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": 73.5,
+      "learning_rate": 0.0001,
+      "loss": 7.9671,
+      "loss/crossentropy": 2.569203495979309,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18632682040333748,
+      "step": 1
+    },
+    {
+      "epoch": 0.0003333333333333333,
+      "grad_norm": 93.0,
+      "learning_rate": 9.999999314610822e-05,
+      "loss": 7.8943,
+      "loss/crossentropy": 1.845904678106308,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19127092137932777,
+      "step": 2
+    },
+    {
+      "epoch": 0.0005,
+      "grad_norm": 40.0,
+      "learning_rate": 9.999997258443473e-05,
+      "loss": 7.7016,
+      "loss/crossentropy": 2.047421157360077,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20494147762656212,
+      "step": 3
+    },
+    {
+      "epoch": 0.0006666666666666666,
+      "grad_norm": 37.5,
+      "learning_rate": 9.999993831498517e-05,
+      "loss": 7.8427,
+      "loss/crossentropy": 1.876516431570053,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24268608912825584,
+      "step": 4
+    },
+    {
+      "epoch": 0.0008333333333333334,
+      "grad_norm": 37.25,
+      "learning_rate": 9.999989033776898e-05,
+      "loss": 7.8733,
+      "loss/crossentropy": 2.017443597316742,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24553321674466133,
+      "step": 5
+    },
+    {
+      "epoch": 0.001,
+      "grad_norm": 58.75,
+      "learning_rate": 9.999982865279924e-05,
+      "loss": 7.9913,
+      "loss/crossentropy": 2.0913399159908295,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18808197975158691,
+      "step": 6
+    },
+    {
+      "epoch": 0.0011666666666666668,
+      "grad_norm": 49.75,
+      "learning_rate": 9.999975326009292e-05,
+      "loss": 7.8335,
+      "loss/crossentropy": 2.122416466474533,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22057654336094856,
+      "step": 7
+    },
+    {
+      "epoch": 0.0013333333333333333,
+      "grad_norm": 37.5,
+      "learning_rate": 9.999966415967066e-05,
+      "loss": 7.5297,
+      "loss/crossentropy": 1.8422361314296722,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17389418557286263,
+      "step": 8
+    },
+    {
+      "epoch": 0.0015,
+      "grad_norm": 34.75,
+      "learning_rate": 9.999956135155687e-05,
+      "loss": 7.747,
+      "loss/crossentropy": 2.0487734377384186,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19654536992311478,
+      "step": 9
+    },
+    {
+      "epoch": 0.0016666666666666668,
+      "grad_norm": 41.75,
+      "learning_rate": 9.999944483577981e-05,
+      "loss": 7.6389,
+      "loss/crossentropy": 2.0523241758346558,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29397766664624214,
+      "step": 10
+    },
+    {
+      "epoch": 0.0018333333333333333,
+      "grad_norm": 36.25,
+      "learning_rate": 9.999931461237134e-05,
+      "loss": 7.7575,
+      "loss/crossentropy": 2.1524189710617065,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20942805707454681,
+      "step": 11
+    },
+    {
+      "epoch": 0.002,
+      "grad_norm": 34.25,
+      "learning_rate": 9.999917068136722e-05,
+      "loss": 7.8672,
+      "loss/crossentropy": 2.24754399061203,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23748969286680222,
+      "step": 12
+    },
+    {
+      "epoch": 0.0021666666666666666,
+      "grad_norm": 33.75,
+      "learning_rate": 9.999901304280685e-05,
+      "loss": 7.7629,
+      "loss/crossentropy": 1.634619951248169,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18291744217276573,
+      "step": 13
+    },
+    {
+      "epoch": 0.0023333333333333335,
+      "grad_norm": 41.0,
+      "learning_rate": 9.999884169673351e-05,
+      "loss": 7.8668,
+      "loss/crossentropy": 1.7690042853355408,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22475926205515862,
+      "step": 14
+    },
+    {
+      "epoch": 0.0025,
+      "grad_norm": 37.25,
+      "learning_rate": 9.999865664319414e-05,
+      "loss": 7.635,
+      "loss/crossentropy": 1.859389066696167,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2027263380587101,
+      "step": 15
+    },
+    {
+      "epoch": 0.0026666666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 280.3080729166667,
+      "learning_rate": 9.999845788223949e-05,
+      "loss": 7.5468,
+      "loss/crossentropy": 2.26497346162796,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.181312408298254,
+      "step": 16
+    },
+    {
+      "epoch": 0.0028333333333333335,
+      "grad_norm": 33.75,
+      "grad_norm_var": 227.43229166666666,
+      "learning_rate": 9.999824541392405e-05,
+      "loss": 7.7265,
+      "loss/crossentropy": 2.273234874010086,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21036671847105026,
+      "step": 17
+    },
+    {
+      "epoch": 0.003,
+      "grad_norm": 41.25,
+      "grad_norm_var": 45.71432291666667,
+      "learning_rate": 9.999801923830603e-05,
+      "loss": 7.7958,
+      "loss/crossentropy": 1.7472622692584991,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22986393049359322,
+      "step": 18
+    },
+    {
+      "epoch": 0.0031666666666666666,
+      "grad_norm": 43.0,
+      "grad_norm_var": 46.608072916666664,
+      "learning_rate": 9.99977793554475e-05,
+      "loss": 7.4669,
+      "loss/crossentropy": 2.0385634303092957,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2178351879119873,
+      "step": 19
+    },
+    {
+      "epoch": 0.0033333333333333335,
+      "grad_norm": 39.25,
+      "grad_norm_var": 46.365625,
+      "learning_rate": 9.999752576541418e-05,
+      "loss": 7.5922,
+      "loss/crossentropy": 1.7339409589767456,
+      "loss/hidden": 3.9453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26616106554865837,
+      "step": 20
+    },
+    {
+      "epoch": 0.0035,
+      "grad_norm": 33.75,
+      "grad_norm_var": 48.166666666666664,
+      "learning_rate": 9.999725846827562e-05,
+      "loss": 7.6232,
+      "loss/crossentropy": 1.7600694447755814,
+      "loss/hidden": 3.79296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23021411895751953,
+      "step": 21
+    },
+    {
+      "epoch": 0.0036666666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 22.983072916666668,
+      "learning_rate": 9.999697746410508e-05,
+      "loss": 7.7923,
+      "loss/crossentropy": 2.362265467643738,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17758158221840858,
+      "step": 22
+    },
+    {
+      "epoch": 0.003833333333333333,
+      "grad_norm": 36.0,
+      "grad_norm_var": 12.541666666666666,
+      "learning_rate": 9.99966827529796e-05,
+      "loss": 7.7335,
+      "loss/crossentropy": 1.964929312467575,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2139221392571926,
+      "step": 23
+    },
+    {
+      "epoch": 0.004,
+      "grad_norm": 33.75,
+      "grad_norm_var": 13.045572916666666,
+      "learning_rate": 9.999637433497999e-05,
+      "loss": 7.7097,
+      "loss/crossentropy": 1.7122240215539932,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2085253857076168,
+      "step": 24
+    },
+    {
+      "epoch": 0.004166666666666667,
+      "grad_norm": 35.25,
+      "grad_norm_var": 12.943489583333333,
+      "learning_rate": 9.999605221019081e-05,
+      "loss": 7.8487,
+      "loss/crossentropy": 1.3764294236898422,
+      "loss/hidden": 4.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25221870094537735,
+      "step": 25
+    },
+    {
+      "epoch": 0.004333333333333333,
+      "grad_norm": 35.5,
+      "grad_norm_var": 11.048958333333333,
+      "learning_rate": 9.999571637870036e-05,
+      "loss": 7.6111,
+      "loss/crossentropy": 2.3533842265605927,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19117553159594536,
+      "step": 26
+    },
+    {
+      "epoch": 0.0045,
+      "grad_norm": 40.25,
+      "grad_norm_var": 12.098958333333334,
+      "learning_rate": 9.99953668406007e-05,
+      "loss": 7.7699,
+      "loss/crossentropy": 1.930314987897873,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24592405930161476,
+      "step": 27
+    },
+    {
+      "epoch": 0.004666666666666667,
+      "grad_norm": 41.25,
+      "grad_norm_var": 13.148958333333333,
+      "learning_rate": 9.999500359598768e-05,
+      "loss": 7.702,
+      "loss/crossentropy": 2.1383188664913177,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1850673332810402,
+      "step": 28
+    },
+    {
+      "epoch": 0.004833333333333334,
+      "grad_norm": 36.75,
+      "grad_norm_var": 12.473958333333334,
+      "learning_rate": 9.999462664496088e-05,
+      "loss": 7.4131,
+      "loss/crossentropy": 2.1065402925014496,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20391792058944702,
+      "step": 29
+    },
+    {
+      "epoch": 0.005,
+      "grad_norm": 34.75,
+      "grad_norm_var": 11.608072916666666,
+      "learning_rate": 9.999423598762363e-05,
+      "loss": 7.5963,
+      "loss/crossentropy": 2.387889564037323,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18003453686833382,
+      "step": 30
+    },
+    {
+      "epoch": 0.005166666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 11.926822916666667,
+      "learning_rate": 9.999383162408304e-05,
+      "loss": 7.7149,
+      "loss/crossentropy": 1.8544995188713074,
+      "loss/hidden": 3.84765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22001322731375694,
+      "step": 31
+    },
+    {
+      "epoch": 0.005333333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 10.907291666666667,
+      "learning_rate": 9.999341355444995e-05,
+      "loss": 7.7558,
+      "loss/crossentropy": 1.8552018105983734,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19524798914790154,
+      "step": 32
+    },
+    {
+      "epoch": 0.0055,
+      "grad_norm": 37.25,
+      "grad_norm_var": 10.345833333333333,
+      "learning_rate": 9.999298177883903e-05,
+      "loss": 7.7273,
+      "loss/crossentropy": 1.7090170681476593,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20817053690552711,
+      "step": 33
+    },
+    {
+      "epoch": 0.005666666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 9.718489583333334,
+      "learning_rate": 9.99925362973686e-05,
+      "loss": 7.5989,
+      "loss/crossentropy": 2.094772696495056,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21507051959633827,
+      "step": 34
+    },
+    {
+      "epoch": 0.005833333333333334,
+      "grad_norm": 38.25,
+      "grad_norm_var": 6.883333333333334,
+      "learning_rate": 9.999207711016081e-05,
+      "loss": 7.8209,
+      "loss/crossentropy": 1.6093821227550507,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21887024864554405,
+      "step": 35
+    },
+    {
+      "epoch": 0.006,
+      "grad_norm": 45.5,
+      "grad_norm_var": 12.033072916666667,
+      "learning_rate": 9.999160421734155e-05,
+      "loss": 7.4932,
+      "loss/crossentropy": 2.328737795352936,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18808270618319511,
+      "step": 36
+    },
+    {
+      "epoch": 0.006166666666666667,
+      "grad_norm": 42.25,
+      "grad_norm_var": 13.555989583333334,
+      "learning_rate": 9.999111761904046e-05,
+      "loss": 7.6479,
+      "loss/crossentropy": 2.154049277305603,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20653583109378815,
+      "step": 37
+    },
+    {
+      "epoch": 0.006333333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 14.1197265625,
+      "learning_rate": 9.999061731539094e-05,
+      "loss": 7.4967,
+      "loss/crossentropy": 2.116226941347122,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20157190039753914,
+      "step": 38
+    },
+    {
+      "epoch": 0.0065,
+      "grad_norm": 31.0,
+      "grad_norm_var": 16.2603515625,
+      "learning_rate": 9.999010330653018e-05,
+      "loss": 7.4664,
+      "loss/crossentropy": 2.212589740753174,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19786111637949944,
+      "step": 39
+    },
+    {
+      "epoch": 0.006666666666666667,
+      "grad_norm": 40.25,
+      "grad_norm_var": 16.470247395833333,
+      "learning_rate": 9.998957559259906e-05,
+      "loss": 7.6662,
+      "loss/crossentropy": 2.4179359674453735,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23757457360625267,
+      "step": 40
+    },
+    {
+      "epoch": 0.006833333333333334,
+      "grad_norm": 37.25,
+      "grad_norm_var": 16.263997395833332,
+      "learning_rate": 9.998903417374228e-05,
+      "loss": 7.6428,
+      "loss/crossentropy": 2.329847365617752,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19405298680067062,
+      "step": 41
+    },
+    {
+      "epoch": 0.007,
+      "grad_norm": 35.0,
+      "grad_norm_var": 16.3853515625,
+      "learning_rate": 9.998847905010826e-05,
+      "loss": 7.6838,
+      "loss/crossentropy": 1.7561518549919128,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1967286802828312,
+      "step": 42
+    },
+    {
+      "epoch": 0.007166666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 16.256705729166665,
+      "learning_rate": 9.998791022184922e-05,
+      "loss": 7.5731,
+      "loss/crossentropy": 1.820853978395462,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2205936461687088,
+      "step": 43
+    },
+    {
+      "epoch": 0.007333333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 15.2556640625,
+      "learning_rate": 9.998732768912104e-05,
+      "loss": 7.8459,
+      "loss/crossentropy": 1.3915326595306396,
+      "loss/hidden": 3.96484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2971978336572647,
+      "step": 44
+    },
+    {
+      "epoch": 0.0075,
+      "grad_norm": 35.25,
+      "grad_norm_var": 15.2791015625,
+      "learning_rate": 9.99867314520835e-05,
+      "loss": 7.5113,
+      "loss/crossentropy": 1.7233994901180267,
+      "loss/hidden": 3.73828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19584518671035767,
+      "step": 45
+    },
+    {
+      "epoch": 0.007666666666666666,
+      "grad_norm": 39.75,
+      "grad_norm_var": 15.961393229166667,
+      "learning_rate": 9.998612151090003e-05,
+      "loss": 7.6346,
+      "loss/crossentropy": 1.3916895985603333,
+      "loss/hidden": 3.96484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24776612222194672,
+      "step": 46
+    },
+    {
+      "epoch": 0.007833333333333333,
+      "grad_norm": 37.5,
+      "grad_norm_var": 15.697330729166667,
+      "learning_rate": 9.998549786573785e-05,
+      "loss": 7.5357,
+      "loss/crossentropy": 1.95862877368927,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2236441969871521,
+      "step": 47
+    },
+    {
+      "epoch": 0.008,
+      "grad_norm": 35.5,
+      "grad_norm_var": 15.4056640625,
+      "learning_rate": 9.998486051676792e-05,
+      "loss": 7.4207,
+      "loss/crossentropy": 1.8571364283561707,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.201379906386137,
+      "step": 48
+    },
+    {
+      "epoch": 0.008166666666666666,
+      "grad_norm": 37.0,
+      "grad_norm_var": 15.3900390625,
+      "learning_rate": 9.9984209464165e-05,
+      "loss": 7.6778,
+      "loss/crossentropy": 2.2022317349910736,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18588031828403473,
+      "step": 49
+    },
+    {
+      "epoch": 0.008333333333333333,
+      "grad_norm": 34.5,
+      "grad_norm_var": 14.8009765625,
+      "learning_rate": 9.998354470810757e-05,
+      "loss": 7.7697,
+      "loss/crossentropy": 1.8480720520019531,
+      "loss/hidden": 3.73828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3351767808198929,
+      "step": 50
+    },
+    {
+      "epoch": 0.0085,
+      "grad_norm": 35.75,
+      "grad_norm_var": 14.688997395833333,
+      "learning_rate": 9.998286624877786e-05,
+      "loss": 7.5311,
+      "loss/crossentropy": 1.880394697189331,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20984028279781342,
+      "step": 51
+    },
+    {
+      "epoch": 0.008666666666666666,
+      "grad_norm": 38.75,
+      "grad_norm_var": 9.513997395833334,
+      "learning_rate": 9.99821740863619e-05,
+      "loss": 7.6409,
+      "loss/crossentropy": 2.4952712059020996,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2372022569179535,
+      "step": 52
+    },
+    {
+      "epoch": 0.008833333333333334,
+      "grad_norm": 3892314112.0,
+      "grad_norm_var": 9.468818042568137e+17,
+      "learning_rate": 9.998146822104943e-05,
+      "loss": 9.0543,
+      "loss/crossentropy": 1.5961681604385376,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18118682876229286,
+      "step": 53
+    },
+    {
+      "epoch": 0.009,
+      "grad_norm": 44.0,
+      "grad_norm_var": 9.468818038554188e+17,
+      "learning_rate": 9.998074865303399e-05,
+      "loss": 7.5862,
+      "loss/crossentropy": 2.1867645978927612,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2127695530653,
+      "step": 54
+    },
+    {
+      "epoch": 0.009166666666666667,
+      "grad_norm": 36.75,
+      "grad_norm_var": 9.468818036689121e+17,
+      "learning_rate": 9.998001538251282e-05,
+      "loss": 7.6058,
+      "loss/crossentropy": 2.333953231573105,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1927742250263691,
+      "step": 55
+    },
+    {
+      "epoch": 0.009333333333333334,
+      "grad_norm": 37.25,
+      "grad_norm_var": 9.468818037662199e+17,
+      "learning_rate": 9.997926840968699e-05,
+      "loss": 7.3744,
+      "loss/crossentropy": 2.0817086696624756,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21321595087647438,
+      "step": 56
+    },
+    {
+      "epoch": 0.0095,
+      "grad_norm": 35.5,
+      "grad_norm_var": 9.468818038229829e+17,
+      "learning_rate": 9.997850773476126e-05,
+      "loss": 7.5238,
+      "loss/crossentropy": 2.0765497386455536,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20579104870557785,
+      "step": 57
+    },
+    {
+      "epoch": 0.009666666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 9.468818038554188e+17,
+      "learning_rate": 9.997773335794416e-05,
+      "loss": 7.4419,
+      "loss/crossentropy": 1.7816000133752823,
+      "loss/hidden": 3.71484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20738175511360168,
+      "step": 58
+    },
+    {
+      "epoch": 0.009833333333333333,
+      "grad_norm": 34.5,
+      "grad_norm_var": 9.468818038310918e+17,
+      "learning_rate": 9.997694527944803e-05,
+      "loss": 7.5833,
+      "loss/crossentropy": 2.397744208574295,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16810957714915276,
+      "step": 59
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 36.75,
+      "grad_norm_var": 9.46881803725675e+17,
+      "learning_rate": 9.99761434994889e-05,
+      "loss": 7.5848,
+      "loss/crossentropy": 1.4827385395765305,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18737642094492912,
+      "step": 60
+    },
+    {
+      "epoch": 0.010166666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 9.468818038229829e+17,
+      "learning_rate": 9.997532801828658e-05,
+      "loss": 7.4993,
+      "loss/crossentropy": 2.182910054922104,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20538238808512688,
+      "step": 61
+    },
+    {
+      "epoch": 0.010333333333333333,
+      "grad_norm": 34.75,
+      "grad_norm_var": 9.468818039851626e+17,
+      "learning_rate": 9.997449883606466e-05,
+      "loss": 7.6736,
+      "loss/crossentropy": 2.0504411458969116,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19961906969547272,
+      "step": 62
+    },
+    {
+      "epoch": 0.0105,
+      "grad_norm": 31.875,
+      "grad_norm_var": 9.468818041676148e+17,
+      "learning_rate": 9.997365595305044e-05,
+      "loss": 7.5312,
+      "loss/crossentropy": 1.8536975383758545,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1750146523118019,
+      "step": 63
+    },
+    {
+      "epoch": 0.010666666666666666,
+      "grad_norm": 35.5,
+      "grad_norm_var": 9.468818041676148e+17,
+      "learning_rate": 9.997279936947502e-05,
+      "loss": 7.6283,
+      "loss/crossentropy": 1.9070446193218231,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1809525452554226,
+      "step": 64
+    },
+    {
+      "epoch": 0.010833333333333334,
+      "grad_norm": 34.0,
+      "grad_norm_var": 9.468818042649226e+17,
+      "learning_rate": 9.997192908557323e-05,
+      "loss": 7.5086,
+      "loss/crossentropy": 1.9207460284233093,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17440685257315636,
+      "step": 65
+    },
+    {
+      "epoch": 0.011,
+      "grad_norm": 36.75,
+      "grad_norm_var": 9.468818041919418e+17,
+      "learning_rate": 9.997104510158365e-05,
+      "loss": 7.3597,
+      "loss/crossentropy": 2.178062379360199,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20515967160463333,
+      "step": 66
+    },
+    {
+      "epoch": 0.011166666666666667,
+      "grad_norm": 39.5,
+      "grad_norm_var": 9.46881804070307e+17,
+      "learning_rate": 9.997014741774866e-05,
+      "loss": 7.7332,
+      "loss/crossentropy": 2.0632802546024323,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2130327671766281,
+      "step": 67
+    },
+    {
+      "epoch": 0.011333333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 9.468818042568137e+17,
+      "learning_rate": 9.996923603431433e-05,
+      "loss": 7.4813,
+      "loss/crossentropy": 2.3944068551063538,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.190116498619318,
+      "step": 68
+    },
+    {
+      "epoch": 0.0115,
+      "grad_norm": 34.0,
+      "grad_norm_var": 8.9275390625,
+      "learning_rate": 9.996831095153055e-05,
+      "loss": 7.4656,
+      "loss/crossentropy": 1.6360968053340912,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19409537687897682,
+      "step": 69
+    },
+    {
+      "epoch": 0.011666666666666667,
+      "grad_norm": 42.25,
+      "grad_norm_var": 7.170247395833333,
+      "learning_rate": 9.996737216965092e-05,
+      "loss": 7.6666,
+      "loss/crossentropy": 1.496591791510582,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2146080546081066,
+      "step": 70
+    },
+    {
+      "epoch": 0.011833333333333333,
+      "grad_norm": 35.75,
+      "grad_norm_var": 7.0712890625,
+      "learning_rate": 9.996641968893282e-05,
+      "loss": 7.4798,
+      "loss/crossentropy": 2.3171346783638,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18420026078820229,
+      "step": 71
+    },
+    {
+      "epoch": 0.012,
+      "grad_norm": 33.75,
+      "grad_norm_var": 7.009309895833334,
+      "learning_rate": 9.996545350963738e-05,
+      "loss": 7.5509,
+      "loss/crossentropy": 1.4588509052991867,
+      "loss/hidden": 3.7109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2755301594734192,
+      "step": 72
+    },
+    {
+      "epoch": 0.012166666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 7.713541666666667,
+      "learning_rate": 9.996447363202946e-05,
+      "loss": 7.4458,
+      "loss/crossentropy": 2.4704632461071014,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2312110774219036,
+      "step": 73
+    },
+    {
+      "epoch": 0.012333333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 8.960416666666667,
+      "learning_rate": 9.996348005637775e-05,
+      "loss": 7.3695,
+      "loss/crossentropy": 2.1731217801570892,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19828997924923897,
+      "step": 74
+    },
+    {
+      "epoch": 0.0125,
+      "grad_norm": 37.25,
+      "grad_norm_var": 9.318489583333333,
+      "learning_rate": 9.996247278295458e-05,
+      "loss": 7.5469,
+      "loss/crossentropy": 1.7659152448177338,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23145361244678497,
+      "step": 75
+    },
+    {
+      "epoch": 0.012666666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 9.446875,
+      "learning_rate": 9.996145181203615e-05,
+      "loss": 7.678,
+      "loss/crossentropy": 2.506075084209442,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22724350914359093,
+      "step": 76
+    },
+    {
+      "epoch": 0.012833333333333334,
+      "grad_norm": 35.5,
+      "grad_norm_var": 9.037239583333333,
+      "learning_rate": 9.996041714390235e-05,
+      "loss": 7.7375,
+      "loss/crossentropy": 1.8532120883464813,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2014121375977993,
+      "step": 77
+    },
+    {
+      "epoch": 0.013,
+      "grad_norm": 34.25,
+      "grad_norm_var": 9.064322916666667,
+      "learning_rate": 9.995936877883682e-05,
+      "loss": 7.6128,
+      "loss/crossentropy": 2.0054582357406616,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2382129207253456,
+      "step": 78
+    },
+    {
+      "epoch": 0.013166666666666667,
+      "grad_norm": 34.75,
+      "grad_norm_var": 8.424934895833333,
+      "learning_rate": 9.9958306717127e-05,
+      "loss": 7.636,
+      "loss/crossentropy": 2.062336191534996,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2009948529303074,
+      "step": 79
+    },
+    {
+      "epoch": 0.013333333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 8.815559895833333,
+      "learning_rate": 9.995723095906407e-05,
+      "loss": 7.8122,
+      "loss/crossentropy": 2.1655342876911163,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1831391118466854,
+      "step": 80
+    },
+    {
+      "epoch": 0.0135,
+      "grad_norm": 34.75,
+      "grad_norm_var": 8.762434895833334,
+      "learning_rate": 9.995614150494293e-05,
+      "loss": 7.9923,
+      "loss/crossentropy": 1.973993867635727,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22885655611753464,
+      "step": 81
+    },
+    {
+      "epoch": 0.013666666666666667,
+      "grad_norm": 34.5,
+      "grad_norm_var": 8.532747395833333,
+      "learning_rate": 9.995503835506226e-05,
+      "loss": 7.5239,
+      "loss/crossentropy": 1.8596991300582886,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2005896307528019,
+      "step": 82
+    },
+    {
+      "epoch": 0.013833333333333333,
+      "grad_norm": 37.5,
+      "grad_norm_var": 7.526497395833333,
+      "learning_rate": 9.995392150972451e-05,
+      "loss": 7.7656,
+      "loss/crossentropy": 1.86503666639328,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2173938862979412,
+      "step": 83
+    },
+    {
+      "epoch": 0.014,
+      "grad_norm": 35.0,
+      "grad_norm_var": 7.332747395833334,
+      "learning_rate": 9.995279096923585e-05,
+      "loss": 7.5828,
+      "loss/crossentropy": 1.962416633963585,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18989747390151024,
+      "step": 84
+    },
+    {
+      "epoch": 0.014166666666666666,
+      "grad_norm": 41.0,
+      "grad_norm_var": 9.6587890625,
+      "learning_rate": 9.995164673390625e-05,
+      "loss": 7.5712,
+      "loss/crossentropy": 2.5914260745048523,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20670515671372414,
+      "step": 85
+    },
+    {
+      "epoch": 0.014333333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 6.2931640625,
+      "learning_rate": 9.995048880404938e-05,
+      "loss": 7.3591,
+      "loss/crossentropy": 2.336932420730591,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18983342498540878,
+      "step": 86
+    },
+    {
+      "epoch": 0.0145,
+      "grad_norm": 35.25,
+      "grad_norm_var": 6.236393229166667,
+      "learning_rate": 9.994931717998272e-05,
+      "loss": 7.3482,
+      "loss/crossentropy": 1.9553337693214417,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19011887535452843,
+      "step": 87
+    },
+    {
+      "epoch": 0.014666666666666666,
+      "grad_norm": 37.5,
+      "grad_norm_var": 6.673893229166667,
+      "learning_rate": 9.994813186202747e-05,
+      "loss": 7.478,
+      "loss/crossentropy": 2.452618956565857,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21071349829435349,
+      "step": 88
+    },
+    {
+      "epoch": 0.014833333333333334,
+      "grad_norm": 34.25,
+      "grad_norm_var": 6.07890625,
+      "learning_rate": 9.994693285050857e-05,
+      "loss": 7.6646,
+      "loss/crossentropy": 2.002704292535782,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19795797765254974,
+      "step": 89
+    },
+    {
+      "epoch": 0.015,
+      "grad_norm": 33.5,
+      "grad_norm_var": 4.83515625,
+      "learning_rate": 9.994572014575476e-05,
+      "loss": 7.6191,
+      "loss/crossentropy": 2.026648461818695,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2158929705619812,
+      "step": 90
+    },
+    {
+      "epoch": 0.015166666666666667,
+      "grad_norm": 39.5,
+      "grad_norm_var": 5.765625,
+      "learning_rate": 9.994449374809851e-05,
+      "loss": 7.6762,
+      "loss/crossentropy": 1.731523722410202,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21968171373009682,
+      "step": 91
+    },
+    {
+      "epoch": 0.015333333333333332,
+      "grad_norm": 35.25,
+      "grad_norm_var": 5.195572916666666,
+      "learning_rate": 9.994325365787602e-05,
+      "loss": 7.5381,
+      "loss/crossentropy": 1.7570662796497345,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1964312382042408,
+      "step": 92
+    },
+    {
+      "epoch": 0.0155,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.968489583333334,
+      "learning_rate": 9.99419998754273e-05,
+      "loss": 7.6536,
+      "loss/crossentropy": 2.0444860458374023,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18911006674170494,
+      "step": 93
+    },
+    {
+      "epoch": 0.015666666666666666,
+      "grad_norm": 36.5,
+      "grad_norm_var": 5.970833333333333,
+      "learning_rate": 9.994073240109606e-05,
+      "loss": 7.289,
+      "loss/crossentropy": 1.5085659623146057,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21490458399057388,
+      "step": 94
+    },
+    {
+      "epoch": 0.015833333333333335,
+      "grad_norm": 33.0,
+      "grad_norm_var": 6.32265625,
+      "learning_rate": 9.993945123522978e-05,
+      "loss": 7.5972,
+      "loss/crossentropy": 2.1204091608524323,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23001987114548683,
+      "step": 95
+    },
+    {
+      "epoch": 0.016,
+      "grad_norm": 32.25,
+      "grad_norm_var": 6.420833333333333,
+      "learning_rate": 9.993815637817974e-05,
+      "loss": 7.5514,
+      "loss/crossentropy": 1.894325166940689,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2360989861190319,
+      "step": 96
+    },
+    {
+      "epoch": 0.016166666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 7.324739583333334,
+      "learning_rate": 9.993684783030088e-05,
+      "loss": 7.5417,
+      "loss/crossentropy": 2.4975918531417847,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18466202914714813,
+      "step": 97
+    },
+    {
+      "epoch": 0.01633333333333333,
+      "grad_norm": 36.25,
+      "grad_norm_var": 7.373958333333333,
+      "learning_rate": 9.993552559195197e-05,
+      "loss": 7.6344,
+      "loss/crossentropy": 1.989092081785202,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1912154033780098,
+      "step": 98
+    },
+    {
+      "epoch": 0.0165,
+      "grad_norm": 32.5,
+      "grad_norm_var": 7.415625,
+      "learning_rate": 9.993418966349552e-05,
+      "loss": 7.5109,
+      "loss/crossentropy": 1.8851015269756317,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22884799912571907,
+      "step": 99
+    },
+    {
+      "epoch": 0.016666666666666666,
+      "grad_norm": 33.75,
+      "grad_norm_var": 7.49765625,
+      "learning_rate": 9.993284004529775e-05,
+      "loss": 7.626,
+      "loss/crossentropy": 2.0821548998355865,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20658370107412338,
+      "step": 100
+    },
+    {
+      "epoch": 0.016833333333333332,
+      "grad_norm": 37.0,
+      "grad_norm_var": 5.205989583333333,
+      "learning_rate": 9.99314767377287e-05,
+      "loss": 7.6134,
+      "loss/crossentropy": 1.843286544084549,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17371157929301262,
+      "step": 101
+    },
+    {
+      "epoch": 0.017,
+      "grad_norm": 32.25,
+      "grad_norm_var": 5.445572916666666,
+      "learning_rate": 9.993009974116211e-05,
+      "loss": 7.4303,
+      "loss/crossentropy": 1.877230018377304,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1641155332326889,
+      "step": 102
+    },
+    {
+      "epoch": 0.017166666666666667,
+      "grad_norm": 34.75,
+      "grad_norm_var": 5.412239583333333,
+      "learning_rate": 9.992870905597548e-05,
+      "loss": 7.6667,
+      "loss/crossentropy": 1.8796460777521133,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2002221643924713,
+      "step": 103
+    },
+    {
+      "epoch": 0.017333333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 4.9125,
+      "learning_rate": 9.992730468255011e-05,
+      "loss": 7.4821,
+      "loss/crossentropy": 2.3127494156360626,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23374580964446068,
+      "step": 104
+    },
+    {
+      "epoch": 0.0175,
+      "grad_norm": 32.75,
+      "grad_norm_var": 5.040625,
+      "learning_rate": 9.9925886621271e-05,
+      "loss": 7.5845,
+      "loss/crossentropy": 2.3329984545707703,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18831797316670418,
+      "step": 105
+    },
+    {
+      "epoch": 0.017666666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 5.01640625,
+      "learning_rate": 9.992445487252691e-05,
+      "loss": 7.7497,
+      "loss/crossentropy": 1.9553033113479614,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20208366215229034,
+      "step": 106
+    },
+    {
+      "epoch": 0.017833333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.120833333333333,
+      "learning_rate": 9.992300943671036e-05,
+      "loss": 7.747,
+      "loss/crossentropy": 2.0242276191711426,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20480553060770035,
+      "step": 107
+    },
+    {
+      "epoch": 0.018,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.990625,
+      "learning_rate": 9.992155031421764e-05,
+      "loss": 7.6461,
+      "loss/crossentropy": 2.40230530500412,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20634019374847412,
+      "step": 108
+    },
+    {
+      "epoch": 0.018166666666666668,
+      "grad_norm": 35.75,
+      "grad_norm_var": 3.10390625,
+      "learning_rate": 9.992007750544876e-05,
+      "loss": 7.7688,
+      "loss/crossentropy": 2.1990231573581696,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2241741679608822,
+      "step": 109
+    },
+    {
+      "epoch": 0.018333333333333333,
+      "grad_norm": 39.75,
+      "grad_norm_var": 4.948958333333334,
+      "learning_rate": 9.991859101080751e-05,
+      "loss": 7.5167,
+      "loss/crossentropy": 2.2589714527130127,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19577842950820923,
+      "step": 110
+    },
+    {
+      "epoch": 0.0185,
+      "grad_norm": 37.5,
+      "grad_norm_var": 5.633333333333334,
+      "learning_rate": 9.991709083070143e-05,
+      "loss": 7.633,
+      "loss/crossentropy": 1.8160150349140167,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19760573282837868,
+      "step": 111
+    },
+    {
+      "epoch": 0.018666666666666668,
+      "grad_norm": 35.5,
+      "grad_norm_var": 5.426822916666667,
+      "learning_rate": 9.991557696554177e-05,
+      "loss": 7.5491,
+      "loss/crossentropy": 1.7978153824806213,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20516641810536385,
+      "step": 112
+    },
+    {
+      "epoch": 0.018833333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.929166666666666,
+      "learning_rate": 9.991404941574361e-05,
+      "loss": 7.5255,
+      "loss/crossentropy": 1.8384543657302856,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21306686475872993,
+      "step": 113
+    },
+    {
+      "epoch": 0.019,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.739322916666667,
+      "learning_rate": 9.99125081817257e-05,
+      "loss": 7.632,
+      "loss/crossentropy": 2.0940029323101044,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1923641823232174,
+      "step": 114
+    },
+    {
+      "epoch": 0.019166666666666665,
+      "grad_norm": 34.75,
+      "grad_norm_var": 4.479166666666667,
+      "learning_rate": 9.99109532639106e-05,
+      "loss": 7.6103,
+      "loss/crossentropy": 1.9544528722763062,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20644604787230492,
+      "step": 115
+    },
+    {
+      "epoch": 0.019333333333333334,
+      "grad_norm": 36.75,
+      "grad_norm_var": 4.716666666666667,
+      "learning_rate": 9.990938466272459e-05,
+      "loss": 7.9127,
+      "loss/crossentropy": 1.4137312471866608,
+      "loss/hidden": 4.03125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2494000941514969,
+      "step": 116
+    },
+    {
+      "epoch": 0.0195,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.379166666666666,
+      "learning_rate": 9.990780237859769e-05,
+      "loss": 7.438,
+      "loss/crossentropy": 2.170451670885086,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21506188064813614,
+      "step": 117
+    },
+    {
+      "epoch": 0.019666666666666666,
+      "grad_norm": 35.0,
+      "grad_norm_var": 4.00390625,
+      "learning_rate": 9.990620641196374e-05,
+      "loss": 7.6142,
+      "loss/crossentropy": 1.825675517320633,
+      "loss/hidden": 3.6484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22746910154819489,
+      "step": 118
+    },
+    {
+      "epoch": 0.019833333333333335,
+      "grad_norm": 35.25,
+      "grad_norm_var": 4.020572916666667,
+      "learning_rate": 9.990459676326024e-05,
+      "loss": 7.5651,
+      "loss/crossentropy": 1.8035025298595428,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19320307672023773,
+      "step": 119
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 34.5,
+      "grad_norm_var": 3.7416666666666667,
+      "learning_rate": 9.990297343292851e-05,
+      "loss": 7.5561,
+      "loss/crossentropy": 1.9253360033035278,
+      "loss/hidden": 3.80078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2591675706207752,
+      "step": 120
+    },
+    {
+      "epoch": 0.020166666666666666,
+      "grad_norm": 31.75,
+      "grad_norm_var": 4.0875,
+      "learning_rate": 9.990133642141359e-05,
+      "loss": 7.6589,
+      "loss/crossentropy": 1.8767026662826538,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21068845689296722,
+      "step": 121
+    },
+    {
+      "epoch": 0.02033333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 5.3875,
+      "learning_rate": 9.989968572916426e-05,
+      "loss": 7.3897,
+      "loss/crossentropy": 2.0437765419483185,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18460320681333542,
+      "step": 122
+    },
+    {
+      "epoch": 0.0205,
+      "grad_norm": 34.0,
+      "grad_norm_var": 5.039322916666666,
+      "learning_rate": 9.989802135663308e-05,
+      "loss": 7.7519,
+      "loss/crossentropy": 2.150317758321762,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17747429013252258,
+      "step": 123
+    },
+    {
+      "epoch": 0.020666666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.926822916666667,
+      "learning_rate": 9.989634330427636e-05,
+      "loss": 7.5183,
+      "loss/crossentropy": 2.179547756910324,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19693326577544212,
+      "step": 124
+    },
+    {
+      "epoch": 0.020833333333333332,
+      "grad_norm": 38.0,
+      "grad_norm_var": 5.557291666666667,
+      "learning_rate": 9.989465157255412e-05,
+      "loss": 7.6732,
+      "loss/crossentropy": 1.4548255801200867,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21135981380939484,
+      "step": 125
+    },
+    {
+      "epoch": 0.021,
+      "grad_norm": 32.75,
+      "grad_norm_var": 4.040625,
+      "learning_rate": 9.989294616193017e-05,
+      "loss": 7.5373,
+      "loss/crossentropy": 1.7414255142211914,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19686686992645264,
+      "step": 126
+    },
+    {
+      "epoch": 0.021166666666666667,
+      "grad_norm": 34.5,
+      "grad_norm_var": 3.365625,
+      "learning_rate": 9.989122707287208e-05,
+      "loss": 7.429,
+      "loss/crossentropy": 1.8234748244285583,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21972696483135223,
+      "step": 127
+    },
+    {
+      "epoch": 0.021333333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.470572916666667,
+      "learning_rate": 9.988949430585111e-05,
+      "loss": 7.3455,
+      "loss/crossentropy": 2.2222295701503754,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.198114313185215,
+      "step": 128
+    },
+    {
+      "epoch": 0.0215,
+      "grad_norm": 34.75,
+      "grad_norm_var": 3.4580729166666666,
+      "learning_rate": 9.988774786134234e-05,
+      "loss": 7.241,
+      "loss/crossentropy": 1.9216758608818054,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19330840185284615,
+      "step": 129
+    },
+    {
+      "epoch": 0.021666666666666667,
+      "grad_norm": 34.75,
+      "grad_norm_var": 3.482291666666667,
+      "learning_rate": 9.988598773982454e-05,
+      "loss": 7.4434,
+      "loss/crossentropy": 2.1256333589553833,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18051563948392868,
+      "step": 130
+    },
+    {
+      "epoch": 0.021833333333333333,
+      "grad_norm": 35.75,
+      "grad_norm_var": 3.6239583333333334,
+      "learning_rate": 9.988421394178027e-05,
+      "loss": 7.4113,
+      "loss/crossentropy": 1.7447483539581299,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17246215045452118,
+      "step": 131
+    },
+    {
+      "epoch": 0.022,
+      "grad_norm": 4764729344.0,
+      "grad_norm_var": 1.4189153373185377e+18,
+      "learning_rate": 9.988242646769584e-05,
+      "loss": 8.3899,
+      "loss/crossentropy": 2.1333084404468536,
+      "loss/hidden": 3.12890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16892877966165543,
+      "step": 132
+    },
+    {
+      "epoch": 0.022166666666666668,
+      "grad_norm": 51.5,
+      "grad_norm_var": 1.4189153366236813e+18,
+      "learning_rate": 9.988062531806126e-05,
+      "loss": 7.651,
+      "loss/crossentropy": 2.1121655106544495,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21801284700632095,
+      "step": 133
+    },
+    {
+      "epoch": 0.022333333333333334,
+      "grad_norm": 36.0,
+      "grad_norm_var": 1.4189153365839752e+18,
+      "learning_rate": 9.987881049337037e-05,
+      "loss": 7.6569,
+      "loss/crossentropy": 2.126715511083603,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17747924476861954,
+      "step": 134
+    },
+    {
+      "epoch": 0.0225,
+      "grad_norm": 36.5,
+      "grad_norm_var": 1.4189153365343427e+18,
+      "learning_rate": 9.98769819941207e-05,
+      "loss": 7.5295,
+      "loss/crossentropy": 1.5165286213159561,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18346380814909935,
+      "step": 135
+    },
+    {
+      "epoch": 0.02266666666666667,
+      "grad_norm": 36.0,
+      "grad_norm_var": 1.4189153364747835e+18,
+      "learning_rate": 9.987513982081351e-05,
+      "loss": 7.6529,
+      "loss/crossentropy": 2.180388480424881,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1846754401922226,
+      "step": 136
+    },
+    {
+      "epoch": 0.022833333333333334,
+      "grad_norm": 38.75,
+      "grad_norm_var": 1.418915336196841e+18,
+      "learning_rate": 9.987328397395387e-05,
+      "loss": 7.5096,
+      "loss/crossentropy": 1.550091952085495,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21606271341443062,
+      "step": 137
+    },
+    {
+      "epoch": 0.023,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.4189153360876493e+18,
+      "learning_rate": 9.98714144540506e-05,
+      "loss": 7.3931,
+      "loss/crossentropy": 1.9342282861471176,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17885055765509605,
+      "step": 138
+    },
+    {
+      "epoch": 0.023166666666666665,
+      "grad_norm": 34.0,
+      "grad_norm_var": 1.4189153360876493e+18,
+      "learning_rate": 9.986953126161619e-05,
+      "loss": 7.7534,
+      "loss/crossentropy": 2.200897455215454,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18954262882471085,
+      "step": 139
+    },
+    {
+      "epoch": 0.023333333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.418915336097576e+18,
+      "learning_rate": 9.986763439716696e-05,
+      "loss": 7.4196,
+      "loss/crossentropy": 1.475202739238739,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23384485766291618,
+      "step": 140
+    },
+    {
+      "epoch": 0.0235,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.4189153362961062e+18,
+      "learning_rate": 9.986572386122291e-05,
+      "loss": 7.4049,
+      "loss/crossentropy": 2.3447420597076416,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17187193408608437,
+      "step": 141
+    },
+    {
+      "epoch": 0.023666666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.4189153362861796e+18,
+      "learning_rate": 9.986379965430786e-05,
+      "loss": 7.4261,
+      "loss/crossentropy": 1.6609918773174286,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21596155688166618,
+      "step": 142
+    },
+    {
+      "epoch": 0.023833333333333335,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.4189153363358124e+18,
+      "learning_rate": 9.986186177694933e-05,
+      "loss": 7.4133,
+      "loss/crossentropy": 1.975971281528473,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23497482761740685,
+      "step": 143
+    },
+    {
+      "epoch": 0.024,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.4189153363705551e+18,
+      "learning_rate": 9.98599102296786e-05,
+      "loss": 7.3945,
+      "loss/crossentropy": 2.0418002009391785,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16329874098300934,
+      "step": 144
+    },
+    {
+      "epoch": 0.024166666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.4189153364400407e+18,
+      "learning_rate": 9.98579450130307e-05,
+      "loss": 7.3843,
+      "loss/crossentropy": 2.586931586265564,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1923389546573162,
+      "step": 145
+    },
+    {
+      "epoch": 0.024333333333333332,
+      "grad_norm": 34.5,
+      "grad_norm_var": 1.4189153364499674e+18,
+      "learning_rate": 9.985596612754439e-05,
+      "loss": 7.4617,
+      "loss/crossentropy": 1.9469241201877594,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20261790975928307,
+      "step": 146
+    },
+    {
+      "epoch": 0.0245,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.4189153366286446e+18,
+      "learning_rate": 9.985397357376222e-05,
+      "loss": 7.3902,
+      "loss/crossentropy": 1.696748986840248,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19723523408174515,
+      "step": 147
+    },
+    {
+      "epoch": 0.024666666666666667,
+      "grad_norm": 31.25,
+      "grad_norm_var": 23.741080729166665,
+      "learning_rate": 9.985196735223045e-05,
+      "loss": 7.4262,
+      "loss/crossentropy": 2.287366271018982,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26205524802207947,
+      "step": 148
+    },
+    {
+      "epoch": 0.024833333333333332,
+      "grad_norm": 31.75,
+      "grad_norm_var": 4.567122395833334,
+      "learning_rate": 9.98499474634991e-05,
+      "loss": 7.4602,
+      "loss/crossentropy": 1.4745359420776367,
+      "loss/hidden": 3.83984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2730640843510628,
+      "step": 149
+    },
+    {
+      "epoch": 0.025,
+      "grad_norm": 35.25,
+      "grad_norm_var": 4.374934895833333,
+      "learning_rate": 9.98479139081219e-05,
+      "loss": 7.3909,
+      "loss/crossentropy": 1.870820164680481,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18368100002408028,
+      "step": 150
+    },
+    {
+      "epoch": 0.025166666666666667,
+      "grad_norm": 38.75,
+      "grad_norm_var": 5.537434895833333,
+      "learning_rate": 9.98458666866564e-05,
+      "loss": 7.4975,
+      "loss/crossentropy": 2.1191659569740295,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16690556704998016,
+      "step": 151
+    },
+    {
+      "epoch": 0.025333333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 5.220247395833334,
+      "learning_rate": 9.984380579966385e-05,
+      "loss": 7.7115,
+      "loss/crossentropy": 2.406974196434021,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17666110768914223,
+      "step": 152
+    },
+    {
+      "epoch": 0.0255,
+      "grad_norm": 43.25,
+      "grad_norm_var": 9.509309895833333,
+      "learning_rate": 9.984173124770923e-05,
+      "loss": 7.4866,
+      "loss/crossentropy": 1.3131984770298004,
+      "loss/hidden": 3.80078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21505622193217278,
+      "step": 153
+    },
+    {
+      "epoch": 0.025666666666666667,
+      "grad_norm": 62.25,
+      "grad_norm_var": 59.11243489583333,
+      "learning_rate": 9.983964303136133e-05,
+      "loss": 7.5887,
+      "loss/crossentropy": 2.2332249581813812,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1927255503833294,
+      "step": 154
+    },
+    {
+      "epoch": 0.025833333333333333,
+      "grad_norm": 36.0,
+      "grad_norm_var": 58.877018229166666,
+      "learning_rate": 9.983754115119261e-05,
+      "loss": 7.3739,
+      "loss/crossentropy": 2.257663667201996,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19335689023137093,
+      "step": 155
+    },
+    {
+      "epoch": 0.026,
+      "grad_norm": 35.25,
+      "grad_norm_var": 58.30983072916667,
+      "learning_rate": 9.983542560777935e-05,
+      "loss": 7.366,
+      "loss/crossentropy": 2.3022145330905914,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19260798767209053,
+      "step": 156
+    },
+    {
+      "epoch": 0.026166666666666668,
+      "grad_norm": 32.75,
+      "grad_norm_var": 58.4166015625,
+      "learning_rate": 9.983329640170149e-05,
+      "loss": 7.2797,
+      "loss/crossentropy": 2.1898127794265747,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17687095701694489,
+      "step": 157
+    },
+    {
+      "epoch": 0.026333333333333334,
+      "grad_norm": 35.0,
+      "grad_norm_var": 57.8478515625,
+      "learning_rate": 9.983115353354281e-05,
+      "loss": 7.2904,
+      "loss/crossentropy": 1.9066053628921509,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18648065626621246,
+      "step": 158
+    },
+    {
+      "epoch": 0.0265,
+      "grad_norm": 32.5,
+      "grad_norm_var": 58.1775390625,
+      "learning_rate": 9.982899700389076e-05,
+      "loss": 7.3832,
+      "loss/crossentropy": 2.1338255405426025,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19225289300084114,
+      "step": 159
+    },
+    {
+      "epoch": 0.02666666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 57.10729166666667,
+      "learning_rate": 9.982682681333658e-05,
+      "loss": 7.545,
+      "loss/crossentropy": 1.81405907869339,
+      "loss/hidden": 3.73046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26101674512028694,
+      "step": 160
+    },
+    {
+      "epoch": 0.026833333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 58.23229166666667,
+      "learning_rate": 9.982464296247522e-05,
+      "loss": 7.3965,
+      "loss/crossentropy": 1.7653635740280151,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16505191847682,
+      "step": 161
+    },
+    {
+      "epoch": 0.027,
+      "grad_norm": 33.75,
+      "grad_norm_var": 58.43307291666667,
+      "learning_rate": 9.982244545190542e-05,
+      "loss": 7.6282,
+      "loss/crossentropy": 2.5411963760852814,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17197951674461365,
+      "step": 162
+    },
+    {
+      "epoch": 0.027166666666666665,
+      "grad_norm": 34.5,
+      "grad_norm_var": 56.9875,
+      "learning_rate": 9.982023428222962e-05,
+      "loss": 7.5649,
+      "loss/crossentropy": 2.2835691571235657,
+      "loss/hidden": 3.16015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.170682180672884,
+      "step": 163
+    },
+    {
+      "epoch": 0.027333333333333334,
+      "grad_norm": 34.5,
+      "grad_norm_var": 55.45390625,
+      "learning_rate": 9.981800945405403e-05,
+      "loss": 7.3372,
+      "loss/crossentropy": 1.8672804236412048,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18279783055186272,
+      "step": 164
+    },
+    {
+      "epoch": 0.0275,
+      "grad_norm": 32.5,
+      "grad_norm_var": 55.0125,
+      "learning_rate": 9.981577096798863e-05,
+      "loss": 7.4292,
+      "loss/crossentropy": 2.196858048439026,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20737188681960106,
+      "step": 165
+    },
+    {
+      "epoch": 0.027666666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 55.72265625,
+      "learning_rate": 9.981351882464706e-05,
+      "loss": 7.3918,
+      "loss/crossentropy": 1.4880334436893463,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18119842186570168,
+      "step": 166
+    },
+    {
+      "epoch": 0.027833333333333335,
+      "grad_norm": 33.5,
+      "grad_norm_var": 55.815625,
+      "learning_rate": 9.98112530246468e-05,
+      "loss": 7.6041,
+      "loss/crossentropy": 1.750911384820938,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22705931961536407,
+      "step": 167
+    },
+    {
+      "epoch": 0.028,
+      "grad_norm": 32.25,
+      "grad_norm_var": 56.557291666666664,
+      "learning_rate": 9.980897356860901e-05,
+      "loss": 7.3669,
+      "loss/crossentropy": 2.3639559745788574,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18918512016534805,
+      "step": 168
+    },
+    {
+      "epoch": 0.028166666666666666,
+      "grad_norm": 34.5,
+      "grad_norm_var": 52.84765625,
+      "learning_rate": 9.980668045715864e-05,
+      "loss": 7.1801,
+      "loss/crossentropy": 1.6124536395072937,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18172302469611168,
+      "step": 169
+    },
+    {
+      "epoch": 0.028333333333333332,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.78515625,
+      "learning_rate": 9.980437369092431e-05,
+      "loss": 7.416,
+      "loss/crossentropy": 1.8924128711223602,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21137376129627228,
+      "step": 170
+    },
+    {
+      "epoch": 0.0285,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.7504557291666667,
+      "learning_rate": 9.980205327053848e-05,
+      "loss": 7.4528,
+      "loss/crossentropy": 2.1068821847438812,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18312924355268478,
+      "step": 171
+    },
+    {
+      "epoch": 0.028666666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.5035807291666667,
+      "learning_rate": 9.97997191966373e-05,
+      "loss": 7.4162,
+      "loss/crossentropy": 2.0754112005233765,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22328588366508484,
+      "step": 172
+    },
+    {
+      "epoch": 0.028833333333333332,
+      "grad_norm": 35.0,
+      "grad_norm_var": 1.7332682291666666,
+      "learning_rate": 9.979737146986064e-05,
+      "loss": 7.5125,
+      "loss/crossentropy": 2.3532181680202484,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22559839859604836,
+      "step": 173
+    },
+    {
+      "epoch": 0.029,
+      "grad_norm": 35.25,
+      "grad_norm_var": 1.7978515625,
+      "learning_rate": 9.979501009085219e-05,
+      "loss": 7.4562,
+      "loss/crossentropy": 1.6273760348558426,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21030885353684425,
+      "step": 174
+    },
+    {
+      "epoch": 0.029166666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.7785807291666667,
+      "learning_rate": 9.979263506025929e-05,
+      "loss": 7.3699,
+      "loss/crossentropy": 2.250314712524414,
+      "loss/hidden": 3.12109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18598228693008423,
+      "step": 175
+    },
+    {
+      "epoch": 0.029333333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.8280598958333334,
+      "learning_rate": 9.97902463787331e-05,
+      "loss": 7.6712,
+      "loss/crossentropy": 2.358814984560013,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2859685495495796,
+      "step": 176
+    },
+    {
+      "epoch": 0.0295,
+      "grad_norm": 36.25,
+      "grad_norm_var": 2.057747395833333,
+      "learning_rate": 9.978784404692847e-05,
+      "loss": 7.6947,
+      "loss/crossentropy": 2.178271234035492,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26804234459996223,
+      "step": 177
+    },
+    {
+      "epoch": 0.029666666666666668,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.052018229166667,
+      "learning_rate": 9.978542806550402e-05,
+      "loss": 7.4962,
+      "loss/crossentropy": 2.2716058492660522,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2189769595861435,
+      "step": 178
+    },
+    {
+      "epoch": 0.029833333333333333,
+      "grad_norm": 34.75,
+      "grad_norm_var": 2.091080729166667,
+      "learning_rate": 9.97829984351221e-05,
+      "loss": 7.4731,
+      "loss/crossentropy": 1.6302326619625092,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2534990794956684,
+      "step": 179
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.091080729166667,
+      "learning_rate": 9.978055515644882e-05,
+      "loss": 7.4943,
+      "loss/crossentropy": 2.0229054391384125,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1985248290002346,
+      "step": 180
+    },
+    {
+      "epoch": 0.030166666666666668,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.551041666666667,
+      "learning_rate": 9.977809823015401e-05,
+      "loss": 7.4271,
+      "loss/crossentropy": 2.0079747140407562,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20629949867725372,
+      "step": 181
+    },
+    {
+      "epoch": 0.030333333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.551041666666667,
+      "learning_rate": 9.977562765691124e-05,
+      "loss": 7.6382,
+      "loss/crossentropy": 1.7329024523496628,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19635221734642982,
+      "step": 182
+    },
+    {
+      "epoch": 0.0305,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.634375,
+      "learning_rate": 9.977314343739786e-05,
+      "loss": 7.3619,
+      "loss/crossentropy": 2.4734097719192505,
+      "loss/hidden": 3.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1726227030158043,
+      "step": 183
+    },
+    {
+      "epoch": 0.030666666666666665,
+      "grad_norm": 42.25,
+      "grad_norm_var": 7.342708333333333,
+      "learning_rate": 9.977064557229492e-05,
+      "loss": 7.2131,
+      "loss/crossentropy": 2.2143907845020294,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1918102391064167,
+      "step": 184
+    },
+    {
+      "epoch": 0.030833333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 7.643489583333333,
+      "learning_rate": 9.97681340622872e-05,
+      "loss": 7.2733,
+      "loss/crossentropy": 2.179398685693741,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2045028991997242,
+      "step": 185
+    },
+    {
+      "epoch": 0.031,
+      "grad_norm": 31.75,
+      "grad_norm_var": 7.76640625,
+      "learning_rate": 9.976560890806328e-05,
+      "loss": 7.3599,
+      "loss/crossentropy": 1.8209953606128693,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1990818865597248,
+      "step": 186
+    },
+    {
+      "epoch": 0.031166666666666665,
+      "grad_norm": 35.75,
+      "grad_norm_var": 7.3322265625,
+      "learning_rate": 9.976307011031542e-05,
+      "loss": 7.4989,
+      "loss/crossentropy": 1.6905051469802856,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2631871812045574,
+      "step": 187
+    },
+    {
+      "epoch": 0.03133333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 7.273372395833333,
+      "learning_rate": 9.976051766973966e-05,
+      "loss": 7.5566,
+      "loss/crossentropy": 1.8864645957946777,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16269812360405922,
+      "step": 188
+    },
+    {
+      "epoch": 0.0315,
+      "grad_norm": 33.25,
+      "grad_norm_var": 7.266080729166666,
+      "learning_rate": 9.975795158703576e-05,
+      "loss": 7.4747,
+      "loss/crossentropy": 1.8467691838741302,
+      "loss/hidden": 3.92578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24105952680110931,
+      "step": 189
+    },
+    {
+      "epoch": 0.03166666666666667,
+      "grad_norm": 34.75,
+      "grad_norm_var": 7.2009765625,
+      "learning_rate": 9.975537186290724e-05,
+      "loss": 7.5958,
+      "loss/crossentropy": 2.065937250852585,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19649240002036095,
+      "step": 190
+    },
+    {
+      "epoch": 0.03183333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 7.162955729166667,
+      "learning_rate": 9.975277849806133e-05,
+      "loss": 7.5776,
+      "loss/crossentropy": 1.760479062795639,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2253318689763546,
+      "step": 191
+    },
+    {
+      "epoch": 0.032,
+      "grad_norm": 32.5,
+      "grad_norm_var": 7.107747395833333,
+      "learning_rate": 9.9750171493209e-05,
+      "loss": 7.7319,
+      "loss/crossentropy": 2.4354325234889984,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19807074591517448,
+      "step": 192
+    },
+    {
+      "epoch": 0.03216666666666667,
+      "grad_norm": 34.5,
+      "grad_norm_var": 6.783268229166667,
+      "learning_rate": 9.974755084906502e-05,
+      "loss": 7.5737,
+      "loss/crossentropy": 1.8597578406333923,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2125672809779644,
+      "step": 193
+    },
+    {
+      "epoch": 0.03233333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 6.783268229166667,
+      "learning_rate": 9.974491656634782e-05,
+      "loss": 7.4921,
+      "loss/crossentropy": 2.4780385494232178,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19840094447135925,
+      "step": 194
+    },
+    {
+      "epoch": 0.0325,
+      "grad_norm": 31.625,
+      "grad_norm_var": 7.051822916666667,
+      "learning_rate": 9.974226864577961e-05,
+      "loss": 7.2698,
+      "loss/crossentropy": 1.88395856320858,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18918826058506966,
+      "step": 195
+    },
+    {
+      "epoch": 0.03266666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 7.138541666666667,
+      "learning_rate": 9.973960708808633e-05,
+      "loss": 7.5145,
+      "loss/crossentropy": 2.3066524267196655,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17494326829910278,
+      "step": 196
+    },
+    {
+      "epoch": 0.03283333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 6.524934895833334,
+      "learning_rate": 9.973693189399766e-05,
+      "loss": 7.453,
+      "loss/crossentropy": 2.067493826150894,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19816205650568008,
+      "step": 197
+    },
+    {
+      "epoch": 0.033,
+      "grad_norm": 34.25,
+      "grad_norm_var": 6.485872395833334,
+      "learning_rate": 9.973424306424705e-05,
+      "loss": 7.3912,
+      "loss/crossentropy": 1.5059811472892761,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23100421205163002,
+      "step": 198
+    },
+    {
+      "epoch": 0.033166666666666664,
+      "grad_norm": 33.5,
+      "grad_norm_var": 6.4681640625,
+      "learning_rate": 9.973154059957162e-05,
+      "loss": 7.3631,
+      "loss/crossentropy": 2.0883964896202087,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19126541167497635,
+      "step": 199
+    },
+    {
+      "epoch": 0.03333333333333333,
+      "grad_norm": 35.25,
+      "grad_norm_var": 1.6775390625,
+      "learning_rate": 9.972882450071228e-05,
+      "loss": 7.5897,
+      "loss/crossentropy": 2.4213720560073853,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1968086063861847,
+      "step": 200
+    },
+    {
+      "epoch": 0.0335,
+      "grad_norm": 34.5,
+      "grad_norm_var": 1.5457682291666666,
+      "learning_rate": 9.972609476841367e-05,
+      "loss": 7.4746,
+      "loss/crossentropy": 1.8749464750289917,
+      "loss/hidden": 3.7265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31945592910051346,
+      "step": 201
+    },
+    {
+      "epoch": 0.033666666666666664,
+      "grad_norm": 35.75,
+      "grad_norm_var": 1.5749348958333333,
+      "learning_rate": 9.972335140342415e-05,
+      "loss": 7.711,
+      "loss/crossentropy": 1.717276781797409,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20567739009857178,
+      "step": 202
+    },
+    {
+      "epoch": 0.03383333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.6103515625,
+      "learning_rate": 9.972059440649584e-05,
+      "loss": 7.5691,
+      "loss/crossentropy": 2.07498636841774,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22483434528112411,
+      "step": 203
+    },
+    {
+      "epoch": 0.034,
+      "grad_norm": 35.25,
+      "grad_norm_var": 1.6962890625,
+      "learning_rate": 9.971782377838457e-05,
+      "loss": 7.7671,
+      "loss/crossentropy": 1.6273568868637085,
+      "loss/hidden": 3.82421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20050818100571632,
+      "step": 204
+    },
+    {
+      "epoch": 0.034166666666666665,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.7160807291666667,
+      "learning_rate": 9.971503951984995e-05,
+      "loss": 7.4089,
+      "loss/crossentropy": 2.2226982414722443,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19824209809303284,
+      "step": 205
+    },
+    {
+      "epoch": 0.034333333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.6889973958333333,
+      "learning_rate": 9.971224163165527e-05,
+      "loss": 7.3443,
+      "loss/crossentropy": 1.8570466339588165,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17482981458306313,
+      "step": 206
+    },
+    {
+      "epoch": 0.0345,
+      "grad_norm": 36.25,
+      "grad_norm_var": 2.0952473958333333,
+      "learning_rate": 9.970943011456761e-05,
+      "loss": 7.6851,
+      "loss/crossentropy": 1.996934950351715,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22505202144384384,
+      "step": 207
+    },
+    {
+      "epoch": 0.034666666666666665,
+      "grad_norm": 33.5,
+      "grad_norm_var": 1.9858723958333333,
+      "learning_rate": 9.970660496935776e-05,
+      "loss": 7.2883,
+      "loss/crossentropy": 2.104332596063614,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.170057974755764,
+      "step": 208
+    },
+    {
+      "epoch": 0.034833333333333334,
+      "grad_norm": 35.25,
+      "grad_norm_var": 2.0858723958333334,
+      "learning_rate": 9.970376619680024e-05,
+      "loss": 7.5088,
+      "loss/crossentropy": 2.0839968621730804,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19400381669402122,
+      "step": 209
+    },
+    {
+      "epoch": 0.035,
+      "grad_norm": 39.0,
+      "grad_norm_var": 3.684309895833333,
+      "learning_rate": 9.970091379767331e-05,
+      "loss": 7.4452,
+      "loss/crossentropy": 1.7162790298461914,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19829977676272392,
+      "step": 210
+    },
+    {
+      "epoch": 0.035166666666666666,
+      "grad_norm": 43.25,
+      "grad_norm_var": 8.073958333333334,
+      "learning_rate": 9.9698047772759e-05,
+      "loss": 7.408,
+      "loss/crossentropy": 2.0295865535736084,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2112085185945034,
+      "step": 211
+    },
+    {
+      "epoch": 0.035333333333333335,
+      "grad_norm": 34.25,
+      "grad_norm_var": 7.598958333333333,
+      "learning_rate": 9.969516812284301e-05,
+      "loss": 7.5912,
+      "loss/crossentropy": 2.076532691717148,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20468730852007866,
+      "step": 212
+    },
+    {
+      "epoch": 0.0355,
+      "grad_norm": 33.0,
+      "grad_norm_var": 7.837239583333333,
+      "learning_rate": 9.969227484871484e-05,
+      "loss": 7.5325,
+      "loss/crossentropy": 2.139839291572571,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2352946475148201,
+      "step": 213
+    },
+    {
+      "epoch": 0.035666666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 8.829166666666667,
+      "learning_rate": 9.968936795116768e-05,
+      "loss": 7.4982,
+      "loss/crossentropy": 2.0447988510131836,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18019738048315048,
+      "step": 214
+    },
+    {
+      "epoch": 0.035833333333333335,
+      "grad_norm": 35.75,
+      "grad_norm_var": 8.751822916666667,
+      "learning_rate": 9.968644743099848e-05,
+      "loss": 7.4297,
+      "loss/crossentropy": 2.0359120666980743,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19392414391040802,
+      "step": 215
+    },
+    {
+      "epoch": 0.036,
+      "grad_norm": 31.75,
+      "grad_norm_var": 9.37890625,
+      "learning_rate": 9.968351328900794e-05,
+      "loss": 7.419,
+      "loss/crossentropy": 1.6464682668447495,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19690733775496483,
+      "step": 216
+    },
+    {
+      "epoch": 0.036166666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 10.03515625,
+      "learning_rate": 9.968056552600043e-05,
+      "loss": 7.4091,
+      "loss/crossentropy": 1.7111313343048096,
+      "loss/hidden": 3.85546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21444109827280045,
+      "step": 217
+    },
+    {
+      "epoch": 0.036333333333333336,
+      "grad_norm": 34.0,
+      "grad_norm_var": 9.945833333333333,
+      "learning_rate": 9.967760414278411e-05,
+      "loss": 7.3835,
+      "loss/crossentropy": 2.2186812460422516,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17553316056728363,
+      "step": 218
+    },
+    {
+      "epoch": 0.0365,
+      "grad_norm": 36.75,
+      "grad_norm_var": 9.612239583333333,
+      "learning_rate": 9.967462914017088e-05,
+      "loss": 7.5934,
+      "loss/crossentropy": 1.7609408795833588,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24462126940488815,
+      "step": 219
+    },
+    {
+      "epoch": 0.03666666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 11.070768229166667,
+      "learning_rate": 9.967164051897633e-05,
+      "loss": 7.4723,
+      "loss/crossentropy": 1.9127337038516998,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18390759825706482,
+      "step": 220
+    },
+    {
+      "epoch": 0.036833333333333336,
+      "grad_norm": 36.75,
+      "grad_norm_var": 11.234830729166667,
+      "learning_rate": 9.966863828001982e-05,
+      "loss": 7.4864,
+      "loss/crossentropy": 1.8919442296028137,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19531705975532532,
+      "step": 221
+    },
+    {
+      "epoch": 0.037,
+      "grad_norm": 32.75,
+      "grad_norm_var": 11.234830729166667,
+      "learning_rate": 9.966562242412442e-05,
+      "loss": 7.4549,
+      "loss/crossentropy": 2.555088996887207,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19128287956118584,
+      "step": 222
+    },
+    {
+      "epoch": 0.03716666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 11.207747395833334,
+      "learning_rate": 9.966259295211697e-05,
+      "loss": 7.3277,
+      "loss/crossentropy": 1.6132709085941315,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2020581029355526,
+      "step": 223
+    },
+    {
+      "epoch": 0.037333333333333336,
+      "grad_norm": 33.0,
+      "grad_norm_var": 11.287434895833334,
+      "learning_rate": 9.965954986482799e-05,
+      "loss": 7.5761,
+      "loss/crossentropy": 1.9417942464351654,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20881640166044235,
+      "step": 224
+    },
+    {
+      "epoch": 0.0375,
+      "grad_norm": 32.5,
+      "grad_norm_var": 11.459309895833334,
+      "learning_rate": 9.965649316309178e-05,
+      "loss": 7.4902,
+      "loss/crossentropy": 2.1505030691623688,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1994522586464882,
+      "step": 225
+    },
+    {
+      "epoch": 0.03766666666666667,
+      "grad_norm": 34.75,
+      "grad_norm_var": 9.9009765625,
+      "learning_rate": 9.965342284774632e-05,
+      "loss": 7.3804,
+      "loss/crossentropy": 1.7770483791828156,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18692020699381828,
+      "step": 226
+    },
+    {
+      "epoch": 0.03783333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.8072265625,
+      "learning_rate": 9.965033891963338e-05,
+      "loss": 7.1566,
+      "loss/crossentropy": 2.0759385526180267,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1766762211918831,
+      "step": 227
+    },
+    {
+      "epoch": 0.038,
+      "grad_norm": 34.75,
+      "grad_norm_var": 3.8806640625,
+      "learning_rate": 9.964724137959843e-05,
+      "loss": 7.6882,
+      "loss/crossentropy": 2.1585110276937485,
+      "loss/hidden": 3.8671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2864663489162922,
+      "step": 228
+    },
+    {
+      "epoch": 0.03816666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.887955729166667,
+      "learning_rate": 9.964413022849068e-05,
+      "loss": 7.3158,
+      "loss/crossentropy": 1.7837422788143158,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17648790404200554,
+      "step": 229
+    },
+    {
+      "epoch": 0.03833333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.459830729166667,
+      "learning_rate": 9.964100546716309e-05,
+      "loss": 7.5637,
+      "loss/crossentropy": 1.929616540670395,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20552103966474533,
+      "step": 230
+    },
+    {
+      "epoch": 0.0385,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.1504557291666666,
+      "learning_rate": 9.963786709647228e-05,
+      "loss": 7.6261,
+      "loss/crossentropy": 1.9666523337364197,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19753441214561462,
+      "step": 231
+    },
+    {
+      "epoch": 0.03866666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.9275390625,
+      "learning_rate": 9.963471511727868e-05,
+      "loss": 7.322,
+      "loss/crossentropy": 1.9215552508831024,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25897734239697456,
+      "step": 232
+    },
+    {
+      "epoch": 0.03883333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 2.6020182291666667,
+      "learning_rate": 9.963154953044645e-05,
+      "loss": 7.3288,
+      "loss/crossentropy": 2.1817697286605835,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.216877032071352,
+      "step": 233
+    },
+    {
+      "epoch": 0.039,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.003059895833333,
+      "learning_rate": 9.962837033684343e-05,
+      "loss": 7.5657,
+      "loss/crossentropy": 2.0929472744464874,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23406041413545609,
+      "step": 234
+    },
+    {
+      "epoch": 0.03916666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.3186848958333335,
+      "learning_rate": 9.96251775373412e-05,
+      "loss": 7.3698,
+      "loss/crossentropy": 2.2823018431663513,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17581793293356895,
+      "step": 235
+    },
+    {
+      "epoch": 0.03933333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 1.4125,
+      "learning_rate": 9.962197113281509e-05,
+      "loss": 7.4896,
+      "loss/crossentropy": 2.5378648042678833,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18384571373462677,
+      "step": 236
+    },
+    {
+      "epoch": 0.0395,
+      "grad_norm": 31.5,
+      "grad_norm_var": 0.99140625,
+      "learning_rate": 9.961875112414416e-05,
+      "loss": 7.6075,
+      "loss/crossentropy": 1.6965321004390717,
+      "loss/hidden": 4.05078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3610421270132065,
+      "step": 237
+    },
+    {
+      "epoch": 0.03966666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.40390625,
+      "learning_rate": 9.961551751221121e-05,
+      "loss": 7.4004,
+      "loss/crossentropy": 1.801620066165924,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2508755698800087,
+      "step": 238
+    },
+    {
+      "epoch": 0.03983333333333333,
+      "grad_norm": 34.5,
+      "grad_norm_var": 1.49765625,
+      "learning_rate": 9.961227029790272e-05,
+      "loss": 7.4518,
+      "loss/crossentropy": 1.9990509450435638,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17706014588475227,
+      "step": 239
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 34.75,
+      "grad_norm_var": 1.6125,
+      "learning_rate": 9.960900948210896e-05,
+      "loss": 7.4336,
+      "loss/crossentropy": 1.602626696228981,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19834685325622559,
+      "step": 240
+    },
+    {
+      "epoch": 0.04016666666666667,
+      "grad_norm": 36.75,
+      "grad_norm_var": 2.21015625,
+      "learning_rate": 9.96057350657239e-05,
+      "loss": 7.6562,
+      "loss/crossentropy": 1.7482210397720337,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2308633252978325,
+      "step": 241
+    },
+    {
+      "epoch": 0.04033333333333333,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.1791666666666667,
+      "learning_rate": 9.960244704964521e-05,
+      "loss": 7.3834,
+      "loss/crossentropy": 2.2258542478084564,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19244441762566566,
+      "step": 242
+    },
+    {
+      "epoch": 0.0405,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.3080729166666667,
+      "learning_rate": 9.959914543477435e-05,
+      "loss": 7.7208,
+      "loss/crossentropy": 2.0081526041030884,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1961013823747635,
+      "step": 243
+    },
+    {
+      "epoch": 0.04066666666666666,
+      "grad_norm": 34.0,
+      "grad_norm_var": 2.2291666666666665,
+      "learning_rate": 9.959583022201647e-05,
+      "loss": 7.4501,
+      "loss/crossentropy": 2.088036686182022,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18969382345676422,
+      "step": 244
+    },
+    {
+      "epoch": 0.04083333333333333,
+      "grad_norm": 35.0,
+      "grad_norm_var": 2.35,
+      "learning_rate": 9.959250141228045e-05,
+      "loss": 7.4074,
+      "loss/crossentropy": 1.8287396430969238,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21593837812542915,
+      "step": 245
+    },
+    {
+      "epoch": 0.041,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.7625,
+      "learning_rate": 9.95891590064789e-05,
+      "loss": 7.4715,
+      "loss/crossentropy": 1.7634437531232834,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18323732912540436,
+      "step": 246
+    },
+    {
+      "epoch": 0.041166666666666664,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.849739583333333,
+      "learning_rate": 9.958580300552815e-05,
+      "loss": 7.3321,
+      "loss/crossentropy": 1.5442231893539429,
+      "loss/hidden": 3.7890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2090124376118183,
+      "step": 247
+    },
+    {
+      "epoch": 0.04133333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.048958333333333,
+      "learning_rate": 9.958243341034827e-05,
+      "loss": 7.4577,
+      "loss/crossentropy": 2.544583737850189,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18567397445440292,
+      "step": 248
+    },
+    {
+      "epoch": 0.0415,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.07890625,
+      "learning_rate": 9.957905022186309e-05,
+      "loss": 7.6299,
+      "loss/crossentropy": 1.9056779742240906,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1823662333190441,
+      "step": 249
+    },
+    {
+      "epoch": 0.041666666666666664,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.9884765625,
+      "learning_rate": 9.957565344100009e-05,
+      "loss": 7.2394,
+      "loss/crossentropy": 1.4222186654806137,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1690158024430275,
+      "step": 250
+    },
+    {
+      "epoch": 0.041833333333333333,
+      "grad_norm": 35.25,
+      "grad_norm_var": 3.2275390625,
+      "learning_rate": 9.957224306869053e-05,
+      "loss": 7.1863,
+      "loss/crossentropy": 2.054836720228195,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2182355523109436,
+      "step": 251
+    },
+    {
+      "epoch": 0.042,
+      "grad_norm": 39.75,
+      "grad_norm_var": 5.7962890625,
+      "learning_rate": 9.956881910586937e-05,
+      "loss": 7.5122,
+      "loss/crossentropy": 1.9823282361030579,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21302693337202072,
+      "step": 252
+    },
+    {
+      "epoch": 0.042166666666666665,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.4916015625,
+      "learning_rate": 9.956538155347534e-05,
+      "loss": 7.3569,
+      "loss/crossentropy": 1.9061083495616913,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18881781026721,
+      "step": 253
+    },
+    {
+      "epoch": 0.042333333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 5.3931640625,
+      "learning_rate": 9.956193041245084e-05,
+      "loss": 7.764,
+      "loss/crossentropy": 2.4186649322509766,
+      "loss/hidden": 3.640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23042699694633484,
+      "step": 254
+    },
+    {
+      "epoch": 0.0425,
+      "grad_norm": 31.5,
+      "grad_norm_var": 5.6900390625,
+      "learning_rate": 9.955846568374201e-05,
+      "loss": 7.4741,
+      "loss/crossentropy": 1.7218818068504333,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20200594142079353,
+      "step": 255
+    },
+    {
+      "epoch": 0.042666666666666665,
+      "grad_norm": 34.25,
+      "grad_norm_var": 5.6322265625,
+      "learning_rate": 9.955498736829875e-05,
+      "loss": 7.3703,
+      "loss/crossentropy": 1.7708523571491241,
+      "loss/hidden": 3.6484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21957089006900787,
+      "step": 256
+    },
+    {
+      "epoch": 0.042833333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 5.192708333333333,
+      "learning_rate": 9.955149546707465e-05,
+      "loss": 7.2327,
+      "loss/crossentropy": 1.588808387517929,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20313889160752296,
+      "step": 257
+    },
+    {
+      "epoch": 0.043,
+      "grad_norm": 31.125,
+      "grad_norm_var": 5.356184895833334,
+      "learning_rate": 9.954798998102702e-05,
+      "loss": 7.694,
+      "loss/crossentropy": 2.2256319522857666,
+      "loss/hidden": 3.7265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2320820689201355,
+      "step": 258
+    },
+    {
+      "epoch": 0.043166666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 5.317122395833334,
+      "learning_rate": 9.954447091111694e-05,
+      "loss": 7.5972,
+      "loss/crossentropy": 2.2975550293922424,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22221146896481514,
+      "step": 259
+    },
+    {
+      "epoch": 0.043333333333333335,
+      "grad_norm": 33.25,
+      "grad_norm_var": 5.267122395833334,
+      "learning_rate": 9.954093825830917e-05,
+      "loss": 7.4645,
+      "loss/crossentropy": 1.7964718043804169,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22343726828694344,
+      "step": 260
+    },
+    {
+      "epoch": 0.0435,
+      "grad_norm": 35.5,
+      "grad_norm_var": 5.409309895833333,
+      "learning_rate": 9.953739202357218e-05,
+      "loss": 7.6442,
+      "loss/crossentropy": 2.3244612216949463,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1996532864868641,
+      "step": 261
+    },
+    {
+      "epoch": 0.043666666666666666,
+      "grad_norm": 34.25,
+      "grad_norm_var": 5.145247395833334,
+      "learning_rate": 9.953383220787824e-05,
+      "loss": 7.3658,
+      "loss/crossentropy": 1.8924684524536133,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1912608966231346,
+      "step": 262
+    },
+    {
+      "epoch": 0.043833333333333335,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.1853515625,
+      "learning_rate": 9.953025881220325e-05,
+      "loss": 7.3769,
+      "loss/crossentropy": 2.1403828859329224,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19196165725588799,
+      "step": 263
+    },
+    {
+      "epoch": 0.044,
+      "grad_norm": 31.625,
+      "grad_norm_var": 5.155989583333334,
+      "learning_rate": 9.952667183752689e-05,
+      "loss": 7.4753,
+      "loss/crossentropy": 2.1855861246585846,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17956650257110596,
+      "step": 264
+    },
+    {
+      "epoch": 0.04416666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.1,
+      "learning_rate": 9.952307128483256e-05,
+      "loss": 7.5188,
+      "loss/crossentropy": 2.1611409187316895,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21880854666233063,
+      "step": 265
+    },
+    {
+      "epoch": 0.044333333333333336,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.912955729166667,
+      "learning_rate": 9.951945715510738e-05,
+      "loss": 7.2769,
+      "loss/crossentropy": 1.5910672396421432,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1987658105790615,
+      "step": 266
+    },
+    {
+      "epoch": 0.0445,
+      "grad_norm": 36.5,
+      "grad_norm_var": 5.327018229166667,
+      "learning_rate": 9.951582944934215e-05,
+      "loss": 7.4072,
+      "loss/crossentropy": 1.8843984603881836,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1851150095462799,
+      "step": 267
+    },
+    {
+      "epoch": 0.04466666666666667,
+      "grad_norm": 37.25,
+      "grad_norm_var": 3.6108723958333333,
+      "learning_rate": 9.951218816853145e-05,
+      "loss": 7.3747,
+      "loss/crossentropy": 2.051399737596512,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17208177596330643,
+      "step": 268
+    },
+    {
+      "epoch": 0.044833333333333336,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.6447265625,
+      "learning_rate": 9.950853331367356e-05,
+      "loss": 7.5431,
+      "loss/crossentropy": 1.8954854309558868,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23021245375275612,
+      "step": 269
+    },
+    {
+      "epoch": 0.045,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.3369140625,
+      "learning_rate": 9.950486488577045e-05,
+      "loss": 7.086,
+      "loss/crossentropy": 2.1651638746261597,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16940529271960258,
+      "step": 270
+    },
+    {
+      "epoch": 0.04516666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.1546223958333335,
+      "learning_rate": 9.950118288582788e-05,
+      "loss": 7.4365,
+      "loss/crossentropy": 2.0743221640586853,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2509123831987381,
+      "step": 271
+    },
+    {
+      "epoch": 0.04533333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.1775390625,
+      "learning_rate": 9.949748731485527e-05,
+      "loss": 7.4719,
+      "loss/crossentropy": 1.6460443139076233,
+      "loss/hidden": 3.734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23979851976037025,
+      "step": 272
+    },
+    {
+      "epoch": 0.0455,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.94765625,
+      "learning_rate": 9.949377817386579e-05,
+      "loss": 7.3823,
+      "loss/crossentropy": 2.06423482298851,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18897506222128868,
+      "step": 273
+    },
+    {
+      "epoch": 0.04566666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.6285807291666665,
+      "learning_rate": 9.949005546387631e-05,
+      "loss": 7.4061,
+      "loss/crossentropy": 1.8803717195987701,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19544275850057602,
+      "step": 274
+    },
+    {
+      "epoch": 0.04583333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.78515625,
+      "learning_rate": 9.948631918590746e-05,
+      "loss": 7.2924,
+      "loss/crossentropy": 2.3207769989967346,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.204855278134346,
+      "step": 275
+    },
+    {
+      "epoch": 0.046,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.973893229166667,
+      "learning_rate": 9.948256934098352e-05,
+      "loss": 7.2179,
+      "loss/crossentropy": 2.212691694498062,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1843850538134575,
+      "step": 276
+    },
+    {
+      "epoch": 0.04616666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.6171223958333334,
+      "learning_rate": 9.947880593013255e-05,
+      "loss": 7.389,
+      "loss/crossentropy": 1.4388393610715866,
+      "loss/hidden": 3.86328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20000701770186424,
+      "step": 277
+    },
+    {
+      "epoch": 0.04633333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.5405598958333333,
+      "learning_rate": 9.947502895438631e-05,
+      "loss": 7.4375,
+      "loss/crossentropy": 1.6164623200893402,
+      "loss/hidden": 3.71875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24112718924880028,
+      "step": 278
+    },
+    {
+      "epoch": 0.0465,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.5580729166666667,
+      "learning_rate": 9.94712384147803e-05,
+      "loss": 7.5675,
+      "loss/crossentropy": 2.080635279417038,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2437608428299427,
+      "step": 279
+    },
+    {
+      "epoch": 0.04666666666666667,
+      "grad_norm": 37.25,
+      "grad_norm_var": 3.5160807291666667,
+      "learning_rate": 9.94674343123537e-05,
+      "loss": 7.4194,
+      "loss/crossentropy": 1.9315711557865143,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20482220873236656,
+      "step": 280
+    },
+    {
+      "epoch": 0.04683333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.5093098958333333,
+      "learning_rate": 9.946361664814943e-05,
+      "loss": 7.3005,
+      "loss/crossentropy": 2.334685444831848,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1906958930194378,
+      "step": 281
+    },
+    {
+      "epoch": 0.047,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.7280598958333333,
+      "learning_rate": 9.945978542321411e-05,
+      "loss": 7.6145,
+      "loss/crossentropy": 2.053435295820236,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18392417207360268,
+      "step": 282
+    },
+    {
+      "epoch": 0.04716666666666667,
+      "grad_norm": 35.5,
+      "grad_norm_var": 3.358268229166667,
+      "learning_rate": 9.945594063859809e-05,
+      "loss": 7.5437,
+      "loss/crossentropy": 2.135278433561325,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18531183525919914,
+      "step": 283
+    },
+    {
+      "epoch": 0.04733333333333333,
+      "grad_norm": 37.25,
+      "grad_norm_var": 3.358268229166667,
+      "learning_rate": 9.945208229535548e-05,
+      "loss": 7.5657,
+      "loss/crossentropy": 2.2949063777923584,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20849039778113365,
+      "step": 284
+    },
+    {
+      "epoch": 0.0475,
+      "grad_norm": 35.0,
+      "grad_norm_var": 3.5171223958333333,
+      "learning_rate": 9.944821039454402e-05,
+      "loss": 7.4337,
+      "loss/crossentropy": 2.347493678331375,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22633015364408493,
+      "step": 285
+    },
+    {
+      "epoch": 0.04766666666666667,
+      "grad_norm": 36.0,
+      "grad_norm_var": 3.8853515625,
+      "learning_rate": 9.944432493722524e-05,
+      "loss": 7.408,
+      "loss/crossentropy": 1.8182571828365326,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1729656383395195,
+      "step": 286
+    },
+    {
+      "epoch": 0.04783333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.8296223958333333,
+      "learning_rate": 9.944042592446434e-05,
+      "loss": 7.518,
+      "loss/crossentropy": 2.2782211005687714,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19678113237023354,
+      "step": 287
+    },
+    {
+      "epoch": 0.048,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.7884765625,
+      "learning_rate": 9.943651335733028e-05,
+      "loss": 7.7055,
+      "loss/crossentropy": 1.8382735848426819,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29597097635269165,
+      "step": 288
+    },
+    {
+      "epoch": 0.04816666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.7462890625,
+      "learning_rate": 9.94325872368957e-05,
+      "loss": 7.5491,
+      "loss/crossentropy": 1.9217680990695953,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1935409978032112,
+      "step": 289
+    },
+    {
+      "epoch": 0.04833333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.694205729166667,
+      "learning_rate": 9.942864756423697e-05,
+      "loss": 7.3539,
+      "loss/crossentropy": 2.2482245564460754,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1774788796901703,
+      "step": 290
+    },
+    {
+      "epoch": 0.0485,
+      "grad_norm": 37.25,
+      "grad_norm_var": 4.172916666666667,
+      "learning_rate": 9.942469434043418e-05,
+      "loss": 7.5497,
+      "loss/crossentropy": 1.7184914350509644,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17235538363456726,
+      "step": 291
+    },
+    {
+      "epoch": 0.048666666666666664,
+      "grad_norm": 29.5,
+      "grad_norm_var": 5.145768229166666,
+      "learning_rate": 9.942072756657112e-05,
+      "loss": 7.4965,
+      "loss/crossentropy": 2.0929549038410187,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1646399423480034,
+      "step": 292
+    },
+    {
+      "epoch": 0.04883333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 5.084309895833333,
+      "learning_rate": 9.941674724373531e-05,
+      "loss": 7.3894,
+      "loss/crossentropy": 1.9612452387809753,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1791134923696518,
+      "step": 293
+    },
+    {
+      "epoch": 0.049,
+      "grad_norm": 31.625,
+      "grad_norm_var": 5.30625,
+      "learning_rate": 9.941275337301796e-05,
+      "loss": 7.2659,
+      "loss/crossentropy": 2.0830692052841187,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20017437636852264,
+      "step": 294
+    },
+    {
+      "epoch": 0.049166666666666664,
+      "grad_norm": 31.5,
+      "grad_norm_var": 5.4181640625,
+      "learning_rate": 9.940874595551404e-05,
+      "loss": 7.3887,
+      "loss/crossentropy": 1.7737980782985687,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18077390268445015,
+      "step": 295
+    },
+    {
+      "epoch": 0.04933333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 4.6822265625,
+      "learning_rate": 9.940472499232217e-05,
+      "loss": 7.533,
+      "loss/crossentropy": 2.0955342948436737,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18926679342985153,
+      "step": 296
+    },
+    {
+      "epoch": 0.0495,
+      "grad_norm": 30.875,
+      "grad_norm_var": 5.159375,
+      "learning_rate": 9.940069048454476e-05,
+      "loss": 7.3838,
+      "loss/crossentropy": 1.9292489290237427,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17382750287652016,
+      "step": 297
+    },
+    {
+      "epoch": 0.049666666666666665,
+      "grad_norm": 35.0,
+      "grad_norm_var": 5.00625,
+      "learning_rate": 9.939664243328788e-05,
+      "loss": 7.5044,
+      "loss/crossentropy": 1.8965256214141846,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21535596996545792,
+      "step": 298
+    },
+    {
+      "epoch": 0.049833333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.926041666666666,
+      "learning_rate": 9.939258083966131e-05,
+      "loss": 7.3787,
+      "loss/crossentropy": 2.2116329669952393,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19921951368451118,
+      "step": 299
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.9677083333333334,
+      "learning_rate": 9.938850570477858e-05,
+      "loss": 7.776,
+      "loss/crossentropy": 1.7931984663009644,
+      "loss/hidden": 4.07421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20052238181233406,
+      "step": 300
+    },
+    {
+      "epoch": 0.050166666666666665,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.984375,
+      "learning_rate": 9.938441702975689e-05,
+      "loss": 7.5502,
+      "loss/crossentropy": 2.151143029332161,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2542959935963154,
+      "step": 301
+    },
+    {
+      "epoch": 0.050333333333333334,
+      "grad_norm": 36.0,
+      "grad_norm_var": 3.984375,
+      "learning_rate": 9.93803148157172e-05,
+      "loss": 7.3519,
+      "loss/crossentropy": 1.8495058715343475,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19523714110255241,
+      "step": 302
+    },
+    {
+      "epoch": 0.0505,
+      "grad_norm": 33.5,
+      "grad_norm_var": 4.00625,
+      "learning_rate": 9.937619906378413e-05,
+      "loss": 7.6,
+      "loss/crossentropy": 1.4866109788417816,
+      "loss/hidden": 3.86328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24062253534793854,
+      "step": 303
+    },
+    {
+      "epoch": 0.050666666666666665,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.99765625,
+      "learning_rate": 9.937206977508604e-05,
+      "loss": 7.532,
+      "loss/crossentropy": 2.360676884651184,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17827368155121803,
+      "step": 304
+    },
+    {
+      "epoch": 0.050833333333333335,
+      "grad_norm": 35.25,
+      "grad_norm_var": 4.318489583333333,
+      "learning_rate": 9.936792695075502e-05,
+      "loss": 7.2543,
+      "loss/crossentropy": 1.8312220871448517,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15719518810510635,
+      "step": 305
+    },
+    {
+      "epoch": 0.051,
+      "grad_norm": 34.75,
+      "grad_norm_var": 4.442708333333333,
+      "learning_rate": 9.936377059192683e-05,
+      "loss": 7.4177,
+      "loss/crossentropy": 1.8039703965187073,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17907331138849258,
+      "step": 306
+    },
+    {
+      "epoch": 0.051166666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.3145182291666666,
+      "learning_rate": 9.935960069974096e-05,
+      "loss": 7.5267,
+      "loss/crossentropy": 2.2493822872638702,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22437208890914917,
+      "step": 307
+    },
+    {
+      "epoch": 0.051333333333333335,
+      "grad_norm": 36.0,
+      "grad_norm_var": 3.0775390625,
+      "learning_rate": 9.935541727534062e-05,
+      "loss": 7.3628,
+      "loss/crossentropy": 1.6787460148334503,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18720288947224617,
+      "step": 308
+    },
+    {
+      "epoch": 0.0515,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.1212890625,
+      "learning_rate": 9.93512203198727e-05,
+      "loss": 7.3632,
+      "loss/crossentropy": 1.9651407599449158,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16669607907533646,
+      "step": 309
+    },
+    {
+      "epoch": 0.051666666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.05625,
+      "learning_rate": 9.934700983448785e-05,
+      "loss": 7.4242,
+      "loss/crossentropy": 1.7720632553100586,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17070115730166435,
+      "step": 310
+    },
+    {
+      "epoch": 0.051833333333333335,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.92890625,
+      "learning_rate": 9.934278582034037e-05,
+      "loss": 7.4942,
+      "loss/crossentropy": 1.9744703769683838,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17627298459410667,
+      "step": 311
+    },
+    {
+      "epoch": 0.052,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.97265625,
+      "learning_rate": 9.93385482785883e-05,
+      "loss": 7.4852,
+      "loss/crossentropy": 1.939420759677887,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20203501358628273,
+      "step": 312
+    },
+    {
+      "epoch": 0.05216666666666667,
+      "grad_norm": 34.75,
+      "grad_norm_var": 2.7405598958333335,
+      "learning_rate": 9.93342972103934e-05,
+      "loss": 7.2799,
+      "loss/crossentropy": 1.8921250402927399,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19524874910712242,
+      "step": 313
+    },
+    {
+      "epoch": 0.052333333333333336,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.724739583333333,
+      "learning_rate": 9.933003261692113e-05,
+      "loss": 7.4116,
+      "loss/crossentropy": 1.8156531155109406,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1783577762544155,
+      "step": 314
+    },
+    {
+      "epoch": 0.0525,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.627083333333333,
+      "learning_rate": 9.932575449934062e-05,
+      "loss": 7.6664,
+      "loss/crossentropy": 1.611064851284027,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18313537165522575,
+      "step": 315
+    },
+    {
+      "epoch": 0.05266666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.627083333333333,
+      "learning_rate": 9.932146285882477e-05,
+      "loss": 7.517,
+      "loss/crossentropy": 2.3886789083480835,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17311230301856995,
+      "step": 316
+    },
+    {
+      "epoch": 0.052833333333333336,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.277083333333333,
+      "learning_rate": 9.931715769655015e-05,
+      "loss": 7.3245,
+      "loss/crossentropy": 2.062618851661682,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16256008669734,
+      "step": 317
+    },
+    {
+      "epoch": 0.053,
+      "grad_norm": 37.25,
+      "grad_norm_var": 2.812239583333333,
+      "learning_rate": 9.931283901369706e-05,
+      "loss": 7.4205,
+      "loss/crossentropy": 2.2248376309871674,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2207106053829193,
+      "step": 318
+    },
+    {
+      "epoch": 0.05316666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.8684895833333335,
+      "learning_rate": 9.930850681144945e-05,
+      "loss": 7.4183,
+      "loss/crossentropy": 1.6167858988046646,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18527067452669144,
+      "step": 319
+    },
+    {
+      "epoch": 0.05333333333333334,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.4389973958333333,
+      "learning_rate": 9.930416109099505e-05,
+      "loss": 7.2778,
+      "loss/crossentropy": 1.5095842480659485,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21223530173301697,
+      "step": 320
+    },
+    {
+      "epoch": 0.0535,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.1858723958333335,
+      "learning_rate": 9.929980185352526e-05,
+      "loss": 7.1886,
+      "loss/crossentropy": 1.7327702194452286,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17669162154197693,
+      "step": 321
+    },
+    {
+      "epoch": 0.05366666666666667,
+      "grad_norm": 37.25,
+      "grad_norm_var": 4.152018229166667,
+      "learning_rate": 9.929542910023517e-05,
+      "loss": 7.5761,
+      "loss/crossentropy": 2.0194932520389557,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1981862224638462,
+      "step": 322
+    },
+    {
+      "epoch": 0.05383333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.030989583333334,
+      "learning_rate": 9.929104283232362e-05,
+      "loss": 7.4121,
+      "loss/crossentropy": 2.0976805686950684,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2352442853152752,
+      "step": 323
+    },
+    {
+      "epoch": 0.054,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.5205729166666666,
+      "learning_rate": 9.928664305099314e-05,
+      "loss": 7.3845,
+      "loss/crossentropy": 1.7154293060302734,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18793410435318947,
+      "step": 324
+    },
+    {
+      "epoch": 0.05416666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.443489583333333,
+      "learning_rate": 9.928222975744991e-05,
+      "loss": 7.4372,
+      "loss/crossentropy": 1.7047147303819656,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18225349113345146,
+      "step": 325
+    },
+    {
+      "epoch": 0.05433333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.35625,
+      "learning_rate": 9.927780295290389e-05,
+      "loss": 7.3018,
+      "loss/crossentropy": 1.9299534857273102,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17289981245994568,
+      "step": 326
+    },
+    {
+      "epoch": 0.0545,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.309375,
+      "learning_rate": 9.927336263856872e-05,
+      "loss": 7.4296,
+      "loss/crossentropy": 1.5677206218242645,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21524176374077797,
+      "step": 327
+    },
+    {
+      "epoch": 0.05466666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.2604166666666665,
+      "learning_rate": 9.926890881566171e-05,
+      "loss": 7.3113,
+      "loss/crossentropy": 1.9055284261703491,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21018758416175842,
+      "step": 328
+    },
+    {
+      "epoch": 0.05483333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.101822916666667,
+      "learning_rate": 9.926444148540393e-05,
+      "loss": 7.3262,
+      "loss/crossentropy": 1.8975631594657898,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1867729350924492,
+      "step": 329
+    },
+    {
+      "epoch": 0.055,
+      "grad_norm": 29.375,
+      "grad_norm_var": 3.872916666666667,
+      "learning_rate": 9.925996064902011e-05,
+      "loss": 7.3047,
+      "loss/crossentropy": 2.149957090616226,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1669253148138523,
+      "step": 330
+    },
+    {
+      "epoch": 0.05516666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.005989583333333,
+      "learning_rate": 9.92554663077387e-05,
+      "loss": 7.3005,
+      "loss/crossentropy": 1.8946216106414795,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21060358732938766,
+      "step": 331
+    },
+    {
+      "epoch": 0.05533333333333333,
+      "grad_norm": 36.5,
+      "grad_norm_var": 4.771875,
+      "learning_rate": 9.925095846279184e-05,
+      "loss": 7.2897,
+      "loss/crossentropy": 1.8477730453014374,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1942521631717682,
+      "step": 332
+    },
+    {
+      "epoch": 0.0555,
+      "grad_norm": 36.25,
+      "grad_norm_var": 5.364322916666667,
+      "learning_rate": 9.924643711541539e-05,
+      "loss": 7.3068,
+      "loss/crossentropy": 2.038343906402588,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20785771310329437,
+      "step": 333
+    },
+    {
+      "epoch": 0.05566666666666667,
+      "grad_norm": 34.5,
+      "grad_norm_var": 4.410416666666666,
+      "learning_rate": 9.92419022668489e-05,
+      "loss": 7.6384,
+      "loss/crossentropy": 2.468900591135025,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20980802923440933,
+      "step": 334
+    },
+    {
+      "epoch": 0.05583333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 4.9806640625,
+      "learning_rate": 9.923735391833564e-05,
+      "loss": 7.3485,
+      "loss/crossentropy": 2.2117743492126465,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18478307873010635,
+      "step": 335
+    },
+    {
+      "epoch": 0.056,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.66015625,
+      "learning_rate": 9.923279207112255e-05,
+      "loss": 7.4143,
+      "loss/crossentropy": 1.931143194437027,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20408664271235466,
+      "step": 336
+    },
+    {
+      "epoch": 0.05616666666666666,
+      "grad_norm": 35.5,
+      "grad_norm_var": 4.947916666666667,
+      "learning_rate": 9.922821672646027e-05,
+      "loss": 7.3851,
+      "loss/crossentropy": 1.9041670262813568,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18027332797646523,
+      "step": 337
+    },
+    {
+      "epoch": 0.05633333333333333,
+      "grad_norm": 34.5,
+      "grad_norm_var": 3.976822916666667,
+      "learning_rate": 9.922362788560319e-05,
+      "loss": 7.4256,
+      "loss/crossentropy": 2.080709308385849,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20545602589845657,
+      "step": 338
+    },
+    {
+      "epoch": 0.0565,
+      "grad_norm": 31.625,
+      "grad_norm_var": 4.1181640625,
+      "learning_rate": 9.921902554980934e-05,
+      "loss": 7.4433,
+      "loss/crossentropy": 2.085605651140213,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21113378182053566,
+      "step": 339
+    },
+    {
+      "epoch": 0.056666666666666664,
+      "grad_norm": 34.5,
+      "grad_norm_var": 4.2244140625,
+      "learning_rate": 9.921440972034049e-05,
+      "loss": 7.3598,
+      "loss/crossentropy": 1.9490495473146439,
+      "loss/hidden": 3.87109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2463916763663292,
+      "step": 340
+    },
+    {
+      "epoch": 0.05683333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.2650390625,
+      "learning_rate": 9.92097803984621e-05,
+      "loss": 7.4158,
+      "loss/crossentropy": 1.8314751535654068,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19313548505306244,
+      "step": 341
+    },
+    {
+      "epoch": 0.057,
+      "grad_norm": 35.75,
+      "grad_norm_var": 4.6634765625,
+      "learning_rate": 9.920513758544332e-05,
+      "loss": 7.2865,
+      "loss/crossentropy": 1.5422937273979187,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1795356124639511,
+      "step": 342
+    },
+    {
+      "epoch": 0.057166666666666664,
+      "grad_norm": 36.5,
+      "grad_norm_var": 5.2259765625,
+      "learning_rate": 9.920048128255699e-05,
+      "loss": 7.3166,
+      "loss/crossentropy": 1.8613337278366089,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19935959205031395,
+      "step": 343
+    },
+    {
+      "epoch": 0.05733333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 5.43515625,
+      "learning_rate": 9.919581149107968e-05,
+      "loss": 7.541,
+      "loss/crossentropy": 2.4002679884433746,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19639062508940697,
+      "step": 344
+    },
+    {
+      "epoch": 0.0575,
+      "grad_norm": 30.125,
+      "grad_norm_var": 6.161393229166666,
+      "learning_rate": 9.919112821229163e-05,
+      "loss": 7.3566,
+      "loss/crossentropy": 2.0356940031051636,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17380831763148308,
+      "step": 345
+    },
+    {
+      "epoch": 0.057666666666666665,
+      "grad_norm": 31.0,
+      "grad_norm_var": 5.461458333333334,
+      "learning_rate": 9.918643144747681e-05,
+      "loss": 7.2969,
+      "loss/crossentropy": 1.829444795846939,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2013947330415249,
+      "step": 346
+    },
+    {
+      "epoch": 0.057833333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 5.399739583333333,
+      "learning_rate": 9.918172119792282e-05,
+      "loss": 7.405,
+      "loss/crossentropy": 2.1244198977947235,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21309788897633553,
+      "step": 347
+    },
+    {
+      "epoch": 0.058,
+      "grad_norm": 36.25,
+      "grad_norm_var": 5.303125,
+      "learning_rate": 9.917699746492104e-05,
+      "loss": 7.6752,
+      "loss/crossentropy": 2.3279683887958527,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2461264804005623,
+      "step": 348
+    },
+    {
+      "epoch": 0.058166666666666665,
+      "grad_norm": 34.5,
+      "grad_norm_var": 4.845572916666667,
+      "learning_rate": 9.917226024976649e-05,
+      "loss": 7.3917,
+      "loss/crossentropy": 1.751452475786209,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22498203814029694,
+      "step": 349
+    },
+    {
+      "epoch": 0.058333333333333334,
+      "grad_norm": 47.75,
+      "grad_norm_var": 17.833333333333332,
+      "learning_rate": 9.91675095537579e-05,
+      "loss": 7.5346,
+      "loss/crossentropy": 1.5791819542646408,
+      "loss/hidden": 3.7109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2855018340051174,
+      "step": 350
+    },
+    {
+      "epoch": 0.0585,
+      "grad_norm": 3892314112.0,
+      "grad_norm_var": 9.468818048893147e+17,
+      "learning_rate": 9.916274537819775e-05,
+      "loss": 8.3129,
+      "loss/crossentropy": 1.8232090175151825,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22548828274011612,
+      "step": 351
+    },
+    {
+      "epoch": 0.058666666666666666,
+      "grad_norm": 41.5,
+      "grad_norm_var": 9.468818045649553e+17,
+      "learning_rate": 9.915796772439207e-05,
+      "loss": 7.5496,
+      "loss/crossentropy": 2.0420787036418915,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.197432741522789,
+      "step": 352
+    },
+    {
+      "epoch": 0.058833333333333335,
+      "grad_norm": 34.25,
+      "grad_norm_var": 9.468818046055002e+17,
+      "learning_rate": 9.915317659365077e-05,
+      "loss": 7.5245,
+      "loss/crossentropy": 1.5124047994613647,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1743464544415474,
+      "step": 353
+    },
+    {
+      "epoch": 0.059,
+      "grad_norm": 33.0,
+      "grad_norm_var": 9.468818046541541e+17,
+      "learning_rate": 9.914837198728733e-05,
+      "loss": 7.604,
+      "loss/crossentropy": 2.2793531715869904,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19110491126775742,
+      "step": 354
+    },
+    {
+      "epoch": 0.059166666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 9.468818046257727e+17,
+      "learning_rate": 9.914355390661896e-05,
+      "loss": 7.3176,
+      "loss/crossentropy": 1.2015042752027512,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19891759008169174,
+      "step": 355
+    },
+    {
+      "epoch": 0.059333333333333335,
+      "grad_norm": 32.0,
+      "grad_norm_var": 9.468818047068625e+17,
+      "learning_rate": 9.913872235296657e-05,
+      "loss": 7.1928,
+      "loss/crossentropy": 1.9171126186847687,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18359432741999626,
+      "step": 356
+    },
+    {
+      "epoch": 0.0595,
+      "grad_norm": 30.5,
+      "grad_norm_var": 9.468818048203884e+17,
+      "learning_rate": 9.913387732765475e-05,
+      "loss": 7.53,
+      "loss/crossentropy": 1.981977254152298,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1935758851468563,
+      "step": 357
+    },
+    {
+      "epoch": 0.059666666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 9.468818049014783e+17,
+      "learning_rate": 9.91290188320118e-05,
+      "loss": 7.2952,
+      "loss/crossentropy": 1.9780019223690033,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17839540541172028,
+      "step": 358
+    },
+    {
+      "epoch": 0.059833333333333336,
+      "grad_norm": 32.75,
+      "grad_norm_var": 9.468818050231131e+17,
+      "learning_rate": 9.91241468673697e-05,
+      "loss": 7.5333,
+      "loss/crossentropy": 2.5614745020866394,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20059003308415413,
+      "step": 359
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2583691264.0,
+      "grad_norm_var": 1.2802935968532966e+18,
+      "learning_rate": 9.911926143506412e-05,
+      "loss": 8.5701,
+      "loss/crossentropy": 1.9112935066223145,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19570891931653023,
+      "step": 360
+    },
+    {
+      "epoch": 0.06016666666666667,
+      "grad_norm": 46.25,
+      "grad_norm_var": 1.2802935959830835e+18,
+      "learning_rate": 9.911436253643445e-05,
+      "loss": 7.2566,
+      "loss/crossentropy": 1.8176748156547546,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17950082197785378,
+      "step": 361
+    },
+    {
+      "epoch": 0.060333333333333336,
+      "grad_norm": 34.75,
+      "grad_norm_var": 1.2802935957807084e+18,
+      "learning_rate": 9.910945017282372e-05,
+      "loss": 7.3139,
+      "loss/crossentropy": 1.928529977798462,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19035305082798004,
+      "step": 362
+    },
+    {
+      "epoch": 0.0605,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.2802935957402335e+18,
+      "learning_rate": 9.91045243455787e-05,
+      "loss": 7.5532,
+      "loss/crossentropy": 1.8307278454303741,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26346417516469955,
+      "step": 363
+    },
+    {
+      "epoch": 0.06066666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.2802935959561e+18,
+      "learning_rate": 9.909958505604984e-05,
+      "loss": 7.0528,
+      "loss/crossentropy": 2.0617458522319794,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.184691421687603,
+      "step": 364
+    },
+    {
+      "epoch": 0.060833333333333336,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.280293596091017e+18,
+      "learning_rate": 9.909463230559127e-05,
+      "loss": 7.6179,
+      "loss/crossentropy": 2.155223786830902,
+      "loss/hidden": 3.71484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24829794466495514,
+      "step": 365
+    },
+    {
+      "epoch": 0.061,
+      "grad_norm": 35.5,
+      "grad_norm_var": 1.280293596752109e+18,
+      "learning_rate": 9.908966609556079e-05,
+      "loss": 7.6505,
+      "loss/crossentropy": 2.0439819991588593,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29273102432489395,
+      "step": 366
+    },
+    {
+      "epoch": 0.06116666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.172162731141148e+17,
+      "learning_rate": 9.908468642731995e-05,
+      "loss": 7.3032,
+      "loss/crossentropy": 2.034709244966507,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2327149398624897,
+      "step": 367
+    },
+    {
+      "epoch": 0.06133333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.172162732917436e+17,
+      "learning_rate": 9.907969330223395e-05,
+      "loss": 7.4659,
+      "loss/crossentropy": 1.8745178282260895,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1996840313076973,
+      "step": 368
+    },
+    {
+      "epoch": 0.0615,
+      "grad_norm": 31.75,
+      "grad_norm_var": 4.172162733455705e+17,
+      "learning_rate": 9.907468672167165e-05,
+      "loss": 7.4144,
+      "loss/crossentropy": 2.0388287007808685,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2378864921629429,
+      "step": 369
+    },
+    {
+      "epoch": 0.06166666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 4.1721627340208877e+17,
+      "learning_rate": 9.906966668700567e-05,
+      "loss": 7.2032,
+      "loss/crossentropy": 2.2473433315753937,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1772308126091957,
+      "step": 370
+    },
+    {
+      "epoch": 0.06183333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.172162734074714e+17,
+      "learning_rate": 9.906463319961225e-05,
+      "loss": 7.3783,
+      "loss/crossentropy": 2.0179379880428314,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21671415120363235,
+      "step": 371
+    },
+    {
+      "epoch": 0.062,
+      "grad_norm": 33.75,
+      "grad_norm_var": 4.172162733697926e+17,
+      "learning_rate": 9.90595862608714e-05,
+      "loss": 7.3767,
+      "loss/crossentropy": 1.8956446200609207,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17291199415922165,
+      "step": 372
+    },
+    {
+      "epoch": 0.06216666666666667,
+      "grad_norm": 35.25,
+      "grad_norm_var": 4.172162732675215e+17,
+      "learning_rate": 9.90545258721667e-05,
+      "loss": 7.0952,
+      "loss/crossentropy": 1.9084904491901398,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17385511845350266,
+      "step": 373
+    },
+    {
+      "epoch": 0.06233333333333333,
+      "grad_norm": 43.25,
+      "grad_norm_var": 4.172162730522139e+17,
+      "learning_rate": 9.904945203488554e-05,
+      "loss": 7.4098,
+      "loss/crossentropy": 2.5498337745666504,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22701657190918922,
+      "step": 374
+    },
+    {
+      "epoch": 0.0625,
+      "grad_norm": 31.25,
+      "grad_norm_var": 4.1721627308451e+17,
+      "learning_rate": 9.904436475041891e-05,
+      "loss": 7.4016,
+      "loss/crossentropy": 2.4548984467983246,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18526265025138855,
+      "step": 375
+    },
+    {
+      "epoch": 0.06266666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 18.427018229166666,
+      "learning_rate": 9.903926402016153e-05,
+      "loss": 7.5127,
+      "loss/crossentropy": 2.308915287256241,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20047524943947792,
+      "step": 376
+    },
+    {
+      "epoch": 0.06283333333333334,
+      "grad_norm": 35.25,
+      "grad_norm_var": 8.6759765625,
+      "learning_rate": 9.903414984551179e-05,
+      "loss": 7.2701,
+      "loss/crossentropy": 1.520705834031105,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16450045630335808,
+      "step": 377
+    },
+    {
+      "epoch": 0.063,
+      "grad_norm": 31.25,
+      "grad_norm_var": 8.978580729166667,
+      "learning_rate": 9.902902222787175e-05,
+      "loss": 7.4852,
+      "loss/crossentropy": 2.291164994239807,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2040424570441246,
+      "step": 378
+    },
+    {
+      "epoch": 0.06316666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 9.327018229166667,
+      "learning_rate": 9.902388116864722e-05,
+      "loss": 7.4674,
+      "loss/crossentropy": 1.819806694984436,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17983737587928772,
+      "step": 379
+    },
+    {
+      "epoch": 0.06333333333333334,
+      "grad_norm": 34.25,
+      "grad_norm_var": 9.258268229166667,
+      "learning_rate": 9.901872666924764e-05,
+      "loss": 7.6163,
+      "loss/crossentropy": 2.0015306621789932,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24792687594890594,
+      "step": 380
+    },
+    {
+      "epoch": 0.0635,
+      "grad_norm": 33.0,
+      "grad_norm_var": 9.111393229166667,
+      "learning_rate": 9.901355873108609e-05,
+      "loss": 7.6637,
+      "loss/crossentropy": 2.000072866678238,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23586535453796387,
+      "step": 381
+    },
+    {
+      "epoch": 0.06366666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 8.867643229166667,
+      "learning_rate": 9.900837735557947e-05,
+      "loss": 7.2072,
+      "loss/crossentropy": 2.2368247509002686,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17863766849040985,
+      "step": 382
+    },
+    {
+      "epoch": 0.06383333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 8.8994140625,
+      "learning_rate": 9.900318254414821e-05,
+      "loss": 7.6932,
+      "loss/crossentropy": 2.405503123998642,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19412217661738396,
+      "step": 383
+    },
+    {
+      "epoch": 0.064,
+      "grad_norm": 46.75,
+      "grad_norm_var": 19.9103515625,
+      "learning_rate": 9.899797429821656e-05,
+      "loss": 7.2723,
+      "loss/crossentropy": 1.9972580075263977,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2110944762825966,
+      "step": 384
+    },
+    {
+      "epoch": 0.06416666666666666,
+      "grad_norm": 35.5,
+      "grad_norm_var": 19.5119140625,
+      "learning_rate": 9.899275261921234e-05,
+      "loss": 7.3453,
+      "loss/crossentropy": 1.6727921962738037,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20363643392920494,
+      "step": 385
+    },
+    {
+      "epoch": 0.06433333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 18.690625,
+      "learning_rate": 9.898751750856713e-05,
+      "loss": 7.3233,
+      "loss/crossentropy": 2.2225599586963654,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17307283729314804,
+      "step": 386
+    },
+    {
+      "epoch": 0.0645,
+      "grad_norm": 32.5,
+      "grad_norm_var": 18.614322916666666,
+      "learning_rate": 9.898226896771619e-05,
+      "loss": 7.3348,
+      "loss/crossentropy": 2.3554776310920715,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19947733730077744,
+      "step": 387
+    },
+    {
+      "epoch": 0.06466666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 19.2587890625,
+      "learning_rate": 9.897700699809837e-05,
+      "loss": 7.5742,
+      "loss/crossentropy": 2.0066860020160675,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20728905498981476,
+      "step": 388
+    },
+    {
+      "epoch": 0.06483333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 19.4650390625,
+      "learning_rate": 9.897173160115632e-05,
+      "loss": 7.4815,
+      "loss/crossentropy": 1.7207957953214645,
+      "loss/hidden": 3.90234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17158129811286926,
+      "step": 389
+    },
+    {
+      "epoch": 0.065,
+      "grad_norm": 31.75,
+      "grad_norm_var": 14.086393229166667,
+      "learning_rate": 9.896644277833631e-05,
+      "loss": 7.6021,
+      "loss/crossentropy": 2.282930314540863,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1811271533370018,
+      "step": 390
+    },
+    {
+      "epoch": 0.06516666666666666,
+      "grad_norm": 34.0,
+      "grad_norm_var": 13.6853515625,
+      "learning_rate": 9.896114053108829e-05,
+      "loss": 7.3217,
+      "loss/crossentropy": 2.172467976808548,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18695824220776558,
+      "step": 391
+    },
+    {
+      "epoch": 0.06533333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 13.7509765625,
+      "learning_rate": 9.895582486086592e-05,
+      "loss": 7.363,
+      "loss/crossentropy": 2.033529132604599,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21254323050379753,
+      "step": 392
+    },
+    {
+      "epoch": 0.0655,
+      "grad_norm": 33.5,
+      "grad_norm_var": 13.594205729166667,
+      "learning_rate": 9.89504957691265e-05,
+      "loss": 7.3398,
+      "loss/crossentropy": 1.848247617483139,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18556710332632065,
+      "step": 393
+    },
+    {
+      "epoch": 0.06566666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 13.869205729166667,
+      "learning_rate": 9.894515325733103e-05,
+      "loss": 7.3527,
+      "loss/crossentropy": 2.0109578371047974,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1525113210082054,
+      "step": 394
+    },
+    {
+      "epoch": 0.06583333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 13.391080729166667,
+      "learning_rate": 9.893979732694421e-05,
+      "loss": 7.3288,
+      "loss/crossentropy": 2.3585988879203796,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17590664327144623,
+      "step": 395
+    },
+    {
+      "epoch": 0.066,
+      "grad_norm": 33.25,
+      "grad_norm_var": 13.392122395833333,
+      "learning_rate": 9.89344279794344e-05,
+      "loss": 7.3622,
+      "loss/crossentropy": 2.0552004277706146,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2060481533408165,
+      "step": 396
+    },
+    {
+      "epoch": 0.06616666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 13.551497395833334,
+      "learning_rate": 9.892904521627361e-05,
+      "loss": 7.2862,
+      "loss/crossentropy": 2.1628974080085754,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22357721626758575,
+      "step": 397
+    },
+    {
+      "epoch": 0.06633333333333333,
+      "grad_norm": 35.75,
+      "grad_norm_var": 13.8041015625,
+      "learning_rate": 9.892364903893759e-05,
+      "loss": 7.2218,
+      "loss/crossentropy": 1.8713282942771912,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1839936561882496,
+      "step": 398
+    },
+    {
+      "epoch": 0.0665,
+      "grad_norm": 34.25,
+      "grad_norm_var": 13.7306640625,
+      "learning_rate": 9.891823944890568e-05,
+      "loss": 7.3538,
+      "loss/crossentropy": 1.3194625228643417,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19643132388591766,
+      "step": 399
+    },
+    {
+      "epoch": 0.06666666666666667,
+      "grad_norm": 41.75,
+      "grad_norm_var": 6.735872395833334,
+      "learning_rate": 9.8912816447661e-05,
+      "loss": 7.445,
+      "loss/crossentropy": 1.8842482566833496,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19740041717886925,
+      "step": 400
+    },
+    {
+      "epoch": 0.06683333333333333,
+      "grad_norm": 37.0,
+      "grad_norm_var": 7.256184895833333,
+      "learning_rate": 9.890738003669029e-05,
+      "loss": 7.232,
+      "loss/crossentropy": 1.4622259736061096,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21342862397432327,
+      "step": 401
+    },
+    {
+      "epoch": 0.067,
+      "grad_norm": 32.0,
+      "grad_norm_var": 7.308268229166667,
+      "learning_rate": 9.890193021748395e-05,
+      "loss": 7.2306,
+      "loss/crossentropy": 1.8058496713638306,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1889837346971035,
+      "step": 402
+    },
+    {
+      "epoch": 0.06716666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 7.2134765625,
+      "learning_rate": 9.88964669915361e-05,
+      "loss": 7.5215,
+      "loss/crossentropy": 1.8579385876655579,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19213734194636345,
+      "step": 403
+    },
+    {
+      "epoch": 0.06733333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 6.81640625,
+      "learning_rate": 9.889099036034451e-05,
+      "loss": 7.5444,
+      "loss/crossentropy": 2.120022624731064,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2287553995847702,
+      "step": 404
+    },
+    {
+      "epoch": 0.0675,
+      "grad_norm": 33.0,
+      "grad_norm_var": 6.739322916666667,
+      "learning_rate": 9.888550032541059e-05,
+      "loss": 7.3545,
+      "loss/crossentropy": 1.3314496725797653,
+      "loss/hidden": 3.765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18857388570904732,
+      "step": 405
+    },
+    {
+      "epoch": 0.06766666666666667,
+      "grad_norm": 38.5,
+      "grad_norm_var": 7.632291666666666,
+      "learning_rate": 9.887999688823955e-05,
+      "loss": 7.2507,
+      "loss/crossentropy": 2.1126837134361267,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17216311395168304,
+      "step": 406
+    },
+    {
+      "epoch": 0.06783333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 8.0119140625,
+      "learning_rate": 9.88744800503401e-05,
+      "loss": 7.1781,
+      "loss/crossentropy": 2.1377086341381073,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1858111023902893,
+      "step": 407
+    },
+    {
+      "epoch": 0.068,
+      "grad_norm": 29.625,
+      "grad_norm_var": 9.230989583333333,
+      "learning_rate": 9.886894981322476e-05,
+      "loss": 7.3993,
+      "loss/crossentropy": 2.1618523001670837,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20276832208037376,
+      "step": 408
+    },
+    {
+      "epoch": 0.06816666666666667,
+      "grad_norm": 37.25,
+      "grad_norm_var": 9.852083333333333,
+      "learning_rate": 9.886340617840968e-05,
+      "loss": 7.1343,
+      "loss/crossentropy": 1.9829298257827759,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1895209550857544,
+      "step": 409
+    },
+    {
+      "epoch": 0.06833333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 9.043489583333333,
+      "learning_rate": 9.885784914741465e-05,
+      "loss": 7.3848,
+      "loss/crossentropy": 1.9673973619937897,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1718750186264515,
+      "step": 410
+    },
+    {
+      "epoch": 0.0685,
+      "grad_norm": 31.75,
+      "grad_norm_var": 9.477083333333333,
+      "learning_rate": 9.88522787217632e-05,
+      "loss": 7.1923,
+      "loss/crossentropy": 2.023388981819153,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17376725003123283,
+      "step": 411
+    },
+    {
+      "epoch": 0.06866666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 9.672916666666667,
+      "learning_rate": 9.884669490298244e-05,
+      "loss": 7.3261,
+      "loss/crossentropy": 1.9144360721111298,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1796419359743595,
+      "step": 412
+    },
+    {
+      "epoch": 0.06883333333333333,
+      "grad_norm": 35.0,
+      "grad_norm_var": 9.360416666666667,
+      "learning_rate": 9.884109769260325e-05,
+      "loss": 7.4866,
+      "loss/crossentropy": 1.9528678357601166,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21739878505468369,
+      "step": 413
+    },
+    {
+      "epoch": 0.069,
+      "grad_norm": 33.5,
+      "grad_norm_var": 9.264322916666666,
+      "learning_rate": 9.883548709216013e-05,
+      "loss": 7.3382,
+      "loss/crossentropy": 1.7033936232328415,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19289804250001907,
+      "step": 414
+    },
+    {
+      "epoch": 0.06916666666666667,
+      "grad_norm": 35.0,
+      "grad_norm_var": 9.301041666666666,
+      "learning_rate": 9.882986310319124e-05,
+      "loss": 7.2687,
+      "loss/crossentropy": 1.8803034126758575,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16312895342707634,
+      "step": 415
+    },
+    {
+      "epoch": 0.06933333333333333,
+      "grad_norm": 39.5,
+      "grad_norm_var": 7.376822916666667,
+      "learning_rate": 9.882422572723844e-05,
+      "loss": 7.5184,
+      "loss/crossentropy": 1.9304080605506897,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2324250303208828,
+      "step": 416
+    },
+    {
+      "epoch": 0.0695,
+      "grad_norm": 32.5,
+      "grad_norm_var": 6.926822916666667,
+      "learning_rate": 9.881857496584726e-05,
+      "loss": 7.458,
+      "loss/crossentropy": 2.000861704349518,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2086937390267849,
+      "step": 417
+    },
+    {
+      "epoch": 0.06966666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 7.06640625,
+      "learning_rate": 9.881291082056685e-05,
+      "loss": 7.5073,
+      "loss/crossentropy": 1.7774852812290192,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19647850096225739,
+      "step": 418
+    },
+    {
+      "epoch": 0.06983333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 7.48125,
+      "learning_rate": 9.880723329295012e-05,
+      "loss": 7.5581,
+      "loss/crossentropy": 1.8696882873773575,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24500614032149315,
+      "step": 419
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 31.625,
+      "grad_norm_var": 7.745768229166667,
+      "learning_rate": 9.880154238455356e-05,
+      "loss": 7.2781,
+      "loss/crossentropy": 2.131507694721222,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19151924923062325,
+      "step": 420
+    },
+    {
+      "epoch": 0.07016666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 7.745768229166667,
+      "learning_rate": 9.879583809693738e-05,
+      "loss": 7.5388,
+      "loss/crossentropy": 2.105042278766632,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1929212473332882,
+      "step": 421
+    },
+    {
+      "epoch": 0.07033333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 6.020768229166666,
+      "learning_rate": 9.879012043166542e-05,
+      "loss": 7.2608,
+      "loss/crossentropy": 2.0129053592681885,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.161848496645689,
+      "step": 422
+    },
+    {
+      "epoch": 0.0705,
+      "grad_norm": 34.25,
+      "grad_norm_var": 5.955208333333333,
+      "learning_rate": 9.878438939030526e-05,
+      "loss": 7.4747,
+      "loss/crossentropy": 1.9779498279094696,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17465418577194214,
+      "step": 423
+    },
+    {
+      "epoch": 0.07066666666666667,
+      "grad_norm": 38.0,
+      "grad_norm_var": 6.186393229166667,
+      "learning_rate": 9.877864497442804e-05,
+      "loss": 7.4397,
+      "loss/crossentropy": 2.0635344684123993,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18920517340302467,
+      "step": 424
+    },
+    {
+      "epoch": 0.07083333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 5.3822265625,
+      "learning_rate": 9.877288718560866e-05,
+      "loss": 7.3153,
+      "loss/crossentropy": 1.782182663679123,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2136215791106224,
+      "step": 425
+    },
+    {
+      "epoch": 0.071,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.5041015625,
+      "learning_rate": 9.876711602542563e-05,
+      "loss": 7.4077,
+      "loss/crossentropy": 1.6254926919937134,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.194427989423275,
+      "step": 426
+    },
+    {
+      "epoch": 0.07116666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 6.0087890625,
+      "learning_rate": 9.876133149546118e-05,
+      "loss": 7.3599,
+      "loss/crossentropy": 2.198939800262451,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18414749205112457,
+      "step": 427
+    },
+    {
+      "epoch": 0.07133333333333333,
+      "grad_norm": 35.0,
+      "grad_norm_var": 6.031705729166666,
+      "learning_rate": 9.875553359730114e-05,
+      "loss": 7.4082,
+      "loss/crossentropy": 2.247727543115616,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19302746653556824,
+      "step": 428
+    },
+    {
+      "epoch": 0.0715,
+      "grad_norm": 34.0,
+      "grad_norm_var": 5.913997395833333,
+      "learning_rate": 9.874972233253504e-05,
+      "loss": 7.3478,
+      "loss/crossentropy": 2.195002317428589,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17825636267662048,
+      "step": 429
+    },
+    {
+      "epoch": 0.07166666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 7.378580729166667,
+      "learning_rate": 9.874389770275607e-05,
+      "loss": 7.2517,
+      "loss/crossentropy": 2.2952789962291718,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16881508007645607,
+      "step": 430
+    },
+    {
+      "epoch": 0.07183333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 7.660416666666666,
+      "learning_rate": 9.87380597095611e-05,
+      "loss": 7.3159,
+      "loss/crossentropy": 1.8355306088924408,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17771482467651367,
+      "step": 431
+    },
+    {
+      "epoch": 0.072,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.65625,
+      "learning_rate": 9.873220835455064e-05,
+      "loss": 7.3736,
+      "loss/crossentropy": 2.0866390466690063,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1750289537012577,
+      "step": 432
+    },
+    {
+      "epoch": 0.07216666666666667,
+      "grad_norm": 34.75,
+      "grad_norm_var": 4.95390625,
+      "learning_rate": 9.872634363932887e-05,
+      "loss": 7.3727,
+      "loss/crossentropy": 2.025203436613083,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1730334535241127,
+      "step": 433
+    },
+    {
+      "epoch": 0.07233333333333333,
+      "grad_norm": 36.25,
+      "grad_norm_var": 5.602083333333334,
+      "learning_rate": 9.872046556550363e-05,
+      "loss": 7.2171,
+      "loss/crossentropy": 2.128805488348007,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19256963953375816,
+      "step": 434
+    },
+    {
+      "epoch": 0.0725,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.38515625,
+      "learning_rate": 9.871457413468644e-05,
+      "loss": 7.3632,
+      "loss/crossentropy": 1.7909742146730423,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18350538611412048,
+      "step": 435
+    },
+    {
+      "epoch": 0.07266666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.259830729166667,
+      "learning_rate": 9.870866934849248e-05,
+      "loss": 7.2637,
+      "loss/crossentropy": 1.7241093814373016,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2180027849972248,
+      "step": 436
+    },
+    {
+      "epoch": 0.07283333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 5.269205729166667,
+      "learning_rate": 9.870275120854054e-05,
+      "loss": 7.4072,
+      "loss/crossentropy": 1.9474639296531677,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18808619305491447,
+      "step": 437
+    },
+    {
+      "epoch": 0.073,
+      "grad_norm": 38.75,
+      "grad_norm_var": 7.200455729166666,
+      "learning_rate": 9.869681971645315e-05,
+      "loss": 7.3156,
+      "loss/crossentropy": 1.4366422295570374,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16360682621598244,
+      "step": 438
+    },
+    {
+      "epoch": 0.07316666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 7.167643229166667,
+      "learning_rate": 9.869087487385644e-05,
+      "loss": 7.226,
+      "loss/crossentropy": 1.7024643123149872,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15661080181598663,
+      "step": 439
+    },
+    {
+      "epoch": 0.07333333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 5.876041666666667,
+      "learning_rate": 9.868491668238025e-05,
+      "loss": 7.4396,
+      "loss/crossentropy": 1.8631815910339355,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2046206295490265,
+      "step": 440
+    },
+    {
+      "epoch": 0.0735,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.874739583333334,
+      "learning_rate": 9.867894514365802e-05,
+      "loss": 7.2107,
+      "loss/crossentropy": 1.690757930278778,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1710291989147663,
+      "step": 441
+    },
+    {
+      "epoch": 0.07366666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 6.048372395833334,
+      "learning_rate": 9.867296025932688e-05,
+      "loss": 7.2471,
+      "loss/crossentropy": 2.181719124317169,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1899368017911911,
+      "step": 442
+    },
+    {
+      "epoch": 0.07383333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.5041015625,
+      "learning_rate": 9.866696203102766e-05,
+      "loss": 7.5269,
+      "loss/crossentropy": 2.421279162168503,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21067209541797638,
+      "step": 443
+    },
+    {
+      "epoch": 0.074,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.2931640625,
+      "learning_rate": 9.866095046040478e-05,
+      "loss": 7.5283,
+      "loss/crossentropy": 2.0766140818595886,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19953646883368492,
+      "step": 444
+    },
+    {
+      "epoch": 0.07416666666666667,
+      "grad_norm": 31.875,
+      "grad_norm_var": 5.303125,
+      "learning_rate": 9.865492554910633e-05,
+      "loss": 7.5428,
+      "loss/crossentropy": 1.8468139171600342,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21182066947221756,
+      "step": 445
+    },
+    {
+      "epoch": 0.07433333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.162239583333333,
+      "learning_rate": 9.86488872987841e-05,
+      "loss": 7.3663,
+      "loss/crossentropy": 2.3212408125400543,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20349783077836037,
+      "step": 446
+    },
+    {
+      "epoch": 0.0745,
+      "grad_norm": 36.75,
+      "grad_norm_var": 4.378059895833333,
+      "learning_rate": 9.864283571109352e-05,
+      "loss": 7.4066,
+      "loss/crossentropy": 2.3919509053230286,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20775995030999184,
+      "step": 447
+    },
+    {
+      "epoch": 0.07466666666666667,
+      "grad_norm": 39.25,
+      "grad_norm_var": 6.318684895833333,
+      "learning_rate": 9.863677078769362e-05,
+      "loss": 7.4543,
+      "loss/crossentropy": 2.1256485879421234,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2432199865579605,
+      "step": 448
+    },
+    {
+      "epoch": 0.07483333333333334,
+      "grad_norm": 33.75,
+      "grad_norm_var": 6.271809895833333,
+      "learning_rate": 9.863069253024719e-05,
+      "loss": 7.5944,
+      "loss/crossentropy": 1.751033991575241,
+      "loss/hidden": 3.8828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2695496417582035,
+      "step": 449
+    },
+    {
+      "epoch": 0.075,
+      "grad_norm": 33.75,
+      "grad_norm_var": 5.8681640625,
+      "learning_rate": 9.862460094042056e-05,
+      "loss": 7.2885,
+      "loss/crossentropy": 2.0045522451400757,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21929249167442322,
+      "step": 450
+    },
+    {
+      "epoch": 0.07516666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 6.1509765625,
+      "learning_rate": 9.861849601988383e-05,
+      "loss": 7.3548,
+      "loss/crossentropy": 2.2150400578975677,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1829955205321312,
+      "step": 451
+    },
+    {
+      "epoch": 0.07533333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 6.329166666666667,
+      "learning_rate": 9.861237777031068e-05,
+      "loss": 7.5843,
+      "loss/crossentropy": 2.0443906784057617,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1967557854950428,
+      "step": 452
+    },
+    {
+      "epoch": 0.0755,
+      "grad_norm": 32.75,
+      "grad_norm_var": 6.329166666666667,
+      "learning_rate": 9.860624619337844e-05,
+      "loss": 7.2489,
+      "loss/crossentropy": 2.106353849172592,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19701764360070229,
+      "step": 453
+    },
+    {
+      "epoch": 0.07566666666666666,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.45390625,
+      "learning_rate": 9.860010129076813e-05,
+      "loss": 7.3707,
+      "loss/crossentropy": 1.9924251735210419,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18526853993535042,
+      "step": 454
+    },
+    {
+      "epoch": 0.07583333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.532291666666667,
+      "learning_rate": 9.859394306416444e-05,
+      "loss": 7.3933,
+      "loss/crossentropy": 1.6768793761730194,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23321601003408432,
+      "step": 455
+    },
+    {
+      "epoch": 0.076,
+      "grad_norm": 34.5,
+      "grad_norm_var": 4.4619140625,
+      "learning_rate": 9.858777151525564e-05,
+      "loss": 7.2093,
+      "loss/crossentropy": 2.154035836458206,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1648601647466421,
+      "step": 456
+    },
+    {
+      "epoch": 0.07616666666666666,
+      "grad_norm": 34.5,
+      "grad_norm_var": 4.5353515625,
+      "learning_rate": 9.85815866457337e-05,
+      "loss": 7.4823,
+      "loss/crossentropy": 2.151414453983307,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2598721943795681,
+      "step": 457
+    },
+    {
+      "epoch": 0.07633333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.314322916666667,
+      "learning_rate": 9.857538845729426e-05,
+      "loss": 7.142,
+      "loss/crossentropy": 2.3205729722976685,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19651631638407707,
+      "step": 458
+    },
+    {
+      "epoch": 0.0765,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.44140625,
+      "learning_rate": 9.856917695163658e-05,
+      "loss": 7.4089,
+      "loss/crossentropy": 1.6730692237615585,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20352322235703468,
+      "step": 459
+    },
+    {
+      "epoch": 0.07666666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 4.542643229166667,
+      "learning_rate": 9.856295213046357e-05,
+      "loss": 7.3371,
+      "loss/crossentropy": 2.4532963037490845,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21194115281105042,
+      "step": 460
+    },
+    {
+      "epoch": 0.07683333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 5.435872395833333,
+      "learning_rate": 9.855671399548181e-05,
+      "loss": 7.4147,
+      "loss/crossentropy": 1.6418677270412445,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20057519152760506,
+      "step": 461
+    },
+    {
+      "epoch": 0.077,
+      "grad_norm": 32.25,
+      "grad_norm_var": 5.398893229166666,
+      "learning_rate": 9.855046254840151e-05,
+      "loss": 7.4987,
+      "loss/crossentropy": 1.9779995381832123,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2105957791209221,
+      "step": 462
+    },
+    {
+      "epoch": 0.07716666666666666,
+      "grad_norm": 39.5,
+      "grad_norm_var": 7.157747395833334,
+      "learning_rate": 9.854419779093655e-05,
+      "loss": 7.7277,
+      "loss/crossentropy": 1.5468579232692719,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17518384009599686,
+      "step": 463
+    },
+    {
+      "epoch": 0.07733333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 4.858333333333333,
+      "learning_rate": 9.853791972480445e-05,
+      "loss": 7.2777,
+      "loss/crossentropy": 2.683764636516571,
+      "loss/hidden": 3.13671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17364051938056946,
+      "step": 464
+    },
+    {
+      "epoch": 0.0775,
+      "grad_norm": 30.625,
+      "grad_norm_var": 5.130143229166666,
+      "learning_rate": 9.853162835172637e-05,
+      "loss": 7.1146,
+      "loss/crossentropy": 1.7351168394088745,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17206433415412903,
+      "step": 465
+    },
+    {
+      "epoch": 0.07766666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 5.078580729166666,
+      "learning_rate": 9.852532367342713e-05,
+      "loss": 7.3142,
+      "loss/crossentropy": 2.0744443237781525,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19953011348843575,
+      "step": 466
+    },
+    {
+      "epoch": 0.07783333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 5.20390625,
+      "learning_rate": 9.851900569163519e-05,
+      "loss": 7.3702,
+      "loss/crossentropy": 2.4033637046813965,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19170591607689857,
+      "step": 467
+    },
+    {
+      "epoch": 0.078,
+      "grad_norm": 31.875,
+      "grad_norm_var": 5.172916666666667,
+      "learning_rate": 9.851267440808265e-05,
+      "loss": 7.3081,
+      "loss/crossentropy": 2.3863994479179382,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18752896040678024,
+      "step": 468
+    },
+    {
+      "epoch": 0.07816666666666666,
+      "grad_norm": 37.0,
+      "grad_norm_var": 6.337239583333333,
+      "learning_rate": 9.85063298245053e-05,
+      "loss": 7.4378,
+      "loss/crossentropy": 1.724489688873291,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21728365495800972,
+      "step": 469
+    },
+    {
+      "epoch": 0.07833333333333334,
+      "grad_norm": 38.75,
+      "grad_norm_var": 8.410416666666666,
+      "learning_rate": 9.84999719426425e-05,
+      "loss": 7.491,
+      "loss/crossentropy": 1.7358942180871964,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2069646678864956,
+      "step": 470
+    },
+    {
+      "epoch": 0.0785,
+      "grad_norm": 34.75,
+      "grad_norm_var": 8.424739583333333,
+      "learning_rate": 9.849360076423734e-05,
+      "loss": 7.1822,
+      "loss/crossentropy": 1.9611319601535797,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16901595145463943,
+      "step": 471
+    },
+    {
+      "epoch": 0.07866666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 8.349739583333333,
+      "learning_rate": 9.84872162910365e-05,
+      "loss": 7.4088,
+      "loss/crossentropy": 1.8835695087909698,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22427978739142418,
+      "step": 472
+    },
+    {
+      "epoch": 0.07883333333333334,
+      "grad_norm": 30.5,
+      "grad_norm_var": 8.724739583333333,
+      "learning_rate": 9.84808185247903e-05,
+      "loss": 7.3324,
+      "loss/crossentropy": 2.1350006461143494,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19744988158345222,
+      "step": 473
+    },
+    {
+      "epoch": 0.079,
+      "grad_norm": 35.25,
+      "grad_norm_var": 8.917708333333334,
+      "learning_rate": 9.847440746725275e-05,
+      "loss": 7.1986,
+      "loss/crossentropy": 1.997681975364685,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1898837462067604,
+      "step": 474
+    },
+    {
+      "epoch": 0.07916666666666666,
+      "grad_norm": 39.75,
+      "grad_norm_var": 11.34765625,
+      "learning_rate": 9.846798312018146e-05,
+      "loss": 7.3897,
+      "loss/crossentropy": 2.253058522939682,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22173089161515236,
+      "step": 475
+    },
+    {
+      "epoch": 0.07933333333333334,
+      "grad_norm": 34.0,
+      "grad_norm_var": 11.094205729166667,
+      "learning_rate": 9.846154548533773e-05,
+      "loss": 7.1971,
+      "loss/crossentropy": 2.1781056821346283,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2001049444079399,
+      "step": 476
+    },
+    {
+      "epoch": 0.0795,
+      "grad_norm": 29.625,
+      "grad_norm_var": 10.947330729166667,
+      "learning_rate": 9.845509456448643e-05,
+      "loss": 7.2308,
+      "loss/crossentropy": 2.23237344622612,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18285806477069855,
+      "step": 477
+    },
+    {
+      "epoch": 0.07966666666666666,
+      "grad_norm": 4060086272.0,
+      "grad_norm_var": 1.0302687662371242e+18,
+      "learning_rate": 9.844863035939615e-05,
+      "loss": 8.7708,
+      "loss/crossentropy": 2.050332546234131,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1951880268752575,
+      "step": 478
+    },
+    {
+      "epoch": 0.07983333333333334,
+      "grad_norm": 42.75,
+      "grad_norm_var": 1.0302687661271636e+18,
+      "learning_rate": 9.844215287183909e-05,
+      "loss": 7.4037,
+      "loss/crossentropy": 1.9212382435798645,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1802588328719139,
+      "step": 479
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 39.25,
+      "grad_norm_var": 1.030268765869179e+18,
+      "learning_rate": 9.843566210359106e-05,
+      "loss": 7.5005,
+      "loss/crossentropy": 2.1036358773708344,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23785079643130302,
+      "step": 480
+    },
+    {
+      "epoch": 0.08016666666666666,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.0302687658522619e+18,
+      "learning_rate": 9.842915805643155e-05,
+      "loss": 7.2304,
+      "loss/crossentropy": 1.495644435286522,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2004375085234642,
+      "step": 481
+    },
+    {
+      "epoch": 0.08033333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.0302687659283886e+18,
+      "learning_rate": 9.842264073214371e-05,
+      "loss": 7.3683,
+      "loss/crossentropy": 1.248510017991066,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19620612636208534,
+      "step": 482
+    },
+    {
+      "epoch": 0.0805,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.0302687659241594e+18,
+      "learning_rate": 9.841611013251429e-05,
+      "loss": 7.5022,
+      "loss/crossentropy": 2.183932900428772,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18435726314783096,
+      "step": 483
+    },
+    {
+      "epoch": 0.08066666666666666,
+      "grad_norm": 34.5,
+      "grad_norm_var": 1.0302687658353449e+18,
+      "learning_rate": 9.840956625933367e-05,
+      "loss": 7.2567,
+      "loss/crossentropy": 2.0957889556884766,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20050502195954323,
+      "step": 484
+    },
+    {
+      "epoch": 0.08083333333333333,
+      "grad_norm": 39.5,
+      "grad_norm_var": 1.0302687657507598e+18,
+      "learning_rate": 9.840300911439591e-05,
+      "loss": 7.2279,
+      "loss/crossentropy": 2.1745176017284393,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17116260156035423,
+      "step": 485
+    },
+    {
+      "epoch": 0.081,
+      "grad_norm": 35.75,
+      "grad_norm_var": 1.0302687658522619e+18,
+      "learning_rate": 9.839643869949866e-05,
+      "loss": 7.1858,
+      "loss/crossentropy": 2.020654261112213,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1931912563741207,
+      "step": 486
+    },
+    {
+      "epoch": 0.08116666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.030268766000286e+18,
+      "learning_rate": 9.838985501644328e-05,
+      "loss": 7.579,
+      "loss/crossentropy": 1.9844678938388824,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19125735759735107,
+      "step": 487
+    },
+    {
+      "epoch": 0.08133333333333333,
+      "grad_norm": 38.0,
+      "grad_norm_var": 1.0302687658311156e+18,
+      "learning_rate": 9.83832580670347e-05,
+      "loss": 7.2854,
+      "loss/crossentropy": 1.9453465342521667,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2013980746269226,
+      "step": 488
+    },
+    {
+      "epoch": 0.0815,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.030268765754989e+18,
+      "learning_rate": 9.837664785308149e-05,
+      "loss": 7.1819,
+      "loss/crossentropy": 1.9938696026802063,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16968103870749474,
+      "step": 489
+    },
+    {
+      "epoch": 0.08166666666666667,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.0302687658734083e+18,
+      "learning_rate": 9.837002437639593e-05,
+      "loss": 7.2293,
+      "loss/crossentropy": 1.6919759213924408,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17779743671417236,
+      "step": 490
+    },
+    {
+      "epoch": 0.08183333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.030268766215978e+18,
+      "learning_rate": 9.836338763879385e-05,
+      "loss": 7.2308,
+      "loss/crossentropy": 1.9851961731910706,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22075363248586655,
+      "step": 491
+    },
+    {
+      "epoch": 0.082,
+      "grad_norm": 33.75,
+      "grad_norm_var": 1.0302687662244365e+18,
+      "learning_rate": 9.835673764209474e-05,
+      "loss": 7.3698,
+      "loss/crossentropy": 1.759193554520607,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1959233544766903,
+      "step": 492
+    },
+    {
+      "epoch": 0.08216666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 1.030268766067954e+18,
+      "learning_rate": 9.835007438812177e-05,
+      "loss": 7.3148,
+      "loss/crossentropy": 1.7496148943901062,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19785473495721817,
+      "step": 493
+    },
+    {
+      "epoch": 0.08233333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 15.7978515625,
+      "learning_rate": 9.834339787870166e-05,
+      "loss": 7.5355,
+      "loss/crossentropy": 2.0986676067113876,
+      "loss/hidden": 3.7734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19847061857581139,
+      "step": 494
+    },
+    {
+      "epoch": 0.0825,
+      "grad_norm": 32.25,
+      "grad_norm_var": 10.5369140625,
+      "learning_rate": 9.833670811566485e-05,
+      "loss": 7.1154,
+      "loss/crossentropy": 1.5699402540922165,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1794753484427929,
+      "step": 495
+    },
+    {
+      "epoch": 0.08266666666666667,
+      "grad_norm": 37.0,
+      "grad_norm_var": 9.1025390625,
+      "learning_rate": 9.833000510084537e-05,
+      "loss": 7.3676,
+      "loss/crossentropy": 1.7302229404449463,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19138918817043304,
+      "step": 496
+    },
+    {
+      "epoch": 0.08283333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 8.817708333333334,
+      "learning_rate": 9.832328883608088e-05,
+      "loss": 7.2645,
+      "loss/crossentropy": 2.1858323514461517,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18926378339529037,
+      "step": 497
+    },
+    {
+      "epoch": 0.083,
+      "grad_norm": 31.75,
+      "grad_norm_var": 8.605989583333333,
+      "learning_rate": 9.83165593232127e-05,
+      "loss": 7.3413,
+      "loss/crossentropy": 1.4862932115793228,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19968301057815552,
+      "step": 498
+    },
+    {
+      "epoch": 0.08316666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 8.18515625,
+      "learning_rate": 9.830981656408574e-05,
+      "loss": 7.2423,
+      "loss/crossentropy": 1.6016323864459991,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.197268046438694,
+      "step": 499
+    },
+    {
+      "epoch": 0.08333333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 8.372330729166666,
+      "learning_rate": 9.830306056054858e-05,
+      "loss": 7.3029,
+      "loss/crossentropy": 1.7606691420078278,
+      "loss/hidden": 3.640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20658674463629723,
+      "step": 500
+    },
+    {
+      "epoch": 0.0835,
+      "grad_norm": 30.875,
+      "grad_norm_var": 6.076822916666667,
+      "learning_rate": 9.829629131445342e-05,
+      "loss": 7.2979,
+      "loss/crossentropy": 1.6005878746509552,
+      "loss/hidden": 3.88671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24985505267977715,
+      "step": 501
+    },
+    {
+      "epoch": 0.08366666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 5.651822916666666,
+      "learning_rate": 9.828950882765608e-05,
+      "loss": 7.3027,
+      "loss/crossentropy": 1.9283245205879211,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17439668625593185,
+      "step": 502
+    },
+    {
+      "epoch": 0.08383333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 5.228059895833334,
+      "learning_rate": 9.828271310201601e-05,
+      "loss": 7.2721,
+      "loss/crossentropy": 1.546176627278328,
+      "loss/hidden": 3.734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22614171728491783,
+      "step": 503
+    },
+    {
+      "epoch": 0.084,
+      "grad_norm": 33.75,
+      "grad_norm_var": 3.528059895833333,
+      "learning_rate": 9.827590413939632e-05,
+      "loss": 7.3988,
+      "loss/crossentropy": 2.066196322441101,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23446843028068542,
+      "step": 504
+    },
+    {
+      "epoch": 0.08416666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.5317057291666667,
+      "learning_rate": 9.82690819416637e-05,
+      "loss": 7.2985,
+      "loss/crossentropy": 1.7853144407272339,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1875152923166752,
+      "step": 505
+    },
+    {
+      "epoch": 0.08433333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.792122395833333,
+      "learning_rate": 9.826224651068852e-05,
+      "loss": 7.2318,
+      "loss/crossentropy": 2.338760584592819,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20884433016180992,
+      "step": 506
+    },
+    {
+      "epoch": 0.0845,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.1497395833333335,
+      "learning_rate": 9.825539784834472e-05,
+      "loss": 7.2198,
+      "loss/crossentropy": 2.0317736864089966,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19717568531632423,
+      "step": 507
+    },
+    {
+      "epoch": 0.08466666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.5677083333333335,
+      "learning_rate": 9.824853595650991e-05,
+      "loss": 7.3425,
+      "loss/crossentropy": 2.238571286201477,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1741107814013958,
+      "step": 508
+    },
+    {
+      "epoch": 0.08483333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.3760416666666666,
+      "learning_rate": 9.824166083706534e-05,
+      "loss": 7.5655,
+      "loss/crossentropy": 2.2807562947273254,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20003702491521835,
+      "step": 509
+    },
+    {
+      "epoch": 0.085,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.9010416666666665,
+      "learning_rate": 9.823477249189586e-05,
+      "loss": 7.5241,
+      "loss/crossentropy": 1.9748588800430298,
+      "loss/hidden": 3.12109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15270168893039227,
+      "step": 510
+    },
+    {
+      "epoch": 0.08516666666666667,
+      "grad_norm": 36.0,
+      "grad_norm_var": 3.6080729166666665,
+      "learning_rate": 9.822787092288991e-05,
+      "loss": 7.175,
+      "loss/crossentropy": 1.509592205286026,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17142293229699135,
+      "step": 511
+    },
+    {
+      "epoch": 0.08533333333333333,
+      "grad_norm": 39.75,
+      "grad_norm_var": 5.610416666666667,
+      "learning_rate": 9.822095613193962e-05,
+      "loss": 7.2398,
+      "loss/crossentropy": 1.9376066327095032,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1732519268989563,
+      "step": 512
+    },
+    {
+      "epoch": 0.0855,
+      "grad_norm": 31.625,
+      "grad_norm_var": 5.603580729166667,
+      "learning_rate": 9.821402812094073e-05,
+      "loss": 7.417,
+      "loss/crossentropy": 2.128629118204117,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1934521086513996,
+      "step": 513
+    },
+    {
+      "epoch": 0.08566666666666667,
+      "grad_norm": 34.5,
+      "grad_norm_var": 5.678059895833333,
+      "learning_rate": 9.820708689179259e-05,
+      "loss": 7.3364,
+      "loss/crossentropy": 1.7944347858428955,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14837288111448288,
+      "step": 514
+    },
+    {
+      "epoch": 0.08583333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 5.713997395833333,
+      "learning_rate": 9.820013244639816e-05,
+      "loss": 7.4981,
+      "loss/crossentropy": 1.5384874045848846,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15830077044665813,
+      "step": 515
+    },
+    {
+      "epoch": 0.086,
+      "grad_norm": 33.5,
+      "grad_norm_var": 5.599739583333333,
+      "learning_rate": 9.819316478666405e-05,
+      "loss": 7.5107,
+      "loss/crossentropy": 1.8282313644886017,
+      "loss/hidden": 3.71875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24127519503235817,
+      "step": 516
+    },
+    {
+      "epoch": 0.08616666666666667,
+      "grad_norm": 36.25,
+      "grad_norm_var": 5.826497395833333,
+      "learning_rate": 9.81861839145005e-05,
+      "loss": 7.5059,
+      "loss/crossentropy": 1.7223950922489166,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21317560225725174,
+      "step": 517
+    },
+    {
+      "epoch": 0.08633333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 5.799934895833333,
+      "learning_rate": 9.817918983182132e-05,
+      "loss": 7.4321,
+      "loss/crossentropy": 2.056610018014908,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2198064923286438,
+      "step": 518
+    },
+    {
+      "epoch": 0.0865,
+      "grad_norm": 31.625,
+      "grad_norm_var": 5.980208333333334,
+      "learning_rate": 9.8172182540544e-05,
+      "loss": 7.099,
+      "loss/crossentropy": 1.5413986295461655,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17947009205818176,
+      "step": 519
+    },
+    {
+      "epoch": 0.08666666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 6.03125,
+      "learning_rate": 9.816516204258963e-05,
+      "loss": 7.3258,
+      "loss/crossentropy": 1.9781705439090729,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21023594588041306,
+      "step": 520
+    },
+    {
+      "epoch": 0.08683333333333333,
+      "grad_norm": 34.5,
+      "grad_norm_var": 6.08125,
+      "learning_rate": 9.815812833988291e-05,
+      "loss": 7.1305,
+      "loss/crossentropy": 2.3283976316452026,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20475587621331215,
+      "step": 521
+    },
+    {
+      "epoch": 0.087,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.513541666666667,
+      "learning_rate": 9.815108143435218e-05,
+      "loss": 7.4251,
+      "loss/crossentropy": 1.9967394471168518,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17605987191200256,
+      "step": 522
+    },
+    {
+      "epoch": 0.08716666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.597916666666666,
+      "learning_rate": 9.814402132792939e-05,
+      "loss": 7.24,
+      "loss/crossentropy": 1.959173321723938,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1960609070956707,
+      "step": 523
+    },
+    {
+      "epoch": 0.08733333333333333,
+      "grad_norm": 34.75,
+      "grad_norm_var": 4.79140625,
+      "learning_rate": 9.81369480225501e-05,
+      "loss": 7.4309,
+      "loss/crossentropy": 1.61872236430645,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17693763226270676,
+      "step": 524
+    },
+    {
+      "epoch": 0.0875,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.621875,
+      "learning_rate": 9.812986152015348e-05,
+      "loss": 7.1741,
+      "loss/crossentropy": 1.8704243898391724,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16461655870079994,
+      "step": 525
+    },
+    {
+      "epoch": 0.08766666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.33125,
+      "learning_rate": 9.812276182268236e-05,
+      "loss": 6.9185,
+      "loss/crossentropy": 1.6973167210817337,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16682424396276474,
+      "step": 526
+    },
+    {
+      "epoch": 0.08783333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 4.04765625,
+      "learning_rate": 9.811564893208318e-05,
+      "loss": 7.2562,
+      "loss/crossentropy": 2.335329443216324,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21264615282416344,
+      "step": 527
+    },
+    {
+      "epoch": 0.088,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.7384765625,
+      "learning_rate": 9.810852285030593e-05,
+      "loss": 7.2408,
+      "loss/crossentropy": 1.8945126831531525,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18671311624348164,
+      "step": 528
+    },
+    {
+      "epoch": 0.08816666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 1.5393229166666667,
+      "learning_rate": 9.81013835793043e-05,
+      "loss": 7.327,
+      "loss/crossentropy": 1.7044986188411713,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1740158013999462,
+      "step": 529
+    },
+    {
+      "epoch": 0.08833333333333333,
+      "grad_norm": 37.0,
+      "grad_norm_var": 2.258072916666667,
+      "learning_rate": 9.809423112103554e-05,
+      "loss": 7.3285,
+      "loss/crossentropy": 1.4971919655799866,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17978889867663383,
+      "step": 530
+    },
+    {
+      "epoch": 0.0885,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.11875,
+      "learning_rate": 9.808706547746057e-05,
+      "loss": 7.5005,
+      "loss/crossentropy": 2.047203630208969,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18218868225812912,
+      "step": 531
+    },
+    {
+      "epoch": 0.08866666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.309375,
+      "learning_rate": 9.807988665054386e-05,
+      "loss": 7.1867,
+      "loss/crossentropy": 1.9481738805770874,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18189242109656334,
+      "step": 532
+    },
+    {
+      "epoch": 0.08883333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.2979166666666666,
+      "learning_rate": 9.807269464225355e-05,
+      "loss": 7.4141,
+      "loss/crossentropy": 2.143236815929413,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1998455822467804,
+      "step": 533
+    },
+    {
+      "epoch": 0.089,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.0854166666666667,
+      "learning_rate": 9.806548945456134e-05,
+      "loss": 7.3471,
+      "loss/crossentropy": 1.6931321620941162,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16084032505750656,
+      "step": 534
+    },
+    {
+      "epoch": 0.08916666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.7926432291666665,
+      "learning_rate": 9.80582710894426e-05,
+      "loss": 7.5005,
+      "loss/crossentropy": 2.3337481021881104,
+      "loss/hidden": 3.05859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1558638084679842,
+      "step": 535
+    },
+    {
+      "epoch": 0.08933333333333333,
+      "grad_norm": 36.5,
+      "grad_norm_var": 4.486393229166667,
+      "learning_rate": 9.805103954887627e-05,
+      "loss": 7.4294,
+      "loss/crossentropy": 1.9169566333293915,
+      "loss/hidden": 3.77734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2434011958539486,
+      "step": 536
+    },
+    {
+      "epoch": 0.0895,
+      "grad_norm": 40.0,
+      "grad_norm_var": 7.379622395833334,
+      "learning_rate": 9.804379483484494e-05,
+      "loss": 7.6072,
+      "loss/crossentropy": 1.5788866728544235,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1909298300743103,
+      "step": 537
+    },
+    {
+      "epoch": 0.08966666666666667,
+      "grad_norm": 31.75,
+      "grad_norm_var": 7.556705729166667,
+      "learning_rate": 9.803653694933476e-05,
+      "loss": 7.3319,
+      "loss/crossentropy": 1.990968108177185,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18440885841846466,
+      "step": 538
+    },
+    {
+      "epoch": 0.08983333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 7.767708333333333,
+      "learning_rate": 9.802926589433553e-05,
+      "loss": 7.4037,
+      "loss/crossentropy": 2.2752568125724792,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1688903495669365,
+      "step": 539
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 30.125,
+      "grad_norm_var": 8.237434895833333,
+      "learning_rate": 9.802198167184067e-05,
+      "loss": 7.0915,
+      "loss/crossentropy": 1.9928709864616394,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1749878227710724,
+      "step": 540
+    },
+    {
+      "epoch": 0.09016666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 8.237434895833333,
+      "learning_rate": 9.801468428384716e-05,
+      "loss": 7.3821,
+      "loss/crossentropy": 1.349032387137413,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26186642050743103,
+      "step": 541
+    },
+    {
+      "epoch": 0.09033333333333333,
+      "grad_norm": 35.0,
+      "grad_norm_var": 8.4744140625,
+      "learning_rate": 9.800737373235565e-05,
+      "loss": 7.3504,
+      "loss/crossentropy": 2.0728733241558075,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17149914801120758,
+      "step": 542
+    },
+    {
+      "epoch": 0.0905,
+      "grad_norm": 37.25,
+      "grad_norm_var": 9.5134765625,
+      "learning_rate": 9.800005001937034e-05,
+      "loss": 7.2083,
+      "loss/crossentropy": 2.1556628346443176,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21120557934045792,
+      "step": 543
+    },
+    {
+      "epoch": 0.09066666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 10.165559895833333,
+      "learning_rate": 9.799271314689908e-05,
+      "loss": 7.3664,
+      "loss/crossentropy": 1.8796964585781097,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22796564176678658,
+      "step": 544
+    },
+    {
+      "epoch": 0.09083333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 10.1837890625,
+      "learning_rate": 9.798536311695334e-05,
+      "loss": 7.2641,
+      "loss/crossentropy": 1.7632421851158142,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16169985756278038,
+      "step": 545
+    },
+    {
+      "epoch": 0.091,
+      "grad_norm": 31.125,
+      "grad_norm_var": 9.323958333333334,
+      "learning_rate": 9.797799993154814e-05,
+      "loss": 7.3923,
+      "loss/crossentropy": 1.725706309080124,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16693437471985817,
+      "step": 546
+    },
+    {
+      "epoch": 0.09116666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 9.382291666666667,
+      "learning_rate": 9.797062359270215e-05,
+      "loss": 7.8711,
+      "loss/crossentropy": 2.060252219438553,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21199437975883484,
+      "step": 547
+    },
+    {
+      "epoch": 0.09133333333333334,
+      "grad_norm": 34.75,
+      "grad_norm_var": 9.614322916666667,
+      "learning_rate": 9.796323410243763e-05,
+      "loss": 7.2769,
+      "loss/crossentropy": 2.35904923081398,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21982023119926453,
+      "step": 548
+    },
+    {
+      "epoch": 0.0915,
+      "grad_norm": 33.75,
+      "grad_norm_var": 9.345572916666667,
+      "learning_rate": 9.795583146278046e-05,
+      "loss": 7.2131,
+      "loss/crossentropy": 2.113744080066681,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17868803814053535,
+      "step": 549
+    },
+    {
+      "epoch": 0.09166666666666666,
+      "grad_norm": 34.25,
+      "grad_norm_var": 8.651822916666667,
+      "learning_rate": 9.794841567576011e-05,
+      "loss": 7.2687,
+      "loss/crossentropy": 2.0138320922851562,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19911404326558113,
+      "step": 550
+    },
+    {
+      "epoch": 0.09183333333333334,
+      "grad_norm": 33.75,
+      "grad_norm_var": 7.629166666666666,
+      "learning_rate": 9.794098674340965e-05,
+      "loss": 7.2922,
+      "loss/crossentropy": 1.5388011932373047,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18451552465558052,
+      "step": 551
+    },
+    {
+      "epoch": 0.092,
+      "grad_norm": 32.25,
+      "grad_norm_var": 7.093489583333334,
+      "learning_rate": 9.793354466776579e-05,
+      "loss": 7.4383,
+      "loss/crossentropy": 1.9522328674793243,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1844920516014099,
+      "step": 552
+    },
+    {
+      "epoch": 0.09216666666666666,
+      "grad_norm": 33.75,
+      "grad_norm_var": 3.948958333333333,
+      "learning_rate": 9.79260894508688e-05,
+      "loss": 7.1703,
+      "loss/crossentropy": 1.3523796498775482,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17598746716976166,
+      "step": 553
+    },
+    {
+      "epoch": 0.09233333333333334,
+      "grad_norm": 36.0,
+      "grad_norm_var": 4.42265625,
+      "learning_rate": 9.791862109476257e-05,
+      "loss": 7.5446,
+      "loss/crossentropy": 1.8268292248249054,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19541719183325768,
+      "step": 554
+    },
+    {
+      "epoch": 0.0925,
+      "grad_norm": 32.75,
+      "grad_norm_var": 4.144205729166667,
+      "learning_rate": 9.791113960149458e-05,
+      "loss": 7.4347,
+      "loss/crossentropy": 2.058392435312271,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17756398767232895,
+      "step": 555
+    },
+    {
+      "epoch": 0.09266666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.7171223958333335,
+      "learning_rate": 9.790364497311597e-05,
+      "loss": 7.3495,
+      "loss/crossentropy": 1.9875744879245758,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18157724291086197,
+      "step": 556
+    },
+    {
+      "epoch": 0.09283333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.8916015625,
+      "learning_rate": 9.789613721168139e-05,
+      "loss": 7.4422,
+      "loss/crossentropy": 2.0252414643764496,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1976994201540947,
+      "step": 557
+    },
+    {
+      "epoch": 0.093,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.7134765625,
+      "learning_rate": 9.788861631924913e-05,
+      "loss": 7.4124,
+      "loss/crossentropy": 1.737688809633255,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19330373406410217,
+      "step": 558
+    },
+    {
+      "epoch": 0.09316666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.5160807291666667,
+      "learning_rate": 9.788108229788111e-05,
+      "loss": 7.378,
+      "loss/crossentropy": 2.3494531214237213,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17403588816523552,
+      "step": 559
+    },
+    {
+      "epoch": 0.09333333333333334,
+      "grad_norm": 37.5,
+      "grad_norm_var": 3.1395833333333334,
+      "learning_rate": 9.787353514964284e-05,
+      "loss": 7.3312,
+      "loss/crossentropy": 2.2165339291095734,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1995975784957409,
+      "step": 560
+    },
+    {
+      "epoch": 0.0935,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.1080729166666665,
+      "learning_rate": 9.786597487660337e-05,
+      "loss": 7.2734,
+      "loss/crossentropy": 2.126331925392151,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2150227539241314,
+      "step": 561
+    },
+    {
+      "epoch": 0.09366666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.8223307291666666,
+      "learning_rate": 9.785840148083543e-05,
+      "loss": 7.3413,
+      "loss/crossentropy": 2.0839502215385437,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1780762765556574,
+      "step": 562
+    },
+    {
+      "epoch": 0.09383333333333334,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.6192057291666666,
+      "learning_rate": 9.785081496441527e-05,
+      "loss": 7.5268,
+      "loss/crossentropy": 2.3646166920661926,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17533158883452415,
+      "step": 563
+    },
+    {
+      "epoch": 0.094,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.4760416666666667,
+      "learning_rate": 9.784321532942282e-05,
+      "loss": 7.1331,
+      "loss/crossentropy": 2.0853044986724854,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18720799684524536,
+      "step": 564
+    },
+    {
+      "epoch": 0.09416666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.5229166666666667,
+      "learning_rate": 9.783560257794154e-05,
+      "loss": 7.2869,
+      "loss/crossentropy": 2.1955504715442657,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19676203280687332,
+      "step": 565
+    },
+    {
+      "epoch": 0.09433333333333334,
+      "grad_norm": 35.25,
+      "grad_norm_var": 3.7270833333333333,
+      "learning_rate": 9.78279767120585e-05,
+      "loss": 7.2674,
+      "loss/crossentropy": 1.9641484469175339,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17377197369933128,
+      "step": 566
+    },
+    {
+      "epoch": 0.0945,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.767708333333333,
+      "learning_rate": 9.782033773386439e-05,
+      "loss": 7.3658,
+      "loss/crossentropy": 2.190814882516861,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21332580596208572,
+      "step": 567
+    },
+    {
+      "epoch": 0.09466666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 4.4072265625,
+      "learning_rate": 9.781268564545348e-05,
+      "loss": 7.2034,
+      "loss/crossentropy": 1.9772469699382782,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18993888422846794,
+      "step": 568
+    },
+    {
+      "epoch": 0.09483333333333334,
+      "grad_norm": 33.5,
+      "grad_norm_var": 4.386393229166667,
+      "learning_rate": 9.780502044892362e-05,
+      "loss": 7.0386,
+      "loss/crossentropy": 1.7807036340236664,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17942193523049355,
+      "step": 569
+    },
+    {
+      "epoch": 0.095,
+      "grad_norm": 36.5,
+      "grad_norm_var": 4.6025390625,
+      "learning_rate": 9.779734214637628e-05,
+      "loss": 7.4366,
+      "loss/crossentropy": 1.63731350004673,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18429255858063698,
+      "step": 570
+    },
+    {
+      "epoch": 0.09516666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.890625,
+      "learning_rate": 9.778965073991651e-05,
+      "loss": 7.2578,
+      "loss/crossentropy": 1.4358558505773544,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1632411926984787,
+      "step": 571
+    },
+    {
+      "epoch": 0.09533333333333334,
+      "grad_norm": 30.375,
+      "grad_norm_var": 5.157291666666667,
+      "learning_rate": 9.778194623165296e-05,
+      "loss": 7.2741,
+      "loss/crossentropy": 1.9152775704860687,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2016092799603939,
+      "step": 572
+    },
+    {
+      "epoch": 0.0955,
+      "grad_norm": 31.5,
+      "grad_norm_var": 5.157291666666667,
+      "learning_rate": 9.777422862369783e-05,
+      "loss": 7.4085,
+      "loss/crossentropy": 2.0969631373882294,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18521985039114952,
+      "step": 573
+    },
+    {
+      "epoch": 0.09566666666666666,
+      "grad_norm": 34.5,
+      "grad_norm_var": 5.25,
+      "learning_rate": 9.776649791816698e-05,
+      "loss": 7.4702,
+      "loss/crossentropy": 2.282189190387726,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18798528239130974,
+      "step": 574
+    },
+    {
+      "epoch": 0.09583333333333334,
+      "grad_norm": 37.5,
+      "grad_norm_var": 6.5625,
+      "learning_rate": 9.77587541171798e-05,
+      "loss": 7.1049,
+      "loss/crossentropy": 1.7100197970867157,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17929638177156448,
+      "step": 575
+    },
+    {
+      "epoch": 0.096,
+      "grad_norm": 36.0,
+      "grad_norm_var": 5.840625,
+      "learning_rate": 9.775099722285935e-05,
+      "loss": 7.3286,
+      "loss/crossentropy": 2.1536933183670044,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1613265983760357,
+      "step": 576
+    },
+    {
+      "epoch": 0.09616666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.840625,
+      "learning_rate": 9.774322723733216e-05,
+      "loss": 7.5203,
+      "loss/crossentropy": 1.5367402732372284,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1888187788426876,
+      "step": 577
+    },
+    {
+      "epoch": 0.09633333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 5.982291666666667,
+      "learning_rate": 9.773544416272845e-05,
+      "loss": 7.3827,
+      "loss/crossentropy": 2.4369155168533325,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20236976817250252,
+      "step": 578
+    },
+    {
+      "epoch": 0.0965,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.829166666666667,
+      "learning_rate": 9.772764800118199e-05,
+      "loss": 7.4226,
+      "loss/crossentropy": 2.500710666179657,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1806088238954544,
+      "step": 579
+    },
+    {
+      "epoch": 0.09666666666666666,
+      "grad_norm": 35.25,
+      "grad_norm_var": 5.322330729166667,
+      "learning_rate": 9.771983875483013e-05,
+      "loss": 7.3541,
+      "loss/crossentropy": 1.5330149829387665,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23990492895245552,
+      "step": 580
+    },
+    {
+      "epoch": 0.09683333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 5.557291666666667,
+      "learning_rate": 9.771201642581385e-05,
+      "loss": 7.2871,
+      "loss/crossentropy": 2.1098267138004303,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18698688223958015,
+      "step": 581
+    },
+    {
+      "epoch": 0.097,
+      "grad_norm": 30.625,
+      "grad_norm_var": 5.6416015625,
+      "learning_rate": 9.770418101627765e-05,
+      "loss": 7.1604,
+      "loss/crossentropy": 1.893375426530838,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1971338987350464,
+      "step": 582
+    },
+    {
+      "epoch": 0.09716666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 5.6134765625,
+      "learning_rate": 9.769633252836969e-05,
+      "loss": 7.5679,
+      "loss/crossentropy": 1.917736440896988,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20975341275334358,
+      "step": 583
+    },
+    {
+      "epoch": 0.09733333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 5.16015625,
+      "learning_rate": 9.768847096424164e-05,
+      "loss": 7.4809,
+      "loss/crossentropy": 2.0464819073677063,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1731277070939541,
+      "step": 584
+    },
+    {
+      "epoch": 0.0975,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.16640625,
+      "learning_rate": 9.76805963260488e-05,
+      "loss": 7.2994,
+      "loss/crossentropy": 2.2766425013542175,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20238212123513222,
+      "step": 585
+    },
+    {
+      "epoch": 0.09766666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.316666666666666,
+      "learning_rate": 9.767270861595005e-05,
+      "loss": 7.4116,
+      "loss/crossentropy": 1.4148157089948654,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1840950809419155,
+      "step": 586
+    },
+    {
+      "epoch": 0.09783333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 4.228580729166667,
+      "learning_rate": 9.766480783610788e-05,
+      "loss": 7.4112,
+      "loss/crossentropy": 1.929806500673294,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22025159001350403,
+      "step": 587
+    },
+    {
+      "epoch": 0.098,
+      "grad_norm": 35.75,
+      "grad_norm_var": 4.270572916666667,
+      "learning_rate": 9.765689398868831e-05,
+      "loss": 7.3646,
+      "loss/crossentropy": 1.3431737124919891,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1964380256831646,
+      "step": 588
+    },
+    {
+      "epoch": 0.09816666666666667,
+      "grad_norm": 35.0,
+      "grad_norm_var": 4.255989583333333,
+      "learning_rate": 9.764896707586096e-05,
+      "loss": 7.6459,
+      "loss/crossentropy": 2.5541348457336426,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2149493768811226,
+      "step": 589
+    },
+    {
+      "epoch": 0.09833333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 4.298372395833334,
+      "learning_rate": 9.764102709979902e-05,
+      "loss": 7.4885,
+      "loss/crossentropy": 1.5293695777654648,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20554304495453835,
+      "step": 590
+    },
+    {
+      "epoch": 0.0985,
+      "grad_norm": 1904214016.0,
+      "grad_norm_var": 2.2662693082968016e+17,
+      "learning_rate": 9.763307406267932e-05,
+      "loss": 7.5986,
+      "loss/crossentropy": 1.9793291985988617,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24488302320241928,
+      "step": 591
+    },
+    {
+      "epoch": 0.09866666666666667,
+      "grad_norm": 41.0,
+      "grad_norm_var": 2.2662693075033792e+17,
+      "learning_rate": 9.76251079666822e-05,
+      "loss": 7.4414,
+      "loss/crossentropy": 1.9010595381259918,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19852081686258316,
+      "step": 592
+    },
+    {
+      "epoch": 0.09883333333333333,
+      "grad_norm": 37.0,
+      "grad_norm_var": 2.266269306789299e+17,
+      "learning_rate": 9.761712881399164e-05,
+      "loss": 7.1544,
+      "loss/crossentropy": 1.7115428000688553,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2066071517765522,
+      "step": 593
+    },
+    {
+      "epoch": 0.099,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.2662693070669968e+17,
+      "learning_rate": 9.760913660679515e-05,
+      "loss": 7.1712,
+      "loss/crossentropy": 2.307830184698105,
+      "loss/hidden": 3.078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17552950605750084,
+      "step": 594
+    },
+    {
+      "epoch": 0.09916666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.2662693075827216e+17,
+      "learning_rate": 9.760113134728384e-05,
+      "loss": 7.1973,
+      "loss/crossentropy": 2.09252592921257,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19455183297395706,
+      "step": 595
+    },
+    {
+      "epoch": 0.09933333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.266269308415815e+17,
+      "learning_rate": 9.75931130376524e-05,
+      "loss": 7.2217,
+      "loss/crossentropy": 2.44236558675766,
+      "loss/hidden": 3.10546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17834673076868057,
+      "step": 596
+    },
+    {
+      "epoch": 0.0995,
+      "grad_norm": 35.75,
+      "grad_norm_var": 2.2662693076818992e+17,
+      "learning_rate": 9.75850816800991e-05,
+      "loss": 7.5668,
+      "loss/crossentropy": 2.2520895898342133,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22190653532743454,
+      "step": 597
+    },
+    {
+      "epoch": 0.09966666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.2662693072653523e+17,
+      "learning_rate": 9.757703727682574e-05,
+      "loss": 7.2988,
+      "loss/crossentropy": 1.8116507530212402,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2566680610179901,
+      "step": 598
+    },
+    {
+      "epoch": 0.09983333333333333,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.2662693070669968e+17,
+      "learning_rate": 9.756897983003781e-05,
+      "loss": 7.1157,
+      "loss/crossentropy": 1.840036928653717,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1869695521891117,
+      "step": 599
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.2662693065512723e+17,
+      "learning_rate": 9.756090934194427e-05,
+      "loss": 7.3596,
+      "loss/crossentropy": 2.069575548171997,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18857147172093391,
+      "step": 600
+    },
+    {
+      "epoch": 0.10016666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.266269306789299e+17,
+      "learning_rate": 9.755282581475769e-05,
+      "loss": 7.0587,
+      "loss/crossentropy": 1.478778824210167,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17617631517350674,
+      "step": 601
+    },
+    {
+      "epoch": 0.10033333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.2662693072058458e+17,
+      "learning_rate": 9.75447292506942e-05,
+      "loss": 7.3075,
+      "loss/crossentropy": 2.0797568261623383,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27910368144512177,
+      "step": 602
+    },
+    {
+      "epoch": 0.1005,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.2662693070074902e+17,
+      "learning_rate": 9.753661965197354e-05,
+      "loss": 7.3651,
+      "loss/crossentropy": 2.084309071302414,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2038382701575756,
+      "step": 603
+    },
+    {
+      "epoch": 0.10066666666666667,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.2662693076422282e+17,
+      "learning_rate": 9.752849702081901e-05,
+      "loss": 7.4663,
+      "loss/crossentropy": 2.4660587310791016,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2053012251853943,
+      "step": 604
+    },
+    {
+      "epoch": 0.10083333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.2662693085348285e+17,
+      "learning_rate": 9.752036135945744e-05,
+      "loss": 7.1663,
+      "loss/crossentropy": 2.302393853664398,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17444464564323425,
+      "step": 605
+    },
+    {
+      "epoch": 0.101,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.266269308514993e+17,
+      "learning_rate": 9.751221267011929e-05,
+      "loss": 7.2215,
+      "loss/crossentropy": 1.6606972068548203,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17002883926033974,
+      "step": 606
+    },
+    {
+      "epoch": 0.10116666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 9.884375,
+      "learning_rate": 9.750405095503859e-05,
+      "loss": 7.4077,
+      "loss/crossentropy": 1.8132504224777222,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1946292594075203,
+      "step": 607
+    },
+    {
+      "epoch": 0.10133333333333333,
+      "grad_norm": 35.0,
+      "grad_norm_var": 5.659375,
+      "learning_rate": 9.749587621645288e-05,
+      "loss": 7.2269,
+      "loss/crossentropy": 2.005624294281006,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17027952894568443,
+      "step": 608
+    },
+    {
+      "epoch": 0.1015,
+      "grad_norm": 30.25,
+      "grad_norm_var": 4.48515625,
+      "learning_rate": 9.748768845660334e-05,
+      "loss": 7.3604,
+      "loss/crossentropy": 1.998530238866806,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19627553597092628,
+      "step": 609
+    },
+    {
+      "epoch": 0.10166666666666667,
+      "grad_norm": 35.25,
+      "grad_norm_var": 4.645572916666667,
+      "learning_rate": 9.74794876777347e-05,
+      "loss": 7.331,
+      "loss/crossentropy": 1.8363763093948364,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19507349282503128,
+      "step": 610
+    },
+    {
+      "epoch": 0.10183333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 4.378059895833333,
+      "learning_rate": 9.74712738820952e-05,
+      "loss": 7.2472,
+      "loss/crossentropy": 1.8793150782585144,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16764680296182632,
+      "step": 611
+    },
+    {
+      "epoch": 0.102,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.9311848958333333,
+      "learning_rate": 9.746304707193675e-05,
+      "loss": 7.4801,
+      "loss/crossentropy": 1.9134803712368011,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2131960727274418,
+      "step": 612
+    },
+    {
+      "epoch": 0.10216666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.4208333333333334,
+      "learning_rate": 9.745480724951473e-05,
+      "loss": 7.2178,
+      "loss/crossentropy": 2.210248053073883,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1874031201004982,
+      "step": 613
+    },
+    {
+      "epoch": 0.10233333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.3955729166666666,
+      "learning_rate": 9.744655441708818e-05,
+      "loss": 7.1421,
+      "loss/crossentropy": 1.9995661079883575,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16971386969089508,
+      "step": 614
+    },
+    {
+      "epoch": 0.1025,
+      "grad_norm": 36.25,
+      "grad_norm_var": 4.086458333333334,
+      "learning_rate": 9.743828857691963e-05,
+      "loss": 7.2352,
+      "loss/crossentropy": 2.1933152973651886,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21801244840025902,
+      "step": 615
+    },
+    {
+      "epoch": 0.10266666666666667,
+      "grad_norm": 34.75,
+      "grad_norm_var": 4.158072916666667,
+      "learning_rate": 9.743000973127523e-05,
+      "loss": 7.3893,
+      "loss/crossentropy": 1.9973363876342773,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17263447493314743,
+      "step": 616
+    },
+    {
+      "epoch": 0.10283333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 4.1125,
+      "learning_rate": 9.742171788242466e-05,
+      "loss": 7.2193,
+      "loss/crossentropy": 1.671396553516388,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16563132032752037,
+      "step": 617
+    },
+    {
+      "epoch": 0.103,
+      "grad_norm": 31.625,
+      "grad_norm_var": 3.925,
+      "learning_rate": 9.741341303264118e-05,
+      "loss": 7.3605,
+      "loss/crossentropy": 1.9448569267988205,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17194196209311485,
+      "step": 618
+    },
+    {
+      "epoch": 0.10316666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 4.224934895833333,
+      "learning_rate": 9.74050951842016e-05,
+      "loss": 7.3728,
+      "loss/crossentropy": 2.0389251559972763,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20530594512820244,
+      "step": 619
+    },
+    {
+      "epoch": 0.10333333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 4.699739583333334,
+      "learning_rate": 9.739676433938633e-05,
+      "loss": 7.4521,
+      "loss/crossentropy": 1.8410753458738327,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2363889515399933,
+      "step": 620
+    },
+    {
+      "epoch": 0.1035,
+      "grad_norm": 29.75,
+      "grad_norm_var": 4.562434895833333,
+      "learning_rate": 9.73884205004793e-05,
+      "loss": 7.2949,
+      "loss/crossentropy": 1.9485596120357513,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2630031704902649,
+      "step": 621
+    },
+    {
+      "epoch": 0.10366666666666667,
+      "grad_norm": 37.0,
+      "grad_norm_var": 5.911393229166666,
+      "learning_rate": 9.7380063669768e-05,
+      "loss": 7.3083,
+      "loss/crossentropy": 2.08263298869133,
+      "loss/hidden": 3.1171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15711040794849396,
+      "step": 622
+    },
+    {
+      "epoch": 0.10383333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 6.001497395833334,
+      "learning_rate": 9.737169384954355e-05,
+      "loss": 7.4823,
+      "loss/crossentropy": 1.850400984287262,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2069677822291851,
+      "step": 623
+    },
+    {
+      "epoch": 0.104,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.628580729166667,
+      "learning_rate": 9.736331104210056e-05,
+      "loss": 7.183,
+      "loss/crossentropy": 2.005878895521164,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18370692059397697,
+      "step": 624
+    },
+    {
+      "epoch": 0.10416666666666667,
+      "grad_norm": 31.875,
+      "grad_norm_var": 5.29765625,
+      "learning_rate": 9.735491524973722e-05,
+      "loss": 7.3535,
+      "loss/crossentropy": 2.2689976394176483,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1807100549340248,
+      "step": 625
+    },
+    {
+      "epoch": 0.10433333333333333,
+      "grad_norm": 37.75,
+      "grad_norm_var": 6.558072916666666,
+      "learning_rate": 9.73465064747553e-05,
+      "loss": 7.2869,
+      "loss/crossentropy": 1.604978010058403,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16992727294564247,
+      "step": 626
+    },
+    {
+      "epoch": 0.1045,
+      "grad_norm": 44.25,
+      "grad_norm_var": 14.2150390625,
+      "learning_rate": 9.73380847194601e-05,
+      "loss": 7.4146,
+      "loss/crossentropy": 1.9762012362480164,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1730644293129444,
+      "step": 627
+    },
+    {
+      "epoch": 0.10466666666666667,
+      "grad_norm": 37.5,
+      "grad_norm_var": 15.056184895833333,
+      "learning_rate": 9.732964998616046e-05,
+      "loss": 7.2546,
+      "loss/crossentropy": 1.5720670521259308,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19047683849930763,
+      "step": 628
+    },
+    {
+      "epoch": 0.10483333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 15.665559895833333,
+      "learning_rate": 9.732120227716888e-05,
+      "loss": 7.7217,
+      "loss/crossentropy": 2.6308372616767883,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18224241584539413,
+      "step": 629
+    },
+    {
+      "epoch": 0.105,
+      "grad_norm": 31.125,
+      "grad_norm_var": 16.102083333333333,
+      "learning_rate": 9.73127415948013e-05,
+      "loss": 7.2075,
+      "loss/crossentropy": 2.029748886823654,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19252733886241913,
+      "step": 630
+    },
+    {
+      "epoch": 0.10516666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 16.159375,
+      "learning_rate": 9.730426794137727e-05,
+      "loss": 7.2245,
+      "loss/crossentropy": 2.165195047855377,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17404165491461754,
+      "step": 631
+    },
+    {
+      "epoch": 0.10533333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 16.2666015625,
+      "learning_rate": 9.72957813192199e-05,
+      "loss": 7.0976,
+      "loss/crossentropy": 2.231869101524353,
+      "loss/hidden": 3.11328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.159787118434906,
+      "step": 632
+    },
+    {
+      "epoch": 0.1055,
+      "grad_norm": 31.25,
+      "grad_norm_var": 16.2666015625,
+      "learning_rate": 9.728728173065585e-05,
+      "loss": 7.1263,
+      "loss/crossentropy": 1.78048437833786,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1777312085032463,
+      "step": 633
+    },
+    {
+      "epoch": 0.10566666666666667,
+      "grad_norm": 36.0,
+      "grad_norm_var": 16.546875,
+      "learning_rate": 9.72787691780153e-05,
+      "loss": 7.5738,
+      "loss/crossentropy": 2.1110564172267914,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2632845491170883,
+      "step": 634
+    },
+    {
+      "epoch": 0.10583333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 15.9931640625,
+      "learning_rate": 9.727024366363206e-05,
+      "loss": 7.3205,
+      "loss/crossentropy": 1.8406849801540375,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2111925520002842,
+      "step": 635
+    },
+    {
+      "epoch": 0.106,
+      "grad_norm": 30.75,
+      "grad_norm_var": 15.478125,
+      "learning_rate": 9.726170518984341e-05,
+      "loss": 7.315,
+      "loss/crossentropy": 2.237843692302704,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18497232720255852,
+      "step": 636
+    },
+    {
+      "epoch": 0.10616666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 15.2916015625,
+      "learning_rate": 9.725315375899024e-05,
+      "loss": 7.4921,
+      "loss/crossentropy": 2.02110955119133,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20984560251235962,
+      "step": 637
+    },
+    {
+      "epoch": 0.10633333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 15.426041666666666,
+      "learning_rate": 9.724458937341698e-05,
+      "loss": 7.1764,
+      "loss/crossentropy": 2.0535950362682343,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19004171341657639,
+      "step": 638
+    },
+    {
+      "epoch": 0.1065,
+      "grad_norm": 35.75,
+      "grad_norm_var": 15.772916666666667,
+      "learning_rate": 9.723601203547158e-05,
+      "loss": 7.0975,
+      "loss/crossentropy": 2.0000297725200653,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1943572536110878,
+      "step": 639
+    },
+    {
+      "epoch": 0.10666666666666667,
+      "grad_norm": 52.5,
+      "grad_norm_var": 38.72604166666667,
+      "learning_rate": 9.722742174750558e-05,
+      "loss": 7.1918,
+      "loss/crossentropy": 1.8211422562599182,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17954260483384132,
+      "step": 640
+    },
+    {
+      "epoch": 0.10683333333333334,
+      "grad_norm": 38.0,
+      "grad_norm_var": 38.90149739583333,
+      "learning_rate": 9.721881851187406e-05,
+      "loss": 7.2372,
+      "loss/crossentropy": 1.7122246026992798,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2225099690258503,
+      "step": 641
+    },
+    {
+      "epoch": 0.107,
+      "grad_norm": 34.0,
+      "grad_norm_var": 38.36243489583333,
+      "learning_rate": 9.721020233093563e-05,
+      "loss": 7.3873,
+      "loss/crossentropy": 2.2953185439109802,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21673983708024025,
+      "step": 642
+    },
+    {
+      "epoch": 0.10716666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 31.917122395833335,
+      "learning_rate": 9.72015732070525e-05,
+      "loss": 7.276,
+      "loss/crossentropy": 1.8881819546222687,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18341781198978424,
+      "step": 643
+    },
+    {
+      "epoch": 0.10733333333333334,
+      "grad_norm": 37.0,
+      "grad_norm_var": 31.6978515625,
+      "learning_rate": 9.719293114259033e-05,
+      "loss": 7.1846,
+      "loss/crossentropy": 1.9236509799957275,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14150086417794228,
+      "step": 644
+    },
+    {
+      "epoch": 0.1075,
+      "grad_norm": 32.25,
+      "grad_norm_var": 30.61640625,
+      "learning_rate": 9.718427613991848e-05,
+      "loss": 7.3035,
+      "loss/crossentropy": 2.1269415616989136,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18655722215771675,
+      "step": 645
+    },
+    {
+      "epoch": 0.10766666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 30.520833333333332,
+      "learning_rate": 9.717560820140969e-05,
+      "loss": 7.3986,
+      "loss/crossentropy": 2.3212895691394806,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17602985352277756,
+      "step": 646
+    },
+    {
+      "epoch": 0.10783333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 31.180989583333332,
+      "learning_rate": 9.716692732944035e-05,
+      "loss": 7.254,
+      "loss/crossentropy": 2.286354273557663,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17863034084439278,
+      "step": 647
+    },
+    {
+      "epoch": 0.108,
+      "grad_norm": 31.125,
+      "grad_norm_var": 31.273958333333333,
+      "learning_rate": 9.715823352639037e-05,
+      "loss": 7.1363,
+      "loss/crossentropy": 2.284827947616577,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16959453374147415,
+      "step": 648
+    },
+    {
+      "epoch": 0.10816666666666666,
+      "grad_norm": 34.75,
+      "grad_norm_var": 30.741666666666667,
+      "learning_rate": 9.714952679464323e-05,
+      "loss": 7.4843,
+      "loss/crossentropy": 2.267686277627945,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19646358862519264,
+      "step": 649
+    },
+    {
+      "epoch": 0.10833333333333334,
+      "grad_norm": 29.875,
+      "grad_norm_var": 31.6572265625,
+      "learning_rate": 9.71408071365859e-05,
+      "loss": 7.2573,
+      "loss/crossentropy": 1.6790443360805511,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14582885429263115,
+      "step": 650
+    },
+    {
+      "epoch": 0.1085,
+      "grad_norm": 33.75,
+      "grad_norm_var": 31.4744140625,
+      "learning_rate": 9.713207455460894e-05,
+      "loss": 7.2008,
+      "loss/crossentropy": 1.847610518336296,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20972991362214088,
+      "step": 651
+    },
+    {
+      "epoch": 0.10866666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 30.9728515625,
+      "learning_rate": 9.71233290511064e-05,
+      "loss": 7.3211,
+      "loss/crossentropy": 1.8943575620651245,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22945917397737503,
+      "step": 652
+    },
+    {
+      "epoch": 0.10883333333333334,
+      "grad_norm": 34.5,
+      "grad_norm_var": 29.876822916666665,
+      "learning_rate": 9.711457062847595e-05,
+      "loss": 7.396,
+      "loss/crossentropy": 2.315647840499878,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19224681705236435,
+      "step": 653
+    },
+    {
+      "epoch": 0.109,
+      "grad_norm": 35.25,
+      "grad_norm_var": 28.327018229166665,
+      "learning_rate": 9.710579928911876e-05,
+      "loss": 7.4095,
+      "loss/crossentropy": 2.0344275534152985,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22277657315135002,
+      "step": 654
+    },
+    {
+      "epoch": 0.10916666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 28.322330729166666,
+      "learning_rate": 9.709701503543954e-05,
+      "loss": 7.3268,
+      "loss/crossentropy": 1.7724144458770752,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25478213280439377,
+      "step": 655
+    },
+    {
+      "epoch": 0.10933333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 5.382747395833333,
+      "learning_rate": 9.708821786984652e-05,
+      "loss": 7.3553,
+      "loss/crossentropy": 1.7411191165447235,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2752874232828617,
+      "step": 656
+    },
+    {
+      "epoch": 0.1095,
+      "grad_norm": 34.75,
+      "grad_norm_var": 4.021809895833333,
+      "learning_rate": 9.707940779475151e-05,
+      "loss": 7.2225,
+      "loss/crossentropy": 2.1497268676757812,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19079606607556343,
+      "step": 657
+    },
+    {
+      "epoch": 0.10966666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.9749348958333335,
+      "learning_rate": 9.707058481256985e-05,
+      "loss": 7.3662,
+      "loss/crossentropy": 2.0671571493148804,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17754127457737923,
+      "step": 658
+    },
+    {
+      "epoch": 0.10983333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 4.151822916666666,
+      "learning_rate": 9.706174892572039e-05,
+      "loss": 7.1936,
+      "loss/crossentropy": 1.9862104654312134,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21077127382159233,
+      "step": 659
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.0416666666666665,
+      "learning_rate": 9.705290013662556e-05,
+      "loss": 7.4406,
+      "loss/crossentropy": 1.9781703352928162,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19248376414179802,
+      "step": 660
+    },
+    {
+      "epoch": 0.11016666666666666,
+      "grad_norm": 36.5,
+      "grad_norm_var": 3.9580729166666666,
+      "learning_rate": 9.704403844771128e-05,
+      "loss": 7.49,
+      "loss/crossentropy": 1.9751454293727875,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2400129921734333,
+      "step": 661
+    },
+    {
+      "epoch": 0.11033333333333334,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.87265625,
+      "learning_rate": 9.703516386140705e-05,
+      "loss": 7.2844,
+      "loss/crossentropy": 1.5694085359573364,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18795155733823776,
+      "step": 662
+    },
+    {
+      "epoch": 0.1105,
+      "grad_norm": 36.25,
+      "grad_norm_var": 3.640625,
+      "learning_rate": 9.70262763801459e-05,
+      "loss": 7.0861,
+      "loss/crossentropy": 1.470748633146286,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17022135108709335,
+      "step": 663
+    },
+    {
+      "epoch": 0.11066666666666666,
+      "grad_norm": 29.75,
+      "grad_norm_var": 4.165559895833334,
+      "learning_rate": 9.701737600636436e-05,
+      "loss": 7.1957,
+      "loss/crossentropy": 2.195081427693367,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18825620040297508,
+      "step": 664
+    },
+    {
+      "epoch": 0.11083333333333334,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.051497395833334,
+      "learning_rate": 9.700846274250251e-05,
+      "loss": 7.3358,
+      "loss/crossentropy": 1.578473910689354,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17406513914465904,
+      "step": 665
+    },
+    {
+      "epoch": 0.111,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.271875,
+      "learning_rate": 9.699953659100401e-05,
+      "loss": 7.3955,
+      "loss/crossentropy": 1.5875136256217957,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2344185709953308,
+      "step": 666
+    },
+    {
+      "epoch": 0.11116666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.312239583333333,
+      "learning_rate": 9.699059755431598e-05,
+      "loss": 7.3451,
+      "loss/crossentropy": 2.2021476924419403,
+      "loss/hidden": 3.12109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16028309613466263,
+      "step": 667
+    },
+    {
+      "epoch": 0.11133333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.9520833333333334,
+      "learning_rate": 9.698164563488914e-05,
+      "loss": 7.3484,
+      "loss/crossentropy": 2.122020661830902,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1891457512974739,
+      "step": 668
+    },
+    {
+      "epoch": 0.1115,
+      "grad_norm": 31.0,
+      "grad_norm_var": 4.105208333333334,
+      "learning_rate": 9.697268083517767e-05,
+      "loss": 7.1805,
+      "loss/crossentropy": 2.2841413617134094,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18636149913072586,
+      "step": 669
+    },
+    {
+      "epoch": 0.11166666666666666,
+      "grad_norm": 36.0,
+      "grad_norm_var": 4.368489583333333,
+      "learning_rate": 9.696370315763936e-05,
+      "loss": 7.1746,
+      "loss/crossentropy": 1.792080044746399,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19854754582047462,
+      "step": 670
+    },
+    {
+      "epoch": 0.11183333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.36640625,
+      "learning_rate": 9.695471260473545e-05,
+      "loss": 7.2425,
+      "loss/crossentropy": 1.911046177148819,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19043425098061562,
+      "step": 671
+    },
+    {
+      "epoch": 0.112,
+      "grad_norm": 31.875,
+      "grad_norm_var": 4.430143229166666,
+      "learning_rate": 9.69457091789308e-05,
+      "loss": 6.9889,
+      "loss/crossentropy": 1.8222531825304031,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1720294989645481,
+      "step": 672
+    },
+    {
+      "epoch": 0.11216666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 4.213997395833333,
+      "learning_rate": 9.693669288269372e-05,
+      "loss": 7.4069,
+      "loss/crossentropy": 2.1999444365501404,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1844763681292534,
+      "step": 673
+    },
+    {
+      "epoch": 0.11233333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 4.3875,
+      "learning_rate": 9.692766371849606e-05,
+      "loss": 7.1498,
+      "loss/crossentropy": 1.7127140760421753,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19976293668150902,
+      "step": 674
+    },
+    {
+      "epoch": 0.1125,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.5212890625,
+      "learning_rate": 9.691862168881325e-05,
+      "loss": 7.0117,
+      "loss/crossentropy": 1.9354283511638641,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20179572328925133,
+      "step": 675
+    },
+    {
+      "epoch": 0.11266666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.477018229166666,
+      "learning_rate": 9.690956679612421e-05,
+      "loss": 7.2363,
+      "loss/crossentropy": 2.0463519990444183,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2051066905260086,
+      "step": 676
+    },
+    {
+      "epoch": 0.11283333333333333,
+      "grad_norm": 50.0,
+      "grad_norm_var": 22.744205729166666,
+      "learning_rate": 9.690049904291139e-05,
+      "loss": 7.4351,
+      "loss/crossentropy": 1.8881178200244904,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.204398512840271,
+      "step": 677
+    },
+    {
+      "epoch": 0.113,
+      "grad_norm": 37.5,
+      "grad_norm_var": 23.485416666666666,
+      "learning_rate": 9.689141843166074e-05,
+      "loss": 7.2911,
+      "loss/crossentropy": 2.4391735792160034,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18365934491157532,
+      "step": 678
+    },
+    {
+      "epoch": 0.11316666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 23.089322916666667,
+      "learning_rate": 9.688232496486178e-05,
+      "loss": 7.1742,
+      "loss/crossentropy": 2.3844321966171265,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2512863278388977,
+      "step": 679
+    },
+    {
+      "epoch": 0.11333333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 22.210416666666667,
+      "learning_rate": 9.687321864500755e-05,
+      "loss": 7.1451,
+      "loss/crossentropy": 1.5501232147216797,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17280655726790428,
+      "step": 680
+    },
+    {
+      "epoch": 0.1135,
+      "grad_norm": 30.375,
+      "grad_norm_var": 22.9712890625,
+      "learning_rate": 9.686409947459458e-05,
+      "loss": 7.3644,
+      "loss/crossentropy": 1.7302386611700058,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18569517135620117,
+      "step": 681
+    },
+    {
+      "epoch": 0.11366666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 23.120247395833335,
+      "learning_rate": 9.685496745612295e-05,
+      "loss": 7.1641,
+      "loss/crossentropy": 2.097352534532547,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1809883527457714,
+      "step": 682
+    },
+    {
+      "epoch": 0.11383333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 23.294791666666665,
+      "learning_rate": 9.684582259209624e-05,
+      "loss": 7.2062,
+      "loss/crossentropy": 1.916637271642685,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20078307017683983,
+      "step": 683
+    },
+    {
+      "epoch": 0.114,
+      "grad_norm": 32.75,
+      "grad_norm_var": 22.47265625,
+      "learning_rate": 9.683666488502158e-05,
+      "loss": 7.2848,
+      "loss/crossentropy": 1.6298768222332,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21732109412550926,
+      "step": 684
+    },
+    {
+      "epoch": 0.11416666666666667,
+      "grad_norm": 34.5,
+      "grad_norm_var": 21.976822916666666,
+      "learning_rate": 9.682749433740962e-05,
+      "loss": 7.4269,
+      "loss/crossentropy": 2.025507390499115,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.258936308324337,
+      "step": 685
+    },
+    {
+      "epoch": 0.11433333333333333,
+      "grad_norm": 38.0,
+      "grad_norm_var": 22.780989583333334,
+      "learning_rate": 9.68183109517745e-05,
+      "loss": 7.3737,
+      "loss/crossentropy": 1.5819874107837677,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23186847567558289,
+      "step": 686
+    },
+    {
+      "epoch": 0.1145,
+      "grad_norm": 34.25,
+      "grad_norm_var": 22.611458333333335,
+      "learning_rate": 9.68091147306339e-05,
+      "loss": 7.0753,
+      "loss/crossentropy": 1.5637894421815872,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17735859379172325,
+      "step": 687
+    },
+    {
+      "epoch": 0.11466666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 22.348372395833334,
+      "learning_rate": 9.6799905676509e-05,
+      "loss": 7.3931,
+      "loss/crossentropy": 2.0656840205192566,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18083598092198372,
+      "step": 688
+    },
+    {
+      "epoch": 0.11483333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 22.376497395833333,
+      "learning_rate": 9.679068379192456e-05,
+      "loss": 7.3676,
+      "loss/crossentropy": 2.242631584405899,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1938803270459175,
+      "step": 689
+    },
+    {
+      "epoch": 0.115,
+      "grad_norm": 32.75,
+      "grad_norm_var": 21.872916666666665,
+      "learning_rate": 9.678144907940876e-05,
+      "loss": 7.2436,
+      "loss/crossentropy": 2.259352743625641,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19465599209070206,
+      "step": 690
+    },
+    {
+      "epoch": 0.11516666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 21.233072916666668,
+      "learning_rate": 9.677220154149337e-05,
+      "loss": 7.2727,
+      "loss/crossentropy": 1.8682022988796234,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19458742812275887,
+      "step": 691
+    },
+    {
+      "epoch": 0.11533333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 21.868684895833333,
+      "learning_rate": 9.676294118071367e-05,
+      "loss": 7.0673,
+      "loss/crossentropy": 2.093078315258026,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17927749827504158,
+      "step": 692
+    },
+    {
+      "epoch": 0.1155,
+      "grad_norm": 31.375,
+      "grad_norm_var": 4.611458333333333,
+      "learning_rate": 9.675366799960841e-05,
+      "loss": 7.2623,
+      "loss/crossentropy": 1.6467059701681137,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14942791871726513,
+      "step": 693
+    },
+    {
+      "epoch": 0.11566666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.2934895833333333,
+      "learning_rate": 9.674438200071991e-05,
+      "loss": 7.2217,
+      "loss/crossentropy": 2.0415206253528595,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1769421324133873,
+      "step": 694
+    },
+    {
+      "epoch": 0.11583333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 3.2749348958333333,
+      "learning_rate": 9.6735083186594e-05,
+      "loss": 7.2758,
+      "loss/crossentropy": 2.1222026348114014,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19957909733057022,
+      "step": 695
+    },
+    {
+      "epoch": 0.116,
+      "grad_norm": 45.5,
+      "grad_norm_var": 13.442122395833334,
+      "learning_rate": 9.672577155977993e-05,
+      "loss": 7.4773,
+      "loss/crossentropy": 2.405282974243164,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21936316788196564,
+      "step": 696
+    },
+    {
+      "epoch": 0.11616666666666667,
+      "grad_norm": 34.75,
+      "grad_norm_var": 12.801822916666667,
+      "learning_rate": 9.671644712283061e-05,
+      "loss": 7.1548,
+      "loss/crossentropy": 1.4405837059020996,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16677791625261307,
+      "step": 697
+    },
+    {
+      "epoch": 0.11633333333333333,
+      "grad_norm": 37.25,
+      "grad_norm_var": 13.266666666666667,
+      "learning_rate": 9.670710987830233e-05,
+      "loss": 7.3013,
+      "loss/crossentropy": 1.5431812107563019,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18322549387812614,
+      "step": 698
+    },
+    {
+      "epoch": 0.1165,
+      "grad_norm": 31.875,
+      "grad_norm_var": 13.187239583333334,
+      "learning_rate": 9.669775982875501e-05,
+      "loss": 7.1375,
+      "loss/crossentropy": 1.9528156220912933,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1746544986963272,
+      "step": 699
+    },
+    {
+      "epoch": 0.11666666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 13.980143229166666,
+      "learning_rate": 9.668839697675196e-05,
+      "loss": 7.2395,
+      "loss/crossentropy": 1.6688774526119232,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19912731274962425,
+      "step": 700
+    },
+    {
+      "epoch": 0.11683333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 15.91015625,
+      "learning_rate": 9.667902132486009e-05,
+      "loss": 7.2943,
+      "loss/crossentropy": 1.9123640358448029,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2353433296084404,
+      "step": 701
+    },
+    {
+      "epoch": 0.117,
+      "grad_norm": 30.5,
+      "grad_norm_var": 15.03515625,
+      "learning_rate": 9.666963287564979e-05,
+      "loss": 7.189,
+      "loss/crossentropy": 1.7621385157108307,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16866527497768402,
+      "step": 702
+    },
+    {
+      "epoch": 0.11716666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 15.535416666666666,
+      "learning_rate": 9.666023163169493e-05,
+      "loss": 7.0641,
+      "loss/crossentropy": 2.004618912935257,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18865032866597176,
+      "step": 703
+    },
+    {
+      "epoch": 0.11733333333333333,
+      "grad_norm": 36.25,
+      "grad_norm_var": 16.249739583333334,
+      "learning_rate": 9.665081759557295e-05,
+      "loss": 7.2017,
+      "loss/crossentropy": 1.782902330160141,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18442142754793167,
+      "step": 704
+    },
+    {
+      "epoch": 0.1175,
+      "grad_norm": 35.0,
+      "grad_norm_var": 16.48125,
+      "learning_rate": 9.664139076986473e-05,
+      "loss": 7.3359,
+      "loss/crossentropy": 2.1698412895202637,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14698780700564384,
+      "step": 705
+    },
+    {
+      "epoch": 0.11766666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 16.485416666666666,
+      "learning_rate": 9.663195115715471e-05,
+      "loss": 7.3171,
+      "loss/crossentropy": 1.7291880697011948,
+      "loss/hidden": 3.94140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21997908502817154,
+      "step": 706
+    },
+    {
+      "epoch": 0.11783333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 16.595572916666665,
+      "learning_rate": 9.66224987600308e-05,
+      "loss": 7.305,
+      "loss/crossentropy": 1.7600916624069214,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24830835685133934,
+      "step": 707
+    },
+    {
+      "epoch": 0.118,
+      "grad_norm": 34.25,
+      "grad_norm_var": 16.170768229166665,
+      "learning_rate": 9.661303358108445e-05,
+      "loss": 7.2735,
+      "loss/crossentropy": 2.0899295806884766,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18403327465057373,
+      "step": 708
+    },
+    {
+      "epoch": 0.11816666666666667,
+      "grad_norm": 41.0,
+      "grad_norm_var": 19.323958333333334,
+      "learning_rate": 9.660355562291055e-05,
+      "loss": 7.0471,
+      "loss/crossentropy": 1.8371227979660034,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1795869842171669,
+      "step": 709
+    },
+    {
+      "epoch": 0.11833333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 19.387239583333333,
+      "learning_rate": 9.659406488810759e-05,
+      "loss": 7.2137,
+      "loss/crossentropy": 2.0583205223083496,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17614539712667465,
+      "step": 710
+    },
+    {
+      "epoch": 0.1185,
+      "grad_norm": 32.5,
+      "grad_norm_var": 19.156184895833334,
+      "learning_rate": 9.658456137927745e-05,
+      "loss": 7.2021,
+      "loss/crossentropy": 1.9199265837669373,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15964625403285027,
+      "step": 711
+    },
+    {
+      "epoch": 0.11866666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 10.275455729166667,
+      "learning_rate": 9.657504509902562e-05,
+      "loss": 7.3052,
+      "loss/crossentropy": 1.8404278457164764,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17447392269968987,
+      "step": 712
+    },
+    {
+      "epoch": 0.11883333333333333,
+      "grad_norm": 35.0,
+      "grad_norm_var": 10.332747395833334,
+      "learning_rate": 9.656551604996102e-05,
+      "loss": 7.2903,
+      "loss/crossentropy": 2.249763637781143,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20632215961813927,
+      "step": 713
+    },
+    {
+      "epoch": 0.119,
+      "grad_norm": 33.25,
+      "grad_norm_var": 9.153580729166666,
+      "learning_rate": 9.655597423469609e-05,
+      "loss": 7.3072,
+      "loss/crossentropy": 2.0143617391586304,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1969807855784893,
+      "step": 714
+    },
+    {
+      "epoch": 0.11916666666666667,
+      "grad_norm": 36.0,
+      "grad_norm_var": 9.645572916666667,
+      "learning_rate": 9.654641965584678e-05,
+      "loss": 7.1574,
+      "loss/crossentropy": 1.9726352393627167,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20835205540060997,
+      "step": 715
+    },
+    {
+      "epoch": 0.11933333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 9.098893229166666,
+      "learning_rate": 9.653685231603256e-05,
+      "loss": 7.3108,
+      "loss/crossentropy": 2.2339956909418106,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17952323704957962,
+      "step": 716
+    },
+    {
+      "epoch": 0.1195,
+      "grad_norm": 29.0,
+      "grad_norm_var": 8.705989583333333,
+      "learning_rate": 9.652727221787631e-05,
+      "loss": 7.0571,
+      "loss/crossentropy": 1.9769413471221924,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18933235481381416,
+      "step": 717
+    },
+    {
+      "epoch": 0.11966666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 8.1,
+      "learning_rate": 9.65176793640045e-05,
+      "loss": 7.7582,
+      "loss/crossentropy": 2.0757074654102325,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20930679887533188,
+      "step": 718
+    },
+    {
+      "epoch": 0.11983333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 7.282291666666667,
+      "learning_rate": 9.650807375704708e-05,
+      "loss": 7.3997,
+      "loss/crossentropy": 1.870364248752594,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20642438158392906,
+      "step": 719
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 29.25,
+      "grad_norm_var": 8.040625,
+      "learning_rate": 9.649845539963747e-05,
+      "loss": 7.4299,
+      "loss/crossentropy": 1.994875967502594,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16478855349123478,
+      "step": 720
+    },
+    {
+      "epoch": 0.12016666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 8.123372395833334,
+      "learning_rate": 9.648882429441257e-05,
+      "loss": 7.298,
+      "loss/crossentropy": 2.005363404750824,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17638836801052094,
+      "step": 721
+    },
+    {
+      "epoch": 0.12033333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 8.0994140625,
+      "learning_rate": 9.647918044401285e-05,
+      "loss": 7.2297,
+      "loss/crossentropy": 2.101778954267502,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20618782192468643,
+      "step": 722
+    },
+    {
+      "epoch": 0.1205,
+      "grad_norm": 33.25,
+      "grad_norm_var": 7.9822265625,
+      "learning_rate": 9.646952385108218e-05,
+      "loss": 7.2682,
+      "loss/crossentropy": 1.9291025698184967,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2194034680724144,
+      "step": 723
+    },
+    {
+      "epoch": 0.12066666666666667,
+      "grad_norm": 35.0,
+      "grad_norm_var": 8.1291015625,
+      "learning_rate": 9.645985451826803e-05,
+      "loss": 7.3166,
+      "loss/crossentropy": 1.787375569343567,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24283263459801674,
+      "step": 724
+    },
+    {
+      "epoch": 0.12083333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.8244140625,
+      "learning_rate": 9.645017244822123e-05,
+      "loss": 7.3247,
+      "loss/crossentropy": 1.7216472327709198,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17585238814353943,
+      "step": 725
+    },
+    {
+      "epoch": 0.121,
+      "grad_norm": 3556769792.0,
+      "grad_norm_var": 7.906631950160014e+17,
+      "learning_rate": 9.644047764359622e-05,
+      "loss": 8.5727,
+      "loss/crossentropy": 1.454335242509842,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20076533406972885,
+      "step": 726
+    },
+    {
+      "epoch": 0.12116666666666667,
+      "grad_norm": 42.5,
+      "grad_norm_var": 7.90663194719604e+17,
+      "learning_rate": 9.643077010705087e-05,
+      "loss": 7.4779,
+      "loss/crossentropy": 1.6177605986595154,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1870615854859352,
+      "step": 727
+    },
+    {
+      "epoch": 0.12133333333333333,
+      "grad_norm": 36.5,
+      "grad_norm_var": 7.906631945491754e+17,
+      "learning_rate": 9.642104984124656e-05,
+      "loss": 7.076,
+      "loss/crossentropy": 1.951613113284111,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19707249477505684,
+      "step": 728
+    },
+    {
+      "epoch": 0.1215,
+      "grad_norm": 33.0,
+      "grad_norm_var": 7.906631946084549e+17,
+      "learning_rate": 9.641131684884817e-05,
+      "loss": 7.3959,
+      "loss/crossentropy": 2.394118160009384,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22449449822306633,
+      "step": 729
+    },
+    {
+      "epoch": 0.12166666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 7.906631946566195e+17,
+      "learning_rate": 9.640157113252403e-05,
+      "loss": 7.265,
+      "loss/crossentropy": 1.5260854363441467,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16729702427983284,
+      "step": 730
+    },
+    {
+      "epoch": 0.12183333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 7.906631948270481e+17,
+      "learning_rate": 9.6391812694946e-05,
+      "loss": 7.0343,
+      "loss/crossentropy": 1.9055115580558777,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15993406996130943,
+      "step": 731
+    },
+    {
+      "epoch": 0.122,
+      "grad_norm": 36.75,
+      "grad_norm_var": 7.906631947381289e+17,
+      "learning_rate": 9.63820415387894e-05,
+      "loss": 7.1845,
+      "loss/crossentropy": 2.015164703130722,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19183053076267242,
+      "step": 732
+    },
+    {
+      "epoch": 0.12216666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 7.906631946195698e+17,
+      "learning_rate": 9.637225766673307e-05,
+      "loss": 7.2563,
+      "loss/crossentropy": 2.0142166912555695,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17613041400909424,
+      "step": 733
+    },
+    {
+      "epoch": 0.12233333333333334,
+      "grad_norm": 37.25,
+      "grad_norm_var": 7.906631945158308e+17,
+      "learning_rate": 9.636246108145929e-05,
+      "loss": 7.4622,
+      "loss/crossentropy": 1.5445748269557953,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19925610348582268,
+      "step": 734
+    },
+    {
+      "epoch": 0.1225,
+      "grad_norm": 33.5,
+      "grad_norm_var": 7.90663194486191e+17,
+      "learning_rate": 9.635265178565385e-05,
+      "loss": 7.1286,
+      "loss/crossentropy": 1.6356547176837921,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1801084205508232,
+      "step": 735
+    },
+    {
+      "epoch": 0.12266666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 7.906631944454363e+17,
+      "learning_rate": 9.634282978200604e-05,
+      "loss": 7.0731,
+      "loss/crossentropy": 1.4879166930913925,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18008296936750412,
+      "step": 736
+    },
+    {
+      "epoch": 0.12283333333333334,
+      "grad_norm": 38.0,
+      "grad_norm_var": 7.906631942416631e+17,
+      "learning_rate": 9.63329950732086e-05,
+      "loss": 7.2036,
+      "loss/crossentropy": 2.29751256108284,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1693776212632656,
+      "step": 737
+    },
+    {
+      "epoch": 0.123,
+      "grad_norm": 31.625,
+      "grad_norm_var": 7.906631942750077e+17,
+      "learning_rate": 9.632314766195781e-05,
+      "loss": 7.3466,
+      "loss/crossentropy": 2.0535372495651245,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18832634389400482,
+      "step": 738
+    },
+    {
+      "epoch": 0.12316666666666666,
+      "grad_norm": 40.0,
+      "grad_norm_var": 7.906631940749395e+17,
+      "learning_rate": 9.631328755095333e-05,
+      "loss": 7.2628,
+      "loss/crossentropy": 1.8776691854000092,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24790114909410477,
+      "step": 739
+    },
+    {
+      "epoch": 0.12333333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 7.906631941749736e+17,
+      "learning_rate": 9.630341474289842e-05,
+      "loss": 7.0856,
+      "loss/crossentropy": 2.3005475103855133,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18241657316684723,
+      "step": 740
+    },
+    {
+      "epoch": 0.1235,
+      "grad_norm": 36.75,
+      "grad_norm_var": 7.906631940786445e+17,
+      "learning_rate": 9.629352924049975e-05,
+      "loss": 7.391,
+      "loss/crossentropy": 1.8251358270645142,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17540963739156723,
+      "step": 741
+    },
+    {
+      "epoch": 0.12366666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 13.064322916666667,
+      "learning_rate": 9.628363104646747e-05,
+      "loss": 7.4362,
+      "loss/crossentropy": 1.4832358211278915,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22334732487797737,
+      "step": 742
+    },
+    {
+      "epoch": 0.12383333333333334,
+      "grad_norm": 33.75,
+      "grad_norm_var": 8.753125,
+      "learning_rate": 9.627372016351524e-05,
+      "loss": 7.2842,
+      "loss/crossentropy": 1.409521147608757,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18547486886382103,
+      "step": 743
+    },
+    {
+      "epoch": 0.124,
+      "grad_norm": 32.25,
+      "grad_norm_var": 8.55390625,
+      "learning_rate": 9.626379659436017e-05,
+      "loss": 7.1275,
+      "loss/crossentropy": 2.0276291966438293,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15947329252958298,
+      "step": 744
+    },
+    {
+      "epoch": 0.12416666666666666,
+      "grad_norm": 35.0,
+      "grad_norm_var": 8.56640625,
+      "learning_rate": 9.62538603417229e-05,
+      "loss": 6.8613,
+      "loss/crossentropy": 2.1070659160614014,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1781615987420082,
+      "step": 745
+    },
+    {
+      "epoch": 0.12433333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 9.45390625,
+      "learning_rate": 9.624391140832749e-05,
+      "loss": 7.1686,
+      "loss/crossentropy": 1.7076525390148163,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19607660174369812,
+      "step": 746
+    },
+    {
+      "epoch": 0.1245,
+      "grad_norm": 27.875,
+      "grad_norm_var": 10.959309895833334,
+      "learning_rate": 9.623394979690147e-05,
+      "loss": 6.9328,
+      "loss/crossentropy": 1.667178988456726,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1780499890446663,
+      "step": 747
+    },
+    {
+      "epoch": 0.12466666666666666,
+      "grad_norm": 40.75,
+      "grad_norm_var": 13.5634765625,
+      "learning_rate": 9.622397551017592e-05,
+      "loss": 7.3289,
+      "loss/crossentropy": 2.0472422540187836,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25567299500107765,
+      "step": 748
+    },
+    {
+      "epoch": 0.12483333333333334,
+      "grad_norm": 34.0,
+      "grad_norm_var": 13.493684895833333,
+      "learning_rate": 9.62139885508853e-05,
+      "loss": 7.1938,
+      "loss/crossentropy": 1.935056746006012,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18027054890990257,
+      "step": 749
+    },
+    {
+      "epoch": 0.125,
+      "grad_norm": 34.0,
+      "grad_norm_var": 12.769205729166666,
+      "learning_rate": 9.620398892176762e-05,
+      "loss": 7.1023,
+      "loss/crossentropy": 1.9850616455078125,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1702703759074211,
+      "step": 750
+    },
+    {
+      "epoch": 0.12516666666666668,
+      "grad_norm": 32.25,
+      "grad_norm_var": 12.925455729166666,
+      "learning_rate": 9.619397662556435e-05,
+      "loss": 7.0518,
+      "loss/crossentropy": 1.9070358872413635,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18442638218402863,
+      "step": 751
+    },
+    {
+      "epoch": 0.12533333333333332,
+      "grad_norm": 32.25,
+      "grad_norm_var": 12.408333333333333,
+      "learning_rate": 9.618395166502037e-05,
+      "loss": 7.477,
+      "loss/crossentropy": 2.2646956145763397,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17256101593375206,
+      "step": 752
+    },
+    {
+      "epoch": 0.1255,
+      "grad_norm": 31.0,
+      "grad_norm_var": 11.620833333333334,
+      "learning_rate": 9.617391404288412e-05,
+      "loss": 7.3544,
+      "loss/crossentropy": 1.724146842956543,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22952701151371002,
+      "step": 753
+    },
+    {
+      "epoch": 0.12566666666666668,
+      "grad_norm": 33.75,
+      "grad_norm_var": 11.389518229166667,
+      "learning_rate": 9.616386376190745e-05,
+      "loss": 6.9786,
+      "loss/crossentropy": 2.112118750810623,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20843658223748207,
+      "step": 754
+    },
+    {
+      "epoch": 0.12583333333333332,
+      "grad_norm": 47.5,
+      "grad_norm_var": 21.334830729166665,
+      "learning_rate": 9.615380082484571e-05,
+      "loss": 6.9438,
+      "loss/crossentropy": 1.8424568474292755,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17724787816405296,
+      "step": 755
+    },
+    {
+      "epoch": 0.126,
+      "grad_norm": 32.75,
+      "grad_norm_var": 21.051822916666666,
+      "learning_rate": 9.614372523445771e-05,
+      "loss": 7.0792,
+      "loss/crossentropy": 1.963213711977005,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22874099016189575,
+      "step": 756
+    },
+    {
+      "epoch": 0.12616666666666668,
+      "grad_norm": 32.5,
+      "grad_norm_var": 20.684375,
+      "learning_rate": 9.613363699350575e-05,
+      "loss": 7.3902,
+      "loss/crossentropy": 1.9653111547231674,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25649160146713257,
+      "step": 757
+    },
+    {
+      "epoch": 0.12633333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 20.87890625,
+      "learning_rate": 9.612353610475555e-05,
+      "loss": 7.3985,
+      "loss/crossentropy": 2.0834298133850098,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16825050488114357,
+      "step": 758
+    },
+    {
+      "epoch": 0.1265,
+      "grad_norm": 30.625,
+      "grad_norm_var": 21.5087890625,
+      "learning_rate": 9.611342257097632e-05,
+      "loss": 7.2984,
+      "loss/crossentropy": 1.9898212254047394,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18303432315587997,
+      "step": 759
+    },
+    {
+      "epoch": 0.12666666666666668,
+      "grad_norm": 37.0,
+      "grad_norm_var": 22.062955729166667,
+      "learning_rate": 9.610329639494076e-05,
+      "loss": 7.0416,
+      "loss/crossentropy": 1.7550754845142365,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20484184846282005,
+      "step": 760
+    },
+    {
+      "epoch": 0.12683333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 22.019205729166668,
+      "learning_rate": 9.609315757942503e-05,
+      "loss": 7.2014,
+      "loss/crossentropy": 1.6708803474903107,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1901095025241375,
+      "step": 761
+    },
+    {
+      "epoch": 0.127,
+      "grad_norm": 27.875,
+      "grad_norm_var": 23.178580729166665,
+      "learning_rate": 9.608300612720873e-05,
+      "loss": 7.0374,
+      "loss/crossentropy": 1.5478509664535522,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16465886682271957,
+      "step": 762
+    },
+    {
+      "epoch": 0.12716666666666668,
+      "grad_norm": 33.0,
+      "grad_norm_var": 20.864322916666666,
+      "learning_rate": 9.607284204107493e-05,
+      "loss": 7.0719,
+      "loss/crossentropy": 1.9549244046211243,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1916433423757553,
+      "step": 763
+    },
+    {
+      "epoch": 0.12733333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 17.712239583333332,
+      "learning_rate": 9.606266532381018e-05,
+      "loss": 7.0771,
+      "loss/crossentropy": 1.9018628597259521,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17975174635648727,
+      "step": 764
+    },
+    {
+      "epoch": 0.1275,
+      "grad_norm": 36.5,
+      "grad_norm_var": 18.28515625,
+      "learning_rate": 9.605247597820448e-05,
+      "loss": 7.3256,
+      "loss/crossentropy": 2.133163273334503,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2208230309188366,
+      "step": 765
+    },
+    {
+      "epoch": 0.12766666666666668,
+      "grad_norm": 31.25,
+      "grad_norm_var": 18.614583333333332,
+      "learning_rate": 9.604227400705133e-05,
+      "loss": 7.3763,
+      "loss/crossentropy": 1.6012031733989716,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21127447858452797,
+      "step": 766
+    },
+    {
+      "epoch": 0.12783333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 18.835416666666667,
+      "learning_rate": 9.603205941314758e-05,
+      "loss": 7.3093,
+      "loss/crossentropy": 1.6147058308124542,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16550498083233833,
+      "step": 767
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 33.5,
+      "grad_norm_var": 18.745572916666667,
+      "learning_rate": 9.602183219929371e-05,
+      "loss": 7.6064,
+      "loss/crossentropy": 2.2897163033485413,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17588834092020988,
+      "step": 768
+    },
+    {
+      "epoch": 0.12816666666666668,
+      "grad_norm": 30.75,
+      "grad_norm_var": 18.83125,
+      "learning_rate": 9.601159236829352e-05,
+      "loss": 7.3156,
+      "loss/crossentropy": 1.849684089422226,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19435883313417435,
+      "step": 769
+    },
+    {
+      "epoch": 0.12833333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 19.664583333333333,
+      "learning_rate": 9.600133992295433e-05,
+      "loss": 7.2441,
+      "loss/crossentropy": 2.2538318634033203,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19374402984976768,
+      "step": 770
+    },
+    {
+      "epoch": 0.1285,
+      "grad_norm": 34.5,
+      "grad_norm_var": 5.41875,
+      "learning_rate": 9.599107486608689e-05,
+      "loss": 7.2588,
+      "loss/crossentropy": 1.5848983824253082,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17050610482692719,
+      "step": 771
+    },
+    {
+      "epoch": 0.12866666666666668,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.41640625,
+      "learning_rate": 9.598079720050544e-05,
+      "loss": 7.2554,
+      "loss/crossentropy": 1.688281625509262,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22045187279582024,
+      "step": 772
+    },
+    {
+      "epoch": 0.12883333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 5.59140625,
+      "learning_rate": 9.597050692902765e-05,
+      "loss": 7.4084,
+      "loss/crossentropy": 2.116013526916504,
+      "loss/hidden": 3.87109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35107842087745667,
+      "step": 773
+    },
+    {
+      "epoch": 0.129,
+      "grad_norm": 31.75,
+      "grad_norm_var": 5.564583333333333,
+      "learning_rate": 9.596020405447466e-05,
+      "loss": 7.2689,
+      "loss/crossentropy": 1.9616560637950897,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17355723679065704,
+      "step": 774
+    },
+    {
+      "epoch": 0.12916666666666668,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.350455729166667,
+      "learning_rate": 9.594988857967106e-05,
+      "loss": 7.1446,
+      "loss/crossentropy": 1.962799996137619,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1704629361629486,
+      "step": 775
+    },
+    {
+      "epoch": 0.12933333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 4.270572916666667,
+      "learning_rate": 9.593956050744492e-05,
+      "loss": 7.0291,
+      "loss/crossentropy": 2.1874040067195892,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16180355846881866,
+      "step": 776
+    },
+    {
+      "epoch": 0.1295,
+      "grad_norm": 35.0,
+      "grad_norm_var": 4.76640625,
+      "learning_rate": 9.59292198406277e-05,
+      "loss": 7.3201,
+      "loss/crossentropy": 2.0684027671813965,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1866193450987339,
+      "step": 777
+    },
+    {
+      "epoch": 0.12966666666666668,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.450455729166667,
+      "learning_rate": 9.591886658205438e-05,
+      "loss": 7.2141,
+      "loss/crossentropy": 1.798172414302826,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1641489341855049,
+      "step": 778
+    },
+    {
+      "epoch": 0.12983333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.434309895833333,
+      "learning_rate": 9.590850073456336e-05,
+      "loss": 7.3282,
+      "loss/crossentropy": 1.8041448891162872,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21086344867944717,
+      "step": 779
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.434309895833333,
+      "learning_rate": 9.589812230099649e-05,
+      "loss": 7.2079,
+      "loss/crossentropy": 1.9510250985622406,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1650061495602131,
+      "step": 780
+    },
+    {
+      "epoch": 0.13016666666666668,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.3677083333333333,
+      "learning_rate": 9.588773128419906e-05,
+      "loss": 7.1809,
+      "loss/crossentropy": 2.310162901878357,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2582610249519348,
+      "step": 781
+    },
+    {
+      "epoch": 0.13033333333333333,
+      "grad_norm": 36.0,
+      "grad_norm_var": 3.20390625,
+      "learning_rate": 9.587732768701986e-05,
+      "loss": 6.9808,
+      "loss/crossentropy": 1.50637586414814,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16680185496807098,
+      "step": 782
+    },
+    {
+      "epoch": 0.1305,
+      "grad_norm": 35.0,
+      "grad_norm_var": 3.48125,
+      "learning_rate": 9.586691151231107e-05,
+      "loss": 7.1922,
+      "loss/crossentropy": 1.9600141942501068,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19956029206514359,
+      "step": 783
+    },
+    {
+      "epoch": 0.13066666666666665,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.686393229166667,
+      "learning_rate": 9.585648276292836e-05,
+      "loss": 7.3053,
+      "loss/crossentropy": 1.8970597684383392,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18671169504523277,
+      "step": 784
+    },
+    {
+      "epoch": 0.13083333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.783072916666667,
+      "learning_rate": 9.584604144173083e-05,
+      "loss": 7.3104,
+      "loss/crossentropy": 1.6864672750234604,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17203770950436592,
+      "step": 785
+    },
+    {
+      "epoch": 0.131,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.294791666666667,
+      "learning_rate": 9.5835587551581e-05,
+      "loss": 6.9471,
+      "loss/crossentropy": 1.7447805106639862,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1637143399566412,
+      "step": 786
+    },
+    {
+      "epoch": 0.13116666666666665,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.283072916666667,
+      "learning_rate": 9.58251210953449e-05,
+      "loss": 7.1848,
+      "loss/crossentropy": 1.606904461979866,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16485954821109772,
+      "step": 787
+    },
+    {
+      "epoch": 0.13133333333333333,
+      "grad_norm": 36.5,
+      "grad_norm_var": 4.258072916666666,
+      "learning_rate": 9.581464207589195e-05,
+      "loss": 7.1967,
+      "loss/crossentropy": 2.123050779104233,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18045473843812943,
+      "step": 788
+    },
+    {
+      "epoch": 0.1315,
+      "grad_norm": 34.75,
+      "grad_norm_var": 4.416666666666667,
+      "learning_rate": 9.580415049609503e-05,
+      "loss": 7.3665,
+      "loss/crossentropy": 2.0198948085308075,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1940045841038227,
+      "step": 789
+    },
+    {
+      "epoch": 0.13166666666666665,
+      "grad_norm": 38.25,
+      "grad_norm_var": 6.136458333333334,
+      "learning_rate": 9.579364635883048e-05,
+      "loss": 7.3307,
+      "loss/crossentropy": 2.293046534061432,
+      "loss/hidden": 3.08984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1891222782433033,
+      "step": 790
+    },
+    {
+      "epoch": 0.13183333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 6.049739583333333,
+      "learning_rate": 9.578312966697807e-05,
+      "loss": 7.081,
+      "loss/crossentropy": 2.1809073388576508,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18149369955062866,
+      "step": 791
+    },
+    {
+      "epoch": 0.132,
+      "grad_norm": 30.75,
+      "grad_norm_var": 5.702018229166667,
+      "learning_rate": 9.577260042342097e-05,
+      "loss": 7.2362,
+      "loss/crossentropy": 1.84629625082016,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1877918615937233,
+      "step": 792
+    },
+    {
+      "epoch": 0.13216666666666665,
+      "grad_norm": 29.25,
+      "grad_norm_var": 6.4806640625,
+      "learning_rate": 9.576205863104588e-05,
+      "loss": 7.4503,
+      "loss/crossentropy": 1.9817317426204681,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18423109501600266,
+      "step": 793
+    },
+    {
+      "epoch": 0.13233333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 7.123893229166667,
+      "learning_rate": 9.575150429274287e-05,
+      "loss": 7.3884,
+      "loss/crossentropy": 2.3506908416748047,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1764533445239067,
+      "step": 794
+    },
+    {
+      "epoch": 0.1325,
+      "grad_norm": 33.75,
+      "grad_norm_var": 7.178580729166667,
+      "learning_rate": 9.574093741140549e-05,
+      "loss": 6.9683,
+      "loss/crossentropy": 1.5706563293933868,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17185845598578453,
+      "step": 795
+    },
+    {
+      "epoch": 0.13266666666666665,
+      "grad_norm": 37.25,
+      "grad_norm_var": 8.350455729166667,
+      "learning_rate": 9.573035798993069e-05,
+      "loss": 7.2168,
+      "loss/crossentropy": 1.1439642012119293,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17750319838523865,
+      "step": 796
+    },
+    {
+      "epoch": 0.13283333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 8.09765625,
+      "learning_rate": 9.571976603121888e-05,
+      "loss": 7.1198,
+      "loss/crossentropy": 2.444544196128845,
+      "loss/hidden": 3.14453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16872760653495789,
+      "step": 797
+    },
+    {
+      "epoch": 0.133,
+      "grad_norm": 36.0,
+      "grad_norm_var": 8.09765625,
+      "learning_rate": 9.570916153817391e-05,
+      "loss": 7.2076,
+      "loss/crossentropy": 2.22321480512619,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.170970369130373,
+      "step": 798
+    },
+    {
+      "epoch": 0.13316666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 8.0150390625,
+      "learning_rate": 9.569854451370307e-05,
+      "loss": 7.0286,
+      "loss/crossentropy": 2.003016710281372,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19756201654672623,
+      "step": 799
+    },
+    {
+      "epoch": 0.13333333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 7.90390625,
+      "learning_rate": 9.568791496071706e-05,
+      "loss": 7.2873,
+      "loss/crossentropy": 1.6287779659032822,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20879724621772766,
+      "step": 800
+    },
+    {
+      "epoch": 0.1335,
+      "grad_norm": 30.75,
+      "grad_norm_var": 7.7791015625,
+      "learning_rate": 9.567727288213005e-05,
+      "loss": 7.1588,
+      "loss/crossentropy": 1.7264407873153687,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18642090260982513,
+      "step": 801
+    },
+    {
+      "epoch": 0.13366666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 7.805143229166666,
+      "learning_rate": 9.56666182808596e-05,
+      "loss": 7.1695,
+      "loss/crossentropy": 2.137233257293701,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18422173708677292,
+      "step": 802
+    },
+    {
+      "epoch": 0.13383333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 7.843489583333334,
+      "learning_rate": 9.565595115982678e-05,
+      "loss": 7.5146,
+      "loss/crossentropy": 1.9141822457313538,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20264451205730438,
+      "step": 803
+    },
+    {
+      "epoch": 0.134,
+      "grad_norm": 30.125,
+      "grad_norm_var": 7.395247395833334,
+      "learning_rate": 9.5645271521956e-05,
+      "loss": 7.0942,
+      "loss/crossentropy": 1.2032550573349,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15775632485747337,
+      "step": 804
+    },
+    {
+      "epoch": 0.13416666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 7.857291666666667,
+      "learning_rate": 9.563457937017515e-05,
+      "loss": 6.9968,
+      "loss/crossentropy": 1.5480999499559402,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14649275690317154,
+      "step": 805
+    },
+    {
+      "epoch": 0.13433333333333333,
+      "grad_norm": 36.5,
+      "grad_norm_var": 6.64140625,
+      "learning_rate": 9.562387470741554e-05,
+      "loss": 7.1829,
+      "loss/crossentropy": 1.913821130990982,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21699776872992516,
+      "step": 806
+    },
+    {
+      "epoch": 0.1345,
+      "grad_norm": 32.5,
+      "grad_norm_var": 6.623958333333333,
+      "learning_rate": 9.561315753661194e-05,
+      "loss": 7.3507,
+      "loss/crossentropy": 2.378206968307495,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17058103904128075,
+      "step": 807
+    },
+    {
+      "epoch": 0.13466666666666666,
+      "grad_norm": 37.5,
+      "grad_norm_var": 8.262239583333333,
+      "learning_rate": 9.560242786070249e-05,
+      "loss": 7.1721,
+      "loss/crossentropy": 2.3090766072273254,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20279062166810036,
+      "step": 808
+    },
+    {
+      "epoch": 0.13483333333333333,
+      "grad_norm": 36.0,
+      "grad_norm_var": 8.170833333333333,
+      "learning_rate": 9.55916856826288e-05,
+      "loss": 7.158,
+      "loss/crossentropy": 2.150182008743286,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16158411651849747,
+      "step": 809
+    },
+    {
+      "epoch": 0.135,
+      "grad_norm": 27.75,
+      "grad_norm_var": 9.270833333333334,
+      "learning_rate": 9.558093100533591e-05,
+      "loss": 6.866,
+      "loss/crossentropy": 1.5633295774459839,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15401308983564377,
+      "step": 810
+    },
+    {
+      "epoch": 0.13516666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 9.598958333333334,
+      "learning_rate": 9.557016383177227e-05,
+      "loss": 7.4332,
+      "loss/crossentropy": 1.5982379913330078,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15583708882331848,
+      "step": 811
+    },
+    {
+      "epoch": 0.13533333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 8.089322916666667,
+      "learning_rate": 9.555938416488977e-05,
+      "loss": 7.3273,
+      "loss/crossentropy": 1.6491760313510895,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16479353606700897,
+      "step": 812
+    },
+    {
+      "epoch": 0.1355,
+      "grad_norm": 47.75,
+      "grad_norm_var": 22.973958333333332,
+      "learning_rate": 9.55485920076437e-05,
+      "loss": 7.2593,
+      "loss/crossentropy": 1.887940376996994,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18628187850117683,
+      "step": 813
+    },
+    {
+      "epoch": 0.13566666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 22.542122395833335,
+      "learning_rate": 9.553778736299279e-05,
+      "loss": 7.2753,
+      "loss/crossentropy": 1.8376013785600662,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17566929385066032,
+      "step": 814
+    },
+    {
+      "epoch": 0.13583333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 22.483333333333334,
+      "learning_rate": 9.552697023389922e-05,
+      "loss": 7.1116,
+      "loss/crossentropy": 2.0512495040893555,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17322762310504913,
+      "step": 815
+    },
+    {
+      "epoch": 0.136,
+      "grad_norm": 32.0,
+      "grad_norm_var": 22.254166666666666,
+      "learning_rate": 9.551614062332856e-05,
+      "loss": 7.2802,
+      "loss/crossentropy": 2.209246516227722,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1635245606303215,
+      "step": 816
+    },
+    {
+      "epoch": 0.13616666666666666,
+      "grad_norm": 28.125,
+      "grad_norm_var": 23.559830729166666,
+      "learning_rate": 9.550529853424979e-05,
+      "loss": 7.1244,
+      "loss/crossentropy": 1.5150603652000427,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16150140762329102,
+      "step": 817
+    },
+    {
+      "epoch": 0.13633333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 23.705208333333335,
+      "learning_rate": 9.549444396963534e-05,
+      "loss": 7.1604,
+      "loss/crossentropy": 1.7344059646129608,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19508787989616394,
+      "step": 818
+    },
+    {
+      "epoch": 0.1365,
+      "grad_norm": 30.375,
+      "grad_norm_var": 23.789322916666666,
+      "learning_rate": 9.548357693246105e-05,
+      "loss": 7.312,
+      "loss/crossentropy": 1.9685882925987244,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19253920391201973,
+      "step": 819
+    },
+    {
+      "epoch": 0.13666666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 23.612239583333334,
+      "learning_rate": 9.547269742570619e-05,
+      "loss": 7.4232,
+      "loss/crossentropy": 1.7538487613201141,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.193400289863348,
+      "step": 820
+    },
+    {
+      "epoch": 0.13683333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 22.484309895833334,
+      "learning_rate": 9.546180545235344e-05,
+      "loss": 7.1951,
+      "loss/crossentropy": 1.9168739020824432,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21507158875465393,
+      "step": 821
+    },
+    {
+      "epoch": 0.137,
+      "grad_norm": 31.875,
+      "grad_norm_var": 21.812239583333334,
+      "learning_rate": 9.545090101538887e-05,
+      "loss": 7.2019,
+      "loss/crossentropy": 2.167972207069397,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16965579986572266,
+      "step": 822
+    },
+    {
+      "epoch": 0.13716666666666666,
+      "grad_norm": 35.25,
+      "grad_norm_var": 22.11875,
+      "learning_rate": 9.543998411780201e-05,
+      "loss": 7.2713,
+      "loss/crossentropy": 2.357577830553055,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2042151391506195,
+      "step": 823
+    },
+    {
+      "epoch": 0.13733333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 20.77890625,
+      "learning_rate": 9.54290547625858e-05,
+      "loss": 7.2059,
+      "loss/crossentropy": 2.018874019384384,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1911584846675396,
+      "step": 824
+    },
+    {
+      "epoch": 0.1375,
+      "grad_norm": 32.5,
+      "grad_norm_var": 20.049739583333334,
+      "learning_rate": 9.541811295273656e-05,
+      "loss": 7.5932,
+      "loss/crossentropy": 2.1227717101573944,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15998363122344017,
+      "step": 825
+    },
+    {
+      "epoch": 0.13766666666666666,
+      "grad_norm": 34.25,
+      "grad_norm_var": 18.505989583333335,
+      "learning_rate": 9.540715869125407e-05,
+      "loss": 7.2702,
+      "loss/crossentropy": 1.5410704165697098,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14585284516215324,
+      "step": 826
+    },
+    {
+      "epoch": 0.13783333333333334,
+      "grad_norm": 28.75,
+      "grad_norm_var": 19.193489583333335,
+      "learning_rate": 9.53961919811415e-05,
+      "loss": 7.1976,
+      "loss/crossentropy": 1.3984507620334625,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20053860172629356,
+      "step": 827
+    },
+    {
+      "epoch": 0.138,
+      "grad_norm": 30.75,
+      "grad_norm_var": 19.477083333333333,
+      "learning_rate": 9.538521282540542e-05,
+      "loss": 7.3064,
+      "loss/crossentropy": 1.939309448003769,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18332596495747566,
+      "step": 828
+    },
+    {
+      "epoch": 0.13816666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.6968098958333333,
+      "learning_rate": 9.537422122705585e-05,
+      "loss": 7.3088,
+      "loss/crossentropy": 2.401898682117462,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17688017711043358,
+      "step": 829
+    },
+    {
+      "epoch": 0.13833333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.812239583333333,
+      "learning_rate": 9.536321718910619e-05,
+      "loss": 6.9041,
+      "loss/crossentropy": 1.9290994107723236,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1785241961479187,
+      "step": 830
+    },
+    {
+      "epoch": 0.1385,
+      "grad_norm": 35.5,
+      "grad_norm_var": 4.334375,
+      "learning_rate": 9.535220071457325e-05,
+      "loss": 7.4805,
+      "loss/crossentropy": 2.228693962097168,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15913259238004684,
+      "step": 831
+    },
+    {
+      "epoch": 0.13866666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.426041666666666,
+      "learning_rate": 9.534117180647728e-05,
+      "loss": 7.2876,
+      "loss/crossentropy": 2.1069608330726624,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18579137325286865,
+      "step": 832
+    },
+    {
+      "epoch": 0.13883333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.444205729166667,
+      "learning_rate": 9.533013046784189e-05,
+      "loss": 7.2957,
+      "loss/crossentropy": 2.0110799372196198,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17723548784852028,
+      "step": 833
+    },
+    {
+      "epoch": 0.139,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.6322916666666667,
+      "learning_rate": 9.531907670169415e-05,
+      "loss": 7.1961,
+      "loss/crossentropy": 2.0143231451511383,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28889353573322296,
+      "step": 834
+    },
+    {
+      "epoch": 0.13916666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.69140625,
+      "learning_rate": 9.530801051106449e-05,
+      "loss": 7.1965,
+      "loss/crossentropy": 1.8046370297670364,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18984964862465858,
+      "step": 835
+    },
+    {
+      "epoch": 0.13933333333333334,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.715559895833333,
+      "learning_rate": 9.52969318989868e-05,
+      "loss": 7.3937,
+      "loss/crossentropy": 1.8575055748224258,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18653538450598717,
+      "step": 836
+    },
+    {
+      "epoch": 0.1395,
+      "grad_norm": 35.0,
+      "grad_norm_var": 4.274934895833334,
+      "learning_rate": 9.528584086849832e-05,
+      "loss": 7.5886,
+      "loss/crossentropy": 2.180445581674576,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1793573908507824,
+      "step": 837
+    },
+    {
+      "epoch": 0.13966666666666666,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.402083333333334,
+      "learning_rate": 9.527473742263973e-05,
+      "loss": 7.1074,
+      "loss/crossentropy": 1.3522489666938782,
+      "loss/hidden": 3.78515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2427074834704399,
+      "step": 838
+    },
+    {
+      "epoch": 0.13983333333333334,
+      "grad_norm": 34.5,
+      "grad_norm_var": 4.124739583333334,
+      "learning_rate": 9.526362156445507e-05,
+      "loss": 7.2462,
+      "loss/crossentropy": 1.5952425301074982,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17379407957196236,
+      "step": 839
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.575455729166666,
+      "learning_rate": 9.525249329699188e-05,
+      "loss": 7.2486,
+      "loss/crossentropy": 1.8740590512752533,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21034130454063416,
+      "step": 840
+    },
+    {
+      "epoch": 0.14016666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 4.897330729166667,
+      "learning_rate": 9.524135262330098e-05,
+      "loss": 7.128,
+      "loss/crossentropy": 1.7866149991750717,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1871703304350376,
+      "step": 841
+    },
+    {
+      "epoch": 0.14033333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.571809895833334,
+      "learning_rate": 9.523019954643669e-05,
+      "loss": 7.0836,
+      "loss/crossentropy": 1.6441450268030167,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16951755061745644,
+      "step": 842
+    },
+    {
+      "epoch": 0.1405,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.276497395833333,
+      "learning_rate": 9.521903406945664e-05,
+      "loss": 7.2593,
+      "loss/crossentropy": 1.6699831187725067,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20064623653888702,
+      "step": 843
+    },
+    {
+      "epoch": 0.14066666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.174934895833333,
+      "learning_rate": 9.520785619542196e-05,
+      "loss": 7.3072,
+      "loss/crossentropy": 1.8868179321289062,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16898546740412712,
+      "step": 844
+    },
+    {
+      "epoch": 0.14083333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 4.139583333333333,
+      "learning_rate": 9.519666592739709e-05,
+      "loss": 7.272,
+      "loss/crossentropy": 1.7591368407011032,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1614004783332348,
+      "step": 845
+    },
+    {
+      "epoch": 0.141,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.156184895833333,
+      "learning_rate": 9.518546326844993e-05,
+      "loss": 7.3079,
+      "loss/crossentropy": 1.8068215101957321,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19330871850252151,
+      "step": 846
+    },
+    {
+      "epoch": 0.14116666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.428580729166667,
+      "learning_rate": 9.517424822165175e-05,
+      "loss": 7.5827,
+      "loss/crossentropy": 2.0385962426662445,
+      "loss/hidden": 3.7890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1921292021870613,
+      "step": 847
+    },
+    {
+      "epoch": 0.14133333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.479622395833333,
+      "learning_rate": 9.516302079007719e-05,
+      "loss": 7.2749,
+      "loss/crossentropy": 1.9993867874145508,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1716153398156166,
+      "step": 848
+    },
+    {
+      "epoch": 0.1415,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.5197265625,
+      "learning_rate": 9.515178097680437e-05,
+      "loss": 6.9822,
+      "loss/crossentropy": 1.8454234600067139,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15885218605399132,
+      "step": 849
+    },
+    {
+      "epoch": 0.14166666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.46640625,
+      "learning_rate": 9.51405287849147e-05,
+      "loss": 7.4828,
+      "loss/crossentropy": 2.2536956071853638,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19160786271095276,
+      "step": 850
+    },
+    {
+      "epoch": 0.14183333333333334,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.423958333333333,
+      "learning_rate": 9.512926421749304e-05,
+      "loss": 7.3405,
+      "loss/crossentropy": 1.9627227783203125,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17236102744936943,
+      "step": 851
+    },
+    {
+      "epoch": 0.142,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.442122395833333,
+      "learning_rate": 9.511798727762764e-05,
+      "loss": 7.1229,
+      "loss/crossentropy": 2.0343536138534546,
+      "loss/hidden": 3.66015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.209948617964983,
+      "step": 852
+    },
+    {
+      "epoch": 0.14216666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.8223307291666666,
+      "learning_rate": 9.510669796841014e-05,
+      "loss": 7.0988,
+      "loss/crossentropy": 1.7543722093105316,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17832380533218384,
+      "step": 853
+    },
+    {
+      "epoch": 0.14233333333333334,
+      "grad_norm": 37.25,
+      "grad_norm_var": 4.887434895833334,
+      "learning_rate": 9.509539629293558e-05,
+      "loss": 7.25,
+      "loss/crossentropy": 1.3975646048784256,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19633839651942253,
+      "step": 854
+    },
+    {
+      "epoch": 0.1425,
+      "grad_norm": 38.5,
+      "grad_norm_var": 7.3166015625,
+      "learning_rate": 9.508408225430237e-05,
+      "loss": 7.2774,
+      "loss/crossentropy": 2.2165739834308624,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21087035164237022,
+      "step": 855
+    },
+    {
+      "epoch": 0.14266666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 6.83515625,
+      "learning_rate": 9.507275585561229e-05,
+      "loss": 6.9727,
+      "loss/crossentropy": 1.4398342072963715,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16120069846510887,
+      "step": 856
+    },
+    {
+      "epoch": 0.14283333333333334,
+      "grad_norm": 38.25,
+      "grad_norm_var": 8.357291666666667,
+      "learning_rate": 9.506141709997057e-05,
+      "loss": 7.0002,
+      "loss/crossentropy": 1.8449595719575882,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20784858986735344,
+      "step": 857
+    },
+    {
+      "epoch": 0.143,
+      "grad_norm": 33.25,
+      "grad_norm_var": 8.36640625,
+      "learning_rate": 9.505006599048579e-05,
+      "loss": 7.2836,
+      "loss/crossentropy": 2.059192806482315,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1859123781323433,
+      "step": 858
+    },
+    {
+      "epoch": 0.14316666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 8.312239583333334,
+      "learning_rate": 9.503870253026991e-05,
+      "loss": 7.1371,
+      "loss/crossentropy": 2.141940951347351,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1802339181303978,
+      "step": 859
+    },
+    {
+      "epoch": 0.14333333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 8.549739583333333,
+      "learning_rate": 9.50273267224383e-05,
+      "loss": 7.3665,
+      "loss/crossentropy": 2.0429071485996246,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17789635062217712,
+      "step": 860
+    },
+    {
+      "epoch": 0.1435,
+      "grad_norm": 30.125,
+      "grad_norm_var": 8.506705729166667,
+      "learning_rate": 9.501593857010969e-05,
+      "loss": 7.3528,
+      "loss/crossentropy": 1.8638008832931519,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2563036270439625,
+      "step": 861
+    },
+    {
+      "epoch": 0.14366666666666666,
+      "grad_norm": 35.25,
+      "grad_norm_var": 8.668489583333333,
+      "learning_rate": 9.50045380764062e-05,
+      "loss": 7.4546,
+      "loss/crossentropy": 2.127779573202133,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2628929652273655,
+      "step": 862
+    },
+    {
+      "epoch": 0.14383333333333334,
+      "grad_norm": 3875536896.0,
+      "grad_norm_var": 9.387366234729743e+17,
+      "learning_rate": 9.499312524445336e-05,
+      "loss": 8.1199,
+      "loss/crossentropy": 2.3370486199855804,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17788996547460556,
+      "step": 863
+    },
+    {
+      "epoch": 0.144,
+      "grad_norm": 43.0,
+      "grad_norm_var": 9.387366231580869e+17,
+      "learning_rate": 9.498170007738005e-05,
+      "loss": 6.8451,
+      "loss/crossentropy": 2.1375666558742523,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17391236871480942,
+      "step": 864
+    },
+    {
+      "epoch": 0.14416666666666667,
+      "grad_norm": 35.0,
+      "grad_norm_var": 9.387366230127543e+17,
+      "learning_rate": 9.497026257831855e-05,
+      "loss": 7.4045,
+      "loss/crossentropy": 1.5083353966474533,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18334266915917397,
+      "step": 865
+    },
+    {
+      "epoch": 0.14433333333333334,
+      "grad_norm": 33.75,
+      "grad_norm_var": 9.387366229118289e+17,
+      "learning_rate": 9.495881275040453e-05,
+      "loss": 7.363,
+      "loss/crossentropy": 2.0846097469329834,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.176199559122324,
+      "step": 866
+    },
+    {
+      "epoch": 0.1445,
+      "grad_norm": 30.625,
+      "grad_norm_var": 9.387366229037549e+17,
+      "learning_rate": 9.494735059677699e-05,
+      "loss": 7.4368,
+      "loss/crossentropy": 1.7406192421913147,
+      "loss/hidden": 3.8046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26504576206207275,
+      "step": 867
+    },
+    {
+      "epoch": 0.14466666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 9.387366228835698e+17,
+      "learning_rate": 9.493587612057837e-05,
+      "loss": 7.3233,
+      "loss/crossentropy": 2.186119467020035,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18371190875768661,
+      "step": 868
+    },
+    {
+      "epoch": 0.14483333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 9.387366228997178e+17,
+      "learning_rate": 9.492438932495444e-05,
+      "loss": 7.2279,
+      "loss/crossentropy": 2.286859691143036,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1991770826280117,
+      "step": 869
+    },
+    {
+      "epoch": 0.145,
+      "grad_norm": 32.0,
+      "grad_norm_var": 9.387366230692726e+17,
+      "learning_rate": 9.491289021305441e-05,
+      "loss": 7.5501,
+      "loss/crossentropy": 2.273844063282013,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21323252469301224,
+      "step": 870
+    },
+    {
+      "epoch": 0.14516666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 9.387366233841599e+17,
+      "learning_rate": 9.490137878803079e-05,
+      "loss": 7.0821,
+      "loss/crossentropy": 1.7654433399438858,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1789274625480175,
+      "step": 871
+    },
+    {
+      "epoch": 0.14533333333333334,
+      "grad_norm": 34.5,
+      "grad_norm_var": 9.387366233357157e+17,
+      "learning_rate": 9.488985505303951e-05,
+      "loss": 7.1429,
+      "loss/crossentropy": 2.2276743054389954,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1722644343972206,
+      "step": 872
+    },
+    {
+      "epoch": 0.1455,
+      "grad_norm": 31.125,
+      "grad_norm_var": 9.387366235658257e+17,
+      "learning_rate": 9.487831901123988e-05,
+      "loss": 7.2318,
+      "loss/crossentropy": 1.354034349322319,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22395050898194313,
+      "step": 873
+    },
+    {
+      "epoch": 0.14566666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 9.387366236506031e+17,
+      "learning_rate": 9.486677066579456e-05,
+      "loss": 7.0956,
+      "loss/crossentropy": 1.8272830247879028,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22604960203170776,
+      "step": 874
+    },
+    {
+      "epoch": 0.14583333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 9.387366237071213e+17,
+      "learning_rate": 9.485521001986962e-05,
+      "loss": 7.4001,
+      "loss/crossentropy": 2.2510926127433777,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17866208404302597,
+      "step": 875
+    },
+    {
+      "epoch": 0.146,
+      "grad_norm": 29.875,
+      "grad_norm_var": 9.387366237353805e+17,
+      "learning_rate": 9.484363707663442e-05,
+      "loss": 7.1694,
+      "loss/crossentropy": 2.0665951669216156,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20089091360569,
+      "step": 876
+    },
+    {
+      "epoch": 0.14616666666666667,
+      "grad_norm": 27.625,
+      "grad_norm_var": 9.387366238161208e+17,
+      "learning_rate": 9.483205183926181e-05,
+      "loss": 7.2439,
+      "loss/crossentropy": 2.1436045169830322,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19738992303609848,
+      "step": 877
+    },
+    {
+      "epoch": 0.14633333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 9.387366239574164e+17,
+      "learning_rate": 9.48204543109279e-05,
+      "loss": 7.0731,
+      "loss/crossentropy": 1.8049315214157104,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20513951405882835,
+      "step": 878
+    },
+    {
+      "epoch": 0.1465,
+      "grad_norm": 34.75,
+      "grad_norm_var": 12.684375,
+      "learning_rate": 9.480884449481225e-05,
+      "loss": 7.4375,
+      "loss/crossentropy": 2.4671058654785156,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20915862172842026,
+      "step": 879
+    },
+    {
+      "epoch": 0.14666666666666667,
+      "grad_norm": 50.0,
+      "grad_norm_var": 25.751041666666666,
+      "learning_rate": 9.479722239409775e-05,
+      "loss": 7.3419,
+      "loss/crossentropy": 1.6614379286766052,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18616421520709991,
+      "step": 880
+    },
+    {
+      "epoch": 0.14683333333333334,
+      "grad_norm": 33.75,
+      "grad_norm_var": 25.468489583333334,
+      "learning_rate": 9.478558801197065e-05,
+      "loss": 7.2587,
+      "loss/crossentropy": 2.0950733423233032,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16770224273204803,
+      "step": 881
+    },
+    {
+      "epoch": 0.147,
+      "grad_norm": 30.125,
+      "grad_norm_var": 25.753580729166668,
+      "learning_rate": 9.47739413516206e-05,
+      "loss": 6.9264,
+      "loss/crossentropy": 2.3005232512950897,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19224124774336815,
+      "step": 882
+    },
+    {
+      "epoch": 0.14716666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 26.1494140625,
+      "learning_rate": 9.476228241624059e-05,
+      "loss": 7.4149,
+      "loss/crossentropy": 2.254081070423126,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1655355878174305,
+      "step": 883
+    },
+    {
+      "epoch": 0.14733333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 26.17265625,
+      "learning_rate": 9.475061120902698e-05,
+      "loss": 7.1466,
+      "loss/crossentropy": 2.30178964138031,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20386098325252533,
+      "step": 884
+    },
+    {
+      "epoch": 0.1475,
+      "grad_norm": 30.75,
+      "grad_norm_var": 26.31015625,
+      "learning_rate": 9.473892773317952e-05,
+      "loss": 7.2337,
+      "loss/crossentropy": 1.8351454138755798,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1995384581387043,
+      "step": 885
+    },
+    {
+      "epoch": 0.14766666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 26.454166666666666,
+      "learning_rate": 9.472723199190125e-05,
+      "loss": 7.3051,
+      "loss/crossentropy": 1.2930150479078293,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16403597220778465,
+      "step": 886
+    },
+    {
+      "epoch": 0.14783333333333334,
+      "grad_norm": 2902458368.0,
+      "grad_norm_var": 5.2651652431394906e+17,
+      "learning_rate": 9.47155239883987e-05,
+      "loss": 8.8761,
+      "loss/crossentropy": 1.4680196046829224,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2003433182835579,
+      "step": 887
+    },
+    {
+      "epoch": 0.148,
+      "grad_norm": 38.5,
+      "grad_norm_var": 5.2651652421720045e+17,
+      "learning_rate": 9.470380372588162e-05,
+      "loss": 7.361,
+      "loss/crossentropy": 1.9177829772233963,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17988650873303413,
+      "step": 888
+    },
+    {
+      "epoch": 0.14816666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 5.265165241537092e+17,
+      "learning_rate": 9.46920712075632e-05,
+      "loss": 7.3953,
+      "loss/crossentropy": 1.8802359998226166,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19930091500282288,
+      "step": 889
+    },
+    {
+      "epoch": 0.14833333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.2651652412045184e+17,
+      "learning_rate": 9.468032643665998e-05,
+      "loss": 7.5635,
+      "loss/crossentropy": 2.278567373752594,
+      "loss/hidden": 3.76171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22147762402892113,
+      "step": 890
+    },
+    {
+      "epoch": 0.1485,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.2651652405393715e+17,
+      "learning_rate": 9.466856941639188e-05,
+      "loss": 7.3267,
+      "loss/crossentropy": 2.0632762610912323,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18990402668714523,
+      "step": 891
+    },
+    {
+      "epoch": 0.14866666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 5.2651652401765645e+17,
+      "learning_rate": 9.465680014998213e-05,
+      "loss": 7.1451,
+      "loss/crossentropy": 1.4889424741268158,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19683867692947388,
+      "step": 892
+    },
+    {
+      "epoch": 0.14883333333333335,
+      "grad_norm": 31.625,
+      "grad_norm_var": 5.2651652392090784e+17,
+      "learning_rate": 9.464501864065735e-05,
+      "loss": 7.2155,
+      "loss/crossentropy": 1.9928219020366669,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17085601761937141,
+      "step": 893
+    },
+    {
+      "epoch": 0.149,
+      "grad_norm": 29.25,
+      "grad_norm_var": 5.26516523960212e+17,
+      "learning_rate": 9.46332248916475e-05,
+      "loss": 7.2801,
+      "loss/crossentropy": 1.4565714597702026,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1648852750658989,
+      "step": 894
+    },
+    {
+      "epoch": 0.14916666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 5.265165240357968e+17,
+      "learning_rate": 9.46214189061859e-05,
+      "loss": 7.1637,
+      "loss/crossentropy": 2.0909490883350372,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1700812503695488,
+      "step": 895
+    },
+    {
+      "epoch": 0.14933333333333335,
+      "grad_norm": 34.0,
+      "grad_norm_var": 5.2651652442279123e+17,
+      "learning_rate": 9.460960068750924e-05,
+      "loss": 7.0791,
+      "loss/crossentropy": 1.9466978311538696,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19675631448626518,
+      "step": 896
+    },
+    {
+      "epoch": 0.1495,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.265165244409316e+17,
+      "learning_rate": 9.459777023885755e-05,
+      "loss": 7.3352,
+      "loss/crossentropy": 2.006341964006424,
+      "loss/hidden": 3.98828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17353980988264084,
+      "step": 897
+    },
+    {
+      "epoch": 0.14966666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.265165243955807e+17,
+      "learning_rate": 9.458592756347419e-05,
+      "loss": 7.1341,
+      "loss/crossentropy": 2.1368420124053955,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26709073036909103,
+      "step": 898
+    },
+    {
+      "epoch": 0.14983333333333335,
+      "grad_norm": 31.375,
+      "grad_norm_var": 5.265165243472064e+17,
+      "learning_rate": 9.457407266460593e-05,
+      "loss": 7.173,
+      "loss/crossentropy": 1.8693583607673645,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15850002318620682,
+      "step": 899
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 30.75,
+      "grad_norm_var": 5.2651652435022976e+17,
+      "learning_rate": 9.456220554550285e-05,
+      "loss": 7.3084,
+      "loss/crossentropy": 1.6289705336093903,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14496482908725739,
+      "step": 900
+    },
+    {
+      "epoch": 0.15016666666666667,
+      "grad_norm": 37.25,
+      "grad_norm_var": 5.265165241930133e+17,
+      "learning_rate": 9.45503262094184e-05,
+      "loss": 7.1026,
+      "loss/crossentropy": 2.0200472474098206,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19197852537035942,
+      "step": 901
+    },
+    {
+      "epoch": 0.15033333333333335,
+      "grad_norm": 33.25,
+      "grad_norm_var": 5.265165242051069e+17,
+      "learning_rate": 9.453843465960933e-05,
+      "loss": 7.0761,
+      "loss/crossentropy": 2.1350898146629333,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23190056532621384,
+      "step": 902
+    },
+    {
+      "epoch": 0.1505,
+      "grad_norm": 30.625,
+      "grad_norm_var": 5.628580729166667,
+      "learning_rate": 9.45265308993358e-05,
+      "loss": 7.2962,
+      "loss/crossentropy": 1.882240429520607,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18290970101952553,
+      "step": 903
+    },
+    {
+      "epoch": 0.15066666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.5479166666666666,
+      "learning_rate": 9.451461493186129e-05,
+      "loss": 7.1687,
+      "loss/crossentropy": 2.004317596554756,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19303890317678452,
+      "step": 904
+    },
+    {
+      "epoch": 0.15083333333333335,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.3770182291666666,
+      "learning_rate": 9.450268676045262e-05,
+      "loss": 7.2162,
+      "loss/crossentropy": 2.0007368624210358,
+      "loss/hidden": 3.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16945723816752434,
+      "step": 905
+    },
+    {
+      "epoch": 0.151,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.5572916666666665,
+      "learning_rate": 9.449074638837999e-05,
+      "loss": 7.0704,
+      "loss/crossentropy": 1.30573008954525,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15130788832902908,
+      "step": 906
+    },
+    {
+      "epoch": 0.15116666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 4.053580729166667,
+      "learning_rate": 9.447879381891692e-05,
+      "loss": 7.2743,
+      "loss/crossentropy": 2.307137757539749,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17470206320285797,
+      "step": 907
+    },
+    {
+      "epoch": 0.15133333333333332,
+      "grad_norm": 29.875,
+      "grad_norm_var": 4.261393229166667,
+      "learning_rate": 9.446682905534023e-05,
+      "loss": 7.4444,
+      "loss/crossentropy": 2.1639389097690582,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20223156735301018,
+      "step": 908
+    },
+    {
+      "epoch": 0.1515,
+      "grad_norm": 29.125,
+      "grad_norm_var": 4.6494140625,
+      "learning_rate": 9.445485210093017e-05,
+      "loss": 7.0866,
+      "loss/crossentropy": 2.106477588415146,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19078096374869347,
+      "step": 909
+    },
+    {
+      "epoch": 0.15166666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 4.727018229166666,
+      "learning_rate": 9.444286295897028e-05,
+      "loss": 7.2998,
+      "loss/crossentropy": 1.8042952716350555,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19603901356458664,
+      "step": 910
+    },
+    {
+      "epoch": 0.15183333333333332,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.989518229166666,
+      "learning_rate": 9.443086163274745e-05,
+      "loss": 7.2626,
+      "loss/crossentropy": 1.7532295882701874,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16379737108945847,
+      "step": 911
+    },
+    {
+      "epoch": 0.152,
+      "grad_norm": 29.875,
+      "grad_norm_var": 4.570572916666666,
+      "learning_rate": 9.44188481255519e-05,
+      "loss": 7.177,
+      "loss/crossentropy": 2.192300647497177,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17354663461446762,
+      "step": 912
+    },
+    {
+      "epoch": 0.15216666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.410416666666666,
+      "learning_rate": 9.440682244067724e-05,
+      "loss": 7.2171,
+      "loss/crossentropy": 1.4452708065509796,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1899765394628048,
+      "step": 913
+    },
+    {
+      "epoch": 0.15233333333333332,
+      "grad_norm": 34.5,
+      "grad_norm_var": 5.134375,
+      "learning_rate": 9.439478458142033e-05,
+      "loss": 7.3901,
+      "loss/crossentropy": 1.7175801694393158,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23136239498853683,
+      "step": 914
+    },
+    {
+      "epoch": 0.1525,
+      "grad_norm": 30.875,
+      "grad_norm_var": 5.135416666666667,
+      "learning_rate": 9.438273455108144e-05,
+      "loss": 7.2527,
+      "loss/crossentropy": 2.262073040008545,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1802721694111824,
+      "step": 915
+    },
+    {
+      "epoch": 0.15266666666666667,
+      "grad_norm": 3590324224.0,
+      "grad_norm_var": 8.056517381102502e+17,
+      "learning_rate": 9.437067235296418e-05,
+      "loss": 8.8204,
+      "loss/crossentropy": 1.7655930519104004,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16529207304120064,
+      "step": 916
+    },
+    {
+      "epoch": 0.15283333333333332,
+      "grad_norm": 40.25,
+      "grad_norm_var": 8.056517380204922e+17,
+      "learning_rate": 9.43585979903754e-05,
+      "loss": 7.2085,
+      "loss/crossentropy": 1.8950283229351044,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20699721947312355,
+      "step": 917
+    },
+    {
+      "epoch": 0.153,
+      "grad_norm": 33.0,
+      "grad_norm_var": 8.05651738027972e+17,
+      "learning_rate": 9.434651146662543e-05,
+      "loss": 7.5518,
+      "loss/crossentropy": 2.4590421617031097,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1780913919210434,
+      "step": 918
+    },
+    {
+      "epoch": 0.15316666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 8.056517380653713e+17,
+      "learning_rate": 9.433441278502783e-05,
+      "loss": 7.1849,
+      "loss/crossentropy": 2.6016831398010254,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18316638097167015,
+      "step": 919
+    },
+    {
+      "epoch": 0.15333333333333332,
+      "grad_norm": 31.125,
+      "grad_norm_var": 8.056517380354518e+17,
+      "learning_rate": 9.43223019488995e-05,
+      "loss": 7.4292,
+      "loss/crossentropy": 2.240295112133026,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19262642040848732,
+      "step": 920
+    },
+    {
+      "epoch": 0.1535,
+      "grad_norm": 31.5,
+      "grad_norm_var": 8.056517380466716e+17,
+      "learning_rate": 9.431017896156074e-05,
+      "loss": 7.4752,
+      "loss/crossentropy": 2.078158289194107,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1713547259569168,
+      "step": 921
+    },
+    {
+      "epoch": 0.15366666666666667,
+      "grad_norm": 31.875,
+      "grad_norm_var": 8.056517380017925e+17,
+      "learning_rate": 9.42980438263351e-05,
+      "loss": 7.6637,
+      "loss/crossentropy": 2.522828459739685,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20312832668423653,
+      "step": 922
+    },
+    {
+      "epoch": 0.15383333333333332,
+      "grad_norm": 31.125,
+      "grad_norm_var": 8.05651737934474e+17,
+      "learning_rate": 9.428589654654951e-05,
+      "loss": 7.3971,
+      "loss/crossentropy": 2.212211459875107,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23165380582213402,
+      "step": 923
+    },
+    {
+      "epoch": 0.154,
+      "grad_norm": 31.375,
+      "grad_norm_var": 8.056517378895949e+17,
+      "learning_rate": 9.42737371255342e-05,
+      "loss": 7.1652,
+      "loss/crossentropy": 1.238950178027153,
+      "loss/hidden": 3.734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20931247621774673,
+      "step": 924
+    },
+    {
+      "epoch": 0.15416666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 8.056517377886171e+17,
+      "learning_rate": 9.426156556662276e-05,
+      "loss": 7.2505,
+      "loss/crossentropy": 1.6552457213401794,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1599828153848648,
+      "step": 925
+    },
+    {
+      "epoch": 0.15433333333333332,
+      "grad_norm": 32.0,
+      "grad_norm_var": 8.056517376988589e+17,
+      "learning_rate": 9.42493818731521e-05,
+      "loss": 7.1577,
+      "loss/crossentropy": 2.1574874818325043,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18953270465135574,
+      "step": 926
+    },
+    {
+      "epoch": 0.1545,
+      "grad_norm": 30.5,
+      "grad_norm_var": 8.056517376651996e+17,
+      "learning_rate": 9.423718604846243e-05,
+      "loss": 7.3593,
+      "loss/crossentropy": 1.9759330451488495,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1907046176493168,
+      "step": 927
+    },
+    {
+      "epoch": 0.15466666666666667,
+      "grad_norm": 31.25,
+      "grad_norm_var": 8.056517376240605e+17,
+      "learning_rate": 9.422497809589731e-05,
+      "loss": 7.2156,
+      "loss/crossentropy": 2.0266083776950836,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.239223662763834,
+      "step": 928
+    },
+    {
+      "epoch": 0.15483333333333332,
+      "grad_norm": 31.25,
+      "grad_norm_var": 8.056517376539799e+17,
+      "learning_rate": 9.421275801880362e-05,
+      "loss": 7.289,
+      "loss/crossentropy": 1.959371656179428,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18589720129966736,
+      "step": 929
+    },
+    {
+      "epoch": 0.155,
+      "grad_norm": 33.5,
+      "grad_norm_var": 8.056517376838993e+17,
+      "learning_rate": 9.420052582053157e-05,
+      "loss": 7.1632,
+      "loss/crossentropy": 1.6841667294502258,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1735333874821663,
+      "step": 930
+    },
+    {
+      "epoch": 0.15516666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 8.056517377399981e+17,
+      "learning_rate": 9.418828150443469e-05,
+      "loss": 6.9572,
+      "loss/crossentropy": 1.8017113208770752,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22501782700419426,
+      "step": 931
+    },
+    {
+      "epoch": 0.15533333333333332,
+      "grad_norm": 30.5,
+      "grad_norm_var": 6.336393229166666,
+      "learning_rate": 9.417602507386981e-05,
+      "loss": 7.1585,
+      "loss/crossentropy": 2.3576722145080566,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17171021178364754,
+      "step": 932
+    },
+    {
+      "epoch": 0.1555,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.4848307291666667,
+      "learning_rate": 9.416375653219709e-05,
+      "loss": 7.14,
+      "loss/crossentropy": 1.5370495319366455,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19083039835095406,
+      "step": 933
+    },
+    {
+      "epoch": 0.15566666666666668,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.3613932291666666,
+      "learning_rate": 9.415147588278005e-05,
+      "loss": 7.3455,
+      "loss/crossentropy": 2.0491169691085815,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19669483602046967,
+      "step": 934
+    },
+    {
+      "epoch": 0.15583333333333332,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.2197916666666666,
+      "learning_rate": 9.413918312898551e-05,
+      "loss": 7.0114,
+      "loss/crossentropy": 1.5280220061540604,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16729918867349625,
+      "step": 935
+    },
+    {
+      "epoch": 0.156,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.2285807291666666,
+      "learning_rate": 9.412687827418356e-05,
+      "loss": 7.0038,
+      "loss/crossentropy": 1.895880103111267,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16356949508190155,
+      "step": 936
+    },
+    {
+      "epoch": 0.15616666666666668,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.228125,
+      "learning_rate": 9.411456132174767e-05,
+      "loss": 7.0697,
+      "loss/crossentropy": 1.4379627853631973,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15815794467926025,
+      "step": 937
+    },
+    {
+      "epoch": 0.15633333333333332,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.2759765625,
+      "learning_rate": 9.410223227505459e-05,
+      "loss": 7.3898,
+      "loss/crossentropy": 1.8750842809677124,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17857589572668076,
+      "step": 938
+    },
+    {
+      "epoch": 0.1565,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.6613932291666667,
+      "learning_rate": 9.408989113748442e-05,
+      "loss": 7.1955,
+      "loss/crossentropy": 1.7537573277950287,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20464301854372025,
+      "step": 939
+    },
+    {
+      "epoch": 0.15666666666666668,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.7205729166666666,
+      "learning_rate": 9.407753791242051e-05,
+      "loss": 7.2697,
+      "loss/crossentropy": 1.9765137135982513,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18495013564825058,
+      "step": 940
+    },
+    {
+      "epoch": 0.15683333333333332,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.8285807291666667,
+      "learning_rate": 9.40651726032496e-05,
+      "loss": 6.9618,
+      "loss/crossentropy": 1.9569795429706573,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16958444565534592,
+      "step": 941
+    },
+    {
+      "epoch": 0.157,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.8080729166666667,
+      "learning_rate": 9.405279521336173e-05,
+      "loss": 7.1809,
+      "loss/crossentropy": 1.4055269658565521,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1510596126317978,
+      "step": 942
+    },
+    {
+      "epoch": 0.15716666666666668,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.0160807291666667,
+      "learning_rate": 9.404040574615018e-05,
+      "loss": 6.9664,
+      "loss/crossentropy": 1.967713549733162,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17161940410733223,
+      "step": 943
+    },
+    {
+      "epoch": 0.15733333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.1488932291666667,
+      "learning_rate": 9.402800420501164e-05,
+      "loss": 7.2314,
+      "loss/crossentropy": 1.6457428634166718,
+      "loss/hidden": 3.76953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23352399468421936,
+      "step": 944
+    },
+    {
+      "epoch": 0.1575,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.29140625,
+      "learning_rate": 9.401559059334601e-05,
+      "loss": 7.1579,
+      "loss/crossentropy": 1.495479092001915,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19215314090251923,
+      "step": 945
+    },
+    {
+      "epoch": 0.15766666666666668,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.9634765625,
+      "learning_rate": 9.400316491455661e-05,
+      "loss": 7.2171,
+      "loss/crossentropy": 1.8108823597431183,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1718391627073288,
+      "step": 946
+    },
+    {
+      "epoch": 0.15783333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.9291666666666667,
+      "learning_rate": 9.399072717204995e-05,
+      "loss": 7.1178,
+      "loss/crossentropy": 2.2023033797740936,
+      "loss/hidden": 3.125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17143452540040016,
+      "step": 947
+    },
+    {
+      "epoch": 0.158,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.17265625,
+      "learning_rate": 9.397827736923596e-05,
+      "loss": 7.3485,
+      "loss/crossentropy": 2.208144634962082,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18023496493697166,
+      "step": 948
+    },
+    {
+      "epoch": 0.15816666666666668,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.15390625,
+      "learning_rate": 9.396581550952781e-05,
+      "loss": 7.2341,
+      "loss/crossentropy": 2.068796455860138,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1868870034813881,
+      "step": 949
+    },
+    {
+      "epoch": 0.15833333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.06640625,
+      "learning_rate": 9.395334159634199e-05,
+      "loss": 6.9557,
+      "loss/crossentropy": 1.999780148267746,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19010787084698677,
+      "step": 950
+    },
+    {
+      "epoch": 0.1585,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.2143229166666667,
+      "learning_rate": 9.394085563309827e-05,
+      "loss": 7.1712,
+      "loss/crossentropy": 2.0135373175144196,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2353024035692215,
+      "step": 951
+    },
+    {
+      "epoch": 0.15866666666666668,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.254166666666667,
+      "learning_rate": 9.392835762321977e-05,
+      "loss": 7.0585,
+      "loss/crossentropy": 1.9209633469581604,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18967651948332787,
+      "step": 952
+    },
+    {
+      "epoch": 0.15883333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 2.8046223958333334,
+      "learning_rate": 9.391584757013289e-05,
+      "loss": 7.2089,
+      "loss/crossentropy": 1.7196340262889862,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16397151723504066,
+      "step": 953
+    },
+    {
+      "epoch": 0.159,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.0077473958333334,
+      "learning_rate": 9.390332547726733e-05,
+      "loss": 7.1139,
+      "loss/crossentropy": 1.9726295173168182,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15174923464655876,
+      "step": 954
+    },
+    {
+      "epoch": 0.15916666666666668,
+      "grad_norm": 28.375,
+      "grad_norm_var": 3.221809895833333,
+      "learning_rate": 9.389079134805609e-05,
+      "loss": 7.1402,
+      "loss/crossentropy": 2.0523089468479156,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14974992722272873,
+      "step": 955
+    },
+    {
+      "epoch": 0.15933333333333333,
+      "grad_norm": 35.5,
+      "grad_norm_var": 4.437434895833333,
+      "learning_rate": 9.387824518593546e-05,
+      "loss": 7.2187,
+      "loss/crossentropy": 2.301606833934784,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1863223984837532,
+      "step": 956
+    },
+    {
+      "epoch": 0.1595,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.346875,
+      "learning_rate": 9.386568699434508e-05,
+      "loss": 6.9746,
+      "loss/crossentropy": 2.2219241559505463,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18025089800357819,
+      "step": 957
+    },
+    {
+      "epoch": 0.15966666666666668,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.391080729166666,
+      "learning_rate": 9.385311677672781e-05,
+      "loss": 7.1517,
+      "loss/crossentropy": 2.006164014339447,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20917531847953796,
+      "step": 958
+    },
+    {
+      "epoch": 0.15983333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.17265625,
+      "learning_rate": 9.384053453652986e-05,
+      "loss": 7.2699,
+      "loss/crossentropy": 2.1049175560474396,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19741756469011307,
+      "step": 959
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 30.5,
+      "grad_norm_var": 4.053125,
+      "learning_rate": 9.382794027720073e-05,
+      "loss": 7.2579,
+      "loss/crossentropy": 2.1736232936382294,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1689480096101761,
+      "step": 960
+    },
+    {
+      "epoch": 0.16016666666666668,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.937239583333333,
+      "learning_rate": 9.381533400219318e-05,
+      "loss": 7.1442,
+      "loss/crossentropy": 1.8950579166412354,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1553553342819214,
+      "step": 961
+    },
+    {
+      "epoch": 0.16033333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 4.012434895833334,
+      "learning_rate": 9.380271571496334e-05,
+      "loss": 7.2108,
+      "loss/crossentropy": 1.5403998047113419,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18360819295048714,
+      "step": 962
+    },
+    {
+      "epoch": 0.1605,
+      "grad_norm": 28.625,
+      "grad_norm_var": 4.166080729166667,
+      "learning_rate": 9.379008541897054e-05,
+      "loss": 7.2215,
+      "loss/crossentropy": 1.4331855326890945,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21428582817316055,
+      "step": 963
+    },
+    {
+      "epoch": 0.16066666666666668,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.8643229166666666,
+      "learning_rate": 9.377744311767746e-05,
+      "loss": 7.086,
+      "loss/crossentropy": 1.837290346622467,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18247783184051514,
+      "step": 964
+    },
+    {
+      "epoch": 0.16083333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.769791666666667,
+      "learning_rate": 9.376478881455009e-05,
+      "loss": 7.0426,
+      "loss/crossentropy": 2.4413352608680725,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17528174817562103,
+      "step": 965
+    },
+    {
+      "epoch": 0.161,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.841666666666667,
+      "learning_rate": 9.375212251305763e-05,
+      "loss": 7.2408,
+      "loss/crossentropy": 1.86382594704628,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1978016346693039,
+      "step": 966
+    },
+    {
+      "epoch": 0.16116666666666668,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.4643229166666667,
+      "learning_rate": 9.373944421667265e-05,
+      "loss": 7.0779,
+      "loss/crossentropy": 1.6653753519058228,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1587580218911171,
+      "step": 967
+    },
+    {
+      "epoch": 0.16133333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.4697916666666666,
+      "learning_rate": 9.372675392887096e-05,
+      "loss": 7.4843,
+      "loss/crossentropy": 2.347218543291092,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17683962360024452,
+      "step": 968
+    },
+    {
+      "epoch": 0.1615,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.9614583333333333,
+      "learning_rate": 9.371405165313169e-05,
+      "loss": 7.0695,
+      "loss/crossentropy": 1.6909139156341553,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17361067980527878,
+      "step": 969
+    },
+    {
+      "epoch": 0.16166666666666665,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.6212890625,
+      "learning_rate": 9.370133739293723e-05,
+      "loss": 7.1812,
+      "loss/crossentropy": 1.865398108959198,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22350451350212097,
+      "step": 970
+    },
+    {
+      "epoch": 0.16183333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.05,
+      "learning_rate": 9.368861115177327e-05,
+      "loss": 7.0703,
+      "loss/crossentropy": 1.7714774459600449,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1964346505701542,
+      "step": 971
+    },
+    {
+      "epoch": 0.162,
+      "grad_norm": 33.75,
+      "grad_norm_var": 1.2934895833333333,
+      "learning_rate": 9.367587293312878e-05,
+      "loss": 7.1107,
+      "loss/crossentropy": 2.3415546119213104,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18950436264276505,
+      "step": 972
+    },
+    {
+      "epoch": 0.16216666666666665,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.3468098958333334,
+      "learning_rate": 9.366312274049602e-05,
+      "loss": 7.1331,
+      "loss/crossentropy": 1.1628313958644867,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1966600064188242,
+      "step": 973
+    },
+    {
+      "epoch": 0.16233333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.3754557291666667,
+      "learning_rate": 9.36503605773705e-05,
+      "loss": 6.9945,
+      "loss/crossentropy": 1.8536908328533173,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15904531627893448,
+      "step": 974
+    },
+    {
+      "epoch": 0.1625,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.4113932291666667,
+      "learning_rate": 9.36375864472511e-05,
+      "loss": 7.4911,
+      "loss/crossentropy": 1.9807639420032501,
+      "loss/hidden": 3.79296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32628800719976425,
+      "step": 975
+    },
+    {
+      "epoch": 0.16266666666666665,
+      "grad_norm": 33.75,
+      "grad_norm_var": 1.6957682291666667,
+      "learning_rate": 9.362480035363986e-05,
+      "loss": 7.0892,
+      "loss/crossentropy": 1.8294160664081573,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17708183079957962,
+      "step": 976
+    },
+    {
+      "epoch": 0.16283333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.7145182291666667,
+      "learning_rate": 9.36120023000422e-05,
+      "loss": 7.4688,
+      "loss/crossentropy": 2.1574412882328033,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17348594963550568,
+      "step": 977
+    },
+    {
+      "epoch": 0.163,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.8337890625,
+      "learning_rate": 9.359919228996674e-05,
+      "loss": 7.2222,
+      "loss/crossentropy": 1.945350557565689,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18768493086099625,
+      "step": 978
+    },
+    {
+      "epoch": 0.16316666666666665,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.3488932291666667,
+      "learning_rate": 9.358637032692545e-05,
+      "loss": 7.2284,
+      "loss/crossentropy": 1.9553067982196808,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17745291441679,
+      "step": 979
+    },
+    {
+      "epoch": 0.16333333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.4580729166666666,
+      "learning_rate": 9.357353641443354e-05,
+      "loss": 7.3811,
+      "loss/crossentropy": 2.3098992109298706,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17864799872040749,
+      "step": 980
+    },
+    {
+      "epoch": 0.1635,
+      "grad_norm": 33.5,
+      "grad_norm_var": 1.5268229166666667,
+      "learning_rate": 9.356069055600948e-05,
+      "loss": 7.2363,
+      "loss/crossentropy": 1.9222070574760437,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2259671874344349,
+      "step": 981
+    },
+    {
+      "epoch": 0.16366666666666665,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.7160807291666667,
+      "learning_rate": 9.354783275517504e-05,
+      "loss": 7.1659,
+      "loss/crossentropy": 2.079308956861496,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19740625843405724,
+      "step": 982
+    },
+    {
+      "epoch": 0.16383333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.8268229166666667,
+      "learning_rate": 9.353496301545529e-05,
+      "loss": 7.1756,
+      "loss/crossentropy": 1.689118355512619,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18044323846697807,
+      "step": 983
+    },
+    {
+      "epoch": 0.164,
+      "grad_norm": 35.0,
+      "grad_norm_var": 2.4955729166666667,
+      "learning_rate": 9.352208134037851e-05,
+      "loss": 7.6591,
+      "loss/crossentropy": 2.4075648486614227,
+      "loss/hidden": 3.10546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17038994655013084,
+      "step": 984
+    },
+    {
+      "epoch": 0.16416666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.505989583333333,
+      "learning_rate": 9.35091877334763e-05,
+      "loss": 7.3304,
+      "loss/crossentropy": 1.6309797316789627,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18557706847786903,
+      "step": 985
+    },
+    {
+      "epoch": 0.16433333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.5104166666666665,
+      "learning_rate": 9.349628219828349e-05,
+      "loss": 7.3376,
+      "loss/crossentropy": 1.7576228082180023,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19469567760825157,
+      "step": 986
+    },
+    {
+      "epoch": 0.1645,
+      "grad_norm": 28.375,
+      "grad_norm_var": 3.278580729166667,
+      "learning_rate": 9.348336473833823e-05,
+      "loss": 7.2064,
+      "loss/crossentropy": 1.5236790180206299,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1657111458480358,
+      "step": 987
+    },
+    {
+      "epoch": 0.16466666666666666,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.9791666666666665,
+      "learning_rate": 9.347043535718192e-05,
+      "loss": 7.2516,
+      "loss/crossentropy": 1.9423125088214874,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18766283243894577,
+      "step": 988
+    },
+    {
+      "epoch": 0.16483333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.9004557291666666,
+      "learning_rate": 9.34574940583592e-05,
+      "loss": 7.2238,
+      "loss/crossentropy": 2.072887897491455,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17064030840992928,
+      "step": 989
+    },
+    {
+      "epoch": 0.165,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.2796223958333335,
+      "learning_rate": 9.344454084541803e-05,
+      "loss": 7.2918,
+      "loss/crossentropy": 2.1878859102725983,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2114771418273449,
+      "step": 990
+    },
+    {
+      "epoch": 0.16516666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.255143229166667,
+      "learning_rate": 9.343157572190957e-05,
+      "loss": 7.0094,
+      "loss/crossentropy": 2.0703546702861786,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17606284841895103,
+      "step": 991
+    },
+    {
+      "epoch": 0.16533333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.8613932291666666,
+      "learning_rate": 9.341859869138831e-05,
+      "loss": 7.1841,
+      "loss/crossentropy": 1.6200998723506927,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1958397924900055,
+      "step": 992
+    },
+    {
+      "epoch": 0.1655,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.408072916666667,
+      "learning_rate": 9.340560975741197e-05,
+      "loss": 7.2225,
+      "loss/crossentropy": 2.042604982852936,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1897963061928749,
+      "step": 993
+    },
+    {
+      "epoch": 0.16566666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.194205729166667,
+      "learning_rate": 9.339260892354153e-05,
+      "loss": 7.1794,
+      "loss/crossentropy": 1.8934089541435242,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19708381965756416,
+      "step": 994
+    },
+    {
+      "epoch": 0.16583333333333333,
+      "grad_norm": 38.25,
+      "grad_norm_var": 5.864322916666667,
+      "learning_rate": 9.337959619334125e-05,
+      "loss": 7.5795,
+      "loss/crossentropy": 1.9645236432552338,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.184968501329422,
+      "step": 995
+    },
+    {
+      "epoch": 0.166,
+      "grad_norm": 29.625,
+      "grad_norm_var": 6.133268229166666,
+      "learning_rate": 9.336657157037866e-05,
+      "loss": 7.067,
+      "loss/crossentropy": 1.8105159401893616,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18072937428951263,
+      "step": 996
+    },
+    {
+      "epoch": 0.16616666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 5.930989583333333,
+      "learning_rate": 9.33535350582245e-05,
+      "loss": 7.1263,
+      "loss/crossentropy": 1.9204743206501007,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1801401823759079,
+      "step": 997
+    },
+    {
+      "epoch": 0.16633333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 5.773372395833333,
+      "learning_rate": 9.334048666045285e-05,
+      "loss": 7.324,
+      "loss/crossentropy": 1.9125606715679169,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.181853249669075,
+      "step": 998
+    },
+    {
+      "epoch": 0.1665,
+      "grad_norm": 30.875,
+      "grad_norm_var": 5.695768229166666,
+      "learning_rate": 9.332742638064094e-05,
+      "loss": 7.0851,
+      "loss/crossentropy": 1.5902134627103806,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15430122800171375,
+      "step": 999
+    },
+    {
+      "epoch": 0.16666666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 5.389518229166667,
+      "learning_rate": 9.331435422236938e-05,
+      "loss": 7.1695,
+      "loss/crossentropy": 2.58576363325119,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17759808152914047,
+      "step": 1000
+    },
+    {
+      "epoch": 0.16683333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 5.417643229166667,
+      "learning_rate": 9.330127018922194e-05,
+      "loss": 7.2752,
+      "loss/crossentropy": 2.087014377117157,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24709226936101913,
+      "step": 1001
+    },
+    {
+      "epoch": 0.167,
+      "grad_norm": 31.25,
+      "grad_norm_var": 5.414322916666666,
+      "learning_rate": 9.328817428478569e-05,
+      "loss": 7.4509,
+      "loss/crossentropy": 1.6802764534950256,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1839093528687954,
+      "step": 1002
+    },
+    {
+      "epoch": 0.16716666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 4.835416666666666,
+      "learning_rate": 9.327506651265095e-05,
+      "loss": 7.2616,
+      "loss/crossentropy": 1.66881962120533,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19290011748671532,
+      "step": 1003
+    },
+    {
+      "epoch": 0.16733333333333333,
+      "grad_norm": 28.125,
+      "grad_norm_var": 5.547916666666667,
+      "learning_rate": 9.32619468764113e-05,
+      "loss": 7.1466,
+      "loss/crossentropy": 1.910039633512497,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17122110351920128,
+      "step": 1004
+    },
+    {
+      "epoch": 0.1675,
+      "grad_norm": 31.625,
+      "grad_norm_var": 5.522330729166667,
+      "learning_rate": 9.324881537966354e-05,
+      "loss": 7.202,
+      "loss/crossentropy": 2.0192385017871857,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16851748898625374,
+      "step": 1005
+    },
+    {
+      "epoch": 0.16766666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 5.2384765625,
+      "learning_rate": 9.323567202600776e-05,
+      "loss": 7.1523,
+      "loss/crossentropy": 1.832692712545395,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19196533411741257,
+      "step": 1006
+    },
+    {
+      "epoch": 0.16783333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 5.2119140625,
+      "learning_rate": 9.322251681904728e-05,
+      "loss": 7.1467,
+      "loss/crossentropy": 2.532587170600891,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1897234544157982,
+      "step": 1007
+    },
+    {
+      "epoch": 0.168,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.2962890625,
+      "learning_rate": 9.320934976238867e-05,
+      "loss": 6.9639,
+      "loss/crossentropy": 1.7438288033008575,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1728871464729309,
+      "step": 1008
+    },
+    {
+      "epoch": 0.16816666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 4.829166666666667,
+      "learning_rate": 9.319617085964176e-05,
+      "loss": 7.1181,
+      "loss/crossentropy": 2.017861932516098,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16254071705043316,
+      "step": 1009
+    },
+    {
+      "epoch": 0.16833333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 4.9353515625,
+      "learning_rate": 9.318298011441964e-05,
+      "loss": 7.2094,
+      "loss/crossentropy": 1.9356443881988525,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1496536247432232,
+      "step": 1010
+    },
+    {
+      "epoch": 0.1685,
+      "grad_norm": 33.75,
+      "grad_norm_var": 1.8931640625,
+      "learning_rate": 9.316977753033859e-05,
+      "loss": 7.2573,
+      "loss/crossentropy": 2.3118956387043,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19966011494398117,
+      "step": 1011
+    },
+    {
+      "epoch": 0.16866666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.8004557291666667,
+      "learning_rate": 9.31565631110182e-05,
+      "loss": 6.9932,
+      "loss/crossentropy": 1.9621807038784027,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17853105068206787,
+      "step": 1012
+    },
+    {
+      "epoch": 0.16883333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.7260416666666667,
+      "learning_rate": 9.314333686008125e-05,
+      "loss": 7.3292,
+      "loss/crossentropy": 1.593971699476242,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23572874069213867,
+      "step": 1013
+    },
+    {
+      "epoch": 0.169,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.7895182291666667,
+      "learning_rate": 9.313009878115381e-05,
+      "loss": 7.3445,
+      "loss/crossentropy": 2.31576144695282,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21944650635123253,
+      "step": 1014
+    },
+    {
+      "epoch": 0.16916666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.0249348958333333,
+      "learning_rate": 9.31168488778652e-05,
+      "loss": 7.3206,
+      "loss/crossentropy": 1.6449679732322693,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16713885590434074,
+      "step": 1015
+    },
+    {
+      "epoch": 0.16933333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.8228515625,
+      "learning_rate": 9.310358715384793e-05,
+      "loss": 7.0185,
+      "loss/crossentropy": 1.6693122684955597,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2041461020708084,
+      "step": 1016
+    },
+    {
+      "epoch": 0.1695,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.8337890625,
+      "learning_rate": 9.309031361273775e-05,
+      "loss": 7.1052,
+      "loss/crossentropy": 1.2949364483356476,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16708527132868767,
+      "step": 1017
+    },
+    {
+      "epoch": 0.16966666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.825,
+      "learning_rate": 9.307702825817373e-05,
+      "loss": 7.3125,
+      "loss/crossentropy": 2.2340132296085358,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18231170624494553,
+      "step": 1018
+    },
+    {
+      "epoch": 0.16983333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.84140625,
+      "learning_rate": 9.306373109379809e-05,
+      "loss": 7.1191,
+      "loss/crossentropy": 2.127659171819687,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16875912249088287,
+      "step": 1019
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.4426432291666667,
+      "learning_rate": 9.305042212325634e-05,
+      "loss": 7.3397,
+      "loss/crossentropy": 2.045904368162155,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18133864179253578,
+      "step": 1020
+    },
+    {
+      "epoch": 0.17016666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.4629557291666666,
+      "learning_rate": 9.30371013501972e-05,
+      "loss": 7.3434,
+      "loss/crossentropy": 2.24924436211586,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24430305883288383,
+      "step": 1021
+    },
+    {
+      "epoch": 0.17033333333333334,
+      "grad_norm": 34.0,
+      "grad_norm_var": 1.8926432291666666,
+      "learning_rate": 9.302376877827263e-05,
+      "loss": 7.2228,
+      "loss/crossentropy": 1.7824518978595734,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20318450778722763,
+      "step": 1022
+    },
+    {
+      "epoch": 0.1705,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.9160807291666666,
+      "learning_rate": 9.301042441113783e-05,
+      "loss": 7.1712,
+      "loss/crossentropy": 2.2989481389522552,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21584442257881165,
+      "step": 1023
+    },
+    {
+      "epoch": 0.17066666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.8837890625,
+      "learning_rate": 9.299706825245126e-05,
+      "loss": 7.3148,
+      "loss/crossentropy": 2.2267412543296814,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21536936983466148,
+      "step": 1024
+    },
+    {
+      "epoch": 0.17083333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.0009765625,
+      "learning_rate": 9.298370030587456e-05,
+      "loss": 7.1826,
+      "loss/crossentropy": 1.9287384748458862,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15461841970682144,
+      "step": 1025
+    },
+    {
+      "epoch": 0.171,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.8181640625,
+      "learning_rate": 9.297032057507264e-05,
+      "loss": 7.1634,
+      "loss/crossentropy": 2.0097956359386444,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2209189385175705,
+      "step": 1026
+    },
+    {
+      "epoch": 0.17116666666666666,
+      "grad_norm": 34.0,
+      "grad_norm_var": 1.8983723958333334,
+      "learning_rate": 9.295692906371363e-05,
+      "loss": 6.9612,
+      "loss/crossentropy": 2.0483106672763824,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.180673498660326,
+      "step": 1027
+    },
+    {
+      "epoch": 0.17133333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.98125,
+      "learning_rate": 9.294352577546888e-05,
+      "loss": 7.0964,
+      "loss/crossentropy": 1.5798965096473694,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1735404171049595,
+      "step": 1028
+    },
+    {
+      "epoch": 0.1715,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.9327473958333334,
+      "learning_rate": 9.293011071401298e-05,
+      "loss": 7.0009,
+      "loss/crossentropy": 2.467678666114807,
+      "loss/hidden": 3.05859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16103530675172806,
+      "step": 1029
+    },
+    {
+      "epoch": 0.17166666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.1264973958333333,
+      "learning_rate": 9.291668388302374e-05,
+      "loss": 7.1786,
+      "loss/crossentropy": 1.899499461054802,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19169988110661507,
+      "step": 1030
+    },
+    {
+      "epoch": 0.17183333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.6885416666666666,
+      "learning_rate": 9.290324528618224e-05,
+      "loss": 7.107,
+      "loss/crossentropy": 1.5165728628635406,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22037753462791443,
+      "step": 1031
+    },
+    {
+      "epoch": 0.172,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.7363932291666666,
+      "learning_rate": 9.28897949271727e-05,
+      "loss": 7.3123,
+      "loss/crossentropy": 1.9432778656482697,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2261851504445076,
+      "step": 1032
+    },
+    {
+      "epoch": 0.17216666666666666,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.176497395833333,
+      "learning_rate": 9.287633280968261e-05,
+      "loss": 7.2204,
+      "loss/crossentropy": 2.029956191778183,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1663634330034256,
+      "step": 1033
+    },
+    {
+      "epoch": 0.17233333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.0780598958333334,
+      "learning_rate": 9.286285893740274e-05,
+      "loss": 7.1274,
+      "loss/crossentropy": 1.7812095880508423,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18223801627755165,
+      "step": 1034
+    },
+    {
+      "epoch": 0.1725,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.1809895833333335,
+      "learning_rate": 9.284937331402697e-05,
+      "loss": 7.3825,
+      "loss/crossentropy": 2.34644615650177,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2404446192085743,
+      "step": 1035
+    },
+    {
+      "epoch": 0.17266666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.2681640625,
+      "learning_rate": 9.283587594325249e-05,
+      "loss": 7.4599,
+      "loss/crossentropy": 1.9660449028015137,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23820456862449646,
+      "step": 1036
+    },
+    {
+      "epoch": 0.17283333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.596875,
+      "learning_rate": 9.282236682877967e-05,
+      "loss": 7.3944,
+      "loss/crossentropy": 2.454943358898163,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18604904040694237,
+      "step": 1037
+    },
+    {
+      "epoch": 0.173,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.42890625,
+      "learning_rate": 9.280884597431212e-05,
+      "loss": 7.201,
+      "loss/crossentropy": 1.7944923341274261,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19807931408286095,
+      "step": 1038
+    },
+    {
+      "epoch": 0.17316666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.54375,
+      "learning_rate": 9.279531338355666e-05,
+      "loss": 7.369,
+      "loss/crossentropy": 2.254125028848648,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18860766291618347,
+      "step": 1039
+    },
+    {
+      "epoch": 0.17333333333333334,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.5145833333333334,
+      "learning_rate": 9.27817690602233e-05,
+      "loss": 7.2031,
+      "loss/crossentropy": 1.7020633667707443,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16921743378043175,
+      "step": 1040
+    },
+    {
+      "epoch": 0.1735,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.4712890625,
+      "learning_rate": 9.276821300802534e-05,
+      "loss": 7.4906,
+      "loss/crossentropy": 2.5524336099624634,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21683865785598755,
+      "step": 1041
+    },
+    {
+      "epoch": 0.17366666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.463997395833333,
+      "learning_rate": 9.27546452306792e-05,
+      "loss": 7.1542,
+      "loss/crossentropy": 1.6130549311637878,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16316590085625648,
+      "step": 1042
+    },
+    {
+      "epoch": 0.17383333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.13125,
+      "learning_rate": 9.274106573190459e-05,
+      "loss": 7.2674,
+      "loss/crossentropy": 2.2276190519332886,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17132963985204697,
+      "step": 1043
+    },
+    {
+      "epoch": 0.174,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.0228515625,
+      "learning_rate": 9.272747451542441e-05,
+      "loss": 7.2025,
+      "loss/crossentropy": 2.537658542394638,
+      "loss/hidden": 3.05859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15953578054904938,
+      "step": 1044
+    },
+    {
+      "epoch": 0.17416666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.153059895833333,
+      "learning_rate": 9.271387158496476e-05,
+      "loss": 7.3926,
+      "loss/crossentropy": 1.8676083832979202,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1780277006328106,
+      "step": 1045
+    },
+    {
+      "epoch": 0.17433333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.9934895833333333,
+      "learning_rate": 9.270025694425497e-05,
+      "loss": 7.1114,
+      "loss/crossentropy": 1.362665131688118,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2049396112561226,
+      "step": 1046
+    },
+    {
+      "epoch": 0.1745,
+      "grad_norm": 34.25,
+      "grad_norm_var": 2.3559895833333333,
+      "learning_rate": 9.268663059702753e-05,
+      "loss": 7.298,
+      "loss/crossentropy": 1.7508555352687836,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20855866745114326,
+      "step": 1047
+    },
+    {
+      "epoch": 0.17466666666666666,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.56640625,
+      "learning_rate": 9.267299254701824e-05,
+      "loss": 7.0742,
+      "loss/crossentropy": 1.7969558238983154,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17054210230708122,
+      "step": 1048
+    },
+    {
+      "epoch": 0.17483333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.2434895833333335,
+      "learning_rate": 9.265934279796602e-05,
+      "loss": 7.1526,
+      "loss/crossentropy": 2.483209639787674,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18839319422841072,
+      "step": 1049
+    },
+    {
+      "epoch": 0.175,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.6009765625,
+      "learning_rate": 9.264568135361302e-05,
+      "loss": 6.9266,
+      "loss/crossentropy": 1.7531154453754425,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20728101953864098,
+      "step": 1050
+    },
+    {
+      "epoch": 0.17516666666666666,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.6712890625,
+      "learning_rate": 9.263200821770461e-05,
+      "loss": 7.1656,
+      "loss/crossentropy": 1.6701923459768295,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15576695650815964,
+      "step": 1051
+    },
+    {
+      "epoch": 0.17533333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.72890625,
+      "learning_rate": 9.261832339398938e-05,
+      "loss": 6.9576,
+      "loss/crossentropy": 2.0287326872348785,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17575950920581818,
+      "step": 1052
+    },
+    {
+      "epoch": 0.1755,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.4893229166666666,
+      "learning_rate": 9.260462688621905e-05,
+      "loss": 7.4147,
+      "loss/crossentropy": 2.1414792835712433,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22248150035738945,
+      "step": 1053
+    },
+    {
+      "epoch": 0.17566666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.939322916666667,
+      "learning_rate": 9.259091869814864e-05,
+      "loss": 7.1442,
+      "loss/crossentropy": 2.3280774652957916,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16715997830033302,
+      "step": 1054
+    },
+    {
+      "epoch": 0.17583333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.879166666666667,
+      "learning_rate": 9.257719883353631e-05,
+      "loss": 7.29,
+      "loss/crossentropy": 2.034582108259201,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20363850891590118,
+      "step": 1055
+    },
+    {
+      "epoch": 0.176,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.853125,
+      "learning_rate": 9.256346729614342e-05,
+      "loss": 7.5493,
+      "loss/crossentropy": 2.168720304965973,
+      "loss/hidden": 3.76171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2940753549337387,
+      "step": 1056
+    },
+    {
+      "epoch": 0.17616666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.6497395833333335,
+      "learning_rate": 9.254972408973461e-05,
+      "loss": 7.2788,
+      "loss/crossentropy": 2.018580198287964,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17006657645106316,
+      "step": 1057
+    },
+    {
+      "epoch": 0.17633333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.687239583333333,
+      "learning_rate": 9.253596921807759e-05,
+      "loss": 7.2668,
+      "loss/crossentropy": 2.0598830580711365,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18338441848754883,
+      "step": 1058
+    },
+    {
+      "epoch": 0.1765,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.7875,
+      "learning_rate": 9.252220268494337e-05,
+      "loss": 7.0292,
+      "loss/crossentropy": 2.1577613055706024,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21034742519259453,
+      "step": 1059
+    },
+    {
+      "epoch": 0.17666666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.0473307291666667,
+      "learning_rate": 9.250842449410611e-05,
+      "loss": 7.2715,
+      "loss/crossentropy": 1.483291283249855,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15924222394824028,
+      "step": 1060
+    },
+    {
+      "epoch": 0.17683333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.951041666666667,
+      "learning_rate": 9.249463464934321e-05,
+      "loss": 7.3362,
+      "loss/crossentropy": 1.979320913553238,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17921079322695732,
+      "step": 1061
+    },
+    {
+      "epoch": 0.177,
+      "grad_norm": 29.25,
+      "grad_norm_var": 3.2510416666666666,
+      "learning_rate": 9.248083315443518e-05,
+      "loss": 7.146,
+      "loss/crossentropy": 1.8284825682640076,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1878691166639328,
+      "step": 1062
+    },
+    {
+      "epoch": 0.17716666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 3.018684895833333,
+      "learning_rate": 9.246702001316583e-05,
+      "loss": 6.9519,
+      "loss/crossentropy": 1.9803139865398407,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16849515587091446,
+      "step": 1063
+    },
+    {
+      "epoch": 0.17733333333333334,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.8705729166666667,
+      "learning_rate": 9.245319522932209e-05,
+      "loss": 7.333,
+      "loss/crossentropy": 1.8942987620830536,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15474717319011688,
+      "step": 1064
+    },
+    {
+      "epoch": 0.1775,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.6723307291666667,
+      "learning_rate": 9.24393588066941e-05,
+      "loss": 7.3183,
+      "loss/crossentropy": 1.9049183428287506,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1924520954489708,
+      "step": 1065
+    },
+    {
+      "epoch": 0.17766666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.5238932291666667,
+      "learning_rate": 9.242551074907519e-05,
+      "loss": 7.038,
+      "loss/crossentropy": 1.4033511132001877,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19038311392068863,
+      "step": 1066
+    },
+    {
+      "epoch": 0.17783333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.294205729166667,
+      "learning_rate": 9.241165106026189e-05,
+      "loss": 6.9859,
+      "loss/crossentropy": 1.9976651966571808,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1772223636507988,
+      "step": 1067
+    },
+    {
+      "epoch": 0.178,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.13515625,
+      "learning_rate": 9.239777974405393e-05,
+      "loss": 7.3643,
+      "loss/crossentropy": 1.8822968304157257,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17623241245746613,
+      "step": 1068
+    },
+    {
+      "epoch": 0.17816666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.012434895833333,
+      "learning_rate": 9.238389680425416e-05,
+      "loss": 6.9632,
+      "loss/crossentropy": 1.845107764005661,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17053649201989174,
+      "step": 1069
+    },
+    {
+      "epoch": 0.17833333333333334,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.7809895833333333,
+      "learning_rate": 9.237000224466872e-05,
+      "loss": 6.9587,
+      "loss/crossentropy": 1.9095114171504974,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18886420503258705,
+      "step": 1070
+    },
+    {
+      "epoch": 0.1785,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.4889973958333333,
+      "learning_rate": 9.235609606910687e-05,
+      "loss": 7.3356,
+      "loss/crossentropy": 2.0150366127490997,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19185121729969978,
+      "step": 1071
+    },
+    {
+      "epoch": 0.17866666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.4458333333333333,
+      "learning_rate": 9.234217828138104e-05,
+      "loss": 7.094,
+      "loss/crossentropy": 2.384048730134964,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17805096879601479,
+      "step": 1072
+    },
+    {
+      "epoch": 0.17883333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.4962890625,
+      "learning_rate": 9.23282488853069e-05,
+      "loss": 7.3047,
+      "loss/crossentropy": 2.1515243649482727,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17858536168932915,
+      "step": 1073
+    },
+    {
+      "epoch": 0.179,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.496875,
+      "learning_rate": 9.231430788470326e-05,
+      "loss": 7.1484,
+      "loss/crossentropy": 1.7612971067428589,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1948094740509987,
+      "step": 1074
+    },
+    {
+      "epoch": 0.17916666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.7405598958333333,
+      "learning_rate": 9.230035528339211e-05,
+      "loss": 7.3864,
+      "loss/crossentropy": 1.7598745822906494,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2074742130935192,
+      "step": 1075
+    },
+    {
+      "epoch": 0.17933333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.6988932291666667,
+      "learning_rate": 9.228639108519868e-05,
+      "loss": 7.2073,
+      "loss/crossentropy": 2.006108194589615,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16674261540174484,
+      "step": 1076
+    },
+    {
+      "epoch": 0.1795,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.5791015625,
+      "learning_rate": 9.227241529395127e-05,
+      "loss": 7.2712,
+      "loss/crossentropy": 1.856663167476654,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18792196735739708,
+      "step": 1077
+    },
+    {
+      "epoch": 0.17966666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.4041015625,
+      "learning_rate": 9.225842791348149e-05,
+      "loss": 7.1093,
+      "loss/crossentropy": 1.85087850689888,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18496759235858917,
+      "step": 1078
+    },
+    {
+      "epoch": 0.17983333333333335,
+      "grad_norm": 33.75,
+      "grad_norm_var": 1.55390625,
+      "learning_rate": 9.224442894762401e-05,
+      "loss": 6.9361,
+      "loss/crossentropy": 2.15621554851532,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17378008365631104,
+      "step": 1079
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.8080729166666667,
+      "learning_rate": 9.223041840021674e-05,
+      "loss": 7.1794,
+      "loss/crossentropy": 2.170605331659317,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20242366567254066,
+      "step": 1080
+    },
+    {
+      "epoch": 0.18016666666666667,
+      "grad_norm": 26.625,
+      "grad_norm_var": 3.2083333333333335,
+      "learning_rate": 9.221639627510076e-05,
+      "loss": 7.226,
+      "loss/crossentropy": 2.1505748331546783,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16657578200101852,
+      "step": 1081
+    },
+    {
+      "epoch": 0.18033333333333335,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.9934895833333335,
+      "learning_rate": 9.220236257612031e-05,
+      "loss": 6.9384,
+      "loss/crossentropy": 1.895026445388794,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19993529841303825,
+      "step": 1082
+    },
+    {
+      "epoch": 0.1805,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.9447916666666667,
+      "learning_rate": 9.21883173071228e-05,
+      "loss": 7.0944,
+      "loss/crossentropy": 1.5978422164916992,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18047995120286942,
+      "step": 1083
+    },
+    {
+      "epoch": 0.18066666666666667,
+      "grad_norm": 28.5,
+      "grad_norm_var": 3.192643229166667,
+      "learning_rate": 9.217426047195882e-05,
+      "loss": 7.1514,
+      "loss/crossentropy": 2.0529713928699493,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1859581246972084,
+      "step": 1084
+    },
+    {
+      "epoch": 0.18083333333333335,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.193489583333333,
+      "learning_rate": 9.216019207448217e-05,
+      "loss": 7.2263,
+      "loss/crossentropy": 1.5950286090373993,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22618724778294563,
+      "step": 1085
+    },
+    {
+      "epoch": 0.181,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.4166015625,
+      "learning_rate": 9.214611211854974e-05,
+      "loss": 7.1725,
+      "loss/crossentropy": 1.5778129696846008,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17347296327352524,
+      "step": 1086
+    },
+    {
+      "epoch": 0.18116666666666667,
+      "grad_norm": 34.75,
+      "grad_norm_var": 4.378125,
+      "learning_rate": 9.213202060802161e-05,
+      "loss": 7.3177,
+      "loss/crossentropy": 1.625305026769638,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21378153562545776,
+      "step": 1087
+    },
+    {
+      "epoch": 0.18133333333333335,
+      "grad_norm": 29.875,
+      "grad_norm_var": 4.412239583333333,
+      "learning_rate": 9.21179175467611e-05,
+      "loss": 7.1323,
+      "loss/crossentropy": 2.235830783843994,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2603518143296242,
+      "step": 1088
+    },
+    {
+      "epoch": 0.1815,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.388997395833333,
+      "learning_rate": 9.210380293863462e-05,
+      "loss": 7.5201,
+      "loss/crossentropy": 1.9198448956012726,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24537065252661705,
+      "step": 1089
+    },
+    {
+      "epoch": 0.18166666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.445572916666666,
+      "learning_rate": 9.208967678751177e-05,
+      "loss": 7.3548,
+      "loss/crossentropy": 1.76581409573555,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21681368723511696,
+      "step": 1090
+    },
+    {
+      "epoch": 0.18183333333333335,
+      "grad_norm": 29.125,
+      "grad_norm_var": 4.443684895833333,
+      "learning_rate": 9.207553909726531e-05,
+      "loss": 7.1191,
+      "loss/crossentropy": 2.404295325279236,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2112221121788025,
+      "step": 1091
+    },
+    {
+      "epoch": 0.182,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.438997395833334,
+      "learning_rate": 9.206138987177118e-05,
+      "loss": 7.4051,
+      "loss/crossentropy": 1.4287064969539642,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2224784456193447,
+      "step": 1092
+    },
+    {
+      "epoch": 0.18216666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 4.637434895833334,
+      "learning_rate": 9.204722911490846e-05,
+      "loss": 7.3927,
+      "loss/crossentropy": 1.435084581375122,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18473000265657902,
+      "step": 1093
+    },
+    {
+      "epoch": 0.18233333333333332,
+      "grad_norm": 28.875,
+      "grad_norm_var": 4.93515625,
+      "learning_rate": 9.20330568305594e-05,
+      "loss": 7.0767,
+      "loss/crossentropy": 2.334080159664154,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19682832062244415,
+      "step": 1094
+    },
+    {
+      "epoch": 0.1825,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.6875,
+      "learning_rate": 9.201887302260943e-05,
+      "loss": 7.1086,
+      "loss/crossentropy": 1.6933359205722809,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1666032113134861,
+      "step": 1095
+    },
+    {
+      "epoch": 0.18266666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 4.29140625,
+      "learning_rate": 9.20046776949471e-05,
+      "loss": 7.0629,
+      "loss/crossentropy": 1.6405772268772125,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1735415756702423,
+      "step": 1096
+    },
+    {
+      "epoch": 0.18283333333333332,
+      "grad_norm": 28.375,
+      "grad_norm_var": 3.5239583333333333,
+      "learning_rate": 9.199047085146415e-05,
+      "loss": 7.0849,
+      "loss/crossentropy": 1.7923431545495987,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16070173308253288,
+      "step": 1097
+    },
+    {
+      "epoch": 0.183,
+      "grad_norm": 34.5,
+      "grad_norm_var": 4.190625,
+      "learning_rate": 9.197625249605546e-05,
+      "loss": 7.2866,
+      "loss/crossentropy": 1.8078107237815857,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19605600833892822,
+      "step": 1098
+    },
+    {
+      "epoch": 0.18316666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 4.1166015625,
+      "learning_rate": 9.196202263261908e-05,
+      "loss": 7.1108,
+      "loss/crossentropy": 1.8644693046808243,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18539436906576157,
+      "step": 1099
+    },
+    {
+      "epoch": 0.18333333333333332,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.7676432291666666,
+      "learning_rate": 9.194778126505621e-05,
+      "loss": 7.2911,
+      "loss/crossentropy": 2.205755800008774,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18408109992742538,
+      "step": 1100
+    },
+    {
+      "epoch": 0.1835,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.762434895833333,
+      "learning_rate": 9.193352839727121e-05,
+      "loss": 7.2636,
+      "loss/crossentropy": 2.352554678916931,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2354946956038475,
+      "step": 1101
+    },
+    {
+      "epoch": 0.18366666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.682747395833333,
+      "learning_rate": 9.191926403317155e-05,
+      "loss": 7.1971,
+      "loss/crossentropy": 1.7942090779542923,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17148781195282936,
+      "step": 1102
+    },
+    {
+      "epoch": 0.18383333333333332,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.2035807291666667,
+      "learning_rate": 9.190498817666793e-05,
+      "loss": 7.3706,
+      "loss/crossentropy": 2.0370034277439117,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19766034930944443,
+      "step": 1103
+    },
+    {
+      "epoch": 0.184,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.107291666666667,
+      "learning_rate": 9.189070083167411e-05,
+      "loss": 7.2171,
+      "loss/crossentropy": 2.3304092288017273,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15748042613267899,
+      "step": 1104
+    },
+    {
+      "epoch": 0.18416666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.595833333333333,
+      "learning_rate": 9.18764020021071e-05,
+      "loss": 7.1927,
+      "loss/crossentropy": 1.6049788445234299,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17319898307323456,
+      "step": 1105
+    },
+    {
+      "epoch": 0.18433333333333332,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.840625,
+      "learning_rate": 9.186209169188695e-05,
+      "loss": 7.301,
+      "loss/crossentropy": 1.8717091083526611,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19568662717938423,
+      "step": 1106
+    },
+    {
+      "epoch": 0.1845,
+      "grad_norm": 29.375,
+      "grad_norm_var": 3.7684895833333334,
+      "learning_rate": 9.184776990493695e-05,
+      "loss": 7.101,
+      "loss/crossentropy": 1.502578467130661,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16818956658244133,
+      "step": 1107
+    },
+    {
+      "epoch": 0.18466666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.769791666666667,
+      "learning_rate": 9.183343664518348e-05,
+      "loss": 7.059,
+      "loss/crossentropy": 1.6985718607902527,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1569894440472126,
+      "step": 1108
+    },
+    {
+      "epoch": 0.18483333333333332,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.747330729166667,
+      "learning_rate": 9.181909191655612e-05,
+      "loss": 7.4008,
+      "loss/crossentropy": 2.136345535516739,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18882112205028534,
+      "step": 1109
+    },
+    {
+      "epoch": 0.185,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.4301432291666667,
+      "learning_rate": 9.180473572298751e-05,
+      "loss": 7.0527,
+      "loss/crossentropy": 1.848597764968872,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1691058613359928,
+      "step": 1110
+    },
+    {
+      "epoch": 0.18516666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.2603515625,
+      "learning_rate": 9.179036806841353e-05,
+      "loss": 7.0831,
+      "loss/crossentropy": 1.4170871824026108,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14410424046218395,
+      "step": 1111
+    },
+    {
+      "epoch": 0.18533333333333332,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.2759765625,
+      "learning_rate": 9.177598895677309e-05,
+      "loss": 7.2329,
+      "loss/crossentropy": 1.8506868183612823,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18226511031389236,
+      "step": 1112
+    },
+    {
+      "epoch": 0.1855,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.693489583333333,
+      "learning_rate": 9.176159839200838e-05,
+      "loss": 7.3381,
+      "loss/crossentropy": 2.072927862405777,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1637607403099537,
+      "step": 1113
+    },
+    {
+      "epoch": 0.18566666666666667,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.0468098958333334,
+      "learning_rate": 9.17471963780646e-05,
+      "loss": 7.1197,
+      "loss/crossentropy": 1.9643321335315704,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.207236610352993,
+      "step": 1114
+    },
+    {
+      "epoch": 0.18583333333333332,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.096809895833333,
+      "learning_rate": 9.173278291889015e-05,
+      "loss": 7.1386,
+      "loss/crossentropy": 1.5390998721122742,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2063824087381363,
+      "step": 1115
+    },
+    {
+      "epoch": 0.186,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.9827473958333333,
+      "learning_rate": 9.171835801843658e-05,
+      "loss": 7.2213,
+      "loss/crossentropy": 1.6902571022510529,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2137136198580265,
+      "step": 1116
+    },
+    {
+      "epoch": 0.18616666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.9684895833333333,
+      "learning_rate": 9.170392168065857e-05,
+      "loss": 7.3127,
+      "loss/crossentropy": 1.8249911814928055,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18652789294719696,
+      "step": 1117
+    },
+    {
+      "epoch": 0.18633333333333332,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.9052083333333334,
+      "learning_rate": 9.168947390951388e-05,
+      "loss": 7.4477,
+      "loss/crossentropy": 2.194087713956833,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1758115589618683,
+      "step": 1118
+    },
+    {
+      "epoch": 0.1865,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.6635416666666667,
+      "learning_rate": 9.167501470896349e-05,
+      "loss": 7.2125,
+      "loss/crossentropy": 1.903648316860199,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22230049595236778,
+      "step": 1119
+    },
+    {
+      "epoch": 0.18666666666666668,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.6354166666666667,
+      "learning_rate": 9.166054408297145e-05,
+      "loss": 6.9454,
+      "loss/crossentropy": 1.9314337074756622,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17697272449731827,
+      "step": 1120
+    },
+    {
+      "epoch": 0.18683333333333332,
+      "grad_norm": 31.75,
+      "grad_norm_var": 0.9635416666666666,
+      "learning_rate": 9.164606203550497e-05,
+      "loss": 6.9815,
+      "loss/crossentropy": 1.6089248955249786,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1548868641257286,
+      "step": 1121
+    },
+    {
+      "epoch": 0.187,
+      "grad_norm": 29.75,
+      "grad_norm_var": 0.9205729166666666,
+      "learning_rate": 9.16315685705344e-05,
+      "loss": 7.1347,
+      "loss/crossentropy": 2.285125106573105,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17942291125655174,
+      "step": 1122
+    },
+    {
+      "epoch": 0.18716666666666668,
+      "grad_norm": 30.5,
+      "grad_norm_var": 0.7676432291666667,
+      "learning_rate": 9.161706369203317e-05,
+      "loss": 7.1366,
+      "loss/crossentropy": 2.4472318291664124,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1744411215186119,
+      "step": 1123
+    },
+    {
+      "epoch": 0.18733333333333332,
+      "grad_norm": 38.0,
+      "grad_norm_var": 3.8186848958333335,
+      "learning_rate": 9.160254740397791e-05,
+      "loss": 6.9888,
+      "loss/crossentropy": 1.8763768374919891,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21799440681934357,
+      "step": 1124
+    },
+    {
+      "epoch": 0.1875,
+      "grad_norm": 38.0,
+      "grad_norm_var": 6.243489583333333,
+      "learning_rate": 9.158801971034832e-05,
+      "loss": 7.3653,
+      "loss/crossentropy": 1.765801727771759,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17796454578638077,
+      "step": 1125
+    },
+    {
+      "epoch": 0.18766666666666668,
+      "grad_norm": 27.375,
+      "grad_norm_var": 7.437239583333334,
+      "learning_rate": 9.157348061512727e-05,
+      "loss": 7.2503,
+      "loss/crossentropy": 2.0462097227573395,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1838400550186634,
+      "step": 1126
+    },
+    {
+      "epoch": 0.18783333333333332,
+      "grad_norm": 30.875,
+      "grad_norm_var": 7.481184895833334,
+      "learning_rate": 9.15589301223007e-05,
+      "loss": 7.2894,
+      "loss/crossentropy": 1.8764572888612747,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1746368706226349,
+      "step": 1127
+    },
+    {
+      "epoch": 0.188,
+      "grad_norm": 29.375,
+      "grad_norm_var": 7.81640625,
+      "learning_rate": 9.154436823585777e-05,
+      "loss": 7.4764,
+      "loss/crossentropy": 2.09167617559433,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19841239601373672,
+      "step": 1128
+    },
+    {
+      "epoch": 0.18816666666666668,
+      "grad_norm": 33.5,
+      "grad_norm_var": 8.04375,
+      "learning_rate": 9.152979495979063e-05,
+      "loss": 7.0105,
+      "loss/crossentropy": 2.3772284984588623,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20802132412791252,
+      "step": 1129
+    },
+    {
+      "epoch": 0.18833333333333332,
+      "grad_norm": 30.125,
+      "grad_norm_var": 8.19140625,
+      "learning_rate": 9.151521029809469e-05,
+      "loss": 7.2106,
+      "loss/crossentropy": 1.7908346056938171,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16295769810676575,
+      "step": 1130
+    },
+    {
+      "epoch": 0.1885,
+      "grad_norm": 29.75,
+      "grad_norm_var": 8.159830729166666,
+      "learning_rate": 9.150061425476838e-05,
+      "loss": 7.2774,
+      "loss/crossentropy": 2.06538662314415,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18629369512200356,
+      "step": 1131
+    },
+    {
+      "epoch": 0.18866666666666668,
+      "grad_norm": 30.25,
+      "grad_norm_var": 8.199934895833334,
+      "learning_rate": 9.14860068338133e-05,
+      "loss": 7.1022,
+      "loss/crossentropy": 1.7420377135276794,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16735665872693062,
+      "step": 1132
+    },
+    {
+      "epoch": 0.18883333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 8.2025390625,
+      "learning_rate": 9.147138803923416e-05,
+      "loss": 7.1235,
+      "loss/crossentropy": 2.0508846938610077,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19641846045851707,
+      "step": 1133
+    },
+    {
+      "epoch": 0.189,
+      "grad_norm": 31.875,
+      "grad_norm_var": 8.20625,
+      "learning_rate": 9.145675787503878e-05,
+      "loss": 7.0063,
+      "loss/crossentropy": 1.808214545249939,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18140819668769836,
+      "step": 1134
+    },
+    {
+      "epoch": 0.18916666666666668,
+      "grad_norm": 29.125,
+      "grad_norm_var": 8.5244140625,
+      "learning_rate": 9.14421163452381e-05,
+      "loss": 7.0805,
+      "loss/crossentropy": 1.9267567694187164,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.173908282071352,
+      "step": 1135
+    },
+    {
+      "epoch": 0.18933333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 8.740625,
+      "learning_rate": 9.142746345384619e-05,
+      "loss": 7.2291,
+      "loss/crossentropy": 2.3334193527698517,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19091741740703583,
+      "step": 1136
+    },
+    {
+      "epoch": 0.1895,
+      "grad_norm": 30.25,
+      "grad_norm_var": 8.8,
+      "learning_rate": 9.141279920488021e-05,
+      "loss": 7.2534,
+      "loss/crossentropy": 1.7937891334295273,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21443713083863258,
+      "step": 1137
+    },
+    {
+      "epoch": 0.18966666666666668,
+      "grad_norm": 32.75,
+      "grad_norm_var": 8.7625,
+      "learning_rate": 9.139812360236046e-05,
+      "loss": 6.8188,
+      "loss/crossentropy": 1.6196817606687546,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17185498401522636,
+      "step": 1138
+    },
+    {
+      "epoch": 0.18983333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 8.7791015625,
+      "learning_rate": 9.138343665031033e-05,
+      "loss": 7.1766,
+      "loss/crossentropy": 2.123296618461609,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18247639387845993,
+      "step": 1139
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 33.25,
+      "grad_norm_var": 6.028059895833334,
+      "learning_rate": 9.136873835275633e-05,
+      "loss": 7.4101,
+      "loss/crossentropy": 2.2366713285446167,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18965138494968414,
+      "step": 1140
+    },
+    {
+      "epoch": 0.19016666666666668,
+      "grad_norm": 39.0,
+      "grad_norm_var": 7.006184895833333,
+      "learning_rate": 9.135402871372808e-05,
+      "loss": 7.2896,
+      "loss/crossentropy": 1.8472765684127808,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2029191255569458,
+      "step": 1141
+    },
+    {
+      "epoch": 0.19033333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 6.205989583333333,
+      "learning_rate": 9.133930773725834e-05,
+      "loss": 7.0556,
+      "loss/crossentropy": 1.9929293394088745,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18635635823011398,
+      "step": 1142
+    },
+    {
+      "epoch": 0.1905,
+      "grad_norm": 30.125,
+      "grad_norm_var": 6.286458333333333,
+      "learning_rate": 9.132457542738292e-05,
+      "loss": 7.3807,
+      "loss/crossentropy": 2.107265204191208,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21145625039935112,
+      "step": 1143
+    },
+    {
+      "epoch": 0.19066666666666668,
+      "grad_norm": 30.25,
+      "grad_norm_var": 6.1119140625,
+      "learning_rate": 9.130983178814077e-05,
+      "loss": 6.9327,
+      "loss/crossentropy": 1.7907715737819672,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.165031298995018,
+      "step": 1144
+    },
+    {
+      "epoch": 0.19083333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 5.867643229166666,
+      "learning_rate": 9.129507682357394e-05,
+      "loss": 6.9596,
+      "loss/crossentropy": 1.617067962884903,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16396093368530273,
+      "step": 1145
+    },
+    {
+      "epoch": 0.191,
+      "grad_norm": 30.125,
+      "grad_norm_var": 5.867643229166666,
+      "learning_rate": 9.128031053772759e-05,
+      "loss": 7.1567,
+      "loss/crossentropy": 2.042255163192749,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17112840339541435,
+      "step": 1146
+    },
+    {
+      "epoch": 0.19116666666666668,
+      "grad_norm": 30.875,
+      "grad_norm_var": 5.741666666666666,
+      "learning_rate": 9.126553293464998e-05,
+      "loss": 6.9984,
+      "loss/crossentropy": 1.9593432247638702,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16826363280415535,
+      "step": 1147
+    },
+    {
+      "epoch": 0.19133333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 5.995572916666666,
+      "learning_rate": 9.125074401839249e-05,
+      "loss": 6.9948,
+      "loss/crossentropy": 2.2592301964759827,
+      "loss/hidden": 3.125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16658739745616913,
+      "step": 1148
+    },
+    {
+      "epoch": 0.1915,
+      "grad_norm": 30.125,
+      "grad_norm_var": 6.008333333333334,
+      "learning_rate": 9.123594379300955e-05,
+      "loss": 7.1862,
+      "loss/crossentropy": 1.6581338793039322,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16667338833212852,
+      "step": 1149
+    },
+    {
+      "epoch": 0.19166666666666668,
+      "grad_norm": 30.5,
+      "grad_norm_var": 5.966080729166666,
+      "learning_rate": 9.122113226255877e-05,
+      "loss": 7.3388,
+      "loss/crossentropy": 1.9912242591381073,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1910179853439331,
+      "step": 1150
+    },
+    {
+      "epoch": 0.19183333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 5.748893229166667,
+      "learning_rate": 9.120630943110077e-05,
+      "loss": 7.321,
+      "loss/crossentropy": 1.9995951056480408,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19255974143743515,
+      "step": 1151
+    },
+    {
+      "epoch": 0.192,
+      "grad_norm": 30.75,
+      "grad_norm_var": 5.567708333333333,
+      "learning_rate": 9.119147530269937e-05,
+      "loss": 7.077,
+      "loss/crossentropy": 1.5149144232273102,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20727255195379257,
+      "step": 1152
+    },
+    {
+      "epoch": 0.19216666666666668,
+      "grad_norm": 30.75,
+      "grad_norm_var": 5.527083333333334,
+      "learning_rate": 9.117662988142138e-05,
+      "loss": 7.0772,
+      "loss/crossentropy": 1.6812386810779572,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1707703210413456,
+      "step": 1153
+    },
+    {
+      "epoch": 0.19233333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 5.439583333333333,
+      "learning_rate": 9.116177317133676e-05,
+      "loss": 7.2581,
+      "loss/crossentropy": 1.8039521276950836,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16479618102312088,
+      "step": 1154
+    },
+    {
+      "epoch": 0.1925,
+      "grad_norm": 29.625,
+      "grad_norm_var": 5.530989583333334,
+      "learning_rate": 9.114690517651859e-05,
+      "loss": 7.1133,
+      "loss/crossentropy": 2.065375417470932,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2033587470650673,
+      "step": 1155
+    },
+    {
+      "epoch": 0.19266666666666668,
+      "grad_norm": 27.875,
+      "grad_norm_var": 5.645768229166666,
+      "learning_rate": 9.1132025901043e-05,
+      "loss": 7.2053,
+      "loss/crossentropy": 1.9802367985248566,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18251686543226242,
+      "step": 1156
+    },
+    {
+      "epoch": 0.19283333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 0.62265625,
+      "learning_rate": 9.111713534898922e-05,
+      "loss": 7.0329,
+      "loss/crossentropy": 2.3393149971961975,
+      "loss/hidden": 3.125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17230282351374626,
+      "step": 1157
+    },
+    {
+      "epoch": 0.193,
+      "grad_norm": 30.625,
+      "grad_norm_var": 0.6197265625,
+      "learning_rate": 9.110223352443958e-05,
+      "loss": 7.2851,
+      "loss/crossentropy": 1.9267436563968658,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28668438643217087,
+      "step": 1158
+    },
+    {
+      "epoch": 0.19316666666666665,
+      "grad_norm": 31.375,
+      "grad_norm_var": 0.7186848958333333,
+      "learning_rate": 9.108732043147952e-05,
+      "loss": 7.2434,
+      "loss/crossentropy": 2.1474104523658752,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18150009214878082,
+      "step": 1159
+    },
+    {
+      "epoch": 0.19333333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 0.846875,
+      "learning_rate": 9.107239607419753e-05,
+      "loss": 7.2608,
+      "loss/crossentropy": 1.780912846326828,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2125644087791443,
+      "step": 1160
+    },
+    {
+      "epoch": 0.1935,
+      "grad_norm": 30.25,
+      "grad_norm_var": 0.84140625,
+      "learning_rate": 9.105746045668521e-05,
+      "loss": 7.1664,
+      "loss/crossentropy": 2.1321937143802643,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21302882954478264,
+      "step": 1161
+    },
+    {
+      "epoch": 0.19366666666666665,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.0749348958333333,
+      "learning_rate": 9.104251358303724e-05,
+      "loss": 7.2137,
+      "loss/crossentropy": 2.0404190123081207,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18157971650362015,
+      "step": 1162
+    },
+    {
+      "epoch": 0.19383333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.1624348958333333,
+      "learning_rate": 9.102755545735141e-05,
+      "loss": 7.2329,
+      "loss/crossentropy": 2.299391061067581,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19177229702472687,
+      "step": 1163
+    },
+    {
+      "epoch": 0.194,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.3119140625,
+      "learning_rate": 9.101258608372856e-05,
+      "loss": 6.7745,
+      "loss/crossentropy": 1.7725453078746796,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18406648188829422,
+      "step": 1164
+    },
+    {
+      "epoch": 0.19416666666666665,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.4768229166666667,
+      "learning_rate": 9.099760546627261e-05,
+      "loss": 7.4388,
+      "loss/crossentropy": 1.3108851462602615,
+      "loss/hidden": 3.73828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2418411336839199,
+      "step": 1165
+    },
+    {
+      "epoch": 0.19433333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.475,
+      "learning_rate": 9.098261360909064e-05,
+      "loss": 7.3005,
+      "loss/crossentropy": 1.936125248670578,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19612173736095428,
+      "step": 1166
+    },
+    {
+      "epoch": 0.1945,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.6791015625,
+      "learning_rate": 9.096761051629268e-05,
+      "loss": 7.1562,
+      "loss/crossentropy": 1.4025073796510696,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24204819649457932,
+      "step": 1167
+    },
+    {
+      "epoch": 0.19466666666666665,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.8577473958333333,
+      "learning_rate": 9.095259619199197e-05,
+      "loss": 7.113,
+      "loss/crossentropy": 1.9666818380355835,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15561643242835999,
+      "step": 1168
+    },
+    {
+      "epoch": 0.19483333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.8864583333333333,
+      "learning_rate": 9.093757064030473e-05,
+      "loss": 7.1962,
+      "loss/crossentropy": 1.8951291888952255,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16155934892594814,
+      "step": 1169
+    },
+    {
+      "epoch": 0.195,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.9614583333333333,
+      "learning_rate": 9.092253386535032e-05,
+      "loss": 7.0512,
+      "loss/crossentropy": 2.00069323182106,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21078437566757202,
+      "step": 1170
+    },
+    {
+      "epoch": 0.19516666666666665,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.9337890625,
+      "learning_rate": 9.090748587125118e-05,
+      "loss": 7.3379,
+      "loss/crossentropy": 1.8147540092468262,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18162329122424126,
+      "step": 1171
+    },
+    {
+      "epoch": 0.19533333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.0760416666666666,
+      "learning_rate": 9.089242666213276e-05,
+      "loss": 7.1343,
+      "loss/crossentropy": 1.1780397295951843,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18952900543808937,
+      "step": 1172
+    },
+    {
+      "epoch": 0.1955,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.0559895833333333,
+      "learning_rate": 9.087735624212365e-05,
+      "loss": 7.5651,
+      "loss/crossentropy": 1.9671280682086945,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19969886913895607,
+      "step": 1173
+    },
+    {
+      "epoch": 0.19566666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.0927083333333334,
+      "learning_rate": 9.08622746153555e-05,
+      "loss": 7.2095,
+      "loss/crossentropy": 1.8983251750469208,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18222284317016602,
+      "step": 1174
+    },
+    {
+      "epoch": 0.19583333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.10390625,
+      "learning_rate": 9.084718178596301e-05,
+      "loss": 7.136,
+      "loss/crossentropy": 1.6355826556682587,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16758504137396812,
+      "step": 1175
+    },
+    {
+      "epoch": 0.196,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.1041015625,
+      "learning_rate": 9.083207775808396e-05,
+      "loss": 7.0999,
+      "loss/crossentropy": 1.8894365727901459,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.148922236636281,
+      "step": 1176
+    },
+    {
+      "epoch": 0.19616666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 1.1921223958333333,
+      "learning_rate": 9.081696253585921e-05,
+      "loss": 7.1252,
+      "loss/crossentropy": 1.5044038444757462,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1827947534620762,
+      "step": 1177
+    },
+    {
+      "epoch": 0.19633333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.3181640625,
+      "learning_rate": 9.080183612343268e-05,
+      "loss": 7.2547,
+      "loss/crossentropy": 2.507762134075165,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19897131621837616,
+      "step": 1178
+    },
+    {
+      "epoch": 0.1965,
+      "grad_norm": 35.0,
+      "grad_norm_var": 1.3518229166666667,
+      "learning_rate": 9.078669852495138e-05,
+      "loss": 7.4047,
+      "loss/crossentropy": 2.420338422060013,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16582558676600456,
+      "step": 1179
+    },
+    {
+      "epoch": 0.19666666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.3747395833333333,
+      "learning_rate": 9.077154974456534e-05,
+      "loss": 7.5926,
+      "loss/crossentropy": 1.992502897977829,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19241167977452278,
+      "step": 1180
+    },
+    {
+      "epoch": 0.19683333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.5160807291666667,
+      "learning_rate": 9.075638978642771e-05,
+      "loss": 7.1995,
+      "loss/crossentropy": 2.0394376814365387,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21910645067691803,
+      "step": 1181
+    },
+    {
+      "epoch": 0.197,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.4223307291666667,
+      "learning_rate": 9.074121865469467e-05,
+      "loss": 7.1043,
+      "loss/crossentropy": 1.9757541120052338,
+      "loss/hidden": 3.14453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14330443926155567,
+      "step": 1182
+    },
+    {
+      "epoch": 0.19716666666666666,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.4238932291666666,
+      "learning_rate": 9.072603635352548e-05,
+      "loss": 7.2873,
+      "loss/crossentropy": 1.9312621057033539,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2252270206809044,
+      "step": 1183
+    },
+    {
+      "epoch": 0.19733333333333333,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.2197265625,
+      "learning_rate": 9.071084288708243e-05,
+      "loss": 7.0756,
+      "loss/crossentropy": 2.153540700674057,
+      "loss/hidden": 3.10546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15975895151495934,
+      "step": 1184
+    },
+    {
+      "epoch": 0.1975,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.2514973958333333,
+      "learning_rate": 9.069563825953092e-05,
+      "loss": 7.3914,
+      "loss/crossentropy": 2.133886218070984,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17746930941939354,
+      "step": 1185
+    },
+    {
+      "epoch": 0.19766666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.378059895833333,
+      "learning_rate": 9.068042247503936e-05,
+      "loss": 7.07,
+      "loss/crossentropy": 1.7272471189498901,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17105481401085854,
+      "step": 1186
+    },
+    {
+      "epoch": 0.19783333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.3968098958333335,
+      "learning_rate": 9.066519553777926e-05,
+      "loss": 7.6655,
+      "loss/crossentropy": 1.818149596452713,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1841256469488144,
+      "step": 1187
+    },
+    {
+      "epoch": 0.198,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.4931640625,
+      "learning_rate": 9.064995745192518e-05,
+      "loss": 7.251,
+      "loss/crossentropy": 2.0524714291095734,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2640038914978504,
+      "step": 1188
+    },
+    {
+      "epoch": 0.19816666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.482291666666667,
+      "learning_rate": 9.06347082216547e-05,
+      "loss": 7.2448,
+      "loss/crossentropy": 1.6076558083295822,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15441237948834896,
+      "step": 1189
+    },
+    {
+      "epoch": 0.19833333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.3587890625,
+      "learning_rate": 9.061944785114851e-05,
+      "loss": 7.2225,
+      "loss/crossentropy": 1.9840098321437836,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18997471034526825,
+      "step": 1190
+    },
+    {
+      "epoch": 0.1985,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.5853515625,
+      "learning_rate": 9.060417634459031e-05,
+      "loss": 7.2616,
+      "loss/crossentropy": 2.2869492769241333,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19546443223953247,
+      "step": 1191
+    },
+    {
+      "epoch": 0.19866666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.5999348958333335,
+      "learning_rate": 9.058889370616689e-05,
+      "loss": 7.1566,
+      "loss/crossentropy": 1.668163388967514,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19207096099853516,
+      "step": 1192
+    },
+    {
+      "epoch": 0.19883333333333333,
+      "grad_norm": 27.875,
+      "grad_norm_var": 3.3059895833333335,
+      "learning_rate": 9.057359994006806e-05,
+      "loss": 6.8752,
+      "loss/crossentropy": 2.23806294798851,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1764606013894081,
+      "step": 1193
+    },
+    {
+      "epoch": 0.199,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.2093098958333335,
+      "learning_rate": 9.055829505048667e-05,
+      "loss": 7.2313,
+      "loss/crossentropy": 2.292779952287674,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1787671446800232,
+      "step": 1194
+    },
+    {
+      "epoch": 0.19916666666666666,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.321875,
+      "learning_rate": 9.054297904161868e-05,
+      "loss": 6.981,
+      "loss/crossentropy": 1.9182047843933105,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15541740506887436,
+      "step": 1195
+    },
+    {
+      "epoch": 0.19933333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.2979166666666666,
+      "learning_rate": 9.052765191766304e-05,
+      "loss": 7.14,
+      "loss/crossentropy": 2.5357794165611267,
+      "loss/hidden": 2.99609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15709705278277397,
+      "step": 1196
+    },
+    {
+      "epoch": 0.1995,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.5171223958333333,
+      "learning_rate": 9.051231368282177e-05,
+      "loss": 6.8715,
+      "loss/crossentropy": 1.8068018555641174,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1656156349927187,
+      "step": 1197
+    },
+    {
+      "epoch": 0.19966666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.5869140625,
+      "learning_rate": 9.049696434129994e-05,
+      "loss": 7.3377,
+      "loss/crossentropy": 2.0673095881938934,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20430799946188927,
+      "step": 1198
+    },
+    {
+      "epoch": 0.19983333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.5869140625,
+      "learning_rate": 9.048160389730566e-05,
+      "loss": 7.2533,
+      "loss/crossentropy": 1.7730631232261658,
+      "loss/hidden": 3.09375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18325388059020042,
+      "step": 1199
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.220572916666667,
+      "learning_rate": 9.046623235505007e-05,
+      "loss": 7.1952,
+      "loss/crossentropy": 1.7203770875930786,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1648796908557415,
+      "step": 1200
+    },
+    {
+      "epoch": 0.20016666666666666,
+      "grad_norm": 35.75,
+      "grad_norm_var": 3.682747395833333,
+      "learning_rate": 9.045084971874738e-05,
+      "loss": 7.1106,
+      "loss/crossentropy": 1.5943018049001694,
+      "loss/hidden": 3.73046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2411169558763504,
+      "step": 1201
+    },
+    {
+      "epoch": 0.20033333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.321875,
+      "learning_rate": 9.043545599261481e-05,
+      "loss": 7.2039,
+      "loss/crossentropy": 1.7558506280183792,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1825847439467907,
+      "step": 1202
+    },
+    {
+      "epoch": 0.2005,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.421875,
+      "learning_rate": 9.042005118087267e-05,
+      "loss": 7.2046,
+      "loss/crossentropy": 2.4395557641983032,
+      "loss/hidden": 3.109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16663893684744835,
+      "step": 1203
+    },
+    {
+      "epoch": 0.20066666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.1354166666666665,
+      "learning_rate": 9.040463528774423e-05,
+      "loss": 7.0709,
+      "loss/crossentropy": 1.9601179361343384,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1911269947886467,
+      "step": 1204
+    },
+    {
+      "epoch": 0.20083333333333334,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.1962890625,
+      "learning_rate": 9.038920831745587e-05,
+      "loss": 7.3041,
+      "loss/crossentropy": 1.9899678826332092,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19194591417908669,
+      "step": 1205
+    },
+    {
+      "epoch": 0.201,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.215625,
+      "learning_rate": 9.0373770274237e-05,
+      "loss": 7.0464,
+      "loss/crossentropy": 1.7225709557533264,
+      "loss/hidden": 3.82421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27960826456546783,
+      "step": 1206
+    },
+    {
+      "epoch": 0.20116666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.1791015625,
+      "learning_rate": 9.035832116232001e-05,
+      "loss": 7.171,
+      "loss/crossentropy": 2.1983217895030975,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16688163951039314,
+      "step": 1207
+    },
+    {
+      "epoch": 0.20133333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.1681640625,
+      "learning_rate": 9.03428609859404e-05,
+      "loss": 7.2815,
+      "loss/crossentropy": 2.3588224053382874,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1658637374639511,
+      "step": 1208
+    },
+    {
+      "epoch": 0.2015,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.6645833333333333,
+      "learning_rate": 9.032738974933664e-05,
+      "loss": 7.3047,
+      "loss/crossentropy": 1.8783065527677536,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18188221007585526,
+      "step": 1209
+    },
+    {
+      "epoch": 0.20166666666666666,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.5738932291666665,
+      "learning_rate": 9.031190745675024e-05,
+      "loss": 6.9593,
+      "loss/crossentropy": 2.06480810046196,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1809498742222786,
+      "step": 1210
+    },
+    {
+      "epoch": 0.20183333333333334,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.6889973958333333,
+      "learning_rate": 9.029641411242579e-05,
+      "loss": 7.0686,
+      "loss/crossentropy": 1.860428050160408,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.192401971668005,
+      "step": 1211
+    },
+    {
+      "epoch": 0.202,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.8229166666666665,
+      "learning_rate": 9.028090972061088e-05,
+      "loss": 7.0726,
+      "loss/crossentropy": 1.8837276697158813,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17082658410072327,
+      "step": 1212
+    },
+    {
+      "epoch": 0.20216666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.888541666666667,
+      "learning_rate": 9.02653942855561e-05,
+      "loss": 6.9024,
+      "loss/crossentropy": 1.633382335305214,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1890016794204712,
+      "step": 1213
+    },
+    {
+      "epoch": 0.20233333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.6811848958333333,
+      "learning_rate": 9.024986781151512e-05,
+      "loss": 6.9439,
+      "loss/crossentropy": 2.256519138813019,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17529591172933578,
+      "step": 1214
+    },
+    {
+      "epoch": 0.2025,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.6947916666666667,
+      "learning_rate": 9.023433030274459e-05,
+      "loss": 7.0088,
+      "loss/crossentropy": 2.3222522139549255,
+      "loss/hidden": 3.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17771752178668976,
+      "step": 1215
+    },
+    {
+      "epoch": 0.20266666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.8988932291666667,
+      "learning_rate": 9.021878176350423e-05,
+      "loss": 7.1987,
+      "loss/crossentropy": 1.9344107806682587,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18068302050232887,
+      "step": 1216
+    },
+    {
+      "epoch": 0.20283333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.5608723958333333,
+      "learning_rate": 9.020322219805674e-05,
+      "loss": 7.2394,
+      "loss/crossentropy": 2.365844488143921,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18970517069101334,
+      "step": 1217
+    },
+    {
+      "epoch": 0.203,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.5520833333333333,
+      "learning_rate": 9.018765161066787e-05,
+      "loss": 7.0407,
+      "loss/crossentropy": 2.0648479759693146,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.186686921864748,
+      "step": 1218
+    },
+    {
+      "epoch": 0.20316666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.4559895833333334,
+      "learning_rate": 9.017207000560639e-05,
+      "loss": 6.9439,
+      "loss/crossentropy": 2.0636188983917236,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16850804537534714,
+      "step": 1219
+    },
+    {
+      "epoch": 0.20333333333333334,
+      "grad_norm": 4110417920.0,
+      "grad_norm_var": 1.0559709513111306e+18,
+      "learning_rate": 9.015647738714408e-05,
+      "loss": 8.1678,
+      "loss/crossentropy": 2.0369352400302887,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1804703250527382,
+      "step": 1220
+    },
+    {
+      "epoch": 0.2035,
+      "grad_norm": 36.0,
+      "grad_norm_var": 1.0559709511698351e+18,
+      "learning_rate": 9.014087375955573e-05,
+      "loss": 7.1254,
+      "loss/crossentropy": 1.598555475473404,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18450486287474632,
+      "step": 1221
+    },
+    {
+      "epoch": 0.20366666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.055970951062793e+18,
+      "learning_rate": 9.012525912711918e-05,
+      "loss": 7.2717,
+      "loss/crossentropy": 1.9923148155212402,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16796409338712692,
+      "step": 1222
+    },
+    {
+      "epoch": 0.20383333333333334,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.0559709510585112e+18,
+      "learning_rate": 9.010963349411529e-05,
+      "loss": 6.9809,
+      "loss/crossentropy": 2.216012567281723,
+      "loss/hidden": 3.16015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17009492218494415,
+      "step": 1223
+    },
+    {
+      "epoch": 0.204,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.0559709510670747e+18,
+      "learning_rate": 9.009399686482787e-05,
+      "loss": 7.2186,
+      "loss/crossentropy": 2.1924517452716827,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1893884502351284,
+      "step": 1224
+    },
+    {
+      "epoch": 0.20416666666666666,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.0559709511998068e+18,
+      "learning_rate": 9.007834924354383e-05,
+      "loss": 7.2308,
+      "loss/crossentropy": 2.026065707206726,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18250852823257446,
+      "step": 1225
+    },
+    {
+      "epoch": 0.20433333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.0559709511655534e+18,
+      "learning_rate": 9.006269063455304e-05,
+      "loss": 7.1048,
+      "loss/crossentropy": 1.966538518667221,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16139357537031174,
+      "step": 1226
+    },
+    {
+      "epoch": 0.2045,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.0559709511141732e+18,
+      "learning_rate": 9.00470210421484e-05,
+      "loss": 6.9923,
+      "loss/crossentropy": 1.996876299381256,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19332023710012436,
+      "step": 1227
+    },
+    {
+      "epoch": 0.20466666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.0559709509900044e+18,
+      "learning_rate": 9.003134047062579e-05,
+      "loss": 7.3161,
+      "loss/crossentropy": 1.58292955160141,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2204379104077816,
+      "step": 1228
+    },
+    {
+      "epoch": 0.20483333333333334,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.0559709510456662e+18,
+      "learning_rate": 9.001564892428415e-05,
+      "loss": 7.1692,
+      "loss/crossentropy": 2.2697527706623077,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2070455104112625,
+      "step": 1229
+    },
+    {
+      "epoch": 0.205,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.0559709511355816e+18,
+      "learning_rate": 8.999994640742543e-05,
+      "loss": 7.12,
+      "loss/crossentropy": 1.7404191195964813,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15038799494504929,
+      "step": 1230
+    },
+    {
+      "epoch": 0.20516666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.0559709510799197e+18,
+      "learning_rate": 8.998423292435454e-05,
+      "loss": 7.4999,
+      "loss/crossentropy": 2.5046131014823914,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29094114527106285,
+      "step": 1231
+    },
+    {
+      "epoch": 0.20533333333333334,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.0559709511869618e+18,
+      "learning_rate": 8.996850847937941e-05,
+      "loss": 7.2368,
+      "loss/crossentropy": 2.3570897579193115,
+      "loss/hidden": 3.12109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1517578884959221,
+      "step": 1232
+    },
+    {
+      "epoch": 0.2055,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.0559709512469055e+18,
+      "learning_rate": 8.995277307681099e-05,
+      "loss": 7.1228,
+      "loss/crossentropy": 1.816855400800705,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17459191754460335,
+      "step": 1233
+    },
+    {
+      "epoch": 0.20566666666666666,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.0559709513111308e+18,
+      "learning_rate": 8.993702672096324e-05,
+      "loss": 6.9661,
+      "loss/crossentropy": 2.1051203310489655,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16508476063609123,
+      "step": 1234
+    },
+    {
+      "epoch": 0.20583333333333334,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.055970951362511e+18,
+      "learning_rate": 8.992126941615313e-05,
+      "loss": 7.2335,
+      "loss/crossentropy": 2.2150754928588867,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16058878600597382,
+      "step": 1235
+    },
+    {
+      "epoch": 0.206,
+      "grad_norm": 29.25,
+      "grad_norm_var": 4.077083333333333,
+      "learning_rate": 8.990550116670057e-05,
+      "loss": 7.2624,
+      "loss/crossentropy": 2.2350084483623505,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1852153055369854,
+      "step": 1236
+    },
+    {
+      "epoch": 0.20616666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.3202473958333334,
+      "learning_rate": 8.988972197692855e-05,
+      "loss": 7.1009,
+      "loss/crossentropy": 2.1296741664409637,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1848640777170658,
+      "step": 1237
+    },
+    {
+      "epoch": 0.20633333333333334,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.8806640625,
+      "learning_rate": 8.987393185116302e-05,
+      "loss": 6.9537,
+      "loss/crossentropy": 1.489246904850006,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15090622007846832,
+      "step": 1238
+    },
+    {
+      "epoch": 0.2065,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.89765625,
+      "learning_rate": 8.985813079373292e-05,
+      "loss": 6.84,
+      "loss/crossentropy": 2.0008981227874756,
+      "loss/hidden": 3.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1607687808573246,
+      "step": 1239
+    },
+    {
+      "epoch": 0.20666666666666667,
+      "grad_norm": 46.5,
+      "grad_norm_var": 17.89375,
+      "learning_rate": 8.98423188089702e-05,
+      "loss": 7.2512,
+      "loss/crossentropy": 1.7720806896686554,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1796865127980709,
+      "step": 1240
+    },
+    {
+      "epoch": 0.20683333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 17.220572916666665,
+      "learning_rate": 8.982649590120982e-05,
+      "loss": 7.2703,
+      "loss/crossentropy": 1.91971156001091,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22106074169278145,
+      "step": 1241
+    },
+    {
+      "epoch": 0.207,
+      "grad_norm": 29.75,
+      "grad_norm_var": 17.449739583333333,
+      "learning_rate": 8.981066207478971e-05,
+      "loss": 7.208,
+      "loss/crossentropy": 1.9688372761011124,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19603705033659935,
+      "step": 1242
+    },
+    {
+      "epoch": 0.20716666666666667,
+      "grad_norm": 38.0,
+      "grad_norm_var": 19.942643229166666,
+      "learning_rate": 8.97948173340508e-05,
+      "loss": 7.1766,
+      "loss/crossentropy": 2.0286186039447784,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17995741963386536,
+      "step": 1243
+    },
+    {
+      "epoch": 0.20733333333333334,
+      "grad_norm": 29.125,
+      "grad_norm_var": 20.322916666666668,
+      "learning_rate": 8.977896168333702e-05,
+      "loss": 7.3233,
+      "loss/crossentropy": 1.7833180129528046,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17228223755955696,
+      "step": 1244
+    },
+    {
+      "epoch": 0.2075,
+      "grad_norm": 32.5,
+      "grad_norm_var": 20.357747395833332,
+      "learning_rate": 8.976309512699528e-05,
+      "loss": 7.1613,
+      "loss/crossentropy": 1.5690025091171265,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16183597594499588,
+      "step": 1245
+    },
+    {
+      "epoch": 0.20766666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 20.088997395833335,
+      "learning_rate": 8.97472176693755e-05,
+      "loss": 7.1242,
+      "loss/crossentropy": 1.9329344630241394,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16291601583361626,
+      "step": 1246
+    },
+    {
+      "epoch": 0.20783333333333334,
+      "grad_norm": 27.5,
+      "grad_norm_var": 21.256184895833332,
+      "learning_rate": 8.973132931483057e-05,
+      "loss": 7.1924,
+      "loss/crossentropy": 2.416533887386322,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2108389437198639,
+      "step": 1247
+    },
+    {
+      "epoch": 0.208,
+      "grad_norm": 30.875,
+      "grad_norm_var": 21.0947265625,
+      "learning_rate": 8.971543006771636e-05,
+      "loss": 6.8263,
+      "loss/crossentropy": 1.3490639328956604,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1566382423043251,
+      "step": 1248
+    },
+    {
+      "epoch": 0.20816666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 21.0759765625,
+      "learning_rate": 8.969951993239177e-05,
+      "loss": 7.2188,
+      "loss/crossentropy": 2.0470789074897766,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18885283544659615,
+      "step": 1249
+    },
+    {
+      "epoch": 0.20833333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 20.783333333333335,
+      "learning_rate": 8.968359891321862e-05,
+      "loss": 7.3137,
+      "loss/crossentropy": 1.695669412612915,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16367409750819206,
+      "step": 1250
+    },
+    {
+      "epoch": 0.2085,
+      "grad_norm": 29.0,
+      "grad_norm_var": 20.783333333333335,
+      "learning_rate": 8.966766701456177e-05,
+      "loss": 7.21,
+      "loss/crossentropy": 1.8349643051624298,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15510903298854828,
+      "step": 1251
+    },
+    {
+      "epoch": 0.20866666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 20.42265625,
+      "learning_rate": 8.965172424078902e-05,
+      "loss": 6.8593,
+      "loss/crossentropy": 2.2177212834358215,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23635417222976685,
+      "step": 1252
+    },
+    {
+      "epoch": 0.20883333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 20.431184895833333,
+      "learning_rate": 8.963577059627118e-05,
+      "loss": 7.1913,
+      "loss/crossentropy": 2.0282241106033325,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1659335009753704,
+      "step": 1253
+    },
+    {
+      "epoch": 0.209,
+      "grad_norm": 30.875,
+      "grad_norm_var": 20.480989583333333,
+      "learning_rate": 8.961980608538203e-05,
+      "loss": 7.1312,
+      "loss/crossentropy": 1.821393758058548,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17700910940766335,
+      "step": 1254
+    },
+    {
+      "epoch": 0.20916666666666667,
+      "grad_norm": 28.125,
+      "grad_norm_var": 21.212434895833333,
+      "learning_rate": 8.960383071249836e-05,
+      "loss": 6.9964,
+      "loss/crossentropy": 1.136364832520485,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16633517667651176,
+      "step": 1255
+    },
+    {
+      "epoch": 0.20933333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 6.116080729166667,
+      "learning_rate": 8.958784448199987e-05,
+      "loss": 6.9569,
+      "loss/crossentropy": 1.9695285856723785,
+      "loss/hidden": 3.11328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16960721462965012,
+      "step": 1256
+    },
+    {
+      "epoch": 0.2095,
+      "grad_norm": 30.375,
+      "grad_norm_var": 6.134830729166667,
+      "learning_rate": 8.95718473982693e-05,
+      "loss": 7.2848,
+      "loss/crossentropy": 1.9908243119716644,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20242078602313995,
+      "step": 1257
+    },
+    {
+      "epoch": 0.20966666666666667,
+      "grad_norm": 26.875,
+      "grad_norm_var": 7.09765625,
+      "learning_rate": 8.955583946569233e-05,
+      "loss": 7.1128,
+      "loss/crossentropy": 2.2287910878658295,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15669751167297363,
+      "step": 1258
+    },
+    {
+      "epoch": 0.20983333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.359375,
+      "learning_rate": 8.95398206886576e-05,
+      "loss": 7.0646,
+      "loss/crossentropy": 1.508285030722618,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19778892770409584,
+      "step": 1259
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.2650390625,
+      "learning_rate": 8.95237910715568e-05,
+      "loss": 7.1306,
+      "loss/crossentropy": 2.029755800962448,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2042919136583805,
+      "step": 1260
+    },
+    {
+      "epoch": 0.21016666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.964583333333333,
+      "learning_rate": 8.950775061878453e-05,
+      "loss": 7.0133,
+      "loss/crossentropy": 1.8159136176109314,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1695379577577114,
+      "step": 1261
+    },
+    {
+      "epoch": 0.21033333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.9311848958333333,
+      "learning_rate": 8.949169933473833e-05,
+      "loss": 7.0799,
+      "loss/crossentropy": 2.2301661372184753,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16201428323984146,
+      "step": 1262
+    },
+    {
+      "epoch": 0.2105,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.482291666666667,
+      "learning_rate": 8.94756372238188e-05,
+      "loss": 7.2399,
+      "loss/crossentropy": 2.077402353286743,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1858767345547676,
+      "step": 1263
+    },
+    {
+      "epoch": 0.21066666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.5455729166666665,
+      "learning_rate": 8.945956429042943e-05,
+      "loss": 7.1386,
+      "loss/crossentropy": 2.3105130195617676,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1865912601351738,
+      "step": 1264
+    },
+    {
+      "epoch": 0.21083333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.2379557291666665,
+      "learning_rate": 8.944348053897671e-05,
+      "loss": 6.9133,
+      "loss/crossentropy": 2.202511876821518,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1948540285229683,
+      "step": 1265
+    },
+    {
+      "epoch": 0.211,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.7122395833333333,
+      "learning_rate": 8.94273859738701e-05,
+      "loss": 7.4478,
+      "loss/crossentropy": 1.9073927849531174,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1643068753182888,
+      "step": 1266
+    },
+    {
+      "epoch": 0.21116666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.5869140625,
+      "learning_rate": 8.941128059952201e-05,
+      "loss": 7.1462,
+      "loss/crossentropy": 1.9705713391304016,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20388410612940788,
+      "step": 1267
+    },
+    {
+      "epoch": 0.21133333333333335,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.6041666666666667,
+      "learning_rate": 8.939516442034781e-05,
+      "loss": 7.0554,
+      "loss/crossentropy": 1.5540541112422943,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19841605052351952,
+      "step": 1268
+    },
+    {
+      "epoch": 0.2115,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.5863932291666667,
+      "learning_rate": 8.937903744076587e-05,
+      "loss": 7.0384,
+      "loss/crossentropy": 1.8669461160898209,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17107466235756874,
+      "step": 1269
+    },
+    {
+      "epoch": 0.21166666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.099739583333333,
+      "learning_rate": 8.936289966519746e-05,
+      "loss": 7.0628,
+      "loss/crossentropy": 2.293793559074402,
+      "loss/hidden": 3.10546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16051578521728516,
+      "step": 1270
+    },
+    {
+      "epoch": 0.21183333333333335,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.7218098958333334,
+      "learning_rate": 8.934675109806688e-05,
+      "loss": 6.8787,
+      "loss/crossentropy": 1.7225153595209122,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14841681346297264,
+      "step": 1271
+    },
+    {
+      "epoch": 0.212,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.9280598958333333,
+      "learning_rate": 8.933059174380131e-05,
+      "loss": 6.9817,
+      "loss/crossentropy": 1.9311038553714752,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17153063416481018,
+      "step": 1272
+    },
+    {
+      "epoch": 0.21216666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.9749348958333333,
+      "learning_rate": 8.931442160683094e-05,
+      "loss": 6.9697,
+      "loss/crossentropy": 2.2986189126968384,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21446756273508072,
+      "step": 1273
+    },
+    {
+      "epoch": 0.21233333333333335,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.09765625,
+      "learning_rate": 8.929824069158894e-05,
+      "loss": 7.0315,
+      "loss/crossentropy": 1.823767900466919,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16026857122778893,
+      "step": 1274
+    },
+    {
+      "epoch": 0.2125,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.1603515625,
+      "learning_rate": 8.928204900251136e-05,
+      "loss": 6.978,
+      "loss/crossentropy": 1.780413269996643,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1813862845301628,
+      "step": 1275
+    },
+    {
+      "epoch": 0.21266666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.1634765625,
+      "learning_rate": 8.926584654403724e-05,
+      "loss": 7.1789,
+      "loss/crossentropy": 2.35905060172081,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1954832375049591,
+      "step": 1276
+    },
+    {
+      "epoch": 0.21283333333333335,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.9134765625,
+      "learning_rate": 8.924963332060863e-05,
+      "loss": 6.8761,
+      "loss/crossentropy": 2.448783278465271,
+      "loss/hidden": 3.0625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16067909821867943,
+      "step": 1277
+    },
+    {
+      "epoch": 0.213,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.9177083333333333,
+      "learning_rate": 8.92334093366704e-05,
+      "loss": 7.2586,
+      "loss/crossentropy": 2.08617702126503,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1942312866449356,
+      "step": 1278
+    },
+    {
+      "epoch": 0.21316666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.003059895833333,
+      "learning_rate": 8.92171745966705e-05,
+      "loss": 6.9743,
+      "loss/crossentropy": 2.039038360118866,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16942651197314262,
+      "step": 1279
+    },
+    {
+      "epoch": 0.21333333333333335,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.890625,
+      "learning_rate": 8.920092910505977e-05,
+      "loss": 7.0371,
+      "loss/crossentropy": 1.5313948392868042,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17466925829648972,
+      "step": 1280
+    },
+    {
+      "epoch": 0.2135,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.9143229166666667,
+      "learning_rate": 8.9184672866292e-05,
+      "loss": 7.1395,
+      "loss/crossentropy": 1.695930540561676,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1709824688732624,
+      "step": 1281
+    },
+    {
+      "epoch": 0.21366666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.4530598958333334,
+      "learning_rate": 8.916840588482392e-05,
+      "loss": 7.2622,
+      "loss/crossentropy": 2.192231595516205,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2439628392457962,
+      "step": 1282
+    },
+    {
+      "epoch": 0.21383333333333332,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.4530598958333334,
+      "learning_rate": 8.915212816511522e-05,
+      "loss": 7.2052,
+      "loss/crossentropy": 1.8892233073711395,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18108197301626205,
+      "step": 1283
+    },
+    {
+      "epoch": 0.214,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.653580729166667,
+      "learning_rate": 8.913583971162852e-05,
+      "loss": 7.0813,
+      "loss/crossentropy": 1.6332352757453918,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15463052317500114,
+      "step": 1284
+    },
+    {
+      "epoch": 0.21416666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.758072916666667,
+      "learning_rate": 8.91195405288294e-05,
+      "loss": 7.0533,
+      "loss/crossentropy": 1.7594351768493652,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17090722918510437,
+      "step": 1285
+    },
+    {
+      "epoch": 0.21433333333333332,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.218489583333333,
+      "learning_rate": 8.910323062118639e-05,
+      "loss": 7.0658,
+      "loss/crossentropy": 1.8605065047740936,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20274737104773521,
+      "step": 1286
+    },
+    {
+      "epoch": 0.2145,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.262434895833333,
+      "learning_rate": 8.908690999317093e-05,
+      "loss": 7.0321,
+      "loss/crossentropy": 1.736699417233467,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17047365754842758,
+      "step": 1287
+    },
+    {
+      "epoch": 0.21466666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.113997395833333,
+      "learning_rate": 8.90705786492574e-05,
+      "loss": 7.2169,
+      "loss/crossentropy": 1.8790197670459747,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19407998397946358,
+      "step": 1288
+    },
+    {
+      "epoch": 0.21483333333333332,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.44765625,
+      "learning_rate": 8.905423659392316e-05,
+      "loss": 6.9602,
+      "loss/crossentropy": 1.623477280139923,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1775852032005787,
+      "step": 1289
+    },
+    {
+      "epoch": 0.215,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.6768229166666666,
+      "learning_rate": 8.903788383164846e-05,
+      "loss": 7.1997,
+      "loss/crossentropy": 1.8577570766210556,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1756720244884491,
+      "step": 1290
+    },
+    {
+      "epoch": 0.21516666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 3.3509765625,
+      "learning_rate": 8.90215203669165e-05,
+      "loss": 6.9301,
+      "loss/crossentropy": 1.453979104757309,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1967424340546131,
+      "step": 1291
+    },
+    {
+      "epoch": 0.21533333333333332,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.465559895833333,
+      "learning_rate": 8.90051462042134e-05,
+      "loss": 6.8198,
+      "loss/crossentropy": 1.7568388879299164,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16247447207570076,
+      "step": 1292
+    },
+    {
+      "epoch": 0.2155,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.72890625,
+      "learning_rate": 8.898876134802826e-05,
+      "loss": 7.2477,
+      "loss/crossentropy": 2.023082450032234,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22672132402658463,
+      "step": 1293
+    },
+    {
+      "epoch": 0.21566666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.70390625,
+      "learning_rate": 8.897236580285308e-05,
+      "loss": 7.0852,
+      "loss/crossentropy": 1.9821708798408508,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17430612444877625,
+      "step": 1294
+    },
+    {
+      "epoch": 0.21583333333333332,
+      "grad_norm": 3456106496.0,
+      "grad_norm_var": 7.46541993790031e+17,
+      "learning_rate": 8.895595957318277e-05,
+      "loss": 7.889,
+      "loss/crossentropy": 2.0135136544704437,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17147020250558853,
+      "step": 1295
+    },
+    {
+      "epoch": 0.216,
+      "grad_norm": 34.5,
+      "grad_norm_var": 7.465419936676273e+17,
+      "learning_rate": 8.893954266351521e-05,
+      "loss": 7.2871,
+      "loss/crossentropy": 1.8673216104507446,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1795753724873066,
+      "step": 1296
+    },
+    {
+      "epoch": 0.21616666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 7.465419936820276e+17,
+      "learning_rate": 8.892311507835119e-05,
+      "loss": 6.9809,
+      "loss/crossentropy": 1.622310757637024,
+      "loss/hidden": 3.13671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1428094319999218,
+      "step": 1297
+    },
+    {
+      "epoch": 0.21633333333333332,
+      "grad_norm": 31.0,
+      "grad_norm_var": 7.465419937468297e+17,
+      "learning_rate": 8.890667682219439e-05,
+      "loss": 7.2261,
+      "loss/crossentropy": 1.6714694499969482,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1609114594757557,
+      "step": 1298
+    },
+    {
+      "epoch": 0.2165,
+      "grad_norm": 31.25,
+      "grad_norm_var": 7.465419937216289e+17,
+      "learning_rate": 8.889022789955151e-05,
+      "loss": 7.1802,
+      "loss/crossentropy": 1.7925153225660324,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1680547334253788,
+      "step": 1299
+    },
+    {
+      "epoch": 0.21666666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 7.465419936892279e+17,
+      "learning_rate": 8.887376831493205e-05,
+      "loss": 7.0988,
+      "loss/crossentropy": 1.8799127638339996,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18989857286214828,
+      "step": 1300
+    },
+    {
+      "epoch": 0.21683333333333332,
+      "grad_norm": 31.5,
+      "grad_norm_var": 7.465419937036283e+17,
+      "learning_rate": 8.885729807284856e-05,
+      "loss": 7.128,
+      "loss/crossentropy": 2.1720742881298065,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1744670458137989,
+      "step": 1301
+    },
+    {
+      "epoch": 0.217,
+      "grad_norm": 29.75,
+      "grad_norm_var": 7.465419937036283e+17,
+      "learning_rate": 8.88408171778164e-05,
+      "loss": 6.9589,
+      "loss/crossentropy": 1.8701461553573608,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17912780493497849,
+      "step": 1302
+    },
+    {
+      "epoch": 0.21716666666666667,
+      "grad_norm": 35.5,
+      "grad_norm_var": 7.465419935272229e+17,
+      "learning_rate": 8.882432563435393e-05,
+      "loss": 7.179,
+      "loss/crossentropy": 1.9762914776802063,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15697065368294716,
+      "step": 1303
+    },
+    {
+      "epoch": 0.21733333333333332,
+      "grad_norm": 33.0,
+      "grad_norm_var": 7.465419934480205e+17,
+      "learning_rate": 8.88078234469824e-05,
+      "loss": 7.2837,
+      "loss/crossentropy": 1.6388632953166962,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2365618571639061,
+      "step": 1304
+    },
+    {
+      "epoch": 0.2175,
+      "grad_norm": 27.875,
+      "grad_norm_var": 7.465419935884248e+17,
+      "learning_rate": 8.879131062022598e-05,
+      "loss": 7.055,
+      "loss/crossentropy": 2.016745537519455,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18361134082078934,
+      "step": 1305
+    },
+    {
+      "epoch": 0.21766666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 7.46541993692828e+17,
+      "learning_rate": 8.877478715861173e-05,
+      "loss": 7.0478,
+      "loss/crossentropy": 1.6777726411819458,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16228575631976128,
+      "step": 1306
+    },
+    {
+      "epoch": 0.21783333333333332,
+      "grad_norm": 27.875,
+      "grad_norm_var": 7.465419936748274e+17,
+      "learning_rate": 8.875825306666968e-05,
+      "loss": 7.0864,
+      "loss/crossentropy": 2.0001308023929596,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17227338999509811,
+      "step": 1307
+    },
+    {
+      "epoch": 0.218,
+      "grad_norm": 31.25,
+      "grad_norm_var": 7.465419936100255e+17,
+      "learning_rate": 8.874170834893272e-05,
+      "loss": 6.9015,
+      "loss/crossentropy": 2.1420152485370636,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18439536541700363,
+      "step": 1308
+    },
+    {
+      "epoch": 0.21816666666666668,
+      "grad_norm": 28.875,
+      "grad_norm_var": 7.465419936712273e+17,
+      "learning_rate": 8.872515300993669e-05,
+      "loss": 6.9725,
+      "loss/crossentropy": 1.688755840063095,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1641828790307045,
+      "step": 1309
+    },
+    {
+      "epoch": 0.21833333333333332,
+      "grad_norm": 29.25,
+      "grad_norm_var": 7.465419937072284e+17,
+      "learning_rate": 8.870858705422033e-05,
+      "loss": 7.2482,
+      "loss/crossentropy": 1.7820774614810944,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20635435730218887,
+      "step": 1310
+    },
+    {
+      "epoch": 0.2185,
+      "grad_norm": 31.0,
+      "grad_norm_var": 4.7134765625,
+      "learning_rate": 8.869201048632532e-05,
+      "loss": 7.1116,
+      "loss/crossentropy": 2.4359863102436066,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.171052236109972,
+      "step": 1311
+    },
+    {
+      "epoch": 0.21866666666666668,
+      "grad_norm": 28.0,
+      "grad_norm_var": 4.097330729166667,
+      "learning_rate": 8.867542331079617e-05,
+      "loss": 6.9759,
+      "loss/crossentropy": 1.9924459159374237,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15327489376068115,
+      "step": 1312
+    },
+    {
+      "epoch": 0.21883333333333332,
+      "grad_norm": 27.125,
+      "grad_norm_var": 4.728059895833334,
+      "learning_rate": 8.865882553218037e-05,
+      "loss": 6.8873,
+      "loss/crossentropy": 1.9398671090602875,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16717145219445229,
+      "step": 1313
+    },
+    {
+      "epoch": 0.219,
+      "grad_norm": 29.0,
+      "grad_norm_var": 4.742643229166666,
+      "learning_rate": 8.864221715502829e-05,
+      "loss": 7.1442,
+      "loss/crossentropy": 1.866513729095459,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19265495613217354,
+      "step": 1314
+    },
+    {
+      "epoch": 0.21916666666666668,
+      "grad_norm": 30.125,
+      "grad_norm_var": 4.633072916666666,
+      "learning_rate": 8.862559818389322e-05,
+      "loss": 6.8746,
+      "loss/crossentropy": 2.090180605649948,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16546057164669037,
+      "step": 1315
+    },
+    {
+      "epoch": 0.21933333333333332,
+      "grad_norm": 28.75,
+      "grad_norm_var": 4.718489583333334,
+      "learning_rate": 8.860896862333134e-05,
+      "loss": 7.1046,
+      "loss/crossentropy": 1.5242722928524017,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17156986892223358,
+      "step": 1316
+    },
+    {
+      "epoch": 0.2195,
+      "grad_norm": 31.875,
+      "grad_norm_var": 4.809309895833334,
+      "learning_rate": 8.859232847790175e-05,
+      "loss": 7.1799,
+      "loss/crossentropy": 2.030907988548279,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20265229977667332,
+      "step": 1317
+    },
+    {
+      "epoch": 0.21966666666666668,
+      "grad_norm": 32.25,
+      "grad_norm_var": 5.1556640625,
+      "learning_rate": 8.857567775216643e-05,
+      "loss": 7.0625,
+      "loss/crossentropy": 1.4974164962768555,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19254015013575554,
+      "step": 1318
+    },
+    {
+      "epoch": 0.21983333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.0353515625,
+      "learning_rate": 8.855901645069026e-05,
+      "loss": 7.0632,
+      "loss/crossentropy": 1.8093978762626648,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16958339512348175,
+      "step": 1319
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.61640625,
+      "learning_rate": 8.854234457804105e-05,
+      "loss": 7.2982,
+      "loss/crossentropy": 2.2763660848140717,
+      "loss/hidden": 3.015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15304754674434662,
+      "step": 1320
+    },
+    {
+      "epoch": 0.22016666666666668,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.5478515625,
+      "learning_rate": 8.852566213878947e-05,
+      "loss": 7.1642,
+      "loss/crossentropy": 1.9220625162124634,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25863387808203697,
+      "step": 1321
+    },
+    {
+      "epoch": 0.22033333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.7322265625,
+      "learning_rate": 8.850896913750911e-05,
+      "loss": 7.2231,
+      "loss/crossentropy": 2.248414307832718,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21608459949493408,
+      "step": 1322
+    },
+    {
+      "epoch": 0.2205,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.49140625,
+      "learning_rate": 8.849226557877646e-05,
+      "loss": 6.8391,
+      "loss/crossentropy": 2.101841062307358,
+      "loss/hidden": 3.09375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15087469294667244,
+      "step": 1323
+    },
+    {
+      "epoch": 0.22066666666666668,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.5119140625,
+      "learning_rate": 8.84755514671709e-05,
+      "loss": 7.2479,
+      "loss/crossentropy": 2.286454677581787,
+      "loss/hidden": 3.16015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1553141064941883,
+      "step": 1324
+    },
+    {
+      "epoch": 0.22083333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.436393229166667,
+      "learning_rate": 8.845882680727469e-05,
+      "loss": 6.9994,
+      "loss/crossentropy": 1.8056836128234863,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15853219106793404,
+      "step": 1325
+    },
+    {
+      "epoch": 0.221,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.4025390625,
+      "learning_rate": 8.844209160367299e-05,
+      "loss": 7.1668,
+      "loss/crossentropy": 2.037298232316971,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15290064364671707,
+      "step": 1326
+    },
+    {
+      "epoch": 0.22116666666666668,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.662955729166667,
+      "learning_rate": 8.842534586095383e-05,
+      "loss": 7.138,
+      "loss/crossentropy": 2.255521208047867,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2043117769062519,
+      "step": 1327
+    },
+    {
+      "epoch": 0.22133333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.3525390625,
+      "learning_rate": 8.840858958370819e-05,
+      "loss": 6.9868,
+      "loss/crossentropy": 2.0564877092838287,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1763647273182869,
+      "step": 1328
+    },
+    {
+      "epoch": 0.2215,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.1184895833333335,
+      "learning_rate": 8.839182277652989e-05,
+      "loss": 6.9926,
+      "loss/crossentropy": 1.7821269035339355,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1580953113734722,
+      "step": 1329
+    },
+    {
+      "epoch": 0.22166666666666668,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.2934895833333333,
+      "learning_rate": 8.837504544401561e-05,
+      "loss": 7.1472,
+      "loss/crossentropy": 2.490646868944168,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18042651563882828,
+      "step": 1330
+    },
+    {
+      "epoch": 0.22183333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.3832682291666667,
+      "learning_rate": 8.8358257590765e-05,
+      "loss": 6.8525,
+      "loss/crossentropy": 1.938695877790451,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16877496242523193,
+      "step": 1331
+    },
+    {
+      "epoch": 0.222,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.2869140625,
+      "learning_rate": 8.834145922138049e-05,
+      "loss": 7.2353,
+      "loss/crossentropy": 2.2627280354499817,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19378405436873436,
+      "step": 1332
+    },
+    {
+      "epoch": 0.22216666666666668,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.191666666666667,
+      "learning_rate": 8.832465034046749e-05,
+      "loss": 6.9395,
+      "loss/crossentropy": 2.0226730406284332,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16587254405021667,
+      "step": 1333
+    },
+    {
+      "epoch": 0.22233333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.9447916666666667,
+      "learning_rate": 8.830783095263425e-05,
+      "loss": 6.9216,
+      "loss/crossentropy": 1.714574009180069,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16929889097809792,
+      "step": 1334
+    },
+    {
+      "epoch": 0.2225,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.4018229166666667,
+      "learning_rate": 8.829100106249189e-05,
+      "loss": 7.0172,
+      "loss/crossentropy": 1.7925970554351807,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22310101985931396,
+      "step": 1335
+    },
+    {
+      "epoch": 0.22266666666666668,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.224739583333333,
+      "learning_rate": 8.827416067465441e-05,
+      "loss": 7.2243,
+      "loss/crossentropy": 2.161846488714218,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1975463330745697,
+      "step": 1336
+    },
+    {
+      "epoch": 0.22283333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.3520182291666667,
+      "learning_rate": 8.825730979373872e-05,
+      "loss": 7.0193,
+      "loss/crossentropy": 1.779839664697647,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1584184467792511,
+      "step": 1337
+    },
+    {
+      "epoch": 0.223,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.1166015625,
+      "learning_rate": 8.824044842436456e-05,
+      "loss": 7.2018,
+      "loss/crossentropy": 1.8944500386714935,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1684500388801098,
+      "step": 1338
+    },
+    {
+      "epoch": 0.22316666666666668,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.080208333333333,
+      "learning_rate": 8.822357657115459e-05,
+      "loss": 7.2954,
+      "loss/crossentropy": 2.237246662378311,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21640732139348984,
+      "step": 1339
+    },
+    {
+      "epoch": 0.22333333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.002083333333333,
+      "learning_rate": 8.82066942387343e-05,
+      "loss": 7.0554,
+      "loss/crossentropy": 2.0690477788448334,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19336192682385445,
+      "step": 1340
+    },
+    {
+      "epoch": 0.2235,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.5395833333333333,
+      "learning_rate": 8.818980143173213e-05,
+      "loss": 7.2203,
+      "loss/crossentropy": 1.7061096876859665,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18845251947641373,
+      "step": 1341
+    },
+    {
+      "epoch": 0.22366666666666668,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.5872395833333335,
+      "learning_rate": 8.817289815477928e-05,
+      "loss": 7.6896,
+      "loss/crossentropy": 1.5505743622779846,
+      "loss/hidden": 3.85546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2833479270339012,
+      "step": 1342
+    },
+    {
+      "epoch": 0.22383333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.2744140625,
+      "learning_rate": 8.815598441250987e-05,
+      "loss": 7.0531,
+      "loss/crossentropy": 1.713552087545395,
+      "loss/hidden": 3.16015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14133765175938606,
+      "step": 1343
+    },
+    {
+      "epoch": 0.224,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.3921223958333333,
+      "learning_rate": 8.813906020956097e-05,
+      "loss": 6.9316,
+      "loss/crossentropy": 2.2636399269104004,
+      "loss/hidden": 3.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16073094308376312,
+      "step": 1344
+    },
+    {
+      "epoch": 0.22416666666666665,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.1499348958333333,
+      "learning_rate": 8.81221255505724e-05,
+      "loss": 7.1873,
+      "loss/crossentropy": 2.3648778796195984,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18794936686754227,
+      "step": 1345
+    },
+    {
+      "epoch": 0.22433333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.6489583333333333,
+      "learning_rate": 8.810518044018689e-05,
+      "loss": 7.2767,
+      "loss/crossentropy": 1.4827014207839966,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1567552611231804,
+      "step": 1346
+    },
+    {
+      "epoch": 0.2245,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.97265625,
+      "learning_rate": 8.808822488305005e-05,
+      "loss": 7.1419,
+      "loss/crossentropy": 2.181717038154602,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17448044940829277,
+      "step": 1347
+    },
+    {
+      "epoch": 0.22466666666666665,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.068489583333333,
+      "learning_rate": 8.807125888381035e-05,
+      "loss": 7.3815,
+      "loss/crossentropy": 1.9000767469406128,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19831407442688942,
+      "step": 1348
+    },
+    {
+      "epoch": 0.22483333333333333,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.1489583333333333,
+      "learning_rate": 8.80542824471191e-05,
+      "loss": 6.9805,
+      "loss/crossentropy": 1.9254313111305237,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19930032268166542,
+      "step": 1349
+    },
+    {
+      "epoch": 0.225,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.2343098958333334,
+      "learning_rate": 8.803729557763047e-05,
+      "loss": 6.9941,
+      "loss/crossentropy": 1.7938042283058167,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16836293041706085,
+      "step": 1350
+    },
+    {
+      "epoch": 0.22516666666666665,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.8936848958333334,
+      "learning_rate": 8.802029828000156e-05,
+      "loss": 7.2638,
+      "loss/crossentropy": 1.8875901997089386,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18301720544695854,
+      "step": 1351
+    },
+    {
+      "epoch": 0.22533333333333333,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.9518229166666667,
+      "learning_rate": 8.800329055889223e-05,
+      "loss": 6.9781,
+      "loss/crossentropy": 1.7926859557628632,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19640637189149857,
+      "step": 1352
+    },
+    {
+      "epoch": 0.2255,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.6520182291666667,
+      "learning_rate": 8.798627241896524e-05,
+      "loss": 7.2083,
+      "loss/crossentropy": 2.0834586918354034,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15520642697811127,
+      "step": 1353
+    },
+    {
+      "epoch": 0.22566666666666665,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.69140625,
+      "learning_rate": 8.796924386488624e-05,
+      "loss": 7.4726,
+      "loss/crossentropy": 1.8181260824203491,
+      "loss/hidden": 4.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33195482194423676,
+      "step": 1354
+    },
+    {
+      "epoch": 0.22583333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.7744140625,
+      "learning_rate": 8.795220490132369e-05,
+      "loss": 7.0995,
+      "loss/crossentropy": 2.0270180106163025,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1946832872927189,
+      "step": 1355
+    },
+    {
+      "epoch": 0.226,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.7744140625,
+      "learning_rate": 8.793515553294891e-05,
+      "loss": 7.0758,
+      "loss/crossentropy": 2.230189561843872,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18500115722417831,
+      "step": 1356
+    },
+    {
+      "epoch": 0.22616666666666665,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.2072265625,
+      "learning_rate": 8.79180957644361e-05,
+      "loss": 7.0507,
+      "loss/crossentropy": 2.0787720680236816,
+      "loss/hidden": 3.125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16317469254136086,
+      "step": 1357
+    },
+    {
+      "epoch": 0.22633333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.201041666666667,
+      "learning_rate": 8.790102560046227e-05,
+      "loss": 7.1832,
+      "loss/crossentropy": 1.994160920381546,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1815122552216053,
+      "step": 1358
+    },
+    {
+      "epoch": 0.2265,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.216080729166667,
+      "learning_rate": 8.788394504570732e-05,
+      "loss": 7.5255,
+      "loss/crossentropy": 1.4656861275434494,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20615416020154953,
+      "step": 1359
+    },
+    {
+      "epoch": 0.22666666666666666,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.28125,
+      "learning_rate": 8.786685410485398e-05,
+      "loss": 7.1196,
+      "loss/crossentropy": 1.8797804713249207,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2133900672197342,
+      "step": 1360
+    },
+    {
+      "epoch": 0.22683333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.309375,
+      "learning_rate": 8.784975278258783e-05,
+      "loss": 7.158,
+      "loss/crossentropy": 1.7950632572174072,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16669543832540512,
+      "step": 1361
+    },
+    {
+      "epoch": 0.227,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.3056640625,
+      "learning_rate": 8.783264108359728e-05,
+      "loss": 6.8924,
+      "loss/crossentropy": 2.453521102666855,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1855342723429203,
+      "step": 1362
+    },
+    {
+      "epoch": 0.22716666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.7634765625,
+      "learning_rate": 8.78155190125736e-05,
+      "loss": 7.0847,
+      "loss/crossentropy": 1.6687047630548477,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1837630569934845,
+      "step": 1363
+    },
+    {
+      "epoch": 0.22733333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.7270182291666667,
+      "learning_rate": 8.779838657421092e-05,
+      "loss": 7.1379,
+      "loss/crossentropy": 1.74602010846138,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1441764123737812,
+      "step": 1364
+    },
+    {
+      "epoch": 0.2275,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.5229166666666667,
+      "learning_rate": 8.778124377320618e-05,
+      "loss": 6.851,
+      "loss/crossentropy": 1.7809283435344696,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16961202025413513,
+      "step": 1365
+    },
+    {
+      "epoch": 0.22766666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.2104166666666667,
+      "learning_rate": 8.776409061425919e-05,
+      "loss": 7.0813,
+      "loss/crossentropy": 2.079925984144211,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.183708768337965,
+      "step": 1366
+    },
+    {
+      "epoch": 0.22783333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.215559895833333,
+      "learning_rate": 8.774692710207257e-05,
+      "loss": 7.0477,
+      "loss/crossentropy": 1.804535448551178,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19650908187031746,
+      "step": 1367
+    },
+    {
+      "epoch": 0.228,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.135872395833333,
+      "learning_rate": 8.772975324135179e-05,
+      "loss": 6.9129,
+      "loss/crossentropy": 1.7298943996429443,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15943126380443573,
+      "step": 1368
+    },
+    {
+      "epoch": 0.22816666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.099739583333333,
+      "learning_rate": 8.771256903680519e-05,
+      "loss": 7.2448,
+      "loss/crossentropy": 1.863993912935257,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16760018840432167,
+      "step": 1369
+    },
+    {
+      "epoch": 0.22833333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.7080729166666666,
+      "learning_rate": 8.769537449314391e-05,
+      "loss": 7.0922,
+      "loss/crossentropy": 1.688790649175644,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1639144904911518,
+      "step": 1370
+    },
+    {
+      "epoch": 0.2285,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.7072916666666667,
+      "learning_rate": 8.76781696150819e-05,
+      "loss": 7.138,
+      "loss/crossentropy": 2.0839433670043945,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21453265100717545,
+      "step": 1371
+    },
+    {
+      "epoch": 0.22866666666666666,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.8020182291666667,
+      "learning_rate": 8.766095440733601e-05,
+      "loss": 7.2556,
+      "loss/crossentropy": 1.6846605837345123,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19930050894618034,
+      "step": 1372
+    },
+    {
+      "epoch": 0.22883333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.7426432291666667,
+      "learning_rate": 8.764372887462586e-05,
+      "loss": 7.0091,
+      "loss/crossentropy": 2.213774800300598,
+      "loss/hidden": 3.109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1519070602953434,
+      "step": 1373
+    },
+    {
+      "epoch": 0.229,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.7270182291666667,
+      "learning_rate": 8.762649302167395e-05,
+      "loss": 7.1046,
+      "loss/crossentropy": 2.119282454252243,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18627038970589638,
+      "step": 1374
+    },
+    {
+      "epoch": 0.22916666666666666,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.7660807291666667,
+      "learning_rate": 8.760924685320557e-05,
+      "loss": 7.0487,
+      "loss/crossentropy": 1.7688325345516205,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18404393270611763,
+      "step": 1375
+    },
+    {
+      "epoch": 0.22933333333333333,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.9223307291666667,
+      "learning_rate": 8.759199037394887e-05,
+      "loss": 7.0125,
+      "loss/crossentropy": 2.181953549385071,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17447860911488533,
+      "step": 1376
+    },
+    {
+      "epoch": 0.2295,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.049934895833333,
+      "learning_rate": 8.757472358863481e-05,
+      "loss": 6.8112,
+      "loss/crossentropy": 2.0696641206741333,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1791696548461914,
+      "step": 1377
+    },
+    {
+      "epoch": 0.22966666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.137434895833333,
+      "learning_rate": 8.755744650199716e-05,
+      "loss": 7.1824,
+      "loss/crossentropy": 2.023148834705353,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24435881525278091,
+      "step": 1378
+    },
+    {
+      "epoch": 0.22983333333333333,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.8708333333333333,
+      "learning_rate": 8.754015911877255e-05,
+      "loss": 6.9685,
+      "loss/crossentropy": 2.015488862991333,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1677897684276104,
+      "step": 1379
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.1205729166666667,
+      "learning_rate": 8.752286144370041e-05,
+      "loss": 7.0386,
+      "loss/crossentropy": 2.144385486841202,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18953008204698563,
+      "step": 1380
+    },
+    {
+      "epoch": 0.23016666666666666,
+      "grad_norm": 40.25,
+      "grad_norm_var": 8.211393229166667,
+      "learning_rate": 8.750555348152298e-05,
+      "loss": 7.2246,
+      "loss/crossentropy": 2.018937200307846,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19441290199756622,
+      "step": 1381
+    },
+    {
+      "epoch": 0.23033333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 8.275455729166667,
+      "learning_rate": 8.748823523698535e-05,
+      "loss": 6.8557,
+      "loss/crossentropy": 2.073577105998993,
+      "loss/hidden": 3.0703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14610829204320908,
+      "step": 1382
+    },
+    {
+      "epoch": 0.2305,
+      "grad_norm": 30.5,
+      "grad_norm_var": 8.267708333333333,
+      "learning_rate": 8.747090671483542e-05,
+      "loss": 7.1882,
+      "loss/crossentropy": 1.808563083410263,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17777547240257263,
+      "step": 1383
+    },
+    {
+      "epoch": 0.23066666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 7.960416666666666,
+      "learning_rate": 8.745356791982391e-05,
+      "loss": 7.1564,
+      "loss/crossentropy": 1.7210944592952728,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20695078000426292,
+      "step": 1384
+    },
+    {
+      "epoch": 0.23083333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 7.960416666666666,
+      "learning_rate": 8.74362188567043e-05,
+      "loss": 7.0791,
+      "loss/crossentropy": 1.6117003560066223,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1556633971631527,
+      "step": 1385
+    },
+    {
+      "epoch": 0.231,
+      "grad_norm": 30.125,
+      "grad_norm_var": 7.919205729166666,
+      "learning_rate": 8.741885953023301e-05,
+      "loss": 7.1246,
+      "loss/crossentropy": 1.8704611659049988,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2107059620320797,
+      "step": 1386
+    },
+    {
+      "epoch": 0.23116666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 7.987239583333333,
+      "learning_rate": 8.740148994516912e-05,
+      "loss": 7.0483,
+      "loss/crossentropy": 1.927421748638153,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1569875031709671,
+      "step": 1387
+    },
+    {
+      "epoch": 0.23133333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 7.705989583333333,
+      "learning_rate": 8.738411010627466e-05,
+      "loss": 7.227,
+      "loss/crossentropy": 2.1431739032268524,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20422163233160973,
+      "step": 1388
+    },
+    {
+      "epoch": 0.2315,
+      "grad_norm": 30.375,
+      "grad_norm_var": 7.720833333333333,
+      "learning_rate": 8.736672001831438e-05,
+      "loss": 7.2047,
+      "loss/crossentropy": 1.5240344256162643,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18182214722037315,
+      "step": 1389
+    },
+    {
+      "epoch": 0.23166666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 7.727018229166666,
+      "learning_rate": 8.734931968605589e-05,
+      "loss": 7.0559,
+      "loss/crossentropy": 1.6655289232730865,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21631991863250732,
+      "step": 1390
+    },
+    {
+      "epoch": 0.23183333333333334,
+      "grad_norm": 29.0,
+      "grad_norm_var": 7.658268229166667,
+      "learning_rate": 8.733190911426958e-05,
+      "loss": 6.987,
+      "loss/crossentropy": 1.955703616142273,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18248649686574936,
+      "step": 1391
+    },
+    {
+      "epoch": 0.232,
+      "grad_norm": 30.875,
+      "grad_norm_var": 7.278580729166666,
+      "learning_rate": 8.731448830772864e-05,
+      "loss": 6.9522,
+      "loss/crossentropy": 2.055171713232994,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18208866566419601,
+      "step": 1392
+    },
+    {
+      "epoch": 0.23216666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 7.434309895833334,
+      "learning_rate": 8.729705727120911e-05,
+      "loss": 6.8244,
+      "loss/crossentropy": 1.920299768447876,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1761598140001297,
+      "step": 1393
+    },
+    {
+      "epoch": 0.23233333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 7.386393229166667,
+      "learning_rate": 8.72796160094898e-05,
+      "loss": 7.0279,
+      "loss/crossentropy": 2.074692815542221,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16406862065196037,
+      "step": 1394
+    },
+    {
+      "epoch": 0.2325,
+      "grad_norm": 31.75,
+      "grad_norm_var": 7.108072916666667,
+      "learning_rate": 8.726216452735232e-05,
+      "loss": 6.9423,
+      "loss/crossentropy": 2.3375765085220337,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18264194950461388,
+      "step": 1395
+    },
+    {
+      "epoch": 0.23266666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 6.689322916666667,
+      "learning_rate": 8.724470282958111e-05,
+      "loss": 7.2743,
+      "loss/crossentropy": 2.2332785427570343,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2182307466864586,
+      "step": 1396
+    },
+    {
+      "epoch": 0.23283333333333334,
+      "grad_norm": 27.625,
+      "grad_norm_var": 1.3171223958333333,
+      "learning_rate": 8.722723092096338e-05,
+      "loss": 7.0684,
+      "loss/crossentropy": 1.6890765130519867,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22477268055081367,
+      "step": 1397
+    },
+    {
+      "epoch": 0.233,
+      "grad_norm": 27.875,
+      "grad_norm_var": 1.6942057291666666,
+      "learning_rate": 8.720974880628916e-05,
+      "loss": 6.95,
+      "loss/crossentropy": 1.2736570984125137,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16321686282753944,
+      "step": 1398
+    },
+    {
+      "epoch": 0.23316666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.6997395833333333,
+      "learning_rate": 8.719225649035126e-05,
+      "loss": 6.9737,
+      "loss/crossentropy": 1.747178003191948,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1727956011891365,
+      "step": 1399
+    },
+    {
+      "epoch": 0.23333333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.69765625,
+      "learning_rate": 8.717475397794531e-05,
+      "loss": 7.2878,
+      "loss/crossentropy": 1.9532324075698853,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18084821850061417,
+      "step": 1400
+    },
+    {
+      "epoch": 0.2335,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.59140625,
+      "learning_rate": 8.715724127386972e-05,
+      "loss": 7.0704,
+      "loss/crossentropy": 2.113216519355774,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17437510937452316,
+      "step": 1401
+    },
+    {
+      "epoch": 0.23366666666666666,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.6895182291666666,
+      "learning_rate": 8.713971838292569e-05,
+      "loss": 6.9268,
+      "loss/crossentropy": 2.245711624622345,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20107146725058556,
+      "step": 1402
+    },
+    {
+      "epoch": 0.23383333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.164322916666667,
+      "learning_rate": 8.712218530991723e-05,
+      "loss": 7.1208,
+      "loss/crossentropy": 1.730014443397522,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22139248251914978,
+      "step": 1403
+    },
+    {
+      "epoch": 0.234,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.164322916666667,
+      "learning_rate": 8.710464205965112e-05,
+      "loss": 6.9937,
+      "loss/crossentropy": 1.6004100143909454,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17204492911696434,
+      "step": 1404
+    },
+    {
+      "epoch": 0.23416666666666666,
+      "grad_norm": 28.25,
+      "grad_norm_var": 2.3358723958333334,
+      "learning_rate": 8.708708863693697e-05,
+      "loss": 7.0352,
+      "loss/crossentropy": 2.1264417320489883,
+      "loss/hidden": 3.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1658274382352829,
+      "step": 1405
+    },
+    {
+      "epoch": 0.23433333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.459375,
+      "learning_rate": 8.706952504658712e-05,
+      "loss": 7.1431,
+      "loss/crossentropy": 1.50221349298954,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15783336013555527,
+      "step": 1406
+    },
+    {
+      "epoch": 0.2345,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.535416666666667,
+      "learning_rate": 8.705195129341672e-05,
+      "loss": 7.1194,
+      "loss/crossentropy": 2.049428880214691,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18712931126356125,
+      "step": 1407
+    },
+    {
+      "epoch": 0.23466666666666666,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.5712890625,
+      "learning_rate": 8.703436738224375e-05,
+      "loss": 6.9781,
+      "loss/crossentropy": 2.3455532789230347,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22837577387690544,
+      "step": 1408
+    },
+    {
+      "epoch": 0.23483333333333334,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.7306640625,
+      "learning_rate": 8.701677331788891e-05,
+      "loss": 7.124,
+      "loss/crossentropy": 1.8043087422847748,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1597243696451187,
+      "step": 1409
+    },
+    {
+      "epoch": 0.235,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.86015625,
+      "learning_rate": 8.699916910517573e-05,
+      "loss": 7.0915,
+      "loss/crossentropy": 1.4584205448627472,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1846204549074173,
+      "step": 1410
+    },
+    {
+      "epoch": 0.23516666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.69375,
+      "learning_rate": 8.69815547489305e-05,
+      "loss": 7.2374,
+      "loss/crossentropy": 1.5898298174142838,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16165027767419815,
+      "step": 1411
+    },
+    {
+      "epoch": 0.23533333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.530143229166667,
+      "learning_rate": 8.696393025398229e-05,
+      "loss": 6.9677,
+      "loss/crossentropy": 1.7190268635749817,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17674877494573593,
+      "step": 1412
+    },
+    {
+      "epoch": 0.2355,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.405143229166667,
+      "learning_rate": 8.694629562516294e-05,
+      "loss": 7.121,
+      "loss/crossentropy": 1.7949635684490204,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19067958369851112,
+      "step": 1413
+    },
+    {
+      "epoch": 0.23566666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.123893229166667,
+      "learning_rate": 8.692865086730713e-05,
+      "loss": 6.89,
+      "loss/crossentropy": 1.7350464463233948,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15222876518964767,
+      "step": 1414
+    },
+    {
+      "epoch": 0.23583333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.1020182291666667,
+      "learning_rate": 8.69109959852522e-05,
+      "loss": 6.9269,
+      "loss/crossentropy": 1.796377807855606,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15365470945835114,
+      "step": 1415
+    },
+    {
+      "epoch": 0.236,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.1041666666666665,
+      "learning_rate": 8.689333098383842e-05,
+      "loss": 7.0796,
+      "loss/crossentropy": 1.9770399034023285,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.198615200817585,
+      "step": 1416
+    },
+    {
+      "epoch": 0.23616666666666666,
+      "grad_norm": 26.875,
+      "grad_norm_var": 2.6375,
+      "learning_rate": 8.68756558679087e-05,
+      "loss": 7.0849,
+      "loss/crossentropy": 2.3155723810195923,
+      "loss/hidden": 3.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16964924335479736,
+      "step": 1417
+    },
+    {
+      "epoch": 0.23633333333333334,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.6561848958333334,
+      "learning_rate": 8.685797064230878e-05,
+      "loss": 7.1044,
+      "loss/crossentropy": 1.8613577038049698,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17155513912439346,
+      "step": 1418
+    },
+    {
+      "epoch": 0.2365,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.8139973958333333,
+      "learning_rate": 8.684027531188717e-05,
+      "loss": 7.3211,
+      "loss/crossentropy": 2.1177275478839874,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22781919687986374,
+      "step": 1419
+    },
+    {
+      "epoch": 0.23666666666666666,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.8546223958333334,
+      "learning_rate": 8.682256988149513e-05,
+      "loss": 6.9696,
+      "loss/crossentropy": 1.833871841430664,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17683470249176025,
+      "step": 1420
+    },
+    {
+      "epoch": 0.23683333333333334,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.8160807291666667,
+      "learning_rate": 8.680485435598673e-05,
+      "loss": 6.937,
+      "loss/crossentropy": 1.6870361864566803,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15399332344532013,
+      "step": 1421
+    },
+    {
+      "epoch": 0.237,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.6155598958333333,
+      "learning_rate": 8.678712874021874e-05,
+      "loss": 7.1699,
+      "loss/crossentropy": 2.2885822653770447,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18261811137199402,
+      "step": 1422
+    },
+    {
+      "epoch": 0.23716666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.9764973958333334,
+      "learning_rate": 8.67693930390508e-05,
+      "loss": 7.036,
+      "loss/crossentropy": 1.8674782514572144,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1692717894911766,
+      "step": 1423
+    },
+    {
+      "epoch": 0.23733333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.8931640625,
+      "learning_rate": 8.67516472573452e-05,
+      "loss": 7.1713,
+      "loss/crossentropy": 1.9137693643569946,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19587906450033188,
+      "step": 1424
+    },
+    {
+      "epoch": 0.2375,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.81875,
+      "learning_rate": 8.673389139996708e-05,
+      "loss": 7.4921,
+      "loss/crossentropy": 2.0193804800510406,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16129348427057266,
+      "step": 1425
+    },
+    {
+      "epoch": 0.23766666666666666,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.80625,
+      "learning_rate": 8.671612547178428e-05,
+      "loss": 7.1952,
+      "loss/crossentropy": 2.0666559040546417,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21362433582544327,
+      "step": 1426
+    },
+    {
+      "epoch": 0.23783333333333334,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.721875,
+      "learning_rate": 8.669834947766746e-05,
+      "loss": 7.0166,
+      "loss/crossentropy": 1.443528801202774,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18207737430930138,
+      "step": 1427
+    },
+    {
+      "epoch": 0.238,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.6302083333333333,
+      "learning_rate": 8.668056342248998e-05,
+      "loss": 7.0718,
+      "loss/crossentropy": 2.0485104620456696,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23772461712360382,
+      "step": 1428
+    },
+    {
+      "epoch": 0.23816666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.3686848958333333,
+      "learning_rate": 8.666276731112801e-05,
+      "loss": 6.907,
+      "loss/crossentropy": 1.6565506756305695,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15394266322255135,
+      "step": 1429
+    },
+    {
+      "epoch": 0.23833333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.9989583333333334,
+      "learning_rate": 8.664496114846044e-05,
+      "loss": 7.1926,
+      "loss/crossentropy": 2.0662891566753387,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17025461420416832,
+      "step": 1430
+    },
+    {
+      "epoch": 0.2385,
+      "grad_norm": 28.25,
+      "grad_norm_var": 2.0884765625,
+      "learning_rate": 8.662714493936895e-05,
+      "loss": 7.0135,
+      "loss/crossentropy": 1.6579667329788208,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1791704222559929,
+      "step": 1431
+    },
+    {
+      "epoch": 0.23866666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.0497395833333334,
+      "learning_rate": 8.660931868873793e-05,
+      "loss": 7.1355,
+      "loss/crossentropy": 1.854185864329338,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20881477370858192,
+      "step": 1432
+    },
+    {
+      "epoch": 0.23883333333333334,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.6561848958333334,
+      "learning_rate": 8.659148240145456e-05,
+      "loss": 7.0677,
+      "loss/crossentropy": 1.779709905385971,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18130938336253166,
+      "step": 1433
+    },
+    {
+      "epoch": 0.239,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.6171223958333334,
+      "learning_rate": 8.657363608240876e-05,
+      "loss": 6.9205,
+      "loss/crossentropy": 1.7048383802175522,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16548937931656837,
+      "step": 1434
+    },
+    {
+      "epoch": 0.23916666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.6796223958333334,
+      "learning_rate": 8.655577973649321e-05,
+      "loss": 7.1432,
+      "loss/crossentropy": 2.16427144408226,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2073451764881611,
+      "step": 1435
+    },
+    {
+      "epoch": 0.23933333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.6389973958333333,
+      "learning_rate": 8.653791336860331e-05,
+      "loss": 7.2272,
+      "loss/crossentropy": 2.327578216791153,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19537092745304108,
+      "step": 1436
+    },
+    {
+      "epoch": 0.2395,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.796875,
+      "learning_rate": 8.652003698363724e-05,
+      "loss": 7.0819,
+      "loss/crossentropy": 1.6819135546684265,
+      "loss/hidden": 3.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15314748883247375,
+      "step": 1437
+    },
+    {
+      "epoch": 0.23966666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.8020833333333333,
+      "learning_rate": 8.65021505864959e-05,
+      "loss": 7.305,
+      "loss/crossentropy": 1.9393266141414642,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17198432981967926,
+      "step": 1438
+    },
+    {
+      "epoch": 0.23983333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.5184895833333334,
+      "learning_rate": 8.648425418208294e-05,
+      "loss": 6.9837,
+      "loss/crossentropy": 2.1693661212921143,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17503316327929497,
+      "step": 1439
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.5552083333333333,
+      "learning_rate": 8.64663477753048e-05,
+      "loss": 7.0159,
+      "loss/crossentropy": 1.4025498777627945,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1701631397008896,
+      "step": 1440
+    },
+    {
+      "epoch": 0.24016666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.9676432291666666,
+      "learning_rate": 8.644843137107059e-05,
+      "loss": 6.9633,
+      "loss/crossentropy": 1.393399029970169,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13988840579986572,
+      "step": 1441
+    },
+    {
+      "epoch": 0.24033333333333334,
+      "grad_norm": 35.25,
+      "grad_norm_var": 3.3666666666666667,
+      "learning_rate": 8.64305049742922e-05,
+      "loss": 6.9885,
+      "loss/crossentropy": 1.7633602619171143,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21782268583774567,
+      "step": 1442
+    },
+    {
+      "epoch": 0.2405,
+      "grad_norm": 33.75,
+      "grad_norm_var": 4.05390625,
+      "learning_rate": 8.641256858988424e-05,
+      "loss": 6.847,
+      "loss/crossentropy": 1.1049831062555313,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17695121467113495,
+      "step": 1443
+    },
+    {
+      "epoch": 0.24066666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 4.008333333333334,
+      "learning_rate": 8.639462222276409e-05,
+      "loss": 6.8518,
+      "loss/crossentropy": 2.1725153625011444,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1662130281329155,
+      "step": 1444
+    },
+    {
+      "epoch": 0.24083333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 4.090625,
+      "learning_rate": 8.637666587785184e-05,
+      "loss": 7.1297,
+      "loss/crossentropy": 1.5501872599124908,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1997702680528164,
+      "step": 1445
+    },
+    {
+      "epoch": 0.241,
+      "grad_norm": 28.25,
+      "grad_norm_var": 4.21015625,
+      "learning_rate": 8.635869956007034e-05,
+      "loss": 7.1231,
+      "loss/crossentropy": 1.423758640885353,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18965114653110504,
+      "step": 1446
+    },
+    {
+      "epoch": 0.24116666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.8791015625,
+      "learning_rate": 8.634072327434515e-05,
+      "loss": 7.098,
+      "loss/crossentropy": 1.4089665859937668,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25003840029239655,
+      "step": 1447
+    },
+    {
+      "epoch": 0.24133333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 3.8135416666666666,
+      "learning_rate": 8.632273702560456e-05,
+      "loss": 7.38,
+      "loss/crossentropy": 2.2259353697299957,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18022334948182106,
+      "step": 1448
+    },
+    {
+      "epoch": 0.2415,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.6518229166666667,
+      "learning_rate": 8.630474081877959e-05,
+      "loss": 7.1316,
+      "loss/crossentropy": 2.1385007202625275,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.177784763276577,
+      "step": 1449
+    },
+    {
+      "epoch": 0.24166666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.4760416666666667,
+      "learning_rate": 8.628673465880404e-05,
+      "loss": 7.0279,
+      "loss/crossentropy": 2.2046436965465546,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1764751896262169,
+      "step": 1450
+    },
+    {
+      "epoch": 0.24183333333333334,
+      "grad_norm": 27.875,
+      "grad_norm_var": 4.048893229166667,
+      "learning_rate": 8.626871855061438e-05,
+      "loss": 7.2016,
+      "loss/crossentropy": 1.8152197003364563,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18991127610206604,
+      "step": 1451
+    },
+    {
+      "epoch": 0.242,
+      "grad_norm": 29.75,
+      "grad_norm_var": 4.048893229166667,
+      "learning_rate": 8.625069249914983e-05,
+      "loss": 7.0934,
+      "loss/crossentropy": 1.2199461609125137,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19496672227978706,
+      "step": 1452
+    },
+    {
+      "epoch": 0.24216666666666667,
+      "grad_norm": 28.625,
+      "grad_norm_var": 4.258268229166666,
+      "learning_rate": 8.623265650935234e-05,
+      "loss": 7.187,
+      "loss/crossentropy": 2.0080507397651672,
+      "loss/hidden": 3.01171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18087223172187805,
+      "step": 1453
+    },
+    {
+      "epoch": 0.24233333333333335,
+      "grad_norm": 29.5,
+      "grad_norm_var": 4.299739583333333,
+      "learning_rate": 8.621461058616656e-05,
+      "loss": 7.1007,
+      "loss/crossentropy": 1.7408147901296616,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17902477830648422,
+      "step": 1454
+    },
+    {
+      "epoch": 0.2425,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.658072916666667,
+      "learning_rate": 8.61965547345399e-05,
+      "loss": 7.1665,
+      "loss/crossentropy": 1.9689427614212036,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1854184102267027,
+      "step": 1455
+    },
+    {
+      "epoch": 0.24266666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 4.540559895833334,
+      "learning_rate": 8.617848895942247e-05,
+      "loss": 7.4566,
+      "loss/crossentropy": 2.1394124925136566,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17221321910619736,
+      "step": 1456
+    },
+    {
+      "epoch": 0.24283333333333335,
+      "grad_norm": 28.0,
+      "grad_norm_var": 4.6791015625,
+      "learning_rate": 8.616041326576711e-05,
+      "loss": 6.9391,
+      "loss/crossentropy": 2.1965531408786774,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17903254553675652,
+      "step": 1457
+    },
+    {
+      "epoch": 0.243,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.0462890625,
+      "learning_rate": 8.614232765852935e-05,
+      "loss": 7.007,
+      "loss/crossentropy": 1.9292385578155518,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18688028305768967,
+      "step": 1458
+    },
+    {
+      "epoch": 0.24316666666666667,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.323372395833333,
+      "learning_rate": 8.612423214266749e-05,
+      "loss": 7.2216,
+      "loss/crossentropy": 2.345898687839508,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18654945492744446,
+      "step": 1459
+    },
+    {
+      "epoch": 0.24333333333333335,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.193489583333333,
+      "learning_rate": 8.610612672314251e-05,
+      "loss": 6.9686,
+      "loss/crossentropy": 1.7359199225902557,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23047712817788124,
+      "step": 1460
+    },
+    {
+      "epoch": 0.2435,
+      "grad_norm": 34.5,
+      "grad_norm_var": 3.4895833333333335,
+      "learning_rate": 8.608801140491811e-05,
+      "loss": 7.5035,
+      "loss/crossentropy": 1.9083283096551895,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17189928144216537,
+      "step": 1461
+    },
+    {
+      "epoch": 0.24366666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.2348307291666667,
+      "learning_rate": 8.606988619296071e-05,
+      "loss": 7.0419,
+      "loss/crossentropy": 1.5378432869911194,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1655520312488079,
+      "step": 1462
+    },
+    {
+      "epoch": 0.24383333333333335,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.2330729166666665,
+      "learning_rate": 8.605175109223944e-05,
+      "loss": 7.2363,
+      "loss/crossentropy": 1.8559139370918274,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21075575798749924,
+      "step": 1463
+    },
+    {
+      "epoch": 0.244,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.1080729166666665,
+      "learning_rate": 8.603360610772612e-05,
+      "loss": 7.0639,
+      "loss/crossentropy": 1.617599070072174,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16611310839653015,
+      "step": 1464
+    },
+    {
+      "epoch": 0.24416666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 3.1354166666666665,
+      "learning_rate": 8.601545124439535e-05,
+      "loss": 7.0136,
+      "loss/crossentropy": 1.7576895207166672,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16732006892561913,
+      "step": 1465
+    },
+    {
+      "epoch": 0.24433333333333335,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.1119140625,
+      "learning_rate": 8.599728650722434e-05,
+      "loss": 7.1847,
+      "loss/crossentropy": 2.0496761202812195,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1775667667388916,
+      "step": 1466
+    },
+    {
+      "epoch": 0.2445,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.78515625,
+      "learning_rate": 8.597911190119308e-05,
+      "loss": 6.8316,
+      "loss/crossentropy": 1.859610378742218,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2175607979297638,
+      "step": 1467
+    },
+    {
+      "epoch": 0.24466666666666667,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.8473307291666665,
+      "learning_rate": 8.596092743128423e-05,
+      "loss": 7.0243,
+      "loss/crossentropy": 1.9550632238388062,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17831335589289665,
+      "step": 1468
+    },
+    {
+      "epoch": 0.24483333333333332,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.7046223958333333,
+      "learning_rate": 8.594273310248318e-05,
+      "loss": 7.2541,
+      "loss/crossentropy": 1.6503151059150696,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15552670508623123,
+      "step": 1469
+    },
+    {
+      "epoch": 0.245,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.693489583333333,
+      "learning_rate": 8.592452891977798e-05,
+      "loss": 7.3435,
+      "loss/crossentropy": 2.0008737444877625,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3390449061989784,
+      "step": 1470
+    },
+    {
+      "epoch": 0.24516666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.046875,
+      "learning_rate": 8.590631488815944e-05,
+      "loss": 7.2786,
+      "loss/crossentropy": 2.1170044243335724,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21889721602201462,
+      "step": 1471
+    },
+    {
+      "epoch": 0.24533333333333332,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.035416666666667,
+      "learning_rate": 8.588809101262103e-05,
+      "loss": 7.2323,
+      "loss/crossentropy": 2.4229221642017365,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18312903493642807,
+      "step": 1472
+    },
+    {
+      "epoch": 0.2455,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.8018229166666666,
+      "learning_rate": 8.586985729815894e-05,
+      "loss": 7.0508,
+      "loss/crossentropy": 1.6384101808071136,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18184493854641914,
+      "step": 1473
+    },
+    {
+      "epoch": 0.24566666666666667,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.113541666666667,
+      "learning_rate": 8.585161374977202e-05,
+      "loss": 6.9332,
+      "loss/crossentropy": 1.8630158305168152,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18213900551199913,
+      "step": 1474
+    },
+    {
+      "epoch": 0.24583333333333332,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.9442057291666666,
+      "learning_rate": 8.583336037246186e-05,
+      "loss": 7.0484,
+      "loss/crossentropy": 1.7091726660728455,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19823084771633148,
+      "step": 1475
+    },
+    {
+      "epoch": 0.246,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.1603515625,
+      "learning_rate": 8.581509717123273e-05,
+      "loss": 7.2434,
+      "loss/crossentropy": 1.754471093416214,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2181721366941929,
+      "step": 1476
+    },
+    {
+      "epoch": 0.24616666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 0.8827473958333333,
+      "learning_rate": 8.579682415109156e-05,
+      "loss": 7.2447,
+      "loss/crossentropy": 1.5663812309503555,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.287110410630703,
+      "step": 1477
+    },
+    {
+      "epoch": 0.24633333333333332,
+      "grad_norm": 28.75,
+      "grad_norm_var": 0.9559895833333333,
+      "learning_rate": 8.577854131704805e-05,
+      "loss": 7.0617,
+      "loss/crossentropy": 1.7967409044504166,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1808944046497345,
+      "step": 1478
+    },
+    {
+      "epoch": 0.2465,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.3645182291666667,
+      "learning_rate": 8.576024867411451e-05,
+      "loss": 7.2348,
+      "loss/crossentropy": 2.039126396179199,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.156598012894392,
+      "step": 1479
+    },
+    {
+      "epoch": 0.24666666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 2.7330729166666665,
+      "learning_rate": 8.574194622730599e-05,
+      "loss": 7.2745,
+      "loss/crossentropy": 1.3324691206216812,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1677563078701496,
+      "step": 1480
+    },
+    {
+      "epoch": 0.24683333333333332,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.725455729166667,
+      "learning_rate": 8.572363398164017e-05,
+      "loss": 7.1396,
+      "loss/crossentropy": 1.4297092854976654,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18744304403662682,
+      "step": 1481
+    },
+    {
+      "epoch": 0.247,
+      "grad_norm": 37.0,
+      "grad_norm_var": 5.767122395833334,
+      "learning_rate": 8.57053119421375e-05,
+      "loss": 7.1088,
+      "loss/crossentropy": 1.9855413734912872,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1613350734114647,
+      "step": 1482
+    },
+    {
+      "epoch": 0.24716666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 5.816080729166667,
+      "learning_rate": 8.568698011382107e-05,
+      "loss": 7.0294,
+      "loss/crossentropy": 2.1739939153194427,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18840103223919868,
+      "step": 1483
+    },
+    {
+      "epoch": 0.24733333333333332,
+      "grad_norm": 28.375,
+      "grad_norm_var": 5.972330729166667,
+      "learning_rate": 8.566863850171663e-05,
+      "loss": 7.2819,
+      "loss/crossentropy": 2.374476373195648,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18744181096553802,
+      "step": 1484
+    },
+    {
+      "epoch": 0.2475,
+      "grad_norm": 29.5,
+      "grad_norm_var": 5.984375,
+      "learning_rate": 8.565028711085265e-05,
+      "loss": 7.0723,
+      "loss/crossentropy": 2.258538693189621,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17334047704935074,
+      "step": 1485
+    },
+    {
+      "epoch": 0.24766666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 5.9541015625,
+      "learning_rate": 8.563192594626027e-05,
+      "loss": 7.0158,
+      "loss/crossentropy": 2.054572969675064,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16874121502041817,
+      "step": 1486
+    },
+    {
+      "epoch": 0.24783333333333332,
+      "grad_norm": 30.5,
+      "grad_norm_var": 5.948372395833333,
+      "learning_rate": 8.56135550129733e-05,
+      "loss": 7.1121,
+      "loss/crossentropy": 1.843277245759964,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16777873411774635,
+      "step": 1487
+    },
+    {
+      "epoch": 0.248,
+      "grad_norm": 29.125,
+      "grad_norm_var": 6.041080729166667,
+      "learning_rate": 8.559517431602824e-05,
+      "loss": 6.967,
+      "loss/crossentropy": 2.1808391213417053,
+      "loss/hidden": 3.11328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16604379191994667,
+      "step": 1488
+    },
+    {
+      "epoch": 0.24816666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 6.089322916666666,
+      "learning_rate": 8.557678386046428e-05,
+      "loss": 7.1192,
+      "loss/crossentropy": 1.953674465417862,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20273207128047943,
+      "step": 1489
+    },
+    {
+      "epoch": 0.24833333333333332,
+      "grad_norm": 29.375,
+      "grad_norm_var": 5.8150390625,
+      "learning_rate": 8.555838365132323e-05,
+      "loss": 7.3797,
+      "loss/crossentropy": 2.531637817621231,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17258794233202934,
+      "step": 1490
+    },
+    {
+      "epoch": 0.2485,
+      "grad_norm": 27.5,
+      "grad_norm_var": 6.267708333333333,
+      "learning_rate": 8.553997369364963e-05,
+      "loss": 7.0006,
+      "loss/crossentropy": 2.1717637181282043,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16751734167337418,
+      "step": 1491
+    },
+    {
+      "epoch": 0.24866666666666667,
+      "grad_norm": 29.125,
+      "grad_norm_var": 6.053580729166667,
+      "learning_rate": 8.552155399249067e-05,
+      "loss": 6.9992,
+      "loss/crossentropy": 1.7864954769611359,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14222882501780987,
+      "step": 1492
+    },
+    {
+      "epoch": 0.24883333333333332,
+      "grad_norm": 33.0,
+      "grad_norm_var": 6.566080729166667,
+      "learning_rate": 8.550312455289625e-05,
+      "loss": 7.2126,
+      "loss/crossentropy": 2.0901902318000793,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16520576551556587,
+      "step": 1493
+    },
+    {
+      "epoch": 0.249,
+      "grad_norm": 27.875,
+      "grad_norm_var": 6.77890625,
+      "learning_rate": 8.548468537991884e-05,
+      "loss": 7.0567,
+      "loss/crossentropy": 1.9554181396961212,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1722060665488243,
+      "step": 1494
+    },
+    {
+      "epoch": 0.24916666666666668,
+      "grad_norm": 29.25,
+      "grad_norm_var": 6.215559895833334,
+      "learning_rate": 8.54662364786137e-05,
+      "loss": 7.0641,
+      "loss/crossentropy": 1.9646504074335098,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22118854150176048,
+      "step": 1495
+    },
+    {
+      "epoch": 0.24933333333333332,
+      "grad_norm": 30.25,
+      "grad_norm_var": 5.078059895833333,
+      "learning_rate": 8.544777785403868e-05,
+      "loss": 7.1731,
+      "loss/crossentropy": 2.0069101750850677,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2539825662970543,
+      "step": 1496
+    },
+    {
+      "epoch": 0.2495,
+      "grad_norm": 29.75,
+      "grad_norm_var": 5.070572916666666,
+      "learning_rate": 8.542930951125432e-05,
+      "loss": 7.3933,
+      "loss/crossentropy": 2.394498735666275,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17703993618488312,
+      "step": 1497
+    },
+    {
+      "epoch": 0.24966666666666668,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.5400390625,
+      "learning_rate": 8.54108314553238e-05,
+      "loss": 7.2543,
+      "loss/crossentropy": 1.8668582439422607,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18204010277986526,
+      "step": 1498
+    },
+    {
+      "epoch": 0.24983333333333332,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.5728515625,
+      "learning_rate": 8.539234369131301e-05,
+      "loss": 7.0307,
+      "loss/crossentropy": 1.8455041944980621,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.156462661921978,
+      "step": 1499
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.6587890625,
+      "learning_rate": 8.53738462242905e-05,
+      "loss": 7.1447,
+      "loss/crossentropy": 2.0016601383686066,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1628295946866274,
+      "step": 1500
+    },
+    {
+      "epoch": 0.25016666666666665,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.7791015625,
+      "learning_rate": 8.535533905932738e-05,
+      "loss": 7.2238,
+      "loss/crossentropy": 1.8938115537166595,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26563864573836327,
+      "step": 1501
+    },
+    {
+      "epoch": 0.25033333333333335,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.7676432291666666,
+      "learning_rate": 8.533682220149756e-05,
+      "loss": 7.0988,
+      "loss/crossentropy": 1.8613539934158325,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24434947595000267,
+      "step": 1502
+    },
+    {
+      "epoch": 0.2505,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.7447916666666667,
+      "learning_rate": 8.53182956558775e-05,
+      "loss": 7.2005,
+      "loss/crossentropy": 2.223646640777588,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17873406410217285,
+      "step": 1503
+    },
+    {
+      "epoch": 0.25066666666666665,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.7452473958333334,
+      "learning_rate": 8.52997594275464e-05,
+      "loss": 7.1795,
+      "loss/crossentropy": 2.195953756570816,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1816156506538391,
+      "step": 1504
+    },
+    {
+      "epoch": 0.25083333333333335,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.82890625,
+      "learning_rate": 8.528121352158604e-05,
+      "loss": 7.051,
+      "loss/crossentropy": 1.791160225868225,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17693329975008965,
+      "step": 1505
+    },
+    {
+      "epoch": 0.251,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.8270833333333334,
+      "learning_rate": 8.526265794308089e-05,
+      "loss": 7.274,
+      "loss/crossentropy": 2.0660716593265533,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16169629991054535,
+      "step": 1506
+    },
+    {
+      "epoch": 0.25116666666666665,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.6999348958333333,
+      "learning_rate": 8.524409269711807e-05,
+      "loss": 7.224,
+      "loss/crossentropy": 1.6352661848068237,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1959129236638546,
+      "step": 1507
+    },
+    {
+      "epoch": 0.25133333333333335,
+      "grad_norm": 27.625,
+      "grad_norm_var": 1.9764973958333334,
+      "learning_rate": 8.522551778878736e-05,
+      "loss": 6.9803,
+      "loss/crossentropy": 1.7254846692085266,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15572316572070122,
+      "step": 1508
+    },
+    {
+      "epoch": 0.2515,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.2416015625,
+      "learning_rate": 8.520693322318116e-05,
+      "loss": 7.0821,
+      "loss/crossentropy": 2.414563000202179,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1611940562725067,
+      "step": 1509
+    },
+    {
+      "epoch": 0.25166666666666665,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.0775390625,
+      "learning_rate": 8.518833900539454e-05,
+      "loss": 7.1837,
+      "loss/crossentropy": 1.6683246195316315,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17418760433793068,
+      "step": 1510
+    },
+    {
+      "epoch": 0.25183333333333335,
+      "grad_norm": 27.375,
+      "grad_norm_var": 1.365625,
+      "learning_rate": 8.516973514052519e-05,
+      "loss": 6.9838,
+      "loss/crossentropy": 1.6335192620754242,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19281744211912155,
+      "step": 1511
+    },
+    {
+      "epoch": 0.252,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.4166015625,
+      "learning_rate": 8.515112163367351e-05,
+      "loss": 7.0811,
+      "loss/crossentropy": 1.8654418885707855,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1812376230955124,
+      "step": 1512
+    },
+    {
+      "epoch": 0.25216666666666665,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.4363932291666666,
+      "learning_rate": 8.513249848994246e-05,
+      "loss": 7.0429,
+      "loss/crossentropy": 2.045855939388275,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17637160047888756,
+      "step": 1513
+    },
+    {
+      "epoch": 0.25233333333333335,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.5166666666666666,
+      "learning_rate": 8.511386571443771e-05,
+      "loss": 6.9242,
+      "loss/crossentropy": 1.500541314482689,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1793259158730507,
+      "step": 1514
+    },
+    {
+      "epoch": 0.2525,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.4926432291666667,
+      "learning_rate": 8.50952233122675e-05,
+      "loss": 7.0311,
+      "loss/crossentropy": 2.022587776184082,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21084844693541527,
+      "step": 1515
+    },
+    {
+      "epoch": 0.25266666666666665,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.3660807291666666,
+      "learning_rate": 8.50765712885428e-05,
+      "loss": 7.3262,
+      "loss/crossentropy": 2.037146747112274,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1558259353041649,
+      "step": 1516
+    },
+    {
+      "epoch": 0.25283333333333335,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.2072265625,
+      "learning_rate": 8.505790964837713e-05,
+      "loss": 7.0918,
+      "loss/crossentropy": 1.8950757384300232,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16136372834444046,
+      "step": 1517
+    },
+    {
+      "epoch": 0.253,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.2395833333333333,
+      "learning_rate": 8.503923839688667e-05,
+      "loss": 7.14,
+      "loss/crossentropy": 1.9728343486785889,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18228712677955627,
+      "step": 1518
+    },
+    {
+      "epoch": 0.25316666666666665,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.5405598958333333,
+      "learning_rate": 8.502055753919032e-05,
+      "loss": 7.1326,
+      "loss/crossentropy": 1.9609195291996002,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2126048132777214,
+      "step": 1519
+    },
+    {
+      "epoch": 0.25333333333333335,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.5817057291666667,
+      "learning_rate": 8.500186708040949e-05,
+      "loss": 7.0215,
+      "loss/crossentropy": 1.7548165321350098,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16868678107857704,
+      "step": 1520
+    },
+    {
+      "epoch": 0.2535,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.4518229166666667,
+      "learning_rate": 8.498316702566828e-05,
+      "loss": 7.3239,
+      "loss/crossentropy": 1.947168231010437,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19160600751638412,
+      "step": 1521
+    },
+    {
+      "epoch": 0.25366666666666665,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.4879557291666667,
+      "learning_rate": 8.496445738009342e-05,
+      "loss": 7.1672,
+      "loss/crossentropy": 1.758422464132309,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17631743103265762,
+      "step": 1522
+    },
+    {
+      "epoch": 0.25383333333333336,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.3416666666666666,
+      "learning_rate": 8.494573814881426e-05,
+      "loss": 7.0456,
+      "loss/crossentropy": 1.759205311536789,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22805256769061089,
+      "step": 1523
+    },
+    {
+      "epoch": 0.254,
+      "grad_norm": 27.75,
+      "grad_norm_var": 1.3103515625,
+      "learning_rate": 8.49270093369628e-05,
+      "loss": 6.7978,
+      "loss/crossentropy": 2.1120024919509888,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16711837612092495,
+      "step": 1524
+    },
+    {
+      "epoch": 0.25416666666666665,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.26640625,
+      "learning_rate": 8.490827094967363e-05,
+      "loss": 7.0777,
+      "loss/crossentropy": 2.5337154865264893,
+      "loss/hidden": 3.1015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17811939492821693,
+      "step": 1525
+    },
+    {
+      "epoch": 0.25433333333333336,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.31640625,
+      "learning_rate": 8.488952299208401e-05,
+      "loss": 7.0392,
+      "loss/crossentropy": 1.9581322371959686,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2321866899728775,
+      "step": 1526
+    },
+    {
+      "epoch": 0.2545,
+      "grad_norm": 27.875,
+      "grad_norm_var": 1.1830729166666667,
+      "learning_rate": 8.487076546933378e-05,
+      "loss": 6.9305,
+      "loss/crossentropy": 2.109581172466278,
+      "loss/hidden": 3.1015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16646679490804672,
+      "step": 1527
+    },
+    {
+      "epoch": 0.25466666666666665,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.2197916666666666,
+      "learning_rate": 8.485199838656543e-05,
+      "loss": 7.0704,
+      "loss/crossentropy": 1.933333933353424,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18859445303678513,
+      "step": 1528
+    },
+    {
+      "epoch": 0.25483333333333336,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.16640625,
+      "learning_rate": 8.483322174892404e-05,
+      "loss": 7.5793,
+      "loss/crossentropy": 1.7409664392471313,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14843013882637024,
+      "step": 1529
+    },
+    {
+      "epoch": 0.255,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.1958333333333333,
+      "learning_rate": 8.481443556155735e-05,
+      "loss": 7.0619,
+      "loss/crossentropy": 1.4247803539037704,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17381511628627777,
+      "step": 1530
+    },
+    {
+      "epoch": 0.25516666666666665,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.2582682291666667,
+      "learning_rate": 8.479563982961571e-05,
+      "loss": 7.2187,
+      "loss/crossentropy": 2.05974417924881,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16389554366469383,
+      "step": 1531
+    },
+    {
+      "epoch": 0.25533333333333336,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.10625,
+      "learning_rate": 8.477683455825207e-05,
+      "loss": 7.1836,
+      "loss/crossentropy": 2.27956685423851,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18153438344597816,
+      "step": 1532
+    },
+    {
+      "epoch": 0.2555,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.270247395833333,
+      "learning_rate": 8.4758019752622e-05,
+      "loss": 7.1096,
+      "loss/crossentropy": 2.0052456855773926,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18995345756411552,
+      "step": 1533
+    },
+    {
+      "epoch": 0.25566666666666665,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.4927083333333333,
+      "learning_rate": 8.473919541788366e-05,
+      "loss": 7.2549,
+      "loss/crossentropy": 1.7704631984233856,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1560269370675087,
+      "step": 1534
+    },
+    {
+      "epoch": 0.25583333333333336,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.218489583333333,
+      "learning_rate": 8.472036155919791e-05,
+      "loss": 7.2382,
+      "loss/crossentropy": 1.8430329859256744,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3204028345644474,
+      "step": 1535
+    },
+    {
+      "epoch": 0.256,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.3760416666666666,
+      "learning_rate": 8.470151818172809e-05,
+      "loss": 7.0855,
+      "loss/crossentropy": 1.8948449194431305,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17169497162103653,
+      "step": 1536
+    },
+    {
+      "epoch": 0.25616666666666665,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.4176432291666665,
+      "learning_rate": 8.468266529064025e-05,
+      "loss": 6.8015,
+      "loss/crossentropy": 1.9506783783435822,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17541857436299324,
+      "step": 1537
+    },
+    {
+      "epoch": 0.25633333333333336,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.570572916666667,
+      "learning_rate": 8.466380289110303e-05,
+      "loss": 6.8109,
+      "loss/crossentropy": 1.991490662097931,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18353094905614853,
+      "step": 1538
+    },
+    {
+      "epoch": 0.2565,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.605208333333333,
+      "learning_rate": 8.464493098828763e-05,
+      "loss": 7.0865,
+      "loss/crossentropy": 1.59281587600708,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18551462143659592,
+      "step": 1539
+    },
+    {
+      "epoch": 0.25666666666666665,
+      "grad_norm": 33.75,
+      "grad_norm_var": 3.4302083333333333,
+      "learning_rate": 8.462604958736793e-05,
+      "loss": 6.8848,
+      "loss/crossentropy": 1.8229161500930786,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16178708523511887,
+      "step": 1540
+    },
+    {
+      "epoch": 0.25683333333333336,
+      "grad_norm": 36.25,
+      "grad_norm_var": 5.943684895833333,
+      "learning_rate": 8.460715869352035e-05,
+      "loss": 7.0446,
+      "loss/crossentropy": 2.298820734024048,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16869497299194336,
+      "step": 1541
+    },
+    {
+      "epoch": 0.257,
+      "grad_norm": 31.375,
+      "grad_norm_var": 5.800455729166667,
+      "learning_rate": 8.458825831192392e-05,
+      "loss": 7.2513,
+      "loss/crossentropy": 2.020752638578415,
+      "loss/hidden": 2.9609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1494499407708645,
+      "step": 1542
+    },
+    {
+      "epoch": 0.25716666666666665,
+      "grad_norm": 28.625,
+      "grad_norm_var": 5.575455729166666,
+      "learning_rate": 8.456934844776032e-05,
+      "loss": 7.0119,
+      "loss/crossentropy": 1.846586525440216,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20768148824572563,
+      "step": 1543
+    },
+    {
+      "epoch": 0.25733333333333336,
+      "grad_norm": 29.75,
+      "grad_norm_var": 5.601822916666666,
+      "learning_rate": 8.455042910621379e-05,
+      "loss": 7.2219,
+      "loss/crossentropy": 1.9798152446746826,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18746008723974228,
+      "step": 1544
+    },
+    {
+      "epoch": 0.2575,
+      "grad_norm": 30.0,
+      "grad_norm_var": 5.601822916666666,
+      "learning_rate": 8.453150029247114e-05,
+      "loss": 6.8856,
+      "loss/crossentropy": 2.1329750418663025,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1751115620136261,
+      "step": 1545
+    },
+    {
+      "epoch": 0.25766666666666665,
+      "grad_norm": 31.5,
+      "grad_norm_var": 5.666666666666667,
+      "learning_rate": 8.451256201172186e-05,
+      "loss": 7.1745,
+      "loss/crossentropy": 1.7879785001277924,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2214762195944786,
+      "step": 1546
+    },
+    {
+      "epoch": 0.25783333333333336,
+      "grad_norm": 29.25,
+      "grad_norm_var": 5.757291666666666,
+      "learning_rate": 8.449361426915797e-05,
+      "loss": 6.9829,
+      "loss/crossentropy": 2.2098914980888367,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1809137426316738,
+      "step": 1547
+    },
+    {
+      "epoch": 0.258,
+      "grad_norm": 37.75,
+      "grad_norm_var": 8.639322916666666,
+      "learning_rate": 8.447465706997408e-05,
+      "loss": 7.1668,
+      "loss/crossentropy": 1.942916750907898,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.189726110547781,
+      "step": 1548
+    },
+    {
+      "epoch": 0.25816666666666666,
+      "grad_norm": 28.75,
+      "grad_norm_var": 8.533268229166667,
+      "learning_rate": 8.445569041936743e-05,
+      "loss": 6.9844,
+      "loss/crossentropy": 2.1587873995304108,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1877158172428608,
+      "step": 1549
+    },
+    {
+      "epoch": 0.25833333333333336,
+      "grad_norm": 28.25,
+      "grad_norm_var": 8.447330729166667,
+      "learning_rate": 8.443671432253784e-05,
+      "loss": 7.2562,
+      "loss/crossentropy": 1.871480107307434,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17582353577017784,
+      "step": 1550
+    },
+    {
+      "epoch": 0.2585,
+      "grad_norm": 31.625,
+      "grad_norm_var": 8.386458333333334,
+      "learning_rate": 8.44177287846877e-05,
+      "loss": 7.0839,
+      "loss/crossentropy": 2.095507889986038,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1706218384206295,
+      "step": 1551
+    },
+    {
+      "epoch": 0.25866666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 7.855989583333334,
+      "learning_rate": 8.439873381102203e-05,
+      "loss": 7.0598,
+      "loss/crossentropy": 2.2450697124004364,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24325856566429138,
+      "step": 1552
+    },
+    {
+      "epoch": 0.25883333333333336,
+      "grad_norm": 28.375,
+      "grad_norm_var": 8.2666015625,
+      "learning_rate": 8.437972940674838e-05,
+      "loss": 7.1284,
+      "loss/crossentropy": 1.8304429054260254,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19983015209436417,
+      "step": 1553
+    },
+    {
+      "epoch": 0.259,
+      "grad_norm": 28.375,
+      "grad_norm_var": 8.085872395833333,
+      "learning_rate": 8.436071557707692e-05,
+      "loss": 7.1071,
+      "loss/crossentropy": 2.146566927433014,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18399665504693985,
+      "step": 1554
+    },
+    {
+      "epoch": 0.25916666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 8.195247395833333,
+      "learning_rate": 8.434169232722043e-05,
+      "loss": 6.898,
+      "loss/crossentropy": 2.0336706936359406,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18127762898802757,
+      "step": 1555
+    },
+    {
+      "epoch": 0.25933333333333336,
+      "grad_norm": 28.625,
+      "grad_norm_var": 7.813541666666667,
+      "learning_rate": 8.432265966239419e-05,
+      "loss": 7.1962,
+      "loss/crossentropy": 1.6365832686424255,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15058623626828194,
+      "step": 1556
+    },
+    {
+      "epoch": 0.2595,
+      "grad_norm": 28.75,
+      "grad_norm_var": 5.547916666666667,
+      "learning_rate": 8.430361758781616e-05,
+      "loss": 7.2417,
+      "loss/crossentropy": 1.6402669847011566,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16793038323521614,
+      "step": 1557
+    },
+    {
+      "epoch": 0.25966666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.833268229166666,
+      "learning_rate": 8.42845661087068e-05,
+      "loss": 7.2838,
+      "loss/crossentropy": 2.1328300833702087,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2763727866113186,
+      "step": 1558
+    },
+    {
+      "epoch": 0.25983333333333336,
+      "grad_norm": 31.875,
+      "grad_norm_var": 5.867122395833333,
+      "learning_rate": 8.42655052302892e-05,
+      "loss": 6.9443,
+      "loss/crossentropy": 1.8955017030239105,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2807241193950176,
+      "step": 1559
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 32.0,
+      "grad_norm_var": 6.026497395833333,
+      "learning_rate": 8.424643495778902e-05,
+      "loss": 7.0687,
+      "loss/crossentropy": 2.062573164701462,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16407060995697975,
+      "step": 1560
+    },
+    {
+      "epoch": 0.26016666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 6.218684895833333,
+      "learning_rate": 8.422735529643444e-05,
+      "loss": 7.1452,
+      "loss/crossentropy": 2.0199232399463654,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20692768320441246,
+      "step": 1561
+    },
+    {
+      "epoch": 0.26033333333333336,
+      "grad_norm": 28.5,
+      "grad_norm_var": 6.403059895833334,
+      "learning_rate": 8.42082662514563e-05,
+      "loss": 7.0374,
+      "loss/crossentropy": 1.9788072109222412,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16032467409968376,
+      "step": 1562
+    },
+    {
+      "epoch": 0.2605,
+      "grad_norm": 27.25,
+      "grad_norm_var": 6.9509765625,
+      "learning_rate": 8.418916782808795e-05,
+      "loss": 7.1187,
+      "loss/crossentropy": 1.8873691856861115,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15415704622864723,
+      "step": 1563
+    },
+    {
+      "epoch": 0.26066666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.9462890625,
+      "learning_rate": 8.417006003156532e-05,
+      "loss": 7.1693,
+      "loss/crossentropy": 1.7227062284946442,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17569826915860176,
+      "step": 1564
+    },
+    {
+      "epoch": 0.2608333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.8854166666666665,
+      "learning_rate": 8.415094286712694e-05,
+      "loss": 7.1147,
+      "loss/crossentropy": 2.2746866047382355,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24957356229424477,
+      "step": 1565
+    },
+    {
+      "epoch": 0.261,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.796875,
+      "learning_rate": 8.413181634001391e-05,
+      "loss": 7.0441,
+      "loss/crossentropy": 1.9594350010156631,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19983908534049988,
+      "step": 1566
+    },
+    {
+      "epoch": 0.26116666666666666,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.57890625,
+      "learning_rate": 8.411268045546983e-05,
+      "loss": 7.2668,
+      "loss/crossentropy": 1.8899872601032257,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18962673097848892,
+      "step": 1567
+    },
+    {
+      "epoch": 0.2613333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.5863932291666667,
+      "learning_rate": 8.409353521874093e-05,
+      "loss": 7.1443,
+      "loss/crossentropy": 1.925775170326233,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20981797203421593,
+      "step": 1568
+    },
+    {
+      "epoch": 0.2615,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.496875,
+      "learning_rate": 8.4074380635076e-05,
+      "loss": 7.0819,
+      "loss/crossentropy": 2.1734819412231445,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1924075074493885,
+      "step": 1569
+    },
+    {
+      "epoch": 0.26166666666666666,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.5759765625,
+      "learning_rate": 8.405521670972634e-05,
+      "loss": 7.413,
+      "loss/crossentropy": 1.7044086158275604,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21472477540373802,
+      "step": 1570
+    },
+    {
+      "epoch": 0.2618333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.559375,
+      "learning_rate": 8.40360434479459e-05,
+      "loss": 7.0485,
+      "loss/crossentropy": 2.3204782605171204,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.212375920265913,
+      "step": 1571
+    },
+    {
+      "epoch": 0.262,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.4275390625,
+      "learning_rate": 8.40168608549911e-05,
+      "loss": 6.9927,
+      "loss/crossentropy": 1.8797473013401031,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15978467091917992,
+      "step": 1572
+    },
+    {
+      "epoch": 0.26216666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.387434895833333,
+      "learning_rate": 8.399766893612096e-05,
+      "loss": 7.3188,
+      "loss/crossentropy": 2.2637689113616943,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18961792439222336,
+      "step": 1573
+    },
+    {
+      "epoch": 0.2623333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.135872395833333,
+      "learning_rate": 8.397846769659707e-05,
+      "loss": 7.2441,
+      "loss/crossentropy": 2.0096384286880493,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17024771869182587,
+      "step": 1574
+    },
+    {
+      "epoch": 0.2625,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.2239583333333335,
+      "learning_rate": 8.395925714168356e-05,
+      "loss": 7.0673,
+      "loss/crossentropy": 2.017771452665329,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1545296274125576,
+      "step": 1575
+    },
+    {
+      "epoch": 0.26266666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.9358723958333333,
+      "learning_rate": 8.39400372766471e-05,
+      "loss": 6.8609,
+      "loss/crossentropy": 1.6938844323158264,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16759593039751053,
+      "step": 1576
+    },
+    {
+      "epoch": 0.2628333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.5052083333333333,
+      "learning_rate": 8.392080810675691e-05,
+      "loss": 7.1057,
+      "loss/crossentropy": 2.3259546160697937,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18901516124606133,
+      "step": 1577
+    },
+    {
+      "epoch": 0.263,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.4302083333333333,
+      "learning_rate": 8.390156963728482e-05,
+      "loss": 6.8684,
+      "loss/crossentropy": 1.8894447982311249,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18735118955373764,
+      "step": 1578
+    },
+    {
+      "epoch": 0.26316666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.0434895833333333,
+      "learning_rate": 8.388232187350512e-05,
+      "loss": 6.9714,
+      "loss/crossentropy": 1.7255138158798218,
+      "loss/hidden": 3.16015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14618822745978832,
+      "step": 1579
+    },
+    {
+      "epoch": 0.2633333333333333,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.1223307291666667,
+      "learning_rate": 8.386306482069473e-05,
+      "loss": 6.9041,
+      "loss/crossentropy": 1.9056564271450043,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19680579379200935,
+      "step": 1580
+    },
+    {
+      "epoch": 0.2635,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.1645833333333333,
+      "learning_rate": 8.384379848413304e-05,
+      "loss": 7.1365,
+      "loss/crossentropy": 2.132213979959488,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1805483102798462,
+      "step": 1581
+    },
+    {
+      "epoch": 0.26366666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.3457682291666666,
+      "learning_rate": 8.382452286910206e-05,
+      "loss": 7.2889,
+      "loss/crossentropy": 2.1925718784332275,
+      "loss/hidden": 3.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16816026344895363,
+      "step": 1582
+    },
+    {
+      "epoch": 0.2638333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.3353515625,
+      "learning_rate": 8.380523798088631e-05,
+      "loss": 7.0877,
+      "loss/crossentropy": 2.1733668744564056,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.161347858607769,
+      "step": 1583
+    },
+    {
+      "epoch": 0.264,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.3822265625,
+      "learning_rate": 8.378594382477282e-05,
+      "loss": 6.8284,
+      "loss/crossentropy": 2.282644271850586,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1963726505637169,
+      "step": 1584
+    },
+    {
+      "epoch": 0.26416666666666666,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.4473307291666666,
+      "learning_rate": 8.376664040605122e-05,
+      "loss": 7.0791,
+      "loss/crossentropy": 1.775516003370285,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15552521869540215,
+      "step": 1585
+    },
+    {
+      "epoch": 0.2643333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.1629557291666666,
+      "learning_rate": 8.374732773001366e-05,
+      "loss": 7.0561,
+      "loss/crossentropy": 2.43366014957428,
+      "loss/hidden": 3.09375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1624160185456276,
+      "step": 1586
+    },
+    {
+      "epoch": 0.2645,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.1494140625,
+      "learning_rate": 8.372800580195479e-05,
+      "loss": 6.8278,
+      "loss/crossentropy": 1.6613228023052216,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18726734071969986,
+      "step": 1587
+    },
+    {
+      "epoch": 0.26466666666666666,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.2125,
+      "learning_rate": 8.370867462717183e-05,
+      "loss": 7.0035,
+      "loss/crossentropy": 1.9981612265110016,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1990807019174099,
+      "step": 1588
+    },
+    {
+      "epoch": 0.2648333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 0.9212890625,
+      "learning_rate": 8.368933421096454e-05,
+      "loss": 7.1323,
+      "loss/crossentropy": 2.2647982239723206,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18666927888989449,
+      "step": 1589
+    },
+    {
+      "epoch": 0.265,
+      "grad_norm": 28.875,
+      "grad_norm_var": 0.9268229166666667,
+      "learning_rate": 8.366998455863522e-05,
+      "loss": 6.8965,
+      "loss/crossentropy": 1.8666184544563293,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1626325212419033,
+      "step": 1590
+    },
+    {
+      "epoch": 0.26516666666666666,
+      "grad_norm": 29.375,
+      "grad_norm_var": 0.7634765625,
+      "learning_rate": 8.365062567548867e-05,
+      "loss": 7.0776,
+      "loss/crossentropy": 2.2567395865917206,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19338753074407578,
+      "step": 1591
+    },
+    {
+      "epoch": 0.2653333333333333,
+      "grad_norm": 27.25,
+      "grad_norm_var": 1.03515625,
+      "learning_rate": 8.363125756683223e-05,
+      "loss": 6.9266,
+      "loss/crossentropy": 1.82755808532238,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1721055768430233,
+      "step": 1592
+    },
+    {
+      "epoch": 0.2655,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.0546223958333334,
+      "learning_rate": 8.361188023797582e-05,
+      "loss": 7.2423,
+      "loss/crossentropy": 1.8901557624340057,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17523159086704254,
+      "step": 1593
+    },
+    {
+      "epoch": 0.26566666666666666,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.1900390625,
+      "learning_rate": 8.359249369423177e-05,
+      "loss": 7.2821,
+      "loss/crossentropy": 2.219612807035446,
+      "loss/hidden": 3.66015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35253578051924706,
+      "step": 1594
+    },
+    {
+      "epoch": 0.2658333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.2145833333333333,
+      "learning_rate": 8.357309794091507e-05,
+      "loss": 6.8563,
+      "loss/crossentropy": 2.0655614733695984,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16120551526546478,
+      "step": 1595
+    },
+    {
+      "epoch": 0.266,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.1872395833333333,
+      "learning_rate": 8.355369298334316e-05,
+      "loss": 6.9474,
+      "loss/crossentropy": 2.0755234956741333,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17432837560772896,
+      "step": 1596
+    },
+    {
+      "epoch": 0.26616666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.19375,
+      "learning_rate": 8.3534278826836e-05,
+      "loss": 7.15,
+      "loss/crossentropy": 1.813786655664444,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2066824771463871,
+      "step": 1597
+    },
+    {
+      "epoch": 0.2663333333333333,
+      "grad_norm": 28.0,
+      "grad_norm_var": 0.9275390625,
+      "learning_rate": 8.351485547671613e-05,
+      "loss": 7.1337,
+      "loss/crossentropy": 1.9518603384494781,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15583394467830658,
+      "step": 1598
+    },
+    {
+      "epoch": 0.2665,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.3447916666666666,
+      "learning_rate": 8.349542293830855e-05,
+      "loss": 7.1309,
+      "loss/crossentropy": 2.03059983253479,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25205692276358604,
+      "step": 1599
+    },
+    {
+      "epoch": 0.26666666666666666,
+      "grad_norm": 26.625,
+      "grad_norm_var": 1.7830729166666666,
+      "learning_rate": 8.347598121694078e-05,
+      "loss": 6.9445,
+      "loss/crossentropy": 1.7553084194660187,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1512312889099121,
+      "step": 1600
+    },
+    {
+      "epoch": 0.2668333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.7947265625,
+      "learning_rate": 8.345653031794292e-05,
+      "loss": 7.2284,
+      "loss/crossentropy": 2.2806900441646576,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1657196544110775,
+      "step": 1601
+    },
+    {
+      "epoch": 0.267,
+      "grad_norm": 27.625,
+      "grad_norm_var": 1.87265625,
+      "learning_rate": 8.343707024664751e-05,
+      "loss": 7.059,
+      "loss/crossentropy": 2.153120219707489,
+      "loss/hidden": 3.04296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16068052127957344,
+      "step": 1602
+    },
+    {
+      "epoch": 0.26716666666666666,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.8275390625,
+      "learning_rate": 8.341760100838965e-05,
+      "loss": 7.0079,
+      "loss/crossentropy": 1.2237086743116379,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19392723962664604,
+      "step": 1603
+    },
+    {
+      "epoch": 0.2673333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.7853515625,
+      "learning_rate": 8.339812260850696e-05,
+      "loss": 7.0806,
+      "loss/crossentropy": 2.2512892484664917,
+      "loss/hidden": 3.01171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1669567283242941,
+      "step": 1604
+    },
+    {
+      "epoch": 0.2675,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.890625,
+      "learning_rate": 8.337863505233953e-05,
+      "loss": 7.2563,
+      "loss/crossentropy": 1.9990591704845428,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21021423861384392,
+      "step": 1605
+    },
+    {
+      "epoch": 0.26766666666666666,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.9072265625,
+      "learning_rate": 8.335913834522999e-05,
+      "loss": 7.2782,
+      "loss/crossentropy": 2.0735031366348267,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19464392587542534,
+      "step": 1606
+    },
+    {
+      "epoch": 0.2678333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.1212890625,
+      "learning_rate": 8.333963249252348e-05,
+      "loss": 7.056,
+      "loss/crossentropy": 1.7016023844480515,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14383026584982872,
+      "step": 1607
+    },
+    {
+      "epoch": 0.268,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.8806640625,
+      "learning_rate": 8.332011749956763e-05,
+      "loss": 6.8989,
+      "loss/crossentropy": 1.6620410680770874,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1806044951081276,
+      "step": 1608
+    },
+    {
+      "epoch": 0.26816666666666666,
+      "grad_norm": 26.875,
+      "grad_norm_var": 2.07890625,
+      "learning_rate": 8.330059337171258e-05,
+      "loss": 6.9859,
+      "loss/crossentropy": 1.2207449227571487,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1972801610827446,
+      "step": 1609
+    },
+    {
+      "epoch": 0.2683333333333333,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.8916015625,
+      "learning_rate": 8.328106011431101e-05,
+      "loss": 7.1042,
+      "loss/crossentropy": 2.1126300394535065,
+      "loss/hidden": 3.125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13497081398963928,
+      "step": 1610
+    },
+    {
+      "epoch": 0.2685,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.8775390625,
+      "learning_rate": 8.326151773271804e-05,
+      "loss": 7.0583,
+      "loss/crossentropy": 1.7973102629184723,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2072882391512394,
+      "step": 1611
+    },
+    {
+      "epoch": 0.26866666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.1228515625,
+      "learning_rate": 8.324196623229135e-05,
+      "loss": 7.0919,
+      "loss/crossentropy": 2.0168280601501465,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19935346394777298,
+      "step": 1612
+    },
+    {
+      "epoch": 0.2688333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.1228515625,
+      "learning_rate": 8.322240561839109e-05,
+      "loss": 7.1014,
+      "loss/crossentropy": 1.9888157546520233,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19377723336219788,
+      "step": 1613
+    },
+    {
+      "epoch": 0.269,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.198372395833333,
+      "learning_rate": 8.32028358963799e-05,
+      "loss": 6.9772,
+      "loss/crossentropy": 1.6363246142864227,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16422894969582558,
+      "step": 1614
+    },
+    {
+      "epoch": 0.26916666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.737434895833333,
+      "learning_rate": 8.318325707162293e-05,
+      "loss": 7.3598,
+      "loss/crossentropy": 2.0699954330921173,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1845544744282961,
+      "step": 1615
+    },
+    {
+      "epoch": 0.2693333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.2997395833333334,
+      "learning_rate": 8.316366914948783e-05,
+      "loss": 7.1584,
+      "loss/crossentropy": 1.847909152507782,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19590704888105392,
+      "step": 1616
+    },
+    {
+      "epoch": 0.2695,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.231705729166667,
+      "learning_rate": 8.314407213534476e-05,
+      "loss": 7.1871,
+      "loss/crossentropy": 2.3094082474708557,
+      "loss/hidden": 3.14453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1651579700410366,
+      "step": 1617
+    },
+    {
+      "epoch": 0.26966666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.020768229166667,
+      "learning_rate": 8.312446603456632e-05,
+      "loss": 7.0004,
+      "loss/crossentropy": 2.274394154548645,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17561131343245506,
+      "step": 1618
+    },
+    {
+      "epoch": 0.2698333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.9426432291666667,
+      "learning_rate": 8.310485085252767e-05,
+      "loss": 7.0246,
+      "loss/crossentropy": 1.7239133417606354,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1731395162642002,
+      "step": 1619
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.06015625,
+      "learning_rate": 8.308522659460641e-05,
+      "loss": 7.0505,
+      "loss/crossentropy": 1.5815710872411728,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.157865097746253,
+      "step": 1620
+    },
+    {
+      "epoch": 0.27016666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.046875,
+      "learning_rate": 8.306559326618259e-05,
+      "loss": 7.0877,
+      "loss/crossentropy": 2.271878808736801,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17844248563051224,
+      "step": 1621
+    },
+    {
+      "epoch": 0.2703333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.946875,
+      "learning_rate": 8.304595087263889e-05,
+      "loss": 7.0577,
+      "loss/crossentropy": 1.968278557062149,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1579712200909853,
+      "step": 1622
+    },
+    {
+      "epoch": 0.2705,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.8619140625,
+      "learning_rate": 8.30262994193603e-05,
+      "loss": 7.1616,
+      "loss/crossentropy": 2.119562655687332,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17437998950481415,
+      "step": 1623
+    },
+    {
+      "epoch": 0.27066666666666667,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.1104166666666666,
+      "learning_rate": 8.300663891173443e-05,
+      "loss": 6.91,
+      "loss/crossentropy": 2.19699227809906,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18078136816620827,
+      "step": 1624
+    },
+    {
+      "epoch": 0.2708333333333333,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.7244140625,
+      "learning_rate": 8.298696935515132e-05,
+      "loss": 6.8903,
+      "loss/crossentropy": 2.244156152009964,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17901168018579483,
+      "step": 1625
+    },
+    {
+      "epoch": 0.271,
+      "grad_norm": 28.0,
+      "grad_norm_var": 1.7518229166666666,
+      "learning_rate": 8.296729075500344e-05,
+      "loss": 7.1475,
+      "loss/crossentropy": 2.1130511462688446,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14990126341581345,
+      "step": 1626
+    },
+    {
+      "epoch": 0.27116666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.0962890625,
+      "learning_rate": 8.294760311668586e-05,
+      "loss": 7.1017,
+      "loss/crossentropy": 1.4834212064743042,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15275181271135807,
+      "step": 1627
+    },
+    {
+      "epoch": 0.2713333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.9889973958333333,
+      "learning_rate": 8.2927906445596e-05,
+      "loss": 7.2311,
+      "loss/crossentropy": 1.8438728153705597,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1795530691742897,
+      "step": 1628
+    },
+    {
+      "epoch": 0.2715,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.0208333333333335,
+      "learning_rate": 8.290820074713384e-05,
+      "loss": 7.3256,
+      "loss/crossentropy": 1.7539753913879395,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19211795926094055,
+      "step": 1629
+    },
+    {
+      "epoch": 0.27166666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.9593098958333333,
+      "learning_rate": 8.28884860267018e-05,
+      "loss": 7.1493,
+      "loss/crossentropy": 2.1783579289913177,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21033593639731407,
+      "step": 1630
+    },
+    {
+      "epoch": 0.2718333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.1677083333333333,
+      "learning_rate": 8.28687622897048e-05,
+      "loss": 7.0612,
+      "loss/crossentropy": 2.0911701321601868,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19308391213417053,
+      "step": 1631
+    },
+    {
+      "epoch": 0.272,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.2059895833333334,
+      "learning_rate": 8.284902954155019e-05,
+      "loss": 6.9751,
+      "loss/crossentropy": 1.435138761997223,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1768188774585724,
+      "step": 1632
+    },
+    {
+      "epoch": 0.27216666666666667,
+      "grad_norm": 27.75,
+      "grad_norm_var": 1.3184895833333334,
+      "learning_rate": 8.282928778764783e-05,
+      "loss": 6.9796,
+      "loss/crossentropy": 1.8080698549747467,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16440429911017418,
+      "step": 1633
+    },
+    {
+      "epoch": 0.2723333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.3184895833333334,
+      "learning_rate": 8.280953703341004e-05,
+      "loss": 7.1569,
+      "loss/crossentropy": 1.7614698261022568,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18445922434329987,
+      "step": 1634
+    },
+    {
+      "epoch": 0.2725,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.3129557291666667,
+      "learning_rate": 8.278977728425157e-05,
+      "loss": 6.947,
+      "loss/crossentropy": 1.7589438259601593,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15930374339222908,
+      "step": 1635
+    },
+    {
+      "epoch": 0.27266666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.0889973958333334,
+      "learning_rate": 8.27700085455897e-05,
+      "loss": 7.0433,
+      "loss/crossentropy": 1.8300014436244965,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16976384073495865,
+      "step": 1636
+    },
+    {
+      "epoch": 0.2728333333333333,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.0780598958333334,
+      "learning_rate": 8.275023082284413e-05,
+      "loss": 6.9003,
+      "loss/crossentropy": 2.1021377742290497,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15598230808973312,
+      "step": 1637
+    },
+    {
+      "epoch": 0.273,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.18515625,
+      "learning_rate": 8.273044412143704e-05,
+      "loss": 6.9528,
+      "loss/crossentropy": 1.8906866610050201,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18633756786584854,
+      "step": 1638
+    },
+    {
+      "epoch": 0.27316666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.15390625,
+      "learning_rate": 8.271064844679306e-05,
+      "loss": 7.128,
+      "loss/crossentropy": 1.927827924489975,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15265002101659775,
+      "step": 1639
+    },
+    {
+      "epoch": 0.2733333333333333,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.03515625,
+      "learning_rate": 8.269084380433929e-05,
+      "loss": 7.0732,
+      "loss/crossentropy": 2.1406911611557007,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19738755375146866,
+      "step": 1640
+    },
+    {
+      "epoch": 0.2735,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.0125,
+      "learning_rate": 8.267103019950529e-05,
+      "loss": 7.2671,
+      "loss/crossentropy": 2.1308011412620544,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1796349138021469,
+      "step": 1641
+    },
+    {
+      "epoch": 0.27366666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 0.9666015625,
+      "learning_rate": 8.265120763772303e-05,
+      "loss": 7.1244,
+      "loss/crossentropy": 1.9307105541229248,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1456285398453474,
+      "step": 1642
+    },
+    {
+      "epoch": 0.2738333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 0.7434895833333334,
+      "learning_rate": 8.263137612442706e-05,
+      "loss": 7.3059,
+      "loss/crossentropy": 2.0887957215309143,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2207413949072361,
+      "step": 1643
+    },
+    {
+      "epoch": 0.274,
+      "grad_norm": 29.625,
+      "grad_norm_var": 0.7302083333333333,
+      "learning_rate": 8.261153566505424e-05,
+      "loss": 6.9932,
+      "loss/crossentropy": 1.6466211527585983,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15797096490859985,
+      "step": 1644
+    },
+    {
+      "epoch": 0.27416666666666667,
+      "grad_norm": 28.625,
+      "grad_norm_var": 0.6809895833333334,
+      "learning_rate": 8.259168626504395e-05,
+      "loss": 6.9243,
+      "loss/crossentropy": 2.00696924328804,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1935207098722458,
+      "step": 1645
+    },
+    {
+      "epoch": 0.2743333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 0.7093098958333334,
+      "learning_rate": 8.257182792983802e-05,
+      "loss": 7.0676,
+      "loss/crossentropy": 1.649848073720932,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1785385087132454,
+      "step": 1646
+    },
+    {
+      "epoch": 0.2745,
+      "grad_norm": 27.75,
+      "grad_norm_var": 0.7489583333333333,
+      "learning_rate": 8.255196066488075e-05,
+      "loss": 6.8512,
+      "loss/crossentropy": 1.8856600672006607,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14813595823943615,
+      "step": 1647
+    },
+    {
+      "epoch": 0.27466666666666667,
+      "grad_norm": 28.25,
+      "grad_norm_var": 0.7489583333333333,
+      "learning_rate": 8.253208447561882e-05,
+      "loss": 7.0555,
+      "loss/crossentropy": 1.8466354608535767,
+      "loss/hidden": 3.1171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1491059549152851,
+      "step": 1648
+    },
+    {
+      "epoch": 0.2748333333333333,
+      "grad_norm": 28.25,
+      "grad_norm_var": 0.675,
+      "learning_rate": 8.251219936750144e-05,
+      "loss": 7.0008,
+      "loss/crossentropy": 2.132791131734848,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1750194914638996,
+      "step": 1649
+    },
+    {
+      "epoch": 0.275,
+      "grad_norm": 27.75,
+      "grad_norm_var": 0.7916015625,
+      "learning_rate": 8.249230534598021e-05,
+      "loss": 7.0188,
+      "loss/crossentropy": 2.1920908987522125,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1745447963476181,
+      "step": 1650
+    },
+    {
+      "epoch": 0.27516666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 0.7893229166666667,
+      "learning_rate": 8.247240241650918e-05,
+      "loss": 6.9948,
+      "loss/crossentropy": 2.176539570093155,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20117703825235367,
+      "step": 1651
+    },
+    {
+      "epoch": 0.2753333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 0.7108723958333333,
+      "learning_rate": 8.245249058454487e-05,
+      "loss": 6.9395,
+      "loss/crossentropy": 1.8022592663764954,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16233988478779793,
+      "step": 1652
+    },
+    {
+      "epoch": 0.2755,
+      "grad_norm": 28.25,
+      "grad_norm_var": 0.7416015625,
+      "learning_rate": 8.243256985554621e-05,
+      "loss": 6.9914,
+      "loss/crossentropy": 1.8188396096229553,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16636322811245918,
+      "step": 1653
+    },
+    {
+      "epoch": 0.27566666666666667,
+      "grad_norm": 29.125,
+      "grad_norm_var": 0.6009765625,
+      "learning_rate": 8.241264023497457e-05,
+      "loss": 7.2485,
+      "loss/crossentropy": 2.085706055164337,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17571250721812248,
+      "step": 1654
+    },
+    {
+      "epoch": 0.2758333333333333,
+      "grad_norm": 27.0,
+      "grad_norm_var": 0.8197265625,
+      "learning_rate": 8.239270172829379e-05,
+      "loss": 6.751,
+      "loss/crossentropy": 1.9176505208015442,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16868801787495613,
+      "step": 1655
+    },
+    {
+      "epoch": 0.276,
+      "grad_norm": 29.5,
+      "grad_norm_var": 0.8520833333333333,
+      "learning_rate": 8.237275434097012e-05,
+      "loss": 7.0378,
+      "loss/crossentropy": 2.209993228316307,
+      "loss/hidden": 3.12890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.151454858481884,
+      "step": 1656
+    },
+    {
+      "epoch": 0.27616666666666667,
+      "grad_norm": 26.625,
+      "grad_norm_var": 1.1499348958333333,
+      "learning_rate": 8.235279807847223e-05,
+      "loss": 7.0835,
+      "loss/crossentropy": 2.094301849603653,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18753011897206306,
+      "step": 1657
+    },
+    {
+      "epoch": 0.2763333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.06015625,
+      "learning_rate": 8.233283294627125e-05,
+      "loss": 7.0901,
+      "loss/crossentropy": 1.595028579235077,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.168398629873991,
+      "step": 1658
+    },
+    {
+      "epoch": 0.2765,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.0858723958333334,
+      "learning_rate": 8.231285894984076e-05,
+      "loss": 6.9932,
+      "loss/crossentropy": 1.7843102812767029,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1590423509478569,
+      "step": 1659
+    },
+    {
+      "epoch": 0.27666666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.1223307291666667,
+      "learning_rate": 8.22928760946567e-05,
+      "loss": 6.9998,
+      "loss/crossentropy": 1.8459515571594238,
+      "loss/hidden": 3.12109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14566442370414734,
+      "step": 1660
+    },
+    {
+      "epoch": 0.2768333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.27890625,
+      "learning_rate": 8.227288438619754e-05,
+      "loss": 7.2782,
+      "loss/crossentropy": 1.916193664073944,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.204721599817276,
+      "step": 1661
+    },
+    {
+      "epoch": 0.277,
+      "grad_norm": 26.5,
+      "grad_norm_var": 1.4684895833333333,
+      "learning_rate": 8.225288382994407e-05,
+      "loss": 6.9197,
+      "loss/crossentropy": 1.4007163494825363,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13985158130526543,
+      "step": 1662
+    },
+    {
+      "epoch": 0.2771666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.4497395833333333,
+      "learning_rate": 8.223287443137957e-05,
+      "loss": 6.9751,
+      "loss/crossentropy": 2.0829312205314636,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16278714314103127,
+      "step": 1663
+    },
+    {
+      "epoch": 0.2773333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.4416015625,
+      "learning_rate": 8.221285619598975e-05,
+      "loss": 7.0515,
+      "loss/crossentropy": 1.7693645358085632,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1759823076426983,
+      "step": 1664
+    },
+    {
+      "epoch": 0.2775,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.0552083333333333,
+      "learning_rate": 8.21928291292627e-05,
+      "loss": 7.1634,
+      "loss/crossentropy": 2.2715067267417908,
+      "loss/hidden": 3.0703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16273073107004166,
+      "step": 1665
+    },
+    {
+      "epoch": 0.2776666666666667,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.0205729166666666,
+      "learning_rate": 8.217279323668895e-05,
+      "loss": 7.1665,
+      "loss/crossentropy": 2.014183148741722,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20622410997748375,
+      "step": 1666
+    },
+    {
+      "epoch": 0.2778333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.35390625,
+      "learning_rate": 8.215274852376147e-05,
+      "loss": 7.0098,
+      "loss/crossentropy": 2.0632661879062653,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19143903255462646,
+      "step": 1667
+    },
+    {
+      "epoch": 0.278,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.334375,
+      "learning_rate": 8.213269499597565e-05,
+      "loss": 6.9134,
+      "loss/crossentropy": 2.2255958914756775,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16110029816627502,
+      "step": 1668
+    },
+    {
+      "epoch": 0.2781666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.395247395833333,
+      "learning_rate": 8.211263265882923e-05,
+      "loss": 7.3844,
+      "loss/crossentropy": 1.7887534946203232,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22624868899583817,
+      "step": 1669
+    },
+    {
+      "epoch": 0.2783333333333333,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.543684895833333,
+      "learning_rate": 8.209256151782243e-05,
+      "loss": 7.0544,
+      "loss/crossentropy": 2.0111959129571915,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17065033316612244,
+      "step": 1670
+    },
+    {
+      "epoch": 0.2785,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.2393229166666666,
+      "learning_rate": 8.207248157845791e-05,
+      "loss": 6.9352,
+      "loss/crossentropy": 2.0150576531887054,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17338843271136284,
+      "step": 1671
+    },
+    {
+      "epoch": 0.2786666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.8208333333333333,
+      "learning_rate": 8.205239284624062e-05,
+      "loss": 7.1849,
+      "loss/crossentropy": 1.8663091659545898,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19603952020406723,
+      "step": 1672
+    },
+    {
+      "epoch": 0.2788333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.1348307291666666,
+      "learning_rate": 8.203229532667807e-05,
+      "loss": 7.1283,
+      "loss/crossentropy": 1.587388962507248,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17431241273880005,
+      "step": 1673
+    },
+    {
+      "epoch": 0.279,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.1421223958333333,
+      "learning_rate": 8.201218902528009e-05,
+      "loss": 7.1869,
+      "loss/crossentropy": 1.9146455824375153,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2287234514951706,
+      "step": 1674
+    },
+    {
+      "epoch": 0.2791666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 3.1489583333333333,
+      "learning_rate": 8.199207394755893e-05,
+      "loss": 7.0442,
+      "loss/crossentropy": 1.8834614008665085,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16575508192181587,
+      "step": 1675
+    },
+    {
+      "epoch": 0.2793333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 3.1625,
+      "learning_rate": 8.197195009902924e-05,
+      "loss": 6.8398,
+      "loss/crossentropy": 2.144304484128952,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16777660697698593,
+      "step": 1676
+    },
+    {
+      "epoch": 0.2795,
+      "grad_norm": 31.625,
+      "grad_norm_var": 3.3608723958333333,
+      "learning_rate": 8.195181748520811e-05,
+      "loss": 6.9631,
+      "loss/crossentropy": 2.2096819281578064,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16557533666491508,
+      "step": 1677
+    },
+    {
+      "epoch": 0.2796666666666667,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.8921223958333333,
+      "learning_rate": 8.193167611161499e-05,
+      "loss": 6.981,
+      "loss/crossentropy": 1.9671113193035126,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18934257328510284,
+      "step": 1678
+    },
+    {
+      "epoch": 0.2798333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.0405598958333333,
+      "learning_rate": 8.191152598377178e-05,
+      "loss": 7.2532,
+      "loss/crossentropy": 2.003457397222519,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2575335167348385,
+      "step": 1679
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.96015625,
+      "learning_rate": 8.189136710720272e-05,
+      "loss": 6.9302,
+      "loss/crossentropy": 1.7706447839736938,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17788533866405487,
+      "step": 1680
+    },
+    {
+      "epoch": 0.2801666666666667,
+      "grad_norm": 28.25,
+      "grad_norm_var": 2.958268229166667,
+      "learning_rate": 8.18711994874345e-05,
+      "loss": 6.9475,
+      "loss/crossentropy": 1.6021004915237427,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2337457314133644,
+      "step": 1681
+    },
+    {
+      "epoch": 0.2803333333333333,
+      "grad_norm": 27.75,
+      "grad_norm_var": 3.0270182291666665,
+      "learning_rate": 8.185102312999617e-05,
+      "loss": 6.8753,
+      "loss/crossentropy": 1.5681849867105484,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1554229725152254,
+      "step": 1682
+    },
+    {
+      "epoch": 0.2805,
+      "grad_norm": 27.0,
+      "grad_norm_var": 3.4077473958333333,
+      "learning_rate": 8.183083804041921e-05,
+      "loss": 6.9801,
+      "loss/crossentropy": 2.0765576660633087,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18441308662295341,
+      "step": 1683
+    },
+    {
+      "epoch": 0.2806666666666667,
+      "grad_norm": 27.625,
+      "grad_norm_var": 3.6369140625,
+      "learning_rate": 8.181064422423748e-05,
+      "loss": 6.8937,
+      "loss/crossentropy": 1.7815276384353638,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14771001040935516,
+      "step": 1684
+    },
+    {
+      "epoch": 0.2808333333333333,
+      "grad_norm": 27.75,
+      "grad_norm_var": 3.79140625,
+      "learning_rate": 8.179044168698721e-05,
+      "loss": 6.9023,
+      "loss/crossentropy": 1.639664113521576,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1681507918983698,
+      "step": 1685
+    },
+    {
+      "epoch": 0.281,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.771875,
+      "learning_rate": 8.177023043420705e-05,
+      "loss": 7.1269,
+      "loss/crossentropy": 2.0239316821098328,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1882992461323738,
+      "step": 1686
+    },
+    {
+      "epoch": 0.2811666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.80625,
+      "learning_rate": 8.175001047143804e-05,
+      "loss": 7.085,
+      "loss/crossentropy": 2.0507695376873016,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16983158141374588,
+      "step": 1687
+    },
+    {
+      "epoch": 0.2813333333333333,
+      "grad_norm": 28.125,
+      "grad_norm_var": 3.4947265625,
+      "learning_rate": 8.172978180422358e-05,
+      "loss": 7.1007,
+      "loss/crossentropy": 2.7620102167129517,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19572355598211288,
+      "step": 1688
+    },
+    {
+      "epoch": 0.2815,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.4947265625,
+      "learning_rate": 8.170954443810948e-05,
+      "loss": 7.2013,
+      "loss/crossentropy": 1.8416691422462463,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1706191748380661,
+      "step": 1689
+    },
+    {
+      "epoch": 0.2816666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.4947265625,
+      "learning_rate": 8.168929837864395e-05,
+      "loss": 7.1158,
+      "loss/crossentropy": 2.668276786804199,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18149346485733986,
+      "step": 1690
+    },
+    {
+      "epoch": 0.2818333333333333,
+      "grad_norm": 2936012800.0,
+      "grad_norm_var": 5.3876068687544326e+17,
+      "learning_rate": 8.16690436313775e-05,
+      "loss": 8.5703,
+      "loss/crossentropy": 1.6693487614393234,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20438792556524277,
+      "step": 1691
+    },
+    {
+      "epoch": 0.282,
+      "grad_norm": 45.25,
+      "grad_norm_var": 5.387606864870333e+17,
+      "learning_rate": 8.164878020186317e-05,
+      "loss": 7.473,
+      "loss/crossentropy": 2.0202498733997345,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17249231040477753,
+      "step": 1692
+    },
+    {
+      "epoch": 0.2821666666666667,
+      "grad_norm": 42.0,
+      "grad_norm_var": 5.3876068623319046e+17,
+      "learning_rate": 8.162850809565623e-05,
+      "loss": 7.0047,
+      "loss/crossentropy": 2.07115176320076,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1747579462826252,
+      "step": 1693
+    },
+    {
+      "epoch": 0.2823333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 5.38760686193432e+17,
+      "learning_rate": 8.160822731831441e-05,
+      "loss": 7.2528,
+      "loss/crossentropy": 1.8939898312091827,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2111487165093422,
+      "step": 1694
+    },
+    {
+      "epoch": 0.2825,
+      "grad_norm": 30.5,
+      "grad_norm_var": 5.387606862240154e+17,
+      "learning_rate": 8.158793787539782e-05,
+      "loss": 7.3467,
+      "loss/crossentropy": 1.8580593168735504,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18915199115872383,
+      "step": 1695
+    },
+    {
+      "epoch": 0.2826666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 5.387606862087237e+17,
+      "learning_rate": 8.156763977246889e-05,
+      "loss": 6.9253,
+      "loss/crossentropy": 1.6111495792865753,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1660899519920349,
+      "step": 1696
+    },
+    {
+      "epoch": 0.2828333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 5.3876068608027315e+17,
+      "learning_rate": 8.154733301509248e-05,
+      "loss": 7.045,
+      "loss/crossentropy": 1.6356629431247711,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18091236054897308,
+      "step": 1697
+    },
+    {
+      "epoch": 0.283,
+      "grad_norm": 26.5,
+      "grad_norm_var": 5.3876068611085664e+17,
+      "learning_rate": 8.152701760883581e-05,
+      "loss": 6.9901,
+      "loss/crossentropy": 1.8398661017417908,
+      "loss/hidden": 3.05078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16181451827287674,
+      "step": 1698
+    },
+    {
+      "epoch": 0.2831666666666667,
+      "grad_norm": 26.875,
+      "grad_norm_var": 5.3876068611391494e+17,
+      "learning_rate": 8.150669355926846e-05,
+      "loss": 6.9348,
+      "loss/crossentropy": 1.4676976054906845,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14524751529097557,
+      "step": 1699
+    },
+    {
+      "epoch": 0.2833333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 5.387606860772148e+17,
+      "learning_rate": 8.148636087196237e-05,
+      "loss": 7.3202,
+      "loss/crossentropy": 2.0150266885757446,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17759757861495018,
+      "step": 1700
+    },
+    {
+      "epoch": 0.2835,
+      "grad_norm": 35.0,
+      "grad_norm_var": 5.387606858998307e+17,
+      "learning_rate": 8.146601955249188e-05,
+      "loss": 6.9675,
+      "loss/crossentropy": 1.7606062293052673,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14642946049571037,
+      "step": 1701
+    },
+    {
+      "epoch": 0.2836666666666667,
+      "grad_norm": 31.875,
+      "grad_norm_var": 5.387606858875973e+17,
+      "learning_rate": 8.144566960643367e-05,
+      "loss": 6.8975,
+      "loss/crossentropy": 1.55070061981678,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14741745591163635,
+      "step": 1702
+    },
+    {
+      "epoch": 0.2838333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 5.38760685893714e+17,
+      "learning_rate": 8.142531103936678e-05,
+      "loss": 7.0671,
+      "loss/crossentropy": 1.6155748516321182,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17731787636876106,
+      "step": 1703
+    },
+    {
+      "epoch": 0.284,
+      "grad_norm": 28.75,
+      "grad_norm_var": 5.387606858784223e+17,
+      "learning_rate": 8.140494385687265e-05,
+      "loss": 6.9621,
+      "loss/crossentropy": 1.822475105524063,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2010866068303585,
+      "step": 1704
+    },
+    {
+      "epoch": 0.2841666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 5.3876068586618886e+17,
+      "learning_rate": 8.138456806453503e-05,
+      "loss": 7.2334,
+      "loss/crossentropy": 2.269442915916443,
+      "loss/hidden": 3.03125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15679952129721642,
+      "step": 1705
+    },
+    {
+      "epoch": 0.2843333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 5.3876068590900576e+17,
+      "learning_rate": 8.136418366794008e-05,
+      "loss": 7.2577,
+      "loss/crossentropy": 2.489030599594116,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19774607941508293,
+      "step": 1706
+    },
+    {
+      "epoch": 0.2845,
+      "grad_norm": 27.0,
+      "grad_norm_var": 27.508072916666666,
+      "learning_rate": 8.13437906726763e-05,
+      "loss": 6.9963,
+      "loss/crossentropy": 2.202205926179886,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18003078177571297,
+      "step": 1707
+    },
+    {
+      "epoch": 0.2846666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 14.268489583333333,
+      "learning_rate": 8.132338908433454e-05,
+      "loss": 7.0833,
+      "loss/crossentropy": 1.8099579513072968,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18593353033065796,
+      "step": 1708
+    },
+    {
+      "epoch": 0.2848333333333333,
+      "grad_norm": 28.625,
+      "grad_norm_var": 5.135872395833333,
+      "learning_rate": 8.130297890850802e-05,
+      "loss": 6.9187,
+      "loss/crossentropy": 1.9506973773241043,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1553124114871025,
+      "step": 1709
+    },
+    {
+      "epoch": 0.285,
+      "grad_norm": 29.5,
+      "grad_norm_var": 5.130208333333333,
+      "learning_rate": 8.128256015079229e-05,
+      "loss": 7.0218,
+      "loss/crossentropy": 2.1073566675186157,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17732353508472443,
+      "step": 1710
+    },
+    {
+      "epoch": 0.2851666666666667,
+      "grad_norm": 26.75,
+      "grad_norm_var": 5.649739583333333,
+      "learning_rate": 8.126213281678528e-05,
+      "loss": 7.0665,
+      "loss/crossentropy": 2.359982818365097,
+      "loss/hidden": 3.08984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17730343714356422,
+      "step": 1711
+    },
+    {
+      "epoch": 0.2853333333333333,
+      "grad_norm": 27.625,
+      "grad_norm_var": 5.84765625,
+      "learning_rate": 8.124169691208723e-05,
+      "loss": 6.8255,
+      "loss/crossentropy": 1.7185687720775604,
+      "loss/hidden": 3.0859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16415394470095634,
+      "step": 1712
+    },
+    {
+      "epoch": 0.2855,
+      "grad_norm": 27.5,
+      "grad_norm_var": 4.81015625,
+      "learning_rate": 8.122125244230079e-05,
+      "loss": 6.8865,
+      "loss/crossentropy": 2.036595106124878,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15688344836235046,
+      "step": 1713
+    },
+    {
+      "epoch": 0.2856666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.947916666666667,
+      "learning_rate": 8.120079941303094e-05,
+      "loss": 7.3588,
+      "loss/crossentropy": 2.0053426325321198,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32154810056090355,
+      "step": 1714
+    },
+    {
+      "epoch": 0.28583333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.5947265625,
+      "learning_rate": 8.118033782988496e-05,
+      "loss": 7.1988,
+      "loss/crossentropy": 1.8093489557504654,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16956805437803268,
+      "step": 1715
+    },
+    {
+      "epoch": 0.286,
+      "grad_norm": 29.0,
+      "grad_norm_var": 4.60390625,
+      "learning_rate": 8.115986769847252e-05,
+      "loss": 6.9215,
+      "loss/crossentropy": 1.986009955406189,
+      "loss/hidden": 3.13671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19330377504229546,
+      "step": 1716
+    },
+    {
+      "epoch": 0.2861666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.5462890625,
+      "learning_rate": 8.113938902440564e-05,
+      "loss": 6.8977,
+      "loss/crossentropy": 1.7001326978206635,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1447098609060049,
+      "step": 1717
+    },
+    {
+      "epoch": 0.28633333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.17890625,
+      "learning_rate": 8.111890181329863e-05,
+      "loss": 6.9204,
+      "loss/crossentropy": 1.9453524053096771,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15566342696547508,
+      "step": 1718
+    },
+    {
+      "epoch": 0.2865,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.403580729166667,
+      "learning_rate": 8.109840607076821e-05,
+      "loss": 6.9269,
+      "loss/crossentropy": 1.6534964740276337,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21997592598199844,
+      "step": 1719
+    },
+    {
+      "epoch": 0.2866666666666667,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.4879557291666665,
+      "learning_rate": 8.107790180243338e-05,
+      "loss": 6.9981,
+      "loss/crossentropy": 2.1482350528240204,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17027843743562698,
+      "step": 1720
+    },
+    {
+      "epoch": 0.28683333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.669791666666667,
+      "learning_rate": 8.105738901391552e-05,
+      "loss": 6.9931,
+      "loss/crossentropy": 1.7035507559776306,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1677701696753502,
+      "step": 1721
+    },
+    {
+      "epoch": 0.287,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.6791666666666667,
+      "learning_rate": 8.103686771083831e-05,
+      "loss": 6.8921,
+      "loss/crossentropy": 1.9129577577114105,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15843845531344414,
+      "step": 1722
+    },
+    {
+      "epoch": 0.2871666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.6205729166666667,
+      "learning_rate": 8.101633789882781e-05,
+      "loss": 6.9713,
+      "loss/crossentropy": 1.9466756880283356,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18042508512735367,
+      "step": 1723
+    },
+    {
+      "epoch": 0.28733333333333333,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.7504557291666667,
+      "learning_rate": 8.099579958351235e-05,
+      "loss": 6.8711,
+      "loss/crossentropy": 1.822874754667282,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1604704149067402,
+      "step": 1724
+    },
+    {
+      "epoch": 0.2875,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.1858723958333335,
+      "learning_rate": 8.097525277052264e-05,
+      "loss": 7.2213,
+      "loss/crossentropy": 1.8169062584638596,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17779788188636303,
+      "step": 1725
+    },
+    {
+      "epoch": 0.2876666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 4.106705729166666,
+      "learning_rate": 8.095469746549172e-05,
+      "loss": 6.8249,
+      "loss/crossentropy": 1.7158812582492828,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1985117420554161,
+      "step": 1726
+    },
+    {
+      "epoch": 0.28783333333333333,
+      "grad_norm": 27.5,
+      "grad_norm_var": 3.9067057291666667,
+      "learning_rate": 8.093413367405489e-05,
+      "loss": 7.0233,
+      "loss/crossentropy": 1.906646966934204,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15743854269385338,
+      "step": 1727
+    },
+    {
+      "epoch": 0.288,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.9330729166666667,
+      "learning_rate": 8.091356140184991e-05,
+      "loss": 6.9968,
+      "loss/crossentropy": 1.7871026396751404,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16673271730542183,
+      "step": 1728
+    },
+    {
+      "epoch": 0.2881666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.8309895833333334,
+      "learning_rate": 8.089298065451672e-05,
+      "loss": 6.8641,
+      "loss/crossentropy": 1.9663325399160385,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16698668524622917,
+      "step": 1729
+    },
+    {
+      "epoch": 0.28833333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.5093098958333333,
+      "learning_rate": 8.087239143769768e-05,
+      "loss": 7.1978,
+      "loss/crossentropy": 1.7978715002536774,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18057232350111008,
+      "step": 1730
+    },
+    {
+      "epoch": 0.2885,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.4061848958333334,
+      "learning_rate": 8.085179375703744e-05,
+      "loss": 7.3673,
+      "loss/crossentropy": 2.15347021818161,
+      "loss/hidden": 3.16015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16674425080418587,
+      "step": 1731
+    },
+    {
+      "epoch": 0.2886666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.4061848958333334,
+      "learning_rate": 8.083118761818295e-05,
+      "loss": 7.001,
+      "loss/crossentropy": 1.6970942318439484,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18776270374655724,
+      "step": 1732
+    },
+    {
+      "epoch": 0.28883333333333333,
+      "grad_norm": 28.0,
+      "grad_norm_var": 3.5205729166666666,
+      "learning_rate": 8.081057302678352e-05,
+      "loss": 6.8459,
+      "loss/crossentropy": 1.8751520812511444,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19694062322378159,
+      "step": 1733
+    },
+    {
+      "epoch": 0.289,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.5614583333333334,
+      "learning_rate": 8.078994998849076e-05,
+      "loss": 7.0133,
+      "loss/crossentropy": 2.1347344666719437,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16332196071743965,
+      "step": 1734
+    },
+    {
+      "epoch": 0.2891666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 3.4,
+      "learning_rate": 8.076931850895859e-05,
+      "loss": 7.0216,
+      "loss/crossentropy": 1.960142195224762,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17830370739102364,
+      "step": 1735
+    },
+    {
+      "epoch": 0.28933333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.228125,
+      "learning_rate": 8.074867859384322e-05,
+      "loss": 6.9388,
+      "loss/crossentropy": 1.9251831322908401,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1704772673547268,
+      "step": 1736
+    },
+    {
+      "epoch": 0.2895,
+      "grad_norm": 28.375,
+      "grad_norm_var": 3.191666666666667,
+      "learning_rate": 8.072803024880322e-05,
+      "loss": 7.0088,
+      "loss/crossentropy": 2.005739152431488,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1563599295914173,
+      "step": 1737
+    },
+    {
+      "epoch": 0.2896666666666667,
+      "grad_norm": 27.875,
+      "grad_norm_var": 3.294205729166667,
+      "learning_rate": 8.070737347949947e-05,
+      "loss": 6.8871,
+      "loss/crossentropy": 1.703059434890747,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19055244326591492,
+      "step": 1738
+    },
+    {
+      "epoch": 0.28983333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.972916666666667,
+      "learning_rate": 8.068670829159511e-05,
+      "loss": 6.7825,
+      "loss/crossentropy": 1.2570236921310425,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1484133694320917,
+      "step": 1739
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 27.0,
+      "grad_norm_var": 3.014518229166667,
+      "learning_rate": 8.066603469075564e-05,
+      "loss": 6.9674,
+      "loss/crossentropy": 1.5935139805078506,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15925994515419006,
+      "step": 1740
+    },
+    {
+      "epoch": 0.2901666666666667,
+      "grad_norm": 25.5,
+      "grad_norm_var": 3.745833333333333,
+      "learning_rate": 8.064535268264883e-05,
+      "loss": 6.9103,
+      "loss/crossentropy": 1.954742968082428,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17588535323739052,
+      "step": 1741
+    },
+    {
+      "epoch": 0.29033333333333333,
+      "grad_norm": 27.125,
+      "grad_norm_var": 3.051497395833333,
+      "learning_rate": 8.062466227294477e-05,
+      "loss": 6.9033,
+      "loss/crossentropy": 1.9432837516069412,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19504057615995407,
+      "step": 1742
+    },
+    {
+      "epoch": 0.2905,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.937239583333333,
+      "learning_rate": 8.060396346731587e-05,
+      "loss": 7.2303,
+      "loss/crossentropy": 1.8995141685009003,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18443816527724266,
+      "step": 1743
+    },
+    {
+      "epoch": 0.2906666666666667,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.6416015625,
+      "learning_rate": 8.058325627143681e-05,
+      "loss": 7.148,
+      "loss/crossentropy": 2.1386114954948425,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16721056401729584,
+      "step": 1744
+    },
+    {
+      "epoch": 0.29083333333333333,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.3372395833333335,
+      "learning_rate": 8.056254069098459e-05,
+      "loss": 6.9154,
+      "loss/crossentropy": 2.123811662197113,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1626821719110012,
+      "step": 1745
+    },
+    {
+      "epoch": 0.291,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.678580729166667,
+      "learning_rate": 8.05418167316385e-05,
+      "loss": 6.9848,
+      "loss/crossentropy": 2.012488305568695,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18469147756695747,
+      "step": 1746
+    },
+    {
+      "epoch": 0.2911666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.6051432291666665,
+      "learning_rate": 8.052108439908013e-05,
+      "loss": 7.2118,
+      "loss/crossentropy": 2.222733795642853,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17967507988214493,
+      "step": 1747
+    },
+    {
+      "epoch": 0.29133333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.9572265625,
+      "learning_rate": 8.050034369899337e-05,
+      "loss": 7.0723,
+      "loss/crossentropy": 1.35679230093956,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19271300174295902,
+      "step": 1748
+    },
+    {
+      "epoch": 0.2915,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.295768229166667,
+      "learning_rate": 8.04795946370644e-05,
+      "loss": 7.2028,
+      "loss/crossentropy": 2.0392096638679504,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1761065199971199,
+      "step": 1749
+    },
+    {
+      "epoch": 0.2916666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.9082682291666666,
+      "learning_rate": 8.04588372189817e-05,
+      "loss": 6.8845,
+      "loss/crossentropy": 2.246554672718048,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20051004737615585,
+      "step": 1750
+    },
+    {
+      "epoch": 0.29183333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.689322916666667,
+      "learning_rate": 8.043807145043604e-05,
+      "loss": 7.2089,
+      "loss/crossentropy": 2.2228686809539795,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17735590785741806,
+      "step": 1751
+    },
+    {
+      "epoch": 0.292,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.9614583333333333,
+      "learning_rate": 8.041729733712045e-05,
+      "loss": 6.8497,
+      "loss/crossentropy": 1.5177088528871536,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17029394581913948,
+      "step": 1752
+    },
+    {
+      "epoch": 0.2921666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 3.903125,
+      "learning_rate": 8.039651488473028e-05,
+      "loss": 6.8599,
+      "loss/crossentropy": 1.9754594266414642,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19144656881690025,
+      "step": 1753
+    },
+    {
+      "epoch": 0.29233333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.1587890625,
+      "learning_rate": 8.037572409896315e-05,
+      "loss": 7.1755,
+      "loss/crossentropy": 1.9540952146053314,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16342727839946747,
+      "step": 1754
+    },
+    {
+      "epoch": 0.2925,
+      "grad_norm": 30.375,
+      "grad_norm_var": 4.192643229166666,
+      "learning_rate": 8.0354924985519e-05,
+      "loss": 6.7816,
+      "loss/crossentropy": 1.5585548281669617,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18010758981108665,
+      "step": 1755
+    },
+    {
+      "epoch": 0.2926666666666667,
+      "grad_norm": 36.5,
+      "grad_norm_var": 6.498372395833333,
+      "learning_rate": 8.033411755009999e-05,
+      "loss": 7.1499,
+      "loss/crossentropy": 1.5687600076198578,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22139346599578857,
+      "step": 1756
+    },
+    {
+      "epoch": 0.29283333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 4.962239583333333,
+      "learning_rate": 8.031330179841062e-05,
+      "loss": 6.9437,
+      "loss/crossentropy": 1.6486482918262482,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15065807476639748,
+      "step": 1757
+    },
+    {
+      "epoch": 0.293,
+      "grad_norm": 30.125,
+      "grad_norm_var": 4.180989583333333,
+      "learning_rate": 8.029247773615764e-05,
+      "loss": 6.9791,
+      "loss/crossentropy": 1.7986524105072021,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19983235374093056,
+      "step": 1758
+    },
+    {
+      "epoch": 0.2931666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 4.069791666666666,
+      "learning_rate": 8.027164536905008e-05,
+      "loss": 7.0019,
+      "loss/crossentropy": 2.064461871981621,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1813412606716156,
+      "step": 1759
+    },
+    {
+      "epoch": 0.29333333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.825455729166667,
+      "learning_rate": 8.025080470279924e-05,
+      "loss": 6.8873,
+      "loss/crossentropy": 1.9530519247055054,
+      "loss/hidden": 2.94140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.136128269135952,
+      "step": 1760
+    },
+    {
+      "epoch": 0.2935,
+      "grad_norm": 28.5,
+      "grad_norm_var": 3.601822916666667,
+      "learning_rate": 8.022995574311876e-05,
+      "loss": 6.8462,
+      "loss/crossentropy": 2.069970518350601,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15924708545207977,
+      "step": 1761
+    },
+    {
+      "epoch": 0.2936666666666667,
+      "grad_norm": 28.5,
+      "grad_norm_var": 3.849739583333333,
+      "learning_rate": 8.020909849572444e-05,
+      "loss": 7.0438,
+      "loss/crossentropy": 1.4966088980436325,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17960093170404434,
+      "step": 1762
+    },
+    {
+      "epoch": 0.29383333333333334,
+      "grad_norm": 27.125,
+      "grad_norm_var": 4.487434895833333,
+      "learning_rate": 8.018823296633441e-05,
+      "loss": 6.6857,
+      "loss/crossentropy": 1.6587276309728622,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16952040046453476,
+      "step": 1763
+    },
+    {
+      "epoch": 0.294,
+      "grad_norm": 28.0,
+      "grad_norm_var": 4.871809895833334,
+      "learning_rate": 8.016735916066913e-05,
+      "loss": 7.0478,
+      "loss/crossentropy": 1.9921867847442627,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16137176379561424,
+      "step": 1764
+    },
+    {
+      "epoch": 0.2941666666666667,
+      "grad_norm": 29.125,
+      "grad_norm_var": 4.90390625,
+      "learning_rate": 8.014647708445124e-05,
+      "loss": 6.9749,
+      "loss/crossentropy": 2.3098158836364746,
+      "loss/hidden": 3.01171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15011011064052582,
+      "step": 1765
+    },
+    {
+      "epoch": 0.29433333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.570247395833333,
+      "learning_rate": 8.012558674340566e-05,
+      "loss": 7.109,
+      "loss/crossentropy": 1.5782488584518433,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16313642263412476,
+      "step": 1766
+    },
+    {
+      "epoch": 0.2945,
+      "grad_norm": 29.75,
+      "grad_norm_var": 4.564322916666667,
+      "learning_rate": 8.010468814325964e-05,
+      "loss": 7.0989,
+      "loss/crossentropy": 1.715920865535736,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1874961219727993,
+      "step": 1767
+    },
+    {
+      "epoch": 0.2946666666666667,
+      "grad_norm": 26.5,
+      "grad_norm_var": 5.158072916666667,
+      "learning_rate": 8.008378128974262e-05,
+      "loss": 7.2552,
+      "loss/crossentropy": 1.7260702550411224,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18568307533860207,
+      "step": 1768
+    },
+    {
+      "epoch": 0.29483333333333334,
+      "grad_norm": 28.75,
+      "grad_norm_var": 5.212434895833334,
+      "learning_rate": 8.006286618858635e-05,
+      "loss": 6.8589,
+      "loss/crossentropy": 1.4239223450422287,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17964031919836998,
+      "step": 1769
+    },
+    {
+      "epoch": 0.295,
+      "grad_norm": 28.625,
+      "grad_norm_var": 4.887239583333334,
+      "learning_rate": 8.004194284552477e-05,
+      "loss": 6.809,
+      "loss/crossentropy": 2.0484983325004578,
+      "loss/hidden": 3.125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14857875555753708,
+      "step": 1770
+    },
+    {
+      "epoch": 0.2951666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 4.837434895833334,
+      "learning_rate": 8.002101126629421e-05,
+      "loss": 6.9297,
+      "loss/crossentropy": 2.2283668518066406,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15431294217705727,
+      "step": 1771
+    },
+    {
+      "epoch": 0.29533333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.5239583333333333,
+      "learning_rate": 8.000007145663312e-05,
+      "loss": 6.8355,
+      "loss/crossentropy": 1.6497902870178223,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15394964255392551,
+      "step": 1772
+    },
+    {
+      "epoch": 0.2955,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.8764973958333333,
+      "learning_rate": 7.997912342228232e-05,
+      "loss": 6.8759,
+      "loss/crossentropy": 1.8970504701137543,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15557139366865158,
+      "step": 1773
+    },
+    {
+      "epoch": 0.2956666666666667,
+      "grad_norm": 4110417920.0,
+      "grad_norm_var": 1.0559709523387351e+18,
+      "learning_rate": 7.99581671689848e-05,
+      "loss": 7.3196,
+      "loss/crossentropy": 2.6301337480545044,
+      "loss/hidden": 3.015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17043613269925117,
+      "step": 1774
+    },
+    {
+      "epoch": 0.29583333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.05597095232589e+18,
+      "learning_rate": 7.993720270248584e-05,
+      "loss": 6.9568,
+      "loss/crossentropy": 1.719387263059616,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19741066545248032,
+      "step": 1775
+    },
+    {
+      "epoch": 0.296,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.0559709523344535e+18,
+      "learning_rate": 7.991623002853296e-05,
+      "loss": 6.6749,
+      "loss/crossentropy": 1.155877485871315,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13954339548945427,
+      "step": 1776
+    },
+    {
+      "epoch": 0.2961666666666667,
+      "grad_norm": 28.0,
+      "grad_norm_var": 1.0559709523515802e+18,
+      "learning_rate": 7.989524915287595e-05,
+      "loss": 7.1404,
+      "loss/crossentropy": 2.134528160095215,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17305652052164078,
+      "step": 1777
+    },
+    {
+      "epoch": 0.29633333333333334,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.0559709523430168e+18,
+      "learning_rate": 7.987426008126683e-05,
+      "loss": 7.1646,
+      "loss/crossentropy": 2.330888092517853,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17424631491303444,
+      "step": 1778
+    },
+    {
+      "epoch": 0.2965,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.0559709523044817e+18,
+      "learning_rate": 7.985326281945989e-05,
+      "loss": 7.0201,
+      "loss/crossentropy": 2.036402851343155,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16846515238285065,
+      "step": 1779
+    },
+    {
+      "epoch": 0.2966666666666667,
+      "grad_norm": 28.0,
+      "grad_norm_var": 1.0559709523044817e+18,
+      "learning_rate": 7.983225737321163e-05,
+      "loss": 6.9704,
+      "loss/crossentropy": 2.133853703737259,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16769133880734444,
+      "step": 1780
+    },
+    {
+      "epoch": 0.29683333333333334,
+      "grad_norm": 26.25,
+      "grad_norm_var": 1.0559709524029604e+18,
+      "learning_rate": 7.98112437482808e-05,
+      "loss": 6.9933,
+      "loss/crossentropy": 1.7176109552383423,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17507165670394897,
+      "step": 1781
+    },
+    {
+      "epoch": 0.297,
+      "grad_norm": 27.75,
+      "grad_norm_var": 1.0559709524586223e+18,
+      "learning_rate": 7.979022195042842e-05,
+      "loss": 6.9713,
+      "loss/crossentropy": 1.7080380022525787,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1753886453807354,
+      "step": 1782
+    },
+    {
+      "epoch": 0.2971666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.055970952462904e+18,
+      "learning_rate": 7.976919198541776e-05,
+      "loss": 6.9518,
+      "loss/crossentropy": 2.0936281085014343,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2058982402086258,
+      "step": 1783
+    },
+    {
+      "epoch": 0.29733333333333334,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.0559709524029604e+18,
+      "learning_rate": 7.974815385901426e-05,
+      "loss": 6.9159,
+      "loss/crossentropy": 2.0488981008529663,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1739674136042595,
+      "step": 1784
+    },
+    {
+      "epoch": 0.2975,
+      "grad_norm": 27.875,
+      "grad_norm_var": 1.0559709524329322e+18,
+      "learning_rate": 7.972710757698567e-05,
+      "loss": 7.1597,
+      "loss/crossentropy": 2.336401492357254,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17067568749189377,
+      "step": 1785
+    },
+    {
+      "epoch": 0.2976666666666667,
+      "grad_norm": 27.875,
+      "grad_norm_var": 1.0559709524586223e+18,
+      "learning_rate": 7.970605314510194e-05,
+      "loss": 7.0471,
+      "loss/crossentropy": 2.010293275117874,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17855480313301086,
+      "step": 1786
+    },
+    {
+      "epoch": 0.29783333333333334,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.0559709524843124e+18,
+      "learning_rate": 7.968499056913524e-05,
+      "loss": 7.2995,
+      "loss/crossentropy": 2.0745608508586884,
+      "loss/hidden": 3.640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30422502383589745,
+      "step": 1787
+    },
+    {
+      "epoch": 0.298,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.0559709525271293e+18,
+      "learning_rate": 7.966391985486003e-05,
+      "loss": 6.8864,
+      "loss/crossentropy": 2.19223316013813,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16438164189457893,
+      "step": 1788
+    },
+    {
+      "epoch": 0.2981666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.0559709525742278e+18,
+      "learning_rate": 7.964284100805297e-05,
+      "loss": 6.9538,
+      "loss/crossentropy": 1.9663271009922028,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20872559025883675,
+      "step": 1789
+    },
+    {
+      "epoch": 0.29833333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.2556640625,
+      "learning_rate": 7.96217540344929e-05,
+      "loss": 7.1325,
+      "loss/crossentropy": 2.1852712631225586,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18366773799061775,
+      "step": 1790
+    },
+    {
+      "epoch": 0.2985,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.7837890625,
+      "learning_rate": 7.960065893996098e-05,
+      "loss": 7.1386,
+      "loss/crossentropy": 1.9224089086055756,
+      "loss/hidden": 3.12890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1370961107313633,
+      "step": 1791
+    },
+    {
+      "epoch": 0.2986666666666667,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.89140625,
+      "learning_rate": 7.957955573024052e-05,
+      "loss": 6.9062,
+      "loss/crossentropy": 1.969079315662384,
+      "loss/hidden": 3.0390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14765902236104012,
+      "step": 1792
+    },
+    {
+      "epoch": 0.29883333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.6927083333333335,
+      "learning_rate": 7.95584444111171e-05,
+      "loss": 7.0947,
+      "loss/crossentropy": 2.089136302471161,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2019180990755558,
+      "step": 1793
+    },
+    {
+      "epoch": 0.299,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.6968098958333333,
+      "learning_rate": 7.95373249883785e-05,
+      "loss": 6.711,
+      "loss/crossentropy": 1.5635971426963806,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20058362558484077,
+      "step": 1794
+    },
+    {
+      "epoch": 0.2991666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.6712890625,
+      "learning_rate": 7.951619746781474e-05,
+      "loss": 6.7382,
+      "loss/crossentropy": 1.8677619993686676,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1745459921658039,
+      "step": 1795
+    },
+    {
+      "epoch": 0.29933333333333334,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.725,
+      "learning_rate": 7.949506185521802e-05,
+      "loss": 6.9398,
+      "loss/crossentropy": 1.988951563835144,
+      "loss/hidden": 3.09375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16441595181822777,
+      "step": 1796
+    },
+    {
+      "epoch": 0.2995,
+      "grad_norm": 26.625,
+      "grad_norm_var": 2.6025390625,
+      "learning_rate": 7.947391815638284e-05,
+      "loss": 7.0052,
+      "loss/crossentropy": 1.6859848499298096,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2239336222410202,
+      "step": 1797
+    },
+    {
+      "epoch": 0.2996666666666667,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.6447265625,
+      "learning_rate": 7.945276637710582e-05,
+      "loss": 6.7844,
+      "loss/crossentropy": 1.5138292014598846,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17344660498201847,
+      "step": 1798
+    },
+    {
+      "epoch": 0.29983333333333334,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.662955729166667,
+      "learning_rate": 7.943160652318585e-05,
+      "loss": 7.0734,
+      "loss/crossentropy": 1.9331949353218079,
+      "loss/hidden": 3.125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16637888550758362,
+      "step": 1799
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 27.375,
+      "grad_norm_var": 2.771875,
+      "learning_rate": 7.941043860042403e-05,
+      "loss": 6.8693,
+      "loss/crossentropy": 2.1852831840515137,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1701827198266983,
+      "step": 1800
+    },
+    {
+      "epoch": 0.3001666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.7244140625,
+      "learning_rate": 7.938926261462366e-05,
+      "loss": 6.8283,
+      "loss/crossentropy": 2.070968210697174,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.190330121666193,
+      "step": 1801
+    },
+    {
+      "epoch": 0.30033333333333334,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.83515625,
+      "learning_rate": 7.936807857159026e-05,
+      "loss": 6.9224,
+      "loss/crossentropy": 2.1456773579120636,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18744849041104317,
+      "step": 1802
+    },
+    {
+      "epoch": 0.3005,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.9072265625,
+      "learning_rate": 7.934688647713158e-05,
+      "loss": 7.1863,
+      "loss/crossentropy": 1.7114560008049011,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20561952143907547,
+      "step": 1803
+    },
+    {
+      "epoch": 0.3006666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.1962890625,
+      "learning_rate": 7.932568633705752e-05,
+      "loss": 7.0394,
+      "loss/crossentropy": 1.9421786963939667,
+      "loss/hidden": 3.125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1410901565104723,
+      "step": 1804
+    },
+    {
+      "epoch": 0.30083333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.4119140625,
+      "learning_rate": 7.930447815718022e-05,
+      "loss": 7.1017,
+      "loss/crossentropy": 1.1867583096027374,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2174985706806183,
+      "step": 1805
+    },
+    {
+      "epoch": 0.301,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.7509765625,
+      "learning_rate": 7.928326194331404e-05,
+      "loss": 6.9254,
+      "loss/crossentropy": 1.2460341155529022,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15135955065488815,
+      "step": 1806
+    },
+    {
+      "epoch": 0.3011666666666667,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.77265625,
+      "learning_rate": 7.926203770127552e-05,
+      "loss": 6.7957,
+      "loss/crossentropy": 1.5237163305282593,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18431520834565163,
+      "step": 1807
+    },
+    {
+      "epoch": 0.30133333333333334,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.77265625,
+      "learning_rate": 7.924080543688337e-05,
+      "loss": 6.8361,
+      "loss/crossentropy": 1.6605301201343536,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16298114135861397,
+      "step": 1808
+    },
+    {
+      "epoch": 0.3015,
+      "grad_norm": 26.0,
+      "grad_norm_var": 2.284375,
+      "learning_rate": 7.921956515595861e-05,
+      "loss": 7.2457,
+      "loss/crossentropy": 1.8478024005889893,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16555768065154552,
+      "step": 1809
+    },
+    {
+      "epoch": 0.3016666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.6186848958333333,
+      "learning_rate": 7.919831686432433e-05,
+      "loss": 6.9624,
+      "loss/crossentropy": 2.0245045125484467,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17697810754179955,
+      "step": 1810
+    },
+    {
+      "epoch": 0.30183333333333334,
+      "grad_norm": 27.375,
+      "grad_norm_var": 2.6809895833333335,
+      "learning_rate": 7.917706056780587e-05,
+      "loss": 7.0756,
+      "loss/crossentropy": 1.7672982215881348,
+      "loss/hidden": 3.09765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1477302983403206,
+      "step": 1811
+    },
+    {
+      "epoch": 0.302,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.6645182291666667,
+      "learning_rate": 7.915579627223079e-05,
+      "loss": 7.2958,
+      "loss/crossentropy": 1.919701874256134,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18852383643388748,
+      "step": 1812
+    },
+    {
+      "epoch": 0.30216666666666664,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.498893229166667,
+      "learning_rate": 7.913452398342881e-05,
+      "loss": 7.1196,
+      "loss/crossentropy": 1.9730682075023651,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1892014592885971,
+      "step": 1813
+    },
+    {
+      "epoch": 0.30233333333333334,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.4184895833333333,
+      "learning_rate": 7.911324370723183e-05,
+      "loss": 6.8,
+      "loss/crossentropy": 1.5512824952602386,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16365887969732285,
+      "step": 1814
+    },
+    {
+      "epoch": 0.3025,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.36640625,
+      "learning_rate": 7.909195544947398e-05,
+      "loss": 6.9363,
+      "loss/crossentropy": 1.9081394374370575,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16053590923547745,
+      "step": 1815
+    },
+    {
+      "epoch": 0.30266666666666664,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.2393229166666666,
+      "learning_rate": 7.907065921599154e-05,
+      "loss": 7.1773,
+      "loss/crossentropy": 1.6458862572908401,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23366864770650864,
+      "step": 1816
+    },
+    {
+      "epoch": 0.30283333333333334,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.3135416666666666,
+      "learning_rate": 7.9049355012623e-05,
+      "loss": 6.986,
+      "loss/crossentropy": 1.9764463603496552,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17300384491682053,
+      "step": 1817
+    },
+    {
+      "epoch": 0.303,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.1348307291666666,
+      "learning_rate": 7.902804284520903e-05,
+      "loss": 7.1401,
+      "loss/crossentropy": 1.9698999524116516,
+      "loss/hidden": 3.14453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15904537588357925,
+      "step": 1818
+    },
+    {
+      "epoch": 0.30316666666666664,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.396875,
+      "learning_rate": 7.900672271959247e-05,
+      "loss": 6.9962,
+      "loss/crossentropy": 2.1794964373111725,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.178677499294281,
+      "step": 1819
+    },
+    {
+      "epoch": 0.30333333333333334,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.1889973958333333,
+      "learning_rate": 7.898539464161838e-05,
+      "loss": 7.1041,
+      "loss/crossentropy": 1.5868679732084274,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16324621438980103,
+      "step": 1820
+    },
+    {
+      "epoch": 0.3035,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.8764973958333333,
+      "learning_rate": 7.896405861713394e-05,
+      "loss": 6.8576,
+      "loss/crossentropy": 1.6912561655044556,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16031718254089355,
+      "step": 1821
+    },
+    {
+      "epoch": 0.30366666666666664,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.8452473958333333,
+      "learning_rate": 7.894271465198857e-05,
+      "loss": 7.1175,
+      "loss/crossentropy": 1.627418577671051,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18628519400954247,
+      "step": 1822
+    },
+    {
+      "epoch": 0.30383333333333334,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.9525390625,
+      "learning_rate": 7.892136275203383e-05,
+      "loss": 6.8954,
+      "loss/crossentropy": 1.6395491808652878,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1456153430044651,
+      "step": 1823
+    },
+    {
+      "epoch": 0.304,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.8629557291666667,
+      "learning_rate": 7.890000292312346e-05,
+      "loss": 6.9028,
+      "loss/crossentropy": 1.851139396429062,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17918989807367325,
+      "step": 1824
+    },
+    {
+      "epoch": 0.30416666666666664,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.4104166666666667,
+      "learning_rate": 7.887863517111338e-05,
+      "loss": 6.9461,
+      "loss/crossentropy": 2.1613464653491974,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16244490072131157,
+      "step": 1825
+    },
+    {
+      "epoch": 0.30433333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.1643229166666667,
+      "learning_rate": 7.88572595018617e-05,
+      "loss": 6.6463,
+      "loss/crossentropy": 1.636417955160141,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17213625274598598,
+      "step": 1826
+    },
+    {
+      "epoch": 0.3045,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.4260416666666667,
+      "learning_rate": 7.883587592122863e-05,
+      "loss": 6.9601,
+      "loss/crossentropy": 1.8333800733089447,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1667829230427742,
+      "step": 1827
+    },
+    {
+      "epoch": 0.30466666666666664,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.051041666666667,
+      "learning_rate": 7.881448443507664e-05,
+      "loss": 6.9331,
+      "loss/crossentropy": 1.9271169006824493,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15610137209296227,
+      "step": 1828
+    },
+    {
+      "epoch": 0.30483333333333335,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.034375,
+      "learning_rate": 7.879308504927035e-05,
+      "loss": 7.1442,
+      "loss/crossentropy": 1.9399696290493011,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21466120332479477,
+      "step": 1829
+    },
+    {
+      "epoch": 0.305,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.0509765625,
+      "learning_rate": 7.877167776967645e-05,
+      "loss": 7.1896,
+      "loss/crossentropy": 2.351029932498932,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21430783718824387,
+      "step": 1830
+    },
+    {
+      "epoch": 0.30516666666666664,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.0509765625,
+      "learning_rate": 7.875026260216393e-05,
+      "loss": 6.7793,
+      "loss/crossentropy": 2.0612818002700806,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2058713324368,
+      "step": 1831
+    },
+    {
+      "epoch": 0.30533333333333335,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.1087890625,
+      "learning_rate": 7.872883955260387e-05,
+      "loss": 7.0005,
+      "loss/crossentropy": 2.0101443231105804,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21682747825980186,
+      "step": 1832
+    },
+    {
+      "epoch": 0.3055,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.0020182291666666,
+      "learning_rate": 7.87074086268695e-05,
+      "loss": 6.8333,
+      "loss/crossentropy": 2.177095741033554,
+      "loss/hidden": 3.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1732555888593197,
+      "step": 1833
+    },
+    {
+      "epoch": 0.30566666666666664,
+      "grad_norm": 26.625,
+      "grad_norm_var": 2.3572265625,
+      "learning_rate": 7.868596983083623e-05,
+      "loss": 6.9011,
+      "loss/crossentropy": 2.169644057750702,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17804266512393951,
+      "step": 1834
+    },
+    {
+      "epoch": 0.30583333333333335,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.062239583333333,
+      "learning_rate": 7.866452317038164e-05,
+      "loss": 7.1436,
+      "loss/crossentropy": 2.098512828350067,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15170537307858467,
+      "step": 1835
+    },
+    {
+      "epoch": 0.306,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.9546223958333333,
+      "learning_rate": 7.864306865138545e-05,
+      "loss": 7.0301,
+      "loss/crossentropy": 1.9148472547531128,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1804008036851883,
+      "step": 1836
+    },
+    {
+      "epoch": 0.30616666666666664,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.9145182291666667,
+      "learning_rate": 7.862160627972955e-05,
+      "loss": 6.7096,
+      "loss/crossentropy": 1.6348781734704971,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1898255255073309,
+      "step": 1837
+    },
+    {
+      "epoch": 0.30633333333333335,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.3275390625,
+      "learning_rate": 7.860013606129796e-05,
+      "loss": 7.2805,
+      "loss/crossentropy": 1.8533981144428253,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17210083827376366,
+      "step": 1838
+    },
+    {
+      "epoch": 0.3065,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.292708333333333,
+      "learning_rate": 7.857865800197684e-05,
+      "loss": 7.0091,
+      "loss/crossentropy": 2.3175882399082184,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17278634011745453,
+      "step": 1839
+    },
+    {
+      "epoch": 0.30666666666666664,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.551497395833333,
+      "learning_rate": 7.855717210765456e-05,
+      "loss": 7.0248,
+      "loss/crossentropy": 2.3028687834739685,
+      "loss/hidden": 3.1171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15742110460996628,
+      "step": 1840
+    },
+    {
+      "epoch": 0.30683333333333335,
+      "grad_norm": 27.375,
+      "grad_norm_var": 2.738997395833333,
+      "learning_rate": 7.85356783842216e-05,
+      "loss": 6.9937,
+      "loss/crossentropy": 2.003747597336769,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16512829810380936,
+      "step": 1841
+    },
+    {
+      "epoch": 0.307,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.7327473958333335,
+      "learning_rate": 7.851417683757053e-05,
+      "loss": 6.9281,
+      "loss/crossentropy": 1.8588949143886566,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17192811891436577,
+      "step": 1842
+    },
+    {
+      "epoch": 0.30716666666666664,
+      "grad_norm": 26.875,
+      "grad_norm_var": 2.6796223958333334,
+      "learning_rate": 7.849266747359619e-05,
+      "loss": 6.9215,
+      "loss/crossentropy": 2.1861804723739624,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15846161916851997,
+      "step": 1843
+    },
+    {
+      "epoch": 0.30733333333333335,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.9395833333333334,
+      "learning_rate": 7.847115029819547e-05,
+      "loss": 6.6718,
+      "loss/crossentropy": 1.7861296236515045,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18043068796396255,
+      "step": 1844
+    },
+    {
+      "epoch": 0.3075,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.9103515625,
+      "learning_rate": 7.84496253172674e-05,
+      "loss": 6.9739,
+      "loss/crossentropy": 2.0851745307445526,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1976468451321125,
+      "step": 1845
+    },
+    {
+      "epoch": 0.30766666666666664,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.9749348958333335,
+      "learning_rate": 7.84280925367132e-05,
+      "loss": 6.8783,
+      "loss/crossentropy": 1.716008022427559,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15882867947220802,
+      "step": 1846
+    },
+    {
+      "epoch": 0.30783333333333335,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.9983723958333335,
+      "learning_rate": 7.84065519624362e-05,
+      "loss": 6.9803,
+      "loss/crossentropy": 2.172625035047531,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.163181871175766,
+      "step": 1847
+    },
+    {
+      "epoch": 0.308,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.9614583333333333,
+      "learning_rate": 7.838500360034188e-05,
+      "loss": 6.8901,
+      "loss/crossentropy": 1.9254584461450577,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15511753037571907,
+      "step": 1848
+    },
+    {
+      "epoch": 0.30816666666666664,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.987434895833333,
+      "learning_rate": 7.836344745633783e-05,
+      "loss": 6.9298,
+      "loss/crossentropy": 1.5840540379285812,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18015084601938725,
+      "step": 1849
+    },
+    {
+      "epoch": 0.30833333333333335,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.2718098958333335,
+      "learning_rate": 7.83418835363338e-05,
+      "loss": 7.0386,
+      "loss/crossentropy": 2.1513716876506805,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19308988377451897,
+      "step": 1850
+    },
+    {
+      "epoch": 0.3085,
+      "grad_norm": 27.25,
+      "grad_norm_var": 3.42265625,
+      "learning_rate": 7.832031184624164e-05,
+      "loss": 7.2145,
+      "loss/crossentropy": 2.079343467950821,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19979696348309517,
+      "step": 1851
+    },
+    {
+      "epoch": 0.30866666666666664,
+      "grad_norm": 34.0,
+      "grad_norm_var": 5.010872395833333,
+      "learning_rate": 7.829873239197538e-05,
+      "loss": 7.6086,
+      "loss/crossentropy": 2.1194871366024017,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17705166339874268,
+      "step": 1852
+    },
+    {
+      "epoch": 0.30883333333333335,
+      "grad_norm": 28.25,
+      "grad_norm_var": 4.920572916666667,
+      "learning_rate": 7.827714517945115e-05,
+      "loss": 7.0366,
+      "loss/crossentropy": 2.1705428659915924,
+      "loss/hidden": 3.12109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14574232324957848,
+      "step": 1853
+    },
+    {
+      "epoch": 0.309,
+      "grad_norm": 26.5,
+      "grad_norm_var": 4.772916666666666,
+      "learning_rate": 7.825555021458716e-05,
+      "loss": 7.0844,
+      "loss/crossentropy": 1.7408331781625748,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1640627086162567,
+      "step": 1854
+    },
+    {
+      "epoch": 0.30916666666666665,
+      "grad_norm": 25.875,
+      "grad_norm_var": 5.166080729166667,
+      "learning_rate": 7.823394750330387e-05,
+      "loss": 6.7663,
+      "loss/crossentropy": 1.8554280549287796,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15056144446134567,
+      "step": 1855
+    },
+    {
+      "epoch": 0.30933333333333335,
+      "grad_norm": 29.125,
+      "grad_norm_var": 5.032291666666667,
+      "learning_rate": 7.821233705152371e-05,
+      "loss": 6.9016,
+      "loss/crossentropy": 1.9535363018512726,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17445489764213562,
+      "step": 1856
+    },
+    {
+      "epoch": 0.3095,
+      "grad_norm": 29.875,
+      "grad_norm_var": 4.954166666666667,
+      "learning_rate": 7.819071886517134e-05,
+      "loss": 6.95,
+      "loss/crossentropy": 2.073364347219467,
+      "loss/hidden": 3.12109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.153484046459198,
+      "step": 1857
+    },
+    {
+      "epoch": 0.30966666666666665,
+      "grad_norm": 27.375,
+      "grad_norm_var": 5.106705729166666,
+      "learning_rate": 7.816909295017352e-05,
+      "loss": 6.9537,
+      "loss/crossentropy": 1.8409361839294434,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1687594186514616,
+      "step": 1858
+    },
+    {
+      "epoch": 0.30983333333333335,
+      "grad_norm": 26.75,
+      "grad_norm_var": 5.140625,
+      "learning_rate": 7.81474593124591e-05,
+      "loss": 6.9988,
+      "loss/crossentropy": 2.390802174806595,
+      "loss/hidden": 3.06640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16115020215511322,
+      "step": 1859
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 29.875,
+      "grad_norm_var": 5.201822916666667,
+      "learning_rate": 7.812581795795907e-05,
+      "loss": 6.9665,
+      "loss/crossentropy": 2.4145980775356293,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1627429723739624,
+      "step": 1860
+    },
+    {
+      "epoch": 0.31016666666666665,
+      "grad_norm": 27.375,
+      "grad_norm_var": 4.398372395833333,
+      "learning_rate": 7.810416889260653e-05,
+      "loss": 7.1229,
+      "loss/crossentropy": 1.765182375907898,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1748659312725067,
+      "step": 1861
+    },
+    {
+      "epoch": 0.31033333333333335,
+      "grad_norm": 25.75,
+      "grad_norm_var": 4.846809895833333,
+      "learning_rate": 7.80825121223367e-05,
+      "loss": 7.2131,
+      "loss/crossentropy": 1.5575336813926697,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2208685576915741,
+      "step": 1862
+    },
+    {
+      "epoch": 0.3105,
+      "grad_norm": 27.625,
+      "grad_norm_var": 4.8712890625,
+      "learning_rate": 7.80608476530869e-05,
+      "loss": 6.7988,
+      "loss/crossentropy": 2.1647118628025055,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1658165119588375,
+      "step": 1863
+    },
+    {
+      "epoch": 0.31066666666666665,
+      "grad_norm": 27.5,
+      "grad_norm_var": 4.923372395833334,
+      "learning_rate": 7.803917549079655e-05,
+      "loss": 6.7471,
+      "loss/crossentropy": 2.0893578827381134,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1486968044191599,
+      "step": 1864
+    },
+    {
+      "epoch": 0.31083333333333335,
+      "grad_norm": 31.25,
+      "grad_norm_var": 5.387239583333334,
+      "learning_rate": 7.801749564140724e-05,
+      "loss": 6.7533,
+      "loss/crossentropy": 1.664018765091896,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1588103286921978,
+      "step": 1865
+    },
+    {
+      "epoch": 0.311,
+      "grad_norm": 29.875,
+      "grad_norm_var": 4.74140625,
+      "learning_rate": 7.799580811086258e-05,
+      "loss": 7.0641,
+      "loss/crossentropy": 2.0539465248584747,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17147555574774742,
+      "step": 1866
+    },
+    {
+      "epoch": 0.31116666666666665,
+      "grad_norm": 28.25,
+      "grad_norm_var": 4.651822916666666,
+      "learning_rate": 7.797411290510835e-05,
+      "loss": 7.2884,
+      "loss/crossentropy": 2.532195746898651,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17648249864578247,
+      "step": 1867
+    },
+    {
+      "epoch": 0.31133333333333335,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.4916666666666667,
+      "learning_rate": 7.795241003009241e-05,
+      "loss": 6.9523,
+      "loss/crossentropy": 1.7519483268260956,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15154092013835907,
+      "step": 1868
+    },
+    {
+      "epoch": 0.3115,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.5375,
+      "learning_rate": 7.793069949176473e-05,
+      "loss": 7.0951,
+      "loss/crossentropy": 2.42670214176178,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18340852856636047,
+      "step": 1869
+    },
+    {
+      "epoch": 0.31166666666666665,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.370833333333333,
+      "learning_rate": 7.790898129607738e-05,
+      "loss": 7.0135,
+      "loss/crossentropy": 1.7717819213867188,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1786283142864704,
+      "step": 1870
+    },
+    {
+      "epoch": 0.31183333333333335,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.9921223958333334,
+      "learning_rate": 7.788725544898452e-05,
+      "loss": 7.0672,
+      "loss/crossentropy": 2.011943757534027,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17356741055846214,
+      "step": 1871
+    },
+    {
+      "epoch": 0.312,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.589583333333333,
+      "learning_rate": 7.78655219564424e-05,
+      "loss": 7.0741,
+      "loss/crossentropy": 1.7151790857315063,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17846400290727615,
+      "step": 1872
+    },
+    {
+      "epoch": 0.31216666666666665,
+      "grad_norm": 28.25,
+      "grad_norm_var": 2.4567057291666665,
+      "learning_rate": 7.784378082440941e-05,
+      "loss": 7.004,
+      "loss/crossentropy": 1.5368647873401642,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15357893332839012,
+      "step": 1873
+    },
+    {
+      "epoch": 0.31233333333333335,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.4,
+      "learning_rate": 7.782203205884598e-05,
+      "loss": 7.0122,
+      "loss/crossentropy": 1.8639425337314606,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14916876703500748,
+      "step": 1874
+    },
+    {
+      "epoch": 0.3125,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.1822265625,
+      "learning_rate": 7.780027566571465e-05,
+      "loss": 7.1856,
+      "loss/crossentropy": 1.9556578546762466,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17245961725711823,
+      "step": 1875
+    },
+    {
+      "epoch": 0.31266666666666665,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.096809895833333,
+      "learning_rate": 7.777851165098012e-05,
+      "loss": 6.7324,
+      "loss/crossentropy": 2.026822805404663,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18160508200526237,
+      "step": 1876
+    },
+    {
+      "epoch": 0.31283333333333335,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.0634765625,
+      "learning_rate": 7.775674002060905e-05,
+      "loss": 6.8268,
+      "loss/crossentropy": 1.7767840176820755,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1577366553246975,
+      "step": 1877
+    },
+    {
+      "epoch": 0.313,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.5249348958333333,
+      "learning_rate": 7.773496078057028e-05,
+      "loss": 7.064,
+      "loss/crossentropy": 1.944902926683426,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.168742710724473,
+      "step": 1878
+    },
+    {
+      "epoch": 0.31316666666666665,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.4458333333333333,
+      "learning_rate": 7.771317393683471e-05,
+      "loss": 6.9466,
+      "loss/crossentropy": 1.8553851544857025,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14800599589943886,
+      "step": 1879
+    },
+    {
+      "epoch": 0.31333333333333335,
+      "grad_norm": 27.375,
+      "grad_norm_var": 1.4676432291666666,
+      "learning_rate": 7.769137949537532e-05,
+      "loss": 7.0088,
+      "loss/crossentropy": 1.9291760623455048,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1697034388780594,
+      "step": 1880
+    },
+    {
+      "epoch": 0.3135,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.1518229166666667,
+      "learning_rate": 7.766957746216721e-05,
+      "loss": 6.9248,
+      "loss/crossentropy": 1.8612181842327118,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1679140403866768,
+      "step": 1881
+    },
+    {
+      "epoch": 0.31366666666666665,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.0910807291666667,
+      "learning_rate": 7.764776784318751e-05,
+      "loss": 7.2471,
+      "loss/crossentropy": 2.191984862089157,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17184484750032425,
+      "step": 1882
+    },
+    {
+      "epoch": 0.31383333333333335,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.1254557291666667,
+      "learning_rate": 7.762595064441542e-05,
+      "loss": 6.9915,
+      "loss/crossentropy": 1.526157170534134,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1863459274172783,
+      "step": 1883
+    },
+    {
+      "epoch": 0.314,
+      "grad_norm": 26.875,
+      "grad_norm_var": 1.2885416666666667,
+      "learning_rate": 7.76041258718323e-05,
+      "loss": 6.7897,
+      "loss/crossentropy": 2.211887240409851,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18450935557484627,
+      "step": 1884
+    },
+    {
+      "epoch": 0.31416666666666665,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.1942057291666666,
+      "learning_rate": 7.758229353142152e-05,
+      "loss": 6.9123,
+      "loss/crossentropy": 1.8004077523946762,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17727744579315186,
+      "step": 1885
+    },
+    {
+      "epoch": 0.31433333333333335,
+      "grad_norm": 26.25,
+      "grad_norm_var": 1.5035807291666667,
+      "learning_rate": 7.756045362916853e-05,
+      "loss": 7.147,
+      "loss/crossentropy": 2.127860367298126,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17892053723335266,
+      "step": 1886
+    },
+    {
+      "epoch": 0.3145,
+      "grad_norm": 27.375,
+      "grad_norm_var": 1.5580729166666667,
+      "learning_rate": 7.753860617106086e-05,
+      "loss": 7.0284,
+      "loss/crossentropy": 1.8950550258159637,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16944055259227753,
+      "step": 1887
+    },
+    {
+      "epoch": 0.31466666666666665,
+      "grad_norm": 27.5,
+      "grad_norm_var": 0.8330729166666667,
+      "learning_rate": 7.751675116308812e-05,
+      "loss": 6.9634,
+      "loss/crossentropy": 2.330602705478668,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2036605067551136,
+      "step": 1888
+    },
+    {
+      "epoch": 0.31483333333333335,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.0488932291666666,
+      "learning_rate": 7.7494888611242e-05,
+      "loss": 6.9335,
+      "loss/crossentropy": 2.0412495732307434,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15381069295108318,
+      "step": 1889
+    },
+    {
+      "epoch": 0.315,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.9645182291666667,
+      "learning_rate": 7.747301852151621e-05,
+      "loss": 7.384,
+      "loss/crossentropy": 2.0527972877025604,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2628222927451134,
+      "step": 1890
+    },
+    {
+      "epoch": 0.31516666666666665,
+      "grad_norm": 46.75,
+      "grad_norm_var": 23.270833333333332,
+      "learning_rate": 7.74511408999066e-05,
+      "loss": 6.6958,
+      "loss/crossentropy": 1.98073011636734,
+      "loss/hidden": 3.1171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14221810549497604,
+      "step": 1891
+    },
+    {
+      "epoch": 0.31533333333333335,
+      "grad_norm": 35.5,
+      "grad_norm_var": 25.316080729166668,
+      "learning_rate": 7.7429255752411e-05,
+      "loss": 7.1242,
+      "loss/crossentropy": 2.39505735039711,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2730525992810726,
+      "step": 1892
+    },
+    {
+      "epoch": 0.3155,
+      "grad_norm": 34.5,
+      "grad_norm_var": 26.171875,
+      "learning_rate": 7.740736308502938e-05,
+      "loss": 6.9788,
+      "loss/crossentropy": 2.0846723914146423,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1551859974861145,
+      "step": 1893
+    },
+    {
+      "epoch": 0.31566666666666665,
+      "grad_norm": 30.125,
+      "grad_norm_var": 25.937434895833334,
+      "learning_rate": 7.738546290376373e-05,
+      "loss": 7.0442,
+      "loss/crossentropy": 2.0316117107868195,
+      "loss/hidden": 3.11328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1621059998869896,
+      "step": 1894
+    },
+    {
+      "epoch": 0.31583333333333335,
+      "grad_norm": 28.5,
+      "grad_norm_var": 25.9978515625,
+      "learning_rate": 7.736355521461811e-05,
+      "loss": 6.9462,
+      "loss/crossentropy": 2.3520907759666443,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17579379677772522,
+      "step": 1895
+    },
+    {
+      "epoch": 0.316,
+      "grad_norm": 28.0,
+      "grad_norm_var": 25.767708333333335,
+      "learning_rate": 7.734164002359863e-05,
+      "loss": 7.2404,
+      "loss/crossentropy": 2.273442804813385,
+      "loss/hidden": 3.03515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17573808506131172,
+      "step": 1896
+    },
+    {
+      "epoch": 0.31616666666666665,
+      "grad_norm": 28.375,
+      "grad_norm_var": 25.308072916666667,
+      "learning_rate": 7.731971733671346e-05,
+      "loss": 6.9272,
+      "loss/crossentropy": 1.351926103234291,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15564968809485435,
+      "step": 1897
+    },
+    {
+      "epoch": 0.31633333333333336,
+      "grad_norm": 28.5,
+      "grad_norm_var": 25.51015625,
+      "learning_rate": 7.729778715997284e-05,
+      "loss": 7.4024,
+      "loss/crossentropy": 2.081222802400589,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18974365293979645,
+      "step": 1898
+    },
+    {
+      "epoch": 0.3165,
+      "grad_norm": 28.75,
+      "grad_norm_var": 25.608072916666668,
+      "learning_rate": 7.727584949938907e-05,
+      "loss": 7.0304,
+      "loss/crossentropy": 1.9955406785011292,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17021814361214638,
+      "step": 1899
+    },
+    {
+      "epoch": 0.31666666666666665,
+      "grad_norm": 25.875,
+      "grad_norm_var": 26.14765625,
+      "learning_rate": 7.725390436097643e-05,
+      "loss": 6.8998,
+      "loss/crossentropy": 1.763851821422577,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15010467544198036,
+      "step": 1900
+    },
+    {
+      "epoch": 0.31683333333333336,
+      "grad_norm": 28.875,
+      "grad_norm_var": 26.02890625,
+      "learning_rate": 7.723195175075136e-05,
+      "loss": 7.1744,
+      "loss/crossentropy": 1.6632992774248123,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19061791524291039,
+      "step": 1901
+    },
+    {
+      "epoch": 0.317,
+      "grad_norm": 27.5,
+      "grad_norm_var": 25.43125,
+      "learning_rate": 7.720999167473227e-05,
+      "loss": 6.9959,
+      "loss/crossentropy": 2.0862348079681396,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.178208339959383,
+      "step": 1902
+    },
+    {
+      "epoch": 0.31716666666666665,
+      "grad_norm": 30.5,
+      "grad_norm_var": 24.739518229166666,
+      "learning_rate": 7.718802413893963e-05,
+      "loss": 7.058,
+      "loss/crossentropy": 1.7267999649047852,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1783318817615509,
+      "step": 1903
+    },
+    {
+      "epoch": 0.31733333333333336,
+      "grad_norm": 28.25,
+      "grad_norm_var": 24.455143229166666,
+      "learning_rate": 7.716604914939598e-05,
+      "loss": 7.0828,
+      "loss/crossentropy": 1.5275295078754425,
+      "loss/hidden": 3.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16118833236396313,
+      "step": 1904
+    },
+    {
+      "epoch": 0.3175,
+      "grad_norm": 29.375,
+      "grad_norm_var": 24.528580729166666,
+      "learning_rate": 7.714406671212589e-05,
+      "loss": 7.0106,
+      "loss/crossentropy": 1.7048605680465698,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2020227424800396,
+      "step": 1905
+    },
+    {
+      "epoch": 0.31766666666666665,
+      "grad_norm": 33.25,
+      "grad_norm_var": 24.841080729166666,
+      "learning_rate": 7.712207683315594e-05,
+      "loss": 6.9846,
+      "loss/crossentropy": 2.0892388224601746,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1827050819993019,
+      "step": 1906
+    },
+    {
+      "epoch": 0.31783333333333336,
+      "grad_norm": 28.625,
+      "grad_norm_var": 6.801041666666666,
+      "learning_rate": 7.710007951851481e-05,
+      "loss": 6.8979,
+      "loss/crossentropy": 2.1098845303058624,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19750048220157623,
+      "step": 1907
+    },
+    {
+      "epoch": 0.318,
+      "grad_norm": 28.375,
+      "grad_norm_var": 4.422330729166666,
+      "learning_rate": 7.707807477423319e-05,
+      "loss": 6.948,
+      "loss/crossentropy": 1.733251303434372,
+      "loss/hidden": 3.14453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15958409011363983,
+      "step": 1908
+    },
+    {
+      "epoch": 0.31816666666666665,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.7108723958333334,
+      "learning_rate": 7.705606260634379e-05,
+      "loss": 6.8363,
+      "loss/crossentropy": 1.514521986246109,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17583103105425835,
+      "step": 1909
+    },
+    {
+      "epoch": 0.31833333333333336,
+      "grad_norm": 27.375,
+      "grad_norm_var": 2.670768229166667,
+      "learning_rate": 7.703404302088138e-05,
+      "loss": 7.1215,
+      "loss/crossentropy": 1.9598141610622406,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15689461305737495,
+      "step": 1910
+    },
+    {
+      "epoch": 0.3185,
+      "grad_norm": 26.625,
+      "grad_norm_var": 2.904166666666667,
+      "learning_rate": 7.701201602388276e-05,
+      "loss": 6.9337,
+      "loss/crossentropy": 1.842720240354538,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14397680386900902,
+      "step": 1911
+    },
+    {
+      "epoch": 0.31866666666666665,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.9744140625,
+      "learning_rate": 7.698998162138673e-05,
+      "loss": 7.1878,
+      "loss/crossentropy": 2.3376344740390778,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1679596770554781,
+      "step": 1912
+    },
+    {
+      "epoch": 0.31883333333333336,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.9791015625,
+      "learning_rate": 7.696793981943417e-05,
+      "loss": 6.9773,
+      "loss/crossentropy": 1.621836394071579,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1444191113114357,
+      "step": 1913
+    },
+    {
+      "epoch": 0.319,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.04765625,
+      "learning_rate": 7.694589062406796e-05,
+      "loss": 7.1688,
+      "loss/crossentropy": 1.625129222869873,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17418662458658218,
+      "step": 1914
+    },
+    {
+      "epoch": 0.31916666666666665,
+      "grad_norm": 26.5,
+      "grad_norm_var": 3.33125,
+      "learning_rate": 7.692383404133301e-05,
+      "loss": 6.8318,
+      "loss/crossentropy": 1.949428677558899,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14501073956489563,
+      "step": 1915
+    },
+    {
+      "epoch": 0.31933333333333336,
+      "grad_norm": 27.375,
+      "grad_norm_var": 2.946875,
+      "learning_rate": 7.690177007727625e-05,
+      "loss": 6.9557,
+      "loss/crossentropy": 2.4070207476615906,
+      "loss/hidden": 3.08984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16160456836223602,
+      "step": 1916
+    },
+    {
+      "epoch": 0.3195,
+      "grad_norm": 28.25,
+      "grad_norm_var": 2.9478515625,
+      "learning_rate": 7.687969873794667e-05,
+      "loss": 6.844,
+      "loss/crossentropy": 1.9400338232517242,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16128266230225563,
+      "step": 1917
+    },
+    {
+      "epoch": 0.31966666666666665,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.9478515625,
+      "learning_rate": 7.685762002939523e-05,
+      "loss": 6.8573,
+      "loss/crossentropy": 1.428574338555336,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1694641150534153,
+      "step": 1918
+    },
+    {
+      "epoch": 0.31983333333333336,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.134375,
+      "learning_rate": 7.683553395767492e-05,
+      "loss": 7.0193,
+      "loss/crossentropy": 2.0802846550941467,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16207095049321651,
+      "step": 1919
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 27.875,
+      "grad_norm_var": 3.1603515625,
+      "learning_rate": 7.681344052884077e-05,
+      "loss": 6.9388,
+      "loss/crossentropy": 1.8136779963970184,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16628169268369675,
+      "step": 1920
+    },
+    {
+      "epoch": 0.32016666666666665,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.1150390625,
+      "learning_rate": 7.679133974894983e-05,
+      "loss": 7.1336,
+      "loss/crossentropy": 2.026133418083191,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24948590621352196,
+      "step": 1921
+    },
+    {
+      "epoch": 0.32033333333333336,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.5997395833333334,
+      "learning_rate": 7.676923162406115e-05,
+      "loss": 6.9633,
+      "loss/crossentropy": 1.9579945504665375,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1824035346508026,
+      "step": 1922
+    },
+    {
+      "epoch": 0.3205,
+      "grad_norm": 27.625,
+      "grad_norm_var": 1.59765625,
+      "learning_rate": 7.674711616023581e-05,
+      "loss": 7.1175,
+      "loss/crossentropy": 1.739817202091217,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14868185482919216,
+      "step": 1923
+    },
+    {
+      "epoch": 0.32066666666666666,
+      "grad_norm": 26.125,
+      "grad_norm_var": 1.825,
+      "learning_rate": 7.672499336353687e-05,
+      "loss": 6.9857,
+      "loss/crossentropy": 1.5688129663467407,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1715860553085804,
+      "step": 1924
+    },
+    {
+      "epoch": 0.32083333333333336,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.7744140625,
+      "learning_rate": 7.670286324002944e-05,
+      "loss": 6.9274,
+      "loss/crossentropy": 1.1536138653755188,
+      "loss/hidden": 3.7109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17496302723884583,
+      "step": 1925
+    },
+    {
+      "epoch": 0.321,
+      "grad_norm": 26.875,
+      "grad_norm_var": 1.8291015625,
+      "learning_rate": 7.668072579578058e-05,
+      "loss": 6.8287,
+      "loss/crossentropy": 1.8669564425945282,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15517407283186913,
+      "step": 1926
+    },
+    {
+      "epoch": 0.32116666666666666,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.7416666666666667,
+      "learning_rate": 7.665858103685944e-05,
+      "loss": 7.3803,
+      "loss/crossentropy": 2.134288877248764,
+      "loss/hidden": 3.1171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15177594125270844,
+      "step": 1927
+    },
+    {
+      "epoch": 0.32133333333333336,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.5958333333333334,
+      "learning_rate": 7.663642896933712e-05,
+      "loss": 6.9012,
+      "loss/crossentropy": 2.088742196559906,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1822258047759533,
+      "step": 1928
+    },
+    {
+      "epoch": 0.3215,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.8400390625,
+      "learning_rate": 7.66142695992867e-05,
+      "loss": 6.8775,
+      "loss/crossentropy": 1.3785552233457565,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16557445004582405,
+      "step": 1929
+    },
+    {
+      "epoch": 0.32166666666666666,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.762434895833333,
+      "learning_rate": 7.659210293278334e-05,
+      "loss": 6.9856,
+      "loss/crossentropy": 1.8797550201416016,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1681162305176258,
+      "step": 1930
+    },
+    {
+      "epoch": 0.32183333333333336,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.6134765625,
+      "learning_rate": 7.656992897590414e-05,
+      "loss": 7.3109,
+      "loss/crossentropy": 2.0877412259578705,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16942401975393295,
+      "step": 1931
+    },
+    {
+      "epoch": 0.322,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.5629557291666667,
+      "learning_rate": 7.654774773472823e-05,
+      "loss": 6.8469,
+      "loss/crossentropy": 1.7201172709465027,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.138444684445858,
+      "step": 1932
+    },
+    {
+      "epoch": 0.32216666666666666,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.65390625,
+      "learning_rate": 7.65255592153367e-05,
+      "loss": 6.8102,
+      "loss/crossentropy": 2.1534808576107025,
+      "loss/hidden": 3.0625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1507816929370165,
+      "step": 1933
+    },
+    {
+      "epoch": 0.32233333333333336,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.71640625,
+      "learning_rate": 7.650336342381269e-05,
+      "loss": 7.2222,
+      "loss/crossentropy": 1.9072502553462982,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17374767735600471,
+      "step": 1934
+    },
+    {
+      "epoch": 0.3225,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.4634765625,
+      "learning_rate": 7.648116036624126e-05,
+      "loss": 7.1072,
+      "loss/crossentropy": 1.803395539522171,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1871289536356926,
+      "step": 1935
+    },
+    {
+      "epoch": 0.32266666666666666,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.4494140625,
+      "learning_rate": 7.645895004870954e-05,
+      "loss": 6.7908,
+      "loss/crossentropy": 2.277501791715622,
+      "loss/hidden": 3.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16612669080495834,
+      "step": 1936
+    },
+    {
+      "epoch": 0.32283333333333336,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.4462890625,
+      "learning_rate": 7.643673247730658e-05,
+      "loss": 6.9223,
+      "loss/crossentropy": 1.8814960718154907,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18017802387475967,
+      "step": 1937
+    },
+    {
+      "epoch": 0.323,
+      "grad_norm": 26.0,
+      "grad_norm_var": 2.71640625,
+      "learning_rate": 7.64145076581235e-05,
+      "loss": 6.8307,
+      "loss/crossentropy": 1.9739068448543549,
+      "loss/hidden": 3.12890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1549847535789013,
+      "step": 1938
+    },
+    {
+      "epoch": 0.32316666666666666,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.77890625,
+      "learning_rate": 7.639227559725332e-05,
+      "loss": 7.067,
+      "loss/crossentropy": 1.6122899651527405,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17071114107966423,
+      "step": 1939
+    },
+    {
+      "epoch": 0.3233333333333333,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.4635416666666665,
+      "learning_rate": 7.637003630079111e-05,
+      "loss": 7.0653,
+      "loss/crossentropy": 2.1543132662773132,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20121221616864204,
+      "step": 1940
+    },
+    {
+      "epoch": 0.3235,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.488997395833333,
+      "learning_rate": 7.634778977483389e-05,
+      "loss": 6.9078,
+      "loss/crossentropy": 1.7137659192085266,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21240157261490822,
+      "step": 1941
+    },
+    {
+      "epoch": 0.32366666666666666,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.280989583333333,
+      "learning_rate": 7.632553602548065e-05,
+      "loss": 6.8976,
+      "loss/crossentropy": 2.4341808557510376,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1924213208258152,
+      "step": 1942
+    },
+    {
+      "epoch": 0.3238333333333333,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.3530598958333333,
+      "learning_rate": 7.630327505883242e-05,
+      "loss": 6.8271,
+      "loss/crossentropy": 1.8177705109119415,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17030875384807587,
+      "step": 1943
+    },
+    {
+      "epoch": 0.324,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.4309895833333335,
+      "learning_rate": 7.628100688099215e-05,
+      "loss": 7.0423,
+      "loss/crossentropy": 2.3800909519195557,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19012291356921196,
+      "step": 1944
+    },
+    {
+      "epoch": 0.32416666666666666,
+      "grad_norm": 26.75,
+      "grad_norm_var": 1.5864583333333333,
+      "learning_rate": 7.62587314980648e-05,
+      "loss": 6.9541,
+      "loss/crossentropy": 1.9023642838001251,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16839328408241272,
+      "step": 1945
+    },
+    {
+      "epoch": 0.3243333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.6343098958333333,
+      "learning_rate": 7.623644891615727e-05,
+      "loss": 6.9947,
+      "loss/crossentropy": 2.4591980576515198,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18968886137008667,
+      "step": 1946
+    },
+    {
+      "epoch": 0.3245,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.5811848958333334,
+      "learning_rate": 7.621415914137846e-05,
+      "loss": 7.0245,
+      "loss/crossentropy": 2.207227349281311,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2148870788514614,
+      "step": 1947
+    },
+    {
+      "epoch": 0.32466666666666666,
+      "grad_norm": 25.625,
+      "grad_norm_var": 2.0311848958333334,
+      "learning_rate": 7.619186217983924e-05,
+      "loss": 6.8463,
+      "loss/crossentropy": 1.7848967611789703,
+      "loss/hidden": 3.12890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19068965688347816,
+      "step": 1948
+    },
+    {
+      "epoch": 0.3248333333333333,
+      "grad_norm": 26.75,
+      "grad_norm_var": 1.99375,
+      "learning_rate": 7.616955803765249e-05,
+      "loss": 7.0644,
+      "loss/crossentropy": 2.094870924949646,
+      "loss/hidden": 3.13671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15366321429610252,
+      "step": 1949
+    },
+    {
+      "epoch": 0.325,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.99375,
+      "learning_rate": 7.614724672093296e-05,
+      "loss": 6.943,
+      "loss/crossentropy": 2.6652117371559143,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1893642321228981,
+      "step": 1950
+    },
+    {
+      "epoch": 0.32516666666666666,
+      "grad_norm": 26.75,
+      "grad_norm_var": 1.99375,
+      "learning_rate": 7.612492823579745e-05,
+      "loss": 7.0461,
+      "loss/crossentropy": 2.2999298870563507,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19316185265779495,
+      "step": 1951
+    },
+    {
+      "epoch": 0.3253333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.1395833333333334,
+      "learning_rate": 7.61026025883647e-05,
+      "loss": 6.8431,
+      "loss/crossentropy": 1.5892395228147507,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.174812950193882,
+      "step": 1952
+    },
+    {
+      "epoch": 0.3255,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.137955729166667,
+      "learning_rate": 7.60802697847554e-05,
+      "loss": 7.2168,
+      "loss/crossentropy": 1.9761822819709778,
+      "loss/hidden": 2.96484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13490652292966843,
+      "step": 1953
+    },
+    {
+      "epoch": 0.32566666666666666,
+      "grad_norm": 27.5,
+      "grad_norm_var": 1.8738932291666666,
+      "learning_rate": 7.605792983109222e-05,
+      "loss": 6.8934,
+      "loss/crossentropy": 2.2519370019435883,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.180035088211298,
+      "step": 1954
+    },
+    {
+      "epoch": 0.3258333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.9337890625,
+      "learning_rate": 7.60355827334998e-05,
+      "loss": 6.8198,
+      "loss/crossentropy": 1.5578933954238892,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19093408435583115,
+      "step": 1955
+    },
+    {
+      "epoch": 0.326,
+      "grad_norm": 50.75,
+      "grad_norm_var": 33.42265625,
+      "learning_rate": 7.60132284981047e-05,
+      "loss": 6.9115,
+      "loss/crossentropy": 2.2493423223495483,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2549340948462486,
+      "step": 1956
+    },
+    {
+      "epoch": 0.32616666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 33.47180989583333,
+      "learning_rate": 7.599086713103547e-05,
+      "loss": 7.04,
+      "loss/crossentropy": 1.8345628380775452,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19467304274439812,
+      "step": 1957
+    },
+    {
+      "epoch": 0.3263333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 33.4041015625,
+      "learning_rate": 7.596849863842263e-05,
+      "loss": 7.1351,
+      "loss/crossentropy": 2.14158171415329,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15880156680941582,
+      "step": 1958
+    },
+    {
+      "epoch": 0.3265,
+      "grad_norm": 28.0,
+      "grad_norm_var": 33.303125,
+      "learning_rate": 7.594612302639859e-05,
+      "loss": 6.9193,
+      "loss/crossentropy": 1.8796173632144928,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19231846183538437,
+      "step": 1959
+    },
+    {
+      "epoch": 0.32666666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 33.3619140625,
+      "learning_rate": 7.592374030109777e-05,
+      "loss": 6.9745,
+      "loss/crossentropy": 1.9451691210269928,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16280245408415794,
+      "step": 1960
+    },
+    {
+      "epoch": 0.3268333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 32.7666015625,
+      "learning_rate": 7.590135046865651e-05,
+      "loss": 7.049,
+      "loss/crossentropy": 1.925338327884674,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1880045346915722,
+      "step": 1961
+    },
+    {
+      "epoch": 0.327,
+      "grad_norm": 28.375,
+      "grad_norm_var": 32.91015625,
+      "learning_rate": 7.587895353521314e-05,
+      "loss": 6.9306,
+      "loss/crossentropy": 1.6036016643047333,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15349698439240456,
+      "step": 1962
+    },
+    {
+      "epoch": 0.32716666666666666,
+      "grad_norm": 26.5,
+      "grad_norm_var": 33.52265625,
+      "learning_rate": 7.585654950690786e-05,
+      "loss": 6.9653,
+      "loss/crossentropy": 1.9686013758182526,
+      "loss/hidden": 3.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17552976310253143,
+      "step": 1963
+    },
+    {
+      "epoch": 0.3273333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 32.357747395833336,
+      "learning_rate": 7.58341383898829e-05,
+      "loss": 7.0797,
+      "loss/crossentropy": 2.07719424366951,
+      "loss/hidden": 3.11328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22164364904165268,
+      "step": 1964
+    },
+    {
+      "epoch": 0.3275,
+      "grad_norm": 26.5,
+      "grad_norm_var": 32.468684895833334,
+      "learning_rate": 7.581172019028238e-05,
+      "loss": 6.9607,
+      "loss/crossentropy": 1.857707530260086,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1626106183975935,
+      "step": 1965
+    },
+    {
+      "epoch": 0.32766666666666666,
+      "grad_norm": 27.0,
+      "grad_norm_var": 33.00930989583333,
+      "learning_rate": 7.578929491425238e-05,
+      "loss": 7.0241,
+      "loss/crossentropy": 2.13506418466568,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15958191826939583,
+      "step": 1966
+    },
+    {
+      "epoch": 0.3278333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 32.38430989583333,
+      "learning_rate": 7.576686256794091e-05,
+      "loss": 7.1273,
+      "loss/crossentropy": 2.255886882543564,
+      "loss/hidden": 3.07421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1753615103662014,
+      "step": 1967
+    },
+    {
+      "epoch": 0.328,
+      "grad_norm": 29.25,
+      "grad_norm_var": 32.411458333333336,
+      "learning_rate": 7.574442315749793e-05,
+      "loss": 7.112,
+      "loss/crossentropy": 2.3582537472248077,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17308564484119415,
+      "step": 1968
+    },
+    {
+      "epoch": 0.32816666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 32.06223958333333,
+      "learning_rate": 7.572197668907532e-05,
+      "loss": 6.8533,
+      "loss/crossentropy": 1.699579805135727,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1667851246893406,
+      "step": 1969
+    },
+    {
+      "epoch": 0.3283333333333333,
+      "grad_norm": 26.125,
+      "grad_norm_var": 32.6587890625,
+      "learning_rate": 7.569952316882694e-05,
+      "loss": 7.088,
+      "loss/crossentropy": 2.3635705411434174,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1791570819914341,
+      "step": 1970
+    },
+    {
+      "epoch": 0.3285,
+      "grad_norm": 26.25,
+      "grad_norm_var": 33.55,
+      "learning_rate": 7.567706260290851e-05,
+      "loss": 6.7652,
+      "loss/crossentropy": 2.054261475801468,
+      "loss/hidden": 3.0703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1568579562008381,
+      "step": 1971
+    },
+    {
+      "epoch": 0.32866666666666666,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.4155598958333333,
+      "learning_rate": 7.565459499747775e-05,
+      "loss": 6.7638,
+      "loss/crossentropy": 1.8203125,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17247551307082176,
+      "step": 1972
+    },
+    {
+      "epoch": 0.3288333333333333,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.066666666666667,
+      "learning_rate": 7.563212035869425e-05,
+      "loss": 6.5917,
+      "loss/crossentropy": 1.909549057483673,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15480611845850945,
+      "step": 1973
+    },
+    {
+      "epoch": 0.329,
+      "grad_norm": 4630511616.0,
+      "grad_norm_var": 1.340102347874055e+18,
+      "learning_rate": 7.56096386927196e-05,
+      "loss": 8.2574,
+      "loss/crossentropy": 1.8203900903463364,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15148979425430298,
+      "step": 1974
+    },
+    {
+      "epoch": 0.32916666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.3401023477727626e+18,
+      "learning_rate": 7.558715000571726e-05,
+      "loss": 7.0224,
+      "loss/crossentropy": 2.227582097053528,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18255995586514473,
+      "step": 1975
+    },
+    {
+      "epoch": 0.3293333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.3401023476569997e+18,
+      "learning_rate": 7.55646543038526e-05,
+      "loss": 6.6633,
+      "loss/crossentropy": 2.0235438346862793,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14363868162035942,
+      "step": 1976
+    },
+    {
+      "epoch": 0.3295,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.340102347681117e+18,
+      "learning_rate": 7.5542151593293e-05,
+      "loss": 6.7019,
+      "loss/crossentropy": 1.7277594953775406,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15272897109389305,
+      "step": 1977
+    },
+    {
+      "epoch": 0.32966666666666666,
+      "grad_norm": 27.875,
+      "grad_norm_var": 1.340102347700411e+18,
+      "learning_rate": 7.551964188020766e-05,
+      "loss": 6.8864,
+      "loss/crossentropy": 2.0088874995708466,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20240258052945137,
+      "step": 1978
+    },
+    {
+      "epoch": 0.3298333333333333,
+      "grad_norm": 26.75,
+      "grad_norm_var": 1.340102347690764e+18,
+      "learning_rate": 7.549712517076777e-05,
+      "loss": 6.6294,
+      "loss/crossentropy": 2.180763453245163,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1812187097966671,
+      "step": 1979
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.3401023477197046e+18,
+      "learning_rate": 7.547460147114641e-05,
+      "loss": 6.9854,
+      "loss/crossentropy": 1.9047144949436188,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1809837929904461,
+      "step": 1980
+    },
+    {
+      "epoch": 0.33016666666666666,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.3401023476184123e+18,
+      "learning_rate": 7.545207078751857e-05,
+      "loss": 6.9115,
+      "loss/crossentropy": 1.8136049807071686,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17625956237316132,
+      "step": 1981
+    },
+    {
+      "epoch": 0.3303333333333333,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.3401023475508838e+18,
+      "learning_rate": 7.542953312606117e-05,
+      "loss": 6.8522,
+      "loss/crossentropy": 1.8081963956356049,
+      "loss/hidden": 3.1171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19056766852736473,
+      "step": 1982
+    },
+    {
+      "epoch": 0.3305,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.3401023475894715e+18,
+      "learning_rate": 7.540698849295305e-05,
+      "loss": 7.0558,
+      "loss/crossentropy": 1.7935562431812286,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18458115682005882,
+      "step": 1983
+    },
+    {
+      "epoch": 0.33066666666666666,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.3401023476135887e+18,
+      "learning_rate": 7.538443689437492e-05,
+      "loss": 6.7505,
+      "loss/crossentropy": 1.2718239575624466,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14389248006045818,
+      "step": 1984
+    },
+    {
+      "epoch": 0.3308333333333333,
+      "grad_norm": 27.5,
+      "grad_norm_var": 1.3401023477100577e+18,
+      "learning_rate": 7.536187833650947e-05,
+      "loss": 6.8596,
+      "loss/crossentropy": 1.7777176648378372,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14321670681238174,
+      "step": 1985
+    },
+    {
+      "epoch": 0.331,
+      "grad_norm": 27.875,
+      "grad_norm_var": 1.3401023476425295e+18,
+      "learning_rate": 7.53393128255412e-05,
+      "loss": 6.9501,
+      "loss/crossentropy": 2.031786859035492,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1821003295481205,
+      "step": 1986
+    },
+    {
+      "epoch": 0.33116666666666666,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.3401023475219433e+18,
+      "learning_rate": 7.531674036765662e-05,
+      "loss": 7.1937,
+      "loss/crossentropy": 2.1827369332313538,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20355921983718872,
+      "step": 1987
+    },
+    {
+      "epoch": 0.3313333333333333,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.3401023474833556e+18,
+      "learning_rate": 7.52941609690441e-05,
+      "loss": 7.0974,
+      "loss/crossentropy": 1.684326320886612,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14068151265382767,
+      "step": 1988
+    },
+    {
+      "epoch": 0.3315,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.3401023474688852e+18,
+      "learning_rate": 7.52715746358939e-05,
+      "loss": 6.8646,
+      "loss/crossentropy": 1.736901193857193,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1760329306125641,
+      "step": 1989
+    },
+    {
+      "epoch": 0.33166666666666667,
+      "grad_norm": 27.5,
+      "grad_norm_var": 1.5729166666666667,
+      "learning_rate": 7.524898137439814e-05,
+      "loss": 6.7771,
+      "loss/crossentropy": 2.0175153017044067,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18924446776509285,
+      "step": 1990
+    },
+    {
+      "epoch": 0.3318333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.3080729166666667,
+      "learning_rate": 7.522638119075096e-05,
+      "loss": 6.9701,
+      "loss/crossentropy": 1.6442890465259552,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1708950586616993,
+      "step": 1991
+    },
+    {
+      "epoch": 0.332,
+      "grad_norm": 28.125,
+      "grad_norm_var": 0.5229166666666667,
+      "learning_rate": 7.520377409114831e-05,
+      "loss": 6.9788,
+      "loss/crossentropy": 2.2745184898376465,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17724668234586716,
+      "step": 1992
+    },
+    {
+      "epoch": 0.33216666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 0.5926432291666667,
+      "learning_rate": 7.518116008178805e-05,
+      "loss": 6.9768,
+      "loss/crossentropy": 1.7873744368553162,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26323091611266136,
+      "step": 1993
+    },
+    {
+      "epoch": 0.3323333333333333,
+      "grad_norm": 28.25,
+      "grad_norm_var": 0.5830729166666667,
+      "learning_rate": 7.515853916886993e-05,
+      "loss": 6.7853,
+      "loss/crossentropy": 1.8601644933223724,
+      "loss/hidden": 3.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1265878528356552,
+      "step": 1994
+    },
+    {
+      "epoch": 0.3325,
+      "grad_norm": 28.125,
+      "grad_norm_var": 0.42337239583333336,
+      "learning_rate": 7.513591135859561e-05,
+      "loss": 6.8199,
+      "loss/crossentropy": 1.8973293006420135,
+      "loss/hidden": 3.09765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15631063655018806,
+      "step": 1995
+    },
+    {
+      "epoch": 0.33266666666666667,
+      "grad_norm": 28.125,
+      "grad_norm_var": 0.42473958333333334,
+      "learning_rate": 7.511327665716863e-05,
+      "loss": 6.8403,
+      "loss/crossentropy": 1.9851139187812805,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16789129376411438,
+      "step": 1996
+    },
+    {
+      "epoch": 0.3328333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 0.4552083333333333,
+      "learning_rate": 7.509063507079443e-05,
+      "loss": 6.9476,
+      "loss/crossentropy": 1.6235378086566925,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1800040602684021,
+      "step": 1997
+    },
+    {
+      "epoch": 0.333,
+      "grad_norm": 28.625,
+      "grad_norm_var": 0.4494140625,
+      "learning_rate": 7.506798660568031e-05,
+      "loss": 6.9203,
+      "loss/crossentropy": 2.3260062634944916,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20887180417776108,
+      "step": 1998
+    },
+    {
+      "epoch": 0.33316666666666667,
+      "grad_norm": 27.375,
+      "grad_norm_var": 0.40182291666666664,
+      "learning_rate": 7.50453312680355e-05,
+      "loss": 6.9222,
+      "loss/crossentropy": 1.9456888735294342,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16610940545797348,
+      "step": 1999
+    },
+    {
+      "epoch": 0.3333333333333333,
+      "grad_norm": 25.875,
+      "grad_norm_var": 0.7197916666666667,
+      "learning_rate": 7.502266906407107e-05,
+      "loss": 6.87,
+      "loss/crossentropy": 1.652198612689972,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1459857076406479,
+      "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
+}