
+      "loss": 7.2536,
+      "loss/crossentropy": 2.305964082479477,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17266706749796867,
+      "step": 740
+    },
+    {
+      "epoch": 0.1235,
+      "grad_norm": 30.875,
+      "grad_norm_var": 6.803125,
+      "learning_rate": 9.629352924049975e-05,
+      "loss": 6.9382,
+      "loss/crossentropy": 1.4626246094703674,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17279667034745216,
+      "step": 741
+    },
+    {
+      "epoch": 0.12366666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 5.787955729166667,
+      "learning_rate": 9.628363104646747e-05,
+      "loss": 7.1037,
+      "loss/crossentropy": 1.7452305853366852,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15795812010765076,
+      "step": 742
+    },
+    {
+      "epoch": 0.12383333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 5.808072916666666,
+      "learning_rate": 9.627372016351524e-05,
+      "loss": 7.7238,
+      "loss/crossentropy": 1.6585390865802765,
+      "loss/hidden": 4.01171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3405285645276308,
+      "step": 743
+    },
+    {
+      "epoch": 0.124,
+      "grad_norm": 29.5,
+      "grad_norm_var": 5.808072916666666,
+      "learning_rate": 9.626379659436017e-05,
+      "loss": 7.1142,
+      "loss/crossentropy": 1.8966160714626312,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16098260693252087,
+      "step": 744
+    },
+    {
+      "epoch": 0.12416666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 5.922916666666667,
+      "learning_rate": 9.62538603417229e-05,
+      "loss": 7.7227,
+      "loss/crossentropy": 2.1238047182559967,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21814902126789093,
+      "step": 745
+    },
+    {
+      "epoch": 0.12433333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 5.6384765625,
+      "learning_rate": 9.624391140832749e-05,
+      "loss": 7.2153,
+      "loss/crossentropy": 2.1036722362041473,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17869900166988373,
+      "step": 746
+    },
+    {
+      "epoch": 0.1245,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.590559895833334,
+      "learning_rate": 9.623394979690147e-05,
+      "loss": 7.7132,
+      "loss/crossentropy": 2.0859395265579224,
+      "loss/hidden": 3.76171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2453242838382721,
+      "step": 747
+    },
+    {
+      "epoch": 0.12466666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 5.780989583333334,
+      "learning_rate": 9.622397551017592e-05,
+      "loss": 7.3756,
+      "loss/crossentropy": 1.7348367422819138,
+      "loss/hidden": 3.6484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19312499463558197,
+      "step": 748
+    },
+    {
+      "epoch": 0.12483333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.2747395833333333,
+      "learning_rate": 9.62139885508853e-05,
+      "loss": 7.5056,
+      "loss/crossentropy": 1.5878428667783737,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2139638438820839,
+      "step": 749
+    },
+    {
+      "epoch": 0.125,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.2802083333333334,
+      "learning_rate": 9.620398892176762e-05,
+      "loss": 7.2773,
+      "loss/crossentropy": 2.146966904401779,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2128700651228428,
+      "step": 750
+    },
+    {
+      "epoch": 0.12516666666666668,
+      "grad_norm": 36.25,
+      "grad_norm_var": 2.8593098958333334,
+      "learning_rate": 9.619397662556435e-05,
+      "loss": 7.059,
+      "loss/crossentropy": 2.31913298368454,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1911771148443222,
+      "step": 751
+    },
+    {
+      "epoch": 0.12533333333333332,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.87265625,
+      "learning_rate": 9.618395166502037e-05,
+      "loss": 7.4854,
+      "loss/crossentropy": 2.0824431777000427,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17100967466831207,
+      "step": 752
+    },
+    {
+      "epoch": 0.1255,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.9650390625,
+      "learning_rate": 9.617391404288412e-05,
+      "loss": 7.0463,
+      "loss/crossentropy": 1.428141564130783,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15233214199543,
+      "step": 753
+    },
+    {
+      "epoch": 0.12566666666666668,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.64765625,
+      "learning_rate": 9.616386376190745e-05,
+      "loss": 7.5695,
+      "loss/crossentropy": 1.6996584981679916,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2763388231396675,
+      "step": 754
+    },
+    {
+      "epoch": 0.12583333333333332,
+      "grad_norm": 28.0,
+      "grad_norm_var": 3.4791015625,
+      "learning_rate": 9.615380082484571e-05,
+      "loss": 6.7122,
+      "loss/crossentropy": 1.0766998827457428,
+      "loss/hidden": 3.9375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15450757928192616,
+      "step": 755
+    },
+    {
+      "epoch": 0.126,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.4833333333333334,
+      "learning_rate": 9.614372523445771e-05,
+      "loss": 7.1857,
+      "loss/crossentropy": 1.817079246044159,
+      "loss/hidden": 3.734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16611116006970406,
+      "step": 756
+    },
+    {
+      "epoch": 0.12616666666666668,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.5624348958333334,
+      "learning_rate": 9.613363699350575e-05,
+      "loss": 7.4296,
+      "loss/crossentropy": 1.7987887412309647,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26039841026067734,
+      "step": 757
+    },
+    {
+      "epoch": 0.12633333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.7393229166666666,
+      "learning_rate": 9.612353610475555e-05,
+      "loss": 7.2079,
+      "loss/crossentropy": 1.636172592639923,
+      "loss/hidden": 3.7578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28059282898902893,
+      "step": 758
+    },
+    {
+      "epoch": 0.1265,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.7270182291666667,
+      "learning_rate": 9.611342257097632e-05,
+      "loss": 7.6725,
+      "loss/crossentropy": 1.7025777101516724,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22375576198101044,
+      "step": 759
+    },
+    {
+      "epoch": 0.12666666666666668,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.41640625,
+      "learning_rate": 9.610329639494076e-05,
+      "loss": 7.2483,
+      "loss/crossentropy": 1.4852552711963654,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12246158346533775,
+      "step": 760
+    },
+    {
+      "epoch": 0.12683333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.41640625,
+      "learning_rate": 9.609315757942503e-05,
+      "loss": 7.6427,
+      "loss/crossentropy": 1.6396119594573975,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17656655423343182,
+      "step": 761
+    },
+    {
+      "epoch": 0.127,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.426497395833333,
+      "learning_rate": 9.608300612720873e-05,
+      "loss": 6.9386,
+      "loss/crossentropy": 1.6664634048938751,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14652318321168423,
+      "step": 762
+    },
+    {
+      "epoch": 0.12716666666666668,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.426497395833333,
+      "learning_rate": 9.607284204107493e-05,
+      "loss": 7.4406,
+      "loss/crossentropy": 1.963476911187172,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2003205232322216,
+      "step": 763
+    },
+    {
+      "epoch": 0.12733333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.2979166666666666,
+      "learning_rate": 9.606266532381018e-05,
+      "loss": 7.626,
+      "loss/crossentropy": 2.132770359516144,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29830845817923546,
+      "step": 764
+    },
+    {
+      "epoch": 0.1275,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.2979166666666666,
+      "learning_rate": 9.605247597820448e-05,
+      "loss": 7.5946,
+      "loss/crossentropy": 1.7848630100488663,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16466624289751053,
+      "step": 765
+    },
+    {
+      "epoch": 0.12766666666666668,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.5385416666666667,
+      "learning_rate": 9.604227400705133e-05,
+      "loss": 7.128,
+      "loss/crossentropy": 1.4399634450674057,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.132892943918705,
+      "step": 766
+    },
+    {
+      "epoch": 0.12783333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.3988932291666667,
+      "learning_rate": 9.603205941314758e-05,
+      "loss": 7.654,
+      "loss/crossentropy": 1.9748166501522064,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20235471427440643,
+      "step": 767
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.3955729166666666,
+      "learning_rate": 9.602183219929371e-05,
+      "loss": 7.4369,
+      "loss/crossentropy": 1.3751680850982666,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1863851323723793,
+      "step": 768
+    },
+    {
+      "epoch": 0.12816666666666668,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.182747395833333,
+      "learning_rate": 9.601159236829352e-05,
+      "loss": 7.2643,
+      "loss/crossentropy": 1.8310846090316772,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15661721490323544,
+      "step": 769
+    },
+    {
+      "epoch": 0.12833333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.3374348958333333,
+      "learning_rate": 9.600133992295433e-05,
+      "loss": 7.4976,
+      "loss/crossentropy": 1.678751677274704,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26620662584900856,
+      "step": 770
+    },
+    {
+      "epoch": 0.1285,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.3104166666666666,
+      "learning_rate": 9.599107486608689e-05,
+      "loss": 7.1742,
+      "loss/crossentropy": 1.8843847960233688,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14782699570059776,
+      "step": 771
+    },
+    {
+      "epoch": 0.12866666666666668,
+      "grad_norm": 34.5,
+      "grad_norm_var": 1.6455729166666666,
+      "learning_rate": 9.598079720050544e-05,
+      "loss": 7.3332,
+      "loss/crossentropy": 2.1146110594272614,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2215891145169735,
+      "step": 772
+    },
+    {
+      "epoch": 0.12883333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.109375,
+      "learning_rate": 9.597050692902765e-05,
+      "loss": 7.2207,
+      "loss/crossentropy": 2.3083147406578064,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20275061577558517,
+      "step": 773
+    },
+    {
+      "epoch": 0.129,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.0957682291666666,
+      "learning_rate": 9.596020405447466e-05,
+      "loss": 7.2527,
+      "loss/crossentropy": 1.648669645190239,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19592230394482613,
+      "step": 774
+    },
+    {
+      "epoch": 0.12916666666666668,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.20390625,
+      "learning_rate": 9.594988857967106e-05,
+      "loss": 7.4228,
+      "loss/crossentropy": 1.8587908148765564,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28279590234160423,
+      "step": 775
+    },
+    {
+      "epoch": 0.12933333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.6497395833333335,
+      "learning_rate": 9.593956050744492e-05,
+      "loss": 7.4491,
+      "loss/crossentropy": 1.4063522219657898,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2254265546798706,
+      "step": 776
+    },
+    {
+      "epoch": 0.1295,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.9619140625,
+      "learning_rate": 9.59292198406277e-05,
+      "loss": 7.1803,
+      "loss/crossentropy": 2.2121740579605103,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20930473133921623,
+      "step": 777
+    },
+    {
+      "epoch": 0.12966666666666668,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.9559895833333334,
+      "learning_rate": 9.591886658205438e-05,
+      "loss": 7.3083,
+      "loss/crossentropy": 2.3242568969726562,
+      "loss/hidden": 3.16015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18255580589175224,
+      "step": 778
+    },
+    {
+      "epoch": 0.12983333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.88125,
+      "learning_rate": 9.590850073456336e-05,
+      "loss": 7.064,
+      "loss/crossentropy": 1.8441268801689148,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2616273760795593,
+      "step": 779
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 34.5,
+      "grad_norm_var": 3.496809895833333,
+      "learning_rate": 9.589812230099649e-05,
+      "loss": 7.2112,
+      "loss/crossentropy": 1.279086783528328,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2156633324921131,
+      "step": 780
+    },
+    {
+      "epoch": 0.13016666666666668,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.746809895833333,
+      "learning_rate": 9.588773128419906e-05,
+      "loss": 6.8557,
+      "loss/crossentropy": 1.6915069222450256,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1882520094513893,
+      "step": 781
+    },
+    {
+      "epoch": 0.13033333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.116080729166667,
+      "learning_rate": 9.587732768701986e-05,
+      "loss": 6.9332,
+      "loss/crossentropy": 2.2176692485809326,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1980743519961834,
+      "step": 782
+    },
+    {
+      "epoch": 0.1305,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.1059895833333333,
+      "learning_rate": 9.586691151231107e-05,
+      "loss": 7.5209,
+      "loss/crossentropy": 1.8367903232574463,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19657382927834988,
+      "step": 783
+    },
+    {
+      "epoch": 0.13066666666666665,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.4208333333333334,
+      "learning_rate": 9.585648276292836e-05,
+      "loss": 7.2173,
+      "loss/crossentropy": 1.5557065308094025,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20581690594553947,
+      "step": 784
+    },
+    {
+      "epoch": 0.13083333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.405989583333333,
+      "learning_rate": 9.584604144173083e-05,
+      "loss": 7.4191,
+      "loss/crossentropy": 1.658543512225151,
+      "loss/hidden": 3.640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16221168637275696,
+      "step": 785
+    },
+    {
+      "epoch": 0.131,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.981184895833333,
+      "learning_rate": 9.5835587551581e-05,
+      "loss": 7.1631,
+      "loss/crossentropy": 1.5449568629264832,
+      "loss/hidden": 3.74609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1926671303808689,
+      "step": 786
+    },
+    {
+      "epoch": 0.13116666666666665,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.9697916666666666,
+      "learning_rate": 9.58251210953449e-05,
+      "loss": 7.1555,
+      "loss/crossentropy": 1.889558732509613,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21849562227725983,
+      "step": 787
+    },
+    {
+      "epoch": 0.13133333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.9785807291666666,
+      "learning_rate": 9.581464207589195e-05,
+      "loss": 7.1529,
+      "loss/crossentropy": 1.3312578797340393,
+      "loss/hidden": 4.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1593394074589014,
+      "step": 788
+    },
+    {
+      "epoch": 0.1315,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.9723307291666667,
+      "learning_rate": 9.580415049609503e-05,
+      "loss": 7.4802,
+      "loss/crossentropy": 1.859999030828476,
+      "loss/hidden": 3.9140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28401579707860947,
+      "step": 789
+    },
+    {
+      "epoch": 0.13166666666666665,
+      "grad_norm": 34.0,
+      "grad_norm_var": 2.703125,
+      "learning_rate": 9.579364635883048e-05,
+      "loss": 7.448,
+      "loss/crossentropy": 1.7404861450195312,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1754416935145855,
+      "step": 790
+    },
+    {
+      "epoch": 0.13183333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.7864583333333335,
+      "learning_rate": 9.578312966697807e-05,
+      "loss": 7.3509,
+      "loss/crossentropy": 1.383358120918274,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17976827174425125,
+      "step": 791
+    },
+    {
+      "epoch": 0.132,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.7372395833333334,
+      "learning_rate": 9.577260042342097e-05,
+      "loss": 7.2368,
+      "loss/crossentropy": 2.1498415172100067,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.189455796033144,
+      "step": 792
+    },
+    {
+      "epoch": 0.13216666666666665,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.528125,
+      "learning_rate": 9.576205863104588e-05,
+      "loss": 7.4449,
+      "loss/crossentropy": 1.4165163487195969,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16151496767997742,
+      "step": 793
+    },
+    {
+      "epoch": 0.13233333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.59140625,
+      "learning_rate": 9.575150429274287e-05,
+      "loss": 7.2671,
+      "loss/crossentropy": 2.6034965217113495,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18676010519266129,
+      "step": 794
+    },
+    {
+      "epoch": 0.1325,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.6393229166666665,
+      "learning_rate": 9.574093741140549e-05,
+      "loss": 7.2838,
+      "loss/crossentropy": 1.7625453174114227,
+      "loss/hidden": 3.921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28167110681533813,
+      "step": 795
+    },
+    {
+      "epoch": 0.13266666666666665,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.6393229166666665,
+      "learning_rate": 9.573035798993069e-05,
+      "loss": 7.3887,
+      "loss/crossentropy": 1.8252238184213638,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1730077564716339,
+      "step": 796
+    },
+    {
+      "epoch": 0.13283333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.832291666666667,
+      "learning_rate": 9.571976603121888e-05,
+      "loss": 7.0188,
+      "loss/crossentropy": 1.136892981827259,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12606851570308208,
+      "step": 797
+    },
+    {
+      "epoch": 0.133,
+      "grad_norm": 28.375,
+      "grad_norm_var": 3.333268229166667,
+      "learning_rate": 9.570916153817391e-05,
+      "loss": 7.1889,
+      "loss/crossentropy": 1.9445914030075073,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21927091479301453,
+      "step": 798
+    },
+    {
+      "epoch": 0.13316666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.4197265625,
+      "learning_rate": 9.569854451370307e-05,
+      "loss": 7.5546,
+      "loss/crossentropy": 1.389248102903366,
+      "loss/hidden": 3.6875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21863888576626778,
+      "step": 799
+    },
+    {
+      "epoch": 0.13333333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.0994140625,
+      "learning_rate": 9.568791496071706e-05,
+      "loss": 7.0812,
+      "loss/crossentropy": 1.6537730991840363,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1779423952102661,
+      "step": 800
+    },
+    {
+      "epoch": 0.1335,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.10390625,
+      "learning_rate": 9.567727288213005e-05,
+      "loss": 7.2501,
+      "loss/crossentropy": 2.202036827802658,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22605827823281288,
+      "step": 801
+    },
+    {
+      "epoch": 0.13366666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.978125,
+      "learning_rate": 9.56666182808596e-05,
+      "loss": 7.2866,
+      "loss/crossentropy": 1.584087774157524,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17102529108524323,
+      "step": 802
+    },
+    {
+      "epoch": 0.13383333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.9375,
+      "learning_rate": 9.565595115982678e-05,
+      "loss": 7.6211,
+      "loss/crossentropy": 2.211240530014038,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2555403672158718,
+      "step": 803
+    },
+    {
+      "epoch": 0.134,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.2900390625,
+      "learning_rate": 9.5645271521956e-05,
+      "loss": 7.3061,
+      "loss/crossentropy": 1.7899938523769379,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32840781658887863,
+      "step": 804
+    },
+    {
+      "epoch": 0.13416666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.3421223958333335,
+      "learning_rate": 9.563457937017515e-05,
+      "loss": 7.3661,
+      "loss/crossentropy": 1.5019114315509796,
+      "loss/hidden": 3.7734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24091443419456482,
+      "step": 805
+    },
+    {
+      "epoch": 0.13433333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.7639973958333335,
+      "learning_rate": 9.562387470741554e-05,
+      "loss": 7.2827,
+      "loss/crossentropy": 1.9453937411308289,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25887253507971764,
+      "step": 806
+    },
+    {
+      "epoch": 0.1345,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.7270833333333333,
+      "learning_rate": 9.561315753661194e-05,
+      "loss": 6.8771,
+      "loss/crossentropy": 1.449354350566864,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18446829169988632,
+      "step": 807
+    },
+    {
+      "epoch": 0.13466666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.517708333333333,
+      "learning_rate": 9.560242786070249e-05,
+      "loss": 7.3977,
+      "loss/crossentropy": 1.3549923300743103,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1514573572203517,
+      "step": 808
+    },
+    {
+      "epoch": 0.13483333333333333,
+      "grad_norm": 27.625,
+      "grad_norm_var": 3.301041666666667,
+      "learning_rate": 9.55916856826288e-05,
+      "loss": 7.1812,
+      "loss/crossentropy": 1.5787281095981598,
+      "loss/hidden": 3.80859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23648673295974731,
+      "step": 809
+    },
+    {
+      "epoch": 0.135,
+      "grad_norm": 36.5,
+      "grad_norm_var": 5.078059895833333,
+      "learning_rate": 9.558093100533591e-05,
+      "loss": 7.4789,
+      "loss/crossentropy": 2.1049546599388123,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34059297293424606,
+      "step": 810
+    },
+    {
+      "epoch": 0.13516666666666666,
+      "grad_norm": 31.125,
+      "grad_norm_var": 5.071875,
+      "learning_rate": 9.557016383177227e-05,
+      "loss": 7.0125,
+      "loss/crossentropy": 1.8799607157707214,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18389278650283813,
+      "step": 811
+    },
+    {
+      "epoch": 0.13533333333333333,
+      "grad_norm": 56.0,
+      "grad_norm_var": 43.010416666666664,
+      "learning_rate": 9.555938416488977e-05,
+      "loss": 7.2583,
+      "loss/crossentropy": 1.8990149199962616,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20581569522619247,
+      "step": 812
+    },
+    {
+      "epoch": 0.1355,
+      "grad_norm": 33.75,
+      "grad_norm_var": 43.010416666666664,
+      "learning_rate": 9.55485920076437e-05,
+      "loss": 7.3948,
+      "loss/crossentropy": 1.287716194987297,
+      "loss/hidden": 3.76171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17925803363323212,
+      "step": 813
+    },
+    {
+      "epoch": 0.13566666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 42.44264322916667,
+      "learning_rate": 9.553778736299279e-05,
+      "loss": 7.1165,
+      "loss/crossentropy": 1.4526919722557068,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1660779807716608,
+      "step": 814
+    },
+    {
+      "epoch": 0.13583333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 42.549739583333334,
+      "learning_rate": 9.552697023389922e-05,
+      "loss": 7.1825,
+      "loss/crossentropy": 1.9017798006534576,
+      "loss/hidden": 3.1015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1294990535825491,
+      "step": 815
+    },
+    {
+      "epoch": 0.136,
+      "grad_norm": 31.625,
+      "grad_norm_var": 42.08125,
+      "learning_rate": 9.551614062332856e-05,
+      "loss": 7.2658,
+      "loss/crossentropy": 1.729724794626236,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1733120009303093,
+      "step": 816
+    },
+    {
+      "epoch": 0.13616666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 41.96875,
+      "learning_rate": 9.550529853424979e-05,
+      "loss": 6.8834,
+      "loss/crossentropy": 0.9069706723093987,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1328633949160576,
+      "step": 817
+    },
+    {
+      "epoch": 0.13633333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 42.2353515625,
+      "learning_rate": 9.549444396963534e-05,
+      "loss": 7.0983,
+      "loss/crossentropy": 1.7946209907531738,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16930393874645233,
+      "step": 818
+    },
+    {
+      "epoch": 0.1365,
+      "grad_norm": 30.25,
+      "grad_norm_var": 42.4509765625,
+      "learning_rate": 9.548357693246105e-05,
+      "loss": 7.2424,
+      "loss/crossentropy": 1.4191659688949585,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18897976353764534,
+      "step": 819
+    },
+    {
+      "epoch": 0.13666666666666666,
+      "grad_norm": 38.5,
+      "grad_norm_var": 43.331705729166664,
+      "learning_rate": 9.547269742570619e-05,
+      "loss": 7.3048,
+      "loss/crossentropy": 1.757742017507553,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21289073675870895,
+      "step": 820
+    },
+    {
+      "epoch": 0.13683333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 42.94557291666667,
+      "learning_rate": 9.546180545235344e-05,
+      "loss": 7.3922,
+      "loss/crossentropy": 1.9324754476547241,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20474570244550705,
+      "step": 821
+    },
+    {
+      "epoch": 0.137,
+      "grad_norm": 28.5,
+      "grad_norm_var": 44.143489583333334,
+      "learning_rate": 9.545090101538887e-05,
+      "loss": 6.975,
+      "loss/crossentropy": 1.586461067199707,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16012245789170265,
+      "step": 822
+    },
+    {
+      "epoch": 0.13716666666666666,
+      "grad_norm": 31.75,
+      "grad_norm_var": 44.22682291666667,
+      "learning_rate": 9.543998411780201e-05,
+      "loss": 7.2944,
+      "loss/crossentropy": 1.5576333180069923,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17472228594124317,
+      "step": 823
+    },
+    {
+      "epoch": 0.13733333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 44.83932291666667,
+      "learning_rate": 9.54290547625858e-05,
+      "loss": 7.2929,
+      "loss/crossentropy": 1.8102333843708038,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20905954018235207,
+      "step": 824
+    },
+    {
+      "epoch": 0.1375,
+      "grad_norm": 35.25,
+      "grad_norm_var": 42.865559895833336,
+      "learning_rate": 9.541811295273656e-05,
+      "loss": 7.3067,
+      "loss/crossentropy": 0.9989386051893234,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17145405896008015,
+      "step": 825
+    },
+    {
+      "epoch": 0.13766666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 43.06223958333333,
+      "learning_rate": 9.540715869125407e-05,
+      "loss": 7.1798,
+      "loss/crossentropy": 1.7404059767723083,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1699448525905609,
+      "step": 826
+    },
+    {
+      "epoch": 0.13783333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 42.86764322916667,
+      "learning_rate": 9.53961919811415e-05,
+      "loss": 7.2755,
+      "loss/crossentropy": 1.1501479148864746,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19713975861668587,
+      "step": 827
+    },
+    {
+      "epoch": 0.138,
+      "grad_norm": 28.375,
+      "grad_norm_var": 6.796875,
+      "learning_rate": 9.538521282540542e-05,
+      "loss": 7.0676,
+      "loss/crossentropy": 2.0919524431228638,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15730703622102737,
+      "step": 828
+    },
+    {
+      "epoch": 0.13816666666666666,
+      "grad_norm": 29.625,
+      "grad_norm_var": 6.6400390625,
+      "learning_rate": 9.537422122705585e-05,
+      "loss": 7.3261,
+      "loss/crossentropy": 1.4668185710906982,
+      "loss/hidden": 3.74609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1639045923948288,
+      "step": 829
+    },
+    {
+      "epoch": 0.13833333333333334,
+      "grad_norm": 30.5,
+      "grad_norm_var": 6.466080729166666,
+      "learning_rate": 9.536321718910619e-05,
+      "loss": 7.3588,
+      "loss/crossentropy": 2.1604126393795013,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18609627708792686,
+      "step": 830
+    },
+    {
+      "epoch": 0.1385,
+      "grad_norm": 31.0,
+      "grad_norm_var": 6.472916666666666,
+      "learning_rate": 9.535220071457325e-05,
+      "loss": 7.2914,
+      "loss/crossentropy": 1.7398014664649963,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1664908565580845,
+      "step": 831
+    },
+    {
+      "epoch": 0.13866666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 6.770768229166666,
+      "learning_rate": 9.534117180647728e-05,
+      "loss": 7.2502,
+      "loss/crossentropy": 1.3350512720644474,
+      "loss/hidden": 3.765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17889262828975916,
+      "step": 832
+    },
+    {
+      "epoch": 0.13883333333333334,
+      "grad_norm": 29.875,
+      "grad_norm_var": 6.826822916666667,
+      "learning_rate": 9.533013046784189e-05,
+      "loss": 7.3238,
+      "loss/crossentropy": 2.4168621003627777,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22906067222356796,
+      "step": 833
+    },
+    {
+      "epoch": 0.139,
+      "grad_norm": 29.125,
+      "grad_norm_var": 7.0853515625,
+      "learning_rate": 9.531907670169415e-05,
+      "loss": 7.1844,
+      "loss/crossentropy": 1.3794061094522476,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16571270301938057,
+      "step": 834
+    },
+    {
+      "epoch": 0.13916666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 7.0853515625,
+      "learning_rate": 9.530801051106449e-05,
+      "loss": 6.8716,
+      "loss/crossentropy": 1.2973438501358032,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12957068346440792,
+      "step": 835
+    },
+    {
+      "epoch": 0.13933333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.381705729166667,
+      "learning_rate": 9.52969318989868e-05,
+      "loss": 7.2721,
+      "loss/crossentropy": 1.8947395384311676,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18677709624171257,
+      "step": 836
+    },
+    {
+      "epoch": 0.1395,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.392708333333333,
+      "learning_rate": 9.528584086849832e-05,
+      "loss": 7.1715,
+      "loss/crossentropy": 1.7173646092414856,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13541538082063198,
+      "step": 837
+    },
+    {
+      "epoch": 0.13966666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.130989583333333,
+      "learning_rate": 9.527473742263973e-05,
+      "loss": 7.3485,
+      "loss/crossentropy": 1.770973801612854,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16501959785819054,
+      "step": 838
+    },
+    {
+      "epoch": 0.13983333333333334,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.4353515625,
+      "learning_rate": 9.526362156445507e-05,
+      "loss": 7.0468,
+      "loss/crossentropy": 1.4360809028148651,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14953473582863808,
+      "step": 839
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.3910807291666667,
+      "learning_rate": 9.525249329699188e-05,
+      "loss": 6.9655,
+      "loss/crossentropy": 1.7091718018054962,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20031239837408066,
+      "step": 840
+    },
+    {
+      "epoch": 0.14016666666666666,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.0082682291666667,
+      "learning_rate": 9.524135262330098e-05,
+      "loss": 7.5125,
+      "loss/crossentropy": 1.9940660297870636,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1591429878026247,
+      "step": 841
+    },
+    {
+      "epoch": 0.14033333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.9926432291666667,
+      "learning_rate": 9.523019954643669e-05,
+      "loss": 7.2906,
+      "loss/crossentropy": 2.18462273478508,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19053708761930466,
+      "step": 842
+    },
+    {
+      "epoch": 0.1405,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.8624348958333334,
+      "learning_rate": 9.521903406945664e-05,
+      "loss": 7.1004,
+      "loss/crossentropy": 1.8117623925209045,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15475555136799812,
+      "step": 843
+    },
+    {
+      "epoch": 0.14066666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.5434895833333333,
+      "learning_rate": 9.520785619542196e-05,
+      "loss": 7.9422,
+      "loss/crossentropy": 1.4361202418804169,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16791957058012486,
+      "step": 844
+    },
+    {
+      "epoch": 0.14083333333333334,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.7207682291666666,
+      "learning_rate": 9.519666592739709e-05,
+      "loss": 6.9648,
+      "loss/crossentropy": 1.5224987119436264,
+      "loss/hidden": 3.86328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20675836876034737,
+      "step": 845
+    },
+    {
+      "epoch": 0.141,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.7186848958333334,
+      "learning_rate": 9.518546326844993e-05,
+      "loss": 7.3723,
+      "loss/crossentropy": 1.5077387690544128,
+      "loss/hidden": 4.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22337773814797401,
+      "step": 846
+    },
+    {
+      "epoch": 0.14116666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.7327473958333333,
+      "learning_rate": 9.517424822165175e-05,
+      "loss": 7.0724,
+      "loss/crossentropy": 1.1534551531076431,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17478630878031254,
+      "step": 847
+    },
+    {
+      "epoch": 0.14133333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.4233723958333333,
+      "learning_rate": 9.516302079007719e-05,
+      "loss": 6.9591,
+      "loss/crossentropy": 1.4297308921813965,
+      "loss/hidden": 3.8046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20949728786945343,
+      "step": 848
+    },
+    {
+      "epoch": 0.1415,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.4233723958333333,
+      "learning_rate": 9.515178097680437e-05,
+      "loss": 7.5144,
+      "loss/crossentropy": 1.7045030891895294,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14692462421953678,
+      "step": 849
+    },
+    {
+      "epoch": 0.14166666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.3061848958333333,
+      "learning_rate": 9.51405287849147e-05,
+      "loss": 7.146,
+      "loss/crossentropy": 1.9769637882709503,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21150068566203117,
+      "step": 850
+    },
+    {
+      "epoch": 0.14183333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.3427083333333334,
+      "learning_rate": 9.512926421749304e-05,
+      "loss": 7.3401,
+      "loss/crossentropy": 1.919234961271286,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28474786318838596,
+      "step": 851
+    },
+    {
+      "epoch": 0.142,
+      "grad_norm": 34.0,
+      "grad_norm_var": 1.9177083333333333,
+      "learning_rate": 9.511798727762764e-05,
+      "loss": 7.1106,
+      "loss/crossentropy": 1.8584485948085785,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16243590787053108,
+      "step": 852
+    },
+    {
+      "epoch": 0.14216666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.1143229166666666,
+      "learning_rate": 9.510669796841014e-05,
+      "loss": 7.5742,
+      "loss/crossentropy": 2.2767185270786285,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19640379771590233,
+      "step": 853
+    },
+    {
+      "epoch": 0.14233333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.0035807291666665,
+      "learning_rate": 9.509539629293558e-05,
+      "loss": 7.4331,
+      "loss/crossentropy": 1.7738919258117676,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19907983765006065,
+      "step": 854
+    },
+    {
+      "epoch": 0.1425,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.7608723958333334,
+      "learning_rate": 9.508408225430237e-05,
+      "loss": 6.8944,
+      "loss/crossentropy": 1.8578266501426697,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18979492783546448,
+      "step": 855
+    },
+    {
+      "epoch": 0.14266666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.6431640625,
+      "learning_rate": 9.507275585561229e-05,
+      "loss": 7.4694,
+      "loss/crossentropy": 1.788054645061493,
+      "loss/hidden": 4.0859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20594334974884987,
+      "step": 856
+    },
+    {
+      "epoch": 0.14283333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.6489583333333333,
+      "learning_rate": 9.506141709997057e-05,
+      "loss": 7.5055,
+      "loss/crossentropy": 1.4296987056732178,
+      "loss/hidden": 3.9609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32834842428565025,
+      "step": 857
+    },
+    {
+      "epoch": 0.143,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.6483723958333334,
+      "learning_rate": 9.505006599048579e-05,
+      "loss": 7.2555,
+      "loss/crossentropy": 1.8494880199432373,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22622110322117805,
+      "step": 858
+    },
+    {
+      "epoch": 0.14316666666666666,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.9712890625,
+      "learning_rate": 9.503870253026991e-05,
+      "loss": 6.9123,
+      "loss/crossentropy": 1.61825492978096,
+      "loss/hidden": 3.6484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1566757969558239,
+      "step": 859
+    },
+    {
+      "epoch": 0.14333333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.96640625,
+      "learning_rate": 9.50273267224383e-05,
+      "loss": 7.1494,
+      "loss/crossentropy": 2.0159516036510468,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1681060865521431,
+      "step": 860
+    },
+    {
+      "epoch": 0.1435,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.12265625,
+      "learning_rate": 9.501593857010969e-05,
+      "loss": 7.4906,
+      "loss/crossentropy": 1.5668489038944244,
+      "loss/hidden": 3.7265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21499893069267273,
+      "step": 861
+    },
+    {
+      "epoch": 0.14366666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.16640625,
+      "learning_rate": 9.50045380764062e-05,
+      "loss": 7.1057,
+      "loss/crossentropy": 1.2165893912315369,
+      "loss/hidden": 3.9609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23781537264585495,
+      "step": 862
+    },
+    {
+      "epoch": 0.14383333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.1697265625,
+      "learning_rate": 9.499312524445336e-05,
+      "loss": 6.6869,
+      "loss/crossentropy": 1.0919674932956696,
+      "loss/hidden": 3.77734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18617354333400726,
+      "step": 863
+    },
+    {
+      "epoch": 0.144,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.65390625,
+      "learning_rate": 9.498170007738005e-05,
+      "loss": 7.0988,
+      "loss/crossentropy": 1.6911379247903824,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17686204239726067,
+      "step": 864
+    },
+    {
+      "epoch": 0.14416666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.6389973958333335,
+      "learning_rate": 9.497026257831855e-05,
+      "loss": 7.4592,
+      "loss/crossentropy": 1.8146097958087921,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18080419301986694,
+      "step": 865
+    },
+    {
+      "epoch": 0.14433333333333334,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.5809895833333334,
+      "learning_rate": 9.495881275040453e-05,
+      "loss": 7.643,
+      "loss/crossentropy": 2.3058604300022125,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16926423460245132,
+      "step": 866
+    },
+    {
+      "epoch": 0.1445,
+      "grad_norm": 39.75,
+      "grad_norm_var": 7.468684895833333,
+      "learning_rate": 9.494735059677699e-05,
+      "loss": 7.3949,
+      "loss/crossentropy": 1.401999369263649,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15934792906045914,
+      "step": 867
+    },
+    {
+      "epoch": 0.14466666666666667,
+      "grad_norm": 49.5,
+      "grad_norm_var": 27.7962890625,
+      "learning_rate": 9.493587612057837e-05,
+      "loss": 7.5433,
+      "loss/crossentropy": 1.87227264046669,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1974172182381153,
+      "step": 868
+    },
+    {
+      "epoch": 0.14483333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 27.7962890625,
+      "learning_rate": 9.492438932495444e-05,
+      "loss": 7.2661,
+      "loss/crossentropy": 2.4418474435806274,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17274828255176544,
+      "step": 869
+    },
+    {
+      "epoch": 0.145,
+      "grad_norm": 34.5,
+      "grad_norm_var": 27.93515625,
+      "learning_rate": 9.491289021305441e-05,
+      "loss": 7.1889,
+      "loss/crossentropy": 1.7477463483810425,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2018323615193367,
+      "step": 870
+    },
+    {
+      "epoch": 0.14516666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 27.4087890625,
+      "learning_rate": 9.490137878803079e-05,
+      "loss": 7.4709,
+      "loss/crossentropy": 2.011961728334427,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23878171294927597,
+      "step": 871
+    },
+    {
+      "epoch": 0.14533333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 27.3125,
+      "learning_rate": 9.488985505303951e-05,
+      "loss": 7.3868,
+      "loss/crossentropy": 2.2182995676994324,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1687772162258625,
+      "step": 872
+    },
+    {
+      "epoch": 0.1455,
+      "grad_norm": 32.5,
+      "grad_norm_var": 27.0009765625,
+      "learning_rate": 9.487831901123988e-05,
+      "loss": 7.4503,
+      "loss/crossentropy": 1.601034089922905,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19292880222201347,
+      "step": 873
+    },
+    {
+      "epoch": 0.14566666666666667,
+      "grad_norm": 59.75,
+      "grad_norm_var": 71.50358072916667,
+      "learning_rate": 9.486677066579456e-05,
+      "loss": 7.5517,
+      "loss/crossentropy": 2.098694920539856,
+      "loss/hidden": 3.72265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27684100717306137,
+      "step": 874
+    },
+    {
+      "epoch": 0.14583333333333334,
+      "grad_norm": 34.75,
+      "grad_norm_var": 69.02233072916667,
+      "learning_rate": 9.485521001986962e-05,
+      "loss": 7.2567,
+      "loss/crossentropy": 1.3583464622497559,
+      "loss/hidden": 3.91015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28156812116503716,
+      "step": 875
+    },
+    {
+      "epoch": 0.146,
+      "grad_norm": 31.75,
+      "grad_norm_var": 68.43932291666667,
+      "learning_rate": 9.484363707663442e-05,
+      "loss": 7.5572,
+      "loss/crossentropy": 1.850482553243637,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17428970709443092,
+      "step": 876
+    },
+    {
+      "epoch": 0.14616666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 69.99791666666667,
+      "learning_rate": 9.483205183926181e-05,
+      "loss": 6.8266,
+      "loss/crossentropy": 1.5660331845283508,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1589797157794237,
+      "step": 877
+    },
+    {
+      "epoch": 0.14633333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 69.99791666666667,
+      "learning_rate": 9.48204543109279e-05,
+      "loss": 7.1353,
+      "loss/crossentropy": 2.137140065431595,
+      "loss/hidden": 3.109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15021326392889023,
+      "step": 878
+    },
+    {
+      "epoch": 0.1465,
+      "grad_norm": 29.625,
+      "grad_norm_var": 70.76223958333334,
+      "learning_rate": 9.480884449481225e-05,
+      "loss": 7.2885,
+      "loss/crossentropy": 1.8290425837039948,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17951563745737076,
+      "step": 879
+    },
+    {
+      "epoch": 0.14666666666666667,
+      "grad_norm": 27.625,
+      "grad_norm_var": 70.996875,
+      "learning_rate": 9.479722239409775e-05,
+      "loss": 7.0434,
+      "loss/crossentropy": 1.5461790561676025,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19305451214313507,
+      "step": 880
+    },
+    {
+      "epoch": 0.14683333333333334,
+      "grad_norm": 29.0,
+      "grad_norm_var": 71.696875,
+      "learning_rate": 9.478558801197065e-05,
+      "loss": 7.2403,
+      "loss/crossentropy": 1.997674286365509,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24135951325297356,
+      "step": 881
+    },
+    {
+      "epoch": 0.147,
+      "grad_norm": 31.375,
+      "grad_norm_var": 71.6400390625,
+      "learning_rate": 9.47739413516206e-05,
+      "loss": 7.4697,
+      "loss/crossentropy": 1.3830721974372864,
+      "loss/hidden": 3.7890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23223229497671127,
+      "step": 882
+    },
+    {
+      "epoch": 0.14716666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 70.85390625,
+      "learning_rate": 9.476228241624059e-05,
+      "loss": 7.1558,
+      "loss/crossentropy": 1.7543727606534958,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19825738668441772,
+      "step": 883
+    },
+    {
+      "epoch": 0.14733333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 54.19479166666667,
+      "learning_rate": 9.475061120902698e-05,
+      "loss": 7.4787,
+      "loss/crossentropy": 1.4276432394981384,
+      "loss/hidden": 3.921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1671007014811039,
+      "step": 884
+    },
+    {
+      "epoch": 0.1475,
+      "grad_norm": 31.625,
+      "grad_norm_var": 54.316080729166664,
+      "learning_rate": 9.473892773317952e-05,
+      "loss": 7.0177,
+      "loss/crossentropy": 1.1863575875759125,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25031600147485733,
+      "step": 885
+    },
+    {
+      "epoch": 0.14766666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 55.2025390625,
+      "learning_rate": 9.472723199190125e-05,
+      "loss": 7.3024,
+      "loss/crossentropy": 1.396965429186821,
+      "loss/hidden": 3.91796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21523795649409294,
+      "step": 886
+    },
+    {
+      "epoch": 0.14783333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 55.5634765625,
+      "learning_rate": 9.47155239883987e-05,
+      "loss": 7.5251,
+      "loss/crossentropy": 1.6711354553699493,
+      "loss/hidden": 3.8984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2758805975317955,
+      "step": 887
+    },
+    {
+      "epoch": 0.148,
+      "grad_norm": 29.25,
+      "grad_norm_var": 56.10670572916667,
+      "learning_rate": 9.470380372588162e-05,
+      "loss": 7.17,
+      "loss/crossentropy": 1.6715282797813416,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2020905315876007,
+      "step": 888
+    },
+    {
+      "epoch": 0.14816666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 56.10670572916667,
+      "learning_rate": 9.46920712075632e-05,
+      "loss": 7.5769,
+      "loss/crossentropy": 1.5983088314533234,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16166867688298225,
+      "step": 889
+    },
+    {
+      "epoch": 0.14833333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.083268229166667,
+      "learning_rate": 9.468032643665998e-05,
+      "loss": 7.3204,
+      "loss/crossentropy": 2.1481020152568817,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20704419165849686,
+      "step": 890
+    },
+    {
+      "epoch": 0.1485,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.9434895833333334,
+      "learning_rate": 9.466856941639188e-05,
+      "loss": 7.3109,
+      "loss/crossentropy": 1.8172811567783356,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19148655608296394,
+      "step": 891
+    },
+    {
+      "epoch": 0.14866666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.9212890625,
+      "learning_rate": 9.465680014998213e-05,
+      "loss": 7.5316,
+      "loss/crossentropy": 1.9877941012382507,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16703401878476143,
+      "step": 892
+    },
+    {
+      "epoch": 0.14883333333333335,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.9603515625,
+      "learning_rate": 9.464501864065735e-05,
+      "loss": 7.0551,
+      "loss/crossentropy": 1.422202154994011,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19844955578446388,
+      "step": 893
+    },
+    {
+      "epoch": 0.149,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.97890625,
+      "learning_rate": 9.46332248916475e-05,
+      "loss": 7.034,
+      "loss/crossentropy": 1.4157359600067139,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13821185193955898,
+      "step": 894
+    },
+    {
+      "epoch": 0.14916666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.97890625,
+      "learning_rate": 9.46214189061859e-05,
+      "loss": 7.0795,
+      "loss/crossentropy": 1.787828505039215,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1712820678949356,
+      "step": 895
+    },
+    {
+      "epoch": 0.14933333333333335,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.490625,
+      "learning_rate": 9.460960068750924e-05,
+      "loss": 7.2957,
+      "loss/crossentropy": 1.925620973110199,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15570113062858582,
+      "step": 896
+    },
+    {
+      "epoch": 0.1495,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.3119140625,
+      "learning_rate": 9.459777023885755e-05,
+      "loss": 7.2857,
+      "loss/crossentropy": 1.4541048854589462,
+      "loss/hidden": 3.80078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21104299277067184,
+      "step": 897
+    },
+    {
+      "epoch": 0.14966666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.2936848958333333,
+      "learning_rate": 9.458592756347419e-05,
+      "loss": 7.6247,
+      "loss/crossentropy": 1.9160146713256836,
+      "loss/hidden": 3.73046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29131875187158585,
+      "step": 898
+    },
+    {
+      "epoch": 0.14983333333333335,
+      "grad_norm": 27.875,
+      "grad_norm_var": 1.7910807291666666,
+      "learning_rate": 9.457407266460593e-05,
+      "loss": 7.2227,
+      "loss/crossentropy": 1.9192684292793274,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16211543045938015,
+      "step": 899
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.8874348958333333,
+      "learning_rate": 9.456220554550285e-05,
+      "loss": 7.2177,
+      "loss/crossentropy": 1.4649416357278824,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1687542423605919,
+      "step": 900
+    },
+    {
+      "epoch": 0.15016666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.8327473958333333,
+      "learning_rate": 9.45503262094184e-05,
+      "loss": 7.2513,
+      "loss/crossentropy": 1.3999031782150269,
+      "loss/hidden": 4.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28319165110588074,
+      "step": 901
+    },
+    {
+      "epoch": 0.15033333333333335,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.6229166666666666,
+      "learning_rate": 9.453843465960933e-05,
+      "loss": 7.2988,
+      "loss/crossentropy": 1.955479621887207,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2052992843091488,
+      "step": 902
+    },
+    {
+      "epoch": 0.1505,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.5952473958333333,
+      "learning_rate": 9.45265308993358e-05,
+      "loss": 7.0896,
+      "loss/crossentropy": 1.5976478159427643,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15355589985847473,
+      "step": 903
+    },
+    {
+      "epoch": 0.15066666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.4614583333333333,
+      "learning_rate": 9.451461493186129e-05,
+      "loss": 7.3303,
+      "loss/crossentropy": 1.9889639914035797,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1897275187075138,
+      "step": 904
+    },
+    {
+      "epoch": 0.15083333333333335,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.52265625,
+      "learning_rate": 9.450268676045262e-05,
+      "loss": 7.3375,
+      "loss/crossentropy": 1.8655506074428558,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17205933667719364,
+      "step": 905
+    },
+    {
+      "epoch": 0.151,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.4259765625,
+      "learning_rate": 9.449074638837999e-05,
+      "loss": 7.1457,
+      "loss/crossentropy": 1.8997942507266998,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2036127820611,
+      "step": 906
+    },
+    {
+      "epoch": 0.15116666666666667,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.7931640625,
+      "learning_rate": 9.447879381891692e-05,
+      "loss": 6.8358,
+      "loss/crossentropy": 1.8480059802532196,
+      "loss/hidden": 3.12109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18111873418092728,
+      "step": 907
+    },
+    {
+      "epoch": 0.15133333333333332,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.7931640625,
+      "learning_rate": 9.446682905534023e-05,
+      "loss": 6.9958,
+      "loss/crossentropy": 2.2430270612239838,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1975494995713234,
+      "step": 908
+    },
+    {
+      "epoch": 0.1515,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.8389973958333334,
+      "learning_rate": 9.445485210093017e-05,
+      "loss": 7.4726,
+      "loss/crossentropy": 1.7635067105293274,
+      "loss/hidden": 3.76171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25735023617744446,
+      "step": 909
+    },
+    {
+      "epoch": 0.15166666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.8458333333333334,
+      "learning_rate": 9.444286295897028e-05,
+      "loss": 7.2584,
+      "loss/crossentropy": 1.3482669442892075,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18593468144536018,
+      "step": 910
+    },
+    {
+      "epoch": 0.15183333333333332,
+      "grad_norm": 34.0,
+      "grad_norm_var": 2.3494140625,
+      "learning_rate": 9.443086163274745e-05,
+      "loss": 7.2616,
+      "loss/crossentropy": 1.6370986104011536,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1851278953254223,
+      "step": 911
+    },
+    {
+      "epoch": 0.152,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.473372395833333,
+      "learning_rate": 9.44188481255519e-05,
+      "loss": 7.2392,
+      "loss/crossentropy": 1.4670822322368622,
+      "loss/hidden": 3.85546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17405189387500286,
+      "step": 912
+    },
+    {
+      "epoch": 0.15216666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.718489583333333,
+      "learning_rate": 9.440682244067724e-05,
+      "loss": 7.1596,
+      "loss/crossentropy": 2.1715012192726135,
+      "loss/hidden": 3.0859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15273620560765266,
+      "step": 913
+    },
+    {
+      "epoch": 0.15233333333333332,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.7150390625,
+      "learning_rate": 9.439478458142033e-05,
+      "loss": 7.2916,
+      "loss/crossentropy": 1.4728807657957077,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16221768036484718,
+      "step": 914
+    },
+    {
+      "epoch": 0.1525,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.115625,
+      "learning_rate": 9.438273455108144e-05,
+      "loss": 7.2901,
+      "loss/crossentropy": 1.2923684120178223,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2441869042813778,
+      "step": 915
+    },
+    {
+      "epoch": 0.15266666666666667,
+      "grad_norm": 27.375,
+      "grad_norm_var": 2.8494140625,
+      "learning_rate": 9.437067235296418e-05,
+      "loss": 7.207,
+      "loss/crossentropy": 1.625198483467102,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17968312464654446,
+      "step": 916
+    },
+    {
+      "epoch": 0.15283333333333332,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.905989583333333,
+      "learning_rate": 9.43585979903754e-05,
+      "loss": 7.1209,
+      "loss/crossentropy": 1.7370411604642868,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18970072641968727,
+      "step": 917
+    },
+    {
+      "epoch": 0.153,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.9625,
+      "learning_rate": 9.434651146662543e-05,
+      "loss": 7.2779,
+      "loss/crossentropy": 2.1734266877174377,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2356298603117466,
+      "step": 918
+    },
+    {
+      "epoch": 0.15316666666666667,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.02890625,
+      "learning_rate": 9.433441278502783e-05,
+      "loss": 7.6113,
+      "loss/crossentropy": 1.9233836829662323,
+      "loss/hidden": 3.66015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26220428198575974,
+      "step": 919
+    },
+    {
+      "epoch": 0.15333333333333332,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.0494140625,
+      "learning_rate": 9.43223019488995e-05,
+      "loss": 7.1924,
+      "loss/crossentropy": 2.0979688465595245,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21158528700470924,
+      "step": 920
+    },
+    {
+      "epoch": 0.1535,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.789322916666667,
+      "learning_rate": 9.431017896156074e-05,
+      "loss": 7.3643,
+      "loss/crossentropy": 1.5537724941968918,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16366195678710938,
+      "step": 921
+    },
+    {
+      "epoch": 0.15366666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.8499348958333335,
+      "learning_rate": 9.42980438263351e-05,
+      "loss": 6.9375,
+      "loss/crossentropy": 1.477300003170967,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1505614947527647,
+      "step": 922
+    },
+    {
+      "epoch": 0.15383333333333332,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.488541666666667,
+      "learning_rate": 9.428589654654951e-05,
+      "loss": 7.2338,
+      "loss/crossentropy": 1.7841804027557373,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1836482211947441,
+      "step": 923
+    },
+    {
+      "epoch": 0.154,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.4268229166666666,
+      "learning_rate": 9.42737371255342e-05,
+      "loss": 6.8434,
+      "loss/crossentropy": 2.0325754284858704,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17711760848760605,
+      "step": 924
+    },
+    {
+      "epoch": 0.15416666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.7080729166666666,
+      "learning_rate": 9.426156556662276e-05,
+      "loss": 7.6641,
+      "loss/crossentropy": 1.8055310249328613,
+      "loss/hidden": 3.859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2446468248963356,
+      "step": 925
+    },
+    {
+      "epoch": 0.15433333333333332,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.859375,
+      "learning_rate": 9.42493818731521e-05,
+      "loss": 7.6913,
+      "loss/crossentropy": 1.3272765129804611,
+      "loss/hidden": 4.16015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31806548684835434,
+      "step": 926
+    },
+    {
+      "epoch": 0.1545,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.6184895833333335,
+      "learning_rate": 9.423718604846243e-05,
+      "loss": 7.4034,
+      "loss/crossentropy": 1.5088046044111252,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14922618493437767,
+      "step": 927
+    },
+    {
+      "epoch": 0.15466666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.7556640625,
+      "learning_rate": 9.422497809589731e-05,
+      "loss": 7.1531,
+      "loss/crossentropy": 1.4438090920448303,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15729905106127262,
+      "step": 928
+    },
+    {
+      "epoch": 0.15483333333333332,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.6494140625,
+      "learning_rate": 9.421275801880362e-05,
+      "loss": 7.3847,
+      "loss/crossentropy": 1.5666642040014267,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1446811594069004,
+      "step": 929
+    },
+    {
+      "epoch": 0.155,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.7556640625,
+      "learning_rate": 9.420052582053157e-05,
+      "loss": 6.8845,
+      "loss/crossentropy": 2.0941355526447296,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16907286643981934,
+      "step": 930
+    },
+    {
+      "epoch": 0.15516666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.660416666666667,
+      "learning_rate": 9.418828150443469e-05,
+      "loss": 7.1415,
+      "loss/crossentropy": 1.277084156870842,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15413564257323742,
+      "step": 931
+    },
+    {
+      "epoch": 0.15533333333333332,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.101822916666667,
+      "learning_rate": 9.417602507386981e-05,
+      "loss": 7.1045,
+      "loss/crossentropy": 1.610141098499298,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16925229504704475,
+      "step": 932
+    },
+    {
+      "epoch": 0.1555,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.128059895833333,
+      "learning_rate": 9.416375653219709e-05,
+      "loss": 6.7544,
+      "loss/crossentropy": 1.5018357336521149,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1386723332107067,
+      "step": 933
+    },
+    {
+      "epoch": 0.15566666666666668,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.042708333333333,
+      "learning_rate": 9.415147588278005e-05,
+      "loss": 6.9652,
+      "loss/crossentropy": 1.7524566650390625,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15637115016579628,
+      "step": 934
+    },
+    {
+      "epoch": 0.15583333333333332,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.9119140625,
+      "learning_rate": 9.413918312898551e-05,
+      "loss": 7.1223,
+      "loss/crossentropy": 1.4138163179159164,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17044668085873127,
+      "step": 935
+    },
+    {
+      "epoch": 0.156,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.9552083333333334,
+      "learning_rate": 9.412687827418356e-05,
+      "loss": 7.0377,
+      "loss/crossentropy": 1.3508528769016266,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1732781808823347,
+      "step": 936
+    },
+    {
+      "epoch": 0.15616666666666668,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.005989583333333,
+      "learning_rate": 9.411456132174767e-05,
+      "loss": 7.1604,
+      "loss/crossentropy": 2.340648829936981,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19552252441644669,
+      "step": 937
+    },
+    {
+      "epoch": 0.15633333333333332,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.10625,
+      "learning_rate": 9.410223227505459e-05,
+      "loss": 7.428,
+      "loss/crossentropy": 1.7425740659236908,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22830413281917572,
+      "step": 938
+    },
+    {
+      "epoch": 0.1565,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.122330729166667,
+      "learning_rate": 9.408989113748442e-05,
+      "loss": 7.4025,
+      "loss/crossentropy": 0.9290481209754944,
+      "loss/hidden": 4.0234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24749279394745827,
+      "step": 939
+    },
+    {
+      "epoch": 0.15666666666666668,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.6830729166666667,
+      "learning_rate": 9.407753791242051e-05,
+      "loss": 7.016,
+      "loss/crossentropy": 1.5664034932851791,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18499549850821495,
+      "step": 940
+    },
+    {
+      "epoch": 0.15683333333333332,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.124739583333333,
+      "learning_rate": 9.40651726032496e-05,
+      "loss": 7.0839,
+      "loss/crossentropy": 1.66704261302948,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15664922818541527,
+      "step": 941
+    },
+    {
+      "epoch": 0.157,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.7895182291666667,
+      "learning_rate": 9.405279521336173e-05,
+      "loss": 7.4609,
+      "loss/crossentropy": 1.534089207649231,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.172721229493618,
+      "step": 942
+    },
+    {
+      "epoch": 0.15716666666666668,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.5379557291666666,
+      "learning_rate": 9.404040574615018e-05,
+      "loss": 7.5585,
+      "loss/crossentropy": 2.106447756290436,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18444381654262543,
+      "step": 943
+    },
+    {
+      "epoch": 0.15733333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.44765625,
+      "learning_rate": 9.402800420501164e-05,
+      "loss": 7.4477,
+      "loss/crossentropy": 1.6428053379058838,
+      "loss/hidden": 3.92578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21111125126481056,
+      "step": 944
+    },
+    {
+      "epoch": 0.1575,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.4916015625,
+      "learning_rate": 9.401559059334601e-05,
+      "loss": 7.4241,
+      "loss/crossentropy": 1.8717712312936783,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3211796395480633,
+      "step": 945
+    },
+    {
+      "epoch": 0.15766666666666668,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.3947265625,
+      "learning_rate": 9.400316491455661e-05,
+      "loss": 7.1774,
+      "loss/crossentropy": 2.349918931722641,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20182542502880096,
+      "step": 946
+    },
+    {
+      "epoch": 0.15783333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.3791666666666667,
+      "learning_rate": 9.399072717204995e-05,
+      "loss": 7.1154,
+      "loss/crossentropy": 1.8603190779685974,
+      "loss/hidden": 3.7421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20828839763998985,
+      "step": 947
+    },
+    {
+      "epoch": 0.158,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.4244140625,
+      "learning_rate": 9.397827736923596e-05,
+      "loss": 7.3174,
+      "loss/crossentropy": 1.6743919849395752,
+      "loss/hidden": 3.7109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37463751435279846,
+      "step": 948
+    },
+    {
+      "epoch": 0.15816666666666668,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.40390625,
+      "learning_rate": 9.396581550952781e-05,
+      "loss": 7.0963,
+      "loss/crossentropy": 2.0914266109466553,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21582521870732307,
+      "step": 949
+    },
+    {
+      "epoch": 0.15833333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.47265625,
+      "learning_rate": 9.395334159634199e-05,
+      "loss": 7.1344,
+      "loss/crossentropy": 1.8080083131790161,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18114790692925453,
+      "step": 950
+    },
+    {
+      "epoch": 0.1585,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.6061848958333333,
+      "learning_rate": 9.394085563309827e-05,
+      "loss": 7.4973,
+      "loss/crossentropy": 2.214223086833954,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22889979928731918,
+      "step": 951
+    },
+    {
+      "epoch": 0.15866666666666668,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.69765625,
+      "learning_rate": 9.392835762321977e-05,
+      "loss": 7.0854,
+      "loss/crossentropy": 1.8400238454341888,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.248918104916811,
+      "step": 952
+    },
+    {
+      "epoch": 0.15883333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.9666015625,
+      "learning_rate": 9.391584757013289e-05,
+      "loss": 7.5425,
+      "loss/crossentropy": 1.8805403411388397,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17116229981184006,
+      "step": 953
+    },
+    {
+      "epoch": 0.159,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.8587890625,
+      "learning_rate": 9.390332547726733e-05,
+      "loss": 7.1284,
+      "loss/crossentropy": 1.739779755473137,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20969608798623085,
+      "step": 954
+    },
+    {
+      "epoch": 0.15916666666666668,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.99140625,
+      "learning_rate": 9.389079134805609e-05,
+      "loss": 7.1018,
+      "loss/crossentropy": 1.8846152275800705,
+      "loss/hidden": 3.7578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23489586636424065,
+      "step": 955
+    },
+    {
+      "epoch": 0.15933333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.4634765625,
+      "learning_rate": 9.387824518593546e-05,
+      "loss": 6.9657,
+      "loss/crossentropy": 1.6097615212202072,
+      "loss/hidden": 3.734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18240181729197502,
+      "step": 956
+    },
+    {
+      "epoch": 0.1595,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.459375,
+      "learning_rate": 9.386568699434508e-05,
+      "loss": 7.1398,
+      "loss/crossentropy": 2.1050661504268646,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19370339810848236,
+      "step": 957
+    },
+    {
+      "epoch": 0.15966666666666668,
+      "grad_norm": 48.0,
+      "grad_norm_var": 20.103580729166666,
+      "learning_rate": 9.385311677672781e-05,
+      "loss": 7.6893,
+      "loss/crossentropy": 2.1763148307800293,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17503639683127403,
+      "step": 958
+    },
+    {
+      "epoch": 0.15983333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 20.450455729166666,
+      "learning_rate": 9.384053453652986e-05,
+      "loss": 7.4834,
+      "loss/crossentropy": 1.2227577418088913,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12875572219491005,
+      "step": 959
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 30.5,
+      "grad_norm_var": 20.545833333333334,
+      "learning_rate": 9.382794027720073e-05,
+      "loss": 7.1467,
+      "loss/crossentropy": 1.5717070400714874,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16495000198483467,
+      "step": 960
+    },
+    {
+      "epoch": 0.16016666666666668,
+      "grad_norm": 30.5,
+      "grad_norm_var": 20.659309895833335,
+      "learning_rate": 9.381533400219318e-05,
+      "loss": 7.0907,
+      "loss/crossentropy": 1.8306347131729126,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1771262213587761,
+      "step": 961
+    },
+    {
+      "epoch": 0.16033333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 20.386458333333334,
+      "learning_rate": 9.380271571496334e-05,
+      "loss": 7.4557,
+      "loss/crossentropy": 2.4782089591026306,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.213636826723814,
+      "step": 962
+    },
+    {
+      "epoch": 0.1605,
+      "grad_norm": 30.125,
+      "grad_norm_var": 20.205143229166666,
+      "learning_rate": 9.379008541897054e-05,
+      "loss": 7.2899,
+      "loss/crossentropy": 1.599755346775055,
+      "loss/hidden": 3.765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18493517115712166,
+      "step": 963
+    },
+    {
+      "epoch": 0.16066666666666668,
+      "grad_norm": 32.75,
+      "grad_norm_var": 20.194205729166665,
+      "learning_rate": 9.377744311767746e-05,
+      "loss": 7.5192,
+      "loss/crossentropy": 1.746928870677948,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17140696942806244,
+      "step": 964
+    },
+    {
+      "epoch": 0.16083333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 19.730143229166668,
+      "learning_rate": 9.376478881455009e-05,
+      "loss": 7.339,
+      "loss/crossentropy": 2.0143012702465057,
+      "loss/hidden": 4.0078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2578558959066868,
+      "step": 965
+    },
+    {
+      "epoch": 0.161,
+      "grad_norm": 30.25,
+      "grad_norm_var": 19.287434895833332,
+      "learning_rate": 9.375212251305763e-05,
+      "loss": 7.1615,
+      "loss/crossentropy": 1.7171165198087692,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14619379863142967,
+      "step": 966
+    },
+    {
+      "epoch": 0.16116666666666668,
+      "grad_norm": 32.25,
+      "grad_norm_var": 19.273958333333333,
+      "learning_rate": 9.373944421667265e-05,
+      "loss": 7.2861,
+      "loss/crossentropy": 1.4313836991786957,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15895283594727516,
+      "step": 967
+    },
+    {
+      "epoch": 0.16133333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 19.533333333333335,
+      "learning_rate": 9.372675392887096e-05,
+      "loss": 7.0251,
+      "loss/crossentropy": 1.4792975187301636,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20261310786008835,
+      "step": 968
+    },
+    {
+      "epoch": 0.1615,
+      "grad_norm": 29.0,
+      "grad_norm_var": 20.162239583333335,
+      "learning_rate": 9.371405165313169e-05,
+      "loss": 7.0704,
+      "loss/crossentropy": 1.8925665318965912,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16673966497182846,
+      "step": 969
+    },
+    {
+      "epoch": 0.16166666666666665,
+      "grad_norm": 30.0,
+      "grad_norm_var": 20.3875,
+      "learning_rate": 9.370133739293723e-05,
+      "loss": 7.5313,
+      "loss/crossentropy": 2.265043467283249,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18483617156744003,
+      "step": 970
+    },
+    {
+      "epoch": 0.16183333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 19.6697265625,
+      "learning_rate": 9.368861115177327e-05,
+      "loss": 7.63,
+      "loss/crossentropy": 1.73798006772995,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2236182652413845,
+      "step": 971
+    },
+    {
+      "epoch": 0.162,
+      "grad_norm": 31.5,
+      "grad_norm_var": 19.2625,
+      "learning_rate": 9.367587293312878e-05,
+      "loss": 7.5108,
+      "loss/crossentropy": 2.109057456254959,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1911468170583248,
+      "step": 972
+    },
+    {
+      "epoch": 0.16216666666666665,
+      "grad_norm": 29.75,
+      "grad_norm_var": 19.6509765625,
+      "learning_rate": 9.366312274049602e-05,
+      "loss": 7.2784,
+      "loss/crossentropy": 1.9426101446151733,
+      "loss/hidden": 3.76171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21822215616703033,
+      "step": 973
+    },
+    {
+      "epoch": 0.16233333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.8009765625,
+      "learning_rate": 9.36503605773705e-05,
+      "loss": 7.7166,
+      "loss/crossentropy": 1.858081966638565,
+      "loss/hidden": 3.7421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23198555782437325,
+      "step": 974
+    },
+    {
+      "epoch": 0.1625,
+      "grad_norm": 53.0,
+      "grad_norm_var": 31.5666015625,
+      "learning_rate": 9.36375864472511e-05,
+      "loss": 7.848,
+      "loss/crossentropy": 2.2449205219745636,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.175953671336174,
+      "step": 975
+    },
+    {
+      "epoch": 0.16266666666666665,
+      "grad_norm": 33.25,
+      "grad_norm_var": 31.3775390625,
+      "learning_rate": 9.362480035363986e-05,
+      "loss": 7.1444,
+      "loss/crossentropy": 1.4836579263210297,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17005424201488495,
+      "step": 976
+    },
+    {
+      "epoch": 0.16283333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 31.1494140625,
+      "learning_rate": 9.36120023000422e-05,
+      "loss": 7.1445,
+      "loss/crossentropy": 1.46025812625885,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1525864601135254,
+      "step": 977
+    },
+    {
+      "epoch": 0.163,
+      "grad_norm": 32.25,
+      "grad_norm_var": 30.99765625,
+      "learning_rate": 9.359919228996674e-05,
+      "loss": 7.0693,
+      "loss/crossentropy": 1.981631189584732,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19974175468087196,
+      "step": 978
+    },
+    {
+      "epoch": 0.16316666666666665,
+      "grad_norm": 30.375,
+      "grad_norm_var": 30.914583333333333,
+      "learning_rate": 9.358637032692545e-05,
+      "loss": 7.5493,
+      "loss/crossentropy": 1.9146924018859863,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23189393803477287,
+      "step": 979
+    },
+    {
+      "epoch": 0.16333333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 31.055208333333333,
+      "learning_rate": 9.357353641443354e-05,
+      "loss": 7.2102,
+      "loss/crossentropy": 1.9897369742393494,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19221243634819984,
+      "step": 980
+    },
+    {
+      "epoch": 0.1635,
+      "grad_norm": 32.25,
+      "grad_norm_var": 31.010416666666668,
+      "learning_rate": 9.356069055600948e-05,
+      "loss": 7.0991,
+      "loss/crossentropy": 2.268047869205475,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17382950708270073,
+      "step": 981
+    },
+    {
+      "epoch": 0.16366666666666665,
+      "grad_norm": 31.0,
+      "grad_norm_var": 30.801822916666666,
+      "learning_rate": 9.354783275517504e-05,
+      "loss": 7.2934,
+      "loss/crossentropy": 1.6492570638656616,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.270766519010067,
+      "step": 982
+    },
+    {
+      "epoch": 0.16383333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 30.906184895833334,
+      "learning_rate": 9.353496301545529e-05,
+      "loss": 7.1503,
+      "loss/crossentropy": 1.856852948665619,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18430958315730095,
+      "step": 983
+    },
+    {
+      "epoch": 0.164,
+      "grad_norm": 31.375,
+      "grad_norm_var": 30.620833333333334,
+      "learning_rate": 9.352208134037851e-05,
+      "loss": 7.184,
+      "loss/crossentropy": 1.7651639580726624,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15293405205011368,
+      "step": 984
+    },
+    {
+      "epoch": 0.16416666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 30.386458333333334,
+      "learning_rate": 9.35091877334763e-05,
+      "loss": 7.3302,
+      "loss/crossentropy": 1.7698874771595001,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19256792590022087,
+      "step": 985
+    },
+    {
+      "epoch": 0.16433333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 30.7587890625,
+      "learning_rate": 9.349628219828349e-05,
+      "loss": 7.3029,
+      "loss/crossentropy": 2.011773020029068,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17887185513973236,
+      "step": 986
+    },
+    {
+      "epoch": 0.1645,
+      "grad_norm": 30.625,
+      "grad_norm_var": 30.998372395833332,
+      "learning_rate": 9.348336473833823e-05,
+      "loss": 7.1779,
+      "loss/crossentropy": 2.06114062666893,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18867448344826698,
+      "step": 987
+    },
+    {
+      "epoch": 0.16466666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 30.949739583333333,
+      "learning_rate": 9.347043535718192e-05,
+      "loss": 7.1645,
+      "loss/crossentropy": 1.3193330019712448,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14304210245609283,
+      "step": 988
+    },
+    {
+      "epoch": 0.16483333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 30.4041015625,
+      "learning_rate": 9.34574940583592e-05,
+      "loss": 7.0902,
+      "loss/crossentropy": 1.2350816279649734,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1587892509996891,
+      "step": 989
+    },
+    {
+      "epoch": 0.165,
+      "grad_norm": 30.875,
+      "grad_norm_var": 30.623958333333334,
+      "learning_rate": 9.344454084541803e-05,
+      "loss": 7.1701,
+      "loss/crossentropy": 1.8795148134231567,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1856541410088539,
+      "step": 990
+    },
+    {
+      "epoch": 0.16516666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.3747395833333333,
+      "learning_rate": 9.343157572190957e-05,
+      "loss": 7.4424,
+      "loss/crossentropy": 1.6773568987846375,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2251323200762272,
+      "step": 991
+    },
+    {
+      "epoch": 0.16533333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.1541666666666666,
+      "learning_rate": 9.341859869138831e-05,
+      "loss": 7.4683,
+      "loss/crossentropy": 1.85666224360466,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1994481161236763,
+      "step": 992
+    },
+    {
+      "epoch": 0.1655,
+      "grad_norm": 30.125,
+      "grad_norm_var": 0.8254557291666667,
+      "learning_rate": 9.340560975741197e-05,
+      "loss": 7.1673,
+      "loss/crossentropy": 1.798208549618721,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2776002138853073,
+      "step": 993
+    },
+    {
+      "epoch": 0.16566666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 0.8311848958333333,
+      "learning_rate": 9.339260892354153e-05,
+      "loss": 7.2964,
+      "loss/crossentropy": 2.394089937210083,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17311453633010387,
+      "step": 994
+    },
+    {
+      "epoch": 0.16583333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 0.8393229166666667,
+      "learning_rate": 9.337959619334125e-05,
+      "loss": 7.2382,
+      "loss/crossentropy": 1.8724756240844727,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15168115496635437,
+      "step": 995
+    },
+    {
+      "epoch": 0.166,
+      "grad_norm": 29.75,
+      "grad_norm_var": 0.8893229166666666,
+      "learning_rate": 9.336657157037866e-05,
+      "loss": 7.221,
+      "loss/crossentropy": 2.161208003759384,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17218947410583496,
+      "step": 996
+    },
+    {
+      "epoch": 0.16616666666666666,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.1020182291666667,
+      "learning_rate": 9.33535350582245e-05,
+      "loss": 6.94,
+      "loss/crossentropy": 1.6933994144201279,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16723676770925522,
+      "step": 997
+    },
+    {
+      "epoch": 0.16633333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.1639973958333334,
+      "learning_rate": 9.334048666045285e-05,
+      "loss": 7.2939,
+      "loss/crossentropy": 2.274313300848007,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17677492275834084,
+      "step": 998
+    },
+    {
+      "epoch": 0.1665,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.3108723958333333,
+      "learning_rate": 9.332742638064094e-05,
+      "loss": 7.1129,
+      "loss/crossentropy": 1.77187579870224,
+      "loss/hidden": 3.79296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27776718884706497,
+      "step": 999
+    },
+    {
+      "epoch": 0.16666666666666666,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.2455729166666667,
+      "learning_rate": 9.331435422236938e-05,
+      "loss": 7.2087,
+      "loss/crossentropy": 1.9265810996294022,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15580631606280804,
+      "step": 1000
+    },
+    {
+      "epoch": 0.16683333333333333,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.3768229166666666,
+      "learning_rate": 9.330127018922194e-05,
+      "loss": 7.2168,
+      "loss/crossentropy": 1.52646504342556,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2279236763715744,
+      "step": 1001
+    },
+    {
+      "epoch": 0.167,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.4244140625,
+      "learning_rate": 9.328817428478569e-05,
+      "loss": 7.3864,
+      "loss/crossentropy": 1.8362654149532318,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2490648813545704,
+      "step": 1002
+    },
+    {
+      "epoch": 0.16716666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.70625,
+      "learning_rate": 9.327506651265095e-05,
+      "loss": 7.3141,
+      "loss/crossentropy": 1.78467258810997,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1509244404733181,
+      "step": 1003
+    },
+    {
+      "epoch": 0.16733333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.70625,
+      "learning_rate": 9.32619468764113e-05,
+      "loss": 7.2591,
+      "loss/crossentropy": 2.2533966302871704,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19303973764181137,
+      "step": 1004
+    },
+    {
+      "epoch": 0.1675,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.6587890625,
+      "learning_rate": 9.324881537966354e-05,
+      "loss": 6.9683,
+      "loss/crossentropy": 1.7321947365999222,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16967357695102692,
+      "step": 1005
+    },
+    {
+      "epoch": 0.16766666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.7580729166666667,
+      "learning_rate": 9.323567202600776e-05,
+      "loss": 7.177,
+      "loss/crossentropy": 1.7321678400039673,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.202758327126503,
+      "step": 1006
+    },
+    {
+      "epoch": 0.16783333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.6520182291666667,
+      "learning_rate": 9.322251681904728e-05,
+      "loss": 7.3569,
+      "loss/crossentropy": 1.8233984112739563,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2399062104523182,
+      "step": 1007
+    },
+    {
+      "epoch": 0.168,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.72265625,
+      "learning_rate": 9.320934976238867e-05,
+      "loss": 7.2127,
+      "loss/crossentropy": 1.373336136341095,
+      "loss/hidden": 4.1015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25906676054000854,
+      "step": 1008
+    },
+    {
+      "epoch": 0.16816666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.7353515625,
+      "learning_rate": 9.319617085964176e-05,
+      "loss": 6.9572,
+      "loss/crossentropy": 1.6063978597521782,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15655718185007572,
+      "step": 1009
+    },
+    {
+      "epoch": 0.16833333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.7869140625,
+      "learning_rate": 9.318298011441964e-05,
+      "loss": 7.1022,
+      "loss/crossentropy": 1.7690593600273132,
+      "loss/hidden": 3.921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20905645936727524,
+      "step": 1010
+    },
+    {
+      "epoch": 0.1685,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.0931640625,
+      "learning_rate": 9.316977753033859e-05,
+      "loss": 7.5942,
+      "loss/crossentropy": 1.4739122241735458,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18209786340594292,
+      "step": 1011
+    },
+    {
+      "epoch": 0.16866666666666666,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.2327473958333335,
+      "learning_rate": 9.31565631110182e-05,
+      "loss": 7.4409,
+      "loss/crossentropy": 1.5012520253658295,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2104843109846115,
+      "step": 1012
+    },
+    {
+      "epoch": 0.16883333333333334,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.9192057291666667,
+      "learning_rate": 9.314333686008125e-05,
+      "loss": 7.2449,
+      "loss/crossentropy": 1.447302222251892,
+      "loss/hidden": 3.6484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19817490130662918,
+      "step": 1013
+    },
+    {
+      "epoch": 0.169,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.9192057291666667,
+      "learning_rate": 9.313009878115381e-05,
+      "loss": 7.304,
+      "loss/crossentropy": 1.575433075428009,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15146752633154392,
+      "step": 1014
+    },
+    {
+      "epoch": 0.16916666666666666,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.8268229166666667,
+      "learning_rate": 9.31168488778652e-05,
+      "loss": 7.3226,
+      "loss/crossentropy": 1.2666431218385696,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17925303801894188,
+      "step": 1015
+    },
+    {
+      "epoch": 0.16933333333333334,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.8473307291666667,
+      "learning_rate": 9.310358715384793e-05,
+      "loss": 7.4087,
+      "loss/crossentropy": 1.6595781594514847,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17667768523097038,
+      "step": 1016
+    },
+    {
+      "epoch": 0.1695,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.7197265625,
+      "learning_rate": 9.309031361273775e-05,
+      "loss": 7.6083,
+      "loss/crossentropy": 1.8028043806552887,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20026536658406258,
+      "step": 1017
+    },
+    {
+      "epoch": 0.16966666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.9837890625,
+      "learning_rate": 9.307702825817373e-05,
+      "loss": 7.3404,
+      "loss/crossentropy": 2.270669639110565,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16769889742136002,
+      "step": 1018
+    },
+    {
+      "epoch": 0.16983333333333334,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.959375,
+      "learning_rate": 9.306373109379809e-05,
+      "loss": 7.0424,
+      "loss/crossentropy": 1.4852006286382675,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1823214292526245,
+      "step": 1019
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.8332682291666667,
+      "learning_rate": 9.305042212325634e-05,
+      "loss": 7.1053,
+      "loss/crossentropy": 2.016376793384552,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17145168781280518,
+      "step": 1020
+    },
+    {
+      "epoch": 0.17016666666666666,
+      "grad_norm": 61.5,
+      "grad_norm_var": 61.40514322916667,
+      "learning_rate": 9.30371013501972e-05,
+      "loss": 7.454,
+      "loss/crossentropy": 1.6555127948522568,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22747880220413208,
+      "step": 1021
+    },
+    {
+      "epoch": 0.17033333333333334,
+      "grad_norm": 42.25,
+      "grad_norm_var": 67.2056640625,
+      "learning_rate": 9.302376877827263e-05,
+      "loss": 7.209,
+      "loss/crossentropy": 1.6700544655323029,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1897926852107048,
+      "step": 1022
+    },
+    {
+      "epoch": 0.1705,
+      "grad_norm": 35.5,
+      "grad_norm_var": 66.83333333333333,
+      "learning_rate": 9.301042441113783e-05,
+      "loss": 7.0414,
+      "loss/crossentropy": 1.493530809879303,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1704997718334198,
+      "step": 1023
+    },
+    {
+      "epoch": 0.17066666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 66.55618489583334,
+      "learning_rate": 9.299706825245126e-05,
+      "loss": 7.5369,
+      "loss/crossentropy": 1.8359811902046204,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18067612871527672,
+      "step": 1024
+    },
+    {
+      "epoch": 0.17083333333333334,
+      "grad_norm": 30.125,
+      "grad_norm_var": 66.24140625,
+      "learning_rate": 9.298370030587456e-05,
+      "loss": 7.3398,
+      "loss/crossentropy": 1.424869641661644,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17005274444818497,
+      "step": 1025
+    },
+    {
+      "epoch": 0.171,
+      "grad_norm": 30.375,
+      "grad_norm_var": 66.48326822916667,
+      "learning_rate": 9.297032057507264e-05,
+      "loss": 6.9411,
+      "loss/crossentropy": 1.884153574705124,
+      "loss/hidden": 3.01953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14143190905451775,
+      "step": 1026
+    },
+    {
+      "epoch": 0.17116666666666666,
+      "grad_norm": 35.0,
+      "grad_norm_var": 66.47180989583333,
+      "learning_rate": 9.295692906371363e-05,
+      "loss": 7.1936,
+      "loss/crossentropy": 1.7658977210521698,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18475309200584888,
+      "step": 1027
+    },
+    {
+      "epoch": 0.17133333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 67.45520833333333,
+      "learning_rate": 9.294352577546888e-05,
+      "loss": 7.4849,
+      "loss/crossentropy": 1.519232600927353,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14876055624336004,
+      "step": 1028
+    },
+    {
+      "epoch": 0.1715,
+      "grad_norm": 30.75,
+      "grad_norm_var": 67.06920572916667,
+      "learning_rate": 9.293011071401298e-05,
+      "loss": 7.4399,
+      "loss/crossentropy": 1.4621777907013893,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13985679671168327,
+      "step": 1029
+    },
+    {
+      "epoch": 0.17166666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 66.48854166666666,
+      "learning_rate": 9.291668388302374e-05,
+      "loss": 7.5762,
+      "loss/crossentropy": 2.215066432952881,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18505752459168434,
+      "step": 1030
+    },
+    {
+      "epoch": 0.17183333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 64.88307291666666,
+      "learning_rate": 9.290324528618224e-05,
+      "loss": 7.0036,
+      "loss/crossentropy": 2.04854553937912,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19184922613203526,
+      "step": 1031
+    },
+    {
+      "epoch": 0.172,
+      "grad_norm": 31.5,
+      "grad_norm_var": 63.706705729166664,
+      "learning_rate": 9.28897949271727e-05,
+      "loss": 7.403,
+      "loss/crossentropy": 1.6185847520828247,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22312923520803452,
+      "step": 1032
+    },
+    {
+      "epoch": 0.17216666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 64.10104166666666,
+      "learning_rate": 9.287633280968261e-05,
+      "loss": 7.8368,
+      "loss/crossentropy": 1.7106279581785202,
+      "loss/hidden": 3.71875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19282570481300354,
+      "step": 1033
+    },
+    {
+      "epoch": 0.17233333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 65.24524739583333,
+      "learning_rate": 9.286285893740274e-05,
+      "loss": 6.8309,
+      "loss/crossentropy": 1.6567908227443695,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13862809166312218,
+      "step": 1034
+    },
+    {
+      "epoch": 0.1725,
+      "grad_norm": 30.375,
+      "grad_norm_var": 64.55774739583333,
+      "learning_rate": 9.284937331402697e-05,
+      "loss": 7.1656,
+      "loss/crossentropy": 1.8226232081651688,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17071571573615074,
+      "step": 1035
+    },
+    {
+      "epoch": 0.17266666666666666,
+      "grad_norm": 35.0,
+      "grad_norm_var": 63.64733072916667,
+      "learning_rate": 9.283587594325249e-05,
+      "loss": 7.0402,
+      "loss/crossentropy": 1.1709645837545395,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16143991611897945,
+      "step": 1036
+    },
+    {
+      "epoch": 0.17283333333333334,
+      "grad_norm": 30.375,
+      "grad_norm_var": 10.945572916666666,
+      "learning_rate": 9.282236682877967e-05,
+      "loss": 7.8393,
+      "loss/crossentropy": 2.0621816515922546,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19756818562746048,
+      "step": 1037
+    },
+    {
+      "epoch": 0.173,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.8666666666666667,
+      "learning_rate": 9.280884597431212e-05,
+      "loss": 7.5471,
+      "loss/crossentropy": 1.7954013347625732,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16884653270244598,
+      "step": 1038
+    },
+    {
+      "epoch": 0.17316666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.84765625,
+      "learning_rate": 9.279531338355666e-05,
+      "loss": 7.3084,
+      "loss/crossentropy": 2.160328984260559,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16826023906469345,
+      "step": 1039
+    },
+    {
+      "epoch": 0.17333333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.7988932291666666,
+      "learning_rate": 9.27817690602233e-05,
+      "loss": 7.5224,
+      "loss/crossentropy": 2.087209850549698,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19153646752238274,
+      "step": 1040
+    },
+    {
+      "epoch": 0.1735,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.958268229166667,
+      "learning_rate": 9.276821300802534e-05,
+      "loss": 7.2352,
+      "loss/crossentropy": 1.9709742963314056,
+      "loss/hidden": 3.12109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20601647347211838,
+      "step": 1041
+    },
+    {
+      "epoch": 0.17366666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.11640625,
+      "learning_rate": 9.27546452306792e-05,
+      "loss": 7.496,
+      "loss/crossentropy": 2.1215765178203583,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26274339854717255,
+      "step": 1042
+    },
+    {
+      "epoch": 0.17383333333333334,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.1744140625,
+      "learning_rate": 9.274106573190459e-05,
+      "loss": 6.9722,
+      "loss/crossentropy": 1.3983889818191528,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13132820650935173,
+      "step": 1043
+    },
+    {
+      "epoch": 0.174,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.1254557291666667,
+      "learning_rate": 9.272747451542441e-05,
+      "loss": 7.1143,
+      "loss/crossentropy": 2.529780864715576,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17341240867972374,
+      "step": 1044
+    },
+    {
+      "epoch": 0.17416666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.4018229166666667,
+      "learning_rate": 9.271387158496476e-05,
+      "loss": 7.3285,
+      "loss/crossentropy": 1.8412316143512726,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18098780512809753,
+      "step": 1045
+    },
+    {
+      "epoch": 0.17433333333333334,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.421875,
+      "learning_rate": 9.270025694425497e-05,
+      "loss": 7.7823,
+      "loss/crossentropy": 1.6852163225412369,
+      "loss/hidden": 3.80078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18295212276279926,
+      "step": 1046
+    },
+    {
+      "epoch": 0.1745,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.3643229166666666,
+      "learning_rate": 9.268663059702753e-05,
+      "loss": 6.9429,
+      "loss/crossentropy": 1.4733633249998093,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16742233000695705,
+      "step": 1047
+    },
+    {
+      "epoch": 0.17466666666666666,
+      "grad_norm": 42.25,
+      "grad_norm_var": 10.81875,
+      "learning_rate": 9.267299254701824e-05,
+      "loss": 7.1808,
+      "loss/crossentropy": 1.76565283536911,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15179136954247952,
+      "step": 1048
+    },
+    {
+      "epoch": 0.17483333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 10.812434895833333,
+      "learning_rate": 9.265934279796602e-05,
+      "loss": 7.3945,
+      "loss/crossentropy": 2.493871033191681,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17638367414474487,
+      "step": 1049
+    },
+    {
+      "epoch": 0.175,
+      "grad_norm": 29.25,
+      "grad_norm_var": 10.905989583333334,
+      "learning_rate": 9.264568135361302e-05,
+      "loss": 7.0946,
+      "loss/crossentropy": 1.0718126818537712,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17817819118499756,
+      "step": 1050
+    },
+    {
+      "epoch": 0.17516666666666666,
+      "grad_norm": 37.5,
+      "grad_norm_var": 13.203059895833333,
+      "learning_rate": 9.263200821770461e-05,
+      "loss": 7.144,
+      "loss/crossentropy": 1.7253368496894836,
+      "loss/hidden": 3.73046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22389013692736626,
+      "step": 1051
+    },
+    {
+      "epoch": 0.17533333333333334,
+      "grad_norm": 27.5,
+      "grad_norm_var": 13.460872395833333,
+      "learning_rate": 9.261832339398938e-05,
+      "loss": 7.2942,
+      "loss/crossentropy": 1.919466644525528,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20888322964310646,
+      "step": 1052
+    },
+    {
+      "epoch": 0.1755,
+      "grad_norm": 28.625,
+      "grad_norm_var": 13.8619140625,
+      "learning_rate": 9.260462688621905e-05,
+      "loss": 7.2526,
+      "loss/crossentropy": 2.213284432888031,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19714241474866867,
+      "step": 1053
+    },
+    {
+      "epoch": 0.17566666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 13.826822916666666,
+      "learning_rate": 9.259091869814864e-05,
+      "loss": 7.636,
+      "loss/crossentropy": 1.81241112947464,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28831223770976067,
+      "step": 1054
+    },
+    {
+      "epoch": 0.17583333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 13.979166666666666,
+      "learning_rate": 9.257719883353631e-05,
+      "loss": 7.5489,
+      "loss/crossentropy": 1.992142990231514,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2022876925766468,
+      "step": 1055
+    },
+    {
+      "epoch": 0.176,
+      "grad_norm": 30.0,
+      "grad_norm_var": 14.051497395833334,
+      "learning_rate": 9.256346729614342e-05,
+      "loss": 7.1309,
+      "loss/crossentropy": 1.680229663848877,
+      "loss/hidden": 3.97265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18140022456645966,
+      "step": 1056
+    },
+    {
+      "epoch": 0.17616666666666667,
+      "grad_norm": 28.625,
+      "grad_norm_var": 14.253059895833333,
+      "learning_rate": 9.254972408973461e-05,
+      "loss": 7.122,
+      "loss/crossentropy": 1.4771035015583038,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20891831442713737,
+      "step": 1057
+    },
+    {
+      "epoch": 0.17633333333333334,
+      "grad_norm": 30.375,
+      "grad_norm_var": 14.126822916666667,
+      "learning_rate": 9.253596921807759e-05,
+      "loss": 6.7827,
+      "loss/crossentropy": 1.6926728934049606,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1915021948516369,
+      "step": 1058
+    },
+    {
+      "epoch": 0.1765,
+      "grad_norm": 32.75,
+      "grad_norm_var": 14.234830729166667,
+      "learning_rate": 9.252220268494337e-05,
+      "loss": 7.4013,
+      "loss/crossentropy": 1.3703967481851578,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15270237810909748,
+      "step": 1059
+    },
+    {
+      "epoch": 0.17666666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 14.322916666666666,
+      "learning_rate": 9.250842449410611e-05,
+      "loss": 7.0274,
+      "loss/crossentropy": 2.548193573951721,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19006280973553658,
+      "step": 1060
+    },
+    {
+      "epoch": 0.17683333333333334,
+      "grad_norm": 30.375,
+      "grad_norm_var": 14.001041666666667,
+      "learning_rate": 9.249463464934321e-05,
+      "loss": 6.9488,
+      "loss/crossentropy": 1.3754972219467163,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14112510904669762,
+      "step": 1061
+    },
+    {
+      "epoch": 0.177,
+      "grad_norm": 33.5,
+      "grad_norm_var": 14.192643229166666,
+      "learning_rate": 9.248083315443518e-05,
+      "loss": 7.0641,
+      "loss/crossentropy": 1.5374733209609985,
+      "loss/hidden": 3.078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12394630908966064,
+      "step": 1062
+    },
+    {
+      "epoch": 0.17716666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 14.357747395833334,
+      "learning_rate": 9.246702001316583e-05,
+      "loss": 6.9103,
+      "loss/crossentropy": 2.03630730509758,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1795985884964466,
+      "step": 1063
+    },
+    {
+      "epoch": 0.17733333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 6.176497395833334,
+      "learning_rate": 9.245319522932209e-05,
+      "loss": 7.3536,
+      "loss/crossentropy": 2.339617282152176,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17432091012597084,
+      "step": 1064
+    },
+    {
+      "epoch": 0.1775,
+      "grad_norm": 31.25,
+      "grad_norm_var": 6.1916015625,
+      "learning_rate": 9.24393588066941e-05,
+      "loss": 7.3402,
+      "loss/crossentropy": 1.4889580011367798,
+      "loss/hidden": 3.9609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20106740668416023,
+      "step": 1065
+    },
+    {
+      "epoch": 0.17766666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 6.12890625,
+      "learning_rate": 9.242551074907519e-05,
+      "loss": 7.707,
+      "loss/crossentropy": 1.8807732462882996,
+      "loss/hidden": 3.90234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28517602011561394,
+      "step": 1066
+    },
+    {
+      "epoch": 0.17783333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.84375,
+      "learning_rate": 9.241165106026189e-05,
+      "loss": 7.6607,
+      "loss/crossentropy": 1.5485778898000717,
+      "loss/hidden": 3.859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23636030405759811,
+      "step": 1067
+    },
+    {
+      "epoch": 0.178,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.414322916666667,
+      "learning_rate": 9.239777974405393e-05,
+      "loss": 7.3144,
+      "loss/crossentropy": 1.7089111506938934,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.147063210606575,
+      "step": 1068
+    },
+    {
+      "epoch": 0.17816666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.220572916666667,
+      "learning_rate": 9.238389680425416e-05,
+      "loss": 6.937,
+      "loss/crossentropy": 1.6708162426948547,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15234573185443878,
+      "step": 1069
+    },
+    {
+      "epoch": 0.17833333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.2202473958333333,
+      "learning_rate": 9.237000224466872e-05,
+      "loss": 7.5136,
+      "loss/crossentropy": 1.8323483765125275,
+      "loss/hidden": 3.7578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26327143609523773,
+      "step": 1070
+    },
+    {
+      "epoch": 0.1785,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.8499348958333333,
+      "learning_rate": 9.235609606910687e-05,
+      "loss": 6.8242,
+      "loss/crossentropy": 1.695602685213089,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13594901375472546,
+      "step": 1071
+    },
+    {
+      "epoch": 0.17866666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.5317057291666667,
+      "learning_rate": 9.234217828138104e-05,
+      "loss": 7.1028,
+      "loss/crossentropy": 1.7536567747592926,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18050695396959782,
+      "step": 1072
+    },
+    {
+      "epoch": 0.17883333333333334,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.341666666666667,
+      "learning_rate": 9.23282488853069e-05,
+      "loss": 7.1524,
+      "loss/crossentropy": 1.9407530725002289,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1710616685450077,
+      "step": 1073
+    },
+    {
+      "epoch": 0.179,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.4291666666666667,
+      "learning_rate": 9.231430788470326e-05,
+      "loss": 7.1724,
+      "loss/crossentropy": 1.5481340885162354,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2104547992348671,
+      "step": 1074
+    },
+    {
+      "epoch": 0.17916666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.0768229166666665,
+      "learning_rate": 9.230035528339211e-05,
+      "loss": 7.3203,
+      "loss/crossentropy": 1.7858853340148926,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15118434838950634,
+      "step": 1075
+    },
+    {
+      "epoch": 0.17933333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.9955729166666667,
+      "learning_rate": 9.228639108519868e-05,
+      "loss": 7.2717,
+      "loss/crossentropy": 1.5924931764602661,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2723642699420452,
+      "step": 1076
+    },
+    {
+      "epoch": 0.1795,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.700455729166667,
+      "learning_rate": 9.227241529395127e-05,
+      "loss": 7.5044,
+      "loss/crossentropy": 1.6551357507705688,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2343754842877388,
+      "step": 1077
+    },
+    {
+      "epoch": 0.17966666666666667,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.2059895833333334,
+      "learning_rate": 9.225842791348149e-05,
+      "loss": 6.7824,
+      "loss/crossentropy": 1.294097363948822,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17944151908159256,
+      "step": 1078
+    },
+    {
+      "epoch": 0.17983333333333335,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.040625,
+      "learning_rate": 9.224442894762401e-05,
+      "loss": 7.2718,
+      "loss/crossentropy": 1.9110237658023834,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22192451357841492,
+      "step": 1079
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.46640625,
+      "learning_rate": 9.223041840021674e-05,
+      "loss": 7.1671,
+      "loss/crossentropy": 1.4245515018701553,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17833401262760162,
+      "step": 1080
+    },
+    {
+      "epoch": 0.18016666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.515625,
+      "learning_rate": 9.221639627510076e-05,
+      "loss": 7.2023,
+      "loss/crossentropy": 2.0501048266887665,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20859062112867832,
+      "step": 1081
+    },
+    {
+      "epoch": 0.18033333333333335,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.5317057291666667,
+      "learning_rate": 9.220236257612031e-05,
+      "loss": 6.7608,
+      "loss/crossentropy": 1.6907588839530945,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16623044945299625,
+      "step": 1082
+    },
+    {
+      "epoch": 0.1805,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.526497395833333,
+      "learning_rate": 9.21883173071228e-05,
+      "loss": 7.3174,
+      "loss/crossentropy": 1.2606586515903473,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1489721965044737,
+      "step": 1083
+    },
+    {
+      "epoch": 0.18066666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.440559895833333,
+      "learning_rate": 9.217426047195882e-05,
+      "loss": 7.6918,
+      "loss/crossentropy": 2.292267620563507,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22536378726363182,
+      "step": 1084
+    },
+    {
+      "epoch": 0.18083333333333335,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.417122395833333,
+      "learning_rate": 9.216019207448217e-05,
+      "loss": 7.0151,
+      "loss/crossentropy": 1.3078829050064087,
+      "loss/hidden": 4.015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1706789880990982,
+      "step": 1085
+    },
+    {
+      "epoch": 0.181,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.4921223958333334,
+      "learning_rate": 9.214611211854974e-05,
+      "loss": 7.3649,
+      "loss/crossentropy": 2.1642357409000397,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22923363372683525,
+      "step": 1086
+    },
+    {
+      "epoch": 0.18116666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.4872395833333334,
+      "learning_rate": 9.213202060802161e-05,
+      "loss": 7.0883,
+      "loss/crossentropy": 1.0673165619373322,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16823571175336838,
+      "step": 1087
+    },
+    {
+      "epoch": 0.18133333333333335,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.033333333333333,
+      "learning_rate": 9.21179175467611e-05,
+      "loss": 7.2322,
+      "loss/crossentropy": 1.8726580739021301,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15625272691249847,
+      "step": 1088
+    },
+    {
+      "epoch": 0.1815,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.236458333333333,
+      "learning_rate": 9.210380293863462e-05,
+      "loss": 6.9307,
+      "loss/crossentropy": 1.1443600505590439,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20036842301487923,
+      "step": 1089
+    },
+    {
+      "epoch": 0.18166666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.256705729166667,
+      "learning_rate": 9.208967678751177e-05,
+      "loss": 7.1227,
+      "loss/crossentropy": 1.539658635854721,
+      "loss/hidden": 3.7890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2563717756420374,
+      "step": 1090
+    },
+    {
+      "epoch": 0.18183333333333335,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.5374348958333335,
+      "learning_rate": 9.207553909726531e-05,
+      "loss": 7.55,
+      "loss/crossentropy": 1.9002563059329987,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25357599928975105,
+      "step": 1091
+    },
+    {
+      "epoch": 0.182,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.299934895833333,
+      "learning_rate": 9.206138987177118e-05,
+      "loss": 7.6483,
+      "loss/crossentropy": 1.6798991560935974,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2713388651609421,
+      "step": 1092
+    },
+    {
+      "epoch": 0.18216666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.8827473958333334,
+      "learning_rate": 9.204722911490846e-05,
+      "loss": 7.3469,
+      "loss/crossentropy": 2.151069074869156,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1687615904957056,
+      "step": 1093
+    },
+    {
+      "epoch": 0.18233333333333332,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.814583333333333,
+      "learning_rate": 9.20330568305594e-05,
+      "loss": 7.0395,
+      "loss/crossentropy": 1.7675020694732666,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17254584282636642,
+      "step": 1094
+    },
+    {
+      "epoch": 0.1825,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.7889973958333334,
+      "learning_rate": 9.201887302260943e-05,
+      "loss": 7.6286,
+      "loss/crossentropy": 1.4319088160991669,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1769372746348381,
+      "step": 1095
+    },
+    {
+      "epoch": 0.18266666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.3697265625,
+      "learning_rate": 9.20046776949471e-05,
+      "loss": 7.0073,
+      "loss/crossentropy": 0.9747176170349121,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1367714675143361,
+      "step": 1096
+    },
+    {
+      "epoch": 0.18283333333333332,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.59765625,
+      "learning_rate": 9.199047085146415e-05,
+      "loss": 6.9152,
+      "loss/crossentropy": 1.8448036015033722,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1561015024781227,
+      "step": 1097
+    },
+    {
+      "epoch": 0.183,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.5468098958333334,
+      "learning_rate": 9.197625249605546e-05,
+      "loss": 7.5927,
+      "loss/crossentropy": 1.6145085394382477,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21670781075954437,
+      "step": 1098
+    },
+    {
+      "epoch": 0.18316666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.627083333333333,
+      "learning_rate": 9.196202263261908e-05,
+      "loss": 6.9341,
+      "loss/crossentropy": 1.88466215133667,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.228733841329813,
+      "step": 1099
+    },
+    {
+      "epoch": 0.18333333333333332,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.6702473958333335,
+      "learning_rate": 9.194778126505621e-05,
+      "loss": 7.2048,
+      "loss/crossentropy": 1.3676262497901917,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1408743243664503,
+      "step": 1100
+    },
+    {
+      "epoch": 0.1835,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.627083333333333,
+      "learning_rate": 9.193352839727121e-05,
+      "loss": 7.0562,
+      "loss/crossentropy": 1.430459812283516,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17401222698390484,
+      "step": 1101
+    },
+    {
+      "epoch": 0.18366666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.5468098958333334,
+      "learning_rate": 9.191926403317155e-05,
+      "loss": 7.1413,
+      "loss/crossentropy": 1.951472908258438,
+      "loss/hidden": 3.66015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19395877420902252,
+      "step": 1102
+    },
+    {
+      "epoch": 0.18383333333333332,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.6389973958333335,
+      "learning_rate": 9.190498817666793e-05,
+      "loss": 7.2435,
+      "loss/crossentropy": 1.633792757987976,
+      "loss/hidden": 3.8671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2615926116704941,
+      "step": 1103
+    },
+    {
+      "epoch": 0.184,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.518684895833333,
+      "learning_rate": 9.189070083167411e-05,
+      "loss": 7.6263,
+      "loss/crossentropy": 1.7993624210357666,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2677503116428852,
+      "step": 1104
+    },
+    {
+      "epoch": 0.18416666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.3363932291666667,
+      "learning_rate": 9.18764020021071e-05,
+      "loss": 6.9299,
+      "loss/crossentropy": 1.8217281103134155,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17948501370847225,
+      "step": 1105
+    },
+    {
+      "epoch": 0.18433333333333332,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.2375,
+      "learning_rate": 9.186209169188695e-05,
+      "loss": 7.4359,
+      "loss/crossentropy": 1.954070895910263,
+      "loss/hidden": 3.83984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21222223341464996,
+      "step": 1106
+    },
+    {
+      "epoch": 0.1845,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.9155598958333333,
+      "learning_rate": 9.184776990493695e-05,
+      "loss": 7.0956,
+      "loss/crossentropy": 1.234974890947342,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21905330382287502,
+      "step": 1107
+    },
+    {
+      "epoch": 0.18466666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 0.8559895833333333,
+      "learning_rate": 9.183343664518348e-05,
+      "loss": 7.0289,
+      "loss/crossentropy": 1.3418802618980408,
+      "loss/hidden": 3.96875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18474972620606422,
+      "step": 1108
+    },
+    {
+      "epoch": 0.18483333333333332,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.0747395833333333,
+      "learning_rate": 9.181909191655612e-05,
+      "loss": 7.2423,
+      "loss/crossentropy": 1.1905895471572876,
+      "loss/hidden": 3.78125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17837058566510677,
+      "step": 1109
+    },
+    {
+      "epoch": 0.185,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.0468098958333334,
+      "learning_rate": 9.180473572298751e-05,
+      "loss": 7.2901,
+      "loss/crossentropy": 1.6753973066806793,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17351505905389786,
+      "step": 1110
+    },
+    {
+      "epoch": 0.18516666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.0479166666666666,
+      "learning_rate": 9.179036806841353e-05,
+      "loss": 7.511,
+      "loss/crossentropy": 2.0897219479084015,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24179621413350105,
+      "step": 1111
+    },
+    {
+      "epoch": 0.18533333333333332,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.521875,
+      "learning_rate": 9.177598895677309e-05,
+      "loss": 7.1737,
+      "loss/crossentropy": 1.7134060561656952,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19625892862677574,
+      "step": 1112
+    },
+    {
+      "epoch": 0.1855,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.1718098958333334,
+      "learning_rate": 9.176159839200838e-05,
+      "loss": 6.9251,
+      "loss/crossentropy": 1.5371767282485962,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20032985508441925,
+      "step": 1113
+    },
+    {
+      "epoch": 0.18566666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.1936848958333333,
+      "learning_rate": 9.17471963780646e-05,
+      "loss": 6.7028,
+      "loss/crossentropy": 1.7134927213191986,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18902384862303734,
+      "step": 1114
+    },
+    {
+      "epoch": 0.18583333333333332,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.3608723958333333,
+      "learning_rate": 9.173278291889015e-05,
+      "loss": 7.036,
+      "loss/crossentropy": 1.5526778548955917,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16117258742451668,
+      "step": 1115
+    },
+    {
+      "epoch": 0.186,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.3830729166666667,
+      "learning_rate": 9.171835801843658e-05,
+      "loss": 7.3174,
+      "loss/crossentropy": 1.6507572382688522,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1393357813358307,
+      "step": 1116
+    },
+    {
+      "epoch": 0.18616666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.4098307291666667,
+      "learning_rate": 9.170392168065857e-05,
+      "loss": 7.4685,
+      "loss/crossentropy": 1.1459402590990067,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14188237488269806,
+      "step": 1117
+    },
+    {
+      "epoch": 0.18633333333333332,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.5809895833333334,
+      "learning_rate": 9.168947390951388e-05,
+      "loss": 7.1813,
+      "loss/crossentropy": 1.8640584349632263,
+      "loss/hidden": 3.640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2659045271575451,
+      "step": 1118
+    },
+    {
+      "epoch": 0.1865,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.7509765625,
+      "learning_rate": 9.167501470896349e-05,
+      "loss": 7.7581,
+      "loss/crossentropy": 2.0622823238372803,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29800528287887573,
+      "step": 1119
+    },
+    {
+      "epoch": 0.18666666666666668,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.9462890625,
+      "learning_rate": 9.166054408297145e-05,
+      "loss": 6.8545,
+      "loss/crossentropy": 1.8968719244003296,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19029945507645607,
+      "step": 1120
+    },
+    {
+      "epoch": 0.18683333333333332,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.9369140625,
+      "learning_rate": 9.164606203550497e-05,
+      "loss": 7.1498,
+      "loss/crossentropy": 1.6898181438446045,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14866111241281033,
+      "step": 1121
+    },
+    {
+      "epoch": 0.187,
+      "grad_norm": 37.25,
+      "grad_norm_var": 4.83515625,
+      "learning_rate": 9.16315685705344e-05,
+      "loss": 7.2064,
+      "loss/crossentropy": 1.9527091979980469,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16515155881643295,
+      "step": 1122
+    },
+    {
+      "epoch": 0.18716666666666668,
+      "grad_norm": 35.0,
+      "grad_norm_var": 5.8041015625,
+      "learning_rate": 9.161706369203317e-05,
+      "loss": 7.0782,
+      "loss/crossentropy": 2.214310497045517,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2078092135488987,
+      "step": 1123
+    },
+    {
+      "epoch": 0.18733333333333332,
+      "grad_norm": 29.5,
+      "grad_norm_var": 5.878125,
+      "learning_rate": 9.160254740397791e-05,
+      "loss": 7.0496,
+      "loss/crossentropy": 1.6759832054376602,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23697318509221077,
+      "step": 1124
+    },
+    {
+      "epoch": 0.1875,
+      "grad_norm": 30.375,
+      "grad_norm_var": 5.8244140625,
+      "learning_rate": 9.158801971034832e-05,
+      "loss": 7.1634,
+      "loss/crossentropy": 1.6485983282327652,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1524697132408619,
+      "step": 1125
+    },
+    {
+      "epoch": 0.18766666666666668,
+      "grad_norm": 29.5,
+      "grad_norm_var": 5.97265625,
+      "learning_rate": 9.157348061512727e-05,
+      "loss": 6.9883,
+      "loss/crossentropy": 1.90101557970047,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14513207040727139,
+      "step": 1126
+    },
+    {
+      "epoch": 0.18783333333333332,
+      "grad_norm": 29.875,
+      "grad_norm_var": 6.0166015625,
+      "learning_rate": 9.15589301223007e-05,
+      "loss": 6.9935,
+      "loss/crossentropy": 1.2328452542424202,
+      "loss/hidden": 3.8046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19398222211748362,
+      "step": 1127
+    },
+    {
+      "epoch": 0.188,
+      "grad_norm": 27.125,
+      "grad_norm_var": 6.46640625,
+      "learning_rate": 9.154436823585777e-05,
+      "loss": 6.7857,
+      "loss/crossentropy": 1.3963468447327614,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15472635626792908,
+      "step": 1128
+    },
+    {
+      "epoch": 0.18816666666666668,
+      "grad_norm": 30.0,
+      "grad_norm_var": 6.48125,
+      "learning_rate": 9.152979495979063e-05,
+      "loss": 7.136,
+      "loss/crossentropy": 2.2000677585601807,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17560309544205666,
+      "step": 1129
+    },
+    {
+      "epoch": 0.18833333333333332,
+      "grad_norm": 30.0,
+      "grad_norm_var": 6.4884765625,
+      "learning_rate": 9.151521029809469e-05,
+      "loss": 7.5217,
+      "loss/crossentropy": 1.9590549767017365,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17268436029553413,
+      "step": 1130
+    },
+    {
+      "epoch": 0.1885,
+      "grad_norm": 31.0,
+      "grad_norm_var": 6.249739583333334,
+      "learning_rate": 9.150061425476838e-05,
+      "loss": 7.6719,
+      "loss/crossentropy": 1.4692141264677048,
+      "loss/hidden": 3.96484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22101164050400257,
+      "step": 1131
+    },
+    {
+      "epoch": 0.18866666666666668,
+      "grad_norm": 29.125,
+      "grad_norm_var": 6.273893229166666,
+      "learning_rate": 9.14860068338133e-05,
+      "loss": 7.1382,
+      "loss/crossentropy": 1.3053969889879227,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18069183453917503,
+      "step": 1132
+    },
+    {
+      "epoch": 0.18883333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 6.308072916666666,
+      "learning_rate": 9.147138803923416e-05,
+      "loss": 7.2155,
+      "loss/crossentropy": 1.490529328584671,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1413809834048152,
+      "step": 1133
+    },
+    {
+      "epoch": 0.189,
+      "grad_norm": 30.5,
+      "grad_norm_var": 6.145572916666667,
+      "learning_rate": 9.145675787503878e-05,
+      "loss": 7.4358,
+      "loss/crossentropy": 1.9087640345096588,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24026436731219292,
+      "step": 1134
+    },
+    {
+      "epoch": 0.18916666666666668,
+      "grad_norm": 29.875,
+      "grad_norm_var": 5.914518229166666,
+      "learning_rate": 9.14421163452381e-05,
+      "loss": 7.447,
+      "loss/crossentropy": 1.5070382356643677,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2686901167035103,
+      "step": 1135
+    },
+    {
+      "epoch": 0.18933333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 5.585416666666666,
+      "learning_rate": 9.142746345384619e-05,
+      "loss": 7.2418,
+      "loss/crossentropy": 1.4398823231458664,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1659804955124855,
+      "step": 1136
+    },
+    {
+      "epoch": 0.1895,
+      "grad_norm": 31.125,
+      "grad_norm_var": 5.596809895833333,
+      "learning_rate": 9.141279920488021e-05,
+      "loss": 7.0661,
+      "loss/crossentropy": 1.9212416410446167,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17001407593488693,
+      "step": 1137
+    },
+    {
+      "epoch": 0.18966666666666668,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.4809895833333333,
+      "learning_rate": 9.139812360236046e-05,
+      "loss": 7.1941,
+      "loss/crossentropy": 1.8251923620700836,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14800377935171127,
+      "step": 1138
+    },
+    {
+      "epoch": 0.18983333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 0.8934895833333333,
+      "learning_rate": 9.138343665031033e-05,
+      "loss": 7.3292,
+      "loss/crossentropy": 1.8601129055023193,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17241744697093964,
+      "step": 1139
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 29.75,
+      "grad_norm_var": 0.8864583333333333,
+      "learning_rate": 9.136873835275633e-05,
+      "loss": 6.8716,
+      "loss/crossentropy": 1.8089761137962341,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15974725037813187,
+      "step": 1140
+    },
+    {
+      "epoch": 0.19016666666666668,
+      "grad_norm": 29.875,
+      "grad_norm_var": 0.8666666666666667,
+      "learning_rate": 9.135402871372808e-05,
+      "loss": 7.2061,
+      "loss/crossentropy": 1.6150592416524887,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24301055073738098,
+      "step": 1141
+    },
+    {
+      "epoch": 0.19033333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 0.9639973958333333,
+      "learning_rate": 9.133930773725834e-05,
+      "loss": 7.3628,
+      "loss/crossentropy": 1.563875988125801,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17582198977470398,
+      "step": 1142
+    },
+    {
+      "epoch": 0.1905,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.2035807291666667,
+      "learning_rate": 9.132457542738292e-05,
+      "loss": 7.1857,
+      "loss/crossentropy": 1.9527733027935028,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1798405796289444,
+      "step": 1143
+    },
+    {
+      "epoch": 0.19066666666666668,
+      "grad_norm": 29.125,
+      "grad_norm_var": 0.6764973958333333,
+      "learning_rate": 9.130983178814077e-05,
+      "loss": 7.0687,
+      "loss/crossentropy": 1.369474709033966,
+      "loss/hidden": 3.8671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1982022561132908,
+      "step": 1144
+    },
+    {
+      "epoch": 0.19083333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 0.7802083333333333,
+      "learning_rate": 9.129507682357394e-05,
+      "loss": 7.1149,
+      "loss/crossentropy": 1.894842118024826,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20691284909844398,
+      "step": 1145
+    },
+    {
+      "epoch": 0.191,
+      "grad_norm": 29.0,
+      "grad_norm_var": 0.8552083333333333,
+      "learning_rate": 9.128031053772759e-05,
+      "loss": 6.9229,
+      "loss/crossentropy": 1.3835454285144806,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15358954295516014,
+      "step": 1146
+    },
+    {
+      "epoch": 0.19116666666666668,
+      "grad_norm": 29.75,
+      "grad_norm_var": 0.79140625,
+      "learning_rate": 9.126553293464998e-05,
+      "loss": 7.4438,
+      "loss/crossentropy": 2.2602747678756714,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1918497011065483,
+      "step": 1147
+    },
+    {
+      "epoch": 0.19133333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 0.9905598958333334,
+      "learning_rate": 9.125074401839249e-05,
+      "loss": 7.5072,
+      "loss/crossentropy": 1.7223348915576935,
+      "loss/hidden": 3.8359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20392953604459763,
+      "step": 1148
+    },
+    {
+      "epoch": 0.1915,
+      "grad_norm": 30.5,
+      "grad_norm_var": 0.9686848958333333,
+      "learning_rate": 9.123594379300955e-05,
+      "loss": 6.8108,
+      "loss/crossentropy": 1.5406135469675064,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15648524276912212,
+      "step": 1149
+    },
+    {
+      "epoch": 0.19166666666666668,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.0343098958333334,
+      "learning_rate": 9.122113226255877e-05,
+      "loss": 7.2257,
+      "loss/crossentropy": 1.7266594469547272,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1957474946975708,
+      "step": 1150
+    },
+    {
+      "epoch": 0.19183333333333333,
+      "grad_norm": 26.75,
+      "grad_norm_var": 1.79765625,
+      "learning_rate": 9.120630943110077e-05,
+      "loss": 7.0052,
+      "loss/crossentropy": 2.1635193824768066,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1783377081155777,
+      "step": 1151
+    },
+    {
+      "epoch": 0.192,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.8285807291666667,
+      "learning_rate": 9.119147530269937e-05,
+      "loss": 7.5332,
+      "loss/crossentropy": 1.8073493093252182,
+      "loss/hidden": 3.7890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2292596623301506,
+      "step": 1152
+    },
+    {
+      "epoch": 0.19216666666666668,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.7518229166666666,
+      "learning_rate": 9.117662988142138e-05,
+      "loss": 7.1552,
+      "loss/crossentropy": 2.0263662338256836,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23550377786159515,
+      "step": 1153
+    },
+    {
+      "epoch": 0.19233333333333333,
+      "grad_norm": 36.0,
+      "grad_norm_var": 3.9942057291666666,
+      "learning_rate": 9.116177317133676e-05,
+      "loss": 7.9774,
+      "loss/crossentropy": 1.7365905940532684,
+      "loss/hidden": 3.875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2624204382300377,
+      "step": 1154
+    },
+    {
+      "epoch": 0.1925,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.6556640625,
+      "learning_rate": 9.114690517651859e-05,
+      "loss": 7.4115,
+      "loss/crossentropy": 2.2122086882591248,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19748417660593987,
+      "step": 1155
+    },
+    {
+      "epoch": 0.19266666666666668,
+      "grad_norm": 28.5,
+      "grad_norm_var": 4.9056640625,
+      "learning_rate": 9.1132025901043e-05,
+      "loss": 7.2611,
+      "loss/crossentropy": 1.843263953924179,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14568496495485306,
+      "step": 1156
+    },
+    {
+      "epoch": 0.19283333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 4.918489583333334,
+      "learning_rate": 9.111713534898922e-05,
+      "loss": 7.6664,
+      "loss/crossentropy": 1.7872939705848694,
+      "loss/hidden": 3.99609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27959929779171944,
+      "step": 1157
+    },
+    {
+      "epoch": 0.193,
+      "grad_norm": 29.5,
+      "grad_norm_var": 4.9650390625,
+      "learning_rate": 9.110223352443958e-05,
+      "loss": 7.2427,
+      "loss/crossentropy": 1.67998605966568,
+      "loss/hidden": 3.8359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23278993926942348,
+      "step": 1158
+    },
+    {
+      "epoch": 0.19316666666666665,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.834375,
+      "learning_rate": 9.108732043147952e-05,
+      "loss": 6.9476,
+      "loss/crossentropy": 1.9119806289672852,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1615016683936119,
+      "step": 1159
+    },
+    {
+      "epoch": 0.19333333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 4.726041666666666,
+      "learning_rate": 9.107239607419753e-05,
+      "loss": 6.8848,
+      "loss/crossentropy": 2.1190615743398666,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1532793641090393,
+      "step": 1160
+    },
+    {
+      "epoch": 0.1935,
+      "grad_norm": 30.125,
+      "grad_norm_var": 4.558072916666666,
+      "learning_rate": 9.105746045668521e-05,
+      "loss": 7.4131,
+      "loss/crossentropy": 1.9063195884227753,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19840717688202858,
+      "step": 1161
+    },
+    {
+      "epoch": 0.19366666666666665,
+      "grad_norm": 30.375,
+      "grad_norm_var": 4.392643229166667,
+      "learning_rate": 9.104251358303724e-05,
+      "loss": 7.3765,
+      "loss/crossentropy": 2.0767466723918915,
+      "loss/hidden": 3.78125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19928371161222458,
+      "step": 1162
+    },
+    {
+      "epoch": 0.19383333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 4.3494140625,
+      "learning_rate": 9.102755545735141e-05,
+      "loss": 7.0533,
+      "loss/crossentropy": 1.630551353096962,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20199225470423698,
+      "step": 1163
+    },
+    {
+      "epoch": 0.194,
+      "grad_norm": 31.125,
+      "grad_norm_var": 4.24140625,
+      "learning_rate": 9.101258608372856e-05,
+      "loss": 7.5036,
+      "loss/crossentropy": 1.3316044062376022,
+      "loss/hidden": 3.8125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2553766742348671,
+      "step": 1164
+    },
+    {
+      "epoch": 0.19416666666666665,
+      "grad_norm": 30.5,
+      "grad_norm_var": 4.24140625,
+      "learning_rate": 9.099760546627261e-05,
+      "loss": 7.1157,
+      "loss/crossentropy": 1.532670795917511,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1587374359369278,
+      "step": 1165
+    },
+    {
+      "epoch": 0.19433333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 4.212434895833334,
+      "learning_rate": 9.098261360909064e-05,
+      "loss": 7.2771,
+      "loss/crossentropy": 2.0286777317523956,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19840383157134056,
+      "step": 1166
+    },
+    {
+      "epoch": 0.1945,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.187239583333333,
+      "learning_rate": 9.096761051629268e-05,
+      "loss": 7.3572,
+      "loss/crossentropy": 1.5462962836027145,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1618911810219288,
+      "step": 1167
+    },
+    {
+      "epoch": 0.19466666666666665,
+      "grad_norm": 27.75,
+      "grad_norm_var": 3.7684895833333334,
+      "learning_rate": 9.095259619199197e-05,
+      "loss": 7.0961,
+      "loss/crossentropy": 1.906777799129486,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19005262665450573,
+      "step": 1168
+    },
+    {
+      "epoch": 0.19483333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.807747395833333,
+      "learning_rate": 9.093757064030473e-05,
+      "loss": 6.9761,
+      "loss/crossentropy": 1.4600431770086288,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2671759743243456,
+      "step": 1169
+    },
+    {
+      "epoch": 0.195,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.7264973958333334,
+      "learning_rate": 9.092253386535032e-05,
+      "loss": 7.2576,
+      "loss/crossentropy": 1.9618641138076782,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2775069437921047,
+      "step": 1170
+    },
+    {
+      "epoch": 0.19516666666666665,
+      "grad_norm": 29.125,
+      "grad_norm_var": 0.7489583333333333,
+      "learning_rate": 9.090748587125118e-05,
+      "loss": 7.1906,
+      "loss/crossentropy": 1.6897124648094177,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18094610795378685,
+      "step": 1171
+    },
+    {
+      "epoch": 0.19533333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 0.6122395833333333,
+      "learning_rate": 9.089242666213276e-05,
+      "loss": 7.2124,
+      "loss/crossentropy": 1.7875485718250275,
+      "loss/hidden": 3.765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2697015330195427,
+      "step": 1172
+    },
+    {
+      "epoch": 0.1955,
+      "grad_norm": 30.75,
+      "grad_norm_var": 0.6434895833333333,
+      "learning_rate": 9.087735624212365e-05,
+      "loss": 7.2829,
+      "loss/crossentropy": 1.5107744634151459,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1898435726761818,
+      "step": 1173
+    },
+    {
+      "epoch": 0.19566666666666666,
+      "grad_norm": 31.75,
+      "grad_norm_var": 0.7958333333333333,
+      "learning_rate": 9.08622746153555e-05,
+      "loss": 7.0894,
+      "loss/crossentropy": 1.4352801144123077,
+      "loss/hidden": 3.80078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1781257726252079,
+      "step": 1174
+    },
+    {
+      "epoch": 0.19583333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.1958333333333333,
+      "learning_rate": 9.084718178596301e-05,
+      "loss": 7.277,
+      "loss/crossentropy": 1.860286682844162,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22503403574228287,
+      "step": 1175
+    },
+    {
+      "epoch": 0.196,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.1936848958333333,
+      "learning_rate": 9.083207775808396e-05,
+      "loss": 7.2784,
+      "loss/crossentropy": 1.7108961418271065,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15173084288835526,
+      "step": 1176
+    },
+    {
+      "epoch": 0.19616666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.3020833333333333,
+      "learning_rate": 9.081696253585921e-05,
+      "loss": 7.3191,
+      "loss/crossentropy": 1.9792208075523376,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17990920320153236,
+      "step": 1177
+    },
+    {
+      "epoch": 0.19633333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.309375,
+      "learning_rate": 9.080183612343268e-05,
+      "loss": 7.5557,
+      "loss/crossentropy": 2.3596280813217163,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2835410051047802,
+      "step": 1178
+    },
+    {
+      "epoch": 0.1965,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.3166015625,
+      "learning_rate": 9.078669852495138e-05,
+      "loss": 7.249,
+      "loss/crossentropy": 1.1893167048692703,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.136909831315279,
+      "step": 1179
+    },
+    {
+      "epoch": 0.19666666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.2587890625,
+      "learning_rate": 9.077154974456534e-05,
+      "loss": 7.1863,
+      "loss/crossentropy": 1.8959875404834747,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2790988143533468,
+      "step": 1180
+    },
+    {
+      "epoch": 0.19683333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.5322265625,
+      "learning_rate": 9.075638978642771e-05,
+      "loss": 6.9262,
+      "loss/crossentropy": 1.694694995880127,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15728522092103958,
+      "step": 1181
+    },
+    {
+      "epoch": 0.197,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.5872395833333333,
+      "learning_rate": 9.074121865469467e-05,
+      "loss": 7.1031,
+      "loss/crossentropy": 1.5371995270252228,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15807504951953888,
+      "step": 1182
+    },
+    {
+      "epoch": 0.19716666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.5921223958333333,
+      "learning_rate": 9.072603635352548e-05,
+      "loss": 7.24,
+      "loss/crossentropy": 1.2076034247875214,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13693496398627758,
+      "step": 1183
+    },
+    {
+      "epoch": 0.19733333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.1452473958333333,
+      "learning_rate": 9.071084288708243e-05,
+      "loss": 7.5155,
+      "loss/crossentropy": 1.7932229936122894,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18797968700528145,
+      "step": 1184
+    },
+    {
+      "epoch": 0.1975,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.1723307291666667,
+      "learning_rate": 9.069563825953092e-05,
+      "loss": 7.2881,
+      "loss/crossentropy": 2.2741920351982117,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19049449265003204,
+      "step": 1185
+    },
+    {
+      "epoch": 0.19766666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.2104166666666667,
+      "learning_rate": 9.068042247503936e-05,
+      "loss": 7.3237,
+      "loss/crossentropy": 1.5595572590827942,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22788817808032036,
+      "step": 1186
+    },
+    {
+      "epoch": 0.19783333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.0645833333333334,
+      "learning_rate": 9.066519553777926e-05,
+      "loss": 7.0393,
+      "loss/crossentropy": 2.3322206139564514,
+      "loss/hidden": 3.09375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1544881910085678,
+      "step": 1187
+    },
+    {
+      "epoch": 0.198,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.0056640625,
+      "learning_rate": 9.064995745192518e-05,
+      "loss": 7.3183,
+      "loss/crossentropy": 1.702795296907425,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2040167860686779,
+      "step": 1188
+    },
+    {
+      "epoch": 0.19816666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.0333333333333334,
+      "learning_rate": 9.06347082216547e-05,
+      "loss": 7.2569,
+      "loss/crossentropy": 1.6651965826749802,
+      "loss/hidden": 3.6875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15890128910541534,
+      "step": 1189
+    },
+    {
+      "epoch": 0.19833333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.3739583333333334,
+      "learning_rate": 9.061944785114851e-05,
+      "loss": 7.5007,
+      "loss/crossentropy": 2.084866464138031,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25188354402780533,
+      "step": 1190
+    },
+    {
+      "epoch": 0.1985,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.1291666666666667,
+      "learning_rate": 9.060417634459031e-05,
+      "loss": 7.2586,
+      "loss/crossentropy": 1.3702656924724579,
+      "loss/hidden": 3.78515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1672367975115776,
+      "step": 1191
+    },
+    {
+      "epoch": 0.19866666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.1228515625,
+      "learning_rate": 9.058889370616689e-05,
+      "loss": 7.313,
+      "loss/crossentropy": 1.6381755471229553,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16992971114814281,
+      "step": 1192
+    },
+    {
+      "epoch": 0.19883333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 0.9895833333333334,
+      "learning_rate": 9.057359994006806e-05,
+      "loss": 7.135,
+      "loss/crossentropy": 1.7042577117681503,
+      "loss/hidden": 3.88671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2118433080613613,
+      "step": 1193
+    },
+    {
+      "epoch": 0.199,
+      "grad_norm": 39.25,
+      "grad_norm_var": 5.232747395833333,
+      "learning_rate": 9.055829505048667e-05,
+      "loss": 7.3896,
+      "loss/crossentropy": 1.9344637989997864,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2457873746752739,
+      "step": 1194
+    },
+    {
+      "epoch": 0.19916666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 5.14140625,
+      "learning_rate": 9.054297904161868e-05,
+      "loss": 7.1646,
+      "loss/crossentropy": 1.8423065841197968,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21592585556209087,
+      "step": 1195
+    },
+    {
+      "epoch": 0.19933333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 5.2134765625,
+      "learning_rate": 9.052765191766304e-05,
+      "loss": 6.9913,
+      "loss/crossentropy": 1.961957186460495,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16453684121370316,
+      "step": 1196
+    },
+    {
+      "epoch": 0.1995,
+      "grad_norm": 29.75,
+      "grad_norm_var": 5.3931640625,
+      "learning_rate": 9.051231368282177e-05,
+      "loss": 6.9762,
+      "loss/crossentropy": 1.8208012282848358,
+      "loss/hidden": 3.734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18695107102394104,
+      "step": 1197
+    },
+    {
+      "epoch": 0.19966666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 5.4041015625,
+      "learning_rate": 9.049696434129994e-05,
+      "loss": 7.3369,
+      "loss/crossentropy": 1.4807577729225159,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15388921275734901,
+      "step": 1198
+    },
+    {
+      "epoch": 0.19983333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 5.2556640625,
+      "learning_rate": 9.048160389730566e-05,
+      "loss": 7.2248,
+      "loss/crossentropy": 1.916600614786148,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26941858045756817,
+      "step": 1199
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.221809895833333,
+      "learning_rate": 9.046623235505007e-05,
+      "loss": 7.6083,
+      "loss/crossentropy": 2.1877745985984802,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21679533645510674,
+      "step": 1200
+    },
+    {
+      "epoch": 0.20016666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.230989583333334,
+      "learning_rate": 9.045084971874738e-05,
+      "loss": 7.3891,
+      "loss/crossentropy": 1.656475082039833,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16719500720500946,
+      "step": 1201
+    },
+    {
+      "epoch": 0.20033333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 5.256705729166667,
+      "learning_rate": 9.043545599261481e-05,
+      "loss": 7.1671,
+      "loss/crossentropy": 1.907217413187027,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1764325574040413,
+      "step": 1202
+    },
+    {
+      "epoch": 0.2005,
+      "grad_norm": 28.125,
+      "grad_norm_var": 5.900455729166667,
+      "learning_rate": 9.042005118087267e-05,
+      "loss": 6.9449,
+      "loss/crossentropy": 2.289278507232666,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19305343553423882,
+      "step": 1203
+    },
+    {
+      "epoch": 0.20066666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 5.833072916666667,
+      "learning_rate": 9.040463528774423e-05,
+      "loss": 7.091,
+      "loss/crossentropy": 1.8759718835353851,
+      "loss/hidden": 3.83984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32736699283123016,
+      "step": 1204
+    },
+    {
+      "epoch": 0.20083333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.705143229166667,
+      "learning_rate": 9.038920831745587e-05,
+      "loss": 7.3753,
+      "loss/crossentropy": 1.1482383757829666,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19319440051913261,
+      "step": 1205
+    },
+    {
+      "epoch": 0.201,
+      "grad_norm": 29.125,
+      "grad_norm_var": 5.879166666666666,
+      "learning_rate": 9.0373770274237e-05,
+      "loss": 7.2135,
+      "loss/crossentropy": 1.4531298279762268,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21289758384227753,
+      "step": 1206
+    },
+    {
+      "epoch": 0.20116666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 5.789518229166666,
+      "learning_rate": 9.035832116232001e-05,
+      "loss": 7.0233,
+      "loss/crossentropy": 1.523962140083313,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17138102278113365,
+      "step": 1207
+    },
+    {
+      "epoch": 0.20133333333333334,
+      "grad_norm": 34.0,
+      "grad_norm_var": 6.093489583333334,
+      "learning_rate": 9.03428609859404e-05,
+      "loss": 7.3418,
+      "loss/crossentropy": 1.1653930693864822,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1387856211513281,
+      "step": 1208
+    },
+    {
+      "epoch": 0.2015,
+      "grad_norm": 32.5,
+      "grad_norm_var": 6.134830729166667,
+      "learning_rate": 9.032738974933664e-05,
+      "loss": 7.213,
+      "loss/crossentropy": 1.0771423131227493,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14415764808654785,
+      "step": 1209
+    },
+    {
+      "epoch": 0.20166666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.1504557291666666,
+      "learning_rate": 9.031190745675024e-05,
+      "loss": 6.9565,
+      "loss/crossentropy": 1.8841006457805634,
+      "loss/hidden": 3.91015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23506051301956177,
+      "step": 1210
+    },
+    {
+      "epoch": 0.20183333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.2728515625,
+      "learning_rate": 9.029641411242579e-05,
+      "loss": 7.5154,
+      "loss/crossentropy": 2.3743822276592255,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22775381058454514,
+      "step": 1211
+    },
+    {
+      "epoch": 0.202,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.237434895833333,
+      "learning_rate": 9.028090972061088e-05,
+      "loss": 7.53,
+      "loss/crossentropy": 1.9960069805383682,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1583007611334324,
+      "step": 1212
+    },
+    {
+      "epoch": 0.20216666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.1166666666666667,
+      "learning_rate": 9.02653942855561e-05,
+      "loss": 7.3417,
+      "loss/crossentropy": 1.8124405890703201,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19223373383283615,
+      "step": 1213
+    },
+    {
+      "epoch": 0.20233333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.137955729166667,
+      "learning_rate": 9.024986781151512e-05,
+      "loss": 7.1486,
+      "loss/crossentropy": 1.587627574801445,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19454994052648544,
+      "step": 1214
+    },
+    {
+      "epoch": 0.2025,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.634375,
+      "learning_rate": 9.023433030274459e-05,
+      "loss": 7.1628,
+      "loss/crossentropy": 1.7969483435153961,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21532484143972397,
+      "step": 1215
+    },
+    {
+      "epoch": 0.20266666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.58125,
+      "learning_rate": 9.021878176350423e-05,
+      "loss": 7.0582,
+      "loss/crossentropy": 1.6361202746629715,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17281216382980347,
+      "step": 1216
+    },
+    {
+      "epoch": 0.20283333333333334,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.8150390625,
+      "learning_rate": 9.020322219805674e-05,
+      "loss": 7.1401,
+      "loss/crossentropy": 1.6873044669628143,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17319026961922646,
+      "step": 1217
+    },
+    {
+      "epoch": 0.203,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.8228515625,
+      "learning_rate": 9.018765161066787e-05,
+      "loss": 7.2404,
+      "loss/crossentropy": 2.2303740978240967,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15988197550177574,
+      "step": 1218
+    },
+    {
+      "epoch": 0.20316666666666666,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.434375,
+      "learning_rate": 9.017207000560639e-05,
+      "loss": 6.7894,
+      "loss/crossentropy": 1.663067877292633,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2851884439587593,
+      "step": 1219
+    },
+    {
+      "epoch": 0.20333333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.49140625,
+      "learning_rate": 9.015647738714408e-05,
+      "loss": 7.3317,
+      "loss/crossentropy": 1.6652137711644173,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13950848951935768,
+      "step": 1220
+    },
+    {
+      "epoch": 0.2035,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.42890625,
+      "learning_rate": 9.014087375955573e-05,
+      "loss": 7.423,
+      "loss/crossentropy": 1.8448649644851685,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.143984017893672,
+      "step": 1221
+    },
+    {
+      "epoch": 0.20366666666666666,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.3139973958333333,
+      "learning_rate": 9.012525912711918e-05,
+      "loss": 7.2826,
+      "loss/crossentropy": 1.7346484065055847,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22051877155900002,
+      "step": 1222
+    },
+    {
+      "epoch": 0.20383333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.3358723958333334,
+      "learning_rate": 9.010963349411529e-05,
+      "loss": 7.3522,
+      "loss/crossentropy": 1.7686149775981903,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18443066999316216,
+      "step": 1223
+    },
+    {
+      "epoch": 0.204,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.7520182291666666,
+      "learning_rate": 9.009399686482787e-05,
+      "loss": 7.1644,
+      "loss/crossentropy": 1.4964023679494858,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17550109699368477,
+      "step": 1224
+    },
+    {
+      "epoch": 0.20416666666666666,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.5885416666666667,
+      "learning_rate": 9.007834924354383e-05,
+      "loss": 7.2392,
+      "loss/crossentropy": 2.415601521730423,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19092579558491707,
+      "step": 1225
+    },
+    {
+      "epoch": 0.20433333333333334,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.5893229166666667,
+      "learning_rate": 9.006269063455304e-05,
+      "loss": 7.2053,
+      "loss/crossentropy": 1.259591594338417,
+      "loss/hidden": 4.09765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23988616839051247,
+      "step": 1226
+    },
+    {
+      "epoch": 0.2045,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.1518229166666667,
+      "learning_rate": 9.00470210421484e-05,
+      "loss": 7.3441,
+      "loss/crossentropy": 1.7177473604679108,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19613390043377876,
+      "step": 1227
+    },
+    {
+      "epoch": 0.20466666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.61875,
+      "learning_rate": 9.003134047062579e-05,
+      "loss": 7.2235,
+      "loss/crossentropy": 1.3223965615034103,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2001427672803402,
+      "step": 1228
+    },
+    {
+      "epoch": 0.20483333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.62890625,
+      "learning_rate": 9.001564892428415e-05,
+      "loss": 7.1399,
+      "loss/crossentropy": 1.687671348452568,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1553989164531231,
+      "step": 1229
+    },
+    {
+      "epoch": 0.205,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.6822265625,
+      "learning_rate": 8.999994640742543e-05,
+      "loss": 7.5995,
+      "loss/crossentropy": 2.007145345211029,
+      "loss/hidden": 3.8359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2645701393485069,
+      "step": 1230
+    },
+    {
+      "epoch": 0.20516666666666666,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.5166015625,
+      "learning_rate": 8.998423292435454e-05,
+      "loss": 7.1565,
+      "loss/crossentropy": 1.9185426235198975,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1818624660372734,
+      "step": 1231
+    },
+    {
+      "epoch": 0.20533333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.51875,
+      "learning_rate": 8.996850847937941e-05,
+      "loss": 7.456,
+      "loss/crossentropy": 2.386841744184494,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20238413661718369,
+      "step": 1232
+    },
+    {
+      "epoch": 0.2055,
+      "grad_norm": 53.5,
+      "grad_norm_var": 34.18014322916667,
+      "learning_rate": 8.995277307681099e-05,
+      "loss": 7.2791,
+      "loss/crossentropy": 1.2465152591466904,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14154892787337303,
+      "step": 1233
+    },
+    {
+      "epoch": 0.20566666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 33.967708333333334,
+      "learning_rate": 8.993702672096324e-05,
+      "loss": 7.4117,
+      "loss/crossentropy": 2.1675582230091095,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18371211923658848,
+      "step": 1234
+    },
+    {
+      "epoch": 0.20583333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 33.612239583333334,
+      "learning_rate": 8.992126941615313e-05,
+      "loss": 7.3614,
+      "loss/crossentropy": 1.3925307989120483,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13119136169552803,
+      "step": 1235
+    },
+    {
+      "epoch": 0.206,
+      "grad_norm": 27.875,
+      "grad_norm_var": 34.784830729166664,
+      "learning_rate": 8.990550116670057e-05,
+      "loss": 6.8879,
+      "loss/crossentropy": 1.5162500590085983,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14715932123363018,
+      "step": 1236
+    },
+    {
+      "epoch": 0.20616666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 34.959309895833336,
+      "learning_rate": 8.988972197692855e-05,
+      "loss": 7.2281,
+      "loss/crossentropy": 2.0481223464012146,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15340117923915386,
+      "step": 1237
+    },
+    {
+      "epoch": 0.20633333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 34.83274739583333,
+      "learning_rate": 8.987393185116302e-05,
+      "loss": 7.5103,
+      "loss/crossentropy": 1.9297794103622437,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24415437504649162,
+      "step": 1238
+    },
+    {
+      "epoch": 0.2065,
+      "grad_norm": 30.25,
+      "grad_norm_var": 35.00390625,
+      "learning_rate": 8.985813079373292e-05,
+      "loss": 6.8773,
+      "loss/crossentropy": 1.6119352504611015,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15101561322808266,
+      "step": 1239
+    },
+    {
+      "epoch": 0.20666666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 35.1416015625,
+      "learning_rate": 8.98423188089702e-05,
+      "loss": 7.1719,
+      "loss/crossentropy": 1.3819709122180939,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1600161474198103,
+      "step": 1240
+    },
+    {
+      "epoch": 0.20683333333333334,
+      "grad_norm": 29.0,
+      "grad_norm_var": 35.18723958333333,
+      "learning_rate": 8.982649590120982e-05,
+      "loss": 7.2573,
+      "loss/crossentropy": 1.6761731505393982,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1604536771774292,
+      "step": 1241
+    },
+    {
+      "epoch": 0.207,
+      "grad_norm": 48.75,
+      "grad_norm_var": 52.84791666666667,
+      "learning_rate": 8.981066207478971e-05,
+      "loss": 7.4697,
+      "loss/crossentropy": 1.8623092472553253,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2204706110060215,
+      "step": 1242
+    },
+    {
+      "epoch": 0.20716666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 53.41458333333333,
+      "learning_rate": 8.97948173340508e-05,
+      "loss": 6.8498,
+      "loss/crossentropy": 1.5715491771697998,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18219494447112083,
+      "step": 1243
+    },
+    {
+      "epoch": 0.20733333333333334,
+      "grad_norm": 29.0,
+      "grad_norm_var": 54.31458333333333,
+      "learning_rate": 8.977896168333702e-05,
+      "loss": 6.969,
+      "loss/crossentropy": 1.8242640793323517,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1585025191307068,
+      "step": 1244
+    },
+    {
+      "epoch": 0.2075,
+      "grad_norm": 28.0,
+      "grad_norm_var": 55.47805989583333,
+      "learning_rate": 8.976309512699528e-05,
+      "loss": 6.9946,
+      "loss/crossentropy": 1.7089583277702332,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15191133320331573,
+      "step": 1245
+    },
+    {
+      "epoch": 0.20766666666666667,
+      "grad_norm": 28.125,
+      "grad_norm_var": 56.12473958333333,
+      "learning_rate": 8.97472176693755e-05,
+      "loss": 6.9526,
+      "loss/crossentropy": 1.2086962163448334,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12226183898746967,
+      "step": 1246
+    },
+    {
+      "epoch": 0.20783333333333334,
+      "grad_norm": 28.125,
+      "grad_norm_var": 56.61015625,
+      "learning_rate": 8.973132931483057e-05,
+      "loss": 7.0176,
+      "loss/crossentropy": 1.7964928448200226,
+      "loss/hidden": 3.8359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22055083885788918,
+      "step": 1247
+    },
+    {
+      "epoch": 0.208,
+      "grad_norm": 32.75,
+      "grad_norm_var": 56.43639322916667,
+      "learning_rate": 8.971543006771636e-05,
+      "loss": 7.3351,
+      "loss/crossentropy": 1.9487533867359161,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18315195851027966,
+      "step": 1248
+    },
+    {
+      "epoch": 0.20816666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 24.685416666666665,
+      "learning_rate": 8.969951993239177e-05,
+      "loss": 7.1565,
+      "loss/crossentropy": 1.9017888307571411,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17371082678437233,
+      "step": 1249
+    },
+    {
+      "epoch": 0.20833333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 24.623893229166665,
+      "learning_rate": 8.968359891321862e-05,
+      "loss": 7.0055,
+      "loss/crossentropy": 1.9904980957508087,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.183632992208004,
+      "step": 1250
+    },
+    {
+      "epoch": 0.2085,
+      "grad_norm": 46.5,
+      "grad_norm_var": 39.95201822916667,
+      "learning_rate": 8.966766701456177e-05,
+      "loss": 7.5897,
+      "loss/crossentropy": 1.880692034959793,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17999951913952827,
+      "step": 1251
+    },
+    {
+      "epoch": 0.20866666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 38.962239583333336,
+      "learning_rate": 8.965172424078902e-05,
+      "loss": 7.2985,
+      "loss/crossentropy": 2.001313865184784,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21834712848067284,
+      "step": 1252
+    },
+    {
+      "epoch": 0.20883333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 38.774739583333336,
+      "learning_rate": 8.963577059627118e-05,
+      "loss": 6.9364,
+      "loss/crossentropy": 1.801378846168518,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23835700377821922,
+      "step": 1253
+    },
+    {
+      "epoch": 0.209,
+      "grad_norm": 29.625,
+      "grad_norm_var": 38.956705729166664,
+      "learning_rate": 8.961980608538203e-05,
+      "loss": 7.1429,
+      "loss/crossentropy": 1.9462988674640656,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20825079455971718,
+      "step": 1254
+    },
+    {
+      "epoch": 0.20916666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 38.89889322916667,
+      "learning_rate": 8.960383071249836e-05,
+      "loss": 7.3652,
+      "loss/crossentropy": 1.714004397392273,
+      "loss/hidden": 4.0625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3170866258442402,
+      "step": 1255
+    },
+    {
+      "epoch": 0.20933333333333334,
+      "grad_norm": 27.5,
+      "grad_norm_var": 39.61145833333333,
+      "learning_rate": 8.958784448199987e-05,
+      "loss": 6.8965,
+      "loss/crossentropy": 1.8080734759569168,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14889338053762913,
+      "step": 1256
+    },
+    {
+      "epoch": 0.2095,
+      "grad_norm": 29.875,
+      "grad_norm_var": 39.3056640625,
+      "learning_rate": 8.95718473982693e-05,
+      "loss": 7.1474,
+      "loss/crossentropy": 1.6043298840522766,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14702529087662697,
+      "step": 1257
+    },
+    {
+      "epoch": 0.20966666666666667,
+      "grad_norm": 28.5,
+      "grad_norm_var": 19.9416015625,
+      "learning_rate": 8.955583946569233e-05,
+      "loss": 6.8836,
+      "loss/crossentropy": 2.12226665019989,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2022249810397625,
+      "step": 1258
+    },
+    {
+      "epoch": 0.20983333333333334,
+      "grad_norm": 41.75,
+      "grad_norm_var": 27.2291015625,
+      "learning_rate": 8.95398206886576e-05,
+      "loss": 7.4099,
+      "loss/crossentropy": 1.487533763051033,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20956316590309143,
+      "step": 1259
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 29.75,
+      "grad_norm_var": 27.0072265625,
+      "learning_rate": 8.95237910715568e-05,
+      "loss": 7.2558,
+      "loss/crossentropy": 1.2420831471681595,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1411299854516983,
+      "step": 1260
+    },
+    {
+      "epoch": 0.21016666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 26.1228515625,
+      "learning_rate": 8.950775061878453e-05,
+      "loss": 7.1308,
+      "loss/crossentropy": 1.8089599609375,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16879883781075478,
+      "step": 1261
+    },
+    {
+      "epoch": 0.21033333333333334,
+      "grad_norm": 27.5,
+      "grad_norm_var": 26.45390625,
+      "learning_rate": 8.949169933473833e-05,
+      "loss": 6.6541,
+      "loss/crossentropy": 1.3704565316438675,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12190248817205429,
+      "step": 1262
+    },
+    {
+      "epoch": 0.2105,
+      "grad_norm": 32.25,
+      "grad_norm_var": 25.5150390625,
+      "learning_rate": 8.94756372238188e-05,
+      "loss": 7.4878,
+      "loss/crossentropy": 2.032979369163513,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18531574681401253,
+      "step": 1263
+    },
+    {
+      "epoch": 0.21066666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 25.4947265625,
+      "learning_rate": 8.945956429042943e-05,
+      "loss": 7.1824,
+      "loss/crossentropy": 1.8457638323307037,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18632981181144714,
+      "step": 1264
+    },
+    {
+      "epoch": 0.21083333333333334,
+      "grad_norm": 28.875,
+      "grad_norm_var": 26.0087890625,
+      "learning_rate": 8.944348053897671e-05,
+      "loss": 7.2811,
+      "loss/crossentropy": 1.8484606444835663,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16157929599285126,
+      "step": 1265
+    },
+    {
+      "epoch": 0.211,
+      "grad_norm": 29.75,
+      "grad_norm_var": 26.2244140625,
+      "learning_rate": 8.94273859738701e-05,
+      "loss": 7.2842,
+      "loss/crossentropy": 1.7703493386507034,
+      "loss/hidden": 3.84375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18733574450016022,
+      "step": 1266
+    },
+    {
+      "epoch": 0.21116666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 11.009830729166667,
+      "learning_rate": 8.941128059952201e-05,
+      "loss": 6.9205,
+      "loss/crossentropy": 1.498087391257286,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20384216867387295,
+      "step": 1267
+    },
+    {
+      "epoch": 0.21133333333333335,
+      "grad_norm": 30.75,
+      "grad_norm_var": 11.009830729166667,
+      "learning_rate": 8.939516442034781e-05,
+      "loss": 7.0617,
+      "loss/crossentropy": 1.2050846070051193,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10520477686077356,
+      "step": 1268
+    },
+    {
+      "epoch": 0.2115,
+      "grad_norm": 38.5,
+      "grad_norm_var": 14.579622395833333,
+      "learning_rate": 8.937903744076587e-05,
+      "loss": 7.1833,
+      "loss/crossentropy": 2.160456895828247,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21616069972515106,
+      "step": 1269
+    },
+    {
+      "epoch": 0.21166666666666667,
+      "grad_norm": 39.5,
+      "grad_norm_var": 18.668489583333333,
+      "learning_rate": 8.936289966519746e-05,
+      "loss": 7.3277,
+      "loss/crossentropy": 1.8163149058818817,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26328181475400925,
+      "step": 1270
+    },
+    {
+      "epoch": 0.21183333333333335,
+      "grad_norm": 28.375,
+      "grad_norm_var": 19.309309895833334,
+      "learning_rate": 8.934675109806688e-05,
+      "loss": 7.1807,
+      "loss/crossentropy": 1.779030442237854,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23459625989198685,
+      "step": 1271
+    },
+    {
+      "epoch": 0.212,
+      "grad_norm": 28.875,
+      "grad_norm_var": 18.669791666666665,
+      "learning_rate": 8.933059174380131e-05,
+      "loss": 6.8549,
+      "loss/crossentropy": 1.8299419581890106,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16380924358963966,
+      "step": 1272
+    },
+    {
+      "epoch": 0.21216666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 19.745768229166668,
+      "learning_rate": 8.931442160683094e-05,
+      "loss": 6.9824,
+      "loss/crossentropy": 1.7724792212247849,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18576912954449654,
+      "step": 1273
+    },
+    {
+      "epoch": 0.21233333333333335,
+      "grad_norm": 29.625,
+      "grad_norm_var": 19.366666666666667,
+      "learning_rate": 8.929824069158894e-05,
+      "loss": 7.3272,
+      "loss/crossentropy": 1.7554685175418854,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24172433838248253,
+      "step": 1274
+    },
+    {
+      "epoch": 0.2125,
+      "grad_norm": 29.875,
+      "grad_norm_var": 12.148893229166667,
+      "learning_rate": 8.928204900251136e-05,
+      "loss": 7.3142,
+      "loss/crossentropy": 1.7340553104877472,
+      "loss/hidden": 3.7265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20534256100654602,
+      "step": 1275
+    },
+    {
+      "epoch": 0.21266666666666667,
+      "grad_norm": 39.25,
+      "grad_norm_var": 16.354622395833335,
+      "learning_rate": 8.926584654403724e-05,
+      "loss": 7.157,
+      "loss/crossentropy": 1.920032113790512,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19680794514715672,
+      "step": 1276
+    },
+    {
+      "epoch": 0.21283333333333335,
+      "grad_norm": 30.75,
+      "grad_norm_var": 16.3744140625,
+      "learning_rate": 8.924963332060863e-05,
+      "loss": 7.3995,
+      "loss/crossentropy": 1.841623306274414,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24768600426614285,
+      "step": 1277
+    },
+    {
+      "epoch": 0.213,
+      "grad_norm": 28.25,
+      "grad_norm_var": 16.0134765625,
+      "learning_rate": 8.92334093366704e-05,
+      "loss": 6.9662,
+      "loss/crossentropy": 1.5030003488063812,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14378325268626213,
+      "step": 1278
+    },
+    {
+      "epoch": 0.21316666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 16.352018229166667,
+      "learning_rate": 8.92171745966705e-05,
+      "loss": 6.8504,
+      "loss/crossentropy": 1.8155283033847809,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17595910839736462,
+      "step": 1279
+    },
+    {
+      "epoch": 0.21333333333333335,
+      "grad_norm": 29.75,
+      "grad_norm_var": 16.386393229166668,
+      "learning_rate": 8.920092910505977e-05,
+      "loss": 6.9831,
+      "loss/crossentropy": 1.4898201525211334,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15095620602369308,
+      "step": 1280
+    },
+    {
+      "epoch": 0.2135,
+      "grad_norm": 27.875,
+      "grad_norm_var": 16.749934895833334,
+      "learning_rate": 8.9184672866292e-05,
+      "loss": 7.2761,
+      "loss/crossentropy": 1.2847067043185234,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15730723179876804,
+      "step": 1281
+    },
+    {
+      "epoch": 0.21366666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 16.6775390625,
+      "learning_rate": 8.916840588482392e-05,
+      "loss": 7.0408,
+      "loss/crossentropy": 1.408691182732582,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1533127836883068,
+      "step": 1282
+    },
+    {
+      "epoch": 0.21383333333333332,
+      "grad_norm": 37.25,
+      "grad_norm_var": 18.8416015625,
+      "learning_rate": 8.915212816511522e-05,
+      "loss": 7.0918,
+      "loss/crossentropy": 1.2837118208408356,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23492558486759663,
+      "step": 1283
+    },
+    {
+      "epoch": 0.214,
+      "grad_norm": 30.75,
+      "grad_norm_var": 18.8416015625,
+      "learning_rate": 8.913583971162852e-05,
+      "loss": 7.3199,
+      "loss/crossentropy": 1.5599457621574402,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23544751107692719,
+      "step": 1284
+    },
+    {
+      "epoch": 0.21416666666666667,
+      "grad_norm": 31.875,
+      "grad_norm_var": 15.463541666666666,
+      "learning_rate": 8.91195405288294e-05,
+      "loss": 6.9959,
+      "loss/crossentropy": 2.083773970603943,
+      "loss/hidden": 3.75,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19660097360610962,
+      "step": 1285
+    },
+    {
+      "epoch": 0.21433333333333332,
+      "grad_norm": 27.75,
+      "grad_norm_var": 11.020572916666667,
+      "learning_rate": 8.910323062118639e-05,
+      "loss": 7.0008,
+      "loss/crossentropy": 1.7534039616584778,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.170754324644804,
+      "step": 1286
+    },
+    {
+      "epoch": 0.2145,
+      "grad_norm": 27.5,
+      "grad_norm_var": 11.3072265625,
+      "learning_rate": 8.908690999317093e-05,
+      "loss": 6.8607,
+      "loss/crossentropy": 2.0595730543136597,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19259288534522057,
+      "step": 1287
+    },
+    {
+      "epoch": 0.21466666666666667,
+      "grad_norm": 28.25,
+      "grad_norm_var": 11.455989583333333,
+      "learning_rate": 8.90705786492574e-05,
+      "loss": 7.1113,
+      "loss/crossentropy": 2.0512263476848602,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20212347432971,
+      "step": 1288
+    },
+    {
+      "epoch": 0.21483333333333332,
+      "grad_norm": 30.5,
+      "grad_norm_var": 10.782291666666667,
+      "learning_rate": 8.905423659392316e-05,
+      "loss": 7.2688,
+      "loss/crossentropy": 1.8296646773815155,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20784572884440422,
+      "step": 1289
+    },
+    {
+      "epoch": 0.215,
+      "grad_norm": 29.625,
+      "grad_norm_var": 10.782291666666667,
+      "learning_rate": 8.903788383164846e-05,
+      "loss": 7.2516,
+      "loss/crossentropy": 1.5367623269557953,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25126538798213005,
+      "step": 1290
+    },
+    {
+      "epoch": 0.21516666666666667,
+      "grad_norm": 40.75,
+      "grad_norm_var": 17.222330729166668,
+      "learning_rate": 8.90215203669165e-05,
+      "loss": 7.3099,
+      "loss/crossentropy": 1.4919409900903702,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17416539788246155,
+      "step": 1291
+    },
+    {
+      "epoch": 0.21533333333333332,
+      "grad_norm": 28.625,
+      "grad_norm_var": 12.889322916666666,
+      "learning_rate": 8.90051462042134e-05,
+      "loss": 6.9503,
+      "loss/crossentropy": 1.6225145161151886,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18747080117464066,
+      "step": 1292
+    },
+    {
+      "epoch": 0.2155,
+      "grad_norm": 31.25,
+      "grad_norm_var": 12.918489583333333,
+      "learning_rate": 8.898876134802826e-05,
+      "loss": 7.0613,
+      "loss/crossentropy": 1.424396775662899,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19608118943870068,
+      "step": 1293
+    },
+    {
+      "epoch": 0.21566666666666667,
+      "grad_norm": 27.75,
+      "grad_norm_var": 13.089322916666667,
+      "learning_rate": 8.897236580285308e-05,
+      "loss": 7.2376,
+      "loss/crossentropy": 1.5595709383487701,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15707385540008545,
+      "step": 1294
+    },
+    {
+      "epoch": 0.21583333333333332,
+      "grad_norm": 27.75,
+      "grad_norm_var": 13.444791666666667,
+      "learning_rate": 8.895595957318277e-05,
+      "loss": 7.1893,
+      "loss/crossentropy": 1.6782929003238678,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16293542832136154,
+      "step": 1295
+    },
+    {
+      "epoch": 0.216,
+      "grad_norm": 28.375,
+      "grad_norm_var": 13.6947265625,
+      "learning_rate": 8.893954266351521e-05,
+      "loss": 6.7408,
+      "loss/crossentropy": 2.086051791906357,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17098212614655495,
+      "step": 1296
+    },
+    {
+      "epoch": 0.21616666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 13.928125,
+      "learning_rate": 8.892311507835119e-05,
+      "loss": 6.9317,
+      "loss/crossentropy": 1.6282245367765427,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16926315613090992,
+      "step": 1297
+    },
+    {
+      "epoch": 0.21633333333333332,
+      "grad_norm": 35.75,
+      "grad_norm_var": 15.75,
+      "learning_rate": 8.890667682219439e-05,
+      "loss": 6.9894,
+      "loss/crossentropy": 1.540303349494934,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18785126879811287,
+      "step": 1298
+    },
+    {
+      "epoch": 0.2165,
+      "grad_norm": 31.125,
+      "grad_norm_var": 12.7353515625,
+      "learning_rate": 8.889022789955151e-05,
+      "loss": 6.9911,
+      "loss/crossentropy": 1.662076286971569,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13003813195973635,
+      "step": 1299
+    },
+    {
+      "epoch": 0.21666666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 12.7212890625,
+      "learning_rate": 8.887376831493205e-05,
+      "loss": 7.1687,
+      "loss/crossentropy": 1.9633639752864838,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19742264971137047,
+      "step": 1300
+    },
+    {
+      "epoch": 0.21683333333333332,
+      "grad_norm": 28.625,
+      "grad_norm_var": 12.687434895833333,
+      "learning_rate": 8.885729807284856e-05,
+      "loss": 7.5099,
+      "loss/crossentropy": 1.6773244589567184,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1508673969656229,
+      "step": 1301
+    },
+    {
+      "epoch": 0.217,
+      "grad_norm": 28.625,
+      "grad_norm_var": 12.464583333333334,
+      "learning_rate": 8.88408171778164e-05,
+      "loss": 7.1217,
+      "loss/crossentropy": 1.2117585092782974,
+      "loss/hidden": 3.7421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.140835739672184,
+      "step": 1302
+    },
+    {
+      "epoch": 0.21716666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 12.017122395833333,
+      "learning_rate": 8.882432563435393e-05,
+      "loss": 7.7823,
+      "loss/crossentropy": 1.9693514108657837,
+      "loss/hidden": 3.73828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1932295598089695,
+      "step": 1303
+    },
+    {
+      "epoch": 0.21733333333333332,
+      "grad_norm": 31.25,
+      "grad_norm_var": 11.738997395833334,
+      "learning_rate": 8.88078234469824e-05,
+      "loss": 6.9794,
+      "loss/crossentropy": 1.8739816546440125,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22708043083548546,
+      "step": 1304
+    },
+    {
+      "epoch": 0.2175,
+      "grad_norm": 40.25,
+      "grad_norm_var": 17.629622395833334,
+      "learning_rate": 8.879131062022598e-05,
+      "loss": 7.1087,
+      "loss/crossentropy": 1.7498936653137207,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25982804596424103,
+      "step": 1305
+    },
+    {
+      "epoch": 0.21766666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 17.527083333333334,
+      "learning_rate": 8.877478715861173e-05,
+      "loss": 7.1936,
+      "loss/crossentropy": 1.5598660111427307,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18610762059688568,
+      "step": 1306
+    },
+    {
+      "epoch": 0.21783333333333332,
+      "grad_norm": 32.5,
+      "grad_norm_var": 11.399739583333334,
+      "learning_rate": 8.875825306666968e-05,
+      "loss": 7.2345,
+      "loss/crossentropy": 1.5103994756937027,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1768077164888382,
+      "step": 1307
+    },
+    {
+      "epoch": 0.218,
+      "grad_norm": 29.5,
+      "grad_norm_var": 11.194205729166667,
+      "learning_rate": 8.874170834893272e-05,
+      "loss": 7.214,
+      "loss/crossentropy": 2.020356208086014,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15426049567759037,
+      "step": 1308
+    },
+    {
+      "epoch": 0.21816666666666668,
+      "grad_norm": 29.25,
+      "grad_norm_var": 11.337955729166667,
+      "learning_rate": 8.872515300993669e-05,
+      "loss": 7.2449,
+      "loss/crossentropy": 1.156544879078865,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2502839658409357,
+      "step": 1309
+    },
+    {
+      "epoch": 0.21833333333333332,
+      "grad_norm": 28.75,
+      "grad_norm_var": 11.003580729166666,
+      "learning_rate": 8.870858705422033e-05,
+      "loss": 7.3904,
+      "loss/crossentropy": 1.9905332624912262,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14915032871067524,
+      "step": 1310
+    },
+    {
+      "epoch": 0.2185,
+      "grad_norm": 28.0,
+      "grad_norm_var": 10.906184895833333,
+      "learning_rate": 8.869201048632532e-05,
+      "loss": 7.1195,
+      "loss/crossentropy": 1.7615238428115845,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2768377661705017,
+      "step": 1311
+    },
+    {
+      "epoch": 0.21866666666666668,
+      "grad_norm": 38.25,
+      "grad_norm_var": 13.801822916666667,
+      "learning_rate": 8.867542331079617e-05,
+      "loss": 7.5338,
+      "loss/crossentropy": 1.3941400051116943,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15738719515502453,
+      "step": 1312
+    },
+    {
+      "epoch": 0.21883333333333332,
+      "grad_norm": 31.625,
+      "grad_norm_var": 12.564518229166667,
+      "learning_rate": 8.865882553218037e-05,
+      "loss": 6.9976,
+      "loss/crossentropy": 1.592208281159401,
+      "loss/hidden": 3.73046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19620600156486034,
+      "step": 1313
+    },
+    {
+      "epoch": 0.219,
+      "grad_norm": 30.375,
+      "grad_norm_var": 11.464322916666667,
+      "learning_rate": 8.864221715502829e-05,
+      "loss": 7.2638,
+      "loss/crossentropy": 1.717018038034439,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14654031209647655,
+      "step": 1314
+    },
+    {
+      "epoch": 0.21916666666666668,
+      "grad_norm": 29.0,
+      "grad_norm_var": 11.812955729166667,
+      "learning_rate": 8.862559818389322e-05,
+      "loss": 7.1108,
+      "loss/crossentropy": 1.872963935136795,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17184802889823914,
+      "step": 1315
+    },
+    {
+      "epoch": 0.21933333333333332,
+      "grad_norm": 28.0,
+      "grad_norm_var": 12.422330729166667,
+      "learning_rate": 8.860896862333134e-05,
+      "loss": 6.7489,
+      "loss/crossentropy": 1.442404419183731,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18149691820144653,
+      "step": 1316
+    },
+    {
+      "epoch": 0.2195,
+      "grad_norm": 29.375,
+      "grad_norm_var": 12.211393229166667,
+      "learning_rate": 8.859232847790175e-05,
+      "loss": 7.217,
+      "loss/crossentropy": 1.8750587701797485,
+      "loss/hidden": 3.0546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14721840247511864,
+      "step": 1317
+    },
+    {
+      "epoch": 0.21966666666666668,
+      "grad_norm": 30.875,
+      "grad_norm_var": 11.775455729166667,
+      "learning_rate": 8.857567775216643e-05,
+      "loss": 7.3634,
+      "loss/crossentropy": 1.995005339384079,
+      "loss/hidden": 3.83203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22919053584337234,
+      "step": 1318
+    },
+    {
+      "epoch": 0.21983333333333333,
+      "grad_norm": 39.0,
+      "grad_norm_var": 15.495572916666667,
+      "learning_rate": 8.855901645069026e-05,
+      "loss": 6.8676,
+      "loss/crossentropy": 1.867722600698471,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14688246697187424,
+      "step": 1319
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 31.25,
+      "grad_norm_var": 15.495572916666667,
+      "learning_rate": 8.854234457804105e-05,
+      "loss": 7.1347,
+      "loss/crossentropy": 1.6819516122341156,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1639174222946167,
+      "step": 1320
+    },
+    {
+      "epoch": 0.22016666666666668,
+      "grad_norm": 28.75,
+      "grad_norm_var": 10.751822916666667,
+      "learning_rate": 8.852566213878947e-05,
+      "loss": 6.8401,
+      "loss/crossentropy": 1.3040947020053864,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15195035003125668,
+      "step": 1321
+    },
+    {
+      "epoch": 0.22033333333333333,
+      "grad_norm": 28.125,
+      "grad_norm_var": 11.153580729166666,
+      "learning_rate": 8.850896913750911e-05,
+      "loss": 7.21,
+      "loss/crossentropy": 1.8918107151985168,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19468124955892563,
+      "step": 1322
+    },
+    {
+      "epoch": 0.2205,
+      "grad_norm": 30.125,
+      "grad_norm_var": 10.964322916666667,
+      "learning_rate": 8.849226557877646e-05,
+      "loss": 7.1427,
+      "loss/crossentropy": 1.793198138475418,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1666645836085081,
+      "step": 1323
+    },
+    {
+      "epoch": 0.22066666666666668,
+      "grad_norm": 31.0,
+      "grad_norm_var": 10.876822916666667,
+      "learning_rate": 8.84755514671709e-05,
+      "loss": 7.7351,
+      "loss/crossentropy": 1.8946149945259094,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24298083782196045,
+      "step": 1324
+    },
+    {
+      "epoch": 0.22083333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 10.721875,
+      "learning_rate": 8.845882680727469e-05,
+      "loss": 7.3047,
+      "loss/crossentropy": 1.3693952858448029,
+      "loss/hidden": 3.796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21227648109197617,
+      "step": 1325
+    },
+    {
+      "epoch": 0.221,
+      "grad_norm": 27.625,
+      "grad_norm_var": 11.1150390625,
+      "learning_rate": 8.844209160367299e-05,
+      "loss": 6.9994,
+      "loss/crossentropy": 1.4340689927339554,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14614807814359665,
+      "step": 1326
+    },
+    {
+      "epoch": 0.22116666666666668,
+      "grad_norm": 33.25,
+      "grad_norm_var": 10.8962890625,
+      "learning_rate": 8.842534586095383e-05,
+      "loss": 6.7807,
+      "loss/crossentropy": 1.8267545104026794,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1497194468975067,
+      "step": 1327
+    },
+    {
+      "epoch": 0.22133333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 7.4275390625,
+      "learning_rate": 8.840858958370819e-05,
+      "loss": 7.2799,
+      "loss/crossentropy": 2.064533621072769,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1987077295780182,
+      "step": 1328
+    },
+    {
+      "epoch": 0.2215,
+      "grad_norm": 27.125,
+      "grad_norm_var": 8.1541015625,
+      "learning_rate": 8.839182277652989e-05,
+      "loss": 6.7331,
+      "loss/crossentropy": 1.58609539270401,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14350396767258644,
+      "step": 1329
+    },
+    {
+      "epoch": 0.22166666666666668,
+      "grad_norm": 27.375,
+      "grad_norm_var": 8.7447265625,
+      "learning_rate": 8.837504544401561e-05,
+      "loss": 6.7566,
+      "loss/crossentropy": 1.752264380455017,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16161640733480453,
+      "step": 1330
+    },
+    {
+      "epoch": 0.22183333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 8.637239583333333,
+      "learning_rate": 8.8358257590765e-05,
+      "loss": 6.9655,
+      "loss/crossentropy": 1.5736742317676544,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17882109060883522,
+      "step": 1331
+    },
+    {
+      "epoch": 0.222,
+      "grad_norm": 29.75,
+      "grad_norm_var": 8.278125,
+      "learning_rate": 8.834145922138049e-05,
+      "loss": 7.1689,
+      "loss/crossentropy": 2.366013765335083,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20151379331946373,
+      "step": 1332
+    },
+    {
+      "epoch": 0.22216666666666668,
+      "grad_norm": 27.625,
+      "grad_norm_var": 8.724739583333333,
+      "learning_rate": 8.832465034046749e-05,
+      "loss": 7.1076,
+      "loss/crossentropy": 1.8252016454935074,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19100939109921455,
+      "step": 1333
+    },
+    {
+      "epoch": 0.22233333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 9.335872395833333,
+      "learning_rate": 8.830783095263425e-05,
+      "loss": 6.6759,
+      "loss/crossentropy": 1.0002970695495605,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1451659481972456,
+      "step": 1334
+    },
+    {
+      "epoch": 0.2225,
+      "grad_norm": 31.0,
+      "grad_norm_var": 4.294205729166666,
+      "learning_rate": 8.829100106249189e-05,
+      "loss": 6.8492,
+      "loss/crossentropy": 1.390068769454956,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16063359007239342,
+      "step": 1335
+    },
+    {
+      "epoch": 0.22266666666666668,
+      "grad_norm": 30.0,
+      "grad_norm_var": 4.187434895833333,
+      "learning_rate": 8.827416067465441e-05,
+      "loss": 6.8579,
+      "loss/crossentropy": 1.1625606268644333,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23727696016430855,
+      "step": 1336
+    },
+    {
+      "epoch": 0.22283333333333333,
+      "grad_norm": 28.125,
+      "grad_norm_var": 4.311458333333333,
+      "learning_rate": 8.825730979373872e-05,
+      "loss": 6.8081,
+      "loss/crossentropy": 1.1123109608888626,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1793264076113701,
+      "step": 1337
+    },
+    {
+      "epoch": 0.223,
+      "grad_norm": 29.5,
+      "grad_norm_var": 4.103059895833334,
+      "learning_rate": 8.824044842436456e-05,
+      "loss": 7.0354,
+      "loss/crossentropy": 1.8462108373641968,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19455980882048607,
+      "step": 1338
+    },
+    {
+      "epoch": 0.22316666666666668,
+      "grad_norm": 29.5,
+      "grad_norm_var": 4.11640625,
+      "learning_rate": 8.822357657115459e-05,
+      "loss": 7.1256,
+      "loss/crossentropy": 1.12643700838089,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12856785394251347,
+      "step": 1339
+    },
+    {
+      "epoch": 0.22333333333333333,
+      "grad_norm": 35.0,
+      "grad_norm_var": 5.674739583333333,
+      "learning_rate": 8.82066942387343e-05,
+      "loss": 7.0614,
+      "loss/crossentropy": 1.6229997724294662,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17172613367438316,
+      "step": 1340
+    },
+    {
+      "epoch": 0.2235,
+      "grad_norm": 36.0,
+      "grad_norm_var": 7.768489583333333,
+      "learning_rate": 8.818980143173213e-05,
+      "loss": 7.0696,
+      "loss/crossentropy": 1.7989262640476227,
+      "loss/hidden": 3.7734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1843792088329792,
+      "step": 1341
+    },
+    {
+      "epoch": 0.22366666666666668,
+      "grad_norm": 30.75,
+      "grad_norm_var": 7.1744140625,
+      "learning_rate": 8.817289815477928e-05,
+      "loss": 7.431,
+      "loss/crossentropy": 1.507471725344658,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15506815910339355,
+      "step": 1342
+    },
+    {
+      "epoch": 0.22383333333333333,
+      "grad_norm": 28.75,
+      "grad_norm_var": 6.9166015625,
+      "learning_rate": 8.815598441250987e-05,
+      "loss": 6.9046,
+      "loss/crossentropy": 1.2516491413116455,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13933237828314304,
+      "step": 1343
+    },
+    {
+      "epoch": 0.224,
+      "grad_norm": 27.75,
+      "grad_norm_var": 7.0900390625,
+      "learning_rate": 8.813906020956097e-05,
+      "loss": 7.0237,
+      "loss/crossentropy": 1.6177708208560944,
+      "loss/hidden": 3.7734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30907687544822693,
+      "step": 1344
+    },
+    {
+      "epoch": 0.22416666666666665,
+      "grad_norm": 31.375,
+      "grad_norm_var": 6.5056640625,
+      "learning_rate": 8.81221255505724e-05,
+      "loss": 6.8969,
+      "loss/crossentropy": 1.5588159263134003,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17392415553331375,
+      "step": 1345
+    },
+    {
+      "epoch": 0.22433333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 5.857291666666667,
+      "learning_rate": 8.810518044018689e-05,
+      "loss": 7.1614,
+      "loss/crossentropy": 1.5660995543003082,
+      "loss/hidden": 3.7578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1824370063841343,
+      "step": 1346
+    },
+    {
+      "epoch": 0.2245,
+      "grad_norm": 29.5,
+      "grad_norm_var": 5.931705729166667,
+      "learning_rate": 8.808822488305005e-05,
+      "loss": 7.4345,
+      "loss/crossentropy": 1.461399108171463,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1693030335009098,
+      "step": 1347
+    },
+    {
+      "epoch": 0.22466666666666665,
+      "grad_norm": 31.5,
+      "grad_norm_var": 5.942643229166666,
+      "learning_rate": 8.807125888381035e-05,
+      "loss": 7.3317,
+      "loss/crossentropy": 1.4001652300357819,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15863394737243652,
+      "step": 1348
+    },
+    {
+      "epoch": 0.22483333333333333,
+      "grad_norm": 36.25,
+      "grad_norm_var": 7.133072916666666,
+      "learning_rate": 8.80542824471191e-05,
+      "loss": 7.4203,
+      "loss/crossentropy": 2.1138390600681305,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18512402847409248,
+      "step": 1349
+    },
+    {
+      "epoch": 0.225,
+      "grad_norm": 29.25,
+      "grad_norm_var": 6.942708333333333,
+      "learning_rate": 8.803729557763047e-05,
+      "loss": 6.9968,
+      "loss/crossentropy": 1.7467091381549835,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19800013676285744,
+      "step": 1350
+    },
+    {
+      "epoch": 0.22516666666666665,
+      "grad_norm": 29.25,
+      "grad_norm_var": 7.112239583333333,
+      "learning_rate": 8.802029828000156e-05,
+      "loss": 7.1186,
+      "loss/crossentropy": 1.9953233003616333,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2265888750553131,
+      "step": 1351
+    },
+    {
+      "epoch": 0.22533333333333333,
+      "grad_norm": 28.5,
+      "grad_norm_var": 7.412239583333333,
+      "learning_rate": 8.800329055889223e-05,
+      "loss": 7.0946,
+      "loss/crossentropy": 1.5274888277053833,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20411700941622257,
+      "step": 1352
+    },
+    {
+      "epoch": 0.2255,
+      "grad_norm": 26.875,
+      "grad_norm_var": 7.939583333333333,
+      "learning_rate": 8.798627241896524e-05,
+      "loss": 7.0565,
+      "loss/crossentropy": 1.921573668718338,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16989030689001083,
+      "step": 1353
+    },
+    {
+      "epoch": 0.22566666666666665,
+      "grad_norm": 27.625,
+      "grad_norm_var": 8.440559895833333,
+      "learning_rate": 8.796924386488624e-05,
+      "loss": 6.7477,
+      "loss/crossentropy": 1.69273541867733,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16533562913537025,
+      "step": 1354
+    },
+    {
+      "epoch": 0.22583333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 8.670833333333333,
+      "learning_rate": 8.795220490132369e-05,
+      "loss": 7.1876,
+      "loss/crossentropy": 2.4070510268211365,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17891918495297432,
+      "step": 1355
+    },
+    {
+      "epoch": 0.226,
+      "grad_norm": 31.0,
+      "grad_norm_var": 7.2375,
+      "learning_rate": 8.793515553294891e-05,
+      "loss": 6.8497,
+      "loss/crossentropy": 1.9144990742206573,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21343015506863594,
+      "step": 1356
+    },
+    {
+      "epoch": 0.22616666666666665,
+      "grad_norm": 37.0,
+      "grad_norm_var": 8.075,
+      "learning_rate": 8.79180957644361e-05,
+      "loss": 7.1019,
+      "loss/crossentropy": 2.0029281973838806,
+      "loss/hidden": 3.828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2547861821949482,
+      "step": 1357
+    },
+    {
+      "epoch": 0.22633333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 8.16015625,
+      "learning_rate": 8.790102560046227e-05,
+      "loss": 7.2401,
+      "loss/crossentropy": 1.71112260222435,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18296929448843002,
+      "step": 1358
+    },
+    {
+      "epoch": 0.2265,
+      "grad_norm": 29.5,
+      "grad_norm_var": 8.040625,
+      "learning_rate": 8.788394504570732e-05,
+      "loss": 7.5036,
+      "loss/crossentropy": 1.8826730847358704,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22693874314427376,
+      "step": 1359
+    },
+    {
+      "epoch": 0.22666666666666666,
+      "grad_norm": 28.625,
+      "grad_norm_var": 7.785872395833334,
+      "learning_rate": 8.786685410485398e-05,
+      "loss": 6.9652,
+      "loss/crossentropy": 1.706690937280655,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16132831200957298,
+      "step": 1360
+    },
+    {
+      "epoch": 0.22683333333333333,
+      "grad_norm": 26.625,
+      "grad_norm_var": 8.5775390625,
+      "learning_rate": 8.784975278258783e-05,
+      "loss": 7.0589,
+      "loss/crossentropy": 1.862322524189949,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2016892172396183,
+      "step": 1361
+    },
+    {
+      "epoch": 0.227,
+      "grad_norm": 29.875,
+      "grad_norm_var": 8.57890625,
+      "learning_rate": 8.783264108359728e-05,
+      "loss": 7.1565,
+      "loss/crossentropy": 1.7877943962812424,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20448906905949116,
+      "step": 1362
+    },
+    {
+      "epoch": 0.22716666666666666,
+      "grad_norm": 29.625,
+      "grad_norm_var": 8.570247395833333,
+      "learning_rate": 8.78155190125736e-05,
+      "loss": 7.4146,
+      "loss/crossentropy": 1.9530489146709442,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2042655497789383,
+      "step": 1363
+    },
+    {
+      "epoch": 0.22733333333333333,
+      "grad_norm": 36.75,
+      "grad_norm_var": 11.282747395833333,
+      "learning_rate": 8.779838657421092e-05,
+      "loss": 7.2438,
+      "loss/crossentropy": 2.173504024744034,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20246214047074318,
+      "step": 1364
+    },
+    {
+      "epoch": 0.2275,
+      "grad_norm": 30.375,
+      "grad_norm_var": 8.868489583333334,
+      "learning_rate": 8.778124377320618e-05,
+      "loss": 6.9205,
+      "loss/crossentropy": 1.4049129486083984,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17277532443404198,
+      "step": 1365
+    },
+    {
+      "epoch": 0.22766666666666666,
+      "grad_norm": 27.375,
+      "grad_norm_var": 9.287434895833334,
+      "learning_rate": 8.776409061425919e-05,
+      "loss": 6.9055,
+      "loss/crossentropy": 1.3801761269569397,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14728276431560516,
+      "step": 1366
+    },
+    {
+      "epoch": 0.22783333333333333,
+      "grad_norm": 28.5,
+      "grad_norm_var": 9.390559895833333,
+      "learning_rate": 8.774692710207257e-05,
+      "loss": 6.8519,
+      "loss/crossentropy": 1.3856933414936066,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21369173377752304,
+      "step": 1367
+    },
+    {
+      "epoch": 0.228,
+      "grad_norm": 29.5,
+      "grad_norm_var": 9.268684895833333,
+      "learning_rate": 8.772975324135179e-05,
+      "loss": 6.6001,
+      "loss/crossentropy": 1.3469049483537674,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13380655460059643,
+      "step": 1368
+    },
+    {
+      "epoch": 0.22816666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 8.673958333333333,
+      "learning_rate": 8.771256903680519e-05,
+      "loss": 6.8957,
+      "loss/crossentropy": 1.1714013069868088,
+      "loss/hidden": 3.7578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20499271154403687,
+      "step": 1369
+    },
+    {
+      "epoch": 0.22833333333333333,
+      "grad_norm": 28.5,
+      "grad_norm_var": 8.419205729166666,
+      "learning_rate": 8.769537449314391e-05,
+      "loss": 6.8322,
+      "loss/crossentropy": 1.8354635387659073,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16002275049686432,
+      "step": 1370
+    },
+    {
+      "epoch": 0.2285,
+      "grad_norm": 28.875,
+      "grad_norm_var": 8.308268229166666,
+      "learning_rate": 8.76781696150819e-05,
+      "loss": 6.8736,
+      "loss/crossentropy": 1.9772877991199493,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1661224588751793,
+      "step": 1371
+    },
+    {
+      "epoch": 0.22866666666666666,
+      "grad_norm": 28.75,
+      "grad_norm_var": 8.416080729166667,
+      "learning_rate": 8.766095440733601e-05,
+      "loss": 6.8107,
+      "loss/crossentropy": 1.4746856093406677,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1853000782430172,
+      "step": 1372
+    },
+    {
+      "epoch": 0.22883333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 5.877018229166667,
+      "learning_rate": 8.764372887462586e-05,
+      "loss": 6.9663,
+      "loss/crossentropy": 1.777550995349884,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17858651094138622,
+      "step": 1373
+    },
+    {
+      "epoch": 0.229,
+      "grad_norm": 29.0,
+      "grad_norm_var": 5.744205729166667,
+      "learning_rate": 8.762649302167395e-05,
+      "loss": 6.8985,
+      "loss/crossentropy": 1.3290674686431885,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15025487542152405,
+      "step": 1374
+    },
+    {
+      "epoch": 0.22916666666666666,
+      "grad_norm": 28.0,
+      "grad_norm_var": 5.939518229166667,
+      "learning_rate": 8.760924685320557e-05,
+      "loss": 7.1549,
+      "loss/crossentropy": 1.9009324610233307,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17355919256806374,
+      "step": 1375
+    },
+    {
+      "epoch": 0.22933333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 5.861393229166667,
+      "learning_rate": 8.759199037394887e-05,
+      "loss": 7.1262,
+      "loss/crossentropy": 2.2094116806983948,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1695697419345379,
+      "step": 1376
+    },
+    {
+      "epoch": 0.2295,
+      "grad_norm": 29.875,
+      "grad_norm_var": 5.163997395833333,
+      "learning_rate": 8.757472358863481e-05,
+      "loss": 7.0997,
+      "loss/crossentropy": 2.250872015953064,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20045072957873344,
+      "step": 1377
+    },
+    {
+      "epoch": 0.22966666666666666,
+      "grad_norm": 27.125,
+      "grad_norm_var": 5.6681640625,
+      "learning_rate": 8.755744650199716e-05,
+      "loss": 6.8707,
+      "loss/crossentropy": 1.9199231564998627,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1821548417210579,
+      "step": 1378
+    },
+    {
+      "epoch": 0.22983333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 5.6775390625,
+      "learning_rate": 8.754015911877255e-05,
+      "loss": 7.343,
+      "loss/crossentropy": 1.9021850228309631,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2082943543791771,
+      "step": 1379
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.2455729166666667,
+      "learning_rate": 8.752286144370041e-05,
+      "loss": 7.3073,
+      "loss/crossentropy": 1.9341810941696167,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17061935551464558,
+      "step": 1380
+    },
+    {
+      "epoch": 0.23016666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.164322916666667,
+      "learning_rate": 8.750555348152298e-05,
+      "loss": 7.2699,
+      "loss/crossentropy": 1.358662635087967,
+      "loss/hidden": 3.859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2191124390810728,
+      "step": 1381
+    },
+    {
+      "epoch": 0.23033333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.978125,
+      "learning_rate": 8.748823523698535e-05,
+      "loss": 7.388,
+      "loss/crossentropy": 1.9777554273605347,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.265559408813715,
+      "step": 1382
+    },
+    {
+      "epoch": 0.2305,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.39765625,
+      "learning_rate": 8.747090671483542e-05,
+      "loss": 7.2046,
+      "loss/crossentropy": 1.8751341998577118,
+      "loss/hidden": 3.73828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23485557362437248,
+      "step": 1383
+    },
+    {
+      "epoch": 0.23066666666666666,
+      "grad_norm": 34.75,
+      "grad_norm_var": 4.978125,
+      "learning_rate": 8.745356791982391e-05,
+      "loss": 7.1845,
+      "loss/crossentropy": 1.5384764969348907,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18676547519862652,
+      "step": 1384
+    },
+    {
+      "epoch": 0.23083333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.9125,
+      "learning_rate": 8.74362188567043e-05,
+      "loss": 6.8499,
+      "loss/crossentropy": 1.632831186056137,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14724744111299515,
+      "step": 1385
+    },
+    {
+      "epoch": 0.231,
+      "grad_norm": 28.625,
+      "grad_norm_var": 4.8884765625,
+      "learning_rate": 8.741885953023301e-05,
+      "loss": 7.058,
+      "loss/crossentropy": 1.5368026793003082,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19731464236974716,
+      "step": 1386
+    },
+    {
+      "epoch": 0.23116666666666666,
+      "grad_norm": 28.0,
+      "grad_norm_var": 5.068489583333333,
+      "learning_rate": 8.740148994516912e-05,
+      "loss": 7.3183,
+      "loss/crossentropy": 1.9754838049411774,
+      "loss/hidden": 3.71484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2648444287478924,
+      "step": 1387
+    },
+    {
+      "epoch": 0.23133333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 4.965625,
+      "learning_rate": 8.738411010627466e-05,
+      "loss": 7.2079,
+      "loss/crossentropy": 1.6709902584552765,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1980009637773037,
+      "step": 1388
+    },
+    {
+      "epoch": 0.2315,
+      "grad_norm": 29.5,
+      "grad_norm_var": 4.23515625,
+      "learning_rate": 8.736672001831438e-05,
+      "loss": 6.896,
+      "loss/crossentropy": 1.4954399168491364,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12558046355843544,
+      "step": 1389
+    },
+    {
+      "epoch": 0.23166666666666666,
+      "grad_norm": 29.125,
+      "grad_norm_var": 4.2228515625,
+      "learning_rate": 8.734931968605589e-05,
+      "loss": 7.1168,
+      "loss/crossentropy": 1.7655189633369446,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30897774174809456,
+      "step": 1390
+    },
+    {
+      "epoch": 0.23183333333333334,
+      "grad_norm": 28.25,
+      "grad_norm_var": 4.1666015625,
+      "learning_rate": 8.733190911426958e-05,
+      "loss": 7.2112,
+      "loss/crossentropy": 2.017255276441574,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19111590459942818,
+      "step": 1391
+    },
+    {
+      "epoch": 0.232,
+      "grad_norm": 29.125,
+      "grad_norm_var": 4.1962890625,
+      "learning_rate": 8.731448830772864e-05,
+      "loss": 6.8487,
+      "loss/crossentropy": 1.6656021475791931,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1857664603739977,
+      "step": 1392
+    },
+    {
+      "epoch": 0.23216666666666666,
+      "grad_norm": 28.375,
+      "grad_norm_var": 4.3166015625,
+      "learning_rate": 8.729705727120911e-05,
+      "loss": 7.4448,
+      "loss/crossentropy": 1.529506891965866,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1576174721121788,
+      "step": 1393
+    },
+    {
+      "epoch": 0.23233333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.853125,
+      "learning_rate": 8.72796160094898e-05,
+      "loss": 7.5149,
+      "loss/crossentropy": 1.2481360882520676,
+      "loss/hidden": 3.9609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26495122350752354,
+      "step": 1394
+    },
+    {
+      "epoch": 0.2325,
+      "grad_norm": 36.0,
+      "grad_norm_var": 6.1822265625,
+      "learning_rate": 8.726216452735232e-05,
+      "loss": 7.2105,
+      "loss/crossentropy": 1.9443733096122742,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20606354251503944,
+      "step": 1395
+    },
+    {
+      "epoch": 0.23266666666666666,
+      "grad_norm": 29.25,
+      "grad_norm_var": 6.070572916666666,
+      "learning_rate": 8.724470282958111e-05,
+      "loss": 6.9416,
+      "loss/crossentropy": 1.9610607624053955,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1985795944929123,
+      "step": 1396
+    },
+    {
+      "epoch": 0.23283333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 5.934830729166666,
+      "learning_rate": 8.722723092096338e-05,
+      "loss": 7.2134,
+      "loss/crossentropy": 1.5050616040825844,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.230638038367033,
+      "step": 1397
+    },
+    {
+      "epoch": 0.233,
+      "grad_norm": 29.375,
+      "grad_norm_var": 5.998893229166667,
+      "learning_rate": 8.720974880628916e-05,
+      "loss": 7.1159,
+      "loss/crossentropy": 1.1800313591957092,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2613990902900696,
+      "step": 1398
+    },
+    {
+      "epoch": 0.23316666666666666,
+      "grad_norm": 27.0,
+      "grad_norm_var": 5.5306640625,
+      "learning_rate": 8.719225649035126e-05,
+      "loss": 7.1222,
+      "loss/crossentropy": 1.548551693558693,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19349294528365135,
+      "step": 1399
+    },
+    {
+      "epoch": 0.23333333333333334,
+      "grad_norm": 35.5,
+      "grad_norm_var": 6.0494140625,
+      "learning_rate": 8.717475397794531e-05,
+      "loss": 6.9024,
+      "loss/crossentropy": 1.6936389207839966,
+      "loss/hidden": 3.6484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19783683493733406,
+      "step": 1400
+    },
+    {
+      "epoch": 0.2335,
+      "grad_norm": 30.25,
+      "grad_norm_var": 6.012434895833334,
+      "learning_rate": 8.715724127386972e-05,
+      "loss": 7.1538,
+      "loss/crossentropy": 1.5932972878217697,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16691654548048973,
+      "step": 1401
+    },
+    {
+      "epoch": 0.23366666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 6.203125,
+      "learning_rate": 8.713971838292569e-05,
+      "loss": 7.171,
+      "loss/crossentropy": 1.684373527765274,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20065860822796822,
+      "step": 1402
+    },
+    {
+      "epoch": 0.23383333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 5.9125,
+      "learning_rate": 8.712218530991723e-05,
+      "loss": 7.276,
+      "loss/crossentropy": 1.7293904423713684,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16419371217489243,
+      "step": 1403
+    },
+    {
+      "epoch": 0.234,
+      "grad_norm": 28.0,
+      "grad_norm_var": 6.229166666666667,
+      "learning_rate": 8.710464205965112e-05,
+      "loss": 7.0833,
+      "loss/crossentropy": 1.6907464861869812,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1767747402191162,
+      "step": 1404
+    },
+    {
+      "epoch": 0.23416666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 6.203125,
+      "learning_rate": 8.708708863693697e-05,
+      "loss": 7.1463,
+      "loss/crossentropy": 1.6274442970752716,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2582891583442688,
+      "step": 1405
+    },
+    {
+      "epoch": 0.23433333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 6.1447265625,
+      "learning_rate": 8.706952504658712e-05,
+      "loss": 7.0599,
+      "loss/crossentropy": 1.5662869662046432,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1793067865073681,
+      "step": 1406
+    },
+    {
+      "epoch": 0.2345,
+      "grad_norm": 30.625,
+      "grad_norm_var": 5.861458333333333,
+      "learning_rate": 8.705195129341672e-05,
+      "loss": 7.2891,
+      "loss/crossentropy": 1.635928362607956,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13916829228401184,
+      "step": 1407
+    },
+    {
+      "epoch": 0.23466666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 5.908072916666667,
+      "learning_rate": 8.703436738224375e-05,
+      "loss": 6.9642,
+      "loss/crossentropy": 1.8249442875385284,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18701985105872154,
+      "step": 1408
+    },
+    {
+      "epoch": 0.23483333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 5.623893229166667,
+      "learning_rate": 8.701677331788891e-05,
+      "loss": 7.2283,
+      "loss/crossentropy": 1.5654152929782867,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.247907355427742,
+      "step": 1409
+    },
+    {
+      "epoch": 0.235,
+      "grad_norm": 31.375,
+      "grad_norm_var": 5.624739583333334,
+      "learning_rate": 8.699916910517573e-05,
+      "loss": 7.1302,
+      "loss/crossentropy": 1.9997363090515137,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20151901245117188,
+      "step": 1410
+    },
+    {
+      "epoch": 0.23516666666666666,
+      "grad_norm": 36.5,
+      "grad_norm_var": 5.999739583333334,
+      "learning_rate": 8.69815547489305e-05,
+      "loss": 6.9722,
+      "loss/crossentropy": 1.9183299839496613,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1610793862491846,
+      "step": 1411
+    },
+    {
+      "epoch": 0.23533333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 5.898958333333334,
+      "learning_rate": 8.696393025398229e-05,
+      "loss": 6.8754,
+      "loss/crossentropy": 1.6777728497982025,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1674475595355034,
+      "step": 1412
+    },
+    {
+      "epoch": 0.2355,
+      "grad_norm": 29.375,
+      "grad_norm_var": 6.0228515625,
+      "learning_rate": 8.694629562516294e-05,
+      "loss": 7.1442,
+      "loss/crossentropy": 1.218498095870018,
+      "loss/hidden": 4.015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21704785898327827,
+      "step": 1413
+    },
+    {
+      "epoch": 0.23566666666666666,
+      "grad_norm": 27.25,
+      "grad_norm_var": 6.679166666666666,
+      "learning_rate": 8.692865086730713e-05,
+      "loss": 7.003,
+      "loss/crossentropy": 1.756615787744522,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20079421997070312,
+      "step": 1414
+    },
+    {
+      "epoch": 0.23583333333333334,
+      "grad_norm": 27.625,
+      "grad_norm_var": 6.406705729166666,
+      "learning_rate": 8.69109959852522e-05,
+      "loss": 7.0626,
+      "loss/crossentropy": 2.1831423342227936,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20736485719680786,
+      "step": 1415
+    },
+    {
+      "epoch": 0.236,
+      "grad_norm": 29.875,
+      "grad_norm_var": 4.710416666666666,
+      "learning_rate": 8.689333098383842e-05,
+      "loss": 7.1615,
+      "loss/crossentropy": 1.6375371515750885,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16935132816433907,
+      "step": 1416
+    },
+    {
+      "epoch": 0.23616666666666666,
+      "grad_norm": 29.25,
+      "grad_norm_var": 4.772916666666666,
+      "learning_rate": 8.68756558679087e-05,
+      "loss": 7.1256,
+      "loss/crossentropy": 1.921682059764862,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20466305688023567,
+      "step": 1417
+    },
+    {
+      "epoch": 0.23633333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 4.481705729166666,
+      "learning_rate": 8.685797064230878e-05,
+      "loss": 7.2413,
+      "loss/crossentropy": 1.344342216849327,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22074035368859768,
+      "step": 1418
+    },
+    {
+      "epoch": 0.2365,
+      "grad_norm": 28.375,
+      "grad_norm_var": 4.639322916666667,
+      "learning_rate": 8.684027531188717e-05,
+      "loss": 6.9632,
+      "loss/crossentropy": 1.786535769701004,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16699090600013733,
+      "step": 1419
+    },
+    {
+      "epoch": 0.23666666666666666,
+      "grad_norm": 29.125,
+      "grad_norm_var": 4.425455729166667,
+      "learning_rate": 8.682256988149513e-05,
+      "loss": 7.1572,
+      "loss/crossentropy": 1.7073152661323547,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2525929920375347,
+      "step": 1420
+    },
+    {
+      "epoch": 0.23683333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 4.430143229166666,
+      "learning_rate": 8.680485435598673e-05,
+      "loss": 7.1865,
+      "loss/crossentropy": 1.6526244580745697,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1847163327038288,
+      "step": 1421
+    },
+    {
+      "epoch": 0.237,
+      "grad_norm": 28.125,
+      "grad_norm_var": 4.601041666666666,
+      "learning_rate": 8.678712874021874e-05,
+      "loss": 7.184,
+      "loss/crossentropy": 1.9928257465362549,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21139519289135933,
+      "step": 1422
+    },
+    {
+      "epoch": 0.23716666666666666,
+      "grad_norm": 28.625,
+      "grad_norm_var": 4.642708333333333,
+      "learning_rate": 8.67693930390508e-05,
+      "loss": 7.0479,
+      "loss/crossentropy": 1.5828899145126343,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1637626551091671,
+      "step": 1423
+    },
+    {
+      "epoch": 0.23733333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 4.606184895833334,
+      "learning_rate": 8.67516472573452e-05,
+      "loss": 7.3309,
+      "loss/crossentropy": 2.038231670856476,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15858252346515656,
+      "step": 1424
+    },
+    {
+      "epoch": 0.2375,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.602083333333334,
+      "learning_rate": 8.673389139996708e-05,
+      "loss": 6.7987,
+      "loss/crossentropy": 1.5226624310016632,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13311301730573177,
+      "step": 1425
+    },
+    {
+      "epoch": 0.23766666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 4.422330729166666,
+      "learning_rate": 8.671612547178428e-05,
+      "loss": 7.3592,
+      "loss/crossentropy": 1.7062960714101791,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25913639180362225,
+      "step": 1426
+    },
+    {
+      "epoch": 0.23783333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.1384765625,
+      "learning_rate": 8.669834947766746e-05,
+      "loss": 7.5093,
+      "loss/crossentropy": 1.1424612551927567,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16857817955315113,
+      "step": 1427
+    },
+    {
+      "epoch": 0.238,
+      "grad_norm": 29.25,
+      "grad_norm_var": 0.7822265625,
+      "learning_rate": 8.668056342248998e-05,
+      "loss": 7.3723,
+      "loss/crossentropy": 1.6899997889995575,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20195226930081844,
+      "step": 1428
+    },
+    {
+      "epoch": 0.23816666666666667,
+      "grad_norm": 27.75,
+      "grad_norm_var": 0.89140625,
+      "learning_rate": 8.666276731112801e-05,
+      "loss": 6.9453,
+      "loss/crossentropy": 1.98584845662117,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15120555087924004,
+      "step": 1429
+    },
+    {
+      "epoch": 0.23833333333333334,
+      "grad_norm": 27.75,
+      "grad_norm_var": 0.7893229166666667,
+      "learning_rate": 8.664496114846044e-05,
+      "loss": 6.9196,
+      "loss/crossentropy": 2.0655926167964935,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18351034075021744,
+      "step": 1430
+    },
+    {
+      "epoch": 0.2385,
+      "grad_norm": 29.25,
+      "grad_norm_var": 0.6462890625,
+      "learning_rate": 8.662714493936895e-05,
+      "loss": 7.362,
+      "loss/crossentropy": 1.6687164604663849,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21955528669059277,
+      "step": 1431
+    },
+    {
+      "epoch": 0.23866666666666667,
+      "grad_norm": 28.25,
+      "grad_norm_var": 0.65390625,
+      "learning_rate": 8.660931868873793e-05,
+      "loss": 6.9246,
+      "loss/crossentropy": 1.4787462428212166,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19696292839944363,
+      "step": 1432
+    },
+    {
+      "epoch": 0.23883333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 0.6728515625,
+      "learning_rate": 8.659148240145456e-05,
+      "loss": 7.0976,
+      "loss/crossentropy": 1.5287677347660065,
+      "loss/hidden": 4.1171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22544145584106445,
+      "step": 1433
+    },
+    {
+      "epoch": 0.239,
+      "grad_norm": 30.375,
+      "grad_norm_var": 0.7634765625,
+      "learning_rate": 8.657363608240876e-05,
+      "loss": 7.1543,
+      "loss/crossentropy": 1.9106423556804657,
+      "loss/hidden": 3.6484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2270779311656952,
+      "step": 1434
+    },
+    {
+      "epoch": 0.23916666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 0.73515625,
+      "learning_rate": 8.655577973649321e-05,
+      "loss": 7.0022,
+      "loss/crossentropy": 2.2625996470451355,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19603535532951355,
+      "step": 1435
+    },
+    {
+      "epoch": 0.23933333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 0.7358723958333333,
+      "learning_rate": 8.653791336860331e-05,
+      "loss": 7.0308,
+      "loss/crossentropy": 1.6555210202932358,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20859287679195404,
+      "step": 1436
+    },
+    {
+      "epoch": 0.2395,
+      "grad_norm": 29.625,
+      "grad_norm_var": 0.7268229166666667,
+      "learning_rate": 8.652003698363724e-05,
+      "loss": 7.385,
+      "loss/crossentropy": 1.9578928053379059,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19808416441082954,
+      "step": 1437
+    },
+    {
+      "epoch": 0.23966666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 0.6552083333333333,
+      "learning_rate": 8.65021505864959e-05,
+      "loss": 6.9802,
+      "loss/crossentropy": 1.7287716567516327,
+      "loss/hidden": 3.04296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13078411109745502,
+      "step": 1438
+    },
+    {
+      "epoch": 0.23983333333333334,
+      "grad_norm": 27.875,
+      "grad_norm_var": 0.7497395833333333,
+      "learning_rate": 8.648425418208294e-05,
+      "loss": 6.2799,
+      "loss/crossentropy": 1.2794785052537918,
+      "loss/hidden": 3.12109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10539323464035988,
+      "step": 1439
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 28.5,
+      "grad_norm_var": 0.6895833333333333,
+      "learning_rate": 8.64663477753048e-05,
+      "loss": 6.9211,
+      "loss/crossentropy": 1.483281522989273,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20138464868068695,
+      "step": 1440
+    },
+    {
+      "epoch": 0.24016666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 0.7139973958333333,
+      "learning_rate": 8.644843137107059e-05,
+      "loss": 7.3511,
+      "loss/crossentropy": 1.6645758748054504,
+      "loss/hidden": 3.86328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19149157777428627,
+      "step": 1441
+    },
+    {
+      "epoch": 0.24033333333333334,
+      "grad_norm": 26.875,
+      "grad_norm_var": 0.9434895833333333,
+      "learning_rate": 8.64305049742922e-05,
+      "loss": 6.9328,
+      "loss/crossentropy": 1.7084144353866577,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20016243122518063,
+      "step": 1442
+    },
+    {
+      "epoch": 0.2405,
+      "grad_norm": 28.875,
+      "grad_norm_var": 0.8561848958333333,
+      "learning_rate": 8.641256858988424e-05,
+      "loss": 7.3646,
+      "loss/crossentropy": 2.013769954442978,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23251285776495934,
+      "step": 1443
+    },
+    {
+      "epoch": 0.24066666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.1478515625,
+      "learning_rate": 8.639462222276409e-05,
+      "loss": 7.2612,
+      "loss/crossentropy": 1.9536723494529724,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2065252885222435,
+      "step": 1444
+    },
+    {
+      "epoch": 0.24083333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.0525390625,
+      "learning_rate": 8.637666587785184e-05,
+      "loss": 7.1018,
+      "loss/crossentropy": 1.3268489837646484,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15171853080391884,
+      "step": 1445
+    },
+    {
+      "epoch": 0.241,
+      "grad_norm": 28.625,
+      "grad_norm_var": 0.9518229166666666,
+      "learning_rate": 8.635869956007034e-05,
+      "loss": 7.103,
+      "loss/crossentropy": 1.5661400854587555,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1982056424021721,
+      "step": 1446
+    },
+    {
+      "epoch": 0.24116666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.0197265625,
+      "learning_rate": 8.634072327434515e-05,
+      "loss": 7.0494,
+      "loss/crossentropy": 1.7641910314559937,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17389076203107834,
+      "step": 1447
+    },
+    {
+      "epoch": 0.24133333333333334,
+      "grad_norm": 29.125,
+      "grad_norm_var": 0.9645833333333333,
+      "learning_rate": 8.632273702560456e-05,
+      "loss": 7.1942,
+      "loss/crossentropy": 1.8309511989355087,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1622365154325962,
+      "step": 1448
+    },
+    {
+      "epoch": 0.2415,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.0254557291666666,
+      "learning_rate": 8.630474081877959e-05,
+      "loss": 7.0812,
+      "loss/crossentropy": 1.5874554812908173,
+      "loss/hidden": 3.8046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1834915168583393,
+      "step": 1449
+    },
+    {
+      "epoch": 0.24166666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.1176432291666667,
+      "learning_rate": 8.628673465880404e-05,
+      "loss": 7.0935,
+      "loss/crossentropy": 1.8441435992717743,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2328551672399044,
+      "step": 1450
+    },
+    {
+      "epoch": 0.24183333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.4770182291666667,
+      "learning_rate": 8.626871855061438e-05,
+      "loss": 7.7535,
+      "loss/crossentropy": 1.5417988896369934,
+      "loss/hidden": 3.74609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22341690957546234,
+      "step": 1451
+    },
+    {
+      "epoch": 0.242,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.5020833333333334,
+      "learning_rate": 8.625069249914983e-05,
+      "loss": 6.9252,
+      "loss/crossentropy": 1.7583518028259277,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18824856542050838,
+      "step": 1452
+    },
+    {
+      "epoch": 0.24216666666666667,
+      "grad_norm": 27.75,
+      "grad_norm_var": 1.6905598958333334,
+      "learning_rate": 8.623265650935234e-05,
+      "loss": 7.0858,
+      "loss/crossentropy": 1.8695081770420074,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19029086455702782,
+      "step": 1453
+    },
+    {
+      "epoch": 0.24233333333333335,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.6947265625,
+      "learning_rate": 8.621461058616656e-05,
+      "loss": 7.2882,
+      "loss/crossentropy": 1.8658898174762726,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17425318621098995,
+      "step": 1454
+    },
+    {
+      "epoch": 0.2425,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.6488932291666667,
+      "learning_rate": 8.61965547345399e-05,
+      "loss": 7.2973,
+      "loss/crossentropy": 2.263028144836426,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19143683463335037,
+      "step": 1455
+    },
+    {
+      "epoch": 0.24266666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.6343098958333333,
+      "learning_rate": 8.617848895942247e-05,
+      "loss": 7.3989,
+      "loss/crossentropy": 1.6578317731618881,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22550906985998154,
+      "step": 1456
+    },
+    {
+      "epoch": 0.24283333333333335,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.6436848958333334,
+      "learning_rate": 8.616041326576711e-05,
+      "loss": 7.0332,
+      "loss/crossentropy": 1.3771851062774658,
+      "loss/hidden": 3.13671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13404356315732002,
+      "step": 1457
+    },
+    {
+      "epoch": 0.243,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.5619140625,
+      "learning_rate": 8.614232765852935e-05,
+      "loss": 6.8897,
+      "loss/crossentropy": 1.9517407417297363,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17514695599675179,
+      "step": 1458
+    },
+    {
+      "epoch": 0.24316666666666667,
+      "grad_norm": 27.5,
+      "grad_norm_var": 1.7875,
+      "learning_rate": 8.612423214266749e-05,
+      "loss": 7.2573,
+      "loss/crossentropy": 2.2129026502370834,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18541562743484974,
+      "step": 1459
+    },
+    {
+      "epoch": 0.24333333333333335,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.7150390625,
+      "learning_rate": 8.610612672314251e-05,
+      "loss": 7.1514,
+      "loss/crossentropy": 2.064738839864731,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2067314051091671,
+      "step": 1460
+    },
+    {
+      "epoch": 0.2435,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.72890625,
+      "learning_rate": 8.608801140491811e-05,
+      "loss": 7.0956,
+      "loss/crossentropy": 1.7680425941944122,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18307415023446083,
+      "step": 1461
+    },
+    {
+      "epoch": 0.24366666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.7728515625,
+      "learning_rate": 8.606988619296071e-05,
+      "loss": 7.4754,
+      "loss/crossentropy": 1.802744910120964,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2313058190047741,
+      "step": 1462
+    },
+    {
+      "epoch": 0.24383333333333335,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.7541015625,
+      "learning_rate": 8.605175109223944e-05,
+      "loss": 7.1625,
+      "loss/crossentropy": 1.8716177940368652,
+      "loss/hidden": 3.796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2981249988079071,
+      "step": 1463
+    },
+    {
+      "epoch": 0.244,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.8337890625,
+      "learning_rate": 8.603360610772612e-05,
+      "loss": 7.1777,
+      "loss/crossentropy": 1.2806765884160995,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1341677624732256,
+      "step": 1464
+    },
+    {
+      "epoch": 0.24416666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.8134765625,
+      "learning_rate": 8.601545124439535e-05,
+      "loss": 7.2685,
+      "loss/crossentropy": 1.607251137495041,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16256394609808922,
+      "step": 1465
+    },
+    {
+      "epoch": 0.24433333333333335,
+      "grad_norm": 27.625,
+      "grad_norm_var": 1.8811848958333333,
+      "learning_rate": 8.599728650722434e-05,
+      "loss": 7.2185,
+      "loss/crossentropy": 2.2882717847824097,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18641741573810577,
+      "step": 1466
+    },
+    {
+      "epoch": 0.2445,
+      "grad_norm": 27.75,
+      "grad_norm_var": 1.5770182291666666,
+      "learning_rate": 8.597911190119308e-05,
+      "loss": 7.0735,
+      "loss/crossentropy": 1.8172245025634766,
+      "loss/hidden": 2.99609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12977637723088264,
+      "step": 1467
+    },
+    {
+      "epoch": 0.24466666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.9317057291666666,
+      "learning_rate": 8.596092743128423e-05,
+      "loss": 7.356,
+      "loss/crossentropy": 1.969709873199463,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1908802092075348,
+      "step": 1468
+    },
+    {
+      "epoch": 0.24483333333333332,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.8177083333333333,
+      "learning_rate": 8.594273310248318e-05,
+      "loss": 6.9405,
+      "loss/crossentropy": 1.341288834810257,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14361903257668018,
+      "step": 1469
+    },
+    {
+      "epoch": 0.245,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.0260416666666665,
+      "learning_rate": 8.592452891977798e-05,
+      "loss": 7.1089,
+      "loss/crossentropy": 1.6152670085430145,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17580680549144745,
+      "step": 1470
+    },
+    {
+      "epoch": 0.24516666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.4749348958333335,
+      "learning_rate": 8.590631488815944e-05,
+      "loss": 7.042,
+      "loss/crossentropy": 1.5854146480560303,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21539480239152908,
+      "step": 1471
+    },
+    {
+      "epoch": 0.24533333333333332,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.4775390625,
+      "learning_rate": 8.588809101262103e-05,
+      "loss": 7.6967,
+      "loss/crossentropy": 2.045955151319504,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2777820751070976,
+      "step": 1472
+    },
+    {
+      "epoch": 0.2455,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.5014973958333333,
+      "learning_rate": 8.586985729815894e-05,
+      "loss": 7.0096,
+      "loss/crossentropy": 1.769845336675644,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20903560891747475,
+      "step": 1473
+    },
+    {
+      "epoch": 0.24566666666666667,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.2905598958333333,
+      "learning_rate": 8.585161374977202e-05,
+      "loss": 7.0576,
+      "loss/crossentropy": 1.7164052724838257,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21758262440562248,
+      "step": 1474
+    },
+    {
+      "epoch": 0.24583333333333332,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.9671223958333333,
+      "learning_rate": 8.583336037246186e-05,
+      "loss": 7.3073,
+      "loss/crossentropy": 1.736921712756157,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.184262003749609,
+      "step": 1475
+    },
+    {
+      "epoch": 0.246,
+      "grad_norm": 27.0,
+      "grad_norm_var": 2.376822916666667,
+      "learning_rate": 8.581509717123273e-05,
+      "loss": 6.9367,
+      "loss/crossentropy": 2.058397263288498,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15587017685174942,
+      "step": 1476
+    },
+    {
+      "epoch": 0.24616666666666667,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.403125,
+      "learning_rate": 8.579682415109156e-05,
+      "loss": 7.0199,
+      "loss/crossentropy": 1.1905291378498077,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14508189260959625,
+      "step": 1477
+    },
+    {
+      "epoch": 0.24633333333333332,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.3684895833333335,
+      "learning_rate": 8.577854131704805e-05,
+      "loss": 7.2465,
+      "loss/crossentropy": 1.8980498313903809,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16383585706353188,
+      "step": 1478
+    },
+    {
+      "epoch": 0.2465,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.5416666666666665,
+      "learning_rate": 8.576024867411451e-05,
+      "loss": 7.3609,
+      "loss/crossentropy": 1.4511105865240097,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19967563822865486,
+      "step": 1479
+    },
+    {
+      "epoch": 0.24666666666666667,
+      "grad_norm": 28.25,
+      "grad_norm_var": 2.537955729166667,
+      "learning_rate": 8.574194622730599e-05,
+      "loss": 7.1107,
+      "loss/crossentropy": 1.9621224701404572,
+      "loss/hidden": 3.93359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23716596141457558,
+      "step": 1480
+    },
+    {
+      "epoch": 0.24683333333333332,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.537955729166667,
+      "learning_rate": 8.572363398164017e-05,
+      "loss": 7.4774,
+      "loss/crossentropy": 1.8094448447227478,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2159697487950325,
+      "step": 1481
+    },
+    {
+      "epoch": 0.247,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.3275390625,
+      "learning_rate": 8.57053119421375e-05,
+      "loss": 7.1269,
+      "loss/crossentropy": 1.6197231113910675,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21871723979711533,
+      "step": 1482
+    },
+    {
+      "epoch": 0.24716666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.4780598958333333,
+      "learning_rate": 8.568698011382107e-05,
+      "loss": 7.2147,
+      "loss/crossentropy": 1.4285579323768616,
+      "loss/hidden": 3.86328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1964031122624874,
+      "step": 1483
+    },
+    {
+      "epoch": 0.24733333333333332,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.234375,
+      "learning_rate": 8.566863850171663e-05,
+      "loss": 6.9712,
+      "loss/crossentropy": 1.8760530650615692,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16306671872735023,
+      "step": 1484
+    },
+    {
+      "epoch": 0.2475,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.49140625,
+      "learning_rate": 8.565028711085265e-05,
+      "loss": 7.1485,
+      "loss/crossentropy": 1.5396107137203217,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19017787277698517,
+      "step": 1485
+    },
+    {
+      "epoch": 0.24766666666666667,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.593489583333333,
+      "learning_rate": 8.563192594626027e-05,
+      "loss": 6.8738,
+      "loss/crossentropy": 1.8597404062747955,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16237230971455574,
+      "step": 1486
+    },
+    {
+      "epoch": 0.24783333333333332,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.039322916666667,
+      "learning_rate": 8.56135550129733e-05,
+      "loss": 7.2401,
+      "loss/crossentropy": 1.8960184007883072,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1787395477294922,
+      "step": 1487
+    },
+    {
+      "epoch": 0.248,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.146875,
+      "learning_rate": 8.559517431602824e-05,
+      "loss": 6.8938,
+      "loss/crossentropy": 1.3574792072176933,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1407286524772644,
+      "step": 1488
+    },
+    {
+      "epoch": 0.24816666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.66640625,
+      "learning_rate": 8.557678386046428e-05,
+      "loss": 6.9772,
+      "loss/crossentropy": 1.5204390436410904,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20090862177312374,
+      "step": 1489
+    },
+    {
+      "epoch": 0.24833333333333332,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.2728515625,
+      "learning_rate": 8.555838365132323e-05,
+      "loss": 7.2302,
+      "loss/crossentropy": 1.2873146086931229,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1601069662719965,
+      "step": 1490
+    },
+    {
+      "epoch": 0.2485,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.2728515625,
+      "learning_rate": 8.553997369364963e-05,
+      "loss": 6.9754,
+      "loss/crossentropy": 1.7107758224010468,
+      "loss/hidden": 3.73828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20447992160916328,
+      "step": 1491
+    },
+    {
+      "epoch": 0.24866666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.6184895833333335,
+      "learning_rate": 8.552155399249067e-05,
+      "loss": 7.0746,
+      "loss/crossentropy": 1.730744630098343,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15238011628389359,
+      "step": 1492
+    },
+    {
+      "epoch": 0.24883333333333332,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.8541015625,
+      "learning_rate": 8.550312455289625e-05,
+      "loss": 6.992,
+      "loss/crossentropy": 1.9606659412384033,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17902575619518757,
+      "step": 1493
+    },
+    {
+      "epoch": 0.249,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.8103515625,
+      "learning_rate": 8.548468537991884e-05,
+      "loss": 7.1084,
+      "loss/crossentropy": 1.9717397689819336,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18162487633526325,
+      "step": 1494
+    },
+    {
+      "epoch": 0.24916666666666668,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.753059895833333,
+      "learning_rate": 8.54662364786137e-05,
+      "loss": 7.4182,
+      "loss/crossentropy": 2.2676640450954437,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16843552514910698,
+      "step": 1495
+    },
+    {
+      "epoch": 0.24933333333333332,
+      "grad_norm": 25.625,
+      "grad_norm_var": 3.8372395833333335,
+      "learning_rate": 8.544777785403868e-05,
+      "loss": 6.7763,
+      "loss/crossentropy": 1.6614613831043243,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16282613947987556,
+      "step": 1496
+    },
+    {
+      "epoch": 0.2495,
+      "grad_norm": 27.75,
+      "grad_norm_var": 4.139583333333333,
+      "learning_rate": 8.542930951125432e-05,
+      "loss": 6.7205,
+      "loss/crossentropy": 1.4288576692342758,
+      "loss/hidden": 3.87890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24302856251597404,
+      "step": 1497
+    },
+    {
+      "epoch": 0.24966666666666668,
+      "grad_norm": 27.0,
+      "grad_norm_var": 4.635872395833333,
+      "learning_rate": 8.54108314553238e-05,
+      "loss": 6.9826,
+      "loss/crossentropy": 2.0986453741788864,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17323355376720428,
+      "step": 1498
+    },
+    {
+      "epoch": 0.24983333333333332,
+      "grad_norm": 28.75,
+      "grad_norm_var": 4.277018229166667,
+      "learning_rate": 8.539234369131301e-05,
+      "loss": 7.093,
+      "loss/crossentropy": 1.7604359090328217,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1727566346526146,
+      "step": 1499
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 30.0,
+      "grad_norm_var": 4.292643229166667,
+      "learning_rate": 8.53738462242905e-05,
+      "loss": 7.2285,
+      "loss/crossentropy": 1.557277888059616,
+      "loss/hidden": 3.6484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18158774450421333,
+      "step": 1500
+    },
+    {
+      "epoch": 0.25016666666666665,
+      "grad_norm": 27.25,
+      "grad_norm_var": 4.244791666666667,
+      "learning_rate": 8.535533905932738e-05,
+      "loss": 6.8726,
+      "loss/crossentropy": 1.6883008480072021,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20317288860678673,
+      "step": 1501
+    },
+    {
+      "epoch": 0.25033333333333335,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.064322916666667,
+      "learning_rate": 8.533682220149756e-05,
+      "loss": 6.9371,
+      "loss/crossentropy": 2.300634264945984,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19072575122117996,
+      "step": 1502
+    },
+    {
+      "epoch": 0.2505,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.970572916666667,
+      "learning_rate": 8.53182956558775e-05,
+      "loss": 7.3828,
+      "loss/crossentropy": 2.0842214226722717,
+      "loss/hidden": 3.87890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.43149901181459427,
+      "step": 1503
+    },
+    {
+      "epoch": 0.25066666666666665,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.8559895833333333,
+      "learning_rate": 8.52997594275464e-05,
+      "loss": 7.125,
+      "loss/crossentropy": 1.7222635746002197,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15759752690792084,
+      "step": 1504
+    },
+    {
+      "epoch": 0.25083333333333335,
+      "grad_norm": 28.375,
+      "grad_norm_var": 3.0374348958333335,
+      "learning_rate": 8.528121352158604e-05,
+      "loss": 7.0611,
+      "loss/crossentropy": 2.133837044239044,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1868600696325302,
+      "step": 1505
+    },
+    {
+      "epoch": 0.251,
+      "grad_norm": 26.625,
+      "grad_norm_var": 1.7018229166666667,
+      "learning_rate": 8.526265794308089e-05,
+      "loss": 6.9267,
+      "loss/crossentropy": 2.0259019136428833,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18409772589802742,
+      "step": 1506
+    },
+    {
+      "epoch": 0.25116666666666665,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.6010416666666667,
+      "learning_rate": 8.524409269711807e-05,
+      "loss": 7.3371,
+      "loss/crossentropy": 1.217084676027298,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16649403423070908,
+      "step": 1507
+    },
+    {
+      "epoch": 0.25133333333333335,
+      "grad_norm": 27.625,
+      "grad_norm_var": 1.4580729166666666,
+      "learning_rate": 8.522551778878736e-05,
+      "loss": 7.1716,
+      "loss/crossentropy": 1.7258444428443909,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1639699935913086,
+      "step": 1508
+    },
+    {
+      "epoch": 0.2515,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.6119140625,
+      "learning_rate": 8.520693322318116e-05,
+      "loss": 7.4071,
+      "loss/crossentropy": 1.9455642998218536,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1857447475194931,
+      "step": 1509
+    },
+    {
+      "epoch": 0.25166666666666665,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.1747395833333334,
+      "learning_rate": 8.518833900539454e-05,
+      "loss": 7.0886,
+      "loss/crossentropy": 1.5587359219789505,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1536642387509346,
+      "step": 1510
+    },
+    {
+      "epoch": 0.25183333333333335,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.0869140625,
+      "learning_rate": 8.516973514052519e-05,
+      "loss": 6.7147,
+      "loss/crossentropy": 1.5748094320297241,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16789069026708603,
+      "step": 1511
+    },
+    {
+      "epoch": 0.252,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.8760416666666666,
+      "learning_rate": 8.515112163367351e-05,
+      "loss": 7.1046,
+      "loss/crossentropy": 1.7668732106685638,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2092406563460827,
+      "step": 1512
+    },
+    {
+      "epoch": 0.25216666666666665,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.9645833333333333,
+      "learning_rate": 8.513249848994246e-05,
+      "loss": 7.1355,
+      "loss/crossentropy": 2.4020401537418365,
+      "loss/hidden": 3.0390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15349615179002285,
+      "step": 1513
+    },
+    {
+      "epoch": 0.25233333333333335,
+      "grad_norm": 27.625,
+      "grad_norm_var": 1.8379557291666666,
+      "learning_rate": 8.511386571443771e-05,
+      "loss": 6.7713,
+      "loss/crossentropy": 1.7552968263626099,
+      "loss/hidden": 3.640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22886677831411362,
+      "step": 1514
+    },
+    {
+      "epoch": 0.2525,
+      "grad_norm": 27.625,
+      "grad_norm_var": 1.9322916666666667,
+      "learning_rate": 8.50952233122675e-05,
+      "loss": 6.9603,
+      "loss/crossentropy": 1.7350502014160156,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23234521597623825,
+      "step": 1515
+    },
+    {
+      "epoch": 0.25266666666666665,
+      "grad_norm": 27.25,
+      "grad_norm_var": 1.9580729166666666,
+      "learning_rate": 8.50765712885428e-05,
+      "loss": 7.0924,
+      "loss/crossentropy": 1.6145904064178467,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17047151178121567,
+      "step": 1516
+    },
+    {
+      "epoch": 0.25283333333333335,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.8379557291666666,
+      "learning_rate": 8.505790964837713e-05,
+      "loss": 7.0948,
+      "loss/crossentropy": 1.5313849449157715,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18544949404895306,
+      "step": 1517
+    },
+    {
+      "epoch": 0.253,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.8197916666666667,
+      "learning_rate": 8.503923839688667e-05,
+      "loss": 6.9616,
+      "loss/crossentropy": 1.575784295797348,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20393813773989677,
+      "step": 1518
+    },
+    {
+      "epoch": 0.25316666666666665,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.8583333333333334,
+      "learning_rate": 8.502055753919032e-05,
+      "loss": 7.4483,
+      "loss/crossentropy": 2.1256864070892334,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3466687947511673,
+      "step": 1519
+    },
+    {
+      "epoch": 0.25333333333333335,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.8962890625,
+      "learning_rate": 8.500186708040949e-05,
+      "loss": 6.7947,
+      "loss/crossentropy": 2.2411983013153076,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1734490506350994,
+      "step": 1520
+    },
+    {
+      "epoch": 0.2535,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.8962890625,
+      "learning_rate": 8.498316702566828e-05,
+      "loss": 6.8642,
+      "loss/crossentropy": 1.929250806570053,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1943759173154831,
+      "step": 1521
+    },
+    {
+      "epoch": 0.25366666666666665,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.6593098958333334,
+      "learning_rate": 8.496445738009342e-05,
+      "loss": 7.1926,
+      "loss/crossentropy": 2.005999267101288,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18388929218053818,
+      "step": 1522
+    },
+    {
+      "epoch": 0.25383333333333336,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.6754557291666667,
+      "learning_rate": 8.494573814881426e-05,
+      "loss": 7.1619,
+      "loss/crossentropy": 1.5308670401573181,
+      "loss/hidden": 3.81640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20634648203849792,
+      "step": 1523
+    },
+    {
+      "epoch": 0.254,
+      "grad_norm": 44.75,
+      "grad_norm_var": 16.811458333333334,
+      "learning_rate": 8.49270093369628e-05,
+      "loss": 7.2042,
+      "loss/crossentropy": 2.2177110612392426,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2473539151251316,
+      "step": 1524
+    },
+    {
+      "epoch": 0.25416666666666665,
+      "grad_norm": 33.5,
+      "grad_norm_var": 17.527018229166668,
+      "learning_rate": 8.490827094967363e-05,
+      "loss": 7.0835,
+      "loss/crossentropy": 1.5321835726499557,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17551949247717857,
+      "step": 1525
+    },
+    {
+      "epoch": 0.25433333333333336,
+      "grad_norm": 28.0,
+      "grad_norm_var": 17.764322916666668,
+      "learning_rate": 8.488952299208401e-05,
+      "loss": 7.1661,
+      "loss/crossentropy": 1.5419676154851913,
+      "loss/hidden": 3.125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1426115594804287,
+      "step": 1526
+    },
+    {
+      "epoch": 0.2545,
+      "grad_norm": 29.25,
+      "grad_norm_var": 17.346875,
+      "learning_rate": 8.487076546933378e-05,
+      "loss": 7.1436,
+      "loss/crossentropy": 1.57235686480999,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18731963261961937,
+      "step": 1527
+    },
+    {
+      "epoch": 0.25466666666666665,
+      "grad_norm": 28.0,
+      "grad_norm_var": 17.624739583333334,
+      "learning_rate": 8.485199838656543e-05,
+      "loss": 7.1221,
+      "loss/crossentropy": 1.94024258852005,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19081877917051315,
+      "step": 1528
+    },
+    {
+      "epoch": 0.25483333333333336,
+      "grad_norm": 26.875,
+      "grad_norm_var": 18.245247395833335,
+      "learning_rate": 8.483322174892404e-05,
+      "loss": 6.9484,
+      "loss/crossentropy": 1.678273856639862,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18141436949372292,
+      "step": 1529
+    },
+    {
+      "epoch": 0.255,
+      "grad_norm": 27.625,
+      "grad_norm_var": 18.245247395833335,
+      "learning_rate": 8.481443556155735e-05,
+      "loss": 7.096,
+      "loss/crossentropy": 1.771409660577774,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14950870350003242,
+      "step": 1530
+    },
+    {
+      "epoch": 0.25516666666666665,
+      "grad_norm": 32.0,
+      "grad_norm_var": 18.151822916666667,
+      "learning_rate": 8.479563982961571e-05,
+      "loss": 7.8379,
+      "loss/crossentropy": 1.5515018999576569,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1533292606472969,
+      "step": 1531
+    },
+    {
+      "epoch": 0.25533333333333336,
+      "grad_norm": 30.5,
+      "grad_norm_var": 17.572916666666668,
+      "learning_rate": 8.477683455825207e-05,
+      "loss": 7.1362,
+      "loss/crossentropy": 2.018773838877678,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18661978095769882,
+      "step": 1532
+    },
+    {
+      "epoch": 0.2555,
+      "grad_norm": 29.625,
+      "grad_norm_var": 17.509375,
+      "learning_rate": 8.4758019752622e-05,
+      "loss": 7.117,
+      "loss/crossentropy": 1.2113154977560043,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13429363630712032,
+      "step": 1533
+    },
+    {
+      "epoch": 0.25566666666666665,
+      "grad_norm": 31.5,
+      "grad_norm_var": 17.262239583333333,
+      "learning_rate": 8.473919541788366e-05,
+      "loss": 7.4119,
+      "loss/crossentropy": 1.7601763010025024,
+      "loss/hidden": 3.796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16794865019619465,
+      "step": 1534
+    },
+    {
+      "epoch": 0.25583333333333336,
+      "grad_norm": 43.0,
+      "grad_norm_var": 26.768489583333334,
+      "learning_rate": 8.472036155919791e-05,
+      "loss": 7.3537,
+      "loss/crossentropy": 1.8414527773857117,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23525508493185043,
+      "step": 1535
+    },
+    {
+      "epoch": 0.256,
+      "grad_norm": 30.75,
+      "grad_norm_var": 26.582747395833334,
+      "learning_rate": 8.470151818172809e-05,
+      "loss": 7.0589,
+      "loss/crossentropy": 1.472220093011856,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19062213227152824,
+      "step": 1536
+    },
+    {
+      "epoch": 0.25616666666666665,
+      "grad_norm": 30.25,
+      "grad_norm_var": 26.030989583333334,
+      "learning_rate": 8.468266529064025e-05,
+      "loss": 6.4525,
+      "loss/crossentropy": 0.974083200097084,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13269340246915817,
+      "step": 1537
+    },
+    {
+      "epoch": 0.25633333333333336,
+      "grad_norm": 29.5,
+      "grad_norm_var": 26.176497395833334,
+      "learning_rate": 8.466380289110303e-05,
+      "loss": 7.0886,
+      "loss/crossentropy": 1.259334146976471,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14339469373226166,
+      "step": 1538
+    },
+    {
+      "epoch": 0.2565,
+      "grad_norm": 29.125,
+      "grad_norm_var": 26.287239583333335,
+      "learning_rate": 8.464493098828763e-05,
+      "loss": 7.2628,
+      "loss/crossentropy": 1.6031538248062134,
+      "loss/hidden": 3.8046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1961599849164486,
+      "step": 1539
+    },
+    {
+      "epoch": 0.25666666666666665,
+      "grad_norm": 27.25,
+      "grad_norm_var": 14.54765625,
+      "learning_rate": 8.462604958736793e-05,
+      "loss": 6.8737,
+      "loss/crossentropy": 1.3939393907785416,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15689893811941147,
+      "step": 1540
+    },
+    {
+      "epoch": 0.25683333333333336,
+      "grad_norm": 27.375,
+      "grad_norm_var": 14.378580729166666,
+      "learning_rate": 8.460715869352035e-05,
+      "loss": 6.9708,
+      "loss/crossentropy": 1.4536120295524597,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13832676224410534,
+      "step": 1541
+    },
+    {
+      "epoch": 0.257,
+      "grad_norm": 29.125,
+      "grad_norm_var": 14.151822916666667,
+      "learning_rate": 8.458825831192392e-05,
+      "loss": 7.3376,
+      "loss/crossentropy": 1.9465630948543549,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1429009847342968,
+      "step": 1542
+    },
+    {
+      "epoch": 0.25716666666666665,
+      "grad_norm": 30.25,
+      "grad_norm_var": 14.099739583333333,
+      "learning_rate": 8.456934844776032e-05,
+      "loss": 7.2746,
+      "loss/crossentropy": 1.4517296254634857,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14977369271218777,
+      "step": 1543
+    },
+    {
+      "epoch": 0.25733333333333336,
+      "grad_norm": 26.625,
+      "grad_norm_var": 14.616080729166667,
+      "learning_rate": 8.455042910621379e-05,
+      "loss": 6.8794,
+      "loss/crossentropy": 1.5584948658943176,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16389862447977066,
+      "step": 1544
+    },
+    {
+      "epoch": 0.2575,
+      "grad_norm": 27.625,
+      "grad_norm_var": 14.330143229166667,
+      "learning_rate": 8.453150029247114e-05,
+      "loss": 7.0129,
+      "loss/crossentropy": 1.470071092247963,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15290442667901516,
+      "step": 1545
+    },
+    {
+      "epoch": 0.25766666666666665,
+      "grad_norm": 30.625,
+      "grad_norm_var": 13.889518229166667,
+      "learning_rate": 8.451256201172186e-05,
+      "loss": 6.7035,
+      "loss/crossentropy": 1.3290538638830185,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12361126020550728,
+      "step": 1546
+    },
+    {
+      "epoch": 0.25783333333333336,
+      "grad_norm": 28.25,
+      "grad_norm_var": 13.928580729166667,
+      "learning_rate": 8.449361426915797e-05,
+      "loss": 7.0498,
+      "loss/crossentropy": 1.8176316320896149,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1719118431210518,
+      "step": 1547
+    },
+    {
+      "epoch": 0.258,
+      "grad_norm": 39.25,
+      "grad_norm_var": 19.196809895833333,
+      "learning_rate": 8.447465706997408e-05,
+      "loss": 7.2554,
+      "loss/crossentropy": 2.1948321759700775,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17607615143060684,
+      "step": 1548
+    },
+    {
+      "epoch": 0.25816666666666666,
+      "grad_norm": 27.75,
+      "grad_norm_var": 19.668489583333333,
+      "learning_rate": 8.445569041936743e-05,
+      "loss": 6.9313,
+      "loss/crossentropy": 1.476136177778244,
+      "loss/hidden": 3.8515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1581102553755045,
+      "step": 1549
+    },
+    {
+      "epoch": 0.25833333333333336,
+      "grad_norm": 29.75,
+      "grad_norm_var": 19.630208333333332,
+      "learning_rate": 8.443671432253784e-05,
+      "loss": 7.3193,
+      "loss/crossentropy": 1.8609939515590668,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17737817764282227,
+      "step": 1550
+    },
+    {
+      "epoch": 0.2585,
+      "grad_norm": 28.0,
+      "grad_norm_var": 8.505208333333334,
+      "learning_rate": 8.44177287846877e-05,
+      "loss": 6.6163,
+      "loss/crossentropy": 1.6206251382827759,
+      "loss/hidden": 3.04296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13844903372228146,
+      "step": 1551
+    },
+    {
+      "epoch": 0.25866666666666666,
+      "grad_norm": 27.125,
+      "grad_norm_var": 8.7072265625,
+      "learning_rate": 8.439873381102203e-05,
+      "loss": 6.9954,
+      "loss/crossentropy": 2.0754114985466003,
+      "loss/hidden": 3.1171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15478336066007614,
+      "step": 1552
+    },
+    {
+      "epoch": 0.25883333333333336,
+      "grad_norm": 27.5,
+      "grad_norm_var": 8.8103515625,
+      "learning_rate": 8.437972940674838e-05,
+      "loss": 7.3634,
+      "loss/crossentropy": 1.820733219385147,
+      "loss/hidden": 3.80859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2608935683965683,
+      "step": 1553
+    },
+    {
+      "epoch": 0.259,
+      "grad_norm": 28.5,
+      "grad_norm_var": 8.815559895833333,
+      "learning_rate": 8.436071557707692e-05,
+      "loss": 7.2288,
+      "loss/crossentropy": 1.9718312621116638,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19307303428649902,
+      "step": 1554
+    },
+    {
+      "epoch": 0.25916666666666666,
+      "grad_norm": 27.875,
+      "grad_norm_var": 8.893684895833333,
+      "learning_rate": 8.434169232722043e-05,
+      "loss": 6.8213,
+      "loss/crossentropy": 1.463202804327011,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17362061887979507,
+      "step": 1555
+    },
+    {
+      "epoch": 0.25933333333333336,
+      "grad_norm": 40.5,
+      "grad_norm_var": 16.898893229166667,
+      "learning_rate": 8.432265966239419e-05,
+      "loss": 7.0426,
+      "loss/crossentropy": 1.4735060930252075,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16704573668539524,
+      "step": 1556
+    },
+    {
+      "epoch": 0.2595,
+      "grad_norm": 28.5,
+      "grad_norm_var": 16.620572916666667,
+      "learning_rate": 8.430361758781616e-05,
+      "loss": 7.0873,
+      "loss/crossentropy": 1.4179468750953674,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14967207983136177,
+      "step": 1557
+    },
+    {
+      "epoch": 0.25966666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 16.585416666666667,
+      "learning_rate": 8.42845661087068e-05,
+      "loss": 7.1166,
+      "loss/crossentropy": 1.5888977199792862,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23130910098552704,
+      "step": 1558
+    },
+    {
+      "epoch": 0.25983333333333336,
+      "grad_norm": 28.375,
+      "grad_norm_var": 16.711393229166667,
+      "learning_rate": 8.42655052302892e-05,
+      "loss": 6.9142,
+      "loss/crossentropy": 1.545868456363678,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1576635129749775,
+      "step": 1559
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 27.625,
+      "grad_norm_var": 16.356184895833334,
+      "learning_rate": 8.424643495778902e-05,
+      "loss": 7.203,
+      "loss/crossentropy": 2.0548255443573,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19096845015883446,
+      "step": 1560
+    },
+    {
+      "epoch": 0.26016666666666666,
+      "grad_norm": 29.625,
+      "grad_norm_var": 16.020768229166666,
+      "learning_rate": 8.422735529643444e-05,
+      "loss": 6.9342,
+      "loss/crossentropy": 1.4977477341890335,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1994573064148426,
+      "step": 1561
+    },
+    {
+      "epoch": 0.26033333333333336,
+      "grad_norm": 27.875,
+      "grad_norm_var": 16.244205729166666,
+      "learning_rate": 8.42082662514563e-05,
+      "loss": 7.1278,
+      "loss/crossentropy": 1.5780085325241089,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15711280703544617,
+      "step": 1562
+    },
+    {
+      "epoch": 0.2605,
+      "grad_norm": 30.875,
+      "grad_norm_var": 16.141666666666666,
+      "learning_rate": 8.418916782808795e-05,
+      "loss": 7.1971,
+      "loss/crossentropy": 1.6861535906791687,
+      "loss/hidden": 3.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14266281202435493,
+      "step": 1563
+    },
+    {
+      "epoch": 0.26066666666666666,
+      "grad_norm": 29.625,
+      "grad_norm_var": 9.9806640625,
+      "learning_rate": 8.417006003156532e-05,
+      "loss": 6.7606,
+      "loss/crossentropy": 1.9404644668102264,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17871755361557007,
+      "step": 1564
+    },
+    {
+      "epoch": 0.2608333333333333,
+      "grad_norm": 37.75,
+      "grad_norm_var": 14.116080729166667,
+      "learning_rate": 8.415094286712694e-05,
+      "loss": 7.1155,
+      "loss/crossentropy": 1.7015289962291718,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20424244925379753,
+      "step": 1565
+    },
+    {
+      "epoch": 0.261,
+      "grad_norm": 31.125,
+      "grad_norm_var": 14.195572916666666,
+      "learning_rate": 8.413181634001391e-05,
+      "loss": 7.2127,
+      "loss/crossentropy": 2.0783791840076447,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25549715384840965,
+      "step": 1566
+    },
+    {
+      "epoch": 0.26116666666666666,
+      "grad_norm": 28.5,
+      "grad_norm_var": 14.074739583333333,
+      "learning_rate": 8.411268045546983e-05,
+      "loss": 7.4015,
+      "loss/crossentropy": 1.431897595524788,
+      "loss/hidden": 3.9453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22732597962021828,
+      "step": 1567
+    },
+    {
+      "epoch": 0.2613333333333333,
+      "grad_norm": 26.5,
+      "grad_norm_var": 14.345247395833333,
+      "learning_rate": 8.409353521874093e-05,
+      "loss": 7.0976,
+      "loss/crossentropy": 1.878661870956421,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1574998889118433,
+      "step": 1568
+    },
+    {
+      "epoch": 0.2615,
+      "grad_norm": 28.0,
+      "grad_norm_var": 14.1916015625,
+      "learning_rate": 8.4074380635076e-05,
+      "loss": 7.0022,
+      "loss/crossentropy": 1.534603327512741,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17322967574000359,
+      "step": 1569
+    },
+    {
+      "epoch": 0.26166666666666666,
+      "grad_norm": 26.875,
+      "grad_norm_var": 14.696875,
+      "learning_rate": 8.405521670972634e-05,
+      "loss": 6.8425,
+      "loss/crossentropy": 1.6109719723463058,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16750073432922363,
+      "step": 1570
+    },
+    {
+      "epoch": 0.2618333333333333,
+      "grad_norm": 26.25,
+      "grad_norm_var": 15.315559895833333,
+      "learning_rate": 8.40360434479459e-05,
+      "loss": 7.1104,
+      "loss/crossentropy": 1.5548063814640045,
+      "loss/hidden": 3.90625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23038798198103905,
+      "step": 1571
+    },
+    {
+      "epoch": 0.262,
+      "grad_norm": 26.75,
+      "grad_norm_var": 7.6384765625,
+      "learning_rate": 8.40168608549911e-05,
+      "loss": 6.9809,
+      "loss/crossentropy": 1.2600118517875671,
+      "loss/hidden": 3.88671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18548330664634705,
+      "step": 1572
+    },
+    {
+      "epoch": 0.26216666666666666,
+      "grad_norm": 35.5,
+      "grad_norm_var": 10.227018229166667,
+      "learning_rate": 8.399766893612096e-05,
+      "loss": 7.0494,
+      "loss/crossentropy": 1.1565890908241272,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13933814130723476,
+      "step": 1573
+    },
+    {
+      "epoch": 0.2623333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 10.213997395833333,
+      "learning_rate": 8.397846769659707e-05,
+      "loss": 6.8379,
+      "loss/crossentropy": 1.641678899526596,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15472246706485748,
+      "step": 1574
+    },
+    {
+      "epoch": 0.2625,
+      "grad_norm": 30.25,
+      "grad_norm_var": 10.173958333333333,
+      "learning_rate": 8.395925714168356e-05,
+      "loss": 6.9379,
+      "loss/crossentropy": 1.7203895598649979,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20822539553046227,
+      "step": 1575
+    },
+    {
+      "epoch": 0.26266666666666666,
+      "grad_norm": 2969567232.0,
+      "grad_norm_var": 5.511455855764571e+17,
+      "learning_rate": 8.39400372766471e-05,
+      "loss": 7.0619,
+      "loss/crossentropy": 1.9653513431549072,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15789402648806572,
+      "step": 1576
+    },
+    {
+      "epoch": 0.2628333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 5.5114558555171066e+17,
+      "learning_rate": 8.392080810675691e-05,
+      "loss": 7.0243,
+      "loss/crossentropy": 1.8614015877246857,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15636020712554455,
+      "step": 1577
+    },
+    {
+      "epoch": 0.263,
+      "grad_norm": 29.625,
+      "grad_norm_var": 5.511455855084045e+17,
+      "learning_rate": 8.390156963728482e-05,
+      "loss": 7.1314,
+      "loss/crossentropy": 1.4498876333236694,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15764983743429184,
+      "step": 1578
+    },
+    {
+      "epoch": 0.26316666666666666,
+      "grad_norm": 26.75,
+      "grad_norm_var": 5.511455856104834e+17,
+      "learning_rate": 8.388232187350512e-05,
+      "loss": 6.9675,
+      "loss/crossentropy": 1.3486933410167694,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15304425545036793,
+      "step": 1579
+    },
+    {
+      "epoch": 0.2633333333333333,
+      "grad_norm": 27.875,
+      "grad_norm_var": 5.511455856537896e+17,
+      "learning_rate": 8.386306482069473e-05,
+      "loss": 6.6682,
+      "loss/crossentropy": 1.859202891588211,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17555386573076248,
+      "step": 1580
+    },
+    {
+      "epoch": 0.2635,
+      "grad_norm": 27.625,
+      "grad_norm_var": 5.5114558590434675e+17,
+      "learning_rate": 8.384379848413304e-05,
+      "loss": 6.9377,
+      "loss/crossentropy": 1.013603761792183,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13042955473065376,
+      "step": 1581
+    },
+    {
+      "epoch": 0.26366666666666666,
+      "grad_norm": 40.5,
+      "grad_norm_var": 5.511455856723494e+17,
+      "learning_rate": 8.382452286910206e-05,
+      "loss": 7.3273,
+      "loss/crossentropy": 1.5050248205661774,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15211456269025803,
+      "step": 1582
+    },
+    {
+      "epoch": 0.2638333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 5.511455856259499e+17,
+      "learning_rate": 8.380523798088631e-05,
+      "loss": 7.5001,
+      "loss/crossentropy": 1.8162740468978882,
+      "loss/hidden": 4.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29241993837058544,
+      "step": 1583
+    },
+    {
+      "epoch": 0.264,
+      "grad_norm": 29.0,
+      "grad_norm_var": 5.511455855640839e+17,
+      "learning_rate": 8.378594382477282e-05,
+      "loss": 7.0835,
+      "loss/crossentropy": 1.2985506057739258,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23931736685335636,
+      "step": 1584
+    },
+    {
+      "epoch": 0.26416666666666666,
+      "grad_norm": 26.375,
+      "grad_norm_var": 5.5114558560429677e+17,
+      "learning_rate": 8.376664040605122e-05,
+      "loss": 7.0656,
+      "loss/crossentropy": 1.7525395154953003,
+      "loss/hidden": 3.16015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15040944889187813,
+      "step": 1585
+    },
+    {
+      "epoch": 0.2643333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 5.511455855486174e+17,
+      "learning_rate": 8.374732773001366e-05,
+      "loss": 7.8002,
+      "loss/crossentropy": 2.012197196483612,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18983308225870132,
+      "step": 1586
+    },
+    {
+      "epoch": 0.2645,
+      "grad_norm": 30.75,
+      "grad_norm_var": 5.511455854372586e+17,
+      "learning_rate": 8.372800580195479e-05,
+      "loss": 6.6716,
+      "loss/crossentropy": 1.6122665256261826,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18357956409454346,
+      "step": 1587
+    },
+    {
+      "epoch": 0.26466666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 5.5114558533208646e+17,
+      "learning_rate": 8.370867462717183e-05,
+      "loss": 7.0967,
+      "loss/crossentropy": 1.8041424453258514,
+      "loss/hidden": 3.74609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21673135831952095,
+      "step": 1588
+    },
+    {
+      "epoch": 0.2648333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 5.511455854898447e+17,
+      "learning_rate": 8.368933421096454e-05,
+      "loss": 7.2579,
+      "loss/crossentropy": 1.7340957969427109,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20513366535305977,
+      "step": 1589
+    },
+    {
+      "epoch": 0.265,
+      "grad_norm": 30.5,
+      "grad_norm_var": 5.51145585462005e+17,
+      "learning_rate": 8.366998455863522e-05,
+      "loss": 7.4061,
+      "loss/crossentropy": 2.109828770160675,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1922304593026638,
+      "step": 1590
+    },
+    {
+      "epoch": 0.26516666666666666,
+      "grad_norm": 37.5,
+      "grad_norm_var": 5.5114558528259366e+17,
+      "learning_rate": 8.365062567548867e-05,
+      "loss": 7.3231,
+      "loss/crossentropy": 0.8247495591640472,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14741606824100018,
+      "step": 1591
+    },
+    {
+      "epoch": 0.2653333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 13.895572916666667,
+      "learning_rate": 8.363125756683223e-05,
+      "loss": 7.1474,
+      "loss/crossentropy": 1.5677947998046875,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22204036638140678,
+      "step": 1592
+    },
+    {
+      "epoch": 0.2655,
+      "grad_norm": 29.5,
+      "grad_norm_var": 13.972330729166666,
+      "learning_rate": 8.361188023797582e-05,
+      "loss": 7.1309,
+      "loss/crossentropy": 2.2050703167915344,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20694582164287567,
+      "step": 1593
+    },
+    {
+      "epoch": 0.26566666666666666,
+      "grad_norm": 28.0,
+      "grad_norm_var": 14.335416666666667,
+      "learning_rate": 8.359249369423177e-05,
+      "loss": 6.9907,
+      "loss/crossentropy": 1.7909895777702332,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1994033232331276,
+      "step": 1594
+    },
+    {
+      "epoch": 0.2658333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 13.521875,
+      "learning_rate": 8.357309794091507e-05,
+      "loss": 7.0712,
+      "loss/crossentropy": 1.9561614692211151,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18634473904967308,
+      "step": 1595
+    },
+    {
+      "epoch": 0.266,
+      "grad_norm": 26.75,
+      "grad_norm_var": 14.0369140625,
+      "learning_rate": 8.355369298334316e-05,
+      "loss": 6.6654,
+      "loss/crossentropy": 1.5238149724900723,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13797534350305796,
+      "step": 1596
+    },
+    {
+      "epoch": 0.26616666666666666,
+      "grad_norm": 27.25,
+      "grad_norm_var": 14.2,
+      "learning_rate": 8.3534278826836e-05,
+      "loss": 7.0763,
+      "loss/crossentropy": 1.6212818920612335,
+      "loss/hidden": 3.86328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22221536561846733,
+      "step": 1597
+    },
+    {
+      "epoch": 0.2663333333333333,
+      "grad_norm": 28.0,
+      "grad_norm_var": 7.611458333333333,
+      "learning_rate": 8.351485547671613e-05,
+      "loss": 6.9769,
+      "loss/crossentropy": 1.9699319750070572,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22051548957824707,
+      "step": 1598
+    },
+    {
+      "epoch": 0.2665,
+      "grad_norm": 32.75,
+      "grad_norm_var": 8.112434895833333,
+      "learning_rate": 8.349542293830855e-05,
+      "loss": 6.9139,
+      "loss/crossentropy": 1.276191920042038,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18996757827699184,
+      "step": 1599
+    },
+    {
+      "epoch": 0.26666666666666666,
+      "grad_norm": 29.125,
+      "grad_norm_var": 8.095833333333333,
+      "learning_rate": 8.347598121694078e-05,
+      "loss": 7.0277,
+      "loss/crossentropy": 1.8602609634399414,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1690271534025669,
+      "step": 1600
+    },
+    {
+      "epoch": 0.2668333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 7.257291666666666,
+      "learning_rate": 8.345653031794292e-05,
+      "loss": 7.2089,
+      "loss/crossentropy": 1.927530288696289,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19715483859181404,
+      "step": 1601
+    },
+    {
+      "epoch": 0.267,
+      "grad_norm": 28.625,
+      "grad_norm_var": 7.345833333333333,
+      "learning_rate": 8.343707024664751e-05,
+      "loss": 7.0145,
+      "loss/crossentropy": 1.5379869937896729,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1418660581111908,
+      "step": 1602
+    },
+    {
+      "epoch": 0.26716666666666666,
+      "grad_norm": 26.75,
+      "grad_norm_var": 8.045833333333333,
+      "learning_rate": 8.341760100838965e-05,
+      "loss": 7.2618,
+      "loss/crossentropy": 1.7103917300701141,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23981189355254173,
+      "step": 1603
+    },
+    {
+      "epoch": 0.2673333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 7.969205729166666,
+      "learning_rate": 8.339812260850696e-05,
+      "loss": 6.997,
+      "loss/crossentropy": 2.4385569095611572,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1898985467851162,
+      "step": 1604
+    },
+    {
+      "epoch": 0.2675,
+      "grad_norm": 27.125,
+      "grad_norm_var": 8.412955729166667,
+      "learning_rate": 8.337863505233953e-05,
+      "loss": 6.4637,
+      "loss/crossentropy": 2.0337356328964233,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18543323129415512,
+      "step": 1605
+    },
+    {
+      "epoch": 0.26766666666666666,
+      "grad_norm": 34.25,
+      "grad_norm_var": 9.678580729166667,
+      "learning_rate": 8.335913834522999e-05,
+      "loss": 7.2893,
+      "loss/crossentropy": 1.8543764650821686,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17504257522523403,
+      "step": 1606
+    },
+    {
+      "epoch": 0.2678333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 5.655143229166667,
+      "learning_rate": 8.333963249252348e-05,
+      "loss": 7.0358,
+      "loss/crossentropy": 1.6898139864206314,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1446678638458252,
+      "step": 1607
+    },
+    {
+      "epoch": 0.268,
+      "grad_norm": 27.75,
+      "grad_norm_var": 4.922330729166666,
+      "learning_rate": 8.332011749956763e-05,
+      "loss": 6.9523,
+      "loss/crossentropy": 1.9716004431247711,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17932548001408577,
+      "step": 1608
+    },
+    {
+      "epoch": 0.26816666666666666,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.917708333333334,
+      "learning_rate": 8.330059337171258e-05,
+      "loss": 6.7706,
+      "loss/crossentropy": 1.751436173915863,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14808068424463272,
+      "step": 1609
+    },
+    {
+      "epoch": 0.2683333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 5.2587890625,
+      "learning_rate": 8.328106011431101e-05,
+      "loss": 7.6746,
+      "loss/crossentropy": 2.046698808670044,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17839471623301506,
+      "step": 1610
+    },
+    {
+      "epoch": 0.2685,
+      "grad_norm": 27.5,
+      "grad_norm_var": 4.862955729166667,
+      "learning_rate": 8.326151773271804e-05,
+      "loss": 7.1291,
+      "loss/crossentropy": 1.4974311590194702,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24513380974531174,
+      "step": 1611
+    },
+    {
+      "epoch": 0.26866666666666666,
+      "grad_norm": 29.625,
+      "grad_norm_var": 4.478125,
+      "learning_rate": 8.324196623229135e-05,
+      "loss": 6.7984,
+      "loss/crossentropy": 2.1385327875614166,
+      "loss/hidden": 3.08984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14981893822550774,
+      "step": 1612
+    },
+    {
+      "epoch": 0.2688333333333333,
+      "grad_norm": 28.25,
+      "grad_norm_var": 4.269791666666666,
+      "learning_rate": 8.322240561839109e-05,
+      "loss": 7.1111,
+      "loss/crossentropy": 1.9146039187908173,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14177163504064083,
+      "step": 1613
+    },
+    {
+      "epoch": 0.269,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.2712890625,
+      "learning_rate": 8.32028358963799e-05,
+      "loss": 6.7759,
+      "loss/crossentropy": 1.4329868853092194,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20702739618718624,
+      "step": 1614
+    },
+    {
+      "epoch": 0.26916666666666667,
+      "grad_norm": 35.75,
+      "grad_norm_var": 6.1244140625,
+      "learning_rate": 8.318325707162293e-05,
+      "loss": 7.0046,
+      "loss/crossentropy": 1.8867762684822083,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18985198065638542,
+      "step": 1615
+    },
+    {
+      "epoch": 0.2693333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 6.1478515625,
+      "learning_rate": 8.316366914948783e-05,
+      "loss": 6.7755,
+      "loss/crossentropy": 1.858707919716835,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19729428738355637,
+      "step": 1616
+    },
+    {
+      "epoch": 0.2695,
+      "grad_norm": 28.75,
+      "grad_norm_var": 6.1625,
+      "learning_rate": 8.314407213534476e-05,
+      "loss": 6.9882,
+      "loss/crossentropy": 1.5668911039829254,
+      "loss/hidden": 3.66015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18824711814522743,
+      "step": 1617
+    },
+    {
+      "epoch": 0.26966666666666667,
+      "grad_norm": 27.875,
+      "grad_norm_var": 6.30390625,
+      "learning_rate": 8.312446603456632e-05,
+      "loss": 7.1649,
+      "loss/crossentropy": 1.7864155173301697,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19499209336936474,
+      "step": 1618
+    },
+    {
+      "epoch": 0.2698333333333333,
+      "grad_norm": 28.625,
+      "grad_norm_var": 5.8009765625,
+      "learning_rate": 8.310485085252767e-05,
+      "loss": 6.999,
+      "loss/crossentropy": 1.765356421470642,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16575749590992928,
+      "step": 1619
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 29.125,
+      "grad_norm_var": 5.825455729166666,
+      "learning_rate": 8.308522659460641e-05,
+      "loss": 7.1248,
+      "loss/crossentropy": 1.698109045624733,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16750097274780273,
+      "step": 1620
+    },
+    {
+      "epoch": 0.27016666666666667,
+      "grad_norm": 27.5,
+      "grad_norm_var": 5.704166666666667,
+      "learning_rate": 8.306559326618259e-05,
+      "loss": 6.8828,
+      "loss/crossentropy": 1.8713548481464386,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19754157587885857,
+      "step": 1621
+    },
+    {
+      "epoch": 0.2703333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 4.276822916666666,
+      "learning_rate": 8.304595087263889e-05,
+      "loss": 6.8131,
+      "loss/crossentropy": 1.7399481683969498,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19789675623178482,
+      "step": 1622
+    },
+    {
+      "epoch": 0.2705,
+      "grad_norm": 29.75,
+      "grad_norm_var": 4.261458333333334,
+      "learning_rate": 8.30262994193603e-05,
+      "loss": 6.9586,
+      "loss/crossentropy": 1.316022828221321,
+      "loss/hidden": 3.8046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14393935725092888,
+      "step": 1623
+    },
+    {
+      "epoch": 0.27066666666666667,
+      "grad_norm": 34.75,
+      "grad_norm_var": 5.778125,
+      "learning_rate": 8.300663891173443e-05,
+      "loss": 7.0578,
+      "loss/crossentropy": 1.5755325853824615,
+      "loss/hidden": 3.765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24174069985747337,
+      "step": 1624
+    },
+    {
+      "epoch": 0.2708333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 6.5837890625,
+      "learning_rate": 8.298696935515132e-05,
+      "loss": 7.7362,
+      "loss/crossentropy": 2.2487829625606537,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18154580518603325,
+      "step": 1625
+    },
+    {
+      "epoch": 0.271,
+      "grad_norm": 30.25,
+      "grad_norm_var": 6.364583333333333,
+      "learning_rate": 8.296729075500344e-05,
+      "loss": 6.9045,
+      "loss/crossentropy": 1.4855230748653412,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.276097908616066,
+      "step": 1626
+    },
+    {
+      "epoch": 0.27116666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 5.933268229166667,
+      "learning_rate": 8.294760311668586e-05,
+      "loss": 7.6882,
+      "loss/crossentropy": 1.9179571866989136,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2302505485713482,
+      "step": 1627
+    },
+    {
+      "epoch": 0.2713333333333333,
+      "grad_norm": 26.625,
+      "grad_norm_var": 6.723893229166666,
+      "learning_rate": 8.2927906445596e-05,
+      "loss": 6.6591,
+      "loss/crossentropy": 1.8981229066848755,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15968799963593483,
+      "step": 1628
+    },
+    {
+      "epoch": 0.2715,
+      "grad_norm": 27.375,
+      "grad_norm_var": 6.976822916666666,
+      "learning_rate": 8.290820074713384e-05,
+      "loss": 7.2994,
+      "loss/crossentropy": 1.774809718132019,
+      "loss/hidden": 3.89453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30011754482984543,
+      "step": 1629
+    },
+    {
+      "epoch": 0.27166666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 6.933072916666666,
+      "learning_rate": 8.28884860267018e-05,
+      "loss": 6.9311,
+      "loss/crossentropy": 1.9921766817569733,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21032875776290894,
+      "step": 1630
+    },
+    {
+      "epoch": 0.2718333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.723958333333333,
+      "learning_rate": 8.28687622897048e-05,
+      "loss": 6.7365,
+      "loss/crossentropy": 1.830979347229004,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21521583199501038,
+      "step": 1631
+    },
+    {
+      "epoch": 0.272,
+      "grad_norm": 28.25,
+      "grad_norm_var": 4.808268229166667,
+      "learning_rate": 8.284902954155019e-05,
+      "loss": 7.0567,
+      "loss/crossentropy": 1.6267684698104858,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19743401557207108,
+      "step": 1632
+    },
+    {
+      "epoch": 0.27216666666666667,
+      "grad_norm": 27.875,
+      "grad_norm_var": 4.95,
+      "learning_rate": 8.282928778764783e-05,
+      "loss": 6.5505,
+      "loss/crossentropy": 1.54688461124897,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19320432841777802,
+      "step": 1633
+    },
+    {
+      "epoch": 0.2723333333333333,
+      "grad_norm": 25.75,
+      "grad_norm_var": 5.692643229166666,
+      "learning_rate": 8.280953703341004e-05,
+      "loss": 7.2325,
+      "loss/crossentropy": 1.7405301481485367,
+      "loss/hidden": 3.73828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1903975997120142,
+      "step": 1634
+    },
+    {
+      "epoch": 0.2725,
+      "grad_norm": 31.25,
+      "grad_norm_var": 5.863541666666666,
+      "learning_rate": 8.278977728425157e-05,
+      "loss": 6.9873,
+      "loss/crossentropy": 1.3869994282722473,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1749619785696268,
+      "step": 1635
+    },
+    {
+      "epoch": 0.27266666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 5.857747395833333,
+      "learning_rate": 8.27700085455897e-05,
+      "loss": 6.651,
+      "loss/crossentropy": 1.5207037031650543,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21287047490477562,
+      "step": 1636
+    },
+    {
+      "epoch": 0.2728333333333333,
+      "grad_norm": 26.625,
+      "grad_norm_var": 6.143489583333333,
+      "learning_rate": 8.275023082284413e-05,
+      "loss": 7.072,
+      "loss/crossentropy": 1.956925630569458,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19619474932551384,
+      "step": 1637
+    },
+    {
+      "epoch": 0.273,
+      "grad_norm": 27.875,
+      "grad_norm_var": 6.295247395833333,
+      "learning_rate": 8.273044412143704e-05,
+      "loss": 6.8739,
+      "loss/crossentropy": 1.2441673576831818,
+      "loss/hidden": 3.8515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14691221714019775,
+      "step": 1638
+    },
+    {
+      "epoch": 0.27316666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 7.217643229166667,
+      "learning_rate": 8.271064844679306e-05,
+      "loss": 6.7274,
+      "loss/crossentropy": 2.5751023292541504,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19316158071160316,
+      "step": 1639
+    },
+    {
+      "epoch": 0.2733333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 5.42890625,
+      "learning_rate": 8.269084380433929e-05,
+      "loss": 6.7466,
+      "loss/crossentropy": 1.7475664019584656,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21757815405726433,
+      "step": 1640
+    },
+    {
+      "epoch": 0.2735,
+      "grad_norm": 29.875,
+      "grad_norm_var": 4.248893229166667,
+      "learning_rate": 8.267103019950529e-05,
+      "loss": 7.1966,
+      "loss/crossentropy": 2.1399194598197937,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.212381724268198,
+      "step": 1641
+    },
+    {
+      "epoch": 0.27366666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 4.162434895833333,
+      "learning_rate": 8.265120763772303e-05,
+      "loss": 6.8291,
+      "loss/crossentropy": 2.0715277791023254,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17723426967859268,
+      "step": 1642
+    },
+    {
+      "epoch": 0.2738333333333333,
+      "grad_norm": 27.875,
+      "grad_norm_var": 4.0650390625,
+      "learning_rate": 8.263137612442706e-05,
+      "loss": 7.0393,
+      "loss/crossentropy": 1.6310051381587982,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15922663547098637,
+      "step": 1643
+    },
+    {
+      "epoch": 0.274,
+      "grad_norm": 27.75,
+      "grad_norm_var": 3.803125,
+      "learning_rate": 8.261153566505424e-05,
+      "loss": 7.3023,
+      "loss/crossentropy": 1.6214478015899658,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23836766555905342,
+      "step": 1644
+    },
+    {
+      "epoch": 0.27416666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.7494140625,
+      "learning_rate": 8.259168626504395e-05,
+      "loss": 6.6836,
+      "loss/crossentropy": 1.8864246904850006,
+      "loss/hidden": 3.07421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15652998350560665,
+      "step": 1645
+    },
+    {
+      "epoch": 0.2743333333333333,
+      "grad_norm": 28.0,
+      "grad_norm_var": 3.82890625,
+      "learning_rate": 8.257182792983802e-05,
+      "loss": 6.9393,
+      "loss/crossentropy": 1.140328787267208,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14030886627733707,
+      "step": 1646
+    },
+    {
+      "epoch": 0.2745,
+      "grad_norm": 38.25,
+      "grad_norm_var": 8.85625,
+      "learning_rate": 8.255196066488075e-05,
+      "loss": 7.0095,
+      "loss/crossentropy": 1.549303650856018,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17772993631660938,
+      "step": 1647
+    },
+    {
+      "epoch": 0.27466666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 8.750455729166667,
+      "learning_rate": 8.253208447561882e-05,
+      "loss": 7.2091,
+      "loss/crossentropy": 2.370518445968628,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2150629200041294,
+      "step": 1648
+    },
+    {
+      "epoch": 0.2748333333333333,
+      "grad_norm": 27.625,
+      "grad_norm_var": 8.8119140625,
+      "learning_rate": 8.251219936750144e-05,
+      "loss": 6.8833,
+      "loss/crossentropy": 1.4538423269987106,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14347987715154886,
+      "step": 1649
+    },
+    {
+      "epoch": 0.275,
+      "grad_norm": 27.875,
+      "grad_norm_var": 8.007291666666667,
+      "learning_rate": 8.249230534598021e-05,
+      "loss": 6.9361,
+      "loss/crossentropy": 1.7133109271526337,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2329830899834633,
+      "step": 1650
+    },
+    {
+      "epoch": 0.27516666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 7.841666666666667,
+      "learning_rate": 8.247240241650918e-05,
+      "loss": 7.0463,
+      "loss/crossentropy": 1.3484934717416763,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22996426187455654,
+      "step": 1651
+    },
+    {
+      "epoch": 0.2753333333333333,
+      "grad_norm": 26.5,
+      "grad_norm_var": 8.451822916666666,
+      "learning_rate": 8.245249058454487e-05,
+      "loss": 6.7415,
+      "loss/crossentropy": 1.9782699048519135,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17345220781862736,
+      "step": 1652
+    },
+    {
+      "epoch": 0.2755,
+      "grad_norm": 29.75,
+      "grad_norm_var": 7.8837890625,
+      "learning_rate": 8.243256985554621e-05,
+      "loss": 7.2286,
+      "loss/crossentropy": 1.7965165376663208,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18303445540368557,
+      "step": 1653
+    },
+    {
+      "epoch": 0.27566666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 8.53125,
+      "learning_rate": 8.241264023497457e-05,
+      "loss": 7.0887,
+      "loss/crossentropy": 1.3913812637329102,
+      "loss/hidden": 3.8828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27453312277793884,
+      "step": 1654
+    },
+    {
+      "epoch": 0.2758333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 7.790559895833334,
+      "learning_rate": 8.239270172829379e-05,
+      "loss": 7.1735,
+      "loss/crossentropy": 1.8213527202606201,
+      "loss/hidden": 3.87109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2634476125240326,
+      "step": 1655
+    },
+    {
+      "epoch": 0.276,
+      "grad_norm": 28.125,
+      "grad_norm_var": 7.921809895833333,
+      "learning_rate": 8.237275434097012e-05,
+      "loss": 7.1436,
+      "loss/crossentropy": 1.782892495393753,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20107223093509674,
+      "step": 1656
+    },
+    {
+      "epoch": 0.27616666666666667,
+      "grad_norm": 27.0,
+      "grad_norm_var": 8.339583333333334,
+      "learning_rate": 8.235279807847223e-05,
+      "loss": 7.039,
+      "loss/crossentropy": 1.8401851952075958,
+      "loss/hidden": 3.04296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1714758574962616,
+      "step": 1657
+    },
+    {
+      "epoch": 0.2763333333333333,
+      "grad_norm": 27.875,
+      "grad_norm_var": 8.492122395833333,
+      "learning_rate": 8.233283294627125e-05,
+      "loss": 7.1463,
+      "loss/crossentropy": 1.5835787653923035,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13180573284626007,
+      "step": 1658
+    },
+    {
+      "epoch": 0.2765,
+      "grad_norm": 29.0,
+      "grad_norm_var": 8.349739583333333,
+      "learning_rate": 8.231285894984076e-05,
+      "loss": 7.1677,
+      "loss/crossentropy": 1.4388826042413712,
+      "loss/hidden": 3.84765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2233562022447586,
+      "step": 1659
+    },
+    {
+      "epoch": 0.27666666666666667,
+      "grad_norm": 26.625,
+      "grad_norm_var": 8.679622395833333,
+      "learning_rate": 8.22928760946567e-05,
+      "loss": 6.6824,
+      "loss/crossentropy": 1.5648126006126404,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.130872605368495,
+      "step": 1660
+    },
+    {
+      "epoch": 0.2768333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 9.340559895833334,
+      "learning_rate": 8.227288438619754e-05,
+      "loss": 6.7709,
+      "loss/crossentropy": 1.7629970461130142,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20937180891633034,
+      "step": 1661
+    },
+    {
+      "epoch": 0.277,
+      "grad_norm": 39.0,
+      "grad_norm_var": 14.714518229166666,
+      "learning_rate": 8.225288382994407e-05,
+      "loss": 6.6848,
+      "loss/crossentropy": 2.10350701212883,
+      "loss/hidden": 2.97265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1433289349079132,
+      "step": 1662
+    },
+    {
+      "epoch": 0.2771666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 10.09765625,
+      "learning_rate": 8.223287443137957e-05,
+      "loss": 7.2117,
+      "loss/crossentropy": 2.2686829268932343,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1735362634062767,
+      "step": 1663
+    },
+    {
+      "epoch": 0.2773333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 10.095247395833333,
+      "learning_rate": 8.221285619598975e-05,
+      "loss": 7.2918,
+      "loss/crossentropy": 1.8061031699180603,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23730029165744781,
+      "step": 1664
+    },
+    {
+      "epoch": 0.2775,
+      "grad_norm": 26.75,
+      "grad_norm_var": 10.380989583333333,
+      "learning_rate": 8.21928291292627e-05,
+      "loss": 6.6875,
+      "loss/crossentropy": 1.4453921169042587,
+      "loss/hidden": 3.125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1263937372714281,
+      "step": 1665
+    },
+    {
+      "epoch": 0.2776666666666667,
+      "grad_norm": 27.875,
+      "grad_norm_var": 10.380989583333333,
+      "learning_rate": 8.217279323668895e-05,
+      "loss": 6.92,
+      "loss/crossentropy": 1.6042270958423615,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16979801282286644,
+      "step": 1666
+    },
+    {
+      "epoch": 0.2778333333333333,
+      "grad_norm": 26.875,
+      "grad_norm_var": 10.843684895833333,
+      "learning_rate": 8.215274852376147e-05,
+      "loss": 7.0608,
+      "loss/crossentropy": 1.7220258712768555,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1650107428431511,
+      "step": 1667
+    },
+    {
+      "epoch": 0.278,
+      "grad_norm": 27.375,
+      "grad_norm_var": 10.549739583333333,
+      "learning_rate": 8.213269499597565e-05,
+      "loss": 7.1072,
+      "loss/crossentropy": 1.8538067936897278,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23773017898201942,
+      "step": 1668
+    },
+    {
+      "epoch": 0.2781666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 10.851822916666666,
+      "learning_rate": 8.211263265882923e-05,
+      "loss": 6.6361,
+      "loss/crossentropy": 1.5824365466833115,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17740889638662338,
+      "step": 1669
+    },
+    {
+      "epoch": 0.2783333333333333,
+      "grad_norm": 49.75,
+      "grad_norm_var": 36.39479166666667,
+      "learning_rate": 8.209256151782243e-05,
+      "loss": 7.1143,
+      "loss/crossentropy": 1.8776679635047913,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18182456865906715,
+      "step": 1670
+    },
+    {
+      "epoch": 0.2785,
+      "grad_norm": 36.5,
+      "grad_norm_var": 38.64733072916667,
+      "learning_rate": 8.207248157845791e-05,
+      "loss": 6.8501,
+      "loss/crossentropy": 1.8925500214099884,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1794169433414936,
+      "step": 1671
+    },
+    {
+      "epoch": 0.2786666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 38.145572916666666,
+      "learning_rate": 8.205239284624062e-05,
+      "loss": 7.0071,
+      "loss/crossentropy": 1.4744906574487686,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.155276071280241,
+      "step": 1672
+    },
+    {
+      "epoch": 0.2788333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 37.0416015625,
+      "learning_rate": 8.203229532667807e-05,
+      "loss": 7.2046,
+      "loss/crossentropy": 1.9407618343830109,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1773352026939392,
+      "step": 1673
+    },
+    {
+      "epoch": 0.279,
+      "grad_norm": 30.5,
+      "grad_norm_var": 36.31015625,
+      "learning_rate": 8.201218902528009e-05,
+      "loss": 6.936,
+      "loss/crossentropy": 1.741473138332367,
+      "loss/hidden": 3.7890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22419481724500656,
+      "step": 1674
+    },
+    {
+      "epoch": 0.2791666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 35.9150390625,
+      "learning_rate": 8.199207394755893e-05,
+      "loss": 7.0361,
+      "loss/crossentropy": 1.8393172919750214,
+      "loss/hidden": 3.72265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2207166701555252,
+      "step": 1675
+    },
+    {
+      "epoch": 0.2793333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 34.7619140625,
+      "learning_rate": 8.197195009902924e-05,
+      "loss": 7.0547,
+      "loss/crossentropy": 2.02646142244339,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16825808957219124,
+      "step": 1676
+    },
+    {
+      "epoch": 0.2795,
+      "grad_norm": 26.125,
+      "grad_norm_var": 36.545833333333334,
+      "learning_rate": 8.195181748520811e-05,
+      "loss": 7.1419,
+      "loss/crossentropy": 1.6126716583967209,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18254036456346512,
+      "step": 1677
+    },
+    {
+      "epoch": 0.2796666666666667,
+      "grad_norm": 27.625,
+      "grad_norm_var": 32.878580729166664,
+      "learning_rate": 8.193167611161499e-05,
+      "loss": 6.7845,
+      "loss/crossentropy": 1.7171251773834229,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1587543785572052,
+      "step": 1678
+    },
+    {
+      "epoch": 0.2798333333333333,
+      "grad_norm": 27.25,
+      "grad_norm_var": 33.55390625,
+      "learning_rate": 8.191152598377178e-05,
+      "loss": 6.8725,
+      "loss/crossentropy": 1.5072792768478394,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13671744614839554,
+      "step": 1679
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 34.75,
+      "grad_norm_var": 34.71015625,
+      "learning_rate": 8.189136710720272e-05,
+      "loss": 7.3084,
+      "loss/crossentropy": 1.8664146661758423,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1888437643647194,
+      "step": 1680
+    },
+    {
+      "epoch": 0.2801666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 33.62233072916667,
+      "learning_rate": 8.18711994874345e-05,
+      "loss": 7.3425,
+      "loss/crossentropy": 1.6438325941562653,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17498685792088509,
+      "step": 1681
+    },
+    {
+      "epoch": 0.2803333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 33.0619140625,
+      "learning_rate": 8.185102312999617e-05,
+      "loss": 7.1414,
+      "loss/crossentropy": 1.7217102944850922,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1634957641363144,
+      "step": 1682
+    },
+    {
+      "epoch": 0.2805,
+      "grad_norm": 28.625,
+      "grad_norm_var": 32.28170572916667,
+      "learning_rate": 8.183083804041921e-05,
+      "loss": 7.0944,
+      "loss/crossentropy": 1.051482379436493,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1322503425180912,
+      "step": 1683
+    },
+    {
+      "epoch": 0.2806666666666667,
+      "grad_norm": 26.125,
+      "grad_norm_var": 33.00826822916667,
+      "learning_rate": 8.181064422423748e-05,
+      "loss": 6.837,
+      "loss/crossentropy": 1.4523184299468994,
+      "loss/hidden": 3.86328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17103984951972961,
+      "step": 1684
+    },
+    {
+      "epoch": 0.2808333333333333,
+      "grad_norm": 27.5,
+      "grad_norm_var": 32.884830729166666,
+      "learning_rate": 8.179044168698721e-05,
+      "loss": 6.7853,
+      "loss/crossentropy": 1.4608682692050934,
+      "loss/hidden": 3.12109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15748637914657593,
+      "step": 1685
+    },
+    {
+      "epoch": 0.281,
+      "grad_norm": 31.5,
+      "grad_norm_var": 8.2853515625,
+      "learning_rate": 8.177023043420705e-05,
+      "loss": 7.1377,
+      "loss/crossentropy": 1.8704125732183456,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23739512264728546,
+      "step": 1686
+    },
+    {
+      "epoch": 0.2811666666666667,
+      "grad_norm": 27.625,
+      "grad_norm_var": 5.451822916666667,
+      "learning_rate": 8.175001047143804e-05,
+      "loss": 6.896,
+      "loss/crossentropy": 1.7419240027666092,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16018554382026196,
+      "step": 1687
+    },
+    {
+      "epoch": 0.2813333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 5.322916666666667,
+      "learning_rate": 8.172978180422358e-05,
+      "loss": 7.5015,
+      "loss/crossentropy": 1.5139735639095306,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31741153448820114,
+      "step": 1688
+    },
+    {
+      "epoch": 0.2815,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.826497395833333,
+      "learning_rate": 8.170954443810948e-05,
+      "loss": 7.0073,
+      "loss/crossentropy": 1.8012758195400238,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15870653837919235,
+      "step": 1689
+    },
+    {
+      "epoch": 0.2816666666666667,
+      "grad_norm": 27.5,
+      "grad_norm_var": 5.954622395833334,
+      "learning_rate": 8.168929837864395e-05,
+      "loss": 6.7713,
+      "loss/crossentropy": 1.7050330489873886,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16706456243991852,
+      "step": 1690
+    },
+    {
+      "epoch": 0.2818333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 5.545768229166667,
+      "learning_rate": 8.16690436313775e-05,
+      "loss": 7.1957,
+      "loss/crossentropy": 1.9629746973514557,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1816374473273754,
+      "step": 1691
+    },
+    {
+      "epoch": 0.282,
+      "grad_norm": 26.875,
+      "grad_norm_var": 5.8478515625,
+      "learning_rate": 8.164878020186317e-05,
+      "loss": 6.7097,
+      "loss/crossentropy": 1.6327789425849915,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25625718757510185,
+      "step": 1692
+    },
+    {
+      "epoch": 0.2821666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 5.343684895833333,
+      "learning_rate": 8.162850809565623e-05,
+      "loss": 7.0074,
+      "loss/crossentropy": 1.5538465976715088,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16084902733564377,
+      "step": 1693
+    },
+    {
+      "epoch": 0.2823333333333333,
+      "grad_norm": 27.875,
+      "grad_norm_var": 5.295247395833333,
+      "learning_rate": 8.160822731831441e-05,
+      "loss": 6.9638,
+      "loss/crossentropy": 1.6474287658929825,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18181404285132885,
+      "step": 1694
+    },
+    {
+      "epoch": 0.2825,
+      "grad_norm": 27.625,
+      "grad_norm_var": 5.205989583333333,
+      "learning_rate": 8.158793787539782e-05,
+      "loss": 7.0905,
+      "loss/crossentropy": 1.9614308178424835,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23940258100628853,
+      "step": 1695
+    },
+    {
+      "epoch": 0.2826666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.0785807291666667,
+      "learning_rate": 8.156763977246889e-05,
+      "loss": 7.084,
+      "loss/crossentropy": 1.6980135440826416,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1318059153854847,
+      "step": 1696
+    },
+    {
+      "epoch": 0.2828333333333333,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.9567057291666665,
+      "learning_rate": 8.154733301509248e-05,
+      "loss": 7.2798,
+      "loss/crossentropy": 2.285195052623749,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17841573804616928,
+      "step": 1697
+    },
+    {
+      "epoch": 0.283,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.8854166666666665,
+      "learning_rate": 8.152701760883581e-05,
+      "loss": 6.84,
+      "loss/crossentropy": 1.8738844096660614,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17672141268849373,
+      "step": 1698
+    },
+    {
+      "epoch": 0.2831666666666667,
+      "grad_norm": 37.5,
+      "grad_norm_var": 7.734309895833333,
+      "learning_rate": 8.150669355926846e-05,
+      "loss": 7.2175,
+      "loss/crossentropy": 1.678205981850624,
+      "loss/hidden": 3.85546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1782095469534397,
+      "step": 1699
+    },
+    {
+      "epoch": 0.2833333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 7.338997395833333,
+      "learning_rate": 8.148636087196237e-05,
+      "loss": 7.3357,
+      "loss/crossentropy": 1.5552005618810654,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18490111827850342,
+      "step": 1700
+    },
+    {
+      "epoch": 0.2835,
+      "grad_norm": 30.875,
+      "grad_norm_var": 7.112239583333333,
+      "learning_rate": 8.146601955249188e-05,
+      "loss": 7.1409,
+      "loss/crossentropy": 1.2435659170150757,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18200484290719032,
+      "step": 1701
+    },
+    {
+      "epoch": 0.2836666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 6.908072916666667,
+      "learning_rate": 8.144566960643367e-05,
+      "loss": 7.5389,
+      "loss/crossentropy": 1.9872223436832428,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2747933268547058,
+      "step": 1702
+    },
+    {
+      "epoch": 0.2838333333333333,
+      "grad_norm": 27.25,
+      "grad_norm_var": 7.019205729166667,
+      "learning_rate": 8.142531103936678e-05,
+      "loss": 7.2176,
+      "loss/crossentropy": 1.9800648391246796,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20019137114286423,
+      "step": 1703
+    },
+    {
+      "epoch": 0.284,
+      "grad_norm": 25.0,
+      "grad_norm_var": 8.373893229166667,
+      "learning_rate": 8.140494385687265e-05,
+      "loss": 6.7664,
+      "loss/crossentropy": 1.6927202641963959,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18261635676026344,
+      "step": 1704
+    },
+    {
+      "epoch": 0.2841666666666667,
+      "grad_norm": 38.25,
+      "grad_norm_var": 12.842122395833334,
+      "learning_rate": 8.138456806453503e-05,
+      "loss": 7.0783,
+      "loss/crossentropy": 1.5268690437078476,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13206701911985874,
+      "step": 1705
+    },
+    {
+      "epoch": 0.2843333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 12.524739583333334,
+      "learning_rate": 8.136418366794008e-05,
+      "loss": 6.8297,
+      "loss/crossentropy": 1.8745198845863342,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20556993037462234,
+      "step": 1706
+    },
+    {
+      "epoch": 0.2845,
+      "grad_norm": 27.25,
+      "grad_norm_var": 12.920247395833334,
+      "learning_rate": 8.13437906726763e-05,
+      "loss": 6.8514,
+      "loss/crossentropy": 1.9250228106975555,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2539803571999073,
+      "step": 1707
+    },
+    {
+      "epoch": 0.2846666666666667,
+      "grad_norm": 27.375,
+      "grad_norm_var": 12.746809895833334,
+      "learning_rate": 8.132338908433454e-05,
+      "loss": 7.3852,
+      "loss/crossentropy": 1.8892545402050018,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16167163476347923,
+      "step": 1708
+    },
+    {
+      "epoch": 0.2848333333333333,
+      "grad_norm": 27.75,
+      "grad_norm_var": 12.978125,
+      "learning_rate": 8.130297890850802e-05,
+      "loss": 7.048,
+      "loss/crossentropy": 1.3800558298826218,
+      "loss/hidden": 3.73828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17052343860268593,
+      "step": 1709
+    },
+    {
+      "epoch": 0.285,
+      "grad_norm": 27.375,
+      "grad_norm_var": 13.108333333333333,
+      "learning_rate": 8.128256015079229e-05,
+      "loss": 7.1962,
+      "loss/crossentropy": 1.8310509622097015,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17635131999850273,
+      "step": 1710
+    },
+    {
+      "epoch": 0.2851666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 13.334309895833334,
+      "learning_rate": 8.126213281678528e-05,
+      "loss": 6.7291,
+      "loss/crossentropy": 0.8421642333269119,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.08007671311497688,
+      "step": 1711
+    },
+    {
+      "epoch": 0.2853333333333333,
+      "grad_norm": 35.0,
+      "grad_norm_var": 14.96640625,
+      "learning_rate": 8.124169691208723e-05,
+      "loss": 7.0254,
+      "loss/crossentropy": 1.3119288682937622,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14421034418046474,
+      "step": 1712
+    },
+    {
+      "epoch": 0.2855,
+      "grad_norm": 28.75,
+      "grad_norm_var": 14.6587890625,
+      "learning_rate": 8.122125244230079e-05,
+      "loss": 6.9844,
+      "loss/crossentropy": 1.2713851630687714,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13534276001155376,
+      "step": 1713
+    },
+    {
+      "epoch": 0.2856666666666667,
+      "grad_norm": 27.75,
+      "grad_norm_var": 15.0041015625,
+      "learning_rate": 8.120079941303094e-05,
+      "loss": 6.9775,
+      "loss/crossentropy": 2.193632423877716,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20735756680369377,
+      "step": 1714
+    },
+    {
+      "epoch": 0.28583333333333333,
+      "grad_norm": 26.875,
+      "grad_norm_var": 11.689322916666667,
+      "learning_rate": 8.118033782988496e-05,
+      "loss": 7.1397,
+      "loss/crossentropy": 1.366313859820366,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17892462760210037,
+      "step": 1715
+    },
+    {
+      "epoch": 0.286,
+      "grad_norm": 29.625,
+      "grad_norm_var": 11.376822916666667,
+      "learning_rate": 8.115986769847252e-05,
+      "loss": 7.3585,
+      "loss/crossentropy": 1.5995260626077652,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17269459553062916,
+      "step": 1716
+    },
+    {
+      "epoch": 0.2861666666666667,
+      "grad_norm": 26.375,
+      "grad_norm_var": 11.751822916666667,
+      "learning_rate": 8.113938902440564e-05,
+      "loss": 6.557,
+      "loss/crossentropy": 1.2278249189257622,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09960130974650383,
+      "step": 1717
+    },
+    {
+      "epoch": 0.28633333333333333,
+      "grad_norm": 26.875,
+      "grad_norm_var": 12.045768229166667,
+      "learning_rate": 8.111890181329863e-05,
+      "loss": 6.6644,
+      "loss/crossentropy": 1.2437103986740112,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17286989092826843,
+      "step": 1718
+    },
+    {
+      "epoch": 0.2865,
+      "grad_norm": 28.125,
+      "grad_norm_var": 11.895833333333334,
+      "learning_rate": 8.109840607076821e-05,
+      "loss": 6.9311,
+      "loss/crossentropy": 1.6388500034809113,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18063346669077873,
+      "step": 1719
+    },
+    {
+      "epoch": 0.2866666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 11.1056640625,
+      "learning_rate": 8.107790180243338e-05,
+      "loss": 6.6224,
+      "loss/crossentropy": 1.457134634256363,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1901661977171898,
+      "step": 1720
+    },
+    {
+      "epoch": 0.28683333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 5.81640625,
+      "learning_rate": 8.105738901391552e-05,
+      "loss": 6.8834,
+      "loss/crossentropy": 1.772074967622757,
+      "loss/hidden": 3.78125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2563551850616932,
+      "step": 1721
+    },
+    {
+      "epoch": 0.287,
+      "grad_norm": 26.125,
+      "grad_norm_var": 6.31015625,
+      "learning_rate": 8.103686771083831e-05,
+      "loss": 6.743,
+      "loss/crossentropy": 1.562201976776123,
+      "loss/hidden": 3.765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2726549804210663,
+      "step": 1722
+    },
+    {
+      "epoch": 0.2871666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 6.182747395833333,
+      "learning_rate": 8.101633789882781e-05,
+      "loss": 7.1795,
+      "loss/crossentropy": 1.4713617414236069,
+      "loss/hidden": 3.75,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22786082699894905,
+      "step": 1723
+    },
+    {
+      "epoch": 0.28733333333333333,
+      "grad_norm": 26.125,
+      "grad_norm_var": 6.5369140625,
+      "learning_rate": 8.099579958351235e-05,
+      "loss": 7.1679,
+      "loss/crossentropy": 2.030321568250656,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2075900286436081,
+      "step": 1724
+    },
+    {
+      "epoch": 0.2875,
+      "grad_norm": 25.75,
+      "grad_norm_var": 7.076497395833333,
+      "learning_rate": 8.097525277052264e-05,
+      "loss": 6.9194,
+      "loss/crossentropy": 1.3708519041538239,
+      "loss/hidden": 4.01953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19819754734635353,
+      "step": 1725
+    },
+    {
+      "epoch": 0.2876666666666667,
+      "grad_norm": 26.75,
+      "grad_norm_var": 7.212239583333333,
+      "learning_rate": 8.095469746549172e-05,
+      "loss": 6.8428,
+      "loss/crossentropy": 2.2041515707969666,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1770605370402336,
+      "step": 1726
+    },
+    {
+      "epoch": 0.28783333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 6.213541666666667,
+      "learning_rate": 8.093413367405489e-05,
+      "loss": 7.1614,
+      "loss/crossentropy": 1.310487300157547,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14241445809602737,
+      "step": 1727
+    },
+    {
+      "epoch": 0.288,
+      "grad_norm": 27.5,
+      "grad_norm_var": 3.1979166666666665,
+      "learning_rate": 8.091356140184991e-05,
+      "loss": 6.8751,
+      "loss/crossentropy": 1.2544075399637222,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1395467333495617,
+      "step": 1728
+    },
+    {
+      "epoch": 0.2881666666666667,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.186393229166667,
+      "learning_rate": 8.089298065451672e-05,
+      "loss": 7.0956,
+      "loss/crossentropy": 1.4277315139770508,
+      "loss/hidden": 3.96484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20680862292647362,
+      "step": 1729
+    },
+    {
+      "epoch": 0.28833333333333333,
+      "grad_norm": 27.5,
+      "grad_norm_var": 3.198372395833333,
+      "learning_rate": 8.087239143769768e-05,
+      "loss": 7.1275,
+      "loss/crossentropy": 2.199025958776474,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18593628332018852,
+      "step": 1730
+    },
+    {
+      "epoch": 0.2885,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.802083333333333,
+      "learning_rate": 8.085179375703744e-05,
+      "loss": 7.1663,
+      "loss/crossentropy": 1.5327014029026031,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1512337662279606,
+      "step": 1731
+    },
+    {
+      "epoch": 0.2886666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 4.9306640625,
+      "learning_rate": 8.083118761818295e-05,
+      "loss": 7.0301,
+      "loss/crossentropy": 2.1589335799217224,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1671556904911995,
+      "step": 1732
+    },
+    {
+      "epoch": 0.28883333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 4.7447265625,
+      "learning_rate": 8.081057302678352e-05,
+      "loss": 7.0091,
+      "loss/crossentropy": 1.6066819727420807,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16801932267844677,
+      "step": 1733
+    },
+    {
+      "epoch": 0.289,
+      "grad_norm": 28.375,
+      "grad_norm_var": 4.5337890625,
+      "learning_rate": 8.078994998849076e-05,
+      "loss": 6.8769,
+      "loss/crossentropy": 1.2998749762773514,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22766570188105106,
+      "step": 1734
+    },
+    {
+      "epoch": 0.2891666666666667,
+      "grad_norm": 27.75,
+      "grad_norm_var": 4.57265625,
+      "learning_rate": 8.076931850895859e-05,
+      "loss": 6.9313,
+      "loss/crossentropy": 1.5506720840930939,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21129202470183372,
+      "step": 1735
+    },
+    {
+      "epoch": 0.28933333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 5.0900390625,
+      "learning_rate": 8.074867859384322e-05,
+      "loss": 7.1436,
+      "loss/crossentropy": 1.9704915881156921,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17415691912174225,
+      "step": 1736
+    },
+    {
+      "epoch": 0.2895,
+      "grad_norm": 27.125,
+      "grad_norm_var": 4.9181640625,
+      "learning_rate": 8.072803024880322e-05,
+      "loss": 7.0268,
+      "loss/crossentropy": 1.365415871143341,
+      "loss/hidden": 3.92578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19732052460312843,
+      "step": 1737
+    },
+    {
+      "epoch": 0.2896666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 4.63515625,
+      "learning_rate": 8.070737347949947e-05,
+      "loss": 7.1417,
+      "loss/crossentropy": 1.8379714488983154,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.179561085999012,
+      "step": 1738
+    },
+    {
+      "epoch": 0.28983333333333333,
+      "grad_norm": 26.75,
+      "grad_norm_var": 4.762434895833334,
+      "learning_rate": 8.068670829159511e-05,
+      "loss": 7.0279,
+      "loss/crossentropy": 1.8199576884508133,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16702722012996674,
+      "step": 1739
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 27.125,
+      "grad_norm_var": 4.517643229166667,
+      "learning_rate": 8.066603469075564e-05,
+      "loss": 7.2574,
+      "loss/crossentropy": 1.8788958191871643,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21776703000068665,
+      "step": 1740
+    },
+    {
+      "epoch": 0.2901666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.152018229166667,
+      "learning_rate": 8.064535268264883e-05,
+      "loss": 7.0703,
+      "loss/crossentropy": 1.4729944616556168,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17053206637501717,
+      "step": 1741
+    },
+    {
+      "epoch": 0.29033333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.5353515625,
+      "learning_rate": 8.062466227294477e-05,
+      "loss": 7.3585,
+      "loss/crossentropy": 1.5924350023269653,
+      "loss/hidden": 4.07421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3067648336291313,
+      "step": 1742
+    },
+    {
+      "epoch": 0.2905,
+      "grad_norm": 34.0,
+      "grad_norm_var": 6.910872395833334,
+      "learning_rate": 8.060396346731587e-05,
+      "loss": 7.3979,
+      "loss/crossentropy": 1.8360927402973175,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.171937869861722,
+      "step": 1743
+    },
+    {
+      "epoch": 0.2906666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 6.64375,
+      "learning_rate": 8.058325627143681e-05,
+      "loss": 6.9895,
+      "loss/crossentropy": 2.1419885456562042,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2188923880457878,
+      "step": 1744
+    },
+    {
+      "epoch": 0.29083333333333333,
+      "grad_norm": 27.0,
+      "grad_norm_var": 7.038997395833333,
+      "learning_rate": 8.056254069098459e-05,
+      "loss": 7.085,
+      "loss/crossentropy": 1.8869548439979553,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19913369789719582,
+      "step": 1745
+    },
+    {
+      "epoch": 0.291,
+      "grad_norm": 27.125,
+      "grad_norm_var": 7.152083333333334,
+      "learning_rate": 8.05418167316385e-05,
+      "loss": 6.9107,
+      "loss/crossentropy": 1.7835271060466766,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21418502554297447,
+      "step": 1746
+    },
+    {
+      "epoch": 0.2911666666666667,
+      "grad_norm": 26.875,
+      "grad_norm_var": 6.557747395833333,
+      "learning_rate": 8.052108439908013e-05,
+      "loss": 7.2438,
+      "loss/crossentropy": 1.4813159108161926,
+      "loss/hidden": 3.81640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2985213175415993,
+      "step": 1747
+    },
+    {
+      "epoch": 0.29133333333333333,
+      "grad_norm": 27.375,
+      "grad_norm_var": 6.658072916666667,
+      "learning_rate": 8.050034369899337e-05,
+      "loss": 6.669,
+      "loss/crossentropy": 1.5193503946065903,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15331173688173294,
+      "step": 1748
+    },
+    {
+      "epoch": 0.2915,
+      "grad_norm": 30.125,
+      "grad_norm_var": 6.691666666666666,
+      "learning_rate": 8.04795946370644e-05,
+      "loss": 7.1138,
+      "loss/crossentropy": 1.9641196131706238,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29699448496103287,
+      "step": 1749
+    },
+    {
+      "epoch": 0.2916666666666667,
+      "grad_norm": 29.125,
+      "grad_norm_var": 6.664322916666666,
+      "learning_rate": 8.04588372189817e-05,
+      "loss": 7.4095,
+      "loss/crossentropy": 2.076164960861206,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2311621755361557,
+      "step": 1750
+    },
+    {
+      "epoch": 0.29183333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 7.60390625,
+      "learning_rate": 8.043807145043604e-05,
+      "loss": 6.7987,
+      "loss/crossentropy": 2.2085287868976593,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16484244912862778,
+      "step": 1751
+    },
+    {
+      "epoch": 0.292,
+      "grad_norm": 29.25,
+      "grad_norm_var": 6.801822916666667,
+      "learning_rate": 8.041729733712045e-05,
+      "loss": 6.9456,
+      "loss/crossentropy": 1.555240511894226,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18162141367793083,
+      "step": 1752
+    },
+    {
+      "epoch": 0.2921666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 6.8041015625,
+      "learning_rate": 8.039651488473028e-05,
+      "loss": 7.1927,
+      "loss/crossentropy": 1.4920425415039062,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20788925141096115,
+      "step": 1753
+    },
+    {
+      "epoch": 0.29233333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 6.512955729166666,
+      "learning_rate": 8.037572409896315e-05,
+      "loss": 7.1723,
+      "loss/crossentropy": 1.5984963774681091,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16287313401699066,
+      "step": 1754
+    },
+    {
+      "epoch": 0.2925,
+      "grad_norm": 27.0,
+      "grad_norm_var": 6.420247395833333,
+      "learning_rate": 8.0354924985519e-05,
+      "loss": 6.6847,
+      "loss/crossentropy": 1.2762255370616913,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14174848049879074,
+      "step": 1755
+    },
+    {
+      "epoch": 0.2926666666666667,
+      "grad_norm": 27.5,
+      "grad_norm_var": 6.302083333333333,
+      "learning_rate": 8.033411755009999e-05,
+      "loss": 6.9856,
+      "loss/crossentropy": 1.941700428724289,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21694884449243546,
+      "step": 1756
+    },
+    {
+      "epoch": 0.29283333333333333,
+      "grad_norm": 27.125,
+      "grad_norm_var": 5.864518229166666,
+      "learning_rate": 8.031330179841062e-05,
+      "loss": 7.0468,
+      "loss/crossentropy": 1.6273793876171112,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.282911229878664,
+      "step": 1757
+    },
+    {
+      "epoch": 0.293,
+      "grad_norm": 28.0,
+      "grad_norm_var": 5.222330729166667,
+      "learning_rate": 8.029247773615764e-05,
+      "loss": 6.5805,
+      "loss/crossentropy": 1.6457321643829346,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17904561571776867,
+      "step": 1758
+    },
+    {
+      "epoch": 0.2931666666666667,
+      "grad_norm": 26.625,
+      "grad_norm_var": 3.7434895833333335,
+      "learning_rate": 8.027164536905008e-05,
+      "loss": 7.2678,
+      "loss/crossentropy": 1.6576599180698395,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2535715512931347,
+      "step": 1759
+    },
+    {
+      "epoch": 0.29333333333333333,
+      "grad_norm": 26.125,
+      "grad_norm_var": 4.107291666666667,
+      "learning_rate": 8.025080470279924e-05,
+      "loss": 6.6432,
+      "loss/crossentropy": 1.279824674129486,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.172175120562315,
+      "step": 1760
+    },
+    {
+      "epoch": 0.2935,
+      "grad_norm": 30.625,
+      "grad_norm_var": 4.248893229166667,
+      "learning_rate": 8.022995574311876e-05,
+      "loss": 6.8749,
+      "loss/crossentropy": 1.4916917234659195,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1399658117443323,
+      "step": 1761
+    },
+    {
+      "epoch": 0.2936666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.34140625,
+      "learning_rate": 8.020909849572444e-05,
+      "loss": 6.7808,
+      "loss/crossentropy": 1.6238489598035812,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17868554964661598,
+      "step": 1762
+    },
+    {
+      "epoch": 0.29383333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 4.124934895833333,
+      "learning_rate": 8.018823296633441e-05,
+      "loss": 6.8956,
+      "loss/crossentropy": 1.6868852376937866,
+      "loss/hidden": 3.98046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21477436274290085,
+      "step": 1763
+    },
+    {
+      "epoch": 0.294,
+      "grad_norm": 28.0,
+      "grad_norm_var": 4.009375,
+      "learning_rate": 8.016735916066913e-05,
+      "loss": 6.901,
+      "loss/crossentropy": 1.3365189731121063,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15891896560788155,
+      "step": 1764
+    },
+    {
+      "epoch": 0.2941666666666667,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.934375,
+      "learning_rate": 8.014647708445124e-05,
+      "loss": 7.2008,
+      "loss/crossentropy": 1.694892257452011,
+      "loss/hidden": 3.8046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2946743369102478,
+      "step": 1765
+    },
+    {
+      "epoch": 0.29433333333333334,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.934375,
+      "learning_rate": 8.012558674340566e-05,
+      "loss": 7.4754,
+      "loss/crossentropy": 1.6758044362068176,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2702377289533615,
+      "step": 1766
+    },
+    {
+      "epoch": 0.2945,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.7478515625,
+      "learning_rate": 8.010468814325964e-05,
+      "loss": 6.9268,
+      "loss/crossentropy": 1.418891355395317,
+      "loss/hidden": 3.9765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2032007835805416,
+      "step": 1767
+    },
+    {
+      "epoch": 0.2946666666666667,
+      "grad_norm": 25.125,
+      "grad_norm_var": 3.49765625,
+      "learning_rate": 8.008378128974262e-05,
+      "loss": 6.7153,
+      "loss/crossentropy": 1.5259346812963486,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1468068603426218,
+      "step": 1768
+    },
+    {
+      "epoch": 0.29483333333333334,
+      "grad_norm": 35.0,
+      "grad_norm_var": 5.699739583333334,
+      "learning_rate": 8.006286618858635e-05,
+      "loss": 7.4526,
+      "loss/crossentropy": 1.8264343738555908,
+      "loss/hidden": 3.74609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3308655321598053,
+      "step": 1769
+    },
+    {
+      "epoch": 0.295,
+      "grad_norm": 30.125,
+      "grad_norm_var": 5.615559895833333,
+      "learning_rate": 8.004194284552477e-05,
+      "loss": 7.196,
+      "loss/crossentropy": 1.5043246895074844,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2522316053509712,
+      "step": 1770
+    },
+    {
+      "epoch": 0.2951666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 5.718489583333334,
+      "learning_rate": 8.002101126629421e-05,
+      "loss": 7.0929,
+      "loss/crossentropy": 2.0333942770957947,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16261755488812923,
+      "step": 1771
+    },
+    {
+      "epoch": 0.29533333333333334,
+      "grad_norm": 28.625,
+      "grad_norm_var": 5.593684895833333,
+      "learning_rate": 8.000007145663312e-05,
+      "loss": 6.9917,
+      "loss/crossentropy": 1.465691938996315,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18343962728977203,
+      "step": 1772
+    },
+    {
+      "epoch": 0.2955,
+      "grad_norm": 27.875,
+      "grad_norm_var": 5.448372395833333,
+      "learning_rate": 7.997912342228232e-05,
+      "loss": 6.8448,
+      "loss/crossentropy": 1.5461867153644562,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15889544412493706,
+      "step": 1773
+    },
+    {
+      "epoch": 0.2956666666666667,
+      "grad_norm": 26.875,
+      "grad_norm_var": 5.673958333333333,
+      "learning_rate": 7.99581671689848e-05,
+      "loss": 6.6658,
+      "loss/crossentropy": 1.9842509329319,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1756543032824993,
+      "step": 1774
+    },
+    {
+      "epoch": 0.29583333333333334,
+      "grad_norm": 27.625,
+      "grad_norm_var": 5.432291666666667,
+      "learning_rate": 7.993720270248584e-05,
+      "loss": 6.7945,
+      "loss/crossentropy": 1.4479915797710419,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.164757139980793,
+      "step": 1775
+    },
+    {
+      "epoch": 0.296,
+      "grad_norm": 27.75,
+      "grad_norm_var": 4.981184895833334,
+      "learning_rate": 7.991623002853296e-05,
+      "loss": 6.9274,
+      "loss/crossentropy": 1.6568330824375153,
+      "loss/hidden": 3.640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19808551669120789,
+      "step": 1776
+    },
+    {
+      "epoch": 0.2961666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 5.955989583333333,
+      "learning_rate": 7.989524915287595e-05,
+      "loss": 6.9704,
+      "loss/crossentropy": 1.8999920189380646,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2618193253874779,
+      "step": 1777
+    },
+    {
+      "epoch": 0.29633333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 5.9822265625,
+      "learning_rate": 7.987426008126683e-05,
+      "loss": 6.8001,
+      "loss/crossentropy": 1.420231431722641,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.159466240555048,
+      "step": 1778
+    },
+    {
+      "epoch": 0.2965,
+      "grad_norm": 29.25,
+      "grad_norm_var": 5.9416015625,
+      "learning_rate": 7.985326281945989e-05,
+      "loss": 7.097,
+      "loss/crossentropy": 1.807582527399063,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19036915153265,
+      "step": 1779
+    },
+    {
+      "epoch": 0.2966666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 5.88515625,
+      "learning_rate": 7.983225737321163e-05,
+      "loss": 6.8431,
+      "loss/crossentropy": 1.2770519852638245,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.141956627368927,
+      "step": 1780
+    },
+    {
+      "epoch": 0.29683333333333334,
+      "grad_norm": 28.625,
+      "grad_norm_var": 5.914322916666666,
+      "learning_rate": 7.98112437482808e-05,
+      "loss": 6.9422,
+      "loss/crossentropy": 2.1618843972682953,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22594073042273521,
+      "step": 1781
+    },
+    {
+      "epoch": 0.297,
+      "grad_norm": 28.5,
+      "grad_norm_var": 5.953059895833333,
+      "learning_rate": 7.979022195042842e-05,
+      "loss": 6.9345,
+      "loss/crossentropy": 1.6791729182004929,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1583862453699112,
+      "step": 1782
+    },
+    {
+      "epoch": 0.2971666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 5.779622395833333,
+      "learning_rate": 7.976919198541776e-05,
+      "loss": 7.1093,
+      "loss/crossentropy": 2.0253910422325134,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18104693293571472,
+      "step": 1783
+    },
+    {
+      "epoch": 0.29733333333333334,
+      "grad_norm": 27.125,
+      "grad_norm_var": 4.8900390625,
+      "learning_rate": 7.974815385901426e-05,
+      "loss": 6.8837,
+      "loss/crossentropy": 1.0846698433160782,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19080137088894844,
+      "step": 1784
+    },
+    {
+      "epoch": 0.2975,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.784309895833333,
+      "learning_rate": 7.972710757698567e-05,
+      "loss": 7.0842,
+      "loss/crossentropy": 1.0163829326629639,
+      "loss/hidden": 3.0859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09758484549820423,
+      "step": 1785
+    },
+    {
+      "epoch": 0.2976666666666667,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.723372395833333,
+      "learning_rate": 7.970605314510194e-05,
+      "loss": 6.8777,
+      "loss/crossentropy": 1.3463373482227325,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20978064835071564,
+      "step": 1786
+    },
+    {
+      "epoch": 0.29783333333333334,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.5171223958333333,
+      "learning_rate": 7.968499056913524e-05,
+      "loss": 7.1014,
+      "loss/crossentropy": 1.911300390958786,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20329588279128075,
+      "step": 1787
+    },
+    {
+      "epoch": 0.298,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.22890625,
+      "learning_rate": 7.966391985486003e-05,
+      "loss": 7.289,
+      "loss/crossentropy": 2.324524700641632,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22398141771554947,
+      "step": 1788
+    },
+    {
+      "epoch": 0.2981666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.1363932291666665,
+      "learning_rate": 7.964284100805297e-05,
+      "loss": 6.5837,
+      "loss/crossentropy": 1.631472885608673,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14790612272918224,
+      "step": 1789
+    },
+    {
+      "epoch": 0.29833333333333334,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.8978515625,
+      "learning_rate": 7.96217540344929e-05,
+      "loss": 7.0222,
+      "loss/crossentropy": 1.9974042177200317,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15219398401677608,
+      "step": 1790
+    },
+    {
+      "epoch": 0.2985,
+      "grad_norm": 26.875,
+      "grad_norm_var": 3.0900390625,
+      "learning_rate": 7.960065893996098e-05,
+      "loss": 6.6089,
+      "loss/crossentropy": 1.646654486656189,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16715771704912186,
+      "step": 1791
+    },
+    {
+      "epoch": 0.2986666666666667,
+      "grad_norm": 26.625,
+      "grad_norm_var": 3.37890625,
+      "learning_rate": 7.957955573024052e-05,
+      "loss": 6.6297,
+      "loss/crossentropy": 1.467742219567299,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15904828533530235,
+      "step": 1792
+    },
+    {
+      "epoch": 0.29883333333333334,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.18125,
+      "learning_rate": 7.95584444111171e-05,
+      "loss": 6.942,
+      "loss/crossentropy": 1.7421496510505676,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2290753275156021,
+      "step": 1793
+    },
+    {
+      "epoch": 0.299,
+      "grad_norm": 27.875,
+      "grad_norm_var": 1.89375,
+      "learning_rate": 7.95373249883785e-05,
+      "loss": 7.0584,
+      "loss/crossentropy": 2.0441207587718964,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14781775325536728,
+      "step": 1794
+    },
+    {
+      "epoch": 0.2991666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.9205729166666667,
+      "learning_rate": 7.951619746781474e-05,
+      "loss": 6.9599,
+      "loss/crossentropy": 1.7582679837942123,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19540252909064293,
+      "step": 1795
+    },
+    {
+      "epoch": 0.29933333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.0738932291666665,
+      "learning_rate": 7.949506185521802e-05,
+      "loss": 7.1559,
+      "loss/crossentropy": 2.3156529366970062,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16512399911880493,
+      "step": 1796
+    },
+    {
+      "epoch": 0.2995,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.0989583333333335,
+      "learning_rate": 7.947391815638284e-05,
+      "loss": 6.8516,
+      "loss/crossentropy": 1.6621273458003998,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1719889882951975,
+      "step": 1797
+    },
+    {
+      "epoch": 0.2996666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.1041666666666665,
+      "learning_rate": 7.945276637710582e-05,
+      "loss": 6.9817,
+      "loss/crossentropy": 1.2713432908058167,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18255398236215115,
+      "step": 1798
+    },
+    {
+      "epoch": 0.29983333333333334,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.0270182291666665,
+      "learning_rate": 7.943160652318585e-05,
+      "loss": 7.2958,
+      "loss/crossentropy": 1.586458146572113,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19176992401480675,
+      "step": 1799
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 26.375,
+      "grad_norm_var": 2.2067057291666665,
+      "learning_rate": 7.941043860042403e-05,
+      "loss": 7.0033,
+      "loss/crossentropy": 1.4122734367847443,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17232872918248177,
+      "step": 1800
+    },
+    {
+      "epoch": 0.3001666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.2207682291666666,
+      "learning_rate": 7.938926261462366e-05,
+      "loss": 6.7776,
+      "loss/crossentropy": 1.7719630151987076,
+      "loss/hidden": 3.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1281336434185505,
+      "step": 1801
+    },
+    {
+      "epoch": 0.30033333333333334,
+      "grad_norm": 26.25,
+      "grad_norm_var": 2.551041666666667,
+      "learning_rate": 7.936807857159026e-05,
+      "loss": 6.7018,
+      "loss/crossentropy": 1.3210044652223587,
+      "loss/hidden": 3.83984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17460490390658379,
+      "step": 1802
+    },
+    {
+      "epoch": 0.3005,
+      "grad_norm": 27.0,
+      "grad_norm_var": 2.6723307291666667,
+      "learning_rate": 7.934688647713158e-05,
+      "loss": 7.2574,
+      "loss/crossentropy": 1.5738000869750977,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20022379606962204,
+      "step": 1803
+    },
+    {
+      "epoch": 0.3006666666666667,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.6393229166666667,
+      "learning_rate": 7.932568633705752e-05,
+      "loss": 7.1622,
+      "loss/crossentropy": 2.075514942407608,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1800731047987938,
+      "step": 1804
+    },
+    {
+      "epoch": 0.30083333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.7249348958333333,
+      "learning_rate": 7.930447815718022e-05,
+      "loss": 6.9116,
+      "loss/crossentropy": 1.5317128598690033,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15660550631582737,
+      "step": 1805
+    },
+    {
+      "epoch": 0.301,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.5139973958333335,
+      "learning_rate": 7.928326194331404e-05,
+      "loss": 7.0955,
+      "loss/crossentropy": 1.6554387956857681,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14489647932350636,
+      "step": 1806
+    },
+    {
+      "epoch": 0.3011666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.6197265625,
+      "learning_rate": 7.926203770127552e-05,
+      "loss": 6.3688,
+      "loss/crossentropy": 1.5688863396644592,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17087645828723907,
+      "step": 1807
+    },
+    {
+      "epoch": 0.30133333333333334,
+      "grad_norm": 112.5,
+      "grad_norm_var": 441.96875,
+      "learning_rate": 7.924080543688337e-05,
+      "loss": 7.431,
+      "loss/crossentropy": 2.037807285785675,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17743926122784615,
+      "step": 1808
+    },
+    {
+      "epoch": 0.3015,
+      "grad_norm": 33.5,
+      "grad_norm_var": 439.55104166666666,
+      "learning_rate": 7.921956515595861e-05,
+      "loss": 7.0165,
+      "loss/crossentropy": 1.3565820753574371,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16366836801171303,
+      "step": 1809
+    },
+    {
+      "epoch": 0.3016666666666667,
+      "grad_norm": 26.875,
+      "grad_norm_var": 440.459375,
+      "learning_rate": 7.919831686432433e-05,
+      "loss": 6.531,
+      "loss/crossentropy": 1.2930429875850677,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16433067619800568,
+      "step": 1810
+    },
+    {
+      "epoch": 0.30183333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 439.78098958333334,
+      "learning_rate": 7.917706056780587e-05,
+      "loss": 7.4376,
+      "loss/crossentropy": 1.3591233491897583,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1684689000248909,
+      "step": 1811
+    },
+    {
+      "epoch": 0.302,
+      "grad_norm": 27.5,
+      "grad_norm_var": 441.95104166666664,
+      "learning_rate": 7.915579627223079e-05,
+      "loss": 7.0151,
+      "loss/crossentropy": 2.3508116602897644,
+      "loss/hidden": 2.9140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15554703772068024,
+      "step": 1812
+    },
+    {
+      "epoch": 0.30216666666666664,
+      "grad_norm": 28.5,
+      "grad_norm_var": 442.4643229166667,
+      "learning_rate": 7.913452398342881e-05,
+      "loss": 7.136,
+      "loss/crossentropy": 2.155543863773346,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19816122949123383,
+      "step": 1813
+    },
+    {
+      "epoch": 0.30233333333333334,
+      "grad_norm": 31.25,
+      "grad_norm_var": 441.28541666666666,
+      "learning_rate": 7.911324370723183e-05,
+      "loss": 6.7815,
+      "loss/crossentropy": 1.916477084159851,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18089919164776802,
+      "step": 1814
+    },
+    {
+      "epoch": 0.3025,
+      "grad_norm": 31.375,
+      "grad_norm_var": 439.24108072916664,
+      "learning_rate": 7.909195544947398e-05,
+      "loss": 7.384,
+      "loss/crossentropy": 1.5295850485563278,
+      "loss/hidden": 3.84375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22763513773679733,
+      "step": 1815
+    },
+    {
+      "epoch": 0.30266666666666664,
+      "grad_norm": 30.375,
+      "grad_norm_var": 435.99524739583336,
+      "learning_rate": 7.907065921599154e-05,
+      "loss": 7.2053,
+      "loss/crossentropy": 1.8417341709136963,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23864583484828472,
+      "step": 1816
+    },
+    {
+      "epoch": 0.30283333333333334,
+      "grad_norm": 27.375,
+      "grad_norm_var": 437.3705729166667,
+      "learning_rate": 7.9049355012623e-05,
+      "loss": 6.8166,
+      "loss/crossentropy": 1.5035779774188995,
+      "loss/hidden": 3.75,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21950983256101608,
+      "step": 1817
+    },
+    {
+      "epoch": 0.303,
+      "grad_norm": 28.125,
+      "grad_norm_var": 435.5317057291667,
+      "learning_rate": 7.902804284520903e-05,
+      "loss": 7.103,
+      "loss/crossentropy": 1.697951763868332,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14957010000944138,
+      "step": 1818
+    },
+    {
+      "epoch": 0.30316666666666664,
+      "grad_norm": 28.125,
+      "grad_norm_var": 434.47057291666664,
+      "learning_rate": 7.900672271959247e-05,
+      "loss": 6.9016,
+      "loss/crossentropy": 1.8022080063819885,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14898456819355488,
+      "step": 1819
+    },
+    {
+      "epoch": 0.30333333333333334,
+      "grad_norm": 26.625,
+      "grad_norm_var": 434.98932291666665,
+      "learning_rate": 7.898539464161838e-05,
+      "loss": 7.1498,
+      "loss/crossentropy": 1.259935587644577,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17238236591219902,
+      "step": 1820
+    },
+    {
+      "epoch": 0.3035,
+      "grad_norm": 27.125,
+      "grad_norm_var": 436.8854166666667,
+      "learning_rate": 7.896405861713394e-05,
+      "loss": 6.7154,
+      "loss/crossentropy": 1.450925201177597,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1539447344839573,
+      "step": 1821
+    },
+    {
+      "epoch": 0.30366666666666664,
+      "grad_norm": 32.75,
+      "grad_norm_var": 436.53326822916665,
+      "learning_rate": 7.894271465198857e-05,
+      "loss": 7.1475,
+      "loss/crossentropy": 1.918646514415741,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18164777755737305,
+      "step": 1822
+    },
+    {
+      "epoch": 0.30383333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 436.6041666666667,
+      "learning_rate": 7.892136275203383e-05,
+      "loss": 6.8871,
+      "loss/crossentropy": 1.5400571823120117,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1768193021416664,
+      "step": 1823
+    },
+    {
+      "epoch": 0.304,
+      "grad_norm": 29.25,
+      "grad_norm_var": 4.658072916666667,
+      "learning_rate": 7.890000292312346e-05,
+      "loss": 7.0315,
+      "loss/crossentropy": 1.7772729694843292,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22446472570300102,
+      "step": 1824
+    },
+    {
+      "epoch": 0.30416666666666664,
+      "grad_norm": 26.875,
+      "grad_norm_var": 3.7436848958333333,
+      "learning_rate": 7.887863517111338e-05,
+      "loss": 7.0343,
+      "loss/crossentropy": 1.6882353723049164,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22113287448883057,
+      "step": 1825
+    },
+    {
+      "epoch": 0.30433333333333334,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.452018229166667,
+      "learning_rate": 7.88572595018617e-05,
+      "loss": 6.9748,
+      "loss/crossentropy": 1.8164548575878143,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21961813420057297,
+      "step": 1826
+    },
+    {
+      "epoch": 0.3045,
+      "grad_norm": 28.375,
+      "grad_norm_var": 3.267708333333333,
+      "learning_rate": 7.883587592122863e-05,
+      "loss": 6.8205,
+      "loss/crossentropy": 1.6645339578390121,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1584730539470911,
+      "step": 1827
+    },
+    {
+      "epoch": 0.30466666666666664,
+      "grad_norm": 28.875,
+      "grad_norm_var": 3.128059895833333,
+      "learning_rate": 7.881448443507664e-05,
+      "loss": 7.2591,
+      "loss/crossentropy": 1.973101168870926,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2651126943528652,
+      "step": 1828
+    },
+    {
+      "epoch": 0.30483333333333335,
+      "grad_norm": 27.5,
+      "grad_norm_var": 3.2561848958333335,
+      "learning_rate": 7.879308504927035e-05,
+      "loss": 7.2025,
+      "loss/crossentropy": 1.5946943312883377,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17251763492822647,
+      "step": 1829
+    },
+    {
+      "epoch": 0.305,
+      "grad_norm": 26.875,
+      "grad_norm_var": 3.0989583333333335,
+      "learning_rate": 7.877167776967645e-05,
+      "loss": 7.0168,
+      "loss/crossentropy": 2.1168753504753113,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18964385241270065,
+      "step": 1830
+    },
+    {
+      "epoch": 0.30516666666666664,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.463997395833333,
+      "learning_rate": 7.875026260216393e-05,
+      "loss": 6.6181,
+      "loss/crossentropy": 1.3980006575584412,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10987738519906998,
+      "step": 1831
+    },
+    {
+      "epoch": 0.30533333333333335,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.3177083333333335,
+      "learning_rate": 7.872883955260387e-05,
+      "loss": 6.5008,
+      "loss/crossentropy": 2.172425150871277,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15537427365779877,
+      "step": 1832
+    },
+    {
+      "epoch": 0.3055,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.5942057291666667,
+      "learning_rate": 7.87074086268695e-05,
+      "loss": 7.0536,
+      "loss/crossentropy": 1.8021511733531952,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21503210812807083,
+      "step": 1833
+    },
+    {
+      "epoch": 0.30566666666666664,
+      "grad_norm": 41.5,
+      "grad_norm_var": 13.395572916666667,
+      "learning_rate": 7.868596983083623e-05,
+      "loss": 7.099,
+      "loss/crossentropy": 1.839784860610962,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14783872663974762,
+      "step": 1834
+    },
+    {
+      "epoch": 0.30583333333333335,
+      "grad_norm": 29.5,
+      "grad_norm_var": 13.218684895833333,
+      "learning_rate": 7.866452317038164e-05,
+      "loss": 6.6123,
+      "loss/crossentropy": 0.9922498315572739,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1440199390053749,
+      "step": 1835
+    },
+    {
+      "epoch": 0.306,
+      "grad_norm": 29.125,
+      "grad_norm_var": 12.544205729166666,
+      "learning_rate": 7.864306865138545e-05,
+      "loss": 7.3191,
+      "loss/crossentropy": 1.683628648519516,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1728571616113186,
+      "step": 1836
+    },
+    {
+      "epoch": 0.30616666666666664,
+      "grad_norm": 27.0,
+      "grad_norm_var": 12.592708333333333,
+      "learning_rate": 7.862160627972955e-05,
+      "loss": 6.9569,
+      "loss/crossentropy": 2.0906372666358948,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18486298620700836,
+      "step": 1837
+    },
+    {
+      "epoch": 0.30633333333333335,
+      "grad_norm": 29.25,
+      "grad_norm_var": 12.060416666666667,
+      "learning_rate": 7.860013606129796e-05,
+      "loss": 6.939,
+      "loss/crossentropy": 1.5109025537967682,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13854311406612396,
+      "step": 1838
+    },
+    {
+      "epoch": 0.3065,
+      "grad_norm": 29.0,
+      "grad_norm_var": 12.074739583333333,
+      "learning_rate": 7.857865800197684e-05,
+      "loss": 7.0375,
+      "loss/crossentropy": 1.3523481339216232,
+      "loss/hidden": 3.77734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18698465079069138,
+      "step": 1839
+    },
+    {
+      "epoch": 0.30666666666666664,
+      "grad_norm": 29.625,
+      "grad_norm_var": 12.062434895833333,
+      "learning_rate": 7.855717210765456e-05,
+      "loss": 6.7407,
+      "loss/crossentropy": 1.227504014968872,
+      "loss/hidden": 3.14453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11474656872451305,
+      "step": 1840
+    },
+    {
+      "epoch": 0.30683333333333335,
+      "grad_norm": 29.0,
+      "grad_norm_var": 11.545572916666666,
+      "learning_rate": 7.85356783842216e-05,
+      "loss": 6.8196,
+      "loss/crossentropy": 1.9633170366287231,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21601013839244843,
+      "step": 1841
+    },
+    {
+      "epoch": 0.307,
+      "grad_norm": 30.75,
+      "grad_norm_var": 11.4869140625,
+      "learning_rate": 7.851417683757053e-05,
+      "loss": 7.2679,
+      "loss/crossentropy": 1.1730319112539291,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1535097137093544,
+      "step": 1842
+    },
+    {
+      "epoch": 0.30716666666666664,
+      "grad_norm": 29.125,
+      "grad_norm_var": 11.3634765625,
+      "learning_rate": 7.849266747359619e-05,
+      "loss": 7.1529,
+      "loss/crossentropy": 1.5851309597492218,
+      "loss/hidden": 3.8046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20339439436793327,
+      "step": 1843
+    },
+    {
+      "epoch": 0.30733333333333335,
+      "grad_norm": 26.375,
+      "grad_norm_var": 12.131705729166667,
+      "learning_rate": 7.847115029819547e-05,
+      "loss": 6.9862,
+      "loss/crossentropy": 1.713428407907486,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20932835340499878,
+      "step": 1844
+    },
+    {
+      "epoch": 0.3075,
+      "grad_norm": 31.375,
+      "grad_norm_var": 11.855208333333334,
+      "learning_rate": 7.84496253172674e-05,
+      "loss": 6.9028,
+      "loss/crossentropy": 1.655678853392601,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14287131652235985,
+      "step": 1845
+    },
+    {
+      "epoch": 0.30766666666666664,
+      "grad_norm": 24.75,
+      "grad_norm_var": 13.0494140625,
+      "learning_rate": 7.84280925367132e-05,
+      "loss": 6.6203,
+      "loss/crossentropy": 1.6767279207706451,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13365484960377216,
+      "step": 1846
+    },
+    {
+      "epoch": 0.30783333333333335,
+      "grad_norm": 32.25,
+      "grad_norm_var": 13.0494140625,
+      "learning_rate": 7.84065519624362e-05,
+      "loss": 7.1116,
+      "loss/crossentropy": 1.8239229321479797,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19592266529798508,
+      "step": 1847
+    },
+    {
+      "epoch": 0.308,
+      "grad_norm": 29.25,
+      "grad_norm_var": 13.068684895833334,
+      "learning_rate": 7.838500360034188e-05,
+      "loss": 7.082,
+      "loss/crossentropy": 2.026229441165924,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19996747002005577,
+      "step": 1848
+    },
+    {
+      "epoch": 0.30816666666666664,
+      "grad_norm": 29.875,
+      "grad_norm_var": 12.94765625,
+      "learning_rate": 7.836344745633783e-05,
+      "loss": 7.0678,
+      "loss/crossentropy": 2.0968761444091797,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17616475000977516,
+      "step": 1849
+    },
+    {
+      "epoch": 0.30833333333333335,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.324934895833333,
+      "learning_rate": 7.83418835363338e-05,
+      "loss": 6.8456,
+      "loss/crossentropy": 1.4505693912506104,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13610314577817917,
+      "step": 1850
+    },
+    {
+      "epoch": 0.3085,
+      "grad_norm": 27.0,
+      "grad_norm_var": 3.567122395833333,
+      "learning_rate": 7.832031184624164e-05,
+      "loss": 7.1366,
+      "loss/crossentropy": 2.1420071125030518,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1590719074010849,
+      "step": 1851
+    },
+    {
+      "epoch": 0.30866666666666664,
+      "grad_norm": 27.625,
+      "grad_norm_var": 3.6624348958333335,
+      "learning_rate": 7.829873239197538e-05,
+      "loss": 7.1721,
+      "loss/crossentropy": 1.3285071849822998,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16042914055287838,
+      "step": 1852
+    },
+    {
+      "epoch": 0.30883333333333335,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.432747395833333,
+      "learning_rate": 7.827714517945115e-05,
+      "loss": 7.0415,
+      "loss/crossentropy": 1.7693351209163666,
+      "loss/hidden": 3.0625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13225612975656986,
+      "step": 1853
+    },
+    {
+      "epoch": 0.309,
+      "grad_norm": 27.375,
+      "grad_norm_var": 3.568489583333333,
+      "learning_rate": 7.825555021458716e-05,
+      "loss": 7.0073,
+      "loss/crossentropy": 2.1532363295555115,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22851087525486946,
+      "step": 1854
+    },
+    {
+      "epoch": 0.30916666666666665,
+      "grad_norm": 26.375,
+      "grad_norm_var": 3.9280598958333335,
+      "learning_rate": 7.823394750330387e-05,
+      "loss": 6.9739,
+      "loss/crossentropy": 1.903586357831955,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16680320911109447,
+      "step": 1855
+    },
+    {
+      "epoch": 0.30933333333333335,
+      "grad_norm": 27.0,
+      "grad_norm_var": 4.011458333333334,
+      "learning_rate": 7.821233705152371e-05,
+      "loss": 6.516,
+      "loss/crossentropy": 1.6436423063278198,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14540725946426392,
+      "step": 1856
+    },
+    {
+      "epoch": 0.3095,
+      "grad_norm": 28.375,
+      "grad_norm_var": 3.9916015625,
+      "learning_rate": 7.819071886517134e-05,
+      "loss": 7.0631,
+      "loss/crossentropy": 1.4558165669441223,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.126516193151474,
+      "step": 1857
+    },
+    {
+      "epoch": 0.30966666666666665,
+      "grad_norm": 35.0,
+      "grad_norm_var": 6.4353515625,
+      "learning_rate": 7.816909295017352e-05,
+      "loss": 6.7891,
+      "loss/crossentropy": 1.5153915584087372,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1435499582439661,
+      "step": 1858
+    },
+    {
+      "epoch": 0.30983333333333335,
+      "grad_norm": 30.875,
+      "grad_norm_var": 6.727018229166666,
+      "learning_rate": 7.81474593124591e-05,
+      "loss": 6.8838,
+      "loss/crossentropy": 1.9987106323242188,
+      "loss/hidden": 3.71484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27799108624458313,
+      "step": 1859
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 30.875,
+      "grad_norm_var": 6.534830729166667,
+      "learning_rate": 7.812581795795907e-05,
+      "loss": 6.9019,
+      "loss/crossentropy": 1.8754624426364899,
+      "loss/hidden": 3.03125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13171296194195747,
+      "step": 1860
+    },
+    {
+      "epoch": 0.31016666666666665,
+      "grad_norm": 30.75,
+      "grad_norm_var": 6.368489583333333,
+      "learning_rate": 7.810416889260653e-05,
+      "loss": 7.4034,
+      "loss/crossentropy": 1.5210355818271637,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16199497133493423,
+      "step": 1861
+    },
+    {
+      "epoch": 0.31033333333333335,
+      "grad_norm": 29.625,
+      "grad_norm_var": 5.060872395833333,
+      "learning_rate": 7.80825121223367e-05,
+      "loss": 7.3534,
+      "loss/crossentropy": 2.180351436138153,
+      "loss/hidden": 3.0703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14645655825734138,
+      "step": 1862
+    },
+    {
+      "epoch": 0.3105,
+      "grad_norm": 30.625,
+      "grad_norm_var": 4.597916666666666,
+      "learning_rate": 7.80608476530869e-05,
+      "loss": 6.8991,
+      "loss/crossentropy": 1.7439337968826294,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1728023998439312,
+      "step": 1863
+    },
+    {
+      "epoch": 0.31066666666666665,
+      "grad_norm": 28.75,
+      "grad_norm_var": 4.613541666666666,
+      "learning_rate": 7.803917549079655e-05,
+      "loss": 7.0085,
+      "loss/crossentropy": 1.8750990629196167,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16987328231334686,
+      "step": 1864
+    },
+    {
+      "epoch": 0.31083333333333335,
+      "grad_norm": 31.0,
+      "grad_norm_var": 4.791080729166667,
+      "learning_rate": 7.801749564140724e-05,
+      "loss": 7.0188,
+      "loss/crossentropy": 1.7670414745807648,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1989278867840767,
+      "step": 1865
+    },
+    {
+      "epoch": 0.311,
+      "grad_norm": 29.875,
+      "grad_norm_var": 4.778059895833334,
+      "learning_rate": 7.799580811086258e-05,
+      "loss": 7.2138,
+      "loss/crossentropy": 1.9684894680976868,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20455539226531982,
+      "step": 1866
+    },
+    {
+      "epoch": 0.31116666666666665,
+      "grad_norm": 29.75,
+      "grad_norm_var": 4.382747395833333,
+      "learning_rate": 7.797411290510835e-05,
+      "loss": 7.1259,
+      "loss/crossentropy": 1.046493113040924,
+      "loss/hidden": 3.828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17215987294912338,
+      "step": 1867
+    },
+    {
+      "epoch": 0.31133333333333335,
+      "grad_norm": 29.625,
+      "grad_norm_var": 4.122330729166666,
+      "learning_rate": 7.795241003009241e-05,
+      "loss": 7.2216,
+      "loss/crossentropy": 1.8449428081512451,
+      "loss/hidden": 3.0234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1435481458902359,
+      "step": 1868
+    },
+    {
+      "epoch": 0.3115,
+      "grad_norm": 29.125,
+      "grad_norm_var": 4.085416666666666,
+      "learning_rate": 7.793069949176473e-05,
+      "loss": 7.1263,
+      "loss/crossentropy": 2.1592356860637665,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24637145176529884,
+      "step": 1869
+    },
+    {
+      "epoch": 0.31166666666666665,
+      "grad_norm": 26.875,
+      "grad_norm_var": 4.255208333333333,
+      "learning_rate": 7.790898129607738e-05,
+      "loss": 6.7132,
+      "loss/crossentropy": 1.515554964542389,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16295021399855614,
+      "step": 1870
+    },
+    {
+      "epoch": 0.31183333333333335,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.6546223958333335,
+      "learning_rate": 7.788725544898452e-05,
+      "loss": 7.0407,
+      "loss/crossentropy": 1.7184237241744995,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.162395890802145,
+      "step": 1871
+    },
+    {
+      "epoch": 0.312,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.2900390625,
+      "learning_rate": 7.78655219564424e-05,
+      "loss": 7.2002,
+      "loss/crossentropy": 2.1363507509231567,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19071825221180916,
+      "step": 1872
+    },
+    {
+      "epoch": 0.31216666666666665,
+      "grad_norm": 28.375,
+      "grad_norm_var": 3.2900390625,
+      "learning_rate": 7.784378082440941e-05,
+      "loss": 6.9572,
+      "loss/crossentropy": 1.750045657157898,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24303391948342323,
+      "step": 1873
+    },
+    {
+      "epoch": 0.31233333333333335,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.6530598958333333,
+      "learning_rate": 7.782203205884598e-05,
+      "loss": 7.0691,
+      "loss/crossentropy": 1.3648466914892197,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20632506720721722,
+      "step": 1874
+    },
+    {
+      "epoch": 0.3125,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.5452473958333333,
+      "learning_rate": 7.780027566571465e-05,
+      "loss": 7.3594,
+      "loss/crossentropy": 1.7212892472743988,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23325346782803535,
+      "step": 1875
+    },
+    {
+      "epoch": 0.31266666666666665,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.0989583333333335,
+      "learning_rate": 7.777851165098012e-05,
+      "loss": 7.2349,
+      "loss/crossentropy": 1.2445969730615616,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15521959587931633,
+      "step": 1876
+    },
+    {
+      "epoch": 0.31283333333333335,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.1393229166666665,
+      "learning_rate": 7.775674002060905e-05,
+      "loss": 6.8145,
+      "loss/crossentropy": 1.4034997522830963,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16615463607013226,
+      "step": 1877
+    },
+    {
+      "epoch": 0.313,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.3707682291666665,
+      "learning_rate": 7.773496078057028e-05,
+      "loss": 6.9783,
+      "loss/crossentropy": 1.895322322845459,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19174939766526222,
+      "step": 1878
+    },
+    {
+      "epoch": 0.31316666666666665,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.2900390625,
+      "learning_rate": 7.771317393683471e-05,
+      "loss": 7.0273,
+      "loss/crossentropy": 1.9640153646469116,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23941552266478539,
+      "step": 1879
+    },
+    {
+      "epoch": 0.31333333333333335,
+      "grad_norm": 27.0,
+      "grad_norm_var": 2.6509765625,
+      "learning_rate": 7.769137949537532e-05,
+      "loss": 6.864,
+      "loss/crossentropy": 1.4724474251270294,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15606267377734184,
+      "step": 1880
+    },
+    {
+      "epoch": 0.3135,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.5416666666666665,
+      "learning_rate": 7.766957746216721e-05,
+      "loss": 6.9245,
+      "loss/crossentropy": 1.7987104952335358,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16625307500362396,
+      "step": 1881
+    },
+    {
+      "epoch": 0.31366666666666665,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.8655598958333335,
+      "learning_rate": 7.764776784318751e-05,
+      "loss": 7.205,
+      "loss/crossentropy": 2.092978924512863,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1646638959646225,
+      "step": 1882
+    },
+    {
+      "epoch": 0.31383333333333335,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.9546223958333333,
+      "learning_rate": 7.762595064441542e-05,
+      "loss": 7.0331,
+      "loss/crossentropy": 2.033162772655487,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2481602057814598,
+      "step": 1883
+    },
+    {
+      "epoch": 0.314,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.1811848958333333,
+      "learning_rate": 7.76041258718323e-05,
+      "loss": 6.8271,
+      "loss/crossentropy": 1.8266779333353043,
+      "loss/hidden": 3.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1392462272197008,
+      "step": 1884
+    },
+    {
+      "epoch": 0.31416666666666665,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.2358723958333333,
+      "learning_rate": 7.758229353142152e-05,
+      "loss": 7.1062,
+      "loss/crossentropy": 2.3862681090831757,
+      "loss/hidden": 3.1015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.164220180362463,
+      "step": 1885
+    },
+    {
+      "epoch": 0.31433333333333335,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.746875,
+      "learning_rate": 7.756045362916853e-05,
+      "loss": 7.3689,
+      "loss/crossentropy": 1.8096283078193665,
+      "loss/hidden": 3.8203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26275773346424103,
+      "step": 1886
+    },
+    {
+      "epoch": 0.3145,
+      "grad_norm": 27.625,
+      "grad_norm_var": 3.862434895833333,
+      "learning_rate": 7.753860617106086e-05,
+      "loss": 6.8594,
+      "loss/crossentropy": 1.6941289901733398,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22652960568666458,
+      "step": 1887
+    },
+    {
+      "epoch": 0.31466666666666665,
+      "grad_norm": 28.875,
+      "grad_norm_var": 3.798958333333333,
+      "learning_rate": 7.751675116308812e-05,
+      "loss": 6.6887,
+      "loss/crossentropy": 1.4575351774692535,
+      "loss/hidden": 3.8359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2733895592391491,
+      "step": 1888
+    },
+    {
+      "epoch": 0.31483333333333335,
+      "grad_norm": 27.625,
+      "grad_norm_var": 3.9309895833333335,
+      "learning_rate": 7.7494888611242e-05,
+      "loss": 7.0136,
+      "loss/crossentropy": 1.8295487463474274,
+      "loss/hidden": 3.75,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21708901971578598,
+      "step": 1889
+    },
+    {
+      "epoch": 0.315,
+      "grad_norm": 27.75,
+      "grad_norm_var": 3.7083333333333335,
+      "learning_rate": 7.747301852151621e-05,
+      "loss": 7.0604,
+      "loss/crossentropy": 1.6264638304710388,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14993342012166977,
+      "step": 1890
+    },
+    {
+      "epoch": 0.31516666666666665,
+      "grad_norm": 26.375,
+      "grad_norm_var": 4.145833333333333,
+      "learning_rate": 7.74511408999066e-05,
+      "loss": 6.8725,
+      "loss/crossentropy": 1.2875807881355286,
+      "loss/hidden": 3.8359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23947446048259735,
+      "step": 1891
+    },
+    {
+      "epoch": 0.31533333333333335,
+      "grad_norm": 25.25,
+      "grad_norm_var": 3.7864583333333335,
+      "learning_rate": 7.7429255752411e-05,
+      "loss": 6.5349,
+      "loss/crossentropy": 1.4500753730535507,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1507728062570095,
+      "step": 1892
+    },
+    {
+      "epoch": 0.3155,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.6655598958333333,
+      "learning_rate": 7.740736308502938e-05,
+      "loss": 6.5843,
+      "loss/crossentropy": 1.6536575853824615,
+      "loss/hidden": 3.1015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14045970141887665,
+      "step": 1893
+    },
+    {
+      "epoch": 0.31566666666666665,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.939583333333333,
+      "learning_rate": 7.738546290376373e-05,
+      "loss": 7.2277,
+      "loss/crossentropy": 1.5449070632457733,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19420025497674942,
+      "step": 1894
+    },
+    {
+      "epoch": 0.31583333333333335,
+      "grad_norm": 31.375,
+      "grad_norm_var": 4.40625,
+      "learning_rate": 7.736355521461811e-05,
+      "loss": 7.0307,
+      "loss/crossentropy": 1.3559617400169373,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11303997971117496,
+      "step": 1895
+    },
+    {
+      "epoch": 0.316,
+      "grad_norm": 29.0,
+      "grad_norm_var": 4.20625,
+      "learning_rate": 7.734164002359863e-05,
+      "loss": 6.979,
+      "loss/crossentropy": 1.6296832859516144,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15469217486679554,
+      "step": 1896
+    },
+    {
+      "epoch": 0.31616666666666665,
+      "grad_norm": 29.75,
+      "grad_norm_var": 4.222330729166667,
+      "learning_rate": 7.731971733671346e-05,
+      "loss": 7.1481,
+      "loss/crossentropy": 1.4442989230155945,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15145175904035568,
+      "step": 1897
+    },
+    {
+      "epoch": 0.31633333333333336,
+      "grad_norm": 27.0,
+      "grad_norm_var": 4.1541015625,
+      "learning_rate": 7.729778715997284e-05,
+      "loss": 7.1184,
+      "loss/crossentropy": 1.3691516071557999,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14277916587889194,
+      "step": 1898
+    },
+    {
+      "epoch": 0.3165,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.01640625,
+      "learning_rate": 7.727584949938907e-05,
+      "loss": 7.1911,
+      "loss/crossentropy": 1.8479988873004913,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1650102473795414,
+      "step": 1899
+    },
+    {
+      "epoch": 0.31666666666666665,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.8072265625,
+      "learning_rate": 7.725390436097643e-05,
+      "loss": 7.0348,
+      "loss/crossentropy": 1.6918924301862717,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1792307198047638,
+      "step": 1900
+    },
+    {
+      "epoch": 0.31683333333333336,
+      "grad_norm": 27.75,
+      "grad_norm_var": 3.808333333333333,
+      "learning_rate": 7.723195175075136e-05,
+      "loss": 6.7804,
+      "loss/crossentropy": 1.537046879529953,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14605162106454372,
+      "step": 1901
+    },
+    {
+      "epoch": 0.317,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.443489583333333,
+      "learning_rate": 7.720999167473227e-05,
+      "loss": 7.1965,
+      "loss/crossentropy": 2.006430447101593,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18422998115420341,
+      "step": 1902
+    },
+    {
+      "epoch": 0.31716666666666665,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.4546223958333333,
+      "learning_rate": 7.718802413893963e-05,
+      "loss": 6.7879,
+      "loss/crossentropy": 1.7092276513576508,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19111250340938568,
+      "step": 1903
+    },
+    {
+      "epoch": 0.31733333333333336,
+      "grad_norm": 28.875,
+      "grad_norm_var": 3.4546223958333333,
+      "learning_rate": 7.716604914939598e-05,
+      "loss": 7.1008,
+      "loss/crossentropy": 1.9587385058403015,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2650653049349785,
+      "step": 1904
+    },
+    {
+      "epoch": 0.3175,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.376822916666667,
+      "learning_rate": 7.714406671212589e-05,
+      "loss": 6.848,
+      "loss/crossentropy": 2.069043219089508,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1532590314745903,
+      "step": 1905
+    },
+    {
+      "epoch": 0.31766666666666665,
+      "grad_norm": 28.0,
+      "grad_norm_var": 3.3427083333333334,
+      "learning_rate": 7.712207683315594e-05,
+      "loss": 6.7033,
+      "loss/crossentropy": 1.9057554602622986,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16242307424545288,
+      "step": 1906
+    },
+    {
+      "epoch": 0.31783333333333336,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.9593098958333335,
+      "learning_rate": 7.710007951851481e-05,
+      "loss": 6.8823,
+      "loss/crossentropy": 1.3203213214874268,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2788276895880699,
+      "step": 1907
+    },
+    {
+      "epoch": 0.318,
+      "grad_norm": 26.5,
+      "grad_norm_var": 2.4306640625,
+      "learning_rate": 7.707807477423319e-05,
+      "loss": 6.729,
+      "loss/crossentropy": 1.7957911789417267,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14845603704452515,
+      "step": 1908
+    },
+    {
+      "epoch": 0.31816666666666665,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.3436848958333334,
+      "learning_rate": 7.705606260634379e-05,
+      "loss": 7.045,
+      "loss/crossentropy": 1.6333225816488266,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17810802906751633,
+      "step": 1909
+    },
+    {
+      "epoch": 0.31833333333333336,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.0708333333333333,
+      "learning_rate": 7.703404302088138e-05,
+      "loss": 6.6415,
+      "loss/crossentropy": 2.0423670411109924,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1727237943559885,
+      "step": 1910
+    },
+    {
+      "epoch": 0.3185,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.751822916666667,
+      "learning_rate": 7.701201602388276e-05,
+      "loss": 7.0141,
+      "loss/crossentropy": 1.8547867238521576,
+      "loss/hidden": 3.84765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2650393471121788,
+      "step": 1911
+    },
+    {
+      "epoch": 0.31866666666666665,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.8247395833333333,
+      "learning_rate": 7.698998162138673e-05,
+      "loss": 6.7923,
+      "loss/crossentropy": 1.812669113278389,
+      "loss/hidden": 3.078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1441791821271181,
+      "step": 1912
+    },
+    {
+      "epoch": 0.31883333333333336,
+      "grad_norm": 26.375,
+      "grad_norm_var": 3.206184895833333,
+      "learning_rate": 7.696793981943417e-05,
+      "loss": 7.0773,
+      "loss/crossentropy": 2.0258938521146774,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2104419209063053,
+      "step": 1913
+    },
+    {
+      "epoch": 0.319,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.9934895833333335,
+      "learning_rate": 7.694589062406796e-05,
+      "loss": 6.9504,
+      "loss/crossentropy": 1.8064450919628143,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14459712617099285,
+      "step": 1914
+    },
+    {
+      "epoch": 0.31916666666666665,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.98125,
+      "learning_rate": 7.692383404133301e-05,
+      "loss": 7.6258,
+      "loss/crossentropy": 1.7547499984502792,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16656643897294998,
+      "step": 1915
+    },
+    {
+      "epoch": 0.31933333333333336,
+      "grad_norm": 26.125,
+      "grad_norm_var": 3.440559895833333,
+      "learning_rate": 7.690177007727625e-05,
+      "loss": 6.794,
+      "loss/crossentropy": 1.5933746993541718,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16254575178027153,
+      "step": 1916
+    },
+    {
+      "epoch": 0.3195,
+      "grad_norm": 25.5,
+      "grad_norm_var": 4.054622395833333,
+      "learning_rate": 7.687969873794667e-05,
+      "loss": 6.8675,
+      "loss/crossentropy": 2.262081354856491,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20529960095882416,
+      "step": 1917
+    },
+    {
+      "epoch": 0.31966666666666665,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.5155598958333334,
+      "learning_rate": 7.685762002939523e-05,
+      "loss": 6.8656,
+      "loss/crossentropy": 1.458501249551773,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13611052371561527,
+      "step": 1918
+    },
+    {
+      "epoch": 0.31983333333333336,
+      "grad_norm": 28.375,
+      "grad_norm_var": 3.357291666666667,
+      "learning_rate": 7.683553395767492e-05,
+      "loss": 6.7519,
+      "loss/crossentropy": 2.0998906195163727,
+      "loss/hidden": 3.12109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1625036932528019,
+      "step": 1919
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.506705729166667,
+      "learning_rate": 7.681344052884077e-05,
+      "loss": 6.8568,
+      "loss/crossentropy": 1.7492288947105408,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20232809893786907,
+      "step": 1920
+    },
+    {
+      "epoch": 0.32016666666666665,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.5192057291666665,
+      "learning_rate": 7.679133974894983e-05,
+      "loss": 6.705,
+      "loss/crossentropy": 1.3074803054332733,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1418884601444006,
+      "step": 1921
+    },
+    {
+      "epoch": 0.32033333333333336,
+      "grad_norm": 31.75,
+      "grad_norm_var": 4.136393229166667,
+      "learning_rate": 7.676923162406115e-05,
+      "loss": 7.2457,
+      "loss/crossentropy": 1.7019118815660477,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18821359053254128,
+      "step": 1922
+    },
+    {
+      "epoch": 0.3205,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.115625,
+      "learning_rate": 7.674711616023581e-05,
+      "loss": 7.2184,
+      "loss/crossentropy": 1.7697814851999283,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22327840328216553,
+      "step": 1923
+    },
+    {
+      "epoch": 0.32066666666666666,
+      "grad_norm": 25.625,
+      "grad_norm_var": 4.4369140625,
+      "learning_rate": 7.672499336353687e-05,
+      "loss": 6.836,
+      "loss/crossentropy": 1.687684714794159,
+      "loss/hidden": 3.14453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14532406255602837,
+      "step": 1924
+    },
+    {
+      "epoch": 0.32083333333333336,
+      "grad_norm": 25.25,
+      "grad_norm_var": 5.05390625,
+      "learning_rate": 7.670286324002944e-05,
+      "loss": 6.8472,
+      "loss/crossentropy": 1.735940158367157,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1309076026082039,
+      "step": 1925
+    },
+    {
+      "epoch": 0.321,
+      "grad_norm": 27.5,
+      "grad_norm_var": 3.790625,
+      "learning_rate": 7.668072579578058e-05,
+      "loss": 6.8334,
+      "loss/crossentropy": 1.7964033260941505,
+      "loss/hidden": 3.07421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13687119260430336,
+      "step": 1926
+    },
+    {
+      "epoch": 0.32116666666666666,
+      "grad_norm": 27.75,
+      "grad_norm_var": 3.6431640625,
+      "learning_rate": 7.665858103685944e-05,
+      "loss": 6.7366,
+      "loss/crossentropy": 1.6139466762542725,
+      "loss/hidden": 3.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13353602588176727,
+      "step": 1927
+    },
+    {
+      "epoch": 0.32133333333333336,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.71640625,
+      "learning_rate": 7.663642896933712e-05,
+      "loss": 7.1254,
+      "loss/crossentropy": 2.315617620944977,
+      "loss/hidden": 3.14453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17119132354855537,
+      "step": 1928
+    },
+    {
+      "epoch": 0.3215,
+      "grad_norm": 26.5,
+      "grad_norm_var": 3.6884765625,
+      "learning_rate": 7.66142695992867e-05,
+      "loss": 6.7417,
+      "loss/crossentropy": 1.6333139091730118,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16588200815021992,
+      "step": 1929
+    },
+    {
+      "epoch": 0.32166666666666666,
+      "grad_norm": 26.0,
+      "grad_norm_var": 3.959375,
+      "learning_rate": 7.659210293278334e-05,
+      "loss": 6.747,
+      "loss/crossentropy": 1.4016158431768417,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1492020282894373,
+      "step": 1930
+    },
+    {
+      "epoch": 0.32183333333333336,
+      "grad_norm": 27.75,
+      "grad_norm_var": 3.8655598958333335,
+      "learning_rate": 7.656992897590414e-05,
+      "loss": 7.0579,
+      "loss/crossentropy": 1.520734429359436,
+      "loss/hidden": 3.7578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24859199672937393,
+      "step": 1931
+    },
+    {
+      "epoch": 0.322,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.118489583333333,
+      "learning_rate": 7.654774773472823e-05,
+      "loss": 7.0604,
+      "loss/crossentropy": 1.949371337890625,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23103799670934677,
+      "step": 1932
+    },
+    {
+      "epoch": 0.32216666666666666,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.620833333333333,
+      "learning_rate": 7.65255592153367e-05,
+      "loss": 6.8736,
+      "loss/crossentropy": 1.5065647065639496,
+      "loss/hidden": 3.73046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1492113471031189,
+      "step": 1933
+    },
+    {
+      "epoch": 0.32233333333333336,
+      "grad_norm": 28.5,
+      "grad_norm_var": 3.3041666666666667,
+      "learning_rate": 7.650336342381269e-05,
+      "loss": 6.8634,
+      "loss/crossentropy": 1.7050878405570984,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19006743282079697,
+      "step": 1934
+    },
+    {
+      "epoch": 0.3225,
+      "grad_norm": 28.375,
+      "grad_norm_var": 3.3041666666666667,
+      "learning_rate": 7.648116036624126e-05,
+      "loss": 6.8793,
+      "loss/crossentropy": 1.845796674489975,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17243629321455956,
+      "step": 1935
+    },
+    {
+      "epoch": 0.32266666666666666,
+      "grad_norm": 26.5,
+      "grad_norm_var": 3.253125,
+      "learning_rate": 7.645895004870954e-05,
+      "loss": 6.8049,
+      "loss/crossentropy": 1.1808339804410934,
+      "loss/hidden": 3.66015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1586676798760891,
+      "step": 1936
+    },
+    {
+      "epoch": 0.32283333333333336,
+      "grad_norm": 27.5,
+      "grad_norm_var": 3.2,
+      "learning_rate": 7.643673247730658e-05,
+      "loss": 6.8314,
+      "loss/crossentropy": 1.37615168094635,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15130289644002914,
+      "step": 1937
+    },
+    {
+      "epoch": 0.323,
+      "grad_norm": 26.25,
+      "grad_norm_var": 2.294791666666667,
+      "learning_rate": 7.64145076581235e-05,
+      "loss": 6.912,
+      "loss/crossentropy": 1.5464135706424713,
+      "loss/hidden": 3.84765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2024979665875435,
+      "step": 1938
+    },
+    {
+      "epoch": 0.32316666666666666,
+      "grad_norm": 26.0,
+      "grad_norm_var": 2.2051432291666666,
+      "learning_rate": 7.639227559725332e-05,
+      "loss": 6.6962,
+      "loss/crossentropy": 1.2784469723701477,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1756535191088915,
+      "step": 1939
+    },
+    {
+      "epoch": 0.3233333333333333,
+      "grad_norm": 27.25,
+      "grad_norm_var": 1.9893229166666666,
+      "learning_rate": 7.637003630079111e-05,
+      "loss": 7.1366,
+      "loss/crossentropy": 1.7384035140275955,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13437513262033463,
+      "step": 1940
+    },
+    {
+      "epoch": 0.3235,
+      "grad_norm": 27.25,
+      "grad_norm_var": 1.6434895833333334,
+      "learning_rate": 7.634778977483389e-05,
+      "loss": 6.8072,
+      "loss/crossentropy": 1.5444829165935516,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21928342804312706,
+      "step": 1941
+    },
+    {
+      "epoch": 0.32366666666666666,
+      "grad_norm": 25.5,
+      "grad_norm_var": 1.92265625,
+      "learning_rate": 7.632553602548065e-05,
+      "loss": 6.6469,
+      "loss/crossentropy": 1.824027806520462,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1451159343123436,
+      "step": 1942
+    },
+    {
+      "epoch": 0.3238333333333333,
+      "grad_norm": 41.75,
+      "grad_norm_var": 14.668489583333333,
+      "learning_rate": 7.630327505883242e-05,
+      "loss": 7.3043,
+      "loss/crossentropy": 1.9817378222942352,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2217976227402687,
+      "step": 1943
+    },
+    {
+      "epoch": 0.324,
+      "grad_norm": 35.5,
+      "grad_norm_var": 17.859309895833334,
+      "learning_rate": 7.628100688099215e-05,
+      "loss": 6.8794,
+      "loss/crossentropy": 1.9806363582611084,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2619929276406765,
+      "step": 1944
+    },
+    {
+      "epoch": 0.32416666666666666,
+      "grad_norm": 29.625,
+      "grad_norm_var": 17.52890625,
+      "learning_rate": 7.62587314980648e-05,
+      "loss": 7.0545,
+      "loss/crossentropy": 1.897900551557541,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24334151297807693,
+      "step": 1945
+    },
+    {
+      "epoch": 0.3243333333333333,
+      "grad_norm": 27.375,
+      "grad_norm_var": 17.1056640625,
+      "learning_rate": 7.623644891615727e-05,
+      "loss": 6.9264,
+      "loss/crossentropy": 1.8679122924804688,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14723801240324974,
+      "step": 1946
+    },
+    {
+      "epoch": 0.3245,
+      "grad_norm": 25.875,
+      "grad_norm_var": 17.64765625,
+      "learning_rate": 7.621415914137846e-05,
+      "loss": 7.1159,
+      "loss/crossentropy": 1.3908490687608719,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12980337999761105,
+      "step": 1947
+    },
+    {
+      "epoch": 0.32466666666666666,
+      "grad_norm": 26.0,
+      "grad_norm_var": 17.9,
+      "learning_rate": 7.619186217983924e-05,
+      "loss": 7.1713,
+      "loss/crossentropy": 2.16692191362381,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2991831377148628,
+      "step": 1948
+    },
+    {
+      "epoch": 0.3248333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 17.908072916666665,
+      "learning_rate": 7.616955803765249e-05,
+      "loss": 6.9156,
+      "loss/crossentropy": 1.8130050897598267,
+      "loss/hidden": 3.10546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14834078960120678,
+      "step": 1949
+    },
+    {
+      "epoch": 0.325,
+      "grad_norm": 28.375,
+      "grad_norm_var": 17.911393229166666,
+      "learning_rate": 7.614724672093296e-05,
+      "loss": 6.89,
+      "loss/crossentropy": 1.773101955652237,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18298736214637756,
+      "step": 1950
+    },
+    {
+      "epoch": 0.32516666666666666,
+      "grad_norm": 27.25,
+      "grad_norm_var": 18.029166666666665,
+      "learning_rate": 7.612492823579745e-05,
+      "loss": 6.7182,
+      "loss/crossentropy": 1.7464256584644318,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3043161705136299,
+      "step": 1951
+    },
+    {
+      "epoch": 0.3253333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 17.755143229166666,
+      "learning_rate": 7.61026025883647e-05,
+      "loss": 6.8968,
+      "loss/crossentropy": 1.490617960691452,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16207384690642357,
+      "step": 1952
+    },
+    {
+      "epoch": 0.3255,
+      "grad_norm": 27.25,
+      "grad_norm_var": 17.800455729166668,
+      "learning_rate": 7.60802697847554e-05,
+      "loss": 6.8648,
+      "loss/crossentropy": 2.0520599484443665,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15137195214629173,
+      "step": 1953
+    },
+    {
+      "epoch": 0.32566666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 17.387434895833334,
+      "learning_rate": 7.605792983109222e-05,
+      "loss": 7.2909,
+      "loss/crossentropy": 1.2635005861520767,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11748159304261208,
+      "step": 1954
+    },
+    {
+      "epoch": 0.3258333333333333,
+      "grad_norm": 26.5,
+      "grad_norm_var": 17.207747395833334,
+      "learning_rate": 7.60355827334998e-05,
+      "loss": 6.9264,
+      "loss/crossentropy": 1.6250987499952316,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1348421834409237,
+      "step": 1955
+    },
+    {
+      "epoch": 0.326,
+      "grad_norm": 28.0,
+      "grad_norm_var": 17.071809895833333,
+      "learning_rate": 7.60132284981047e-05,
+      "loss": 7.0407,
+      "loss/crossentropy": 1.3203220516443253,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16120794974267483,
+      "step": 1956
+    },
+    {
+      "epoch": 0.32616666666666666,
+      "grad_norm": 40.25,
+      "grad_norm_var": 24.587434895833333,
+      "learning_rate": 7.599086713103547e-05,
+      "loss": 7.1496,
+      "loss/crossentropy": 2.502292364835739,
+      "loss/hidden": 3.0859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16814671456813812,
+      "step": 1957
+    },
+    {
+      "epoch": 0.3263333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 23.6962890625,
+      "learning_rate": 7.596849863842263e-05,
+      "loss": 6.4611,
+      "loss/crossentropy": 1.696190595626831,
+      "loss/hidden": 3.125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1337653361260891,
+      "step": 1958
+    },
+    {
+      "epoch": 0.3265,
+      "grad_norm": 33.0,
+      "grad_norm_var": 15.092122395833334,
+      "learning_rate": 7.594612302639859e-05,
+      "loss": 7.0354,
+      "loss/crossentropy": 1.6811990141868591,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1704744752496481,
+      "step": 1959
+    },
+    {
+      "epoch": 0.32666666666666666,
+      "grad_norm": 31.75,
+      "grad_norm_var": 13.084309895833334,
+      "learning_rate": 7.592374030109777e-05,
+      "loss": 6.8049,
+      "loss/crossentropy": 1.4199679493904114,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13939975388348103,
+      "step": 1960
+    },
+    {
+      "epoch": 0.3268333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 13.086458333333333,
+      "learning_rate": 7.590135046865651e-05,
+      "loss": 7.0515,
+      "loss/crossentropy": 1.9711971879005432,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1594713032245636,
+      "step": 1961
+    },
+    {
+      "epoch": 0.327,
+      "grad_norm": 25.625,
+      "grad_norm_var": 13.76640625,
+      "learning_rate": 7.587895353521314e-05,
+      "loss": 7.0146,
+      "loss/crossentropy": 1.7343676090240479,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18260842934250832,
+      "step": 1962
+    },
+    {
+      "epoch": 0.32716666666666666,
+      "grad_norm": 37.25,
+      "grad_norm_var": 16.568684895833332,
+      "learning_rate": 7.585654950690786e-05,
+      "loss": 6.9841,
+      "loss/crossentropy": 1.4708878099918365,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1759204864501953,
+      "step": 1963
+    },
+    {
+      "epoch": 0.3273333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 15.4650390625,
+      "learning_rate": 7.58341383898829e-05,
+      "loss": 6.7756,
+      "loss/crossentropy": 1.9903148412704468,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17452102340757847,
+      "step": 1964
+    },
+    {
+      "epoch": 0.3275,
+      "grad_norm": 27.625,
+      "grad_norm_var": 15.816666666666666,
+      "learning_rate": 7.581172019028238e-05,
+      "loss": 7.214,
+      "loss/crossentropy": 1.9448735415935516,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22730185091495514,
+      "step": 1965
+    },
+    {
+      "epoch": 0.32766666666666666,
+      "grad_norm": 26.25,
+      "grad_norm_var": 16.612434895833335,
+      "learning_rate": 7.578929491425238e-05,
+      "loss": 6.7937,
+      "loss/crossentropy": 1.6424171030521393,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14761119335889816,
+      "step": 1966
+    },
+    {
+      "epoch": 0.3278333333333333,
+      "grad_norm": 43.5,
+      "grad_norm_var": 27.039518229166667,
+      "learning_rate": 7.576686256794091e-05,
+      "loss": 7.3124,
+      "loss/crossentropy": 1.8280293941497803,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20526448637247086,
+      "step": 1967
+    },
+    {
+      "epoch": 0.328,
+      "grad_norm": 28.25,
+      "grad_norm_var": 27.372916666666665,
+      "learning_rate": 7.574442315749793e-05,
+      "loss": 6.8749,
+      "loss/crossentropy": 1.377597153186798,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14071182627230883,
+      "step": 1968
+    },
+    {
+      "epoch": 0.32816666666666666,
+      "grad_norm": 35.0,
+      "grad_norm_var": 27.251822916666665,
+      "learning_rate": 7.572197668907532e-05,
+      "loss": 6.492,
+      "loss/crossentropy": 1.859922617673874,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13686381466686726,
+      "step": 1969
+    },
+    {
+      "epoch": 0.3283333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 27.09140625,
+      "learning_rate": 7.569952316882694e-05,
+      "loss": 6.7999,
+      "loss/crossentropy": 1.4993602633476257,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17878538742661476,
+      "step": 1970
+    },
+    {
+      "epoch": 0.3285,
+      "grad_norm": 33.25,
+      "grad_norm_var": 25.25625,
+      "learning_rate": 7.567706260290851e-05,
+      "loss": 7.0913,
+      "loss/crossentropy": 1.3628067672252655,
+      "loss/hidden": 3.859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14887463301420212,
+      "step": 1971
+    },
+    {
+      "epoch": 0.32866666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 24.4447265625,
+      "learning_rate": 7.565459499747775e-05,
+      "loss": 7.0248,
+      "loss/crossentropy": 1.9492396712303162,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21044185012578964,
+      "step": 1972
+    },
+    {
+      "epoch": 0.3288333333333333,
+      "grad_norm": 27.625,
+      "grad_norm_var": 20.926822916666666,
+      "learning_rate": 7.563212035869425e-05,
+      "loss": 6.9809,
+      "loss/crossentropy": 1.8395329862833023,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16357857547700405,
+      "step": 1973
+    },
+    {
+      "epoch": 0.329,
+      "grad_norm": 27.75,
+      "grad_norm_var": 21.624739583333334,
+      "learning_rate": 7.56096386927196e-05,
+      "loss": 7.0102,
+      "loss/crossentropy": 1.7907786071300507,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18460842221975327,
+      "step": 1974
+    },
+    {
+      "epoch": 0.32916666666666666,
+      "grad_norm": 28.75,
+      "grad_norm_var": 21.7,
+      "learning_rate": 7.558715000571726e-05,
+      "loss": 7.0319,
+      "loss/crossentropy": 1.6687232851982117,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2605494260787964,
+      "step": 1975
+    },
+    {
+      "epoch": 0.3293333333333333,
+      "grad_norm": 27.125,
+      "grad_norm_var": 22.497330729166666,
+      "learning_rate": 7.55646543038526e-05,
+      "loss": 6.4879,
+      "loss/crossentropy": 1.8237475007772446,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1547717321664095,
+      "step": 1976
+    },
+    {
+      "epoch": 0.3295,
+      "grad_norm": 27.25,
+      "grad_norm_var": 23.103580729166666,
+      "learning_rate": 7.5542151593293e-05,
+      "loss": 7.1899,
+      "loss/crossentropy": 1.8912782967090607,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19629128649830818,
+      "step": 1977
+    },
+    {
+      "epoch": 0.32966666666666666,
+      "grad_norm": 28.625,
+      "grad_norm_var": 21.731705729166666,
+      "learning_rate": 7.551964188020766e-05,
+      "loss": 6.9527,
+      "loss/crossentropy": 1.7356953918933868,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2349882610142231,
+      "step": 1978
+    },
+    {
+      "epoch": 0.3298333333333333,
+      "grad_norm": 26.875,
+      "grad_norm_var": 19.327083333333334,
+      "learning_rate": 7.549712517076777e-05,
+      "loss": 7.1087,
+      "loss/crossentropy": 1.9298615157604218,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14285527169704437,
+      "step": 1979
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 26.375,
+      "grad_norm_var": 20.131184895833332,
+      "learning_rate": 7.547460147114641e-05,
+      "loss": 7.0014,
+      "loss/crossentropy": 1.7447529137134552,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18788661807775497,
+      "step": 1980
+    },
+    {
+      "epoch": 0.33016666666666666,
+      "grad_norm": 40.25,
+      "grad_norm_var": 26.39765625,
+      "learning_rate": 7.545207078751857e-05,
+      "loss": 7.2777,
+      "loss/crossentropy": 2.3176685869693756,
+      "loss/hidden": 3.7421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.43646563962101936,
+      "step": 1981
+    },
+    {
+      "epoch": 0.3303333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 25.16015625,
+      "learning_rate": 7.542953312606117e-05,
+      "loss": 7.4115,
+      "loss/crossentropy": 1.1465453654527664,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17520064301788807,
+      "step": 1982
+    },
+    {
+      "epoch": 0.3305,
+      "grad_norm": 31.125,
+      "grad_norm_var": 14.0806640625,
+      "learning_rate": 7.540698849295305e-05,
+      "loss": 7.211,
+      "loss/crossentropy": 1.9278732538223267,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1668613739311695,
+      "step": 1983
+    },
+    {
+      "epoch": 0.33066666666666666,
+      "grad_norm": 29.25,
+      "grad_norm_var": 13.881705729166667,
+      "learning_rate": 7.538443689437492e-05,
+      "loss": 7.1892,
+      "loss/crossentropy": 1.958818107843399,
+      "loss/hidden": 3.79296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28305254876613617,
+      "step": 1984
+    },
+    {
+      "epoch": 0.3308333333333333,
+      "grad_norm": 27.75,
+      "grad_norm_var": 12.5978515625,
+      "learning_rate": 7.536187833650947e-05,
+      "loss": 6.5296,
+      "loss/crossentropy": 1.821618765592575,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1746337190270424,
+      "step": 1985
+    },
+    {
+      "epoch": 0.331,
+      "grad_norm": 27.125,
+      "grad_norm_var": 12.264322916666666,
+      "learning_rate": 7.53393128255412e-05,
+      "loss": 6.7454,
+      "loss/crossentropy": 1.7255724370479584,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12872214615345,
+      "step": 1986
+    },
+    {
+      "epoch": 0.33116666666666666,
+      "grad_norm": 28.125,
+      "grad_norm_var": 11.311393229166667,
+      "learning_rate": 7.531674036765662e-05,
+      "loss": 6.9214,
+      "loss/crossentropy": 1.4453624933958054,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1719880998134613,
+      "step": 1987
+    },
+    {
+      "epoch": 0.3313333333333333,
+      "grad_norm": 26.875,
+      "grad_norm_var": 11.577018229166667,
+      "learning_rate": 7.52941609690441e-05,
+      "loss": 7.2681,
+      "loss/crossentropy": 1.7877110838890076,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20354793220758438,
+      "step": 1988
+    },
+    {
+      "epoch": 0.3315,
+      "grad_norm": 28.875,
+      "grad_norm_var": 11.454622395833333,
+      "learning_rate": 7.52715746358939e-05,
+      "loss": 6.8813,
+      "loss/crossentropy": 1.6874558627605438,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16517934203147888,
+      "step": 1989
+    },
+    {
+      "epoch": 0.33166666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 11.555143229166667,
+      "learning_rate": 7.524898137439814e-05,
+      "loss": 7.0432,
+      "loss/crossentropy": 1.3462169617414474,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23449290171265602,
+      "step": 1990
+    },
+    {
+      "epoch": 0.3318333333333333,
+      "grad_norm": 27.75,
+      "grad_norm_var": 11.649934895833333,
+      "learning_rate": 7.522638119075096e-05,
+      "loss": 6.974,
+      "loss/crossentropy": 1.558683604001999,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2526928149163723,
+      "step": 1991
+    },
+    {
+      "epoch": 0.332,
+      "grad_norm": 27.25,
+      "grad_norm_var": 11.620833333333334,
+      "learning_rate": 7.520377409114831e-05,
+      "loss": 7.0326,
+      "loss/crossentropy": 1.7414982914924622,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1637286227196455,
+      "step": 1992
+    },
+    {
+      "epoch": 0.33216666666666667,
+      "grad_norm": 34.5,
+      "grad_norm_var": 13.274739583333334,
+      "learning_rate": 7.518116008178805e-05,
+      "loss": 6.6616,
+      "loss/crossentropy": 1.0742476731538773,
+      "loss/hidden": 3.8515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1765630804002285,
+      "step": 1993
+    },
+    {
+      "epoch": 0.3323333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 13.384830729166667,
+      "learning_rate": 7.515853916886993e-05,
+      "loss": 6.7673,
+      "loss/crossentropy": 1.6484030038118362,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1557282730937004,
+      "step": 1994
+    },
+    {
+      "epoch": 0.3325,
+      "grad_norm": 31.125,
+      "grad_norm_var": 13.0041015625,
+      "learning_rate": 7.513591135859561e-05,
+      "loss": 7.0006,
+      "loss/crossentropy": 1.8683974295854568,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1803109608590603,
+      "step": 1995
+    },
+    {
+      "epoch": 0.33266666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 12.205989583333333,
+      "learning_rate": 7.511327665716863e-05,
+      "loss": 7.0199,
+      "loss/crossentropy": 2.0102295577526093,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19259504601359367,
+      "step": 1996
+    },
+    {
+      "epoch": 0.3328333333333333,
+      "grad_norm": 27.0,
+      "grad_norm_var": 5.042708333333334,
+      "learning_rate": 7.509063507079443e-05,
+      "loss": 6.9124,
+      "loss/crossentropy": 1.9524219632148743,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1803882084786892,
+      "step": 1997
+    },
+    {
+      "epoch": 0.333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 4.380989583333333,
+      "learning_rate": 7.506798660568031e-05,
+      "loss": 7.1799,
+      "loss/crossentropy": 1.6741455495357513,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15316328406333923,
+      "step": 1998
+    },
+    {
+      "epoch": 0.33316666666666667,
+      "grad_norm": 27.375,
+      "grad_norm_var": 4.189583333333333,
+      "learning_rate": 7.50453312680355e-05,
+      "loss": 6.8463,
+      "loss/crossentropy": 1.2846168503165245,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16056417115032673,
+      "step": 1999
+    },
+    {
+      "epoch": 0.3333333333333333,
+      "grad_norm": 26.625,
+      "grad_norm_var": 4.445247395833333,
+      "learning_rate": 7.502266906407107e-05,
+      "loss": 6.4687,
+      "loss/crossentropy": 1.4025170803070068,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1475424598902464,
+      "step": 2000
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 6000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 9223372036854775807,
+  "save_steps": 2000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 5.715019849269248e+18,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}