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": 47.0,
+      "learning_rate": 0.0001,
+      "loss": 8.0388,
+      "loss/crossentropy": 2.015764206647873,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27190398424863815,
+      "step": 1
+    },
+    {
+      "epoch": 0.0003333333333333333,
+      "grad_norm": 51.75,
+      "learning_rate": 9.999999314610822e-05,
+      "loss": 7.1282,
+      "loss/crossentropy": 1.6513436883687973,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21722547337412834,
+      "step": 2
+    },
+    {
+      "epoch": 0.0005,
+      "grad_norm": 43.75,
+      "learning_rate": 9.999997258443473e-05,
+      "loss": 7.8038,
+      "loss/crossentropy": 1.228823497891426,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2009564395993948,
+      "step": 3
+    },
+    {
+      "epoch": 0.0006666666666666666,
+      "grad_norm": 40.75,
+      "learning_rate": 9.999993831498517e-05,
+      "loss": 7.4582,
+      "loss/crossentropy": 2.0953528583049774,
+      "loss/hidden": 3.6875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27634887397289276,
+      "step": 4
+    },
+    {
+      "epoch": 0.0008333333333333334,
+      "grad_norm": 34.5,
+      "learning_rate": 9.999989033776898e-05,
+      "loss": 7.3349,
+      "loss/crossentropy": 1.712344229221344,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21917030215263367,
+      "step": 5
+    },
+    {
+      "epoch": 0.001,
+      "grad_norm": 31.75,
+      "learning_rate": 9.999982865279924e-05,
+      "loss": 7.5036,
+      "loss/crossentropy": 1.8789039552211761,
+      "loss/hidden": 3.76953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20935339480638504,
+      "step": 6
+    },
+    {
+      "epoch": 0.0011666666666666668,
+      "grad_norm": 33.25,
+      "learning_rate": 9.999975326009292e-05,
+      "loss": 7.1155,
+      "loss/crossentropy": 1.37015450745821,
+      "loss/hidden": 3.14453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13620511814951897,
+      "step": 7
+    },
+    {
+      "epoch": 0.0013333333333333333,
+      "grad_norm": 35.0,
+      "learning_rate": 9.999966415967066e-05,
+      "loss": 7.2857,
+      "loss/crossentropy": 1.668631225824356,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21737387031316757,
+      "step": 8
+    },
+    {
+      "epoch": 0.0015,
+      "grad_norm": 35.5,
+      "learning_rate": 9.999956135155687e-05,
+      "loss": 7.1543,
+      "loss/crossentropy": 1.65235635638237,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1770026832818985,
+      "step": 9
+    },
+    {
+      "epoch": 0.0016666666666666668,
+      "grad_norm": 31.0,
+      "learning_rate": 9.999944483577981e-05,
+      "loss": 7.1422,
+      "loss/crossentropy": 1.598803088068962,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17498954012989998,
+      "step": 10
+    },
+    {
+      "epoch": 0.0018333333333333333,
+      "grad_norm": 33.75,
+      "learning_rate": 9.999931461237134e-05,
+      "loss": 7.6142,
+      "loss/crossentropy": 1.65041284263134,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23264329880475998,
+      "step": 11
+    },
+    {
+      "epoch": 0.002,
+      "grad_norm": 38.5,
+      "learning_rate": 9.999917068136722e-05,
+      "loss": 7.426,
+      "loss/crossentropy": 1.4104643166065216,
+      "loss/hidden": 3.765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21118134632706642,
+      "step": 12
+    },
+    {
+      "epoch": 0.0021666666666666666,
+      "grad_norm": 40.25,
+      "learning_rate": 9.999901304280685e-05,
+      "loss": 7.9928,
+      "loss/crossentropy": 1.6555309295654297,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16892194002866745,
+      "step": 13
+    },
+    {
+      "epoch": 0.0023333333333333335,
+      "grad_norm": 37.0,
+      "learning_rate": 9.999884169673351e-05,
+      "loss": 7.2367,
+      "loss/crossentropy": 1.5136004090309143,
+      "loss/hidden": 3.71484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21831507235765457,
+      "step": 14
+    },
+    {
+      "epoch": 0.0025,
+      "grad_norm": 33.75,
+      "learning_rate": 9.999865664319414e-05,
+      "loss": 7.6944,
+      "loss/crossentropy": 1.755567491054535,
+      "loss/hidden": 3.89453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2622127905488014,
+      "step": 15
+    },
+    {
+      "epoch": 0.0026666666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 35.131184895833336,
+      "learning_rate": 9.999845788223949e-05,
+      "loss": 7.4481,
+      "loss/crossentropy": 1.3894422799348831,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17372214049100876,
+      "step": 16
+    },
+    {
+      "epoch": 0.0028333333333333335,
+      "grad_norm": 33.75,
+      "grad_norm_var": 29.223893229166666,
+      "learning_rate": 9.999824541392405e-05,
+      "loss": 7.1243,
+      "loss/crossentropy": 1.823076069355011,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2317724972963333,
+      "step": 17
+    },
+    {
+      "epoch": 0.003,
+      "grad_norm": 38.5,
+      "grad_norm_var": 13.4619140625,
+      "learning_rate": 9.999801923830603e-05,
+      "loss": 7.5944,
+      "loss/crossentropy": 1.8854615688323975,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2611945904791355,
+      "step": 18
+    },
+    {
+      "epoch": 0.0031666666666666666,
+      "grad_norm": 35.75,
+      "grad_norm_var": 8.970247395833333,
+      "learning_rate": 9.99977793554475e-05,
+      "loss": 7.7183,
+      "loss/crossentropy": 1.9868004322052002,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22432727739214897,
+      "step": 19
+    },
+    {
+      "epoch": 0.0033333333333333335,
+      "grad_norm": 33.75,
+      "grad_norm_var": 6.935872395833333,
+      "learning_rate": 9.999752576541418e-05,
+      "loss": 7.4676,
+      "loss/crossentropy": 1.3248331844806671,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20089071616530418,
+      "step": 20
+    },
+    {
+      "epoch": 0.0035,
+      "grad_norm": 34.5,
+      "grad_norm_var": 6.935872395833333,
+      "learning_rate": 9.999725846827562e-05,
+      "loss": 7.5965,
+      "loss/crossentropy": 1.3043542802333832,
+      "loss/hidden": 3.81640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2816320173442364,
+      "step": 21
+    },
+    {
+      "epoch": 0.0036666666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 6.456184895833333,
+      "learning_rate": 9.999697746410508e-05,
+      "loss": 7.5651,
+      "loss/crossentropy": 1.4966092258691788,
+      "loss/hidden": 3.72265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19401345402002335,
+      "step": 22
+    },
+    {
+      "epoch": 0.003833333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 6.403580729166666,
+      "learning_rate": 9.99966827529796e-05,
+      "loss": 7.7272,
+      "loss/crossentropy": 1.5755852162837982,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1464239191263914,
+      "step": 23
+    },
+    {
+      "epoch": 0.004,
+      "grad_norm": 34.0,
+      "grad_norm_var": 6.460872395833333,
+      "learning_rate": 9.999637433497999e-05,
+      "loss": 7.5594,
+      "loss/crossentropy": 1.77267724275589,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26836544647812843,
+      "step": 24
+    },
+    {
+      "epoch": 0.004166666666666667,
+      "grad_norm": 36.75,
+      "grad_norm_var": 6.6587890625,
+      "learning_rate": 9.999605221019081e-05,
+      "loss": 7.9483,
+      "loss/crossentropy": 1.4049716889858246,
+      "loss/hidden": 3.89453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2349778600037098,
+      "step": 25
+    },
+    {
+      "epoch": 0.004333333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 6.242708333333334,
+      "learning_rate": 9.999571637870036e-05,
+      "loss": 7.0353,
+      "loss/crossentropy": 1.6401013135910034,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18000634759664536,
+      "step": 26
+    },
+    {
+      "epoch": 0.0045,
+      "grad_norm": 34.75,
+      "grad_norm_var": 6.134375,
+      "learning_rate": 9.99953668406007e-05,
+      "loss": 7.5181,
+      "loss/crossentropy": 1.5710090696811676,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11938946694135666,
+      "step": 27
+    },
+    {
+      "epoch": 0.004666666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 5.387239583333334,
+      "learning_rate": 9.999500359598768e-05,
+      "loss": 7.174,
+      "loss/crossentropy": 1.7123600542545319,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16798048838973045,
+      "step": 28
+    },
+    {
+      "epoch": 0.004833333333333334,
+      "grad_norm": 35.5,
+      "grad_norm_var": 3.34375,
+      "learning_rate": 9.999462664496088e-05,
+      "loss": 7.4485,
+      "loss/crossentropy": 1.2491168677806854,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17139602452516556,
+      "step": 29
+    },
+    {
+      "epoch": 0.005,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.1705729166666665,
+      "learning_rate": 9.999423598762363e-05,
+      "loss": 7.0512,
+      "loss/crossentropy": 1.4913183897733688,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1743941754102707,
+      "step": 30
+    },
+    {
+      "epoch": 0.005166666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.34375,
+      "learning_rate": 9.999383162408304e-05,
+      "loss": 7.3525,
+      "loss/crossentropy": 2.0380469262599945,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20196463912725449,
+      "step": 31
+    },
+    {
+      "epoch": 0.005333333333333333,
+      "grad_norm": 37.75,
+      "grad_norm_var": 3.6468098958333335,
+      "learning_rate": 9.999341355444995e-05,
+      "loss": 7.477,
+      "loss/crossentropy": 1.167970821261406,
+      "loss/hidden": 3.90625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26106464490294456,
+      "step": 32
+    },
+    {
+      "epoch": 0.0055,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.6254557291666667,
+      "learning_rate": 9.999298177883903e-05,
+      "loss": 7.4047,
+      "loss/crossentropy": 1.9607312381267548,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2703733518719673,
+      "step": 33
+    },
+    {
+      "epoch": 0.005666666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.5171223958333333,
+      "learning_rate": 9.99925362973686e-05,
+      "loss": 7.4611,
+      "loss/crossentropy": 2.214526116847992,
+      "loss/hidden": 3.87109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3339185602962971,
+      "step": 34
+    },
+    {
+      "epoch": 0.005833333333333334,
+      "grad_norm": 34.0,
+      "grad_norm_var": 2.3530598958333333,
+      "learning_rate": 9.999207711016081e-05,
+      "loss": 7.4829,
+      "loss/crossentropy": 1.9158975481987,
+      "loss/hidden": 3.73828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21212821826338768,
+      "step": 35
+    },
+    {
+      "epoch": 0.006,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.3530598958333333,
+      "learning_rate": 9.999160421734155e-05,
+      "loss": 6.8363,
+      "loss/crossentropy": 1.2238537073135376,
+      "loss/hidden": 3.7578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20537541806697845,
+      "step": 36
+    },
+    {
+      "epoch": 0.006166666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.326041666666667,
+      "learning_rate": 9.999111761904046e-05,
+      "loss": 6.5735,
+      "loss/crossentropy": 1.386010155081749,
+      "loss/hidden": 3.796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21487441286444664,
+      "step": 37
+    },
+    {
+      "epoch": 0.006333333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.52265625,
+      "learning_rate": 9.999061731539094e-05,
+      "loss": 7.0498,
+      "loss/crossentropy": 1.6234411001205444,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26318273693323135,
+      "step": 38
+    },
+    {
+      "epoch": 0.0065,
+      "grad_norm": 33.75,
+      "grad_norm_var": 3.517708333333333,
+      "learning_rate": 9.999010330653018e-05,
+      "loss": 7.2467,
+      "loss/crossentropy": 1.775725930929184,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.228952556848526,
+      "step": 39
+    },
+    {
+      "epoch": 0.006666666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.6145833333333335,
+      "learning_rate": 9.998957559259906e-05,
+      "loss": 7.3362,
+      "loss/crossentropy": 1.3429348766803741,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1380001101642847,
+      "step": 40
+    },
+    {
+      "epoch": 0.006833333333333334,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.1093098958333334,
+      "learning_rate": 9.998903417374228e-05,
+      "loss": 7.4727,
+      "loss/crossentropy": 1.2927208542823792,
+      "loss/hidden": 3.765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22973712906241417,
+      "step": 41
+    },
+    {
+      "epoch": 0.007,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.042708333333333,
+      "learning_rate": 9.998847905010826e-05,
+      "loss": 7.3356,
+      "loss/crossentropy": 1.642817735671997,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22616977244615555,
+      "step": 42
+    },
+    {
+      "epoch": 0.007166666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.934375,
+      "learning_rate": 9.998791022184922e-05,
+      "loss": 7.2349,
+      "loss/crossentropy": 1.7549102306365967,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2533755637705326,
+      "step": 43
+    },
+    {
+      "epoch": 0.007333333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.95390625,
+      "learning_rate": 9.998732768912104e-05,
+      "loss": 6.9294,
+      "loss/crossentropy": 2.2454869747161865,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.164547361433506,
+      "step": 44
+    },
+    {
+      "epoch": 0.0075,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.59765625,
+      "learning_rate": 9.99867314520835e-05,
+      "loss": 7.2884,
+      "loss/crossentropy": 1.0857951641082764,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15111834183335304,
+      "step": 45
+    },
+    {
+      "epoch": 0.007666666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.6447265625,
+      "learning_rate": 9.998612151090003e-05,
+      "loss": 7.4896,
+      "loss/crossentropy": 2.3741764277219772,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21800962835550308,
+      "step": 46
+    },
+    {
+      "epoch": 0.007833333333333333,
+      "grad_norm": 35.5,
+      "grad_norm_var": 3.0103515625,
+      "learning_rate": 9.998549786573785e-05,
+      "loss": 7.8114,
+      "loss/crossentropy": 1.4853083044290543,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1981479413807392,
+      "step": 47
+    },
+    {
+      "epoch": 0.008,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.5494140625,
+      "learning_rate": 9.998486051676792e-05,
+      "loss": 7.6955,
+      "loss/crossentropy": 1.8257912993431091,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25801796093583107,
+      "step": 48
+    },
+    {
+      "epoch": 0.008166666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.4572265625,
+      "learning_rate": 9.9984209464165e-05,
+      "loss": 7.5406,
+      "loss/crossentropy": 1.6004481613636017,
+      "loss/hidden": 3.9765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2681163512170315,
+      "step": 49
+    },
+    {
+      "epoch": 0.008333333333333333,
+      "grad_norm": 34.5,
+      "grad_norm_var": 1.5931640625,
+      "learning_rate": 9.998354470810757e-05,
+      "loss": 7.4694,
+      "loss/crossentropy": 1.1797468811273575,
+      "loss/hidden": 3.79296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18932894617319107,
+      "step": 50
+    },
+    {
+      "epoch": 0.0085,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.76015625,
+      "learning_rate": 9.998286624877786e-05,
+      "loss": 7.1043,
+      "loss/crossentropy": 1.4719633758068085,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32244824431836605,
+      "step": 51
+    },
+    {
+      "epoch": 0.008666666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.66640625,
+      "learning_rate": 9.99821740863619e-05,
+      "loss": 7.4195,
+      "loss/crossentropy": 1.6572435796260834,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16776087693870068,
+      "step": 52
+    },
+    {
+      "epoch": 0.008833333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.2619140625,
+      "learning_rate": 9.998146822104943e-05,
+      "loss": 7.0209,
+      "loss/crossentropy": 1.6052251160144806,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16937347874045372,
+      "step": 53
+    },
+    {
+      "epoch": 0.009,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.3872395833333333,
+      "learning_rate": 9.998074865303399e-05,
+      "loss": 7.2891,
+      "loss/crossentropy": 1.534130334854126,
+      "loss/hidden": 3.73046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23601197451353073,
+      "step": 54
+    },
+    {
+      "epoch": 0.009166666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.3083333333333333,
+      "learning_rate": 9.998001538251282e-05,
+      "loss": 7.3219,
+      "loss/crossentropy": 1.6903752535581589,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18934278190135956,
+      "step": 55
+    },
+    {
+      "epoch": 0.009333333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.3143229166666666,
+      "learning_rate": 9.997926840968699e-05,
+      "loss": 7.7928,
+      "loss/crossentropy": 1.417355865240097,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28693941608071327,
+      "step": 56
+    },
+    {
+      "epoch": 0.0095,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.4405598958333334,
+      "learning_rate": 9.997850773476126e-05,
+      "loss": 6.941,
+      "loss/crossentropy": 1.1929711401462555,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18017736449837685,
+      "step": 57
+    },
+    {
+      "epoch": 0.009666666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.9559895833333334,
+      "learning_rate": 9.997773335794416e-05,
+      "loss": 7.2535,
+      "loss/crossentropy": 1.5090252310037613,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18393898010253906,
+      "step": 58
+    },
+    {
+      "epoch": 0.009833333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.174934895833333,
+      "learning_rate": 9.997694527944803e-05,
+      "loss": 6.7798,
+      "loss/crossentropy": 2.142109751701355,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2558054178953171,
+      "step": 59
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.2119140625,
+      "learning_rate": 9.99761434994889e-05,
+      "loss": 6.9992,
+      "loss/crossentropy": 1.6562436819076538,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15309249237179756,
+      "step": 60
+    },
+    {
+      "epoch": 0.010166666666666666,
+      "grad_norm": 41.0,
+      "grad_norm_var": 7.055143229166666,
+      "learning_rate": 9.997532801828658e-05,
+      "loss": 7.7495,
+      "loss/crossentropy": 1.5080684423446655,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32041720300912857,
+      "step": 61
+    },
+    {
+      "epoch": 0.010333333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 7.058072916666666,
+      "learning_rate": 9.997449883606466e-05,
+      "loss": 7.1658,
+      "loss/crossentropy": 0.9691251963376999,
+      "loss/hidden": 3.7734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2008093148469925,
+      "step": 62
+    },
+    {
+      "epoch": 0.0105,
+      "grad_norm": 29.625,
+      "grad_norm_var": 7.146809895833333,
+      "learning_rate": 9.997365595305044e-05,
+      "loss": 6.8092,
+      "loss/crossentropy": 1.3262115865945816,
+      "loss/hidden": 4.01171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24134666472673416,
+      "step": 63
+    },
+    {
+      "epoch": 0.010666666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 7.364322916666667,
+      "learning_rate": 9.997279936947502e-05,
+      "loss": 7.1277,
+      "loss/crossentropy": 1.9158867001533508,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17067686840891838,
+      "step": 64
+    },
+    {
+      "epoch": 0.010833333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 7.3994140625,
+      "learning_rate": 9.997192908557323e-05,
+      "loss": 7.4407,
+      "loss/crossentropy": 1.9335598796606064,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17702890932559967,
+      "step": 65
+    },
+    {
+      "epoch": 0.011,
+      "grad_norm": 32.0,
+      "grad_norm_var": 7.068684895833333,
+      "learning_rate": 9.997104510158365e-05,
+      "loss": 7.3374,
+      "loss/crossentropy": 1.7730918526649475,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1596100777387619,
+      "step": 66
+    },
+    {
+      "epoch": 0.011166666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 7.259309895833334,
+      "learning_rate": 9.997014741774866e-05,
+      "loss": 6.7962,
+      "loss/crossentropy": 1.5432041361927986,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18355126678943634,
+      "step": 67
+    },
+    {
+      "epoch": 0.011333333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 7.2837890625,
+      "learning_rate": 9.996923603431433e-05,
+      "loss": 6.9029,
+      "loss/crossentropy": 1.232265830039978,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24395740032196045,
+      "step": 68
+    },
+    {
+      "epoch": 0.0115,
+      "grad_norm": 32.5,
+      "grad_norm_var": 7.2837890625,
+      "learning_rate": 9.996831095153055e-05,
+      "loss": 7.5835,
+      "loss/crossentropy": 1.7480098903179169,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18551664799451828,
+      "step": 69
+    },
+    {
+      "epoch": 0.011666666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 7.552018229166666,
+      "learning_rate": 9.996737216965092e-05,
+      "loss": 7.2713,
+      "loss/crossentropy": 1.4004269242286682,
+      "loss/hidden": 3.80078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24054613709449768,
+      "step": 70
+    },
+    {
+      "epoch": 0.011833333333333333,
+      "grad_norm": 37.0,
+      "grad_norm_var": 9.0478515625,
+      "learning_rate": 9.996641968893282e-05,
+      "loss": 6.8689,
+      "loss/crossentropy": 1.4935747683048248,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2075977362692356,
+      "step": 71
+    },
+    {
+      "epoch": 0.012,
+      "grad_norm": 31.25,
+      "grad_norm_var": 9.1197265625,
+      "learning_rate": 9.996545350963738e-05,
+      "loss": 7.2183,
+      "loss/crossentropy": 1.565061777830124,
+      "loss/hidden": 3.875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35492730885744095,
+      "step": 72
+    },
+    {
+      "epoch": 0.012166666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 9.007747395833333,
+      "learning_rate": 9.996447363202946e-05,
+      "loss": 7.3531,
+      "loss/crossentropy": 1.3612003773450851,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23692565225064754,
+      "step": 73
+    },
+    {
+      "epoch": 0.012333333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 8.672916666666667,
+      "learning_rate": 9.996348005637775e-05,
+      "loss": 7.3786,
+      "loss/crossentropy": 1.858590006828308,
+      "loss/hidden": 3.7734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31772397831082344,
+      "step": 74
+    },
+    {
+      "epoch": 0.0125,
+      "grad_norm": 39.0,
+      "grad_norm_var": 10.734830729166667,
+      "learning_rate": 9.996247278295458e-05,
+      "loss": 7.6853,
+      "loss/crossentropy": 1.517339512705803,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1719112992286682,
+      "step": 75
+    },
+    {
+      "epoch": 0.012666666666666666,
+      "grad_norm": 31.125,
+      "grad_norm_var": 10.995572916666667,
+      "learning_rate": 9.996145181203615e-05,
+      "loss": 7.5477,
+      "loss/crossentropy": 1.8657438457012177,
+      "loss/hidden": 3.765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2289566993713379,
+      "step": 76
+    },
+    {
+      "epoch": 0.012833333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 6.547330729166666,
+      "learning_rate": 9.996041714390235e-05,
+      "loss": 6.8349,
+      "loss/crossentropy": 1.3593356609344482,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2103591337800026,
+      "step": 77
+    },
+    {
+      "epoch": 0.013,
+      "grad_norm": 37.0,
+      "grad_norm_var": 7.766080729166666,
+      "learning_rate": 9.995936877883682e-05,
+      "loss": 7.7018,
+      "loss/crossentropy": 1.642098844051361,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19433317705988884,
+      "step": 78
+    },
+    {
+      "epoch": 0.013166666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 7.22890625,
+      "learning_rate": 9.9958306717127e-05,
+      "loss": 7.4576,
+      "loss/crossentropy": 1.7163181602954865,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2721835747361183,
+      "step": 79
+    },
+    {
+      "epoch": 0.013333333333333334,
+      "grad_norm": 31.875,
+      "grad_norm_var": 6.938541666666667,
+      "learning_rate": 9.995723095906407e-05,
+      "loss": 7.4261,
+      "loss/crossentropy": 1.9470502138137817,
+      "loss/hidden": 3.8046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23109716176986694,
+      "step": 80
+    },
+    {
+      "epoch": 0.0135,
+      "grad_norm": 34.0,
+      "grad_norm_var": 6.845768229166667,
+      "learning_rate": 9.995614150494293e-05,
+      "loss": 7.6891,
+      "loss/crossentropy": 1.7489202320575714,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1897631213068962,
+      "step": 81
+    },
+    {
+      "epoch": 0.013666666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 6.750455729166666,
+      "learning_rate": 9.995503835506226e-05,
+      "loss": 6.7672,
+      "loss/crossentropy": 1.3573525249958038,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15716634131968021,
+      "step": 82
+    },
+    {
+      "epoch": 0.013833333333333333,
+      "grad_norm": 36.0,
+      "grad_norm_var": 6.319791666666666,
+      "learning_rate": 9.995392150972451e-05,
+      "loss": 7.4197,
+      "loss/crossentropy": 2.055204689502716,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2153688222169876,
+      "step": 83
+    },
+    {
+      "epoch": 0.014,
+      "grad_norm": 32.25,
+      "grad_norm_var": 6.420572916666667,
+      "learning_rate": 9.995279096923585e-05,
+      "loss": 7.6211,
+      "loss/crossentropy": 1.6128068566322327,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18876289576292038,
+      "step": 84
+    },
+    {
+      "epoch": 0.014166666666666666,
+      "grad_norm": 30.75,
+      "grad_norm_var": 6.870833333333334,
+      "learning_rate": 9.995164673390625e-05,
+      "loss": 7.4225,
+      "loss/crossentropy": 1.9184208512306213,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22452763468027115,
+      "step": 85
+    },
+    {
+      "epoch": 0.014333333333333333,
+      "grad_norm": 35.25,
+      "grad_norm_var": 6.078580729166666,
+      "learning_rate": 9.995048880404938e-05,
+      "loss": 7.4224,
+      "loss/crossentropy": 1.450558990240097,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.267075689509511,
+      "step": 86
+    },
+    {
+      "epoch": 0.0145,
+      "grad_norm": 31.75,
+      "grad_norm_var": 5.586393229166666,
+      "learning_rate": 9.994931717998272e-05,
+      "loss": 7.3244,
+      "loss/crossentropy": 1.2475939691066742,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1947660744190216,
+      "step": 87
+    },
+    {
+      "epoch": 0.014666666666666666,
+      "grad_norm": 34.5,
+      "grad_norm_var": 5.2681640625,
+      "learning_rate": 9.994813186202747e-05,
+      "loss": 7.2037,
+      "loss/crossentropy": 1.3766074776649475,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19298407062888145,
+      "step": 88
+    },
+    {
+      "epoch": 0.014833333333333334,
+      "grad_norm": 33.5,
+      "grad_norm_var": 5.122330729166666,
+      "learning_rate": 9.994693285050857e-05,
+      "loss": 7.1216,
+      "loss/crossentropy": 1.3637863248586655,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23826205730438232,
+      "step": 89
+    },
+    {
+      "epoch": 0.015,
+      "grad_norm": 30.625,
+      "grad_norm_var": 5.720833333333333,
+      "learning_rate": 9.994572014575476e-05,
+      "loss": 7.0287,
+      "loss/crossentropy": 1.558560699224472,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17919514700770378,
+      "step": 90
+    },
+    {
+      "epoch": 0.015166666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 4.117122395833333,
+      "learning_rate": 9.994449374809851e-05,
+      "loss": 7.1082,
+      "loss/crossentropy": 1.8318926692008972,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23970575630664825,
+      "step": 91
+    },
+    {
+      "epoch": 0.015333333333333332,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.220833333333333,
+      "learning_rate": 9.994325365787602e-05,
+      "loss": 7.329,
+      "loss/crossentropy": 1.851843684911728,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3288569562137127,
+      "step": 92
+    },
+    {
+      "epoch": 0.0155,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.160872395833334,
+      "learning_rate": 9.99419998754273e-05,
+      "loss": 7.4068,
+      "loss/crossentropy": 1.5664864778518677,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22041853331029415,
+      "step": 93
+    },
+    {
+      "epoch": 0.015666666666666666,
+      "grad_norm": 29.25,
+      "grad_norm_var": 3.8056640625,
+      "learning_rate": 9.994073240109606e-05,
+      "loss": 7.1767,
+      "loss/crossentropy": 1.4907843619585037,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18625176325440407,
+      "step": 94
+    },
+    {
+      "epoch": 0.015833333333333335,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.6509765625,
+      "learning_rate": 9.993945123522978e-05,
+      "loss": 6.9561,
+      "loss/crossentropy": 1.8172418773174286,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20110055431723595,
+      "step": 95
+    },
+    {
+      "epoch": 0.016,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.7791015625,
+      "learning_rate": 9.993815637817974e-05,
+      "loss": 7.2981,
+      "loss/crossentropy": 1.1591381877660751,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18586143851280212,
+      "step": 96
+    },
+    {
+      "epoch": 0.016166666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.5942057291666667,
+      "learning_rate": 9.993684783030088e-05,
+      "loss": 6.9345,
+      "loss/crossentropy": 1.2907428294420242,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21381350234150887,
+      "step": 97
+    },
+    {
+      "epoch": 0.01633333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.4869140625,
+      "learning_rate": 9.993552559195197e-05,
+      "loss": 7.3921,
+      "loss/crossentropy": 2.438232123851776,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2648901715874672,
+      "step": 98
+    },
+    {
+      "epoch": 0.0165,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.5244140625,
+      "learning_rate": 9.993418966349552e-05,
+      "loss": 6.9331,
+      "loss/crossentropy": 1.7527403831481934,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18184861540794373,
+      "step": 99
+    },
+    {
+      "epoch": 0.016666666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.6421223958333333,
+      "learning_rate": 9.993284004529775e-05,
+      "loss": 7.4561,
+      "loss/crossentropy": 1.867360770702362,
+      "loss/hidden": 3.76953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27160242944955826,
+      "step": 100
+    },
+    {
+      "epoch": 0.016833333333333332,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.5530598958333335,
+      "learning_rate": 9.99314767377287e-05,
+      "loss": 7.1804,
+      "loss/crossentropy": 1.718344509601593,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19866780564188957,
+      "step": 101
+    },
+    {
+      "epoch": 0.017,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.323958333333333,
+      "learning_rate": 9.993009974116211e-05,
+      "loss": 7.3187,
+      "loss/crossentropy": 1.3677956759929657,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15728242695331573,
+      "step": 102
+    },
+    {
+      "epoch": 0.017166666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.324934895833333,
+      "learning_rate": 9.992870905597548e-05,
+      "loss": 7.3042,
+      "loss/crossentropy": 1.3308251798152924,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16750549711287022,
+      "step": 103
+    },
+    {
+      "epoch": 0.017333333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.7233723958333333,
+      "learning_rate": 9.992730468255011e-05,
+      "loss": 7.2264,
+      "loss/crossentropy": 1.7275401055812836,
+      "loss/hidden": 3.734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29135602340102196,
+      "step": 104
+    },
+    {
+      "epoch": 0.0175,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.5082682291666667,
+      "learning_rate": 9.9925886621271e-05,
+      "loss": 7.4005,
+      "loss/crossentropy": 1.7864578068256378,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1873011216521263,
+      "step": 105
+    },
+    {
+      "epoch": 0.017666666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.728125,
+      "learning_rate": 9.992445487252691e-05,
+      "loss": 6.9595,
+      "loss/crossentropy": 1.5800124257802963,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2642953395843506,
+      "step": 106
+    },
+    {
+      "epoch": 0.017833333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.69140625,
+      "learning_rate": 9.992300943671036e-05,
+      "loss": 7.284,
+      "loss/crossentropy": 1.1614376306533813,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15753909014165401,
+      "step": 107
+    },
+    {
+      "epoch": 0.018,
+      "grad_norm": 35.25,
+      "grad_norm_var": 2.72265625,
+      "learning_rate": 9.992155031421764e-05,
+      "loss": 6.7385,
+      "loss/crossentropy": 2.0193272531032562,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28752196952700615,
+      "step": 108
+    },
+    {
+      "epoch": 0.018166666666666668,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.841080729166667,
+      "learning_rate": 9.992007750544876e-05,
+      "loss": 6.6446,
+      "loss/crossentropy": 1.359722450375557,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1365048959851265,
+      "step": 109
+    },
+    {
+      "epoch": 0.018333333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.700455729166667,
+      "learning_rate": 9.991859101080751e-05,
+      "loss": 7.7226,
+      "loss/crossentropy": 1.7012291550636292,
+      "loss/hidden": 3.73046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26156413927674294,
+      "step": 110
+    },
+    {
+      "epoch": 0.0185,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.9893229166666666,
+      "learning_rate": 9.991709083070143e-05,
+      "loss": 6.908,
+      "loss/crossentropy": 1.6251331716775894,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1416161134839058,
+      "step": 111
+    },
+    {
+      "epoch": 0.018666666666666668,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.178580729166667,
+      "learning_rate": 9.991557696554177e-05,
+      "loss": 7.8488,
+      "loss/crossentropy": 2.0877310037612915,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2414073385298252,
+      "step": 112
+    },
+    {
+      "epoch": 0.018833333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.0874348958333333,
+      "learning_rate": 9.991404941574361e-05,
+      "loss": 7.1683,
+      "loss/crossentropy": 2.0482693016529083,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2420409545302391,
+      "step": 113
+    },
+    {
+      "epoch": 0.019,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.1270182291666666,
+      "learning_rate": 9.99125081817257e-05,
+      "loss": 7.1673,
+      "loss/crossentropy": 1.921759843826294,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2487247809767723,
+      "step": 114
+    },
+    {
+      "epoch": 0.019166666666666665,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.101822916666667,
+      "learning_rate": 9.99109532639106e-05,
+      "loss": 6.7798,
+      "loss/crossentropy": 1.3495987802743912,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14604422822594643,
+      "step": 115
+    },
+    {
+      "epoch": 0.019333333333333334,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.2113932291666667,
+      "learning_rate": 9.990938466272459e-05,
+      "loss": 7.4022,
+      "loss/crossentropy": 1.8897645175457,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22202493250370026,
+      "step": 116
+    },
+    {
+      "epoch": 0.0195,
+      "grad_norm": 38.75,
+      "grad_norm_var": 6.7119140625,
+      "learning_rate": 9.990780237859769e-05,
+      "loss": 7.2288,
+      "loss/crossentropy": 1.8467685282230377,
+      "loss/hidden": 3.765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19912605360150337,
+      "step": 117
+    },
+    {
+      "epoch": 0.019666666666666666,
+      "grad_norm": 35.25,
+      "grad_norm_var": 6.88125,
+      "learning_rate": 9.990620641196374e-05,
+      "loss": 7.5258,
+      "loss/crossentropy": 1.4544800072908401,
+      "loss/hidden": 3.7578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23149508237838745,
+      "step": 118
+    },
+    {
+      "epoch": 0.019833333333333335,
+      "grad_norm": 36.0,
+      "grad_norm_var": 7.794205729166666,
+      "learning_rate": 9.990459676326024e-05,
+      "loss": 7.3993,
+      "loss/crossentropy": 1.7979627847671509,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2008170560002327,
+      "step": 119
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 33.75,
+      "grad_norm_var": 7.8837890625,
+      "learning_rate": 9.990297343292851e-05,
+      "loss": 7.448,
+      "loss/crossentropy": 1.582246333360672,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22383124008774757,
+      "step": 120
+    },
+    {
+      "epoch": 0.020166666666666666,
+      "grad_norm": 30.75,
+      "grad_norm_var": 7.6712890625,
+      "learning_rate": 9.990133642141359e-05,
+      "loss": 7.1318,
+      "loss/crossentropy": 2.2869274020195007,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26430132053792477,
+      "step": 121
+    },
+    {
+      "epoch": 0.02033333333333333,
+      "grad_norm": 35.0,
+      "grad_norm_var": 7.228059895833334,
+      "learning_rate": 9.989968572916426e-05,
+      "loss": 7.3843,
+      "loss/crossentropy": 1.4673973321914673,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19540218636393547,
+      "step": 122
+    },
+    {
+      "epoch": 0.0205,
+      "grad_norm": 28.375,
+      "grad_norm_var": 8.345247395833333,
+      "learning_rate": 9.989802135663308e-05,
+      "loss": 6.881,
+      "loss/crossentropy": 1.4048103988170624,
+      "loss/hidden": 3.7265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17774714902043343,
+      "step": 123
+    },
+    {
+      "epoch": 0.020666666666666667,
+      "grad_norm": 28.625,
+      "grad_norm_var": 8.845572916666667,
+      "learning_rate": 9.989634330427636e-05,
+      "loss": 7.2466,
+      "loss/crossentropy": 1.162289284169674,
+      "loss/hidden": 3.09765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19122084602713585,
+      "step": 124
+    },
+    {
+      "epoch": 0.020833333333333332,
+      "grad_norm": 31.5,
+      "grad_norm_var": 8.485872395833333,
+      "learning_rate": 9.989465157255412e-05,
+      "loss": 7.2053,
+      "loss/crossentropy": 1.312704622745514,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15034876391291618,
+      "step": 125
+    },
+    {
+      "epoch": 0.021,
+      "grad_norm": 30.625,
+      "grad_norm_var": 8.647916666666667,
+      "learning_rate": 9.989294616193017e-05,
+      "loss": 6.8005,
+      "loss/crossentropy": 1.6380910277366638,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14181393012404442,
+      "step": 126
+    },
+    {
+      "epoch": 0.021166666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 8.079622395833333,
+      "learning_rate": 9.989122707287208e-05,
+      "loss": 7.3312,
+      "loss/crossentropy": 1.211222618818283,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2158495895564556,
+      "step": 127
+    },
+    {
+      "epoch": 0.021333333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 8.187239583333334,
+      "learning_rate": 9.988949430585111e-05,
+      "loss": 7.3088,
+      "loss/crossentropy": 1.9512027502059937,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2397935502231121,
+      "step": 128
+    },
+    {
+      "epoch": 0.0215,
+      "grad_norm": 29.875,
+      "grad_norm_var": 8.465559895833334,
+      "learning_rate": 9.988774786134234e-05,
+      "loss": 7.0164,
+      "loss/crossentropy": 2.2886485755443573,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17705616354942322,
+      "step": 129
+    },
+    {
+      "epoch": 0.021666666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 8.831184895833333,
+      "learning_rate": 9.988598773982454e-05,
+      "loss": 6.7626,
+      "loss/crossentropy": 1.6010637283325195,
+      "loss/hidden": 3.1171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1609804928302765,
+      "step": 130
+    },
+    {
+      "epoch": 0.021833333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 8.859830729166667,
+      "learning_rate": 9.988421394178027e-05,
+      "loss": 7.1795,
+      "loss/crossentropy": 2.1255241334438324,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2915850803256035,
+      "step": 131
+    },
+    {
+      "epoch": 0.022,
+      "grad_norm": 30.5,
+      "grad_norm_var": 8.44375,
+      "learning_rate": 9.988242646769584e-05,
+      "loss": 6.6464,
+      "loss/crossentropy": 1.4530896842479706,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1587641667574644,
+      "step": 132
+    },
+    {
+      "epoch": 0.022166666666666668,
+      "grad_norm": 35.0,
+      "grad_norm_var": 6.01015625,
+      "learning_rate": 9.988062531806126e-05,
+      "loss": 6.8361,
+      "loss/crossentropy": 1.58548042178154,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1621798612177372,
+      "step": 133
+    },
+    {
+      "epoch": 0.022333333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 5.246809895833334,
+      "learning_rate": 9.987881049337037e-05,
+      "loss": 7.2599,
+      "loss/crossentropy": 1.462182879447937,
+      "loss/hidden": 3.7734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27100462280213833,
+      "step": 134
+    },
+    {
+      "epoch": 0.0225,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.8811848958333335,
+      "learning_rate": 9.98769819941207e-05,
+      "loss": 7.1722,
+      "loss/crossentropy": 1.54609115421772,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19119622558355331,
+      "step": 135
+    },
+    {
+      "epoch": 0.02266666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.535872395833333,
+      "learning_rate": 9.987513982081351e-05,
+      "loss": 7.3784,
+      "loss/crossentropy": 1.6093524098396301,
+      "loss/hidden": 3.9921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27235616743564606,
+      "step": 136
+    },
+    {
+      "epoch": 0.022833333333333334,
+      "grad_norm": 28.125,
+      "grad_norm_var": 4.133333333333334,
+      "learning_rate": 9.987328397395387e-05,
+      "loss": 7.2047,
+      "loss/crossentropy": 1.6545589417219162,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15520754642784595,
+      "step": 137
+    },
+    {
+      "epoch": 0.023,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.0327473958333333,
+      "learning_rate": 9.98714144540506e-05,
+      "loss": 7.157,
+      "loss/crossentropy": 1.9186067283153534,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15664204582571983,
+      "step": 138
+    },
+    {
+      "epoch": 0.023166666666666665,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.630989583333333,
+      "learning_rate": 9.986953126161619e-05,
+      "loss": 6.6498,
+      "loss/crossentropy": 1.324135184288025,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16996346972882748,
+      "step": 139
+    },
+    {
+      "epoch": 0.023333333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.284309895833333,
+      "learning_rate": 9.986763439716696e-05,
+      "loss": 6.7863,
+      "loss/crossentropy": 1.7032744586467743,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1454637162387371,
+      "step": 140
+    },
+    {
+      "epoch": 0.0235,
+      "grad_norm": 37.5,
+      "grad_norm_var": 4.840559895833334,
+      "learning_rate": 9.986572386122291e-05,
+      "loss": 7.2084,
+      "loss/crossentropy": 1.7331673800945282,
+      "loss/hidden": 3.71484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24686722084879875,
+      "step": 141
+    },
+    {
+      "epoch": 0.023666666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.817708333333333,
+      "learning_rate": 9.986379965430786e-05,
+      "loss": 7.4599,
+      "loss/crossentropy": 2.2102459371089935,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.247200645506382,
+      "step": 142
+    },
+    {
+      "epoch": 0.023833333333333335,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.959309895833333,
+      "learning_rate": 9.986186177694933e-05,
+      "loss": 6.8554,
+      "loss/crossentropy": 1.3829247057437897,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3295183889567852,
+      "step": 143
+    },
+    {
+      "epoch": 0.024,
+      "grad_norm": 29.125,
+      "grad_norm_var": 5.265559895833333,
+      "learning_rate": 9.98599102296786e-05,
+      "loss": 6.6174,
+      "loss/crossentropy": 1.4767974317073822,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17495453730225563,
+      "step": 144
+    },
+    {
+      "epoch": 0.024166666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 5.188997395833334,
+      "learning_rate": 9.98579450130307e-05,
+      "loss": 7.0007,
+      "loss/crossentropy": 1.502312645316124,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1338094538077712,
+      "step": 145
+    },
+    {
+      "epoch": 0.024333333333333332,
+      "grad_norm": 37.0,
+      "grad_norm_var": 6.986393229166667,
+      "learning_rate": 9.985596612754439e-05,
+      "loss": 7.2911,
+      "loss/crossentropy": 1.6600433588027954,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1484263464808464,
+      "step": 146
+    },
+    {
+      "epoch": 0.0245,
+      "grad_norm": 31.625,
+      "grad_norm_var": 6.978059895833334,
+      "learning_rate": 9.985397357376222e-05,
+      "loss": 6.6697,
+      "loss/crossentropy": 1.9513159692287445,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2093215435743332,
+      "step": 147
+    },
+    {
+      "epoch": 0.024666666666666667,
+      "grad_norm": 40.75,
+      "grad_norm_var": 11.825455729166666,
+      "learning_rate": 9.985196735223045e-05,
+      "loss": 7.3884,
+      "loss/crossentropy": 1.5966178327798843,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15934613719582558,
+      "step": 148
+    },
+    {
+      "epoch": 0.024833333333333332,
+      "grad_norm": 34.5,
+      "grad_norm_var": 11.667643229166666,
+      "learning_rate": 9.98499474634991e-05,
+      "loss": 7.223,
+      "loss/crossentropy": 2.1022060215473175,
+      "loss/hidden": 3.71484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25909359380602837,
+      "step": 149
+    },
+    {
+      "epoch": 0.025,
+      "grad_norm": 31.5,
+      "grad_norm_var": 11.567708333333334,
+      "learning_rate": 9.98479139081219e-05,
+      "loss": 7.5453,
+      "loss/crossentropy": 1.3491213917732239,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31083543226122856,
+      "step": 150
+    },
+    {
+      "epoch": 0.025166666666666667,
+      "grad_norm": 31.25,
+      "grad_norm_var": 11.567708333333334,
+      "learning_rate": 9.98458666866564e-05,
+      "loss": 6.9526,
+      "loss/crossentropy": 1.1305021345615387,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17413905262947083,
+      "step": 151
+    },
+    {
+      "epoch": 0.025333333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 11.609375,
+      "learning_rate": 9.984380579966385e-05,
+      "loss": 7.5569,
+      "loss/crossentropy": 1.7270284742116928,
+      "loss/hidden": 3.80859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.262040700763464,
+      "step": 152
+    },
+    {
+      "epoch": 0.0255,
+      "grad_norm": 30.25,
+      "grad_norm_var": 10.660872395833334,
+      "learning_rate": 9.984173124770923e-05,
+      "loss": 7.0024,
+      "loss/crossentropy": 1.4194546043872833,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18989887833595276,
+      "step": 153
+    },
+    {
+      "epoch": 0.025666666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 10.62890625,
+      "learning_rate": 9.983964303136133e-05,
+      "loss": 6.827,
+      "loss/crossentropy": 1.8736361861228943,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19809295609593391,
+      "step": 154
+    },
+    {
+      "epoch": 0.025833333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 10.31640625,
+      "learning_rate": 9.983754115119261e-05,
+      "loss": 6.8501,
+      "loss/crossentropy": 0.9456725493073463,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1407306743785739,
+      "step": 155
+    },
+    {
+      "epoch": 0.026,
+      "grad_norm": 32.25,
+      "grad_norm_var": 10.26640625,
+      "learning_rate": 9.983542560777935e-05,
+      "loss": 7.1745,
+      "loss/crossentropy": 1.7430013120174408,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18162920884788036,
+      "step": 156
+    },
+    {
+      "epoch": 0.026166666666666668,
+      "grad_norm": 30.375,
+      "grad_norm_var": 8.9416015625,
+      "learning_rate": 9.983329640170149e-05,
+      "loss": 7.198,
+      "loss/crossentropy": 1.6644318103790283,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21564067527651787,
+      "step": 157
+    },
+    {
+      "epoch": 0.026333333333333334,
+      "grad_norm": 27.625,
+      "grad_norm_var": 10.321875,
+      "learning_rate": 9.983115353354281e-05,
+      "loss": 6.6243,
+      "loss/crossentropy": 1.250006690621376,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18062720447778702,
+      "step": 158
+    },
+    {
+      "epoch": 0.0265,
+      "grad_norm": 30.625,
+      "grad_norm_var": 9.976822916666666,
+      "learning_rate": 9.982899700389076e-05,
+      "loss": 6.7883,
+      "loss/crossentropy": 1.4420564621686935,
+      "loss/hidden": 3.8203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2596291061490774,
+      "step": 159
+    },
+    {
+      "epoch": 0.02666666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 9.398893229166667,
+      "learning_rate": 9.982682681333658e-05,
+      "loss": 6.6638,
+      "loss/crossentropy": 1.3785683065652847,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1723865158855915,
+      "step": 160
+    },
+    {
+      "epoch": 0.026833333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 9.120572916666667,
+      "learning_rate": 9.982464296247522e-05,
+      "loss": 7.3516,
+      "loss/crossentropy": 1.4394225776195526,
+      "loss/hidden": 3.734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2615921422839165,
+      "step": 161
+    },
+    {
+      "epoch": 0.027,
+      "grad_norm": 33.5,
+      "grad_norm_var": 7.77890625,
+      "learning_rate": 9.982244545190542e-05,
+      "loss": 7.1089,
+      "loss/crossentropy": 1.884726345539093,
+      "loss/hidden": 3.6875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25164157152175903,
+      "step": 162
+    },
+    {
+      "epoch": 0.027166666666666665,
+      "grad_norm": 33.5,
+      "grad_norm_var": 7.8384765625,
+      "learning_rate": 9.982023428222962e-05,
+      "loss": 7.3006,
+      "loss/crossentropy": 1.3327789902687073,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16302406042814255,
+      "step": 163
+    },
+    {
+      "epoch": 0.027333333333333334,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.8806640625,
+      "learning_rate": 9.981800945405403e-05,
+      "loss": 6.8786,
+      "loss/crossentropy": 2.1047302186489105,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20406359992921352,
+      "step": 164
+    },
+    {
+      "epoch": 0.0275,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.3916666666666666,
+      "learning_rate": 9.981577096798863e-05,
+      "loss": 7.1575,
+      "loss/crossentropy": 1.828908532857895,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2953005023300648,
+      "step": 165
+    },
+    {
+      "epoch": 0.027666666666666666,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.403580729166667,
+      "learning_rate": 9.981351882464706e-05,
+      "loss": 6.7743,
+      "loss/crossentropy": 1.1883389204740524,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12561755441129208,
+      "step": 166
+    },
+    {
+      "epoch": 0.027833333333333335,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.4098307291666665,
+      "learning_rate": 9.98112530246468e-05,
+      "loss": 7.0895,
+      "loss/crossentropy": 1.4930688589811325,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22280864231288433,
+      "step": 167
+    },
+    {
+      "epoch": 0.028,
+      "grad_norm": 27.75,
+      "grad_norm_var": 3.0801432291666666,
+      "learning_rate": 9.980897356860901e-05,
+      "loss": 6.9969,
+      "loss/crossentropy": 1.5407499372959137,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2660684399306774,
+      "step": 168
+    },
+    {
+      "epoch": 0.028166666666666666,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.419205729166667,
+      "learning_rate": 9.980668045715864e-05,
+      "loss": 6.4818,
+      "loss/crossentropy": 1.2468992918729782,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1492381915450096,
+      "step": 169
+    },
+    {
+      "epoch": 0.028333333333333332,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.4801432291666665,
+      "learning_rate": 9.980437369092431e-05,
+      "loss": 6.8676,
+      "loss/crossentropy": 1.8643461614847183,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16300494596362114,
+      "step": 170
+    },
+    {
+      "epoch": 0.0285,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.4119140625,
+      "learning_rate": 9.980205327053848e-05,
+      "loss": 7.237,
+      "loss/crossentropy": 1.6697624623775482,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19493362307548523,
+      "step": 171
+    },
+    {
+      "epoch": 0.028666666666666667,
+      "grad_norm": 35.75,
+      "grad_norm_var": 4.662434895833333,
+      "learning_rate": 9.97997191966373e-05,
+      "loss": 7.4244,
+      "loss/crossentropy": 1.3050570487976074,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16732595954090357,
+      "step": 172
+    },
+    {
+      "epoch": 0.028833333333333332,
+      "grad_norm": 32.75,
+      "grad_norm_var": 4.680989583333333,
+      "learning_rate": 9.979737146986064e-05,
+      "loss": 6.862,
+      "loss/crossentropy": 1.136594980955124,
+      "loss/hidden": 4.0234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20343116112053394,
+      "step": 173
+    },
+    {
+      "epoch": 0.029,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.628125,
+      "learning_rate": 9.979501009085219e-05,
+      "loss": 7.2069,
+      "loss/crossentropy": 1.4369015246629715,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19320333749055862,
+      "step": 174
+    },
+    {
+      "epoch": 0.029166666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.6947265625,
+      "learning_rate": 9.979263506025929e-05,
+      "loss": 6.7702,
+      "loss/crossentropy": 1.0700709372758865,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17891627363860607,
+      "step": 175
+    },
+    {
+      "epoch": 0.029333333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 4.21015625,
+      "learning_rate": 9.97902463787331e-05,
+      "loss": 6.3209,
+      "loss/crossentropy": 1.5019266307353973,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13693124055862427,
+      "step": 176
+    },
+    {
+      "epoch": 0.0295,
+      "grad_norm": 30.5,
+      "grad_norm_var": 4.215625,
+      "learning_rate": 9.978784404692847e-05,
+      "loss": 7.0097,
+      "loss/crossentropy": 1.0866649374365807,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10230135731399059,
+      "step": 177
+    },
+    {
+      "epoch": 0.029666666666666668,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.94140625,
+      "learning_rate": 9.978542806550402e-05,
+      "loss": 7.419,
+      "loss/crossentropy": 1.6656216979026794,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16881580278277397,
+      "step": 178
+    },
+    {
+      "epoch": 0.029833333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.5551432291666667,
+      "learning_rate": 9.97829984351221e-05,
+      "loss": 7.406,
+      "loss/crossentropy": 1.6889117062091827,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2585541270673275,
+      "step": 179
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 8086618112.0,
+      "grad_norm_var": 4.087086999255681e+18,
+      "learning_rate": 9.978055515644882e-05,
+      "loss": 7.2784,
+      "loss/crossentropy": 2.095020294189453,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1817887146025896,
+      "step": 180
+    },
+    {
+      "epoch": 0.030166666666666668,
+      "grad_norm": 44.5,
+      "grad_norm_var": 4.087086998337513e+18,
+      "learning_rate": 9.977809823015401e-05,
+      "loss": 6.9852,
+      "loss/crossentropy": 1.773234412074089,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2178725991398096,
+      "step": 181
+    },
+    {
+      "epoch": 0.030333333333333334,
+      "grad_norm": 36.0,
+      "grad_norm_var": 4.0870869980089943e+18,
+      "learning_rate": 9.977562765691124e-05,
+      "loss": 6.8961,
+      "loss/crossentropy": 1.3541785031557083,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18293461948633194,
+      "step": 182
+    },
+    {
+      "epoch": 0.0305,
+      "grad_norm": 29.625,
+      "grad_norm_var": 4.087086998169042e+18,
+      "learning_rate": 9.977314343739786e-05,
+      "loss": 7.052,
+      "loss/crossentropy": 1.3558832257986069,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20253374427556992,
+      "step": 183
+    },
+    {
+      "epoch": 0.030666666666666665,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.087086997916335e+18,
+      "learning_rate": 9.977064557229492e-05,
+      "loss": 7.2972,
+      "loss/crossentropy": 1.0495585799217224,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1994080375880003,
+      "step": 184
+    },
+    {
+      "epoch": 0.030833333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 4.08708699773944e+18,
+      "learning_rate": 9.97681340622872e-05,
+      "loss": 7.1021,
+      "loss/crossentropy": 1.8053418397903442,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17343341559171677,
+      "step": 185
+    },
+    {
+      "epoch": 0.031,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.0870869978405233e+18,
+      "learning_rate": 9.976560890806328e-05,
+      "loss": 7.177,
+      "loss/crossentropy": 1.5566141307353973,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15112582594156265,
+      "step": 186
+    },
+    {
+      "epoch": 0.031166666666666665,
+      "grad_norm": 30.125,
+      "grad_norm_var": 4.0870869979668767e+18,
+      "learning_rate": 9.976307011031542e-05,
+      "loss": 6.7397,
+      "loss/crossentropy": 1.2723989188671112,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16342337802052498,
+      "step": 187
+    },
+    {
+      "epoch": 0.03133333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.087086998185889e+18,
+      "learning_rate": 9.976051766973966e-05,
+      "loss": 7.3915,
+      "loss/crossentropy": 1.6031639650464058,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17763563804328442,
+      "step": 188
+    },
+    {
+      "epoch": 0.0315,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.0870869981521946e+18,
+      "learning_rate": 9.975795158703576e-05,
+      "loss": 7.0979,
+      "loss/crossentropy": 1.4858632981777191,
+      "loss/hidden": 3.76171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.330683471634984,
+      "step": 189
+    },
+    {
+      "epoch": 0.03166666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.0870869981606185e+18,
+      "learning_rate": 9.975537186290724e-05,
+      "loss": 7.3654,
+      "loss/crossentropy": 2.4191945791244507,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25662462785840034,
+      "step": 190
+    },
+    {
+      "epoch": 0.03183333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 4.0870869981016535e+18,
+      "learning_rate": 9.975277849806133e-05,
+      "loss": 7.2515,
+      "loss/crossentropy": 1.9716623425483704,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19087167084217072,
+      "step": 191
+    },
+    {
+      "epoch": 0.032,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.08708699785737e+18,
+      "learning_rate": 9.9750171493209e-05,
+      "loss": 7.4281,
+      "loss/crossentropy": 1.6307190656661987,
+      "loss/hidden": 3.71484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23531628027558327,
+      "step": 192
+    },
+    {
+      "epoch": 0.03216666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 4.0870869978152525e+18,
+      "learning_rate": 9.974755084906502e-05,
+      "loss": 7.03,
+      "loss/crossentropy": 1.4380917251110077,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3211356829851866,
+      "step": 193
+    },
+    {
+      "epoch": 0.03233333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.087086997629934e+18,
+      "learning_rate": 9.974491656634782e-05,
+      "loss": 7.3389,
+      "loss/crossentropy": 1.0064150243997574,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1723968144506216,
+      "step": 194
+    },
+    {
+      "epoch": 0.0325,
+      "grad_norm": 32.75,
+      "grad_norm_var": 4.087086997537275e+18,
+      "learning_rate": 9.974226864577961e-05,
+      "loss": 7.5133,
+      "loss/crossentropy": 1.7098364979028702,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17002902552485466,
+      "step": 195
+    },
+    {
+      "epoch": 0.03266666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 12.689322916666667,
+      "learning_rate": 9.973960708808633e-05,
+      "loss": 7.0961,
+      "loss/crossentropy": 1.2855606228113174,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14486568979918957,
+      "step": 196
+    },
+    {
+      "epoch": 0.03283333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.542708333333333,
+      "learning_rate": 9.973693189399766e-05,
+      "loss": 7.7159,
+      "loss/crossentropy": 1.4339561015367508,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15074146911501884,
+      "step": 197
+    },
+    {
+      "epoch": 0.033,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.4968098958333333,
+      "learning_rate": 9.973424306424705e-05,
+      "loss": 6.8725,
+      "loss/crossentropy": 1.3229663521051407,
+      "loss/hidden": 3.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11913192272186279,
+      "step": 198
+    },
+    {
+      "epoch": 0.033166666666666664,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.2809895833333333,
+      "learning_rate": 9.973154059957162e-05,
+      "loss": 7.2338,
+      "loss/crossentropy": 1.5000391602516174,
+      "loss/hidden": 3.83203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24410316348075867,
+      "step": 199
+    },
+    {
+      "epoch": 0.03333333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.38125,
+      "learning_rate": 9.972882450071228e-05,
+      "loss": 6.9197,
+      "loss/crossentropy": 1.185929924249649,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19635562598705292,
+      "step": 200
+    },
+    {
+      "epoch": 0.0335,
+      "grad_norm": 33.5,
+      "grad_norm_var": 1.6457682291666667,
+      "learning_rate": 9.972609476841367e-05,
+      "loss": 7.1692,
+      "loss/crossentropy": 1.2175997495651245,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21033136174082756,
+      "step": 201
+    },
+    {
+      "epoch": 0.033666666666666664,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.6223307291666667,
+      "learning_rate": 9.972335140342415e-05,
+      "loss": 7.4232,
+      "loss/crossentropy": 2.04010409116745,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.202102143317461,
+      "step": 202
+    },
+    {
+      "epoch": 0.03383333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.5989583333333333,
+      "learning_rate": 9.972059440649584e-05,
+      "loss": 7.0424,
+      "loss/crossentropy": 1.7747844010591507,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25096169114112854,
+      "step": 203
+    },
+    {
+      "epoch": 0.034,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.0247395833333335,
+      "learning_rate": 9.971782377838457e-05,
+      "loss": 6.8701,
+      "loss/crossentropy": 1.1719542741775513,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14155150577425957,
+      "step": 204
+    },
+    {
+      "epoch": 0.034166666666666665,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.9143229166666667,
+      "learning_rate": 9.971503951984995e-05,
+      "loss": 7.2028,
+      "loss/crossentropy": 1.5867617726325989,
+      "loss/hidden": 3.83984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2683488056063652,
+      "step": 205
+    },
+    {
+      "epoch": 0.034333333333333334,
+      "grad_norm": 35.25,
+      "grad_norm_var": 2.8330729166666666,
+      "learning_rate": 9.971224163165527e-05,
+      "loss": 7.1905,
+      "loss/crossentropy": 1.7028155624866486,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3061375916004181,
+      "step": 206
+    },
+    {
+      "epoch": 0.0345,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.9166666666666665,
+      "learning_rate": 9.970943011456761e-05,
+      "loss": 6.8989,
+      "loss/crossentropy": 2.1334219872951508,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2289668172597885,
+      "step": 207
+    },
+    {
+      "epoch": 0.034666666666666665,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.9614583333333333,
+      "learning_rate": 9.970660496935776e-05,
+      "loss": 7.4457,
+      "loss/crossentropy": 1.2610541135072708,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.224246047437191,
+      "step": 208
+    },
+    {
+      "epoch": 0.034833333333333334,
+      "grad_norm": 40.25,
+      "grad_norm_var": 7.595247395833334,
+      "learning_rate": 9.970376619680024e-05,
+      "loss": 6.7979,
+      "loss/crossentropy": 1.1224466264247894,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16827253252267838,
+      "step": 209
+    },
+    {
+      "epoch": 0.035,
+      "grad_norm": 34.75,
+      "grad_norm_var": 7.981705729166666,
+      "learning_rate": 9.970091379767331e-05,
+      "loss": 6.8035,
+      "loss/crossentropy": 1.7064603418111801,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20265027321875095,
+      "step": 210
+    },
+    {
+      "epoch": 0.035166666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 8.2134765625,
+      "learning_rate": 9.9698047772759e-05,
+      "loss": 6.6525,
+      "loss/crossentropy": 1.3764468729496002,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18533694557845592,
+      "step": 211
+    },
+    {
+      "epoch": 0.035333333333333335,
+      "grad_norm": 30.25,
+      "grad_norm_var": 8.11015625,
+      "learning_rate": 9.969516812284301e-05,
+      "loss": 7.0358,
+      "loss/crossentropy": 1.9165741205215454,
+      "loss/hidden": 3.74609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31283821910619736,
+      "step": 212
+    },
+    {
+      "epoch": 0.0355,
+      "grad_norm": 32.0,
+      "grad_norm_var": 8.084375,
+      "learning_rate": 9.969227484871484e-05,
+      "loss": 7.4682,
+      "loss/crossentropy": 1.740638554096222,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19709094986319542,
+      "step": 213
+    },
+    {
+      "epoch": 0.035666666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 8.314518229166667,
+      "learning_rate": 9.968936795116768e-05,
+      "loss": 6.6945,
+      "loss/crossentropy": 0.8976159617304802,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.08759790286421776,
+      "step": 214
+    },
+    {
+      "epoch": 0.035833333333333335,
+      "grad_norm": 36.5,
+      "grad_norm_var": 9.531705729166667,
+      "learning_rate": 9.968644743099848e-05,
+      "loss": 7.0007,
+      "loss/crossentropy": 1.8110921084880829,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19646160677075386,
+      "step": 215
+    },
+    {
+      "epoch": 0.036,
+      "grad_norm": 30.75,
+      "grad_norm_var": 9.4072265625,
+      "learning_rate": 9.968351328900794e-05,
+      "loss": 6.5056,
+      "loss/crossentropy": 0.9803343713283539,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11365807242691517,
+      "step": 216
+    },
+    {
+      "epoch": 0.036166666666666666,
+      "grad_norm": 30.75,
+      "grad_norm_var": 9.470247395833333,
+      "learning_rate": 9.968056552600043e-05,
+      "loss": 6.9747,
+      "loss/crossentropy": 1.8104292750358582,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20365064218640327,
+      "step": 217
+    },
+    {
+      "epoch": 0.036333333333333336,
+      "grad_norm": 30.25,
+      "grad_norm_var": 9.626497395833333,
+      "learning_rate": 9.967760414278411e-05,
+      "loss": 7.2355,
+      "loss/crossentropy": 1.7058001458644867,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22070244327187538,
+      "step": 218
+    },
+    {
+      "epoch": 0.0365,
+      "grad_norm": 30.0,
+      "grad_norm_var": 9.694205729166667,
+      "learning_rate": 9.967462914017088e-05,
+      "loss": 7.0123,
+      "loss/crossentropy": 1.2381923496723175,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16867449507117271,
+      "step": 219
+    },
+    {
+      "epoch": 0.03666666666666667,
+      "grad_norm": 43.5,
+      "grad_norm_var": 16.608268229166665,
+      "learning_rate": 9.967164051897633e-05,
+      "loss": 7.5328,
+      "loss/crossentropy": 1.5826676338911057,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2712370380759239,
+      "step": 220
+    },
+    {
+      "epoch": 0.036833333333333336,
+      "grad_norm": 31.75,
+      "grad_norm_var": 16.7134765625,
+      "learning_rate": 9.966863828001982e-05,
+      "loss": 6.893,
+      "loss/crossentropy": 1.2962444424629211,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14901776053011417,
+      "step": 221
+    },
+    {
+      "epoch": 0.037,
+      "grad_norm": 32.5,
+      "grad_norm_var": 16.363997395833334,
+      "learning_rate": 9.966562242412442e-05,
+      "loss": 6.9004,
+      "loss/crossentropy": 1.8578623831272125,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23354769870638847,
+      "step": 222
+    },
+    {
+      "epoch": 0.03716666666666667,
+      "grad_norm": 28.25,
+      "grad_norm_var": 17.343489583333334,
+      "learning_rate": 9.966259295211697e-05,
+      "loss": 7.047,
+      "loss/crossentropy": 1.6259884238243103,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1877981834113598,
+      "step": 223
+    },
+    {
+      "epoch": 0.037333333333333336,
+      "grad_norm": 30.25,
+      "grad_norm_var": 17.720833333333335,
+      "learning_rate": 9.965954986482799e-05,
+      "loss": 7.1541,
+      "loss/crossentropy": 1.7171413600444794,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21186752244830132,
+      "step": 224
+    },
+    {
+      "epoch": 0.0375,
+      "grad_norm": 29.5,
+      "grad_norm_var": 13.924739583333333,
+      "learning_rate": 9.965649316309178e-05,
+      "loss": 6.9774,
+      "loss/crossentropy": 1.3424269035458565,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27712389826774597,
+      "step": 225
+    },
+    {
+      "epoch": 0.03766666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 13.385872395833333,
+      "learning_rate": 9.965342284774632e-05,
+      "loss": 7.0364,
+      "loss/crossentropy": 1.363720841705799,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14927429426461458,
+      "step": 226
+    },
+    {
+      "epoch": 0.03783333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 13.345768229166667,
+      "learning_rate": 9.965033891963338e-05,
+      "loss": 7.1098,
+      "loss/crossentropy": 1.637398675084114,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22406315430998802,
+      "step": 227
+    },
+    {
+      "epoch": 0.038,
+      "grad_norm": 32.5,
+      "grad_norm_var": 13.191080729166666,
+      "learning_rate": 9.964724137959843e-05,
+      "loss": 7.4034,
+      "loss/crossentropy": 1.655188411474228,
+      "loss/hidden": 3.9921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2601375915110111,
+      "step": 228
+    },
+    {
+      "epoch": 0.03816666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 13.608072916666666,
+      "learning_rate": 9.964413022849068e-05,
+      "loss": 6.9138,
+      "loss/crossentropy": 1.5690461844205856,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17689584381878376,
+      "step": 229
+    },
+    {
+      "epoch": 0.03833333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 13.051822916666667,
+      "learning_rate": 9.964100546716309e-05,
+      "loss": 6.8051,
+      "loss/crossentropy": 1.7078745067119598,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16476575285196304,
+      "step": 230
+    },
+    {
+      "epoch": 0.0385,
+      "grad_norm": 37.25,
+      "grad_norm_var": 13.538541666666667,
+      "learning_rate": 9.963786709647228e-05,
+      "loss": 7.2079,
+      "loss/crossentropy": 1.1831867694854736,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15216390788555145,
+      "step": 231
+    },
+    {
+      "epoch": 0.03866666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 13.470572916666667,
+      "learning_rate": 9.963471511727868e-05,
+      "loss": 7.4329,
+      "loss/crossentropy": 1.100526675581932,
+      "loss/hidden": 3.9921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3972581662237644,
+      "step": 232
+    },
+    {
+      "epoch": 0.03883333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 13.4478515625,
+      "learning_rate": 9.963154953044645e-05,
+      "loss": 7.0174,
+      "loss/crossentropy": 1.3149209916591644,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17489828169345856,
+      "step": 233
+    },
+    {
+      "epoch": 0.039,
+      "grad_norm": 32.0,
+      "grad_norm_var": 13.188997395833333,
+      "learning_rate": 9.962837033684343e-05,
+      "loss": 7.0444,
+      "loss/crossentropy": 1.6122197806835175,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18191683292388916,
+      "step": 234
+    },
+    {
+      "epoch": 0.03916666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 12.828580729166667,
+      "learning_rate": 9.96251775373412e-05,
+      "loss": 7.2731,
+      "loss/crossentropy": 1.6611078679561615,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18038855865597725,
+      "step": 235
+    },
+    {
+      "epoch": 0.03933333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.198893229166667,
+      "learning_rate": 9.962197113281509e-05,
+      "loss": 7.3323,
+      "loss/crossentropy": 1.4444286078214645,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18717906437814236,
+      "step": 236
+    },
+    {
+      "epoch": 0.0395,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.203059895833333,
+      "learning_rate": 9.961875112414416e-05,
+      "loss": 7.394,
+      "loss/crossentropy": 1.2385891154408455,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16062070801854134,
+      "step": 237
+    },
+    {
+      "epoch": 0.03966666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 4.233333333333333,
+      "learning_rate": 9.961551751221121e-05,
+      "loss": 7.2373,
+      "loss/crossentropy": 1.59829643368721,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18773933500051498,
+      "step": 238
+    },
+    {
+      "epoch": 0.03983333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.7684895833333334,
+      "learning_rate": 9.961227029790272e-05,
+      "loss": 7.4197,
+      "loss/crossentropy": 1.8970681428909302,
+      "loss/hidden": 3.640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2612060457468033,
+      "step": 239
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.6718098958333334,
+      "learning_rate": 9.960900948210896e-05,
+      "loss": 7.1812,
+      "loss/crossentropy": 1.66687873005867,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2203241027891636,
+      "step": 240
+    },
+    {
+      "epoch": 0.04016666666666667,
+      "grad_norm": 35.0,
+      "grad_norm_var": 3.9181640625,
+      "learning_rate": 9.96057350657239e-05,
+      "loss": 6.4739,
+      "loss/crossentropy": 1.127626657485962,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14585050754249096,
+      "step": 241
+    },
+    {
+      "epoch": 0.04033333333333333,
+      "grad_norm": 34.5,
+      "grad_norm_var": 4.154166666666667,
+      "learning_rate": 9.960244704964521e-05,
+      "loss": 7.154,
+      "loss/crossentropy": 1.389030396938324,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1379571333527565,
+      "step": 242
+    },
+    {
+      "epoch": 0.0405,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.157291666666667,
+      "learning_rate": 9.959914543477435e-05,
+      "loss": 7.129,
+      "loss/crossentropy": 1.4154839515686035,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17527304217219353,
+      "step": 243
+    },
+    {
+      "epoch": 0.04066666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 4.208072916666667,
+      "learning_rate": 9.959583022201647e-05,
+      "loss": 7.106,
+      "loss/crossentropy": 1.0321640968322754,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17023729532957077,
+      "step": 244
+    },
+    {
+      "epoch": 0.04083333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.692643229166667,
+      "learning_rate": 9.959250141228045e-05,
+      "loss": 7.0659,
+      "loss/crossentropy": 1.6669679433107376,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35606617107987404,
+      "step": 245
+    },
+    {
+      "epoch": 0.041,
+      "grad_norm": 55.25,
+      "grad_norm_var": 36.339518229166664,
+      "learning_rate": 9.95891590064789e-05,
+      "loss": 6.9787,
+      "loss/crossentropy": 1.5344274044036865,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32492806017398834,
+      "step": 246
+    },
+    {
+      "epoch": 0.041166666666666664,
+      "grad_norm": 33.5,
+      "grad_norm_var": 35.503580729166664,
+      "learning_rate": 9.958580300552815e-05,
+      "loss": 7.3157,
+      "loss/crossentropy": 2.0076797902584076,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2526522632688284,
+      "step": 247
+    },
+    {
+      "epoch": 0.04133333333333333,
+      "grad_norm": 36.5,
+      "grad_norm_var": 35.995768229166664,
+      "learning_rate": 9.958243341034827e-05,
+      "loss": 6.9608,
+      "loss/crossentropy": 1.737393319606781,
+      "loss/hidden": 3.98828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2978449836373329,
+      "step": 248
+    },
+    {
+      "epoch": 0.0415,
+      "grad_norm": 32.5,
+      "grad_norm_var": 35.526041666666664,
+      "learning_rate": 9.957905022186309e-05,
+      "loss": 7.5004,
+      "loss/crossentropy": 1.8952009677886963,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19244135171175003,
+      "step": 249
+    },
+    {
+      "epoch": 0.041666666666666664,
+      "grad_norm": 27.875,
+      "grad_norm_var": 37.63795572916667,
+      "learning_rate": 9.957565344100009e-05,
+      "loss": 6.7315,
+      "loss/crossentropy": 1.7154051959514618,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1966639757156372,
+      "step": 250
+    },
+    {
+      "epoch": 0.041833333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 38.16015625,
+      "learning_rate": 9.957224306869053e-05,
+      "loss": 7.0747,
+      "loss/crossentropy": 1.884273886680603,
+      "loss/hidden": 3.80078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24361001327633858,
+      "step": 251
+    },
+    {
+      "epoch": 0.042,
+      "grad_norm": 30.875,
+      "grad_norm_var": 38.597330729166664,
+      "learning_rate": 9.956881910586937e-05,
+      "loss": 6.7937,
+      "loss/crossentropy": 1.5800000429153442,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19121325388550758,
+      "step": 252
+    },
+    {
+      "epoch": 0.042166666666666665,
+      "grad_norm": 33.5,
+      "grad_norm_var": 38.3369140625,
+      "learning_rate": 9.956538155347534e-05,
+      "loss": 7.4439,
+      "loss/crossentropy": 1.6637271493673325,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2565694823861122,
+      "step": 253
+    },
+    {
+      "epoch": 0.042333333333333334,
+      "grad_norm": 37.25,
+      "grad_norm_var": 38.50598958333333,
+      "learning_rate": 9.956193041245084e-05,
+      "loss": 7.0995,
+      "loss/crossentropy": 1.5276811718940735,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16093368455767632,
+      "step": 254
+    },
+    {
+      "epoch": 0.0425,
+      "grad_norm": 32.5,
+      "grad_norm_var": 37.28723958333333,
+      "learning_rate": 9.955846568374201e-05,
+      "loss": 6.4927,
+      "loss/crossentropy": 1.6154318898916245,
+      "loss/hidden": 3.13671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13068031892180443,
+      "step": 255
+    },
+    {
+      "epoch": 0.042666666666666665,
+      "grad_norm": 28.375,
+      "grad_norm_var": 38.76640625,
+      "learning_rate": 9.955498736829875e-05,
+      "loss": 6.6218,
+      "loss/crossentropy": 1.550460159778595,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17490394413471222,
+      "step": 256
+    },
+    {
+      "epoch": 0.042833333333333334,
+      "grad_norm": 26.25,
+      "grad_norm_var": 42.36666666666667,
+      "learning_rate": 9.955149546707465e-05,
+      "loss": 6.7822,
+      "loss/crossentropy": 1.4549120962619781,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16110243648290634,
+      "step": 257
+    },
+    {
+      "epoch": 0.043,
+      "grad_norm": 31.75,
+      "grad_norm_var": 42.44973958333333,
+      "learning_rate": 9.954798998102702e-05,
+      "loss": 6.7702,
+      "loss/crossentropy": 1.3850528970360756,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1854463703930378,
+      "step": 258
+    },
+    {
+      "epoch": 0.043166666666666666,
+      "grad_norm": 27.875,
+      "grad_norm_var": 44.1244140625,
+      "learning_rate": 9.954447091111694e-05,
+      "loss": 7.0334,
+      "loss/crossentropy": 1.6108764857053757,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20439260825514793,
+      "step": 259
+    },
+    {
+      "epoch": 0.043333333333333335,
+      "grad_norm": 30.5,
+      "grad_norm_var": 44.3384765625,
+      "learning_rate": 9.954093825830917e-05,
+      "loss": 7.574,
+      "loss/crossentropy": 2.0142699629068375,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1655888967216015,
+      "step": 260
+    },
+    {
+      "epoch": 0.0435,
+      "grad_norm": 32.75,
+      "grad_norm_var": 44.342122395833336,
+      "learning_rate": 9.953739202357218e-05,
+      "loss": 7.3679,
+      "loss/crossentropy": 1.7621987462043762,
+      "loss/hidden": 3.7734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2661568485200405,
+      "step": 261
+    },
+    {
+      "epoch": 0.043666666666666666,
+      "grad_norm": 35.75,
+      "grad_norm_var": 10.196809895833333,
+      "learning_rate": 9.953383220787824e-05,
+      "loss": 7.0685,
+      "loss/crossentropy": 1.4321940243244171,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2649080101400614,
+      "step": 262
+    },
+    {
+      "epoch": 0.043833333333333335,
+      "grad_norm": 33.25,
+      "grad_norm_var": 10.142643229166667,
+      "learning_rate": 9.953025881220325e-05,
+      "loss": 7.1322,
+      "loss/crossentropy": 1.6164200007915497,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16854076832532883,
+      "step": 263
+    },
+    {
+      "epoch": 0.044,
+      "grad_norm": 31.5,
+      "grad_norm_var": 8.533268229166667,
+      "learning_rate": 9.952667183752689e-05,
+      "loss": 7.7075,
+      "loss/crossentropy": 1.5773536264896393,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20186477713286877,
+      "step": 264
+    },
+    {
+      "epoch": 0.04416666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 8.567122395833334,
+      "learning_rate": 9.952307128483256e-05,
+      "loss": 7.0641,
+      "loss/crossentropy": 1.9845942556858063,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23343012481927872,
+      "step": 265
+    },
+    {
+      "epoch": 0.044333333333333336,
+      "grad_norm": 33.25,
+      "grad_norm_var": 7.937239583333334,
+      "learning_rate": 9.951945715510738e-05,
+      "loss": 6.9057,
+      "loss/crossentropy": 1.6168481409549713,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16589028760790825,
+      "step": 266
+    },
+    {
+      "epoch": 0.0445,
+      "grad_norm": 30.125,
+      "grad_norm_var": 7.982291666666667,
+      "learning_rate": 9.951582944934215e-05,
+      "loss": 6.9383,
+      "loss/crossentropy": 1.5559693723917007,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25133585184812546,
+      "step": 267
+    },
+    {
+      "epoch": 0.04466666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 7.968684895833333,
+      "learning_rate": 9.951218816853145e-05,
+      "loss": 7.1961,
+      "loss/crossentropy": 1.3414212465286255,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1478818543255329,
+      "step": 268
+    },
+    {
+      "epoch": 0.044833333333333336,
+      "grad_norm": 31.5,
+      "grad_norm_var": 7.733268229166667,
+      "learning_rate": 9.950853331367356e-05,
+      "loss": 7.0719,
+      "loss/crossentropy": 1.5315271243453026,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16057554073631763,
+      "step": 269
+    },
+    {
+      "epoch": 0.045,
+      "grad_norm": 30.125,
+      "grad_norm_var": 5.495572916666666,
+      "learning_rate": 9.950486488577045e-05,
+      "loss": 6.7578,
+      "loss/crossentropy": 1.6037451326847076,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1604137122631073,
+      "step": 270
+    },
+    {
+      "epoch": 0.04516666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.495572916666666,
+      "learning_rate": 9.950118288582788e-05,
+      "loss": 6.8719,
+      "loss/crossentropy": 1.1782431602478027,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18701835349202156,
+      "step": 271
+    },
+    {
+      "epoch": 0.04533333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 4.965625,
+      "learning_rate": 9.949748731485527e-05,
+      "loss": 7.044,
+      "loss/crossentropy": 1.400882750749588,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26591288298368454,
+      "step": 272
+    },
+    {
+      "epoch": 0.0455,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.190625,
+      "learning_rate": 9.949377817386579e-05,
+      "loss": 7.4913,
+      "loss/crossentropy": 0.867219865322113,
+      "loss/hidden": 3.73828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2675766684114933,
+      "step": 273
+    },
+    {
+      "epoch": 0.04566666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.4462890625,
+      "learning_rate": 9.949005546387631e-05,
+      "loss": 7.1206,
+      "loss/crossentropy": 1.874289482831955,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20881149545311928,
+      "step": 274
+    },
+    {
+      "epoch": 0.04583333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.6947916666666667,
+      "learning_rate": 9.948631918590746e-05,
+      "loss": 6.9333,
+      "loss/crossentropy": 1.5097296833992004,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1876576580107212,
+      "step": 275
+    },
+    {
+      "epoch": 0.046,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.98515625,
+      "learning_rate": 9.948256934098352e-05,
+      "loss": 7.4482,
+      "loss/crossentropy": 1.0763804763555527,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27809485141187906,
+      "step": 276
+    },
+    {
+      "epoch": 0.04616666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.8884765625,
+      "learning_rate": 9.947880593013255e-05,
+      "loss": 7.3084,
+      "loss/crossentropy": 2.053309381008148,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22167784348130226,
+      "step": 277
+    },
+    {
+      "epoch": 0.04633333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.67265625,
+      "learning_rate": 9.947502895438631e-05,
+      "loss": 7.3537,
+      "loss/crossentropy": 1.3560047149658203,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16262023150920868,
+      "step": 278
+    },
+    {
+      "epoch": 0.0465,
+      "grad_norm": 37.0,
+      "grad_norm_var": 3.60625,
+      "learning_rate": 9.94712384147803e-05,
+      "loss": 7.1441,
+      "loss/crossentropy": 1.6553308367729187,
+      "loss/hidden": 3.80078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34052540734410286,
+      "step": 279
+    },
+    {
+      "epoch": 0.04666666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.771875,
+      "learning_rate": 9.94674343123537e-05,
+      "loss": 7.0543,
+      "loss/crossentropy": 1.671010285615921,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23619508370757103,
+      "step": 280
+    },
+    {
+      "epoch": 0.04683333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.771875,
+      "learning_rate": 9.946361664814943e-05,
+      "loss": 6.9507,
+      "loss/crossentropy": 1.2728363275527954,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20054782927036285,
+      "step": 281
+    },
+    {
+      "epoch": 0.047,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.66015625,
+      "learning_rate": 9.945978542321411e-05,
+      "loss": 6.5744,
+      "loss/crossentropy": 0.998386912047863,
+      "loss/hidden": 3.03125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09477789141237736,
+      "step": 282
+    },
+    {
+      "epoch": 0.04716666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.569205729166667,
+      "learning_rate": 9.945594063859809e-05,
+      "loss": 6.9131,
+      "loss/crossentropy": 1.7489349246025085,
+      "loss/hidden": 3.640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32071322202682495,
+      "step": 283
+    },
+    {
+      "epoch": 0.04733333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.7514973958333333,
+      "learning_rate": 9.945208229535548e-05,
+      "loss": 7.4689,
+      "loss/crossentropy": 2.0416936576366425,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23788784816861153,
+      "step": 284
+    },
+    {
+      "epoch": 0.0475,
+      "grad_norm": 28.5,
+      "grad_norm_var": 4.279622395833333,
+      "learning_rate": 9.944821039454402e-05,
+      "loss": 7.1895,
+      "loss/crossentropy": 1.7832486033439636,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18590330705046654,
+      "step": 285
+    },
+    {
+      "epoch": 0.04766666666666667,
+      "grad_norm": 37.0,
+      "grad_norm_var": 6.223958333333333,
+      "learning_rate": 9.944432493722524e-05,
+      "loss": 7.046,
+      "loss/crossentropy": 1.468076929450035,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1862525474280119,
+      "step": 286
+    },
+    {
+      "epoch": 0.04783333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 6.462239583333333,
+      "learning_rate": 9.944042592446434e-05,
+      "loss": 7.1954,
+      "loss/crossentropy": 1.3620029240846634,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1722669992595911,
+      "step": 287
+    },
+    {
+      "epoch": 0.048,
+      "grad_norm": 32.25,
+      "grad_norm_var": 6.449934895833334,
+      "learning_rate": 9.943651335733028e-05,
+      "loss": 7.345,
+      "loss/crossentropy": 1.3657430112361908,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1477232426404953,
+      "step": 288
+    },
+    {
+      "epoch": 0.04816666666666667,
+      "grad_norm": 31.875,
+      "grad_norm_var": 6.436458333333333,
+      "learning_rate": 9.94325872368957e-05,
+      "loss": 7.3979,
+      "loss/crossentropy": 1.3430730998516083,
+      "loss/hidden": 3.78515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1977537926286459,
+      "step": 289
+    },
+    {
+      "epoch": 0.04833333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 6.174739583333333,
+      "learning_rate": 9.942864756423697e-05,
+      "loss": 6.9853,
+      "loss/crossentropy": 1.7727776169776917,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17479711771011353,
+      "step": 290
+    },
+    {
+      "epoch": 0.0485,
+      "grad_norm": 30.125,
+      "grad_norm_var": 6.1447265625,
+      "learning_rate": 9.942469434043418e-05,
+      "loss": 7.0402,
+      "loss/crossentropy": 1.56208536028862,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21697872504591942,
+      "step": 291
+    },
+    {
+      "epoch": 0.048666666666666664,
+      "grad_norm": 29.25,
+      "grad_norm_var": 6.1447265625,
+      "learning_rate": 9.942072756657112e-05,
+      "loss": 6.9928,
+      "loss/crossentropy": 1.5282710641622543,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1731567531824112,
+      "step": 292
+    },
+    {
+      "epoch": 0.04883333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 6.353125,
+      "learning_rate": 9.941674724373531e-05,
+      "loss": 6.8876,
+      "loss/crossentropy": 1.3770808428525925,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21321340836584568,
+      "step": 293
+    },
+    {
+      "epoch": 0.049,
+      "grad_norm": 27.75,
+      "grad_norm_var": 7.230143229166667,
+      "learning_rate": 9.941275337301796e-05,
+      "loss": 6.5595,
+      "loss/crossentropy": 1.5444412529468536,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14273350313305855,
+      "step": 294
+    },
+    {
+      "epoch": 0.049166666666666664,
+      "grad_norm": 36.25,
+      "grad_norm_var": 6.728580729166667,
+      "learning_rate": 9.940874595551404e-05,
+      "loss": 7.0873,
+      "loss/crossentropy": 1.8509480655193329,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1753649227321148,
+      "step": 295
+    },
+    {
+      "epoch": 0.04933333333333333,
+      "grad_norm": 42.0,
+      "grad_norm_var": 13.487955729166666,
+      "learning_rate": 9.940472499232217e-05,
+      "loss": 7.2649,
+      "loss/crossentropy": 2.137119799852371,
+      "loss/hidden": 3.828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3466581925749779,
+      "step": 296
+    },
+    {
+      "epoch": 0.0495,
+      "grad_norm": 30.0,
+      "grad_norm_var": 13.487955729166666,
+      "learning_rate": 9.940069048454476e-05,
+      "loss": 6.76,
+      "loss/crossentropy": 2.316546082496643,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24957827851176262,
+      "step": 297
+    },
+    {
+      "epoch": 0.049666666666666665,
+      "grad_norm": 30.125,
+      "grad_norm_var": 13.453125,
+      "learning_rate": 9.939664243328788e-05,
+      "loss": 6.9381,
+      "loss/crossentropy": 1.629784494638443,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2546052001416683,
+      "step": 298
+    },
+    {
+      "epoch": 0.049833333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 13.5775390625,
+      "learning_rate": 9.939258083966131e-05,
+      "loss": 6.7227,
+      "loss/crossentropy": 1.5890116840600967,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1683155559003353,
+      "step": 299
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 29.875,
+      "grad_norm_var": 13.77265625,
+      "learning_rate": 9.938850570477858e-05,
+      "loss": 7.2107,
+      "loss/crossentropy": 1.9451849162578583,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17290722578763962,
+      "step": 300
+    },
+    {
+      "epoch": 0.050166666666666665,
+      "grad_norm": 37.25,
+      "grad_norm_var": 14.602083333333333,
+      "learning_rate": 9.938441702975689e-05,
+      "loss": 6.9981,
+      "loss/crossentropy": 2.071776658296585,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25798892974853516,
+      "step": 301
+    },
+    {
+      "epoch": 0.050333333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 13.343489583333334,
+      "learning_rate": 9.93803148157172e-05,
+      "loss": 6.7827,
+      "loss/crossentropy": 1.1773535460233688,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13183625414967537,
+      "step": 302
+    },
+    {
+      "epoch": 0.0505,
+      "grad_norm": 32.5,
+      "grad_norm_var": 13.152083333333334,
+      "learning_rate": 9.937619906378413e-05,
+      "loss": 6.8614,
+      "loss/crossentropy": 2.0693687200546265,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19452445581555367,
+      "step": 303
+    },
+    {
+      "epoch": 0.050666666666666665,
+      "grad_norm": 29.625,
+      "grad_norm_var": 13.473372395833334,
+      "learning_rate": 9.937206977508604e-05,
+      "loss": 7.1653,
+      "loss/crossentropy": 2.0442996323108673,
+      "loss/hidden": 3.66015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17838554084300995,
+      "step": 304
+    },
+    {
+      "epoch": 0.050833333333333335,
+      "grad_norm": 29.5,
+      "grad_norm_var": 13.79375,
+      "learning_rate": 9.936792695075502e-05,
+      "loss": 7.3782,
+      "loss/crossentropy": 1.3510285913944244,
+      "loss/hidden": 3.14453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10517092235386372,
+      "step": 305
+    },
+    {
+      "epoch": 0.051,
+      "grad_norm": 29.0,
+      "grad_norm_var": 14.2009765625,
+      "learning_rate": 9.936377059192683e-05,
+      "loss": 6.9483,
+      "loss/crossentropy": 1.8398471474647522,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18359501659870148,
+      "step": 306
+    },
+    {
+      "epoch": 0.051166666666666666,
+      "grad_norm": 27.875,
+      "grad_norm_var": 14.9322265625,
+      "learning_rate": 9.935960069974096e-05,
+      "loss": 6.4221,
+      "loss/crossentropy": 1.8597908020019531,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18311693333089352,
+      "step": 307
+    },
+    {
+      "epoch": 0.051333333333333335,
+      "grad_norm": 30.75,
+      "grad_norm_var": 14.6494140625,
+      "learning_rate": 9.935541727534062e-05,
+      "loss": 6.9276,
+      "loss/crossentropy": 1.6094922870397568,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26469697803258896,
+      "step": 308
+    },
+    {
+      "epoch": 0.0515,
+      "grad_norm": 33.25,
+      "grad_norm_var": 14.676497395833334,
+      "learning_rate": 9.93512203198727e-05,
+      "loss": 6.8775,
+      "loss/crossentropy": 1.4130693972110748,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1726217232644558,
+      "step": 309
+    },
+    {
+      "epoch": 0.051666666666666666,
+      "grad_norm": 28.125,
+      "grad_norm_var": 14.489583333333334,
+      "learning_rate": 9.934700983448785e-05,
+      "loss": 6.4815,
+      "loss/crossentropy": 1.8149312138557434,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14228468015789986,
+      "step": 310
+    },
+    {
+      "epoch": 0.051833333333333335,
+      "grad_norm": 27.25,
+      "grad_norm_var": 14.077083333333333,
+      "learning_rate": 9.934278582034037e-05,
+      "loss": 6.7575,
+      "loss/crossentropy": 1.4098316729068756,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15191207826137543,
+      "step": 311
+    },
+    {
+      "epoch": 0.052,
+      "grad_norm": 29.5,
+      "grad_norm_var": 5.717708333333333,
+      "learning_rate": 9.93385482785883e-05,
+      "loss": 7.0661,
+      "loss/crossentropy": 1.2506734728813171,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2104758396744728,
+      "step": 312
+    },
+    {
+      "epoch": 0.05216666666666667,
+      "grad_norm": 28.625,
+      "grad_norm_var": 5.898893229166666,
+      "learning_rate": 9.93342972103934e-05,
+      "loss": 6.8494,
+      "loss/crossentropy": 1.9692685008049011,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17125158198177814,
+      "step": 313
+    },
+    {
+      "epoch": 0.052333333333333336,
+      "grad_norm": 31.5,
+      "grad_norm_var": 5.992708333333334,
+      "learning_rate": 9.933003261692113e-05,
+      "loss": 7.0405,
+      "loss/crossentropy": 1.5915742814540863,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1476550530642271,
+      "step": 314
+    },
+    {
+      "epoch": 0.0525,
+      "grad_norm": 31.875,
+      "grad_norm_var": 6.137239583333334,
+      "learning_rate": 9.932575449934062e-05,
+      "loss": 7.5291,
+      "loss/crossentropy": 2.2194809913635254,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1954730562865734,
+      "step": 315
+    },
+    {
+      "epoch": 0.05266666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 7.014518229166667,
+      "learning_rate": 9.932146285882477e-05,
+      "loss": 6.8653,
+      "loss/crossentropy": 1.9022675156593323,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17530881613492966,
+      "step": 316
+    },
+    {
+      "epoch": 0.052833333333333336,
+      "grad_norm": 29.25,
+      "grad_norm_var": 4.0228515625,
+      "learning_rate": 9.931715769655015e-05,
+      "loss": 6.7117,
+      "loss/crossentropy": 1.8810075521469116,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20105278491973877,
+      "step": 317
+    },
+    {
+      "epoch": 0.053,
+      "grad_norm": 30.25,
+      "grad_norm_var": 4.0228515625,
+      "learning_rate": 9.931283901369706e-05,
+      "loss": 7.1338,
+      "loss/crossentropy": 1.6084867715835571,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17544957250356674,
+      "step": 318
+    },
+    {
+      "epoch": 0.05316666666666667,
+      "grad_norm": 26.75,
+      "grad_norm_var": 4.322330729166667,
+      "learning_rate": 9.930850681144945e-05,
+      "loss": 6.7322,
+      "loss/crossentropy": 1.9372955560684204,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22798743844032288,
+      "step": 319
+    },
+    {
+      "epoch": 0.05333333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.369791666666667,
+      "learning_rate": 9.930416109099505e-05,
+      "loss": 6.7024,
+      "loss/crossentropy": 2.0897410213947296,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20163432508707047,
+      "step": 320
+    },
+    {
+      "epoch": 0.0535,
+      "grad_norm": 31.625,
+      "grad_norm_var": 4.5369140625,
+      "learning_rate": 9.929980185352526e-05,
+      "loss": 7.0626,
+      "loss/crossentropy": 1.799334704875946,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16541759110987186,
+      "step": 321
+    },
+    {
+      "epoch": 0.05366666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 4.483268229166667,
+      "learning_rate": 9.929542910023517e-05,
+      "loss": 6.8316,
+      "loss/crossentropy": 1.4304459393024445,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2662687115371227,
+      "step": 322
+    },
+    {
+      "epoch": 0.05383333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.624739583333334,
+      "learning_rate": 9.929104283232362e-05,
+      "loss": 6.8603,
+      "loss/crossentropy": 1.520107924938202,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1625481117516756,
+      "step": 323
+    },
+    {
+      "epoch": 0.054,
+      "grad_norm": 32.75,
+      "grad_norm_var": 4.970572916666667,
+      "learning_rate": 9.928664305099314e-05,
+      "loss": 7.2975,
+      "loss/crossentropy": 1.922645777463913,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35359594598412514,
+      "step": 324
+    },
+    {
+      "epoch": 0.05416666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 4.571809895833334,
+      "learning_rate": 9.928222975744991e-05,
+      "loss": 6.8766,
+      "loss/crossentropy": 2.0565673410892487,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19727135077118874,
+      "step": 325
+    },
+    {
+      "epoch": 0.05433333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.261458333333334,
+      "learning_rate": 9.927780295290389e-05,
+      "loss": 7.1271,
+      "loss/crossentropy": 1.5576016902923584,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18772463873028755,
+      "step": 326
+    },
+    {
+      "epoch": 0.0545,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.614518229166667,
+      "learning_rate": 9.927336263856872e-05,
+      "loss": 6.9329,
+      "loss/crossentropy": 1.5006282478570938,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23727844608947635,
+      "step": 327
+    },
+    {
+      "epoch": 0.05466666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.700455729166667,
+      "learning_rate": 9.926890881566171e-05,
+      "loss": 6.8211,
+      "loss/crossentropy": 1.223441168665886,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14496134035289288,
+      "step": 328
+    },
+    {
+      "epoch": 0.05483333333333333,
+      "grad_norm": 27.625,
+      "grad_norm_var": 4.016080729166666,
+      "learning_rate": 9.926444148540393e-05,
+      "loss": 6.309,
+      "loss/crossentropy": 0.858293853700161,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10236557200551033,
+      "step": 329
+    },
+    {
+      "epoch": 0.055,
+      "grad_norm": 34.5,
+      "grad_norm_var": 4.994205729166667,
+      "learning_rate": 9.925996064902011e-05,
+      "loss": 7.0001,
+      "loss/crossentropy": 2.172655910253525,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1768443062901497,
+      "step": 330
+    },
+    {
+      "epoch": 0.05516666666666667,
+      "grad_norm": 39.5,
+      "grad_norm_var": 9.875,
+      "learning_rate": 9.92554663077387e-05,
+      "loss": 7.1776,
+      "loss/crossentropy": 1.2981346100568771,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17009487748146057,
+      "step": 331
+    },
+    {
+      "epoch": 0.05533333333333333,
+      "grad_norm": 35.25,
+      "grad_norm_var": 10.354166666666666,
+      "learning_rate": 9.925095846279184e-05,
+      "loss": 7.2192,
+      "loss/crossentropy": 1.36193186044693,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.253113005310297,
+      "step": 332
+    },
+    {
+      "epoch": 0.0555,
+      "grad_norm": 28.25,
+      "grad_norm_var": 10.675,
+      "learning_rate": 9.924643711541539e-05,
+      "loss": 6.9291,
+      "loss/crossentropy": 1.9987045526504517,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23344938457012177,
+      "step": 333
+    },
+    {
+      "epoch": 0.05566666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 10.773958333333333,
+      "learning_rate": 9.92419022668489e-05,
+      "loss": 7.828,
+      "loss/crossentropy": 2.0947715640068054,
+      "loss/hidden": 3.98828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4486834332346916,
+      "step": 334
+    },
+    {
+      "epoch": 0.05583333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 9.553580729166667,
+      "learning_rate": 9.923735391833564e-05,
+      "loss": 6.8631,
+      "loss/crossentropy": 1.9164935052394867,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16560781374573708,
+      "step": 335
+    },
+    {
+      "epoch": 0.056,
+      "grad_norm": 31.125,
+      "grad_norm_var": 9.526822916666667,
+      "learning_rate": 9.923279207112255e-05,
+      "loss": 7.1829,
+      "loss/crossentropy": 2.01135965436697,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3246071469038725,
+      "step": 336
+    },
+    {
+      "epoch": 0.05616666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 9.781705729166667,
+      "learning_rate": 9.922821672646027e-05,
+      "loss": 7.4666,
+      "loss/crossentropy": 1.5571194291114807,
+      "loss/hidden": 3.7109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16909234598279,
+      "step": 337
+    },
+    {
+      "epoch": 0.05633333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 9.98125,
+      "learning_rate": 9.922362788560319e-05,
+      "loss": 7.0297,
+      "loss/crossentropy": 1.3507355153560638,
+      "loss/hidden": 3.8828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23893941938877106,
+      "step": 338
+    },
+    {
+      "epoch": 0.0565,
+      "grad_norm": 30.5,
+      "grad_norm_var": 9.892708333333333,
+      "learning_rate": 9.921902554980934e-05,
+      "loss": 6.7374,
+      "loss/crossentropy": 1.8911243975162506,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2164096813648939,
+      "step": 339
+    },
+    {
+      "epoch": 0.056666666666666664,
+      "grad_norm": 31.125,
+      "grad_norm_var": 9.7666015625,
+      "learning_rate": 9.921440972034049e-05,
+      "loss": 7.1219,
+      "loss/crossentropy": 1.6565758883953094,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21637392789125443,
+      "step": 340
+    },
+    {
+      "epoch": 0.05683333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 9.944205729166667,
+      "learning_rate": 9.92097803984621e-05,
+      "loss": 6.6962,
+      "loss/crossentropy": 1.563918575644493,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1653151847422123,
+      "step": 341
+    },
+    {
+      "epoch": 0.057,
+      "grad_norm": 31.625,
+      "grad_norm_var": 9.930989583333334,
+      "learning_rate": 9.920513758544332e-05,
+      "loss": 7.2076,
+      "loss/crossentropy": 1.9003488719463348,
+      "loss/hidden": 3.82421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2942137196660042,
+      "step": 342
+    },
+    {
+      "epoch": 0.057166666666666664,
+      "grad_norm": 39.25,
+      "grad_norm_var": 13.5353515625,
+      "learning_rate": 9.920048128255699e-05,
+      "loss": 6.8842,
+      "loss/crossentropy": 1.453528344631195,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17757558822631836,
+      "step": 343
+    },
+    {
+      "epoch": 0.05733333333333333,
+      "grad_norm": 36.5,
+      "grad_norm_var": 14.1212890625,
+      "learning_rate": 9.919581149107968e-05,
+      "loss": 6.9399,
+      "loss/crossentropy": 1.048169419169426,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1615643361583352,
+      "step": 344
+    },
+    {
+      "epoch": 0.0575,
+      "grad_norm": 38.75,
+      "grad_norm_var": 14.776041666666666,
+      "learning_rate": 9.919112821229163e-05,
+      "loss": 6.8893,
+      "loss/crossentropy": 1.4191984683275223,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1365694012492895,
+      "step": 345
+    },
+    {
+      "epoch": 0.057666666666666665,
+      "grad_norm": 31.5,
+      "grad_norm_var": 14.776041666666666,
+      "learning_rate": 9.918643144747681e-05,
+      "loss": 7.4587,
+      "loss/crossentropy": 1.7477758526802063,
+      "loss/hidden": 3.734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23377961292862892,
+      "step": 346
+    },
+    {
+      "epoch": 0.057833333333333334,
+      "grad_norm": 27.875,
+      "grad_norm_var": 13.002018229166667,
+      "learning_rate": 9.918172119792282e-05,
+      "loss": 7.1625,
+      "loss/crossentropy": 1.7191437631845474,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14523478224873543,
+      "step": 347
+    },
+    {
+      "epoch": 0.058,
+      "grad_norm": 29.625,
+      "grad_norm_var": 12.676822916666667,
+      "learning_rate": 9.917699746492104e-05,
+      "loss": 7.2374,
+      "loss/crossentropy": 1.711678385734558,
+      "loss/hidden": 3.7578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3540639393031597,
+      "step": 348
+    },
+    {
+      "epoch": 0.058166666666666665,
+      "grad_norm": 29.0,
+      "grad_norm_var": 12.354166666666666,
+      "learning_rate": 9.917226024976649e-05,
+      "loss": 7.0111,
+      "loss/crossentropy": 1.871002122759819,
+      "loss/hidden": 3.80078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.253181129693985,
+      "step": 349
+    },
+    {
+      "epoch": 0.058333333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 12.370833333333334,
+      "learning_rate": 9.91675095537579e-05,
+      "loss": 6.9966,
+      "loss/crossentropy": 1.1223047077655792,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16907763667404652,
+      "step": 350
+    },
+    {
+      "epoch": 0.0585,
+      "grad_norm": 28.875,
+      "grad_norm_var": 12.618489583333334,
+      "learning_rate": 9.916274537819775e-05,
+      "loss": 7.1552,
+      "loss/crossentropy": 1.913307100534439,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2782452665269375,
+      "step": 351
+    },
+    {
+      "epoch": 0.058666666666666666,
+      "grad_norm": 37.25,
+      "grad_norm_var": 14.491080729166667,
+      "learning_rate": 9.915796772439207e-05,
+      "loss": 7.2787,
+      "loss/crossentropy": 2.047320604324341,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16641808673739433,
+      "step": 352
+    },
+    {
+      "epoch": 0.058833333333333335,
+      "grad_norm": 31.5,
+      "grad_norm_var": 14.363997395833334,
+      "learning_rate": 9.915317659365077e-05,
+      "loss": 6.9958,
+      "loss/crossentropy": 1.6410242393612862,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.165967701934278,
+      "step": 353
+    },
+    {
+      "epoch": 0.059,
+      "grad_norm": 30.875,
+      "grad_norm_var": 13.791080729166667,
+      "learning_rate": 9.914837198728733e-05,
+      "loss": 7.2045,
+      "loss/crossentropy": 1.2407794147729874,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1580798588693142,
+      "step": 354
+    },
+    {
+      "epoch": 0.059166666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 13.6181640625,
+      "learning_rate": 9.914355390661896e-05,
+      "loss": 7.0317,
+      "loss/crossentropy": 1.891918033361435,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24348440766334534,
+      "step": 355
+    },
+    {
+      "epoch": 0.059333333333333335,
+      "grad_norm": 31.375,
+      "grad_norm_var": 13.585872395833333,
+      "learning_rate": 9.913872235296657e-05,
+      "loss": 6.7965,
+      "loss/crossentropy": 1.520493820309639,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18387246876955032,
+      "step": 356
+    },
+    {
+      "epoch": 0.0595,
+      "grad_norm": 32.0,
+      "grad_norm_var": 12.545572916666666,
+      "learning_rate": 9.913387732765475e-05,
+      "loss": 7.5963,
+      "loss/crossentropy": 1.6907641142606735,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1569608375430107,
+      "step": 357
+    },
+    {
+      "epoch": 0.059666666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 12.851822916666666,
+      "learning_rate": 9.91290188320118e-05,
+      "loss": 7.325,
+      "loss/crossentropy": 1.5798785090446472,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17809266969561577,
+      "step": 358
+    },
+    {
+      "epoch": 0.059833333333333336,
+      "grad_norm": 31.625,
+      "grad_norm_var": 9.480143229166666,
+      "learning_rate": 9.91241468673697e-05,
+      "loss": 7.1735,
+      "loss/crossentropy": 2.05547434091568,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18853947147727013,
+      "step": 359
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 33.0,
+      "grad_norm_var": 8.091080729166666,
+      "learning_rate": 9.911926143506412e-05,
+      "loss": 7.1808,
+      "loss/crossentropy": 2.253300368785858,
+      "loss/hidden": 4.1015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2493579462170601,
+      "step": 360
+    },
+    {
+      "epoch": 0.06016666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 4.672330729166666,
+      "learning_rate": 9.911436253643445e-05,
+      "loss": 6.5164,
+      "loss/crossentropy": 1.4248720556497574,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18392807617783546,
+      "step": 361
+    },
+    {
+      "epoch": 0.060333333333333336,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.672330729166666,
+      "learning_rate": 9.910945017282372e-05,
+      "loss": 7.3267,
+      "loss/crossentropy": 1.6512728929519653,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14323543198406696,
+      "step": 362
+    },
+    {
+      "epoch": 0.0605,
+      "grad_norm": 28.875,
+      "grad_norm_var": 4.279622395833333,
+      "learning_rate": 9.91045243455787e-05,
+      "loss": 6.9783,
+      "loss/crossentropy": 1.2311359643936157,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13875133730471134,
+      "step": 363
+    },
+    {
+      "epoch": 0.06066666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 4.0681640625,
+      "learning_rate": 9.909958505604984e-05,
+      "loss": 6.9123,
+      "loss/crossentropy": 1.041396975517273,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10275209881365299,
+      "step": 364
+    },
+    {
+      "epoch": 0.060833333333333336,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.82890625,
+      "learning_rate": 9.909463230559127e-05,
+      "loss": 6.9391,
+      "loss/crossentropy": 1.6771114617586136,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1523474808782339,
+      "step": 365
+    },
+    {
+      "epoch": 0.061,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.8072916666666665,
+      "learning_rate": 9.908966609556079e-05,
+      "loss": 7.3729,
+      "loss/crossentropy": 1.1425088867545128,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1245612483471632,
+      "step": 366
+    },
+    {
+      "epoch": 0.06116666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.6103515625,
+      "learning_rate": 9.908468642731995e-05,
+      "loss": 7.018,
+      "loss/crossentropy": 1.5243964493274689,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15938062593340874,
+      "step": 367
+    },
+    {
+      "epoch": 0.06133333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 1.8093098958333333,
+      "learning_rate": 9.907969330223395e-05,
+      "loss": 6.8973,
+      "loss/crossentropy": 0.9294636249542236,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16288216598331928,
+      "step": 368
+    },
+    {
+      "epoch": 0.0615,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.8125,
+      "learning_rate": 9.907468672167165e-05,
+      "loss": 7.2048,
+      "loss/crossentropy": 1.645423173904419,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2051373142749071,
+      "step": 369
+    },
+    {
+      "epoch": 0.06166666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.8051432291666667,
+      "learning_rate": 9.906966668700567e-05,
+      "loss": 7.2549,
+      "loss/crossentropy": 1.6831459701061249,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19335630536079407,
+      "step": 370
+    },
+    {
+      "epoch": 0.06183333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.8952473958333333,
+      "learning_rate": 9.906463319961225e-05,
+      "loss": 7.1289,
+      "loss/crossentropy": 1.8449882715940475,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20429406687617302,
+      "step": 371
+    },
+    {
+      "epoch": 0.062,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.223372395833333,
+      "learning_rate": 9.90595862608714e-05,
+      "loss": 6.4735,
+      "loss/crossentropy": 1.3120336532592773,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1317860446870327,
+      "step": 372
+    },
+    {
+      "epoch": 0.06216666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.187239583333333,
+      "learning_rate": 9.90545258721667e-05,
+      "loss": 7.2773,
+      "loss/crossentropy": 1.5410139858722687,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16073031350970268,
+      "step": 373
+    },
+    {
+      "epoch": 0.06233333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.4723307291666665,
+      "learning_rate": 9.904945203488554e-05,
+      "loss": 7.3947,
+      "loss/crossentropy": 1.5286826118826866,
+      "loss/hidden": 3.109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13874854519963264,
+      "step": 374
+    },
+    {
+      "epoch": 0.0625,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.8848307291666666,
+      "learning_rate": 9.904436475041891e-05,
+      "loss": 6.9468,
+      "loss/crossentropy": 1.3267207443714142,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2536827567964792,
+      "step": 375
+    },
+    {
+      "epoch": 0.06266666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.7051432291666666,
+      "learning_rate": 9.903926402016153e-05,
+      "loss": 6.5394,
+      "loss/crossentropy": 1.119832620024681,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10435830242931843,
+      "step": 376
+    },
+    {
+      "epoch": 0.06283333333333334,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.5697916666666667,
+      "learning_rate": 9.903414984551179e-05,
+      "loss": 7.5438,
+      "loss/crossentropy": 2.038603186607361,
+      "loss/hidden": 3.78515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.322290800511837,
+      "step": 377
+    },
+    {
+      "epoch": 0.063,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.816666666666667,
+      "learning_rate": 9.902902222787175e-05,
+      "loss": 7.1061,
+      "loss/crossentropy": 1.7512010633945465,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3207052852958441,
+      "step": 378
+    },
+    {
+      "epoch": 0.06316666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.4872395833333334,
+      "learning_rate": 9.902388116864722e-05,
+      "loss": 6.8527,
+      "loss/crossentropy": 1.269251748919487,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12383817881345749,
+      "step": 379
+    },
+    {
+      "epoch": 0.06333333333333334,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.8686848958333333,
+      "learning_rate": 9.901872666924764e-05,
+      "loss": 7.2741,
+      "loss/crossentropy": 1.2731651365756989,
+      "loss/hidden": 3.83984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23384994268417358,
+      "step": 380
+    },
+    {
+      "epoch": 0.0635,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.7895833333333333,
+      "learning_rate": 9.901355873108609e-05,
+      "loss": 7.0535,
+      "loss/crossentropy": 1.5953963994979858,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23109585791826248,
+      "step": 381
+    },
+    {
+      "epoch": 0.06366666666666666,
+      "grad_norm": 28.375,
+      "grad_norm_var": 3.285872395833333,
+      "learning_rate": 9.900837735557947e-05,
+      "loss": 6.5947,
+      "loss/crossentropy": 1.602350801229477,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13890548422932625,
+      "step": 382
+    },
+    {
+      "epoch": 0.06383333333333334,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.218489583333333,
+      "learning_rate": 9.900318254414821e-05,
+      "loss": 6.936,
+      "loss/crossentropy": 1.5561874359846115,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23637470044195652,
+      "step": 383
+    },
+    {
+      "epoch": 0.064,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.6083333333333334,
+      "learning_rate": 9.899797429821656e-05,
+      "loss": 7.0894,
+      "loss/crossentropy": 1.3163021504878998,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16271806322038174,
+      "step": 384
+    },
+    {
+      "epoch": 0.06416666666666666,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.6405598958333334,
+      "learning_rate": 9.899275261921234e-05,
+      "loss": 6.5898,
+      "loss/crossentropy": 1.5003906786441803,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16999328136444092,
+      "step": 385
+    },
+    {
+      "epoch": 0.06433333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.7353515625,
+      "learning_rate": 9.898751750856713e-05,
+      "loss": 7.3537,
+      "loss/crossentropy": 1.761914700269699,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21671501360833645,
+      "step": 386
+    },
+    {
+      "epoch": 0.0645,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.4400390625,
+      "learning_rate": 9.898226896771619e-05,
+      "loss": 7.0966,
+      "loss/crossentropy": 1.324119359254837,
+      "loss/hidden": 3.7109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24070443212985992,
+      "step": 387
+    },
+    {
+      "epoch": 0.06466666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.3427083333333334,
+      "learning_rate": 9.897700699809837e-05,
+      "loss": 7.329,
+      "loss/crossentropy": 1.9189965426921844,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16512704640626907,
+      "step": 388
+    },
+    {
+      "epoch": 0.06483333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.3427083333333334,
+      "learning_rate": 9.897173160115632e-05,
+      "loss": 6.7596,
+      "loss/crossentropy": 1.5462406650185585,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12559622339904308,
+      "step": 389
+    },
+    {
+      "epoch": 0.065,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.7997395833333334,
+      "learning_rate": 9.896644277833631e-05,
+      "loss": 6.4505,
+      "loss/crossentropy": 1.1322835385799408,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18489011749625206,
+      "step": 390
+    },
+    {
+      "epoch": 0.06516666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.5728515625,
+      "learning_rate": 9.896114053108829e-05,
+      "loss": 7.0024,
+      "loss/crossentropy": 1.5260199159383774,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2132732141762972,
+      "step": 391
+    },
+    {
+      "epoch": 0.06533333333333333,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.0650390625,
+      "learning_rate": 9.895582486086592e-05,
+      "loss": 6.2697,
+      "loss/crossentropy": 1.1537235230207443,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14999711327254772,
+      "step": 392
+    },
+    {
+      "epoch": 0.0655,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.0872395833333335,
+      "learning_rate": 9.89504957691265e-05,
+      "loss": 7.4023,
+      "loss/crossentropy": 1.409236192703247,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1763436421751976,
+      "step": 393
+    },
+    {
+      "epoch": 0.06566666666666666,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.9322916666666667,
+      "learning_rate": 9.894515325733103e-05,
+      "loss": 6.6669,
+      "loss/crossentropy": 0.7334737330675125,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12552095763385296,
+      "step": 394
+    },
+    {
+      "epoch": 0.06583333333333333,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.1947265625,
+      "learning_rate": 9.893979732694421e-05,
+      "loss": 6.563,
+      "loss/crossentropy": 1.4535393714904785,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14764047786593437,
+      "step": 395
+    },
+    {
+      "epoch": 0.066,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.183333333333333,
+      "learning_rate": 9.89344279794344e-05,
+      "loss": 7.1506,
+      "loss/crossentropy": 2.3290280997753143,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1923549324274063,
+      "step": 396
+    },
+    {
+      "epoch": 0.06616666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.5864583333333333,
+      "learning_rate": 9.892904521627361e-05,
+      "loss": 7.2726,
+      "loss/crossentropy": 1.7218973338603973,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2068917229771614,
+      "step": 397
+    },
+    {
+      "epoch": 0.06633333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.2577473958333334,
+      "learning_rate": 9.892364903893759e-05,
+      "loss": 6.9218,
+      "loss/crossentropy": 1.2974925339221954,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15329620987176895,
+      "step": 398
+    },
+    {
+      "epoch": 0.0665,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.2150390625,
+      "learning_rate": 9.891823944890568e-05,
+      "loss": 7.1408,
+      "loss/crossentropy": 1.6945496201515198,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1738334558904171,
+      "step": 399
+    },
+    {
+      "epoch": 0.06666666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.4018229166666667,
+      "learning_rate": 9.8912816447661e-05,
+      "loss": 7.0689,
+      "loss/crossentropy": 1.1989670246839523,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17318839952349663,
+      "step": 400
+    },
+    {
+      "epoch": 0.06683333333333333,
+      "grad_norm": 37.5,
+      "grad_norm_var": 5.3,
+      "learning_rate": 9.890738003669029e-05,
+      "loss": 7.0358,
+      "loss/crossentropy": 1.6337324529886246,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25416150875389576,
+      "step": 401
+    },
+    {
+      "epoch": 0.067,
+      "grad_norm": 31.75,
+      "grad_norm_var": 5.27265625,
+      "learning_rate": 9.890193021748395e-05,
+      "loss": 7.1105,
+      "loss/crossentropy": 1.3696629405021667,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23377813398838043,
+      "step": 402
+    },
+    {
+      "epoch": 0.06716666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 5.294791666666667,
+      "learning_rate": 9.88964669915361e-05,
+      "loss": 7.3575,
+      "loss/crossentropy": 0.8983005583286285,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17355326376855373,
+      "step": 403
+    },
+    {
+      "epoch": 0.06733333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.294791666666667,
+      "learning_rate": 9.889099036034451e-05,
+      "loss": 7.4992,
+      "loss/crossentropy": 1.9310307800769806,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17338347807526588,
+      "step": 404
+    },
+    {
+      "epoch": 0.0675,
+      "grad_norm": 33.5,
+      "grad_norm_var": 5.692122395833334,
+      "learning_rate": 9.888550032541059e-05,
+      "loss": 7.5245,
+      "loss/crossentropy": 1.7606293559074402,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15558410063385963,
+      "step": 405
+    },
+    {
+      "epoch": 0.06766666666666667,
+      "grad_norm": 44.0,
+      "grad_norm_var": 15.885872395833333,
+      "learning_rate": 9.887999688823955e-05,
+      "loss": 6.9123,
+      "loss/crossentropy": 1.4918632507324219,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20198907144367695,
+      "step": 406
+    },
+    {
+      "epoch": 0.06783333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 15.82890625,
+      "learning_rate": 9.88744800503401e-05,
+      "loss": 6.3646,
+      "loss/crossentropy": 1.2554514855146408,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17770635895431042,
+      "step": 407
+    },
+    {
+      "epoch": 0.068,
+      "grad_norm": 29.125,
+      "grad_norm_var": 15.3056640625,
+      "learning_rate": 9.886894981322476e-05,
+      "loss": 6.7323,
+      "loss/crossentropy": 1.3313074856996536,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13258123397827148,
+      "step": 408
+    },
+    {
+      "epoch": 0.06816666666666667,
+      "grad_norm": 29.125,
+      "grad_norm_var": 15.84375,
+      "learning_rate": 9.886340617840968e-05,
+      "loss": 6.8192,
+      "loss/crossentropy": 1.4146481305360794,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15282718185335398,
+      "step": 409
+    },
+    {
+      "epoch": 0.06833333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 15.29140625,
+      "learning_rate": 9.885784914741465e-05,
+      "loss": 7.2462,
+      "loss/crossentropy": 1.262852057814598,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1464645303785801,
+      "step": 410
+    },
+    {
+      "epoch": 0.0685,
+      "grad_norm": 31.875,
+      "grad_norm_var": 14.138997395833334,
+      "learning_rate": 9.88522787217632e-05,
+      "loss": 6.7425,
+      "loss/crossentropy": 1.6671670973300934,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2631005682051182,
+      "step": 411
+    },
+    {
+      "epoch": 0.06866666666666667,
+      "grad_norm": 27.375,
+      "grad_norm_var": 15.653059895833334,
+      "learning_rate": 9.884669490298244e-05,
+      "loss": 6.8382,
+      "loss/crossentropy": 1.6557044088840485,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.140958396717906,
+      "step": 412
+    },
+    {
+      "epoch": 0.06883333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 15.563997395833333,
+      "learning_rate": 9.884109769260325e-05,
+      "loss": 7.6609,
+      "loss/crossentropy": 0.9135521650314331,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19201525300741196,
+      "step": 413
+    },
+    {
+      "epoch": 0.069,
+      "grad_norm": 30.5,
+      "grad_norm_var": 15.563997395833333,
+      "learning_rate": 9.883548709216013e-05,
+      "loss": 6.6332,
+      "loss/crossentropy": 1.3997588455677032,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13156344182789326,
+      "step": 414
+    },
+    {
+      "epoch": 0.06916666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 15.512239583333333,
+      "learning_rate": 9.882986310319124e-05,
+      "loss": 7.1636,
+      "loss/crossentropy": 1.7076705694198608,
+      "loss/hidden": 3.92578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25658633559942245,
+      "step": 415
+    },
+    {
+      "epoch": 0.06933333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 15.186393229166667,
+      "learning_rate": 9.882422572723844e-05,
+      "loss": 6.8606,
+      "loss/crossentropy": 1.3504041358828545,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2513937735930085,
+      "step": 416
+    },
+    {
+      "epoch": 0.0695,
+      "grad_norm": 29.375,
+      "grad_norm_var": 13.497916666666667,
+      "learning_rate": 9.881857496584726e-05,
+      "loss": 6.7591,
+      "loss/crossentropy": 1.1982027888298035,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1371591743081808,
+      "step": 417
+    },
+    {
+      "epoch": 0.06966666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 13.5931640625,
+      "learning_rate": 9.881291082056685e-05,
+      "loss": 6.6236,
+      "loss/crossentropy": 1.4953482449054718,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2729326821863651,
+      "step": 418
+    },
+    {
+      "epoch": 0.06983333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 13.4759765625,
+      "learning_rate": 9.880723329295012e-05,
+      "loss": 6.9131,
+      "loss/crossentropy": 1.4245893955230713,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15015526488423347,
+      "step": 419
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 30.375,
+      "grad_norm_var": 13.561458333333333,
+      "learning_rate": 9.880154238455356e-05,
+      "loss": 6.7411,
+      "loss/crossentropy": 1.4112870395183563,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16381575912237167,
+      "step": 420
+    },
+    {
+      "epoch": 0.07016666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 13.303059895833334,
+      "learning_rate": 9.879583809693738e-05,
+      "loss": 7.1319,
+      "loss/crossentropy": 1.6169283092021942,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14410164579749107,
+      "step": 421
+    },
+    {
+      "epoch": 0.07033333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.9593098958333333,
+      "learning_rate": 9.879012043166542e-05,
+      "loss": 6.8012,
+      "loss/crossentropy": 1.3054140955209732,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15814785845577717,
+      "step": 422
+    },
+    {
+      "epoch": 0.0705,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.8728515625,
+      "learning_rate": 9.878438939030526e-05,
+      "loss": 6.5904,
+      "loss/crossentropy": 1.4737870246171951,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17665542662143707,
+      "step": 423
+    },
+    {
+      "epoch": 0.07066666666666667,
+      "grad_norm": 26.375,
+      "grad_norm_var": 2.8296223958333333,
+      "learning_rate": 9.877864497442804e-05,
+      "loss": 6.497,
+      "loss/crossentropy": 1.4228816032409668,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19815989956259727,
+      "step": 424
+    },
+    {
+      "epoch": 0.07083333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.7583333333333333,
+      "learning_rate": 9.877288718560866e-05,
+      "loss": 7.0911,
+      "loss/crossentropy": 2.074735075235367,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24289842694997787,
+      "step": 425
+    },
+    {
+      "epoch": 0.071,
+      "grad_norm": 27.75,
+      "grad_norm_var": 3.162239583333333,
+      "learning_rate": 9.876711602542563e-05,
+      "loss": 7.1676,
+      "loss/crossentropy": 1.2838065922260284,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24612978007644415,
+      "step": 426
+    },
+    {
+      "epoch": 0.07116666666666667,
+      "grad_norm": 28.0,
+      "grad_norm_var": 3.2046223958333333,
+      "learning_rate": 9.876133149546118e-05,
+      "loss": 6.6919,
+      "loss/crossentropy": 1.8741240799427032,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16576944291591644,
+      "step": 427
+    },
+    {
+      "epoch": 0.07133333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.7639973958333335,
+      "learning_rate": 9.875553359730114e-05,
+      "loss": 6.7182,
+      "loss/crossentropy": 1.4090934842824936,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15427778661251068,
+      "step": 428
+    },
+    {
+      "epoch": 0.0715,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.8499348958333335,
+      "learning_rate": 9.874972233253504e-05,
+      "loss": 6.7471,
+      "loss/crossentropy": 1.103910744190216,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14564213017001748,
+      "step": 429
+    },
+    {
+      "epoch": 0.07166666666666667,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.0218098958333335,
+      "learning_rate": 9.874389770275607e-05,
+      "loss": 7.2498,
+      "loss/crossentropy": 1.8716143667697906,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.289576917886734,
+      "step": 430
+    },
+    {
+      "epoch": 0.07183333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.6025390625,
+      "learning_rate": 9.87380597095611e-05,
+      "loss": 7.1361,
+      "loss/crossentropy": 1.1144326478242874,
+      "loss/hidden": 3.10546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12110078148543835,
+      "step": 431
+    },
+    {
+      "epoch": 0.072,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.6025390625,
+      "learning_rate": 9.873220835455064e-05,
+      "loss": 6.771,
+      "loss/crossentropy": 1.6715168803930283,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13846534118056297,
+      "step": 432
+    },
+    {
+      "epoch": 0.07216666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.66640625,
+      "learning_rate": 9.872634363932887e-05,
+      "loss": 7.3403,
+      "loss/crossentropy": 1.8598685264587402,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17878054454922676,
+      "step": 433
+    },
+    {
+      "epoch": 0.07233333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.81640625,
+      "learning_rate": 9.872046556550363e-05,
+      "loss": 6.9061,
+      "loss/crossentropy": 1.5577319860458374,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.150806475430727,
+      "step": 434
+    },
+    {
+      "epoch": 0.0725,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.7301432291666665,
+      "learning_rate": 9.871457413468644e-05,
+      "loss": 6.7873,
+      "loss/crossentropy": 1.6065486669540405,
+      "loss/hidden": 3.09765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1471869796514511,
+      "step": 435
+    },
+    {
+      "epoch": 0.07266666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.738541666666667,
+      "learning_rate": 9.870866934849248e-05,
+      "loss": 7.392,
+      "loss/crossentropy": 1.7273681610822678,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2536742892116308,
+      "step": 436
+    },
+    {
+      "epoch": 0.07283333333333333,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.887955729166667,
+      "learning_rate": 9.870275120854054e-05,
+      "loss": 6.7686,
+      "loss/crossentropy": 2.0005056262016296,
+      "loss/hidden": 3.81640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23111211135983467,
+      "step": 437
+    },
+    {
+      "epoch": 0.073,
+      "grad_norm": 29.25,
+      "grad_norm_var": 3.9535807291666667,
+      "learning_rate": 9.869681971645315e-05,
+      "loss": 7.4992,
+      "loss/crossentropy": 1.8039406090974808,
+      "loss/hidden": 3.7578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28976285830140114,
+      "step": 438
+    },
+    {
+      "epoch": 0.07316666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.9416015625,
+      "learning_rate": 9.869087487385644e-05,
+      "loss": 6.8613,
+      "loss/crossentropy": 1.982006013393402,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1841346099972725,
+      "step": 439
+    },
+    {
+      "epoch": 0.07333333333333333,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.1166015625,
+      "learning_rate": 9.868491668238025e-05,
+      "loss": 6.8652,
+      "loss/crossentropy": 1.5234146118164062,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17372388020157814,
+      "step": 440
+    },
+    {
+      "epoch": 0.0735,
+      "grad_norm": 34.25,
+      "grad_norm_var": 4.0400390625,
+      "learning_rate": 9.867894514365802e-05,
+      "loss": 7.6485,
+      "loss/crossentropy": 1.2764847427606583,
+      "loss/hidden": 3.7421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2595882900059223,
+      "step": 441
+    },
+    {
+      "epoch": 0.07366666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.7666015625,
+      "learning_rate": 9.867296025932688e-05,
+      "loss": 7.3161,
+      "loss/crossentropy": 1.551239237189293,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16433005779981613,
+      "step": 442
+    },
+    {
+      "epoch": 0.07383333333333333,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.6728515625,
+      "learning_rate": 9.866696203102766e-05,
+      "loss": 7.177,
+      "loss/crossentropy": 1.9006485342979431,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18515821173787117,
+      "step": 443
+    },
+    {
+      "epoch": 0.074,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.6958333333333333,
+      "learning_rate": 9.866095046040478e-05,
+      "loss": 6.5369,
+      "loss/crossentropy": 1.1699100732803345,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16365684941411018,
+      "step": 444
+    },
+    {
+      "epoch": 0.07416666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.76015625,
+      "learning_rate": 9.865492554910633e-05,
+      "loss": 7.1591,
+      "loss/crossentropy": 1.692724883556366,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18517207354307175,
+      "step": 445
+    },
+    {
+      "epoch": 0.07433333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.790625,
+      "learning_rate": 9.86488872987841e-05,
+      "loss": 7.1471,
+      "loss/crossentropy": 1.5410718321800232,
+      "loss/hidden": 4.01171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.43903179466724396,
+      "step": 446
+    },
+    {
+      "epoch": 0.0745,
+      "grad_norm": 29.125,
+      "grad_norm_var": 4.0025390625,
+      "learning_rate": 9.864283571109352e-05,
+      "loss": 6.5643,
+      "loss/crossentropy": 1.3042651116847992,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13341313041746616,
+      "step": 447
+    },
+    {
+      "epoch": 0.07466666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.0712890625,
+      "learning_rate": 9.863677078769362e-05,
+      "loss": 7.4753,
+      "loss/crossentropy": 1.3810160607099533,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23434478230774403,
+      "step": 448
+    },
+    {
+      "epoch": 0.07483333333333334,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.5353515625,
+      "learning_rate": 9.863069253024719e-05,
+      "loss": 7.2343,
+      "loss/crossentropy": 1.5376935005187988,
+      "loss/hidden": 3.625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23124074935913086,
+      "step": 449
+    },
+    {
+      "epoch": 0.075,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.4400390625,
+      "learning_rate": 9.862460094042056e-05,
+      "loss": 7.0098,
+      "loss/crossentropy": 1.751096397638321,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1434621810913086,
+      "step": 450
+    },
+    {
+      "epoch": 0.07516666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.440625,
+      "learning_rate": 9.861849601988383e-05,
+      "loss": 6.6323,
+      "loss/crossentropy": 1.1928651630878448,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16096126288175583,
+      "step": 451
+    },
+    {
+      "epoch": 0.07533333333333334,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.7583333333333333,
+      "learning_rate": 9.861237777031068e-05,
+      "loss": 7.0513,
+      "loss/crossentropy": 1.9867673516273499,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16949284449219704,
+      "step": 452
+    },
+    {
+      "epoch": 0.0755,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.5712890625,
+      "learning_rate": 9.860624619337844e-05,
+      "loss": 6.9948,
+      "loss/crossentropy": 1.2133909314870834,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18272900208830833,
+      "step": 453
+    },
+    {
+      "epoch": 0.07566666666666666,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.6979166666666665,
+      "learning_rate": 9.860010129076813e-05,
+      "loss": 6.8924,
+      "loss/crossentropy": 1.5619240701198578,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15395177155733109,
+      "step": 454
+    },
+    {
+      "epoch": 0.07583333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.7223307291666665,
+      "learning_rate": 9.859394306416444e-05,
+      "loss": 7.0382,
+      "loss/crossentropy": 1.4500148221850395,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19516615942120552,
+      "step": 455
+    },
+    {
+      "epoch": 0.076,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.9559895833333334,
+      "learning_rate": 9.858777151525564e-05,
+      "loss": 7.3597,
+      "loss/crossentropy": 1.2372301891446114,
+      "loss/hidden": 3.7109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.223042793571949,
+      "step": 456
+    },
+    {
+      "epoch": 0.07616666666666666,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.2733723958333334,
+      "learning_rate": 9.85815866457337e-05,
+      "loss": 6.7725,
+      "loss/crossentropy": 1.7923309803009033,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2751024030148983,
+      "step": 457
+    },
+    {
+      "epoch": 0.07633333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.787239583333333,
+      "learning_rate": 9.857538845729426e-05,
+      "loss": 6.6334,
+      "loss/crossentropy": 1.5725050419569016,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15925869904458523,
+      "step": 458
+    },
+    {
+      "epoch": 0.0765,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.5160807291666667,
+      "learning_rate": 9.856917695163658e-05,
+      "loss": 6.9434,
+      "loss/crossentropy": 1.7009003162384033,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16625344939529896,
+      "step": 459
+    },
+    {
+      "epoch": 0.07666666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.512434895833333,
+      "learning_rate": 9.856295213046357e-05,
+      "loss": 6.8745,
+      "loss/crossentropy": 1.5608884245157242,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2525358907878399,
+      "step": 460
+    },
+    {
+      "epoch": 0.07683333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.0747395833333333,
+      "learning_rate": 9.855671399548181e-05,
+      "loss": 6.8753,
+      "loss/crossentropy": 1.5073769092559814,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20055905729532242,
+      "step": 461
+    },
+    {
+      "epoch": 0.077,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.9,
+      "learning_rate": 9.855046254840151e-05,
+      "loss": 6.6672,
+      "loss/crossentropy": 1.9303827285766602,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20459270849823952,
+      "step": 462
+    },
+    {
+      "epoch": 0.07716666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.8384765625,
+      "learning_rate": 9.854419779093655e-05,
+      "loss": 6.918,
+      "loss/crossentropy": 1.4022010415792465,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14251694455742836,
+      "step": 463
+    },
+    {
+      "epoch": 0.07733333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.4431640625,
+      "learning_rate": 9.853791972480445e-05,
+      "loss": 6.733,
+      "loss/crossentropy": 1.7563574463129044,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1708853840827942,
+      "step": 464
+    },
+    {
+      "epoch": 0.0775,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.6212890625,
+      "learning_rate": 9.853162835172637e-05,
+      "loss": 6.8478,
+      "loss/crossentropy": 1.4509409219026566,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14121680706739426,
+      "step": 465
+    },
+    {
+      "epoch": 0.07766666666666666,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.6268229166666666,
+      "learning_rate": 9.852532367342713e-05,
+      "loss": 7.066,
+      "loss/crossentropy": 2.1067994236946106,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24858925491571426,
+      "step": 466
+    },
+    {
+      "epoch": 0.07783333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.84765625,
+      "learning_rate": 9.851900569163519e-05,
+      "loss": 7.0774,
+      "loss/crossentropy": 1.1014457046985626,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1381321232765913,
+      "step": 467
+    },
+    {
+      "epoch": 0.078,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.5684895833333334,
+      "learning_rate": 9.851267440808265e-05,
+      "loss": 6.965,
+      "loss/crossentropy": 1.9848509430885315,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15665691532194614,
+      "step": 468
+    },
+    {
+      "epoch": 0.07816666666666666,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.5780598958333334,
+      "learning_rate": 9.85063298245053e-05,
+      "loss": 7.4537,
+      "loss/crossentropy": 1.7907701879739761,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3415554091334343,
+      "step": 469
+    },
+    {
+      "epoch": 0.07833333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.7489583333333334,
+      "learning_rate": 9.84999719426425e-05,
+      "loss": 7.0107,
+      "loss/crossentropy": 1.4050543904304504,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2011774256825447,
+      "step": 470
+    },
+    {
+      "epoch": 0.0785,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.8747395833333333,
+      "learning_rate": 9.849360076423734e-05,
+      "loss": 6.8379,
+      "loss/crossentropy": 1.291910782456398,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23814310878515244,
+      "step": 471
+    },
+    {
+      "epoch": 0.07866666666666666,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.6125,
+      "learning_rate": 9.84872162910365e-05,
+      "loss": 7.0224,
+      "loss/crossentropy": 1.6638221144676208,
+      "loss/hidden": 3.80078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2528918646275997,
+      "step": 472
+    },
+    {
+      "epoch": 0.07883333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.440625,
+      "learning_rate": 9.84808185247903e-05,
+      "loss": 7.0542,
+      "loss/crossentropy": 1.4460849463939667,
+      "loss/hidden": 3.6484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24226966872811317,
+      "step": 473
+    },
+    {
+      "epoch": 0.079,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.116080729166667,
+      "learning_rate": 9.847440746725275e-05,
+      "loss": 6.921,
+      "loss/crossentropy": 1.8123798668384552,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.191888689994812,
+      "step": 474
+    },
+    {
+      "epoch": 0.07916666666666666,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.6979166666666665,
+      "learning_rate": 9.846798312018146e-05,
+      "loss": 7.5232,
+      "loss/crossentropy": 1.6415761709213257,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24300210550427437,
+      "step": 475
+    },
+    {
+      "epoch": 0.07933333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.725455729166667,
+      "learning_rate": 9.846154548533773e-05,
+      "loss": 7.0933,
+      "loss/crossentropy": 1.9737498462200165,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20842399820685387,
+      "step": 476
+    },
+    {
+      "epoch": 0.0795,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.8643229166666666,
+      "learning_rate": 9.845509456448643e-05,
+      "loss": 6.9684,
+      "loss/crossentropy": 1.5572169423103333,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19147998467087746,
+      "step": 477
+    },
+    {
+      "epoch": 0.07966666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.6645182291666667,
+      "learning_rate": 9.844863035939615e-05,
+      "loss": 6.9128,
+      "loss/crossentropy": 1.906806856393814,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15526464208960533,
+      "step": 478
+    },
+    {
+      "epoch": 0.07983333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.76640625,
+      "learning_rate": 9.844215287183909e-05,
+      "loss": 7.2578,
+      "loss/crossentropy": 1.443288266658783,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16811441257596016,
+      "step": 479
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.674934895833333,
+      "learning_rate": 9.843566210359106e-05,
+      "loss": 7.1386,
+      "loss/crossentropy": 2.162033826112747,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22076943516731262,
+      "step": 480
+    },
+    {
+      "epoch": 0.08016666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.943489583333333,
+      "learning_rate": 9.842915805643155e-05,
+      "loss": 6.9895,
+      "loss/crossentropy": 1.6815131306648254,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1633237786591053,
+      "step": 481
+    },
+    {
+      "epoch": 0.08033333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.8374348958333333,
+      "learning_rate": 9.842264073214371e-05,
+      "loss": 6.6895,
+      "loss/crossentropy": 1.4824548959732056,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20978729613125324,
+      "step": 482
+    },
+    {
+      "epoch": 0.0805,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.912955729166667,
+      "learning_rate": 9.841611013251429e-05,
+      "loss": 6.8164,
+      "loss/crossentropy": 1.4602741301059723,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17982185631990433,
+      "step": 483
+    },
+    {
+      "epoch": 0.08066666666666666,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.0893229166666667,
+      "learning_rate": 9.840956625933367e-05,
+      "loss": 6.8987,
+      "loss/crossentropy": 1.7982778251171112,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1877780705690384,
+      "step": 484
+    },
+    {
+      "epoch": 0.08083333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 3.4666015625,
+      "learning_rate": 9.840300911439591e-05,
+      "loss": 6.735,
+      "loss/crossentropy": 1.3495407849550247,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15540699288249016,
+      "step": 485
+    },
+    {
+      "epoch": 0.081,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.9509765625,
+      "learning_rate": 9.839643869949866e-05,
+      "loss": 7.5219,
+      "loss/crossentropy": 1.48193821310997,
+      "loss/hidden": 3.77734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2577380947768688,
+      "step": 486
+    },
+    {
+      "epoch": 0.08116666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.9509765625,
+      "learning_rate": 9.838985501644328e-05,
+      "loss": 6.7139,
+      "loss/crossentropy": 1.7966694235801697,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1714860461652279,
+      "step": 487
+    },
+    {
+      "epoch": 0.08133333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 3.7603515625,
+      "learning_rate": 9.83832580670347e-05,
+      "loss": 7.1478,
+      "loss/crossentropy": 1.7598281800746918,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23178521916270256,
+      "step": 488
+    },
+    {
+      "epoch": 0.0815,
+      "grad_norm": 27.0,
+      "grad_norm_var": 4.64765625,
+      "learning_rate": 9.837664785308149e-05,
+      "loss": 6.5713,
+      "loss/crossentropy": 1.6064825356006622,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1570918019860983,
+      "step": 489
+    },
+    {
+      "epoch": 0.08166666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.9181640625,
+      "learning_rate": 9.837002437639593e-05,
+      "loss": 6.8044,
+      "loss/crossentropy": 1.4402455165982246,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23276809975504875,
+      "step": 490
+    },
+    {
+      "epoch": 0.08183333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.1497395833333335,
+      "learning_rate": 9.836338763879385e-05,
+      "loss": 6.35,
+      "loss/crossentropy": 1.475183218717575,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15859464183449745,
+      "step": 491
+    },
+    {
+      "epoch": 0.082,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.0426432291666665,
+      "learning_rate": 9.835673764209474e-05,
+      "loss": 7.1277,
+      "loss/crossentropy": 1.522533044219017,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31567465886473656,
+      "step": 492
+    },
+    {
+      "epoch": 0.08216666666666667,
+      "grad_norm": 27.0,
+      "grad_norm_var": 2.8988932291666667,
+      "learning_rate": 9.835007438812177e-05,
+      "loss": 6.6873,
+      "loss/crossentropy": 2.2898610830307007,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1842830590903759,
+      "step": 493
+    },
+    {
+      "epoch": 0.08233333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.486458333333333,
+      "learning_rate": 9.834339787870166e-05,
+      "loss": 7.2161,
+      "loss/crossentropy": 1.3005068749189377,
+      "loss/hidden": 3.83203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24435850605368614,
+      "step": 494
+    },
+    {
+      "epoch": 0.0825,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.5759765625,
+      "learning_rate": 9.833670811566485e-05,
+      "loss": 6.838,
+      "loss/crossentropy": 1.2712122946977615,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13630807399749756,
+      "step": 495
+    },
+    {
+      "epoch": 0.08266666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.596809895833333,
+      "learning_rate": 9.833000510084537e-05,
+      "loss": 6.9043,
+      "loss/crossentropy": 1.7596306204795837,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14436211995780468,
+      "step": 496
+    },
+    {
+      "epoch": 0.08283333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 3.5552083333333333,
+      "learning_rate": 9.832328883608088e-05,
+      "loss": 6.6913,
+      "loss/crossentropy": 1.151959091424942,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13815506733953953,
+      "step": 497
+    },
+    {
+      "epoch": 0.083,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.5155598958333334,
+      "learning_rate": 9.83165593232127e-05,
+      "loss": 6.9965,
+      "loss/crossentropy": 1.3583546876907349,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16475296951830387,
+      "step": 498
+    },
+    {
+      "epoch": 0.08316666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.51015625,
+      "learning_rate": 9.830981656408574e-05,
+      "loss": 6.7634,
+      "loss/crossentropy": 1.8273972570896149,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17411424219608307,
+      "step": 499
+    },
+    {
+      "epoch": 0.08333333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.480208333333333,
+      "learning_rate": 9.830306056054858e-05,
+      "loss": 6.4987,
+      "loss/crossentropy": 1.2338118851184845,
+      "loss/hidden": 3.75,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30939680337905884,
+      "step": 500
+    },
+    {
+      "epoch": 0.0835,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.3052083333333333,
+      "learning_rate": 9.829629131445342e-05,
+      "loss": 7.0907,
+      "loss/crossentropy": 1.6119868755340576,
+      "loss/hidden": 3.8203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2798030972480774,
+      "step": 501
+    },
+    {
+      "epoch": 0.08366666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.1184895833333335,
+      "learning_rate": 9.828950882765608e-05,
+      "loss": 6.2711,
+      "loss/crossentropy": 1.2915617376565933,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13018281571567059,
+      "step": 502
+    },
+    {
+      "epoch": 0.08383333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.1837890625,
+      "learning_rate": 9.828271310201601e-05,
+      "loss": 7.0456,
+      "loss/crossentropy": 1.2496111392974854,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14296716824173927,
+      "step": 503
+    },
+    {
+      "epoch": 0.084,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.1947265625,
+      "learning_rate": 9.827590413939632e-05,
+      "loss": 6.7848,
+      "loss/crossentropy": 1.1799578815698624,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21814516186714172,
+      "step": 504
+    },
+    {
+      "epoch": 0.08416666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.8910807291666667,
+      "learning_rate": 9.82690819416637e-05,
+      "loss": 7.0063,
+      "loss/crossentropy": 2.227379620075226,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2798748090863228,
+      "step": 505
+    },
+    {
+      "epoch": 0.08433333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.8910807291666667,
+      "learning_rate": 9.826224651068852e-05,
+      "loss": 6.9504,
+      "loss/crossentropy": 1.7532093524932861,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20848191529512405,
+      "step": 506
+    },
+    {
+      "epoch": 0.0845,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.8181640625,
+      "learning_rate": 9.825539784834472e-05,
+      "loss": 7.0495,
+      "loss/crossentropy": 1.6463024020195007,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18096379563212395,
+      "step": 507
+    },
+    {
+      "epoch": 0.08466666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.8697265625,
+      "learning_rate": 9.824853595650991e-05,
+      "loss": 6.4,
+      "loss/crossentropy": 1.6303430795669556,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16679994016885757,
+      "step": 508
+    },
+    {
+      "epoch": 0.08483333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.34765625,
+      "learning_rate": 9.824166083706534e-05,
+      "loss": 6.1906,
+      "loss/crossentropy": 1.5004529505968094,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18870088830590248,
+      "step": 509
+    },
+    {
+      "epoch": 0.085,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.1639973958333334,
+      "learning_rate": 9.823477249189586e-05,
+      "loss": 6.8114,
+      "loss/crossentropy": 1.3908627331256866,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15263560600578785,
+      "step": 510
+    },
+    {
+      "epoch": 0.08516666666666667,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.3622395833333334,
+      "learning_rate": 9.822787092288991e-05,
+      "loss": 6.8574,
+      "loss/crossentropy": 2.0620186924934387,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27069373801350594,
+      "step": 511
+    },
+    {
+      "epoch": 0.08533333333333333,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.4686848958333334,
+      "learning_rate": 9.822095613193962e-05,
+      "loss": 7.0702,
+      "loss/crossentropy": 1.7053555250167847,
+      "loss/hidden": 3.12890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1921213436871767,
+      "step": 512
+    },
+    {
+      "epoch": 0.0855,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.42265625,
+      "learning_rate": 9.821402812094073e-05,
+      "loss": 6.8986,
+      "loss/crossentropy": 1.9131797850131989,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14682536479085684,
+      "step": 513
+    },
+    {
+      "epoch": 0.08566666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.7052083333333334,
+      "learning_rate": 9.820708689179259e-05,
+      "loss": 6.9649,
+      "loss/crossentropy": 1.5448258519172668,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14751841872930527,
+      "step": 514
+    },
+    {
+      "epoch": 0.08583333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.7052083333333334,
+      "learning_rate": 9.820013244639816e-05,
+      "loss": 6.9054,
+      "loss/crossentropy": 0.8975181132555008,
+      "loss/hidden": 3.09765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09918954316526651,
+      "step": 515
+    },
+    {
+      "epoch": 0.086,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.9431640625,
+      "learning_rate": 9.819316478666405e-05,
+      "loss": 6.7323,
+      "loss/crossentropy": 1.1570946723222733,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17046009004116058,
+      "step": 516
+    },
+    {
+      "epoch": 0.08616666666666667,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.090625,
+      "learning_rate": 9.81861839145005e-05,
+      "loss": 6.6729,
+      "loss/crossentropy": 1.2703944444656372,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17309994250535965,
+      "step": 517
+    },
+    {
+      "epoch": 0.08633333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.3309895833333334,
+      "learning_rate": 9.817918983182132e-05,
+      "loss": 7.2607,
+      "loss/crossentropy": 1.4812899976968765,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20999731495976448,
+      "step": 518
+    },
+    {
+      "epoch": 0.0865,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.3124348958333334,
+      "learning_rate": 9.8172182540544e-05,
+      "loss": 7.0058,
+      "loss/crossentropy": 1.1661194860935211,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1877425778657198,
+      "step": 519
+    },
+    {
+      "epoch": 0.08666666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.3583333333333334,
+      "learning_rate": 9.816516204258963e-05,
+      "loss": 6.9387,
+      "loss/crossentropy": 1.927243947982788,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16112736612558365,
+      "step": 520
+    },
+    {
+      "epoch": 0.08683333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.8994140625,
+      "learning_rate": 9.815812833988291e-05,
+      "loss": 6.8382,
+      "loss/crossentropy": 1.5174863189458847,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17156189493834972,
+      "step": 521
+    },
+    {
+      "epoch": 0.087,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.890625,
+      "learning_rate": 9.815108143435218e-05,
+      "loss": 6.6661,
+      "loss/crossentropy": 1.4303490817546844,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21909217163920403,
+      "step": 522
+    },
+    {
+      "epoch": 0.08716666666666667,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.9139973958333334,
+      "learning_rate": 9.814402132792939e-05,
+      "loss": 7.4279,
+      "loss/crossentropy": 1.4629042595624924,
+      "loss/hidden": 3.74609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19645161274820566,
+      "step": 523
+    },
+    {
+      "epoch": 0.08733333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.9434895833333334,
+      "learning_rate": 9.81369480225501e-05,
+      "loss": 7.3962,
+      "loss/crossentropy": 1.8141767233610153,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2287716306746006,
+      "step": 524
+    },
+    {
+      "epoch": 0.0875,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.0233723958333334,
+      "learning_rate": 9.812986152015348e-05,
+      "loss": 6.9655,
+      "loss/crossentropy": 1.3640852272510529,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14019519835710526,
+      "step": 525
+    },
+    {
+      "epoch": 0.08766666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.99375,
+      "learning_rate": 9.812276182268236e-05,
+      "loss": 7.3055,
+      "loss/crossentropy": 1.4165369719266891,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15033666044473648,
+      "step": 526
+    },
+    {
+      "epoch": 0.08783333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.4056640625,
+      "learning_rate": 9.811564893208318e-05,
+      "loss": 6.8433,
+      "loss/crossentropy": 1.1786036044359207,
+      "loss/hidden": 3.8359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16844666376709938,
+      "step": 527
+    },
+    {
+      "epoch": 0.088,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.5113932291666665,
+      "learning_rate": 9.810852285030593e-05,
+      "loss": 6.9332,
+      "loss/crossentropy": 1.6334620118141174,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21338831633329391,
+      "step": 528
+    },
+    {
+      "epoch": 0.08816666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.5113932291666665,
+      "learning_rate": 9.81013835793043e-05,
+      "loss": 6.9617,
+      "loss/crossentropy": 1.8668213188648224,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1473287157714367,
+      "step": 529
+    },
+    {
+      "epoch": 0.08833333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.3598307291666667,
+      "learning_rate": 9.809423112103554e-05,
+      "loss": 7.1205,
+      "loss/crossentropy": 1.5200925767421722,
+      "loss/hidden": 3.76953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22000214457511902,
+      "step": 530
+    },
+    {
+      "epoch": 0.0885,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.3275390625,
+      "learning_rate": 9.808706547746057e-05,
+      "loss": 6.9948,
+      "loss/crossentropy": 1.4339255094528198,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16505714319646358,
+      "step": 531
+    },
+    {
+      "epoch": 0.08866666666666667,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.403059895833333,
+      "learning_rate": 9.807988665054386e-05,
+      "loss": 6.8232,
+      "loss/crossentropy": 2.044094979763031,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17992260307073593,
+      "step": 532
+    },
+    {
+      "epoch": 0.08883333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.1660807291666666,
+      "learning_rate": 9.807269464225355e-05,
+      "loss": 7.3613,
+      "loss/crossentropy": 1.8550164997577667,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18196892365813255,
+      "step": 533
+    },
+    {
+      "epoch": 0.089,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.9785807291666666,
+      "learning_rate": 9.806548945456134e-05,
+      "loss": 6.6134,
+      "loss/crossentropy": 1.6432620882987976,
+      "loss/hidden": 3.0234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13885372690856457,
+      "step": 534
+    },
+    {
+      "epoch": 0.08916666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.959375,
+      "learning_rate": 9.80582710894426e-05,
+      "loss": 6.7517,
+      "loss/crossentropy": 1.702562391757965,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17485880386084318,
+      "step": 535
+    },
+    {
+      "epoch": 0.08933333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.9541015625,
+      "learning_rate": 9.805103954887627e-05,
+      "loss": 7.1787,
+      "loss/crossentropy": 1.9431041926145554,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1696106381714344,
+      "step": 536
+    },
+    {
+      "epoch": 0.0895,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.95,
+      "learning_rate": 9.804379483484494e-05,
+      "loss": 7.5377,
+      "loss/crossentropy": 1.7971654832363129,
+      "loss/hidden": 3.859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30418267473578453,
+      "step": 537
+    },
+    {
+      "epoch": 0.08966666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.190625,
+      "learning_rate": 9.803653694933476e-05,
+      "loss": 7.3215,
+      "loss/crossentropy": 1.9498747289180756,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19328440725803375,
+      "step": 538
+    },
+    {
+      "epoch": 0.08983333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.3046223958333334,
+      "learning_rate": 9.802926589433553e-05,
+      "loss": 6.5449,
+      "loss/crossentropy": 1.51683808863163,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20039740949869156,
+      "step": 539
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.349739583333333,
+      "learning_rate": 9.802198167184067e-05,
+      "loss": 6.3769,
+      "loss/crossentropy": 1.4991832077503204,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14776822179555893,
+      "step": 540
+    },
+    {
+      "epoch": 0.09016666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 2.825,
+      "learning_rate": 9.801468428384716e-05,
+      "loss": 6.9081,
+      "loss/crossentropy": 1.2094965279102325,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15462427772581577,
+      "step": 541
+    },
+    {
+      "epoch": 0.09033333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.83515625,
+      "learning_rate": 9.800737373235565e-05,
+      "loss": 6.8153,
+      "loss/crossentropy": 1.5440810024738312,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18239294737577438,
+      "step": 542
+    },
+    {
+      "epoch": 0.0905,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.8884765625,
+      "learning_rate": 9.800005001937034e-05,
+      "loss": 6.9232,
+      "loss/crossentropy": 1.8635855913162231,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21322701126337051,
+      "step": 543
+    },
+    {
+      "epoch": 0.09066666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.7478515625,
+      "learning_rate": 9.799271314689908e-05,
+      "loss": 7.2039,
+      "loss/crossentropy": 1.5228805541992188,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2353954017162323,
+      "step": 544
+    },
+    {
+      "epoch": 0.09083333333333334,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.63515625,
+      "learning_rate": 9.798536311695334e-05,
+      "loss": 6.512,
+      "loss/crossentropy": 1.5062799155712128,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1575067937374115,
+      "step": 545
+    },
+    {
+      "epoch": 0.091,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.590625,
+      "learning_rate": 9.797799993154814e-05,
+      "loss": 7.3783,
+      "loss/crossentropy": 1.3743330240249634,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14204004779458046,
+      "step": 546
+    },
+    {
+      "epoch": 0.09116666666666666,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.8181640625,
+      "learning_rate": 9.797062359270215e-05,
+      "loss": 6.4564,
+      "loss/crossentropy": 1.2686700522899628,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11861040443181992,
+      "step": 547
+    },
+    {
+      "epoch": 0.09133333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.361458333333333,
+      "learning_rate": 9.796323410243763e-05,
+      "loss": 6.7514,
+      "loss/crossentropy": 1.763284057378769,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16353006660938263,
+      "step": 548
+    },
+    {
+      "epoch": 0.0915,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.361458333333333,
+      "learning_rate": 9.795583146278046e-05,
+      "loss": 7.052,
+      "loss/crossentropy": 1.4608117491006851,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1339084878563881,
+      "step": 549
+    },
+    {
+      "epoch": 0.09166666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.362434895833333,
+      "learning_rate": 9.794841567576011e-05,
+      "loss": 7.4853,
+      "loss/crossentropy": 1.9210692346096039,
+      "loss/hidden": 3.7578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3668474480509758,
+      "step": 550
+    },
+    {
+      "epoch": 0.09183333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.275,
+      "learning_rate": 9.794098674340965e-05,
+      "loss": 7.2024,
+      "loss/crossentropy": 1.3585729598999023,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17329660430550575,
+      "step": 551
+    },
+    {
+      "epoch": 0.092,
+      "grad_norm": 27.0,
+      "grad_norm_var": 3.99140625,
+      "learning_rate": 9.793354466776579e-05,
+      "loss": 6.6092,
+      "loss/crossentropy": 1.609390527009964,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24672024510800838,
+      "step": 552
+    },
+    {
+      "epoch": 0.09216666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.99140625,
+      "learning_rate": 9.79260894508688e-05,
+      "loss": 7.04,
+      "loss/crossentropy": 1.8825242817401886,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19359325990080833,
+      "step": 553
+    },
+    {
+      "epoch": 0.09233333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.853125,
+      "learning_rate": 9.791862109476257e-05,
+      "loss": 6.8277,
+      "loss/crossentropy": 1.3447879999876022,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13356898352503777,
+      "step": 554
+    },
+    {
+      "epoch": 0.0925,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.9625,
+      "learning_rate": 9.791113960149458e-05,
+      "loss": 6.8456,
+      "loss/crossentropy": 1.9514881372451782,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21370084583759308,
+      "step": 555
+    },
+    {
+      "epoch": 0.09266666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.9625,
+      "learning_rate": 9.790364497311597e-05,
+      "loss": 6.7079,
+      "loss/crossentropy": 1.3238005712628365,
+      "loss/hidden": 3.14453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12908729072660208,
+      "step": 556
+    },
+    {
+      "epoch": 0.09283333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.978125,
+      "learning_rate": 9.789613721168139e-05,
+      "loss": 6.6304,
+      "loss/crossentropy": 1.558670699596405,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13811413943767548,
+      "step": 557
+    },
+    {
+      "epoch": 0.093,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.24765625,
+      "learning_rate": 9.788861631924913e-05,
+      "loss": 7.1054,
+      "loss/crossentropy": 2.224619686603546,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21757641062140465,
+      "step": 558
+    },
+    {
+      "epoch": 0.09316666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.9184895833333333,
+      "learning_rate": 9.788108229788111e-05,
+      "loss": 7.1799,
+      "loss/crossentropy": 1.8788108825683594,
+      "loss/hidden": 3.640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18427220731973648,
+      "step": 559
+    },
+    {
+      "epoch": 0.09333333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.8530598958333333,
+      "learning_rate": 9.787353514964284e-05,
+      "loss": 7.0331,
+      "loss/crossentropy": 1.7641720548272133,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19019095040857792,
+      "step": 560
+    },
+    {
+      "epoch": 0.0935,
+      "grad_norm": 26.375,
+      "grad_norm_var": 2.6684895833333333,
+      "learning_rate": 9.786597487660337e-05,
+      "loss": 6.7162,
+      "loss/crossentropy": 1.8242340385913849,
+      "loss/hidden": 3.73046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25750827230513096,
+      "step": 561
+    },
+    {
+      "epoch": 0.09366666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.616080729166667,
+      "learning_rate": 9.785840148083543e-05,
+      "loss": 6.7623,
+      "loss/crossentropy": 1.6919041872024536,
+      "loss/hidden": 3.12890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13569016009569168,
+      "step": 562
+    },
+    {
+      "epoch": 0.09383333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.5764973958333335,
+      "learning_rate": 9.785081496441527e-05,
+      "loss": 7.2596,
+      "loss/crossentropy": 1.72784423828125,
+      "loss/hidden": 3.71484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19006014801561832,
+      "step": 563
+    },
+    {
+      "epoch": 0.094,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.6395833333333334,
+      "learning_rate": 9.784321532942282e-05,
+      "loss": 6.6004,
+      "loss/crossentropy": 1.566937267780304,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1701590158045292,
+      "step": 564
+    },
+    {
+      "epoch": 0.09416666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.5559895833333335,
+      "learning_rate": 9.783560257794154e-05,
+      "loss": 7.1836,
+      "loss/crossentropy": 1.3668065667152405,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1645987294614315,
+      "step": 565
+    },
+    {
+      "epoch": 0.09433333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.7552083333333335,
+      "learning_rate": 9.78279767120585e-05,
+      "loss": 7.1728,
+      "loss/crossentropy": 1.3751739487051964,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1458847690373659,
+      "step": 566
+    },
+    {
+      "epoch": 0.0945,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.5768229166666665,
+      "learning_rate": 9.782033773386439e-05,
+      "loss": 7.1971,
+      "loss/crossentropy": 1.93217933177948,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2505273334681988,
+      "step": 567
+    },
+    {
+      "epoch": 0.09466666666666666,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.075455729166667,
+      "learning_rate": 9.781268564545348e-05,
+      "loss": 6.9163,
+      "loss/crossentropy": 1.3366051390767097,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2988725043833256,
+      "step": 568
+    },
+    {
+      "epoch": 0.09483333333333334,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.1791666666666667,
+      "learning_rate": 9.780502044892362e-05,
+      "loss": 6.5801,
+      "loss/crossentropy": 1.4976085722446442,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14795389771461487,
+      "step": 569
+    },
+    {
+      "epoch": 0.095,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.4457682291666667,
+      "learning_rate": 9.779734214637628e-05,
+      "loss": 6.4807,
+      "loss/crossentropy": 1.67276331782341,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15900277346372604,
+      "step": 570
+    },
+    {
+      "epoch": 0.09516666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.4436848958333335,
+      "learning_rate": 9.778965073991651e-05,
+      "loss": 6.9812,
+      "loss/crossentropy": 1.4228438585996628,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1732054352760315,
+      "step": 571
+    },
+    {
+      "epoch": 0.09533333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.412955729166667,
+      "learning_rate": 9.778194623165296e-05,
+      "loss": 6.6781,
+      "loss/crossentropy": 1.465393453836441,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16653447039425373,
+      "step": 572
+    },
+    {
+      "epoch": 0.0955,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.4229166666666666,
+      "learning_rate": 9.777422862369783e-05,
+      "loss": 6.727,
+      "loss/crossentropy": 1.8058487474918365,
+      "loss/hidden": 3.72265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22924640774726868,
+      "step": 573
+    },
+    {
+      "epoch": 0.09566666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.9395833333333334,
+      "learning_rate": 9.776649791816698e-05,
+      "loss": 7.5235,
+      "loss/crossentropy": 1.7399717271327972,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19607891328632832,
+      "step": 574
+    },
+    {
+      "epoch": 0.09583333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.9488932291666667,
+      "learning_rate": 9.77587541171798e-05,
+      "loss": 7.0177,
+      "loss/crossentropy": 1.519953191280365,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15953503549098969,
+      "step": 575
+    },
+    {
+      "epoch": 0.096,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.8598307291666667,
+      "learning_rate": 9.775099722285935e-05,
+      "loss": 7.1901,
+      "loss/crossentropy": 1.9360006749629974,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17414483055472374,
+      "step": 576
+    },
+    {
+      "epoch": 0.09616666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.3072916666666667,
+      "learning_rate": 9.774322723733216e-05,
+      "loss": 7.2844,
+      "loss/crossentropy": 1.4809782207012177,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16507546417415142,
+      "step": 577
+    },
+    {
+      "epoch": 0.09633333333333334,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.3739583333333334,
+      "learning_rate": 9.773544416272845e-05,
+      "loss": 6.9057,
+      "loss/crossentropy": 1.7025932371616364,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2140498273074627,
+      "step": 578
+    },
+    {
+      "epoch": 0.0965,
+      "grad_norm": 35.5,
+      "grad_norm_var": 3.3580729166666665,
+      "learning_rate": 9.772764800118199e-05,
+      "loss": 6.8644,
+      "loss/crossentropy": 1.443876400589943,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23979828134179115,
+      "step": 579
+    },
+    {
+      "epoch": 0.09666666666666666,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.3072916666666665,
+      "learning_rate": 9.771983875483013e-05,
+      "loss": 7.0165,
+      "loss/crossentropy": 1.5318384170532227,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1742076575756073,
+      "step": 580
+    },
+    {
+      "epoch": 0.09683333333333333,
+      "grad_norm": 27.75,
+      "grad_norm_var": 3.7291666666666665,
+      "learning_rate": 9.771201642581385e-05,
+      "loss": 6.9728,
+      "loss/crossentropy": 2.00956591963768,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17072244361042976,
+      "step": 581
+    },
+    {
+      "epoch": 0.097,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.595833333333333,
+      "learning_rate": 9.770418101627765e-05,
+      "loss": 7.431,
+      "loss/crossentropy": 2.0086843073368073,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20463511534035206,
+      "step": 582
+    },
+    {
+      "epoch": 0.09716666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.698958333333333,
+      "learning_rate": 9.769633252836969e-05,
+      "loss": 6.8939,
+      "loss/crossentropy": 1.2146107405424118,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15556911751627922,
+      "step": 583
+    },
+    {
+      "epoch": 0.09733333333333333,
+      "grad_norm": 27.875,
+      "grad_norm_var": 3.937239583333333,
+      "learning_rate": 9.768847096424164e-05,
+      "loss": 6.6847,
+      "loss/crossentropy": 1.3093099147081375,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1486276388168335,
+      "step": 584
+    },
+    {
+      "epoch": 0.0975,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.7622395833333333,
+      "learning_rate": 9.76805963260488e-05,
+      "loss": 6.9212,
+      "loss/crossentropy": 1.5321282744407654,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18583380058407784,
+      "step": 585
+    },
+    {
+      "epoch": 0.09766666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.323372395833333,
+      "learning_rate": 9.767270861595005e-05,
+      "loss": 6.6329,
+      "loss/crossentropy": 1.4445045590400696,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14285002881661057,
+      "step": 586
+    },
+    {
+      "epoch": 0.09783333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.393684895833333,
+      "learning_rate": 9.766480783610788e-05,
+      "loss": 6.4554,
+      "loss/crossentropy": 1.6248749047517776,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1326998956501484,
+      "step": 587
+    },
+    {
+      "epoch": 0.098,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.4546223958333333,
+      "learning_rate": 9.765689398868831e-05,
+      "loss": 6.8885,
+      "loss/crossentropy": 1.5625443756580353,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19499482214450836,
+      "step": 588
+    },
+    {
+      "epoch": 0.09816666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.5139973958333335,
+      "learning_rate": 9.764896707586096e-05,
+      "loss": 7.162,
+      "loss/crossentropy": 1.857871800661087,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18508696928620338,
+      "step": 589
+    },
+    {
+      "epoch": 0.09833333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.092643229166667,
+      "learning_rate": 9.764102709979902e-05,
+      "loss": 6.9312,
+      "loss/crossentropy": 1.6804062724113464,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15879181027412415,
+      "step": 590
+    },
+    {
+      "epoch": 0.0985,
+      "grad_norm": 29.625,
+      "grad_norm_var": 4.128125,
+      "learning_rate": 9.763307406267932e-05,
+      "loss": 6.902,
+      "loss/crossentropy": 1.4155396074056625,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17896759510040283,
+      "step": 591
+    },
+    {
+      "epoch": 0.09866666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 4.225,
+      "learning_rate": 9.76251079666822e-05,
+      "loss": 7.221,
+      "loss/crossentropy": 1.1623346209526062,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11757522448897362,
+      "step": 592
+    },
+    {
+      "epoch": 0.09883333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 4.148372395833333,
+      "learning_rate": 9.761712881399164e-05,
+      "loss": 6.7943,
+      "loss/crossentropy": 1.711197167634964,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20673281326889992,
+      "step": 593
+    },
+    {
+      "epoch": 0.099,
+      "grad_norm": 29.75,
+      "grad_norm_var": 4.023958333333334,
+      "learning_rate": 9.760913660679515e-05,
+      "loss": 6.6294,
+      "loss/crossentropy": 1.412598729133606,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3168126614764333,
+      "step": 594
+    },
+    {
+      "epoch": 0.09916666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.25,
+      "learning_rate": 9.760113134728384e-05,
+      "loss": 6.8201,
+      "loss/crossentropy": 1.5064578652381897,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24883772060275078,
+      "step": 595
+    },
+    {
+      "epoch": 0.09933333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.0809895833333334,
+      "learning_rate": 9.75931130376524e-05,
+      "loss": 6.7601,
+      "loss/crossentropy": 1.9312743097543716,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19404222816228867,
+      "step": 596
+    },
+    {
+      "epoch": 0.0995,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.77890625,
+      "learning_rate": 9.75850816800991e-05,
+      "loss": 7.3074,
+      "loss/crossentropy": 1.9680878818035126,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2916663810610771,
+      "step": 597
+    },
+    {
+      "epoch": 0.09966666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.7733723958333334,
+      "learning_rate": 9.757703727682574e-05,
+      "loss": 6.8996,
+      "loss/crossentropy": 1.8550788760185242,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1608690246939659,
+      "step": 598
+    },
+    {
+      "epoch": 0.09983333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.6676432291666667,
+      "learning_rate": 9.756897983003781e-05,
+      "loss": 6.9309,
+      "loss/crossentropy": 1.9774325788021088,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1924036145210266,
+      "step": 599
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.3067057291666666,
+      "learning_rate": 9.756090934194427e-05,
+      "loss": 7.0654,
+      "loss/crossentropy": 1.9172890186309814,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17722957953810692,
+      "step": 600
+    },
+    {
+      "epoch": 0.10016666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.2330729166666667,
+      "learning_rate": 9.755282581475769e-05,
+      "loss": 6.8526,
+      "loss/crossentropy": 1.6295377165079117,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15426835790276527,
+      "step": 601
+    },
+    {
+      "epoch": 0.10033333333333333,
+      "grad_norm": 27.375,
+      "grad_norm_var": 1.7197265625,
+      "learning_rate": 9.75447292506942e-05,
+      "loss": 6.655,
+      "loss/crossentropy": 1.492895856499672,
+      "loss/hidden": 3.08984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13759666681289673,
+      "step": 602
+    },
+    {
+      "epoch": 0.1005,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.66875,
+      "learning_rate": 9.753661965197354e-05,
+      "loss": 6.9816,
+      "loss/crossentropy": 1.730408102273941,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20357651636004448,
+      "step": 603
+    },
+    {
+      "epoch": 0.10066666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.6046223958333334,
+      "learning_rate": 9.752849702081901e-05,
+      "loss": 7.0638,
+      "loss/crossentropy": 1.7239017188549042,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1713139358907938,
+      "step": 604
+    },
+    {
+      "epoch": 0.10083333333333333,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.72890625,
+      "learning_rate": 9.752036135945744e-05,
+      "loss": 6.9032,
+      "loss/crossentropy": 1.9123197048902512,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.166269407607615,
+      "step": 605
+    },
+    {
+      "epoch": 0.101,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.1728515625,
+      "learning_rate": 9.751221267011929e-05,
+      "loss": 7.1609,
+      "loss/crossentropy": 2.311406910419464,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22842303290963173,
+      "step": 606
+    },
+    {
+      "epoch": 0.10116666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.2093098958333333,
+      "learning_rate": 9.750405095503859e-05,
+      "loss": 6.8101,
+      "loss/crossentropy": 1.530705913901329,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22688306868076324,
+      "step": 607
+    },
+    {
+      "epoch": 0.10133333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.646875,
+      "learning_rate": 9.749587621645288e-05,
+      "loss": 7.5102,
+      "loss/crossentropy": 1.8076139986515045,
+      "loss/hidden": 3.7734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30030296742916107,
+      "step": 608
+    },
+    {
+      "epoch": 0.1015,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.8018229166666666,
+      "learning_rate": 9.748768845660334e-05,
+      "loss": 6.7849,
+      "loss/crossentropy": 1.4822484999895096,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1463693156838417,
+      "step": 609
+    },
+    {
+      "epoch": 0.10166666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.87890625,
+      "learning_rate": 9.74794876777347e-05,
+      "loss": 6.8185,
+      "loss/crossentropy": 1.566762924194336,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1515646893531084,
+      "step": 610
+    },
+    {
+      "epoch": 0.10183333333333333,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.9275390625,
+      "learning_rate": 9.74712738820952e-05,
+      "loss": 6.5295,
+      "loss/crossentropy": 1.6448956429958344,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21347195468842983,
+      "step": 611
+    },
+    {
+      "epoch": 0.102,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.0119140625,
+      "learning_rate": 9.746304707193675e-05,
+      "loss": 7.4048,
+      "loss/crossentropy": 1.5197457373142242,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21838144212961197,
+      "step": 612
+    },
+    {
+      "epoch": 0.10216666666666667,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.14765625,
+      "learning_rate": 9.745480724951473e-05,
+      "loss": 6.4885,
+      "loss/crossentropy": 2.393069565296173,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19103804975748062,
+      "step": 613
+    },
+    {
+      "epoch": 0.10233333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.5184895833333334,
+      "learning_rate": 9.744655441708818e-05,
+      "loss": 7.4833,
+      "loss/crossentropy": 1.794210433959961,
+      "loss/hidden": 3.71484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2325542066246271,
+      "step": 614
+    },
+    {
+      "epoch": 0.1025,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.6369140625,
+      "learning_rate": 9.743828857691963e-05,
+      "loss": 6.4373,
+      "loss/crossentropy": 1.2132785096764565,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12143177725374699,
+      "step": 615
+    },
+    {
+      "epoch": 0.10266666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.62890625,
+      "learning_rate": 9.743000973127523e-05,
+      "loss": 7.0124,
+      "loss/crossentropy": 1.3035457134246826,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1357813635841012,
+      "step": 616
+    },
+    {
+      "epoch": 0.10283333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.6625,
+      "learning_rate": 9.742171788242466e-05,
+      "loss": 6.5431,
+      "loss/crossentropy": 0.8840593248605728,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15165851637721062,
+      "step": 617
+    },
+    {
+      "epoch": 0.103,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.32265625,
+      "learning_rate": 9.741341303264118e-05,
+      "loss": 6.7935,
+      "loss/crossentropy": 1.244262933731079,
+      "loss/hidden": 4.00390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20905006304383278,
+      "step": 618
+    },
+    {
+      "epoch": 0.10316666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.403580729166667,
+      "learning_rate": 9.74050951842016e-05,
+      "loss": 6.6221,
+      "loss/crossentropy": 1.4201457425951958,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12143450789153576,
+      "step": 619
+    },
+    {
+      "epoch": 0.10333333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.4559895833333334,
+      "learning_rate": 9.739676433938633e-05,
+      "loss": 6.8997,
+      "loss/crossentropy": 1.5559433475136757,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13031989242881536,
+      "step": 620
+    },
+    {
+      "epoch": 0.1035,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.36015625,
+      "learning_rate": 9.73884205004793e-05,
+      "loss": 6.892,
+      "loss/crossentropy": 1.5828931331634521,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13859030604362488,
+      "step": 621
+    },
+    {
+      "epoch": 0.10366666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.1997395833333333,
+      "learning_rate": 9.7380063669768e-05,
+      "loss": 6.6631,
+      "loss/crossentropy": 1.3110764995217323,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1283620810136199,
+      "step": 622
+    },
+    {
+      "epoch": 0.10383333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.36015625,
+      "learning_rate": 9.737169384954355e-05,
+      "loss": 7.5041,
+      "loss/crossentropy": 1.8889169991016388,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20347655937075615,
+      "step": 623
+    },
+    {
+      "epoch": 0.104,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.715625,
+      "learning_rate": 9.736331104210056e-05,
+      "loss": 7.0787,
+      "loss/crossentropy": 2.0271048545837402,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1943119689822197,
+      "step": 624
+    },
+    {
+      "epoch": 0.10416666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.0056640625,
+      "learning_rate": 9.735491524973722e-05,
+      "loss": 7.2658,
+      "loss/crossentropy": 1.3105893656611443,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17542754113674164,
+      "step": 625
+    },
+    {
+      "epoch": 0.10433333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.9353515625,
+      "learning_rate": 9.73465064747553e-05,
+      "loss": 6.9306,
+      "loss/crossentropy": 1.5623179972171783,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35115011036396027,
+      "step": 626
+    },
+    {
+      "epoch": 0.1045,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.8145833333333334,
+      "learning_rate": 9.73380847194601e-05,
+      "loss": 6.618,
+      "loss/crossentropy": 1.7431139945983887,
+      "loss/hidden": 3.7734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23764119669795036,
+      "step": 627
+    },
+    {
+      "epoch": 0.10466666666666667,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.0552083333333333,
+      "learning_rate": 9.732964998616046e-05,
+      "loss": 6.587,
+      "loss/crossentropy": 1.249927893280983,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11416234821081161,
+      "step": 628
+    },
+    {
+      "epoch": 0.10483333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.8921223958333333,
+      "learning_rate": 9.732120227716888e-05,
+      "loss": 6.9894,
+      "loss/crossentropy": 1.6706857830286026,
+      "loss/hidden": 3.7421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.280877910554409,
+      "step": 629
+    },
+    {
+      "epoch": 0.105,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.1499348958333333,
+      "learning_rate": 9.73127415948013e-05,
+      "loss": 6.6645,
+      "loss/crossentropy": 1.8794518113136292,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25239719450473785,
+      "step": 630
+    },
+    {
+      "epoch": 0.10516666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.074934895833333,
+      "learning_rate": 9.730426794137727e-05,
+      "loss": 7.0315,
+      "loss/crossentropy": 1.941453456878662,
+      "loss/hidden": 3.87890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23865931853652,
+      "step": 631
+    },
+    {
+      "epoch": 0.10533333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.1171223958333334,
+      "learning_rate": 9.72957813192199e-05,
+      "loss": 7.1967,
+      "loss/crossentropy": 1.1591122299432755,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2066329289227724,
+      "step": 632
+    },
+    {
+      "epoch": 0.1055,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.502083333333333,
+      "learning_rate": 9.728728173065585e-05,
+      "loss": 6.8588,
+      "loss/crossentropy": 1.926354318857193,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3170504793524742,
+      "step": 633
+    },
+    {
+      "epoch": 0.10566666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.432747395833333,
+      "learning_rate": 9.72787691780153e-05,
+      "loss": 6.9414,
+      "loss/crossentropy": 1.741186648607254,
+      "loss/hidden": 3.66015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19436799362301826,
+      "step": 634
+    },
+    {
+      "epoch": 0.10583333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.4197265625,
+      "learning_rate": 9.727024366363206e-05,
+      "loss": 6.8866,
+      "loss/crossentropy": 1.432238832116127,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22247987799346447,
+      "step": 635
+    },
+    {
+      "epoch": 0.106,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.4309895833333335,
+      "learning_rate": 9.726170518984341e-05,
+      "loss": 6.7905,
+      "loss/crossentropy": 1.2945048660039902,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16999693773686886,
+      "step": 636
+    },
+    {
+      "epoch": 0.10616666666666667,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.5520833333333335,
+      "learning_rate": 9.725315375899024e-05,
+      "loss": 6.5599,
+      "loss/crossentropy": 1.464924231171608,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20916100591421127,
+      "step": 637
+    },
+    {
+      "epoch": 0.10633333333333334,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.5885416666666665,
+      "learning_rate": 9.724458937341698e-05,
+      "loss": 6.6968,
+      "loss/crossentropy": 1.3260242938995361,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21571195125579834,
+      "step": 638
+    },
+    {
+      "epoch": 0.1065,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.287239583333333,
+      "learning_rate": 9.723601203547158e-05,
+      "loss": 6.9984,
+      "loss/crossentropy": 1.5926786959171295,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3045806996524334,
+      "step": 639
+    },
+    {
+      "epoch": 0.10666666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.25625,
+      "learning_rate": 9.722742174750558e-05,
+      "loss": 6.7535,
+      "loss/crossentropy": 1.588733658194542,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2000904083251953,
+      "step": 640
+    },
+    {
+      "epoch": 0.10683333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.5143229166666667,
+      "learning_rate": 9.721881851187406e-05,
+      "loss": 6.9981,
+      "loss/crossentropy": 2.038671374320984,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22879015281796455,
+      "step": 641
+    },
+    {
+      "epoch": 0.107,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.6510416666666667,
+      "learning_rate": 9.721020233093563e-05,
+      "loss": 7.0265,
+      "loss/crossentropy": 1.8578952550888062,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3181113712489605,
+      "step": 642
+    },
+    {
+      "epoch": 0.10716666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.7093098958333333,
+      "learning_rate": 9.72015732070525e-05,
+      "loss": 7.1423,
+      "loss/crossentropy": 1.2508052438497543,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14286888763308525,
+      "step": 643
+    },
+    {
+      "epoch": 0.10733333333333334,
+      "grad_norm": 27.75,
+      "grad_norm_var": 1.7093098958333333,
+      "learning_rate": 9.719293114259033e-05,
+      "loss": 6.7903,
+      "loss/crossentropy": 1.6175063848495483,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18509873375296593,
+      "step": 644
+    },
+    {
+      "epoch": 0.1075,
+      "grad_norm": 27.0,
+      "grad_norm_var": 1.8853515625,
+      "learning_rate": 9.718427613991848e-05,
+      "loss": 6.8594,
+      "loss/crossentropy": 2.0574215948581696,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17839450761675835,
+      "step": 645
+    },
+    {
+      "epoch": 0.10766666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.2806640625,
+      "learning_rate": 9.717560820140969e-05,
+      "loss": 6.9529,
+      "loss/crossentropy": 1.7327706515789032,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20203590393066406,
+      "step": 646
+    },
+    {
+      "epoch": 0.10783333333333334,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.52890625,
+      "learning_rate": 9.716692732944035e-05,
+      "loss": 6.6285,
+      "loss/crossentropy": 2.002866104245186,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1876381952315569,
+      "step": 647
+    },
+    {
+      "epoch": 0.108,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.5822265625,
+      "learning_rate": 9.715823352639037e-05,
+      "loss": 6.8543,
+      "loss/crossentropy": 1.0742419362068176,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13539068680256605,
+      "step": 648
+    },
+    {
+      "epoch": 0.10816666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.33515625,
+      "learning_rate": 9.714952679464323e-05,
+      "loss": 6.6677,
+      "loss/crossentropy": 1.1316107213497162,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20465175807476044,
+      "step": 649
+    },
+    {
+      "epoch": 0.10833333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.14375,
+      "learning_rate": 9.71408071365859e-05,
+      "loss": 7.1131,
+      "loss/crossentropy": 1.1828127279877663,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12344163469970226,
+      "step": 650
+    },
+    {
+      "epoch": 0.1085,
+      "grad_norm": 28.125,
+      "grad_norm_var": 3.160416666666667,
+      "learning_rate": 9.713207455460894e-05,
+      "loss": 6.9142,
+      "loss/crossentropy": 1.5287613421678543,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2282132599502802,
+      "step": 651
+    },
+    {
+      "epoch": 0.10866666666666666,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.17265625,
+      "learning_rate": 9.71233290511064e-05,
+      "loss": 6.8428,
+      "loss/crossentropy": 1.3678704053163528,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18871070258319378,
+      "step": 652
+    },
+    {
+      "epoch": 0.10883333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.238541666666667,
+      "learning_rate": 9.711457062847595e-05,
+      "loss": 7.1566,
+      "loss/crossentropy": 1.1779525876045227,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14346816204488277,
+      "step": 653
+    },
+    {
+      "epoch": 0.109,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.6660807291666666,
+      "learning_rate": 9.710579928911876e-05,
+      "loss": 7.0187,
+      "loss/crossentropy": 1.165409579873085,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14943919517099857,
+      "step": 654
+    },
+    {
+      "epoch": 0.10916666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 3.8218098958333333,
+      "learning_rate": 9.709701503543954e-05,
+      "loss": 6.9463,
+      "loss/crossentropy": 1.7063288986682892,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17271865159273148,
+      "step": 655
+    },
+    {
+      "epoch": 0.10933333333333334,
+      "grad_norm": 28.5,
+      "grad_norm_var": 3.910872395833333,
+      "learning_rate": 9.708821786984652e-05,
+      "loss": 6.9349,
+      "loss/crossentropy": 1.4682232439517975,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20218270272016525,
+      "step": 656
+    },
+    {
+      "epoch": 0.1095,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.94375,
+      "learning_rate": 9.707940779475151e-05,
+      "loss": 6.979,
+      "loss/crossentropy": 2.069106310606003,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2026902660727501,
+      "step": 657
+    },
+    {
+      "epoch": 0.10966666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.8979166666666667,
+      "learning_rate": 9.707058481256985e-05,
+      "loss": 6.8815,
+      "loss/crossentropy": 1.8183580487966537,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22271856293082237,
+      "step": 658
+    },
+    {
+      "epoch": 0.10983333333333334,
+      "grad_norm": 28.5,
+      "grad_norm_var": 3.9546223958333333,
+      "learning_rate": 9.706174892572039e-05,
+      "loss": 6.2977,
+      "loss/crossentropy": 1.6013105809688568,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13599575497210026,
+      "step": 659
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.756705729166667,
+      "learning_rate": 9.705290013662556e-05,
+      "loss": 6.631,
+      "loss/crossentropy": 1.3786100298166275,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18141034804284573,
+      "step": 660
+    },
+    {
+      "epoch": 0.11016666666666666,
+      "grad_norm": 27.5,
+      "grad_norm_var": 3.598893229166667,
+      "learning_rate": 9.704403844771128e-05,
+      "loss": 6.7207,
+      "loss/crossentropy": 1.2611496448516846,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23525800928473473,
+      "step": 661
+    },
+    {
+      "epoch": 0.11033333333333334,
+      "grad_norm": 27.625,
+      "grad_norm_var": 3.220572916666667,
+      "learning_rate": 9.703516386140705e-05,
+      "loss": 6.3873,
+      "loss/crossentropy": 1.2649259716272354,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12177464924752712,
+      "step": 662
+    },
+    {
+      "epoch": 0.1105,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.795768229166667,
+      "learning_rate": 9.70262763801459e-05,
+      "loss": 7.6091,
+      "loss/crossentropy": 1.9886409044265747,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20045031234622002,
+      "step": 663
+    },
+    {
+      "epoch": 0.11066666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.6768229166666666,
+      "learning_rate": 9.701737600636436e-05,
+      "loss": 7.0348,
+      "loss/crossentropy": 2.0670722126960754,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25915420427918434,
+      "step": 664
+    },
+    {
+      "epoch": 0.11083333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.6518229166666667,
+      "learning_rate": 9.700846274250251e-05,
+      "loss": 6.9435,
+      "loss/crossentropy": 1.2922873795032501,
+      "loss/hidden": 3.83203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3342493027448654,
+      "step": 665
+    },
+    {
+      "epoch": 0.111,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.8676432291666667,
+      "learning_rate": 9.699953659100401e-05,
+      "loss": 7.0355,
+      "loss/crossentropy": 1.8357878625392914,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25074972212314606,
+      "step": 666
+    },
+    {
+      "epoch": 0.11116666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.76875,
+      "learning_rate": 9.699059755431598e-05,
+      "loss": 7.0087,
+      "loss/crossentropy": 1.8910188376903534,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1865641139447689,
+      "step": 667
+    },
+    {
+      "epoch": 0.11133333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.7577473958333334,
+      "learning_rate": 9.698164563488914e-05,
+      "loss": 6.9317,
+      "loss/crossentropy": 1.6376636326313019,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32860966958105564,
+      "step": 668
+    },
+    {
+      "epoch": 0.1115,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.7059895833333334,
+      "learning_rate": 9.697268083517767e-05,
+      "loss": 6.826,
+      "loss/crossentropy": 1.811766117811203,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15397534891963005,
+      "step": 669
+    },
+    {
+      "epoch": 0.11166666666666666,
+      "grad_norm": 28.0,
+      "grad_norm_var": 1.2854166666666667,
+      "learning_rate": 9.696370315763936e-05,
+      "loss": 6.5573,
+      "loss/crossentropy": 1.1196317374706268,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12236335687339306,
+      "step": 670
+    },
+    {
+      "epoch": 0.11183333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 0.8858723958333333,
+      "learning_rate": 9.695471260473545e-05,
+      "loss": 6.4967,
+      "loss/crossentropy": 1.5364763289690018,
+      "loss/hidden": 3.046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12572124414145947,
+      "step": 671
+    },
+    {
+      "epoch": 0.112,
+      "grad_norm": 27.5,
+      "grad_norm_var": 1.0285807291666667,
+      "learning_rate": 9.69457091789308e-05,
+      "loss": 6.9021,
+      "loss/crossentropy": 1.548076406121254,
+      "loss/hidden": 3.82421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19756012596189976,
+      "step": 672
+    },
+    {
+      "epoch": 0.11216666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 0.9018229166666667,
+      "learning_rate": 9.693669288269372e-05,
+      "loss": 7.1784,
+      "loss/crossentropy": 1.3275788128376007,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17808063328266144,
+      "step": 673
+    },
+    {
+      "epoch": 0.11233333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 0.8988932291666667,
+      "learning_rate": 9.692766371849606e-05,
+      "loss": 6.564,
+      "loss/crossentropy": 1.3597323894500732,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13040512800216675,
+      "step": 674
+    },
+    {
+      "epoch": 0.1125,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.3931640625,
+      "learning_rate": 9.691862168881325e-05,
+      "loss": 6.83,
+      "loss/crossentropy": 1.1135968118906021,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16545339487493038,
+      "step": 675
+    },
+    {
+      "epoch": 0.11266666666666666,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.3942057291666667,
+      "learning_rate": 9.690956679612421e-05,
+      "loss": 7.2353,
+      "loss/crossentropy": 1.4192567318677902,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1604620646685362,
+      "step": 676
+    },
+    {
+      "epoch": 0.11283333333333333,
+      "grad_norm": 28.0,
+      "grad_norm_var": 1.3030598958333333,
+      "learning_rate": 9.690049904291139e-05,
+      "loss": 6.7055,
+      "loss/crossentropy": 1.5826443284749985,
+      "loss/hidden": 3.10546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13827670831233263,
+      "step": 677
+    },
+    {
+      "epoch": 0.113,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.178125,
+      "learning_rate": 9.689141843166074e-05,
+      "loss": 6.9351,
+      "loss/crossentropy": 1.8830396234989166,
+      "loss/hidden": 3.76953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2651614472270012,
+      "step": 678
+    },
+    {
+      "epoch": 0.11316666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.1455729166666666,
+      "learning_rate": 9.688232496486178e-05,
+      "loss": 6.8014,
+      "loss/crossentropy": 1.044035404920578,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17703981511294842,
+      "step": 679
+    },
+    {
+      "epoch": 0.11333333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.140625,
+      "learning_rate": 9.687321864500755e-05,
+      "loss": 6.8278,
+      "loss/crossentropy": 1.6375316381454468,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15793989598751068,
+      "step": 680
+    },
+    {
+      "epoch": 0.1135,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.1379557291666667,
+      "learning_rate": 9.686409947459458e-05,
+      "loss": 7.328,
+      "loss/crossentropy": 1.8119606971740723,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16568780690431595,
+      "step": 681
+    },
+    {
+      "epoch": 0.11366666666666667,
+      "grad_norm": 26.875,
+      "grad_norm_var": 1.4942057291666666,
+      "learning_rate": 9.685496745612295e-05,
+      "loss": 6.2698,
+      "loss/crossentropy": 1.6398979425430298,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16278844699263573,
+      "step": 682
+    },
+    {
+      "epoch": 0.11383333333333333,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.5104166666666667,
+      "learning_rate": 9.684582259209624e-05,
+      "loss": 6.3844,
+      "loss/crossentropy": 1.6580571979284286,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15163831412792206,
+      "step": 683
+    },
+    {
+      "epoch": 0.114,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.658072916666667,
+      "learning_rate": 9.683666488502158e-05,
+      "loss": 7.7771,
+      "loss/crossentropy": 1.6030221581459045,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2512742020189762,
+      "step": 684
+    },
+    {
+      "epoch": 0.11416666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.6416015625,
+      "learning_rate": 9.682749433740962e-05,
+      "loss": 6.8587,
+      "loss/crossentropy": 2.0054213404655457,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20447374135255814,
+      "step": 685
+    },
+    {
+      "epoch": 0.11433333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.517708333333333,
+      "learning_rate": 9.68183109517745e-05,
+      "loss": 6.6522,
+      "loss/crossentropy": 1.5510950908064842,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15018807910382748,
+      "step": 686
+    },
+    {
+      "epoch": 0.1145,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.85625,
+      "learning_rate": 9.68091147306339e-05,
+      "loss": 7.1964,
+      "loss/crossentropy": 1.603594422340393,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19394610077142715,
+      "step": 687
+    },
+    {
+      "epoch": 0.11466666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.55625,
+      "learning_rate": 9.6799905676509e-05,
+      "loss": 6.596,
+      "loss/crossentropy": 1.203146532177925,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13622648641467094,
+      "step": 688
+    },
+    {
+      "epoch": 0.11483333333333333,
+      "grad_norm": 28.25,
+      "grad_norm_var": 2.66015625,
+      "learning_rate": 9.679068379192456e-05,
+      "loss": 6.8174,
+      "loss/crossentropy": 1.8890211880207062,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19486655294895172,
+      "step": 689
+    },
+    {
+      "epoch": 0.115,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.56875,
+      "learning_rate": 9.678144907940876e-05,
+      "loss": 6.5261,
+      "loss/crossentropy": 1.2591911479830742,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15090861823409796,
+      "step": 690
+    },
+    {
+      "epoch": 0.11516666666666667,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.4416015625,
+      "learning_rate": 9.677220154149337e-05,
+      "loss": 6.9695,
+      "loss/crossentropy": 1.5569715201854706,
+      "loss/hidden": 3.11328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15242954529821873,
+      "step": 691
+    },
+    {
+      "epoch": 0.11533333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.4572265625,
+      "learning_rate": 9.676294118071367e-05,
+      "loss": 7.225,
+      "loss/crossentropy": 1.8722420632839203,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18725543096661568,
+      "step": 692
+    },
+    {
+      "epoch": 0.1155,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.30390625,
+      "learning_rate": 9.675366799960841e-05,
+      "loss": 6.6473,
+      "loss/crossentropy": 1.6560204923152924,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21664336323738098,
+      "step": 693
+    },
+    {
+      "epoch": 0.11566666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.3322265625,
+      "learning_rate": 9.674438200071991e-05,
+      "loss": 6.4056,
+      "loss/crossentropy": 1.1548308283090591,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11235202103853226,
+      "step": 694
+    },
+    {
+      "epoch": 0.11583333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.265625,
+      "learning_rate": 9.6735083186594e-05,
+      "loss": 7.151,
+      "loss/crossentropy": 1.9906552731990814,
+      "loss/hidden": 3.71875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2601942904293537,
+      "step": 695
+    },
+    {
+      "epoch": 0.116,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.249739583333333,
+      "learning_rate": 9.672577155977993e-05,
+      "loss": 7.2406,
+      "loss/crossentropy": 2.064489871263504,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23205005750060081,
+      "step": 696
+    },
+    {
+      "epoch": 0.11616666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.2561848958333335,
+      "learning_rate": 9.671644712283061e-05,
+      "loss": 7.1376,
+      "loss/crossentropy": 1.980436235666275,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2317683883011341,
+      "step": 697
+    },
+    {
+      "epoch": 0.11633333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.6285807291666665,
+      "learning_rate": 9.670710987830233e-05,
+      "loss": 7.002,
+      "loss/crossentropy": 1.6101481318473816,
+      "loss/hidden": 3.7109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2259971722960472,
+      "step": 698
+    },
+    {
+      "epoch": 0.1165,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.4770182291666667,
+      "learning_rate": 9.669775982875501e-05,
+      "loss": 6.5709,
+      "loss/crossentropy": 1.5695571601390839,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20522866770625114,
+      "step": 699
+    },
+    {
+      "epoch": 0.11666666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.6931640625,
+      "learning_rate": 9.668839697675196e-05,
+      "loss": 6.8714,
+      "loss/crossentropy": 1.5800593048334122,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13942819274961948,
+      "step": 700
+    },
+    {
+      "epoch": 0.11683333333333333,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.1400390625,
+      "learning_rate": 9.667902132486009e-05,
+      "loss": 7.086,
+      "loss/crossentropy": 1.6381309926509857,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20893165469169617,
+      "step": 701
+    },
+    {
+      "epoch": 0.117,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.39140625,
+      "learning_rate": 9.666963287564979e-05,
+      "loss": 6.7993,
+      "loss/crossentropy": 1.406240090727806,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14155222102999687,
+      "step": 702
+    },
+    {
+      "epoch": 0.11716666666666667,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.3119140625,
+      "learning_rate": 9.666023163169493e-05,
+      "loss": 6.8117,
+      "loss/crossentropy": 1.5503928065299988,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22186162695288658,
+      "step": 703
+    },
+    {
+      "epoch": 0.11733333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.691666666666667,
+      "learning_rate": 9.665081759557295e-05,
+      "loss": 6.8223,
+      "loss/crossentropy": 1.3800048977136612,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14740336686372757,
+      "step": 704
+    },
+    {
+      "epoch": 0.1175,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.5791666666666666,
+      "learning_rate": 9.664139076986473e-05,
+      "loss": 6.8227,
+      "loss/crossentropy": 1.8590564131736755,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2021271251142025,
+      "step": 705
+    },
+    {
+      "epoch": 0.11766666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.981184895833333,
+      "learning_rate": 9.663195115715471e-05,
+      "loss": 6.9499,
+      "loss/crossentropy": 1.4088309109210968,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2622120473533869,
+      "step": 706
+    },
+    {
+      "epoch": 0.11783333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.7405598958333335,
+      "learning_rate": 9.66224987600308e-05,
+      "loss": 6.8124,
+      "loss/crossentropy": 0.7446690201759338,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.08213549666106701,
+      "step": 707
+    },
+    {
+      "epoch": 0.118,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.7875,
+      "learning_rate": 9.661303358108445e-05,
+      "loss": 6.7138,
+      "loss/crossentropy": 1.4749462604522705,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12841501459479332,
+      "step": 708
+    },
+    {
+      "epoch": 0.11816666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.8583333333333334,
+      "learning_rate": 9.660355562291055e-05,
+      "loss": 6.2732,
+      "loss/crossentropy": 1.4664273858070374,
+      "loss/hidden": 3.0859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18284995667636395,
+      "step": 709
+    },
+    {
+      "epoch": 0.11833333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.8421223958333335,
+      "learning_rate": 9.659406488810759e-05,
+      "loss": 6.935,
+      "loss/crossentropy": 1.71576888859272,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1478842906653881,
+      "step": 710
+    },
+    {
+      "epoch": 0.1185,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.9416015625,
+      "learning_rate": 9.658456137927745e-05,
+      "loss": 6.4814,
+      "loss/crossentropy": 1.682497650384903,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2495732717216015,
+      "step": 711
+    },
+    {
+      "epoch": 0.11866666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.9968098958333333,
+      "learning_rate": 9.657504509902562e-05,
+      "loss": 6.7867,
+      "loss/crossentropy": 1.3865531384944916,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19834261015057564,
+      "step": 712
+    },
+    {
+      "epoch": 0.11883333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.9712890625,
+      "learning_rate": 9.656551604996102e-05,
+      "loss": 6.6116,
+      "loss/crossentropy": 0.7168354988098145,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15031130611896515,
+      "step": 713
+    },
+    {
+      "epoch": 0.119,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.32265625,
+      "learning_rate": 9.655597423469609e-05,
+      "loss": 6.8321,
+      "loss/crossentropy": 1.9286990463733673,
+      "loss/hidden": 3.78125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3007762059569359,
+      "step": 714
+    },
+    {
+      "epoch": 0.11916666666666667,
+      "grad_norm": 41.75,
+      "grad_norm_var": 11.3259765625,
+      "learning_rate": 9.654641965584678e-05,
+      "loss": 7.197,
+      "loss/crossentropy": 1.7300785183906555,
+      "loss/hidden": 3.94921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2913442552089691,
+      "step": 715
+    },
+    {
+      "epoch": 0.11933333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 11.47890625,
+      "learning_rate": 9.653685231603256e-05,
+      "loss": 6.6237,
+      "loss/crossentropy": 1.163270428776741,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15404702350497246,
+      "step": 716
+    },
+    {
+      "epoch": 0.1195,
+      "grad_norm": 30.0,
+      "grad_norm_var": 10.788541666666667,
+      "learning_rate": 9.652727221787631e-05,
+      "loss": 7.0861,
+      "loss/crossentropy": 2.2192281931638718,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23627253249287605,
+      "step": 717
+    },
+    {
+      "epoch": 0.11966666666666667,
+      "grad_norm": 27.75,
+      "grad_norm_var": 10.788541666666667,
+      "learning_rate": 9.65176793640045e-05,
+      "loss": 6.326,
+      "loss/crossentropy": 0.9741474464535713,
+      "loss/hidden": 3.0390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10810017958283424,
+      "step": 718
+    },
+    {
+      "epoch": 0.11983333333333333,
+      "grad_norm": 25.875,
+      "grad_norm_var": 11.67265625,
+      "learning_rate": 9.650807375704708e-05,
+      "loss": 6.5607,
+      "loss/crossentropy": 1.051061324775219,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13475926592946053,
+      "step": 719
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 29.75,
+      "grad_norm_var": 11.560872395833334,
+      "learning_rate": 9.649845539963747e-05,
+      "loss": 7.3885,
+      "loss/crossentropy": 1.7970368564128876,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30203753150999546,
+      "step": 720
+    },
+    {
+      "epoch": 0.12016666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 11.495247395833333,
+      "learning_rate": 9.648882429441257e-05,
+      "loss": 6.5989,
+      "loss/crossentropy": 0.8426260724663734,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.08113336376845837,
+      "step": 721
+    },
+    {
+      "epoch": 0.12033333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 11.285416666666666,
+      "learning_rate": 9.647918044401285e-05,
+      "loss": 7.1849,
+      "loss/crossentropy": 1.6168367713689804,
+      "loss/hidden": 3.71484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25180529057979584,
+      "step": 722
+    },
+    {
+      "epoch": 0.1205,
+      "grad_norm": 28.0,
+      "grad_norm_var": 11.628580729166666,
+      "learning_rate": 9.646952385108218e-05,
+      "loss": 6.2459,
+      "loss/crossentropy": 1.617216557264328,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19698919728398323,
+      "step": 723
+    },
+    {
+      "epoch": 0.12066666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 11.552018229166666,
+      "learning_rate": 9.645985451826803e-05,
+      "loss": 6.6805,
+      "loss/crossentropy": 1.2977196872234344,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14982175081968307,
+      "step": 724
+    },
+    {
+      "epoch": 0.12083333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 11.508268229166667,
+      "learning_rate": 9.645017244822123e-05,
+      "loss": 6.8615,
+      "loss/crossentropy": 1.3951329737901688,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1752911787480116,
+      "step": 725
+    },
+    {
+      "epoch": 0.121,
+      "grad_norm": 28.25,
+      "grad_norm_var": 11.7744140625,
+      "learning_rate": 9.644047764359622e-05,
+      "loss": 6.7321,
+      "loss/crossentropy": 1.494904488325119,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1937835942953825,
+      "step": 726
+    },
+    {
+      "epoch": 0.12116666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 11.751041666666667,
+      "learning_rate": 9.643077010705087e-05,
+      "loss": 6.6861,
+      "loss/crossentropy": 1.0755391269922256,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15914678946137428,
+      "step": 727
+    },
+    {
+      "epoch": 0.12133333333333333,
+      "grad_norm": 28.0,
+      "grad_norm_var": 12.047916666666667,
+      "learning_rate": 9.642104984124656e-05,
+      "loss": 6.609,
+      "loss/crossentropy": 1.1384456008672714,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1375688686966896,
+      "step": 728
+    },
+    {
+      "epoch": 0.1215,
+      "grad_norm": 28.0,
+      "grad_norm_var": 12.308072916666667,
+      "learning_rate": 9.641131684884817e-05,
+      "loss": 6.8089,
+      "loss/crossentropy": 1.7485393583774567,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1841743802651763,
+      "step": 729
+    },
+    {
+      "epoch": 0.12166666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 12.008268229166667,
+      "learning_rate": 9.640157113252403e-05,
+      "loss": 6.8773,
+      "loss/crossentropy": 1.3851639330387115,
+      "loss/hidden": 3.84765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21244927495718002,
+      "step": 730
+    },
+    {
+      "epoch": 0.12183333333333334,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.8655598958333333,
+      "learning_rate": 9.6391812694946e-05,
+      "loss": 6.6945,
+      "loss/crossentropy": 2.034461110830307,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16909070312976837,
+      "step": 731
+    },
+    {
+      "epoch": 0.122,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.8802083333333333,
+      "learning_rate": 9.63820415387894e-05,
+      "loss": 6.5646,
+      "loss/crossentropy": 1.626416176557541,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23018508031964302,
+      "step": 732
+    },
+    {
+      "epoch": 0.12216666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.8186848958333333,
+      "learning_rate": 9.637225766673307e-05,
+      "loss": 7.168,
+      "loss/crossentropy": 1.2721869349479675,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27399044297635555,
+      "step": 733
+    },
+    {
+      "epoch": 0.12233333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.7233723958333333,
+      "learning_rate": 9.636246108145929e-05,
+      "loss": 6.7627,
+      "loss/crossentropy": 2.161185532808304,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2517819069325924,
+      "step": 734
+    },
+    {
+      "epoch": 0.1225,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.1869140625,
+      "learning_rate": 9.635265178565385e-05,
+      "loss": 7.3394,
+      "loss/crossentropy": 1.9593259394168854,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23115667700767517,
+      "step": 735
+    },
+    {
+      "epoch": 0.12266666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.2046223958333333,
+      "learning_rate": 9.634282978200604e-05,
+      "loss": 6.818,
+      "loss/crossentropy": 2.4617882668972015,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2561354413628578,
+      "step": 736
+    },
+    {
+      "epoch": 0.12283333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.6749348958333334,
+      "learning_rate": 9.63329950732086e-05,
+      "loss": 6.6859,
+      "loss/crossentropy": 1.8943971395492554,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19631916657090187,
+      "step": 737
+    },
+    {
+      "epoch": 0.123,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.5671223958333333,
+      "learning_rate": 9.632314766195781e-05,
+      "loss": 6.9489,
+      "loss/crossentropy": 2.0378142297267914,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15763233974575996,
+      "step": 738
+    },
+    {
+      "epoch": 0.12316666666666666,
+      "grad_norm": 28.0,
+      "grad_norm_var": 1.5671223958333333,
+      "learning_rate": 9.631328755095333e-05,
+      "loss": 6.8251,
+      "loss/crossentropy": 1.8512370437383652,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1793438456952572,
+      "step": 739
+    },
+    {
+      "epoch": 0.12333333333333334,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.4723307291666667,
+      "learning_rate": 9.630341474289842e-05,
+      "loss": 6.3814,
+      "loss/crossentropy": 1.370538242161274,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18280507531017065,
+      "step": 740
+    },
+    {
+      "epoch": 0.1235,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.4830729166666667,
+      "learning_rate": 9.629352924049975e-05,
+      "loss": 6.7228,
+      "loss/crossentropy": 2.3160256445407867,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17024001106619835,
+      "step": 741
+    },
+    {
+      "epoch": 0.12366666666666666,
+      "grad_norm": 26.875,
+      "grad_norm_var": 1.7988932291666666,
+      "learning_rate": 9.628363104646747e-05,
+      "loss": 6.1821,
+      "loss/crossentropy": 1.5081074684858322,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16622129827737808,
+      "step": 742
+    },
+    {
+      "epoch": 0.12383333333333334,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.8059895833333333,
+      "learning_rate": 9.627372016351524e-05,
+      "loss": 6.5492,
+      "loss/crossentropy": 1.3975093811750412,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15924087166786194,
+      "step": 743
+    },
+    {
+      "epoch": 0.124,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.8260416666666666,
+      "learning_rate": 9.626379659436017e-05,
+      "loss": 6.8219,
+      "loss/crossentropy": 1.901090830564499,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1702580563724041,
+      "step": 744
+    },
+    {
+      "epoch": 0.12416666666666666,
+      "grad_norm": 27.375,
+      "grad_norm_var": 1.9676432291666666,
+      "learning_rate": 9.62538603417229e-05,
+      "loss": 6.5135,
+      "loss/crossentropy": 1.7890234887599945,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1815632525831461,
+      "step": 745
+    },
+    {
+      "epoch": 0.12433333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.92890625,
+      "learning_rate": 9.624391140832749e-05,
+      "loss": 7.0427,
+      "loss/crossentropy": 1.2604841142892838,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21951996721327305,
+      "step": 746
+    },
+    {
+      "epoch": 0.1245,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.9830729166666667,
+      "learning_rate": 9.623394979690147e-05,
+      "loss": 6.661,
+      "loss/crossentropy": 1.4802487045526505,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14423799887299538,
+      "step": 747
+    },
+    {
+      "epoch": 0.12466666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.4635416666666665,
+      "learning_rate": 9.622397551017592e-05,
+      "loss": 7.041,
+      "loss/crossentropy": 1.2775348275899887,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19484638050198555,
+      "step": 748
+    },
+    {
+      "epoch": 0.12483333333333334,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.6103515625,
+      "learning_rate": 9.62139885508853e-05,
+      "loss": 7.014,
+      "loss/crossentropy": 1.925628885626793,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19798103347420692,
+      "step": 749
+    },
+    {
+      "epoch": 0.125,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.6510416666666665,
+      "learning_rate": 9.620398892176762e-05,
+      "loss": 6.8329,
+      "loss/crossentropy": 1.1196279376745224,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10725859738886356,
+      "step": 750
+    },
+    {
+      "epoch": 0.12516666666666668,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.5520833333333335,
+      "learning_rate": 9.619397662556435e-05,
+      "loss": 6.6299,
+      "loss/crossentropy": 1.1663732826709747,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20117589458823204,
+      "step": 751
+    },
+    {
+      "epoch": 0.12533333333333332,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.093489583333333,
+      "learning_rate": 9.618395166502037e-05,
+      "loss": 7.3826,
+      "loss/crossentropy": 1.6024877727031708,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16880478337407112,
+      "step": 752
+    },
+    {
+      "epoch": 0.1255,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.74140625,
+      "learning_rate": 9.617391404288412e-05,
+      "loss": 6.9638,
+      "loss/crossentropy": 1.6815844774246216,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15550420247018337,
+      "step": 753
+    },
+    {
+      "epoch": 0.12566666666666668,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.9525390625,
+      "learning_rate": 9.616386376190745e-05,
+      "loss": 6.327,
+      "loss/crossentropy": 1.329882264137268,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13500298745930195,
+      "step": 754
+    },
+    {
+      "epoch": 0.12583333333333332,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.86640625,
+      "learning_rate": 9.615380082484571e-05,
+      "loss": 7.2015,
+      "loss/crossentropy": 1.3013105019927025,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17569159716367722,
+      "step": 755
+    },
+    {
+      "epoch": 0.126,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.8962890625,
+      "learning_rate": 9.614372523445771e-05,
+      "loss": 6.6609,
+      "loss/crossentropy": 1.6127808392047882,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1659044213593006,
+      "step": 756
+    },
+    {
+      "epoch": 0.12616666666666668,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.9436848958333335,
+      "learning_rate": 9.613363699350575e-05,
+      "loss": 6.886,
+      "loss/crossentropy": 1.0334024354815483,
+      "loss/hidden": 3.1171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17941672634333372,
+      "step": 757
+    },
+    {
+      "epoch": 0.12633333333333333,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.7080729166666666,
+      "learning_rate": 9.612353610475555e-05,
+      "loss": 6.2095,
+      "loss/crossentropy": 1.0988683477044106,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14567056391388178,
+      "step": 758
+    },
+    {
+      "epoch": 0.1265,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.8955729166666666,
+      "learning_rate": 9.611342257097632e-05,
+      "loss": 7.148,
+      "loss/crossentropy": 1.6511463522911072,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20340903103351593,
+      "step": 759
+    },
+    {
+      "epoch": 0.12666666666666668,
+      "grad_norm": 27.375,
+      "grad_norm_var": 3.052018229166667,
+      "learning_rate": 9.610329639494076e-05,
+      "loss": 6.4546,
+      "loss/crossentropy": 1.1347185745835304,
+      "loss/hidden": 3.05859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10608388483524323,
+      "step": 760
+    },
+    {
+      "epoch": 0.12683333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.789322916666667,
+      "learning_rate": 9.609315757942503e-05,
+      "loss": 6.6186,
+      "loss/crossentropy": 1.6999226212501526,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19528476521372795,
+      "step": 761
+    },
+    {
+      "epoch": 0.127,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.880143229166667,
+      "learning_rate": 9.608300612720873e-05,
+      "loss": 7.0733,
+      "loss/crossentropy": 1.3845355212688446,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20608391612768173,
+      "step": 762
+    },
+    {
+      "epoch": 0.12716666666666668,
+      "grad_norm": 26.625,
+      "grad_norm_var": 3.314322916666667,
+      "learning_rate": 9.607284204107493e-05,
+      "loss": 6.9627,
+      "loss/crossentropy": 1.2844055891036987,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2255815453827381,
+      "step": 763
+    },
+    {
+      "epoch": 0.12733333333333333,
+      "grad_norm": 26.625,
+      "grad_norm_var": 3.147330729166667,
+      "learning_rate": 9.606266532381018e-05,
+      "loss": 6.9287,
+      "loss/crossentropy": 1.5335675477981567,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1263347789645195,
+      "step": 764
+    },
+    {
+      "epoch": 0.1275,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.1166015625,
+      "learning_rate": 9.605247597820448e-05,
+      "loss": 6.9739,
+      "loss/crossentropy": 1.8695217669010162,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1747262366116047,
+      "step": 765
+    },
+    {
+      "epoch": 0.12766666666666668,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.1143229166666666,
+      "learning_rate": 9.604227400705133e-05,
+      "loss": 6.4636,
+      "loss/crossentropy": 1.3276889026165009,
+      "loss/hidden": 3.0859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11597120389342308,
+      "step": 766
+    },
+    {
+      "epoch": 0.12783333333333333,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.0863932291666667,
+      "learning_rate": 9.603205941314758e-05,
+      "loss": 6.2796,
+      "loss/crossentropy": 1.057211548089981,
+      "loss/hidden": 3.0703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10299487039446831,
+      "step": 767
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.395768229166667,
+      "learning_rate": 9.602183219929371e-05,
+      "loss": 6.704,
+      "loss/crossentropy": 1.7421257197856903,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20691610127687454,
+      "step": 768
+    },
+    {
+      "epoch": 0.12816666666666668,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.934375,
+      "learning_rate": 9.601159236829352e-05,
+      "loss": 6.8235,
+      "loss/crossentropy": 1.439538598060608,
+      "loss/hidden": 3.9140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22591623291373253,
+      "step": 769
+    },
+    {
+      "epoch": 0.12833333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.7712890625,
+      "learning_rate": 9.600133992295433e-05,
+      "loss": 6.5832,
+      "loss/crossentropy": 1.238479107618332,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1591899823397398,
+      "step": 770
+    },
+    {
+      "epoch": 0.1285,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.93125,
+      "learning_rate": 9.599107486608689e-05,
+      "loss": 7.1446,
+      "loss/crossentropy": 1.714221179485321,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14866385608911514,
+      "step": 771
+    },
+    {
+      "epoch": 0.12866666666666668,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.9684895833333333,
+      "learning_rate": 9.598079720050544e-05,
+      "loss": 6.8982,
+      "loss/crossentropy": 1.3181781768798828,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16883035376667976,
+      "step": 772
+    },
+    {
+      "epoch": 0.12883333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.3559895833333333,
+      "learning_rate": 9.597050692902765e-05,
+      "loss": 6.9636,
+      "loss/crossentropy": 1.5021403431892395,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16215253993868828,
+      "step": 773
+    },
+    {
+      "epoch": 0.129,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.2264973958333334,
+      "learning_rate": 9.596020405447466e-05,
+      "loss": 6.8601,
+      "loss/crossentropy": 1.1908520609140396,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17125163599848747,
+      "step": 774
+    },
+    {
+      "epoch": 0.12916666666666668,
+      "grad_norm": 26.625,
+      "grad_norm_var": 2.3059895833333335,
+      "learning_rate": 9.594988857967106e-05,
+      "loss": 6.7215,
+      "loss/crossentropy": 0.911146990954876,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15889035165309906,
+      "step": 775
+    },
+    {
+      "epoch": 0.12933333333333333,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.2583333333333333,
+      "learning_rate": 9.593956050744492e-05,
+      "loss": 6.8107,
+      "loss/crossentropy": 1.6832085251808167,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23193244636058807,
+      "step": 776
+    },
+    {
+      "epoch": 0.1295,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.2541015625,
+      "learning_rate": 9.59292198406277e-05,
+      "loss": 6.6668,
+      "loss/crossentropy": 1.5619841068983078,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16906663961708546,
+      "step": 777
+    },
+    {
+      "epoch": 0.12966666666666668,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.1041666666666665,
+      "learning_rate": 9.591886658205438e-05,
+      "loss": 6.9829,
+      "loss/crossentropy": 1.7919222116470337,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19456206634640694,
+      "step": 778
+    },
+    {
+      "epoch": 0.12983333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.1270182291666666,
+      "learning_rate": 9.590850073456336e-05,
+      "loss": 8.0128,
+      "loss/crossentropy": 2.312351554632187,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23978988081216812,
+      "step": 779
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.8598307291666667,
+      "learning_rate": 9.589812230099649e-05,
+      "loss": 7.2349,
+      "loss/crossentropy": 1.8043695986270905,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17244477570056915,
+      "step": 780
+    },
+    {
+      "epoch": 0.13016666666666668,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.7559895833333334,
+      "learning_rate": 9.588773128419906e-05,
+      "loss": 6.6837,
+      "loss/crossentropy": 1.2657390087842941,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1250052247196436,
+      "step": 781
+    },
+    {
+      "epoch": 0.13033333333333333,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.7958333333333334,
+      "learning_rate": 9.587732768701986e-05,
+      "loss": 6.5026,
+      "loss/crossentropy": 1.3608160316944122,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19106113724410534,
+      "step": 782
+    },
+    {
+      "epoch": 0.1305,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.9639973958333334,
+      "learning_rate": 9.586691151231107e-05,
+      "loss": 6.8022,
+      "loss/crossentropy": 1.209273800253868,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1794354785233736,
+      "step": 783
+    },
+    {
+      "epoch": 0.13066666666666665,
+      "grad_norm": 35.25,
+      "grad_norm_var": 3.8520182291666667,
+      "learning_rate": 9.585648276292836e-05,
+      "loss": 7.2437,
+      "loss/crossentropy": 1.7829331159591675,
+      "loss/hidden": 3.640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30202513188123703,
+      "step": 784
+    },
+    {
+      "epoch": 0.13083333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 4.615625,
+      "learning_rate": 9.584604144173083e-05,
+      "loss": 6.6887,
+      "loss/crossentropy": 1.2763479948043823,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13859128206968307,
+      "step": 785
+    },
+    {
+      "epoch": 0.131,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.7181640625,
+      "learning_rate": 9.5835587551581e-05,
+      "loss": 6.6243,
+      "loss/crossentropy": 1.6887988448143005,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21134157478809357,
+      "step": 786
+    },
+    {
+      "epoch": 0.13116666666666665,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.803125,
+      "learning_rate": 9.58251210953449e-05,
+      "loss": 6.591,
+      "loss/crossentropy": 1.6435635685920715,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1841796226799488,
+      "step": 787
+    },
+    {
+      "epoch": 0.13133333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.804166666666666,
+      "learning_rate": 9.581464207589195e-05,
+      "loss": 6.7774,
+      "loss/crossentropy": 1.5818364322185516,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2561847046017647,
+      "step": 788
+    },
+    {
+      "epoch": 0.1315,
+      "grad_norm": 26.375,
+      "grad_norm_var": 5.713997395833333,
+      "learning_rate": 9.580415049609503e-05,
+      "loss": 6.982,
+      "loss/crossentropy": 2.383569300174713,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1847519651055336,
+      "step": 789
+    },
+    {
+      "epoch": 0.13166666666666665,
+      "grad_norm": 30.375,
+      "grad_norm_var": 5.6712890625,
+      "learning_rate": 9.579364635883048e-05,
+      "loss": 6.4862,
+      "loss/crossentropy": 1.369117945432663,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13245585188269615,
+      "step": 790
+    },
+    {
+      "epoch": 0.13183333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.759309895833334,
+      "learning_rate": 9.578312966697807e-05,
+      "loss": 7.1771,
+      "loss/crossentropy": 2.042441338300705,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22358595207333565,
+      "step": 791
+    },
+    {
+      "epoch": 0.132,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.177018229166666,
+      "learning_rate": 9.577260042342097e-05,
+      "loss": 6.8503,
+      "loss/crossentropy": 1.5456407964229584,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15839309990406036,
+      "step": 792
+    },
+    {
+      "epoch": 0.13216666666666665,
+      "grad_norm": 30.0,
+      "grad_norm_var": 4.041666666666667,
+      "learning_rate": 9.576205863104588e-05,
+      "loss": 6.7188,
+      "loss/crossentropy": 1.8247463256120682,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1572584118694067,
+      "step": 793
+    },
+    {
+      "epoch": 0.13233333333333333,
+      "grad_norm": 27.625,
+      "grad_norm_var": 4.6072265625,
+      "learning_rate": 9.575150429274287e-05,
+      "loss": 6.7608,
+      "loss/crossentropy": 1.0372321158647537,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14176041074097157,
+      "step": 794
+    },
+    {
+      "epoch": 0.1325,
+      "grad_norm": 26.25,
+      "grad_norm_var": 5.7119140625,
+      "learning_rate": 9.574093741140549e-05,
+      "loss": 6.6846,
+      "loss/crossentropy": 1.240045703947544,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14266852848231792,
+      "step": 795
+    },
+    {
+      "epoch": 0.13266666666666665,
+      "grad_norm": 30.375,
+      "grad_norm_var": 5.6634765625,
+      "learning_rate": 9.573035798993069e-05,
+      "loss": 6.9076,
+      "loss/crossentropy": 1.8248577117919922,
+      "loss/hidden": 3.859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23838410340249538,
+      "step": 796
+    },
+    {
+      "epoch": 0.13283333333333333,
+      "grad_norm": 4177526784.0,
+      "grad_norm_var": 1.0907331111261569e+18,
+      "learning_rate": 9.571976603121888e-05,
+      "loss": 7.2423,
+      "loss/crossentropy": 1.316101610660553,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2635743021965027,
+      "step": 797
+    },
+    {
+      "epoch": 0.133,
+      "grad_norm": 37.75,
+      "grad_norm_var": 1.0907331108128424e+18,
+      "learning_rate": 9.570916153817391e-05,
+      "loss": 6.223,
+      "loss/crossentropy": 1.5232540667057037,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1748909205198288,
+      "step": 798
+    },
+    {
+      "epoch": 0.13316666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.0907331107475685e+18,
+      "learning_rate": 9.569854451370307e-05,
+      "loss": 6.7389,
+      "loss/crossentropy": 1.303069368004799,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25480288080871105,
+      "step": 799
+    },
+    {
+      "epoch": 0.13333333333333333,
+      "grad_norm": 36.25,
+      "grad_norm_var": 1.0907331107127558e+18,
+      "learning_rate": 9.568791496071706e-05,
+      "loss": 7.1203,
+      "loss/crossentropy": 1.5637471973896027,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16657356172800064,
+      "step": 800
+    },
+    {
+      "epoch": 0.1335,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.0907331107736781e+18,
+      "learning_rate": 9.567727288213005e-05,
+      "loss": 6.4694,
+      "loss/crossentropy": 1.200748085975647,
+      "loss/hidden": 3.01953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10192291811108589,
+      "step": 801
+    },
+    {
+      "epoch": 0.13366666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.090733110817194e+18,
+      "learning_rate": 9.56666182808596e-05,
+      "loss": 6.7045,
+      "loss/crossentropy": 1.2447317093610764,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1798014249652624,
+      "step": 802
+    },
+    {
+      "epoch": 0.13383333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.0907331108215456e+18,
+      "learning_rate": 9.565595115982678e-05,
+      "loss": 6.9973,
+      "loss/crossentropy": 1.0941070914268494,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20473548583686352,
+      "step": 803
+    },
+    {
+      "epoch": 0.134,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.090733110838952e+18,
+      "learning_rate": 9.5645271521956e-05,
+      "loss": 6.9191,
+      "loss/crossentropy": 1.1080967336893082,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13076345063745975,
+      "step": 804
+    },
+    {
+      "epoch": 0.13416666666666666,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.0907331107736781e+18,
+      "learning_rate": 9.563457937017515e-05,
+      "loss": 6.3146,
+      "loss/crossentropy": 1.2985542491078377,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18152809981256723,
+      "step": 805
+    },
+    {
+      "epoch": 0.13433333333333333,
+      "grad_norm": 27.25,
+      "grad_norm_var": 1.0907331108824678e+18,
+      "learning_rate": 9.562387470741554e-05,
+      "loss": 6.3556,
+      "loss/crossentropy": 1.5479900687932968,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15538710355758667,
+      "step": 806
+    },
+    {
+      "epoch": 0.1345,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.0907331109607964e+18,
+      "learning_rate": 9.561315753661194e-05,
+      "loss": 7.2221,
+      "loss/crossentropy": 1.285927951335907,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2129192315042019,
+      "step": 807
+    },
+    {
+      "epoch": 0.13466666666666666,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.0907331109999608e+18,
+      "learning_rate": 9.560242786070249e-05,
+      "loss": 6.9297,
+      "loss/crossentropy": 1.8271296620368958,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20804137364029884,
+      "step": 808
+    },
+    {
+      "epoch": 0.13483333333333333,
+      "grad_norm": 25.875,
+      "grad_norm_var": 1.0907331111435633e+18,
+      "learning_rate": 9.55916856826288e-05,
+      "loss": 6.2083,
+      "loss/crossentropy": 1.7633541822433472,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1793544664978981,
+      "step": 809
+    },
+    {
+      "epoch": 0.135,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.0907331110347735e+18,
+      "learning_rate": 9.558093100533591e-05,
+      "loss": 6.6219,
+      "loss/crossentropy": 1.89547860622406,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18614793568849564,
+      "step": 810
+    },
+    {
+      "epoch": 0.13516666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.0907331108737646e+18,
+      "learning_rate": 9.557016383177227e-05,
+      "loss": 6.849,
+      "loss/crossentropy": 1.5208584070205688,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16671922616660595,
+      "step": 811
+    },
+    {
+      "epoch": 0.13533333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.0907331108911711e+18,
+      "learning_rate": 9.555938416488977e-05,
+      "loss": 6.7214,
+      "loss/crossentropy": 1.7651298642158508,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20645348727703094,
+      "step": 812
+    },
+    {
+      "epoch": 0.1355,
+      "grad_norm": 30.125,
+      "grad_norm_var": 9.2353515625,
+      "learning_rate": 9.55485920076437e-05,
+      "loss": 7.0837,
+      "loss/crossentropy": 0.8462116047739983,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12763971276581287,
+      "step": 813
+    },
+    {
+      "epoch": 0.13566666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 5.81875,
+      "learning_rate": 9.553778736299279e-05,
+      "loss": 6.8846,
+      "loss/crossentropy": 1.8222520351409912,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15145867317914963,
+      "step": 814
+    },
+    {
+      "epoch": 0.13583333333333333,
+      "grad_norm": 26.5,
+      "grad_norm_var": 6.02265625,
+      "learning_rate": 9.552697023389922e-05,
+      "loss": 6.3034,
+      "loss/crossentropy": 1.5191392451524734,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1477720383554697,
+      "step": 815
+    },
+    {
+      "epoch": 0.136,
+      "grad_norm": 25.625,
+      "grad_norm_var": 4.069205729166667,
+      "learning_rate": 9.551614062332856e-05,
+      "loss": 6.3766,
+      "loss/crossentropy": 1.2401023209095001,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1795745911076665,
+      "step": 816
+    },
+    {
+      "epoch": 0.13616666666666666,
+      "grad_norm": 27.375,
+      "grad_norm_var": 3.6958333333333333,
+      "learning_rate": 9.550529853424979e-05,
+      "loss": 6.3193,
+      "loss/crossentropy": 1.1493524312973022,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13225699588656425,
+      "step": 817
+    },
+    {
+      "epoch": 0.13633333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.3958333333333335,
+      "learning_rate": 9.549444396963534e-05,
+      "loss": 7.074,
+      "loss/crossentropy": 1.3883174806833267,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1946331448853016,
+      "step": 818
+    },
+    {
+      "epoch": 0.1365,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.3837890625,
+      "learning_rate": 9.548357693246105e-05,
+      "loss": 6.8662,
+      "loss/crossentropy": 1.4820697605609894,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1775229312479496,
+      "step": 819
+    },
+    {
+      "epoch": 0.13666666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.2353515625,
+      "learning_rate": 9.547269742570619e-05,
+      "loss": 6.7228,
+      "loss/crossentropy": 1.665278211236,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15727989003062248,
+      "step": 820
+    },
+    {
+      "epoch": 0.13683333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.5447265625,
+      "learning_rate": 9.546180545235344e-05,
+      "loss": 6.7407,
+      "loss/crossentropy": 1.2566642463207245,
+      "loss/hidden": 3.67578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18570634350180626,
+      "step": 821
+    },
+    {
+      "epoch": 0.137,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.3587890625,
+      "learning_rate": 9.545090101538887e-05,
+      "loss": 6.8989,
+      "loss/crossentropy": 1.448990821838379,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20253212377429008,
+      "step": 822
+    },
+    {
+      "epoch": 0.13716666666666666,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.3587890625,
+      "learning_rate": 9.543998411780201e-05,
+      "loss": 7.0963,
+      "loss/crossentropy": 1.3227804154157639,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15900520235300064,
+      "step": 823
+    },
+    {
+      "epoch": 0.13733333333333334,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.3181640625,
+      "learning_rate": 9.54290547625858e-05,
+      "loss": 6.8576,
+      "loss/crossentropy": 1.068933516740799,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12162080220878124,
+      "step": 824
+    },
+    {
+      "epoch": 0.1375,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.6973307291666666,
+      "learning_rate": 9.541811295273656e-05,
+      "loss": 6.8121,
+      "loss/crossentropy": 1.1542446911334991,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15490317158401012,
+      "step": 825
+    },
+    {
+      "epoch": 0.13766666666666666,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.51875,
+      "learning_rate": 9.540715869125407e-05,
+      "loss": 6.9212,
+      "loss/crossentropy": 1.4736111089587212,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14284480456262827,
+      "step": 826
+    },
+    {
+      "epoch": 0.13783333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.489518229166667,
+      "learning_rate": 9.53961919811415e-05,
+      "loss": 6.842,
+      "loss/crossentropy": 1.7014836072921753,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17102410830557346,
+      "step": 827
+    },
+    {
+      "epoch": 0.138,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.559830729166667,
+      "learning_rate": 9.538521282540542e-05,
+      "loss": 6.5574,
+      "loss/crossentropy": 1.2606604620814323,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17054615262895823,
+      "step": 828
+    },
+    {
+      "epoch": 0.13816666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.4843098958333334,
+      "learning_rate": 9.537422122705585e-05,
+      "loss": 6.4654,
+      "loss/crossentropy": 1.3435598015785217,
+      "loss/hidden": 3.09375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11619688849896193,
+      "step": 829
+    },
+    {
+      "epoch": 0.13833333333333334,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.028580729166667,
+      "learning_rate": 9.536321718910619e-05,
+      "loss": 6.3738,
+      "loss/crossentropy": 1.076637014746666,
+      "loss/hidden": 3.0703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10972944647073746,
+      "step": 830
+    },
+    {
+      "epoch": 0.1385,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.7104166666666667,
+      "learning_rate": 9.535220071457325e-05,
+      "loss": 6.6418,
+      "loss/crossentropy": 1.7037672102451324,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20552153512835503,
+      "step": 831
+    },
+    {
+      "epoch": 0.13866666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 0.9379557291666667,
+      "learning_rate": 9.534117180647728e-05,
+      "loss": 6.547,
+      "loss/crossentropy": 1.7507948279380798,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20301245152950287,
+      "step": 832
+    },
+    {
+      "epoch": 0.13883333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.4916666666666667,
+      "learning_rate": 9.533013046784189e-05,
+      "loss": 6.952,
+      "loss/crossentropy": 1.8267524242401123,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16741746850311756,
+      "step": 833
+    },
+    {
+      "epoch": 0.139,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.5893229166666667,
+      "learning_rate": 9.531907670169415e-05,
+      "loss": 6.6224,
+      "loss/crossentropy": 1.732950508594513,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18124941363930702,
+      "step": 834
+    },
+    {
+      "epoch": 0.13916666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.5504557291666667,
+      "learning_rate": 9.530801051106449e-05,
+      "loss": 6.8669,
+      "loss/crossentropy": 1.1107327789068222,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16977784968912601,
+      "step": 835
+    },
+    {
+      "epoch": 0.13933333333333334,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.54140625,
+      "learning_rate": 9.52969318989868e-05,
+      "loss": 6.7344,
+      "loss/crossentropy": 1.324594423174858,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14172261767089367,
+      "step": 836
+    },
+    {
+      "epoch": 0.1395,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.40390625,
+      "learning_rate": 9.528584086849832e-05,
+      "loss": 6.7147,
+      "loss/crossentropy": 1.3007304072380066,
+      "loss/hidden": 3.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11234837584197521,
+      "step": 837
+    },
+    {
+      "epoch": 0.13966666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.39765625,
+      "learning_rate": 9.527473742263973e-05,
+      "loss": 6.7973,
+      "loss/crossentropy": 1.5427119061350822,
+      "loss/hidden": 3.82421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24523010104894638,
+      "step": 838
+    },
+    {
+      "epoch": 0.13983333333333334,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.4768229166666667,
+      "learning_rate": 9.526362156445507e-05,
+      "loss": 6.6099,
+      "loss/crossentropy": 1.4119697362184525,
+      "loss/hidden": 3.640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17216363549232483,
+      "step": 839
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.154166666666667,
+      "learning_rate": 9.525249329699188e-05,
+      "loss": 7.2271,
+      "loss/crossentropy": 1.9238676130771637,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23542418703436852,
+      "step": 840
+    },
+    {
+      "epoch": 0.14016666666666666,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.2622395833333333,
+      "learning_rate": 9.524135262330098e-05,
+      "loss": 7.499,
+      "loss/crossentropy": 2.2861229181289673,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19216257333755493,
+      "step": 841
+    },
+    {
+      "epoch": 0.14033333333333334,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.4268229166666666,
+      "learning_rate": 9.523019954643669e-05,
+      "loss": 6.5638,
+      "loss/crossentropy": 1.3084105104207993,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12469115294516087,
+      "step": 842
+    },
+    {
+      "epoch": 0.1405,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.3759765625,
+      "learning_rate": 9.521903406945664e-05,
+      "loss": 6.8629,
+      "loss/crossentropy": 1.6192159950733185,
+      "loss/hidden": 3.76953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20565154030919075,
+      "step": 843
+    },
+    {
+      "epoch": 0.14066666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.3587890625,
+      "learning_rate": 9.520785619542196e-05,
+      "loss": 6.7475,
+      "loss/crossentropy": 1.7351672053337097,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1903889775276184,
+      "step": 844
+    },
+    {
+      "epoch": 0.14083333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.6875,
+      "learning_rate": 9.519666592739709e-05,
+      "loss": 6.9226,
+      "loss/crossentropy": 1.676860898733139,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17689999006688595,
+      "step": 845
+    },
+    {
+      "epoch": 0.141,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.4947265625,
+      "learning_rate": 9.518546326844993e-05,
+      "loss": 6.9766,
+      "loss/crossentropy": 2.105147272348404,
+      "loss/hidden": 3.85546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26574863865971565,
+      "step": 846
+    },
+    {
+      "epoch": 0.14116666666666666,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.5192057291666665,
+      "learning_rate": 9.517424822165175e-05,
+      "loss": 6.7578,
+      "loss/crossentropy": 1.2014988660812378,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22761327400803566,
+      "step": 847
+    },
+    {
+      "epoch": 0.14133333333333334,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.633072916666667,
+      "learning_rate": 9.516302079007719e-05,
+      "loss": 7.1276,
+      "loss/crossentropy": 1.5993420332670212,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2370535023510456,
+      "step": 848
+    },
+    {
+      "epoch": 0.1415,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.9518229166666667,
+      "learning_rate": 9.515178097680437e-05,
+      "loss": 6.9106,
+      "loss/crossentropy": 1.61234450340271,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22792143002152443,
+      "step": 849
+    },
+    {
+      "epoch": 0.14166666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.8442057291666667,
+      "learning_rate": 9.51405287849147e-05,
+      "loss": 6.9403,
+      "loss/crossentropy": 1.6368253529071808,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2259569726884365,
+      "step": 850
+    },
+    {
+      "epoch": 0.14183333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.004622395833333,
+      "learning_rate": 9.512926421749304e-05,
+      "loss": 6.9868,
+      "loss/crossentropy": 1.7041020095348358,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19391244649887085,
+      "step": 851
+    },
+    {
+      "epoch": 0.142,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.9785807291666666,
+      "learning_rate": 9.511798727762764e-05,
+      "loss": 6.4189,
+      "loss/crossentropy": 1.373700737953186,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15583642199635506,
+      "step": 852
+    },
+    {
+      "epoch": 0.14216666666666666,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.0004557291666667,
+      "learning_rate": 9.510669796841014e-05,
+      "loss": 6.9529,
+      "loss/crossentropy": 1.485058069229126,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2598130479454994,
+      "step": 853
+    },
+    {
+      "epoch": 0.14233333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.0104166666666665,
+      "learning_rate": 9.509539629293558e-05,
+      "loss": 6.7833,
+      "loss/crossentropy": 1.528167724609375,
+      "loss/hidden": 3.90234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35026970505714417,
+      "step": 854
+    },
+    {
+      "epoch": 0.1425,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.8785807291666667,
+      "learning_rate": 9.508408225430237e-05,
+      "loss": 6.6249,
+      "loss/crossentropy": 1.8074993342161179,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17642150819301605,
+      "step": 855
+    },
+    {
+      "epoch": 0.14266666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.3910807291666667,
+      "learning_rate": 9.507275585561229e-05,
+      "loss": 6.9863,
+      "loss/crossentropy": 1.498419925570488,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14689111337065697,
+      "step": 856
+    },
+    {
+      "epoch": 0.14283333333333334,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.4879557291666667,
+      "learning_rate": 9.506141709997057e-05,
+      "loss": 6.565,
+      "loss/crossentropy": 1.1097244322299957,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13454408384859562,
+      "step": 857
+    },
+    {
+      "epoch": 0.143,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.4143229166666667,
+      "learning_rate": 9.505006599048579e-05,
+      "loss": 6.8645,
+      "loss/crossentropy": 2.0309173464775085,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21611754968762398,
+      "step": 858
+    },
+    {
+      "epoch": 0.14316666666666666,
+      "grad_norm": 26.625,
+      "grad_norm_var": 2.0416666666666665,
+      "learning_rate": 9.503870253026991e-05,
+      "loss": 6.9038,
+      "loss/crossentropy": 1.1781870126724243,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13299256563186646,
+      "step": 859
+    },
+    {
+      "epoch": 0.14333333333333334,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.265625,
+      "learning_rate": 9.50273267224383e-05,
+      "loss": 6.5733,
+      "loss/crossentropy": 1.3335058093070984,
+      "loss/hidden": 3.109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12909270077943802,
+      "step": 860
+    },
+    {
+      "epoch": 0.1435,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.7160807291666667,
+      "learning_rate": 9.501593857010969e-05,
+      "loss": 6.4952,
+      "loss/crossentropy": 1.3636210933327675,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1412863340228796,
+      "step": 861
+    },
+    {
+      "epoch": 0.14366666666666666,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.8488932291666667,
+      "learning_rate": 9.50045380764062e-05,
+      "loss": 6.9494,
+      "loss/crossentropy": 1.5361263453960419,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25922345742583275,
+      "step": 862
+    },
+    {
+      "epoch": 0.14383333333333334,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.8739583333333334,
+      "learning_rate": 9.499312524445336e-05,
+      "loss": 7.1393,
+      "loss/crossentropy": 1.6427059173583984,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17301714606583118,
+      "step": 863
+    },
+    {
+      "epoch": 0.144,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.8416666666666666,
+      "learning_rate": 9.498170007738005e-05,
+      "loss": 6.7791,
+      "loss/crossentropy": 1.8128132969141006,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15439187735319138,
+      "step": 864
+    },
+    {
+      "epoch": 0.14416666666666667,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.1186848958333333,
+      "learning_rate": 9.497026257831855e-05,
+      "loss": 6.5659,
+      "loss/crossentropy": 2.2501683235168457,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21063398197293282,
+      "step": 865
+    },
+    {
+      "epoch": 0.14433333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.9905598958333333,
+      "learning_rate": 9.495881275040453e-05,
+      "loss": 6.6395,
+      "loss/crossentropy": 1.250039003789425,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12767360918223858,
+      "step": 866
+    },
+    {
+      "epoch": 0.1445,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.53125,
+      "learning_rate": 9.494735059677699e-05,
+      "loss": 6.7208,
+      "loss/crossentropy": 1.3687303513288498,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21333901584148407,
+      "step": 867
+    },
+    {
+      "epoch": 0.14466666666666667,
+      "grad_norm": 27.5,
+      "grad_norm_var": 1.6572265625,
+      "learning_rate": 9.493587612057837e-05,
+      "loss": 6.4006,
+      "loss/crossentropy": 1.7404839247465134,
+      "loss/hidden": 3.13671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1315847784280777,
+      "step": 868
+    },
+    {
+      "epoch": 0.14483333333333334,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.6510416666666667,
+      "learning_rate": 9.492438932495444e-05,
+      "loss": 6.8423,
+      "loss/crossentropy": 1.8640455901622772,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18806974962353706,
+      "step": 869
+    },
+    {
+      "epoch": 0.145,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.1546223958333335,
+      "learning_rate": 9.491289021305441e-05,
+      "loss": 7.3314,
+      "loss/crossentropy": 2.117801755666733,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2098713479936123,
+      "step": 870
+    },
+    {
+      "epoch": 0.14516666666666667,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.7309895833333333,
+      "learning_rate": 9.490137878803079e-05,
+      "loss": 6.7026,
+      "loss/crossentropy": 0.8712976723909378,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19407054781913757,
+      "step": 871
+    },
+    {
+      "epoch": 0.14533333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.0093098958333333,
+      "learning_rate": 9.488985505303951e-05,
+      "loss": 6.7614,
+      "loss/crossentropy": 1.5071170665323734,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20746826520189643,
+      "step": 872
+    },
+    {
+      "epoch": 0.1455,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.0768229166666665,
+      "learning_rate": 9.487831901123988e-05,
+      "loss": 6.8808,
+      "loss/crossentropy": 1.5924786031246185,
+      "loss/hidden": 3.7421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23435182869434357,
+      "step": 873
+    },
+    {
+      "epoch": 0.14566666666666667,
+      "grad_norm": 28.25,
+      "grad_norm_var": 2.097916666666667,
+      "learning_rate": 9.486677066579456e-05,
+      "loss": 6.7799,
+      "loss/crossentropy": 1.5978667587041855,
+      "loss/hidden": 3.14453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14294573664665222,
+      "step": 874
+    },
+    {
+      "epoch": 0.14583333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.6997395833333333,
+      "learning_rate": 9.485521001986962e-05,
+      "loss": 6.7737,
+      "loss/crossentropy": 1.509243592619896,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1823038011789322,
+      "step": 875
+    },
+    {
+      "epoch": 0.146,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.5624348958333334,
+      "learning_rate": 9.484363707663442e-05,
+      "loss": 6.6774,
+      "loss/crossentropy": 1.0428966879844666,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17252317816019058,
+      "step": 876
+    },
+    {
+      "epoch": 0.14616666666666667,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.7958333333333334,
+      "learning_rate": 9.483205183926181e-05,
+      "loss": 6.588,
+      "loss/crossentropy": 0.949081763625145,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10428016912192106,
+      "step": 877
+    },
+    {
+      "epoch": 0.14633333333333334,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.7489583333333334,
+      "learning_rate": 9.48204543109279e-05,
+      "loss": 6.8537,
+      "loss/crossentropy": 1.753394454717636,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21356584876775742,
+      "step": 878
+    },
+    {
+      "epoch": 0.1465,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.7291015625,
+      "learning_rate": 9.480884449481225e-05,
+      "loss": 6.8986,
+      "loss/crossentropy": 1.2782315462827682,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15598896704614162,
+      "step": 879
+    },
+    {
+      "epoch": 0.14666666666666667,
+      "grad_norm": 27.0,
+      "grad_norm_var": 2.066666666666667,
+      "learning_rate": 9.479722239409775e-05,
+      "loss": 6.6731,
+      "loss/crossentropy": 1.4414279013872147,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15914863720536232,
+      "step": 880
+    },
+    {
+      "epoch": 0.14683333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.7645182291666666,
+      "learning_rate": 9.478558801197065e-05,
+      "loss": 6.6562,
+      "loss/crossentropy": 1.2711293995380402,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13687402941286564,
+      "step": 881
+    },
+    {
+      "epoch": 0.147,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.7768229166666667,
+      "learning_rate": 9.47739413516206e-05,
+      "loss": 6.571,
+      "loss/crossentropy": 1.1690870523452759,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1315490622073412,
+      "step": 882
+    },
+    {
+      "epoch": 0.14716666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.934375,
+      "learning_rate": 9.476228241624059e-05,
+      "loss": 7.6772,
+      "loss/crossentropy": 1.6540156453847885,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19950690492987633,
+      "step": 883
+    },
+    {
+      "epoch": 0.14733333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.6942057291666666,
+      "learning_rate": 9.475061120902698e-05,
+      "loss": 6.6939,
+      "loss/crossentropy": 1.4847566783428192,
+      "loss/hidden": 3.12890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1577866580337286,
+      "step": 884
+    },
+    {
+      "epoch": 0.1475,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.6869140625,
+      "learning_rate": 9.473892773317952e-05,
+      "loss": 6.7253,
+      "loss/crossentropy": 1.3767812550067902,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2304423786699772,
+      "step": 885
+    },
+    {
+      "epoch": 0.14766666666666667,
+      "grad_norm": 27.5,
+      "grad_norm_var": 1.5895182291666667,
+      "learning_rate": 9.472723199190125e-05,
+      "loss": 6.7192,
+      "loss/crossentropy": 1.6085245311260223,
+      "loss/hidden": 3.12890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16864492185413837,
+      "step": 886
+    },
+    {
+      "epoch": 0.14783333333333334,
+      "grad_norm": 27.75,
+      "grad_norm_var": 1.6580729166666666,
+      "learning_rate": 9.47155239883987e-05,
+      "loss": 6.7612,
+      "loss/crossentropy": 1.3260292783379555,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14614624343812466,
+      "step": 887
+    },
+    {
+      "epoch": 0.148,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.4869140625,
+      "learning_rate": 9.470380372588162e-05,
+      "loss": 6.8686,
+      "loss/crossentropy": 1.4136118963360786,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24818306602537632,
+      "step": 888
+    },
+    {
+      "epoch": 0.14816666666666667,
+      "grad_norm": 27.5,
+      "grad_norm_var": 1.5785807291666667,
+      "learning_rate": 9.46920712075632e-05,
+      "loss": 6.8164,
+      "loss/crossentropy": 1.8791348934173584,
+      "loss/hidden": 4.0078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27732761949300766,
+      "step": 889
+    },
+    {
+      "epoch": 0.14833333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.6181640625,
+      "learning_rate": 9.468032643665998e-05,
+      "loss": 6.5978,
+      "loss/crossentropy": 1.3766923397779465,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14428966492414474,
+      "step": 890
+    },
+    {
+      "epoch": 0.1485,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.6181640625,
+      "learning_rate": 9.466856941639188e-05,
+      "loss": 6.7711,
+      "loss/crossentropy": 1.2270091325044632,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16888568736612797,
+      "step": 891
+    },
+    {
+      "epoch": 0.14866666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.5764973958333333,
+      "learning_rate": 9.465680014998213e-05,
+      "loss": 7.069,
+      "loss/crossentropy": 1.7997345924377441,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17343921586871147,
+      "step": 892
+    },
+    {
+      "epoch": 0.14883333333333335,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.3952473958333333,
+      "learning_rate": 9.464501864065735e-05,
+      "loss": 6.9376,
+      "loss/crossentropy": 1.4494410380721092,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29051431082189083,
+      "step": 893
+    },
+    {
+      "epoch": 0.149,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.36875,
+      "learning_rate": 9.46332248916475e-05,
+      "loss": 6.652,
+      "loss/crossentropy": 1.5664034485816956,
+      "loss/hidden": 3.6875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35951946675777435,
+      "step": 894
+    },
+    {
+      "epoch": 0.14916666666666667,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.61015625,
+      "learning_rate": 9.46214189061859e-05,
+      "loss": 6.356,
+      "loss/crossentropy": 1.777071237564087,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22660009935498238,
+      "step": 895
+    },
+    {
+      "epoch": 0.14933333333333335,
+      "grad_norm": 25.5,
+      "grad_norm_var": 2.14765625,
+      "learning_rate": 9.460960068750924e-05,
+      "loss": 6.7566,
+      "loss/crossentropy": 1.2509669959545135,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1698252372443676,
+      "step": 896
+    },
+    {
+      "epoch": 0.1495,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.153580729166667,
+      "learning_rate": 9.459777023885755e-05,
+      "loss": 6.7189,
+      "loss/crossentropy": 1.807364284992218,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18868035450577736,
+      "step": 897
+    },
+    {
+      "epoch": 0.14966666666666667,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.155989583333333,
+      "learning_rate": 9.458592756347419e-05,
+      "loss": 6.9761,
+      "loss/crossentropy": 1.522624596953392,
+      "loss/hidden": 3.0078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12400520220398903,
+      "step": 898
+    },
+    {
+      "epoch": 0.14983333333333335,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.86015625,
+      "learning_rate": 9.457407266460593e-05,
+      "loss": 6.648,
+      "loss/crossentropy": 1.2911832630634308,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16489013843238354,
+      "step": 899
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.111458333333333,
+      "learning_rate": 9.456220554550285e-05,
+      "loss": 7.0601,
+      "loss/crossentropy": 2.048378899693489,
+      "loss/hidden": 3.72265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25268127024173737,
+      "step": 900
+    },
+    {
+      "epoch": 0.15016666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.9311848958333333,
+      "learning_rate": 9.45503262094184e-05,
+      "loss": 6.5816,
+      "loss/crossentropy": 1.9151304364204407,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1854175217449665,
+      "step": 901
+    },
+    {
+      "epoch": 0.15033333333333335,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.84140625,
+      "learning_rate": 9.453843465960933e-05,
+      "loss": 6.5712,
+      "loss/crossentropy": 1.3903828710317612,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12931624799966812,
+      "step": 902
+    },
+    {
+      "epoch": 0.1505,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.7931640625,
+      "learning_rate": 9.45265308993358e-05,
+      "loss": 6.6937,
+      "loss/crossentropy": 0.884873166680336,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2095654346048832,
+      "step": 903
+    },
+    {
+      "epoch": 0.15066666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 1.9103515625,
+      "learning_rate": 9.451461493186129e-05,
+      "loss": 6.4043,
+      "loss/crossentropy": 1.432923972606659,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1620604507625103,
+      "step": 904
+    },
+    {
+      "epoch": 0.15083333333333335,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.5634765625,
+      "learning_rate": 9.450268676045262e-05,
+      "loss": 7.1785,
+      "loss/crossentropy": 1.5224296301603317,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21086937934160233,
+      "step": 905
+    },
+    {
+      "epoch": 0.151,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.55390625,
+      "learning_rate": 9.449074638837999e-05,
+      "loss": 6.8505,
+      "loss/crossentropy": 2.0315501987934113,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1887502819299698,
+      "step": 906
+    },
+    {
+      "epoch": 0.15116666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.55390625,
+      "learning_rate": 9.447879381891692e-05,
+      "loss": 6.8196,
+      "loss/crossentropy": 2.0016221404075623,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21317322179675102,
+      "step": 907
+    },
+    {
+      "epoch": 0.15133333333333332,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.5561848958333333,
+      "learning_rate": 9.446682905534023e-05,
+      "loss": 6.6458,
+      "loss/crossentropy": 1.3033087104558945,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18597162887454033,
+      "step": 908
+    },
+    {
+      "epoch": 0.1515,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.4884765625,
+      "learning_rate": 9.445485210093017e-05,
+      "loss": 6.8986,
+      "loss/crossentropy": 1.48745097219944,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17175030894577503,
+      "step": 909
+    },
+    {
+      "epoch": 0.15166666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.47265625,
+      "learning_rate": 9.444286295897028e-05,
+      "loss": 6.796,
+      "loss/crossentropy": 1.6483173668384552,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1788303665816784,
+      "step": 910
+    },
+    {
+      "epoch": 0.15183333333333332,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.3372395833333335,
+      "learning_rate": 9.443086163274745e-05,
+      "loss": 7.2361,
+      "loss/crossentropy": 1.427602395415306,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25849100202322006,
+      "step": 911
+    },
+    {
+      "epoch": 0.152,
+      "grad_norm": 27.625,
+      "grad_norm_var": 1.7207682291666666,
+      "learning_rate": 9.44188481255519e-05,
+      "loss": 6.928,
+      "loss/crossentropy": 1.5816201120615005,
+      "loss/hidden": 3.87109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30472470447421074,
+      "step": 912
+    },
+    {
+      "epoch": 0.15216666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 1.8643229166666666,
+      "learning_rate": 9.440682244067724e-05,
+      "loss": 6.9598,
+      "loss/crossentropy": 1.9130527079105377,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2092103734612465,
+      "step": 913
+    },
+    {
+      "epoch": 0.15233333333333332,
+      "grad_norm": 27.625,
+      "grad_norm_var": 1.94140625,
+      "learning_rate": 9.439478458142033e-05,
+      "loss": 6.9606,
+      "loss/crossentropy": 1.0294331312179565,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15341288968920708,
+      "step": 914
+    },
+    {
+      "epoch": 0.1525,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.9457682291666667,
+      "learning_rate": 9.438273455108144e-05,
+      "loss": 6.7431,
+      "loss/crossentropy": 1.5968154817819595,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1610046774148941,
+      "step": 915
+    },
+    {
+      "epoch": 0.15266666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.9457682291666667,
+      "learning_rate": 9.437067235296418e-05,
+      "loss": 7.1008,
+      "loss/crossentropy": 1.4984619095921516,
+      "loss/hidden": 3.74609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18942087143659592,
+      "step": 916
+    },
+    {
+      "epoch": 0.15283333333333332,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.983268229166667,
+      "learning_rate": 9.43585979903754e-05,
+      "loss": 7.1829,
+      "loss/crossentropy": 1.5713898837566376,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1644783318042755,
+      "step": 917
+    },
+    {
+      "epoch": 0.153,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.9643229166666667,
+      "learning_rate": 9.434651146662543e-05,
+      "loss": 6.81,
+      "loss/crossentropy": 1.022684097290039,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16458085924386978,
+      "step": 918
+    },
+    {
+      "epoch": 0.15316666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.6426432291666666,
+      "learning_rate": 9.433441278502783e-05,
+      "loss": 7.3599,
+      "loss/crossentropy": 1.779938131570816,
+      "loss/hidden": 3.78125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33398906514048576,
+      "step": 919
+    },
+    {
+      "epoch": 0.15333333333333332,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.4056640625,
+      "learning_rate": 9.43223019488995e-05,
+      "loss": 7.3089,
+      "loss/crossentropy": 1.5696324706077576,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16729436069726944,
+      "step": 920
+    },
+    {
+      "epoch": 0.1535,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.9541015625,
+      "learning_rate": 9.431017896156074e-05,
+      "loss": 7.0047,
+      "loss/crossentropy": 1.3372326716780663,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17373233661055565,
+      "step": 921
+    },
+    {
+      "epoch": 0.15366666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.7791015625,
+      "learning_rate": 9.42980438263351e-05,
+      "loss": 6.863,
+      "loss/crossentropy": 1.808920294046402,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24913391843438148,
+      "step": 922
+    },
+    {
+      "epoch": 0.15383333333333332,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.809375,
+      "learning_rate": 9.428589654654951e-05,
+      "loss": 6.683,
+      "loss/crossentropy": 1.5314701944589615,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14396323077380657,
+      "step": 923
+    },
+    {
+      "epoch": 0.154,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.8330729166666666,
+      "learning_rate": 9.42737371255342e-05,
+      "loss": 6.6809,
+      "loss/crossentropy": 1.5369913876056671,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1440065409988165,
+      "step": 924
+    },
+    {
+      "epoch": 0.15416666666666667,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.731705729166667,
+      "learning_rate": 9.426156556662276e-05,
+      "loss": 6.6101,
+      "loss/crossentropy": 1.5807256698608398,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15959331579506397,
+      "step": 925
+    },
+    {
+      "epoch": 0.15433333333333332,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.7348307291666667,
+      "learning_rate": 9.42493818731521e-05,
+      "loss": 6.7241,
+      "loss/crossentropy": 1.9071035385131836,
+      "loss/hidden": 3.7109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2186364307999611,
+      "step": 926
+    },
+    {
+      "epoch": 0.1545,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.6080729166666665,
+      "learning_rate": 9.423718604846243e-05,
+      "loss": 7.0448,
+      "loss/crossentropy": 1.6829068064689636,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26727957278490067,
+      "step": 927
+    },
+    {
+      "epoch": 0.15466666666666667,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.3643229166666666,
+      "learning_rate": 9.422497809589731e-05,
+      "loss": 7.1008,
+      "loss/crossentropy": 1.8336281776428223,
+      "loss/hidden": 3.7734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2665562555193901,
+      "step": 928
+    },
+    {
+      "epoch": 0.15483333333333332,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.044205729166667,
+      "learning_rate": 9.421275801880362e-05,
+      "loss": 6.7488,
+      "loss/crossentropy": 1.202999860048294,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1640343815088272,
+      "step": 929
+    },
+    {
+      "epoch": 0.155,
+      "grad_norm": 26.875,
+      "grad_norm_var": 2.2895182291666667,
+      "learning_rate": 9.420052582053157e-05,
+      "loss": 6.3745,
+      "loss/crossentropy": 0.9804428443312645,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1364411050453782,
+      "step": 930
+    },
+    {
+      "epoch": 0.15516666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.3393229166666667,
+      "learning_rate": 9.418828150443469e-05,
+      "loss": 7.3503,
+      "loss/crossentropy": 1.0129078775644302,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3142009563744068,
+      "step": 931
+    },
+    {
+      "epoch": 0.15533333333333332,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.5004557291666667,
+      "learning_rate": 9.417602507386981e-05,
+      "loss": 6.3487,
+      "loss/crossentropy": 1.5488710403442383,
+      "loss/hidden": 3.078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13036078214645386,
+      "step": 932
+    },
+    {
+      "epoch": 0.1555,
+      "grad_norm": 27.875,
+      "grad_norm_var": 1.9393229166666666,
+      "learning_rate": 9.416375653219709e-05,
+      "loss": 6.5817,
+      "loss/crossentropy": 0.9361780658364296,
+      "loss/hidden": 3.109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1352296294644475,
+      "step": 933
+    },
+    {
+      "epoch": 0.15566666666666668,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.189322916666667,
+      "learning_rate": 9.415147588278005e-05,
+      "loss": 6.886,
+      "loss/crossentropy": 1.7425316870212555,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15156837925314903,
+      "step": 934
+    },
+    {
+      "epoch": 0.15583333333333332,
+      "grad_norm": 25.875,
+      "grad_norm_var": 2.139518229166667,
+      "learning_rate": 9.413918312898551e-05,
+      "loss": 6.7067,
+      "loss/crossentropy": 1.8844291865825653,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15378071926534176,
+      "step": 935
+    },
+    {
+      "epoch": 0.156,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.2427083333333333,
+      "learning_rate": 9.412687827418356e-05,
+      "loss": 6.555,
+      "loss/crossentropy": 1.3648638874292374,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20318766683340073,
+      "step": 936
+    },
+    {
+      "epoch": 0.15616666666666668,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.0113932291666665,
+      "learning_rate": 9.411456132174767e-05,
+      "loss": 6.8804,
+      "loss/crossentropy": 0.8283918276429176,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09471303224563599,
+      "step": 937
+    },
+    {
+      "epoch": 0.15633333333333332,
+      "grad_norm": 27.375,
+      "grad_norm_var": 2.096809895833333,
+      "learning_rate": 9.410223227505459e-05,
+      "loss": 6.8104,
+      "loss/crossentropy": 2.2247671484947205,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.159748837351799,
+      "step": 938
+    },
+    {
+      "epoch": 0.1565,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.003059895833333,
+      "learning_rate": 9.408989113748442e-05,
+      "loss": 6.5397,
+      "loss/crossentropy": 1.5369086861610413,
+      "loss/hidden": 3.078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13781077414751053,
+      "step": 939
+    },
+    {
+      "epoch": 0.15666666666666668,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.2264973958333334,
+      "learning_rate": 9.407753791242051e-05,
+      "loss": 6.4877,
+      "loss/crossentropy": 1.2110577672719955,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12467943131923676,
+      "step": 940
+    },
+    {
+      "epoch": 0.15683333333333332,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.1684895833333333,
+      "learning_rate": 9.40651726032496e-05,
+      "loss": 7.0721,
+      "loss/crossentropy": 1.717575654387474,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16085078194737434,
+      "step": 941
+    },
+    {
+      "epoch": 0.157,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.5483723958333333,
+      "learning_rate": 9.405279521336173e-05,
+      "loss": 7.1052,
+      "loss/crossentropy": 1.2131355553865433,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16941608116030693,
+      "step": 942
+    },
+    {
+      "epoch": 0.15716666666666668,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.5921223958333335,
+      "learning_rate": 9.404040574615018e-05,
+      "loss": 6.6737,
+      "loss/crossentropy": 1.4986532032489777,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24445821717381477,
+      "step": 943
+    },
+    {
+      "epoch": 0.15733333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.5921223958333335,
+      "learning_rate": 9.402800420501164e-05,
+      "loss": 6.9161,
+      "loss/crossentropy": 1.035700723528862,
+      "loss/hidden": 3.12109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11124230176210403,
+      "step": 944
+    },
+    {
+      "epoch": 0.1575,
+      "grad_norm": 25.875,
+      "grad_norm_var": 3.0561848958333333,
+      "learning_rate": 9.401559059334601e-05,
+      "loss": 6.0012,
+      "loss/crossentropy": 1.1177880614995956,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17776408419013023,
+      "step": 945
+    },
+    {
+      "epoch": 0.15766666666666668,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.0962890625,
+      "learning_rate": 9.400316491455661e-05,
+      "loss": 6.8133,
+      "loss/crossentropy": 1.3670192509889603,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18806476704776287,
+      "step": 946
+    },
+    {
+      "epoch": 0.15783333333333333,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.74765625,
+      "learning_rate": 9.399072717204995e-05,
+      "loss": 6.6337,
+      "loss/crossentropy": 2.049343913793564,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19486809894442558,
+      "step": 947
+    },
+    {
+      "epoch": 0.158,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.70625,
+      "learning_rate": 9.397827736923596e-05,
+      "loss": 6.8354,
+      "loss/crossentropy": 1.4676995277404785,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22139647649601102,
+      "step": 948
+    },
+    {
+      "epoch": 0.15816666666666668,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.068489583333333,
+      "learning_rate": 9.396581550952781e-05,
+      "loss": 7.1093,
+      "loss/crossentropy": 1.7508991807699203,
+      "loss/hidden": 3.078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14620482921600342,
+      "step": 949
+    },
+    {
+      "epoch": 0.15833333333333333,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.9520833333333334,
+      "learning_rate": 9.395334159634199e-05,
+      "loss": 6.5735,
+      "loss/crossentropy": 1.998562604188919,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17252274602651596,
+      "step": 950
+    },
+    {
+      "epoch": 0.1585,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.4775390625,
+      "learning_rate": 9.394085563309827e-05,
+      "loss": 6.9826,
+      "loss/crossentropy": 1.6761377900838852,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.152450829744339,
+      "step": 951
+    },
+    {
+      "epoch": 0.15866666666666668,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.4436848958333335,
+      "learning_rate": 9.392835762321977e-05,
+      "loss": 6.7069,
+      "loss/crossentropy": 1.9230597019195557,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18309792131185532,
+      "step": 952
+    },
+    {
+      "epoch": 0.15883333333333333,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.4322916666666665,
+      "learning_rate": 9.391584757013289e-05,
+      "loss": 6.6781,
+      "loss/crossentropy": 2.1945230662822723,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1544794701039791,
+      "step": 953
+    },
+    {
+      "epoch": 0.159,
+      "grad_norm": 27.375,
+      "grad_norm_var": 2.4322916666666665,
+      "learning_rate": 9.390332547726733e-05,
+      "loss": 6.4855,
+      "loss/crossentropy": 1.84402334690094,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16222894191741943,
+      "step": 954
+    },
+    {
+      "epoch": 0.15916666666666668,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.3603515625,
+      "learning_rate": 9.389079134805609e-05,
+      "loss": 6.7115,
+      "loss/crossentropy": 1.1176575869321823,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1567247435450554,
+      "step": 955
+    },
+    {
+      "epoch": 0.15933333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.0337890625,
+      "learning_rate": 9.387824518593546e-05,
+      "loss": 6.8158,
+      "loss/crossentropy": 2.139141410589218,
+      "loss/hidden": 3.8125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2895619533956051,
+      "step": 956
+    },
+    {
+      "epoch": 0.1595,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.0369140625,
+      "learning_rate": 9.386568699434508e-05,
+      "loss": 6.8399,
+      "loss/crossentropy": 1.9488290548324585,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18183168396353722,
+      "step": 957
+    },
+    {
+      "epoch": 0.15966666666666668,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.357291666666667,
+      "learning_rate": 9.385311677672781e-05,
+      "loss": 7.3172,
+      "loss/crossentropy": 1.1987807005643845,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1368956658989191,
+      "step": 958
+    },
+    {
+      "epoch": 0.15983333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.2483723958333335,
+      "learning_rate": 9.384053453652986e-05,
+      "loss": 6.7999,
+      "loss/crossentropy": 2.046382397413254,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18021916970610619,
+      "step": 959
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.325,
+      "learning_rate": 9.382794027720073e-05,
+      "loss": 6.5911,
+      "loss/crossentropy": 1.4518917500972748,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.137871028855443,
+      "step": 960
+    },
+    {
+      "epoch": 0.16016666666666668,
+      "grad_norm": 27.0,
+      "grad_norm_var": 1.9634765625,
+      "learning_rate": 9.381533400219318e-05,
+      "loss": 6.4323,
+      "loss/crossentropy": 1.4430282711982727,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12539849989116192,
+      "step": 961
+    },
+    {
+      "epoch": 0.16033333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.1478515625,
+      "learning_rate": 9.380271571496334e-05,
+      "loss": 6.7533,
+      "loss/crossentropy": 1.4864182621240616,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21022829227149487,
+      "step": 962
+    },
+    {
+      "epoch": 0.1605,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.1650390625,
+      "learning_rate": 9.379008541897054e-05,
+      "loss": 7.4136,
+      "loss/crossentropy": 1.9163076877593994,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26462791860103607,
+      "step": 963
+    },
+    {
+      "epoch": 0.16066666666666668,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.467643229166667,
+      "learning_rate": 9.377744311767746e-05,
+      "loss": 6.4287,
+      "loss/crossentropy": 1.5390331447124481,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1931270845234394,
+      "step": 964
+    },
+    {
+      "epoch": 0.16083333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.363541666666667,
+      "learning_rate": 9.376478881455009e-05,
+      "loss": 6.6009,
+      "loss/crossentropy": 1.6977374702692032,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1623922735452652,
+      "step": 965
+    },
+    {
+      "epoch": 0.161,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.3160807291666665,
+      "learning_rate": 9.375212251305763e-05,
+      "loss": 6.3185,
+      "loss/crossentropy": 0.9074634611606598,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21587281860411167,
+      "step": 966
+    },
+    {
+      "epoch": 0.16116666666666668,
+      "grad_norm": 28.25,
+      "grad_norm_var": 2.1567057291666667,
+      "learning_rate": 9.373944421667265e-05,
+      "loss": 6.8387,
+      "loss/crossentropy": 1.536196306347847,
+      "loss/hidden": 3.05078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12623077537864447,
+      "step": 967
+    },
+    {
+      "epoch": 0.16133333333333333,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.1020182291666667,
+      "learning_rate": 9.372675392887096e-05,
+      "loss": 6.8656,
+      "loss/crossentropy": 2.036444216966629,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15584169328212738,
+      "step": 968
+    },
+    {
+      "epoch": 0.1615,
+      "grad_norm": 28.25,
+      "grad_norm_var": 2.122330729166667,
+      "learning_rate": 9.371405165313169e-05,
+      "loss": 7.0526,
+      "loss/crossentropy": 1.2288314998149872,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20249300450086594,
+      "step": 969
+    },
+    {
+      "epoch": 0.16166666666666665,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.9885416666666667,
+      "learning_rate": 9.370133739293723e-05,
+      "loss": 6.4385,
+      "loss/crossentropy": 1.4051976948976517,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26594360545277596,
+      "step": 970
+    },
+    {
+      "epoch": 0.16183333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.3775390625,
+      "learning_rate": 9.368861115177327e-05,
+      "loss": 7.4186,
+      "loss/crossentropy": 1.9679397344589233,
+      "loss/hidden": 4.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30216044932603836,
+      "step": 971
+    },
+    {
+      "epoch": 0.162,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.51015625,
+      "learning_rate": 9.367587293312878e-05,
+      "loss": 6.3767,
+      "loss/crossentropy": 1.4228357076644897,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12103093415498734,
+      "step": 972
+    },
+    {
+      "epoch": 0.16216666666666665,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.88515625,
+      "learning_rate": 9.366312274049602e-05,
+      "loss": 7.139,
+      "loss/crossentropy": 1.3934602439403534,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16450956650078297,
+      "step": 973
+    },
+    {
+      "epoch": 0.16233333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.444205729166667,
+      "learning_rate": 9.36503605773705e-05,
+      "loss": 7.1216,
+      "loss/crossentropy": 1.4134599715471268,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16329437587410212,
+      "step": 974
+    },
+    {
+      "epoch": 0.1625,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.662955729166667,
+      "learning_rate": 9.36375864472511e-05,
+      "loss": 6.2965,
+      "loss/crossentropy": 0.9992646798491478,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10369177348911762,
+      "step": 975
+    },
+    {
+      "epoch": 0.16266666666666665,
+      "grad_norm": 26.875,
+      "grad_norm_var": 2.859375,
+      "learning_rate": 9.362480035363986e-05,
+      "loss": 6.6671,
+      "loss/crossentropy": 1.825451374053955,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1982804238796234,
+      "step": 976
+    },
+    {
+      "epoch": 0.16283333333333333,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.6775390625,
+      "learning_rate": 9.36120023000422e-05,
+      "loss": 6.6047,
+      "loss/crossentropy": 0.8018264323472977,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1146076861768961,
+      "step": 977
+    },
+    {
+      "epoch": 0.163,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.6113932291666666,
+      "learning_rate": 9.359919228996674e-05,
+      "loss": 7.3768,
+      "loss/crossentropy": 2.1425884664058685,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18394936248660088,
+      "step": 978
+    },
+    {
+      "epoch": 0.16316666666666665,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.6768229166666666,
+      "learning_rate": 9.358637032692545e-05,
+      "loss": 6.6483,
+      "loss/crossentropy": 1.8683583438396454,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19553732126951218,
+      "step": 979
+    },
+    {
+      "epoch": 0.16333333333333333,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.539322916666667,
+      "learning_rate": 9.357353641443354e-05,
+      "loss": 6.6568,
+      "loss/crossentropy": 1.1269803941249847,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12698210403323174,
+      "step": 980
+    },
+    {
+      "epoch": 0.1635,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.5247395833333335,
+      "learning_rate": 9.356069055600948e-05,
+      "loss": 6.7423,
+      "loss/crossentropy": 1.7101454734802246,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2579207345843315,
+      "step": 981
+    },
+    {
+      "epoch": 0.16366666666666665,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.530143229166667,
+      "learning_rate": 9.354783275517504e-05,
+      "loss": 6.978,
+      "loss/crossentropy": 1.6930875182151794,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18442372232675552,
+      "step": 982
+    },
+    {
+      "epoch": 0.16383333333333333,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.6059895833333333,
+      "learning_rate": 9.353496301545529e-05,
+      "loss": 6.0959,
+      "loss/crossentropy": 2.0503551065921783,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1571796014904976,
+      "step": 983
+    },
+    {
+      "epoch": 0.164,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.3291015625,
+      "learning_rate": 9.352208134037851e-05,
+      "loss": 6.9039,
+      "loss/crossentropy": 2.170955538749695,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1941094473004341,
+      "step": 984
+    },
+    {
+      "epoch": 0.16416666666666666,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.4514973958333335,
+      "learning_rate": 9.35091877334763e-05,
+      "loss": 7.1528,
+      "loss/crossentropy": 1.5367806255817413,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18657927587628365,
+      "step": 985
+    },
+    {
+      "epoch": 0.16433333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.4518229166666665,
+      "learning_rate": 9.349628219828349e-05,
+      "loss": 6.7108,
+      "loss/crossentropy": 0.8706261143088341,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11337267328053713,
+      "step": 986
+    },
+    {
+      "epoch": 0.1645,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.082291666666667,
+      "learning_rate": 9.348336473833823e-05,
+      "loss": 6.5357,
+      "loss/crossentropy": 1.13887557387352,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1355858026072383,
+      "step": 987
+    },
+    {
+      "epoch": 0.16466666666666666,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.9155598958333333,
+      "learning_rate": 9.347043535718192e-05,
+      "loss": 6.5553,
+      "loss/crossentropy": 1.5605327934026718,
+      "loss/hidden": 3.109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13332156650722027,
+      "step": 988
+    },
+    {
+      "epoch": 0.16483333333333333,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.7030598958333334,
+      "learning_rate": 9.34574940583592e-05,
+      "loss": 7.0307,
+      "loss/crossentropy": 2.0721123665571213,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20332932844758034,
+      "step": 989
+    },
+    {
+      "epoch": 0.165,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.5559895833333335,
+      "learning_rate": 9.344454084541803e-05,
+      "loss": 7.2336,
+      "loss/crossentropy": 1.893361508846283,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17474518530070782,
+      "step": 990
+    },
+    {
+      "epoch": 0.16516666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.7375,
+      "learning_rate": 9.343157572190957e-05,
+      "loss": 6.8883,
+      "loss/crossentropy": 1.6536499708890915,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1391207305714488,
+      "step": 991
+    },
+    {
+      "epoch": 0.16533333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.408072916666667,
+      "learning_rate": 9.341859869138831e-05,
+      "loss": 6.7059,
+      "loss/crossentropy": 0.8432480245828629,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09941447153687477,
+      "step": 992
+    },
+    {
+      "epoch": 0.1655,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.289322916666667,
+      "learning_rate": 9.340560975741197e-05,
+      "loss": 6.9982,
+      "loss/crossentropy": 2.248078227043152,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19019538909196854,
+      "step": 993
+    },
+    {
+      "epoch": 0.16566666666666666,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.1125,
+      "learning_rate": 9.339260892354153e-05,
+      "loss": 6.7605,
+      "loss/crossentropy": 1.3738857507705688,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.140726363286376,
+      "step": 994
+    },
+    {
+      "epoch": 0.16583333333333333,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.3,
+      "learning_rate": 9.337959619334125e-05,
+      "loss": 7.0867,
+      "loss/crossentropy": 1.4559247940778732,
+      "loss/hidden": 3.63671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16540184617042542,
+      "step": 995
+    },
+    {
+      "epoch": 0.166,
+      "grad_norm": 27.375,
+      "grad_norm_var": 2.2718098958333335,
+      "learning_rate": 9.336657157037866e-05,
+      "loss": 6.4514,
+      "loss/crossentropy": 1.8479040265083313,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15281882509589195,
+      "step": 996
+    },
+    {
+      "epoch": 0.16616666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.5247395833333335,
+      "learning_rate": 9.33535350582245e-05,
+      "loss": 7.0539,
+      "loss/crossentropy": 1.9026743173599243,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1550293043255806,
+      "step": 997
+    },
+    {
+      "epoch": 0.16633333333333333,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.6302083333333335,
+      "learning_rate": 9.334048666045285e-05,
+      "loss": 6.5069,
+      "loss/crossentropy": 1.2742085456848145,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16048939153552055,
+      "step": 998
+    },
+    {
+      "epoch": 0.1665,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.503059895833333,
+      "learning_rate": 9.332742638064094e-05,
+      "loss": 7.0699,
+      "loss/crossentropy": 2.0430028438568115,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16402114555239677,
+      "step": 999
+    },
+    {
+      "epoch": 0.16666666666666666,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.96015625,
+      "learning_rate": 9.331435422236938e-05,
+      "loss": 6.8397,
+      "loss/crossentropy": 1.5021632313728333,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24582071229815483,
+      "step": 1000
+    },
+    {
+      "epoch": 0.16683333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.8202473958333334,
+      "learning_rate": 9.330127018922194e-05,
+      "loss": 6.9787,
+      "loss/crossentropy": 1.2862660512328148,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1655043950304389,
+      "step": 1001
+    },
+    {
+      "epoch": 0.167,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.8205729166666667,
+      "learning_rate": 9.328817428478569e-05,
+      "loss": 6.551,
+      "loss/crossentropy": 1.8244322463870049,
+      "loss/hidden": 3.10546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1513009099289775,
+      "step": 1002
+    },
+    {
+      "epoch": 0.16716666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.0098307291666666,
+      "learning_rate": 9.327506651265095e-05,
+      "loss": 7.0498,
+      "loss/crossentropy": 1.703669548034668,
+      "loss/hidden": 3.7265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2555901035666466,
+      "step": 1003
+    },
+    {
+      "epoch": 0.16733333333333333,
+      "grad_norm": 26.25,
+      "grad_norm_var": 2.559375,
+      "learning_rate": 9.32619468764113e-05,
+      "loss": 6.4082,
+      "loss/crossentropy": 1.5771573781967163,
+      "loss/hidden": 3.8125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2212020792067051,
+      "step": 1004
+    },
+    {
+      "epoch": 0.1675,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.35625,
+      "learning_rate": 9.324881537966354e-05,
+      "loss": 6.5945,
+      "loss/crossentropy": 2.1410138607025146,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18434720113873482,
+      "step": 1005
+    },
+    {
+      "epoch": 0.16766666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.411393229166667,
+      "learning_rate": 9.323567202600776e-05,
+      "loss": 6.6545,
+      "loss/crossentropy": 1.1437709033489227,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1774177122861147,
+      "step": 1006
+    },
+    {
+      "epoch": 0.16783333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.138541666666667,
+      "learning_rate": 9.322251681904728e-05,
+      "loss": 6.7476,
+      "loss/crossentropy": 1.3867665529251099,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23731738328933716,
+      "step": 1007
+    },
+    {
+      "epoch": 0.168,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.314322916666667,
+      "learning_rate": 9.320934976238867e-05,
+      "loss": 6.8831,
+      "loss/crossentropy": 1.3709579110145569,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15835164114832878,
+      "step": 1008
+    },
+    {
+      "epoch": 0.16816666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.421875,
+      "learning_rate": 9.319617085964176e-05,
+      "loss": 6.555,
+      "loss/crossentropy": 1.3165605813264847,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22091330960392952,
+      "step": 1009
+    },
+    {
+      "epoch": 0.16833333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.4184895833333333,
+      "learning_rate": 9.318298011441964e-05,
+      "loss": 6.8214,
+      "loss/crossentropy": 1.665429800748825,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17766010761260986,
+      "step": 1010
+    },
+    {
+      "epoch": 0.1685,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.230208333333333,
+      "learning_rate": 9.316977753033859e-05,
+      "loss": 6.5599,
+      "loss/crossentropy": 1.6475956588983536,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18197769299149513,
+      "step": 1011
+    },
+    {
+      "epoch": 0.16866666666666666,
+      "grad_norm": 25.875,
+      "grad_norm_var": 2.777083333333333,
+      "learning_rate": 9.31565631110182e-05,
+      "loss": 6.5457,
+      "loss/crossentropy": 1.4782060533761978,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1904252003878355,
+      "step": 1012
+    },
+    {
+      "epoch": 0.16883333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.3247395833333333,
+      "learning_rate": 9.314333686008125e-05,
+      "loss": 6.8884,
+      "loss/crossentropy": 1.7327529937028885,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31483493745326996,
+      "step": 1013
+    },
+    {
+      "epoch": 0.169,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.2358723958333333,
+      "learning_rate": 9.313009878115381e-05,
+      "loss": 6.7042,
+      "loss/crossentropy": 1.6143233627080917,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18221945315599442,
+      "step": 1014
+    },
+    {
+      "epoch": 0.16916666666666666,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.2114583333333333,
+      "learning_rate": 9.31168488778652e-05,
+      "loss": 6.4314,
+      "loss/crossentropy": 1.290844403207302,
+      "loss/hidden": 3.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09984238259494305,
+      "step": 1015
+    },
+    {
+      "epoch": 0.16933333333333334,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.175,
+      "learning_rate": 9.310358715384793e-05,
+      "loss": 6.2406,
+      "loss/crossentropy": 1.1085650324821472,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17697811871767044,
+      "step": 1016
+    },
+    {
+      "epoch": 0.1695,
+      "grad_norm": 27.375,
+      "grad_norm_var": 2.3364583333333333,
+      "learning_rate": 9.309031361273775e-05,
+      "loss": 6.3544,
+      "loss/crossentropy": 1.3015744537115097,
+      "loss/hidden": 3.01953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11535845696926117,
+      "step": 1017
+    },
+    {
+      "epoch": 0.16966666666666666,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.662239583333333,
+      "learning_rate": 9.307702825817373e-05,
+      "loss": 6.5172,
+      "loss/crossentropy": 1.759951412677765,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18253059312701225,
+      "step": 1018
+    },
+    {
+      "epoch": 0.16983333333333334,
+      "grad_norm": 27.375,
+      "grad_norm_var": 2.501822916666667,
+      "learning_rate": 9.306373109379809e-05,
+      "loss": 6.0286,
+      "loss/crossentropy": 0.70027294754982,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09618740063160658,
+      "step": 1019
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 25.25,
+      "grad_norm_var": 2.8872395833333333,
+      "learning_rate": 9.305042212325634e-05,
+      "loss": 6.7102,
+      "loss/crossentropy": 1.299631044268608,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22817913629114628,
+      "step": 1020
+    },
+    {
+      "epoch": 0.17016666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.006705729166667,
+      "learning_rate": 9.30371013501972e-05,
+      "loss": 7.0757,
+      "loss/crossentropy": 1.1401289254426956,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16579823568463326,
+      "step": 1021
+    },
+    {
+      "epoch": 0.17033333333333334,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.8629557291666665,
+      "learning_rate": 9.302376877827263e-05,
+      "loss": 6.6939,
+      "loss/crossentropy": 1.1806171089410782,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15253576263785362,
+      "step": 1022
+    },
+    {
+      "epoch": 0.1705,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.7643229166666665,
+      "learning_rate": 9.301042441113783e-05,
+      "loss": 7.164,
+      "loss/crossentropy": 1.7891227304935455,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1749817430973053,
+      "step": 1023
+    },
+    {
+      "epoch": 0.17066666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.8051432291666667,
+      "learning_rate": 9.299706825245126e-05,
+      "loss": 6.8655,
+      "loss/crossentropy": 1.7026661485433578,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14925392344594002,
+      "step": 1024
+    },
+    {
+      "epoch": 0.17083333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.5322916666666666,
+      "learning_rate": 9.298370030587456e-05,
+      "loss": 6.6227,
+      "loss/crossentropy": 1.6058360189199448,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18634888343513012,
+      "step": 1025
+    },
+    {
+      "epoch": 0.171,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.465625,
+      "learning_rate": 9.297032057507264e-05,
+      "loss": 6.9121,
+      "loss/crossentropy": 1.4227608889341354,
+      "loss/hidden": 3.73828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20639438182115555,
+      "step": 1026
+    },
+    {
+      "epoch": 0.17116666666666666,
+      "grad_norm": 25.375,
+      "grad_norm_var": 3.0125,
+      "learning_rate": 9.295692906371363e-05,
+      "loss": 6.3686,
+      "loss/crossentropy": 1.4738323539495468,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1534047992900014,
+      "step": 1027
+    },
+    {
+      "epoch": 0.17133333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.6837890625,
+      "learning_rate": 9.294352577546888e-05,
+      "loss": 6.7086,
+      "loss/crossentropy": 1.8850750923156738,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17676223628222942,
+      "step": 1028
+    },
+    {
+      "epoch": 0.1715,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.582747395833333,
+      "learning_rate": 9.293011071401298e-05,
+      "loss": 6.4771,
+      "loss/crossentropy": 1.3900472968816757,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15907294303178787,
+      "step": 1029
+    },
+    {
+      "epoch": 0.17166666666666666,
+      "grad_norm": 26.125,
+      "grad_norm_var": 2.746809895833333,
+      "learning_rate": 9.291668388302374e-05,
+      "loss": 6.3026,
+      "loss/crossentropy": 1.7207022309303284,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16812225244939327,
+      "step": 1030
+    },
+    {
+      "epoch": 0.17183333333333334,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.7796223958333335,
+      "learning_rate": 9.290324528618224e-05,
+      "loss": 6.737,
+      "loss/crossentropy": 1.1087007820606232,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15751628205180168,
+      "step": 1031
+    },
+    {
+      "epoch": 0.172,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.8436848958333334,
+      "learning_rate": 9.28897949271727e-05,
+      "loss": 7.1036,
+      "loss/crossentropy": 1.5463591516017914,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22198307141661644,
+      "step": 1032
+    },
+    {
+      "epoch": 0.17216666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.11875,
+      "learning_rate": 9.287633280968261e-05,
+      "loss": 6.7399,
+      "loss/crossentropy": 1.8197236359119415,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14384458400309086,
+      "step": 1033
+    },
+    {
+      "epoch": 0.17233333333333334,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.9317057291666666,
+      "learning_rate": 9.286285893740274e-05,
+      "loss": 6.9903,
+      "loss/crossentropy": 1.4223038405179977,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11996994912624359,
+      "step": 1034
+    },
+    {
+      "epoch": 0.1725,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.8457682291666666,
+      "learning_rate": 9.284937331402697e-05,
+      "loss": 6.5254,
+      "loss/crossentropy": 1.8838764131069183,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16456597298383713,
+      "step": 1035
+    },
+    {
+      "epoch": 0.17266666666666666,
+      "grad_norm": 28.25,
+      "grad_norm_var": 2.0863932291666667,
+      "learning_rate": 9.283587594325249e-05,
+      "loss": 7.2951,
+      "loss/crossentropy": 1.690007284283638,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1627235785126686,
+      "step": 1036
+    },
+    {
+      "epoch": 0.17283333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.50390625,
+      "learning_rate": 9.282236682877967e-05,
+      "loss": 6.7527,
+      "loss/crossentropy": 1.6431153267621994,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1587188020348549,
+      "step": 1037
+    },
+    {
+      "epoch": 0.173,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.6337890625,
+      "learning_rate": 9.280884597431212e-05,
+      "loss": 6.8026,
+      "loss/crossentropy": 1.761688381433487,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28126591071486473,
+      "step": 1038
+    },
+    {
+      "epoch": 0.17316666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.64140625,
+      "learning_rate": 9.279531338355666e-05,
+      "loss": 6.8327,
+      "loss/crossentropy": 1.7254878282546997,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23531349375844002,
+      "step": 1039
+    },
+    {
+      "epoch": 0.17333333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.18515625,
+      "learning_rate": 9.27817690602233e-05,
+      "loss": 7.1946,
+      "loss/crossentropy": 1.578020840883255,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24730910547077656,
+      "step": 1040
+    },
+    {
+      "epoch": 0.1735,
+      "grad_norm": 28.375,
+      "grad_norm_var": 3.2129557291666666,
+      "learning_rate": 9.276821300802534e-05,
+      "loss": 6.5041,
+      "loss/crossentropy": 1.6880528777837753,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17321892082691193,
+      "step": 1041
+    },
+    {
+      "epoch": 0.17366666666666666,
+      "grad_norm": 27.625,
+      "grad_norm_var": 3.3129557291666667,
+      "learning_rate": 9.27546452306792e-05,
+      "loss": 6.369,
+      "loss/crossentropy": 1.6766884177923203,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24857143685221672,
+      "step": 1042
+    },
+    {
+      "epoch": 0.17383333333333334,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.3978515625,
+      "learning_rate": 9.274106573190459e-05,
+      "loss": 6.7366,
+      "loss/crossentropy": 1.4286645650863647,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17977101914584637,
+      "step": 1043
+    },
+    {
+      "epoch": 0.174,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.6645833333333333,
+      "learning_rate": 9.272747451542441e-05,
+      "loss": 6.2943,
+      "loss/crossentropy": 1.0265646129846573,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13961484655737877,
+      "step": 1044
+    },
+    {
+      "epoch": 0.17416666666666666,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.725455729166667,
+      "learning_rate": 9.271387158496476e-05,
+      "loss": 6.9447,
+      "loss/crossentropy": 1.6943024098873138,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2832085192203522,
+      "step": 1045
+    },
+    {
+      "epoch": 0.17433333333333334,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.1285807291666665,
+      "learning_rate": 9.270025694425497e-05,
+      "loss": 7.1521,
+      "loss/crossentropy": 1.3861359506845474,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2078065276145935,
+      "step": 1046
+    },
+    {
+      "epoch": 0.1745,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.1973307291666666,
+      "learning_rate": 9.268663059702753e-05,
+      "loss": 6.6814,
+      "loss/crossentropy": 0.8704834654927254,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10470980685204268,
+      "step": 1047
+    },
+    {
+      "epoch": 0.17466666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.269791666666667,
+      "learning_rate": 9.267299254701824e-05,
+      "loss": 6.7021,
+      "loss/crossentropy": 1.3742065727710724,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1371091976761818,
+      "step": 1048
+    },
+    {
+      "epoch": 0.17483333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.0875,
+      "learning_rate": 9.265934279796602e-05,
+      "loss": 7.4382,
+      "loss/crossentropy": 1.875411033630371,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2505655810236931,
+      "step": 1049
+    },
+    {
+      "epoch": 0.175,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.25,
+      "learning_rate": 9.264568135361302e-05,
+      "loss": 6.8456,
+      "loss/crossentropy": 1.2539885640144348,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18397588655352592,
+      "step": 1050
+    },
+    {
+      "epoch": 0.17516666666666666,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.328125,
+      "learning_rate": 9.263200821770461e-05,
+      "loss": 7.1828,
+      "loss/crossentropy": 1.8055228888988495,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19033204391598701,
+      "step": 1051
+    },
+    {
+      "epoch": 0.17533333333333334,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.174934895833333,
+      "learning_rate": 9.261832339398938e-05,
+      "loss": 6.8599,
+      "loss/crossentropy": 2.30813068151474,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1916029490530491,
+      "step": 1052
+    },
+    {
+      "epoch": 0.1755,
+      "grad_norm": 28.375,
+      "grad_norm_var": 3.0184895833333334,
+      "learning_rate": 9.260462688621905e-05,
+      "loss": 6.3348,
+      "loss/crossentropy": 1.7150506675243378,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15869835764169693,
+      "step": 1053
+    },
+    {
+      "epoch": 0.17566666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.0184895833333334,
+      "learning_rate": 9.259091869814864e-05,
+      "loss": 6.7278,
+      "loss/crossentropy": 1.5105496421456337,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18605177477002144,
+      "step": 1054
+    },
+    {
+      "epoch": 0.17583333333333334,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.064583333333333,
+      "learning_rate": 9.257719883353631e-05,
+      "loss": 6.3432,
+      "loss/crossentropy": 1.0289728492498398,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13243985548615456,
+      "step": 1055
+    },
+    {
+      "epoch": 0.176,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.53125,
+      "learning_rate": 9.256346729614342e-05,
+      "loss": 7.0523,
+      "loss/crossentropy": 1.4567533880472183,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14434895291924477,
+      "step": 1056
+    },
+    {
+      "epoch": 0.17616666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.4853515625,
+      "learning_rate": 9.254972408973461e-05,
+      "loss": 6.9902,
+      "loss/crossentropy": 1.5392310470342636,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1923016645014286,
+      "step": 1057
+    },
+    {
+      "epoch": 0.17633333333333334,
+      "grad_norm": 27.0,
+      "grad_norm_var": 2.6705729166666665,
+      "learning_rate": 9.253596921807759e-05,
+      "loss": 6.607,
+      "loss/crossentropy": 1.7072811126708984,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15615876764059067,
+      "step": 1058
+    },
+    {
+      "epoch": 0.1765,
+      "grad_norm": 26.5,
+      "grad_norm_var": 3.2259765625,
+      "learning_rate": 9.252220268494337e-05,
+      "loss": 6.7976,
+      "loss/crossentropy": 1.2664166390895844,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20431069657206535,
+      "step": 1059
+    },
+    {
+      "epoch": 0.17666666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.8942057291666665,
+      "learning_rate": 9.250842449410611e-05,
+      "loss": 6.3915,
+      "loss/crossentropy": 1.7454079389572144,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16402440890669823,
+      "step": 1060
+    },
+    {
+      "epoch": 0.17683333333333334,
+      "grad_norm": 28.25,
+      "grad_norm_var": 2.848893229166667,
+      "learning_rate": 9.249463464934321e-05,
+      "loss": 6.7293,
+      "loss/crossentropy": 0.8675141632556915,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10183609277009964,
+      "step": 1061
+    },
+    {
+      "epoch": 0.177,
+      "grad_norm": 26.25,
+      "grad_norm_var": 3.50625,
+      "learning_rate": 9.248083315443518e-05,
+      "loss": 6.311,
+      "loss/crossentropy": 1.3239275515079498,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1458367221057415,
+      "step": 1062
+    },
+    {
+      "epoch": 0.17716666666666667,
+      "grad_norm": 27.375,
+      "grad_norm_var": 3.66015625,
+      "learning_rate": 9.246702001316583e-05,
+      "loss": 6.8022,
+      "loss/crossentropy": 1.54161936044693,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20145102217793465,
+      "step": 1063
+    },
+    {
+      "epoch": 0.17733333333333334,
+      "grad_norm": 27.625,
+      "grad_norm_var": 3.7462890625,
+      "learning_rate": 9.245319522932209e-05,
+      "loss": 6.7071,
+      "loss/crossentropy": 1.0704002380371094,
+      "loss/hidden": 2.94140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1460615172982216,
+      "step": 1064
+    },
+    {
+      "epoch": 0.1775,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.6759765625,
+      "learning_rate": 9.24393588066941e-05,
+      "loss": 6.828,
+      "loss/crossentropy": 1.2201762646436691,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12776055745780468,
+      "step": 1065
+    },
+    {
+      "epoch": 0.17766666666666667,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.2291015625,
+      "learning_rate": 9.242551074907519e-05,
+      "loss": 7.0489,
+      "loss/crossentropy": 1.1659608334302902,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12117037083953619,
+      "step": 1066
+    },
+    {
+      "epoch": 0.17783333333333334,
+      "grad_norm": 28.0,
+      "grad_norm_var": 1.8270833333333334,
+      "learning_rate": 9.241165106026189e-05,
+      "loss": 6.9507,
+      "loss/crossentropy": 1.7780919075012207,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18536414206027985,
+      "step": 1067
+    },
+    {
+      "epoch": 0.178,
+      "grad_norm": 28.0,
+      "grad_norm_var": 1.7030598958333334,
+      "learning_rate": 9.239777974405393e-05,
+      "loss": 6.4759,
+      "loss/crossentropy": 1.3345106691122055,
+      "loss/hidden": 3.04296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12685726210474968,
+      "step": 1068
+    },
+    {
+      "epoch": 0.17816666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.7645182291666666,
+      "learning_rate": 9.238389680425416e-05,
+      "loss": 6.8943,
+      "loss/crossentropy": 2.1657201945781708,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16807297617197037,
+      "step": 1069
+    },
+    {
+      "epoch": 0.17833333333333334,
+      "grad_norm": 26.75,
+      "grad_norm_var": 1.6160807291666666,
+      "learning_rate": 9.237000224466872e-05,
+      "loss": 6.7557,
+      "loss/crossentropy": 1.762036293745041,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24878711998462677,
+      "step": 1070
+    },
+    {
+      "epoch": 0.1785,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.5957682291666666,
+      "learning_rate": 9.235609606910687e-05,
+      "loss": 6.7522,
+      "loss/crossentropy": 1.821693480014801,
+      "loss/hidden": 3.76171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3557448387145996,
+      "step": 1071
+    },
+    {
+      "epoch": 0.17866666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.3268229166666667,
+      "learning_rate": 9.234217828138104e-05,
+      "loss": 7.1196,
+      "loss/crossentropy": 1.8081410229206085,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20041708461940289,
+      "step": 1072
+    },
+    {
+      "epoch": 0.17883333333333334,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.0113932291666667,
+      "learning_rate": 9.23282488853069e-05,
+      "loss": 6.2645,
+      "loss/crossentropy": 1.2684693336486816,
+      "loss/hidden": 3.12890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14612330868840218,
+      "step": 1073
+    },
+    {
+      "epoch": 0.179,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.0497395833333334,
+      "learning_rate": 9.231430788470326e-05,
+      "loss": 6.3198,
+      "loss/crossentropy": 1.3656416982412338,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1677461415529251,
+      "step": 1074
+    },
+    {
+      "epoch": 0.17916666666666667,
+      "grad_norm": 26.875,
+      "grad_norm_var": 0.9764973958333333,
+      "learning_rate": 9.230035528339211e-05,
+      "loss": 6.5185,
+      "loss/crossentropy": 1.0957149416208267,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1348528154194355,
+      "step": 1075
+    },
+    {
+      "epoch": 0.17933333333333334,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.1337890625,
+      "learning_rate": 9.228639108519868e-05,
+      "loss": 7.3823,
+      "loss/crossentropy": 1.8525986969470978,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17726282589137554,
+      "step": 1076
+    },
+    {
+      "epoch": 0.1795,
+      "grad_norm": 34.75,
+      "grad_norm_var": 3.7947265625,
+      "learning_rate": 9.227241529395127e-05,
+      "loss": 6.1566,
+      "loss/crossentropy": 1.348332166671753,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18071135319769382,
+      "step": 1077
+    },
+    {
+      "epoch": 0.17966666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.4643229166666667,
+      "learning_rate": 9.225842791348149e-05,
+      "loss": 6.563,
+      "loss/crossentropy": 1.061979040503502,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15336202550679445,
+      "step": 1078
+    },
+    {
+      "epoch": 0.17983333333333335,
+      "grad_norm": 27.75,
+      "grad_norm_var": 3.3988932291666667,
+      "learning_rate": 9.224442894762401e-05,
+      "loss": 6.5426,
+      "loss/crossentropy": 1.5598224252462387,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21232370659708977,
+      "step": 1079
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 28.5,
+      "grad_norm_var": 3.3,
+      "learning_rate": 9.223041840021674e-05,
+      "loss": 6.7749,
+      "loss/crossentropy": 1.7184421867132187,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15667770244181156,
+      "step": 1080
+    },
+    {
+      "epoch": 0.18016666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.564322916666667,
+      "learning_rate": 9.221639627510076e-05,
+      "loss": 6.6496,
+      "loss/crossentropy": 1.9194800555706024,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19381767511367798,
+      "step": 1081
+    },
+    {
+      "epoch": 0.18033333333333335,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.5518229166666666,
+      "learning_rate": 9.220236257612031e-05,
+      "loss": 6.8568,
+      "loss/crossentropy": 1.6562037765979767,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19203035160899162,
+      "step": 1082
+    },
+    {
+      "epoch": 0.1805,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.069791666666666,
+      "learning_rate": 9.21883173071228e-05,
+      "loss": 6.8741,
+      "loss/crossentropy": 1.8158580362796783,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.178849708288908,
+      "step": 1083
+    },
+    {
+      "epoch": 0.18066666666666667,
+      "grad_norm": 43.0,
+      "grad_norm_var": 15.444791666666667,
+      "learning_rate": 9.217426047195882e-05,
+      "loss": 6.9982,
+      "loss/crossentropy": 1.7587174326181412,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20827142894268036,
+      "step": 1084
+    },
+    {
+      "epoch": 0.18083333333333335,
+      "grad_norm": 28.125,
+      "grad_norm_var": 15.693489583333333,
+      "learning_rate": 9.216019207448217e-05,
+      "loss": 6.5129,
+      "loss/crossentropy": 1.8337259590625763,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1768171191215515,
+      "step": 1085
+    },
+    {
+      "epoch": 0.181,
+      "grad_norm": 27.75,
+      "grad_norm_var": 15.295572916666666,
+      "learning_rate": 9.214611211854974e-05,
+      "loss": 6.5272,
+      "loss/crossentropy": 2.1704297438263893,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15514697041362524,
+      "step": 1086
+    },
+    {
+      "epoch": 0.18116666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 15.100455729166667,
+      "learning_rate": 9.213202060802161e-05,
+      "loss": 6.8904,
+      "loss/crossentropy": 1.6175484359264374,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18295736238360405,
+      "step": 1087
+    },
+    {
+      "epoch": 0.18133333333333335,
+      "grad_norm": 26.625,
+      "grad_norm_var": 15.977018229166667,
+      "learning_rate": 9.21179175467611e-05,
+      "loss": 6.6092,
+      "loss/crossentropy": 0.8297821581363678,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13756988383829594,
+      "step": 1088
+    },
+    {
+      "epoch": 0.1815,
+      "grad_norm": 36.0,
+      "grad_norm_var": 17.648958333333333,
+      "learning_rate": 9.210380293863462e-05,
+      "loss": 6.7444,
+      "loss/crossentropy": 1.4588651955127716,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.195820190012455,
+      "step": 1089
+    },
+    {
+      "epoch": 0.18166666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 17.938997395833333,
+      "learning_rate": 9.208967678751177e-05,
+      "loss": 6.7921,
+      "loss/crossentropy": 1.406354010105133,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12879913859069347,
+      "step": 1090
+    },
+    {
+      "epoch": 0.18183333333333335,
+      "grad_norm": 28.375,
+      "grad_norm_var": 17.246809895833334,
+      "learning_rate": 9.207553909726531e-05,
+      "loss": 6.7435,
+      "loss/crossentropy": 1.4283540695905685,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21265421621501446,
+      "step": 1091
+    },
+    {
+      "epoch": 0.182,
+      "grad_norm": 28.25,
+      "grad_norm_var": 17.684375,
+      "learning_rate": 9.206138987177118e-05,
+      "loss": 6.6358,
+      "loss/crossentropy": 1.9023324400186539,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23484686017036438,
+      "step": 1092
+    },
+    {
+      "epoch": 0.18216666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 16.784830729166668,
+      "learning_rate": 9.204722911490846e-05,
+      "loss": 6.7327,
+      "loss/crossentropy": 1.3453277945518494,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20566649548709393,
+      "step": 1093
+    },
+    {
+      "epoch": 0.18233333333333332,
+      "grad_norm": 29.25,
+      "grad_norm_var": 16.87890625,
+      "learning_rate": 9.20330568305594e-05,
+      "loss": 6.7803,
+      "loss/crossentropy": 1.8893288373947144,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2047295831143856,
+      "step": 1094
+    },
+    {
+      "epoch": 0.1825,
+      "grad_norm": 26.5,
+      "grad_norm_var": 17.463541666666668,
+      "learning_rate": 9.201887302260943e-05,
+      "loss": 7.0077,
+      "loss/crossentropy": 2.0443789064884186,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37492508813738823,
+      "step": 1095
+    },
+    {
+      "epoch": 0.18266666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 17.174934895833335,
+      "learning_rate": 9.20046776949471e-05,
+      "loss": 6.8071,
+      "loss/crossentropy": 1.9399181604385376,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2062360793352127,
+      "step": 1096
+    },
+    {
+      "epoch": 0.18283333333333332,
+      "grad_norm": 28.5,
+      "grad_norm_var": 17.463997395833335,
+      "learning_rate": 9.199047085146415e-05,
+      "loss": 7.0174,
+      "loss/crossentropy": 2.3190993070602417,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16586605831980705,
+      "step": 1097
+    },
+    {
+      "epoch": 0.183,
+      "grad_norm": 27.5,
+      "grad_norm_var": 17.935416666666665,
+      "learning_rate": 9.197625249605546e-05,
+      "loss": 6.8453,
+      "loss/crossentropy": 1.9766651391983032,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23800357058644295,
+      "step": 1098
+    },
+    {
+      "epoch": 0.18316666666666667,
+      "grad_norm": 27.625,
+      "grad_norm_var": 18.154622395833332,
+      "learning_rate": 9.196202263261908e-05,
+      "loss": 6.2618,
+      "loss/crossentropy": 1.3636380434036255,
+      "loss/hidden": 3.8828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2844836488366127,
+      "step": 1099
+    },
+    {
+      "epoch": 0.18333333333333332,
+      "grad_norm": 30.25,
+      "grad_norm_var": 6.467122395833333,
+      "learning_rate": 9.194778126505621e-05,
+      "loss": 6.6721,
+      "loss/crossentropy": 1.7554650008678436,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19685994274914265,
+      "step": 1100
+    },
+    {
+      "epoch": 0.1835,
+      "grad_norm": 29.75,
+      "grad_norm_var": 6.36640625,
+      "learning_rate": 9.193352839727121e-05,
+      "loss": 6.508,
+      "loss/crossentropy": 1.418417051434517,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1817461997270584,
+      "step": 1101
+    },
+    {
+      "epoch": 0.18366666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 6.179622395833333,
+      "learning_rate": 9.191926403317155e-05,
+      "loss": 6.7888,
+      "loss/crossentropy": 1.9059070199728012,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17574192211031914,
+      "step": 1102
+    },
+    {
+      "epoch": 0.18383333333333332,
+      "grad_norm": 27.25,
+      "grad_norm_var": 6.195572916666666,
+      "learning_rate": 9.190498817666793e-05,
+      "loss": 6.7849,
+      "loss/crossentropy": 1.3170613497495651,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21735167875885963,
+      "step": 1103
+    },
+    {
+      "epoch": 0.184,
+      "grad_norm": 28.625,
+      "grad_norm_var": 5.724739583333333,
+      "learning_rate": 9.189070083167411e-05,
+      "loss": 6.7249,
+      "loss/crossentropy": 1.7239506840705872,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22732261009514332,
+      "step": 1104
+    },
+    {
+      "epoch": 0.18416666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.699934895833333,
+      "learning_rate": 9.18764020021071e-05,
+      "loss": 7.4892,
+      "loss/crossentropy": 2.229788452386856,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24073180928826332,
+      "step": 1105
+    },
+    {
+      "epoch": 0.18433333333333332,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.4337890625,
+      "learning_rate": 9.186209169188695e-05,
+      "loss": 6.3854,
+      "loss/crossentropy": 1.439317725598812,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14441492035984993,
+      "step": 1106
+    },
+    {
+      "epoch": 0.1845,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.4327473958333334,
+      "learning_rate": 9.184776990493695e-05,
+      "loss": 6.6847,
+      "loss/crossentropy": 1.704232081770897,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1967526078224182,
+      "step": 1107
+    },
+    {
+      "epoch": 0.18466666666666667,
+      "grad_norm": 27.875,
+      "grad_norm_var": 1.4747395833333334,
+      "learning_rate": 9.183343664518348e-05,
+      "loss": 6.3027,
+      "loss/crossentropy": 1.175614818930626,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13120904564857483,
+      "step": 1108
+    },
+    {
+      "epoch": 0.18483333333333332,
+      "grad_norm": 26.25,
+      "grad_norm_var": 1.8561848958333333,
+      "learning_rate": 9.181909191655612e-05,
+      "loss": 7.0494,
+      "loss/crossentropy": 1.6754688024520874,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1912320740520954,
+      "step": 1109
+    },
+    {
+      "epoch": 0.185,
+      "grad_norm": 25.125,
+      "grad_norm_var": 2.6059895833333333,
+      "learning_rate": 9.180473572298751e-05,
+      "loss": 6.5747,
+      "loss/crossentropy": 1.5051643997430801,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10529914731159806,
+      "step": 1110
+    },
+    {
+      "epoch": 0.18516666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.4155598958333333,
+      "learning_rate": 9.179036806841353e-05,
+      "loss": 6.7311,
+      "loss/crossentropy": 1.602803036570549,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18250281177461147,
+      "step": 1111
+    },
+    {
+      "epoch": 0.18533333333333332,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.4155598958333333,
+      "learning_rate": 9.177598895677309e-05,
+      "loss": 6.9634,
+      "loss/crossentropy": 1.5102469325065613,
+      "loss/hidden": 3.859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3009185716509819,
+      "step": 1112
+    },
+    {
+      "epoch": 0.1855,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.4389973958333333,
+      "learning_rate": 9.176159839200838e-05,
+      "loss": 6.4181,
+      "loss/crossentropy": 1.632187157869339,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15377922169864178,
+      "step": 1113
+    },
+    {
+      "epoch": 0.18566666666666667,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.3510416666666667,
+      "learning_rate": 9.17471963780646e-05,
+      "loss": 6.5475,
+      "loss/crossentropy": 1.6477246582508087,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13139150105416775,
+      "step": 1114
+    },
+    {
+      "epoch": 0.18583333333333332,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.448372395833333,
+      "learning_rate": 9.173278291889015e-05,
+      "loss": 6.7408,
+      "loss/crossentropy": 1.353840321302414,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23572589084506035,
+      "step": 1115
+    },
+    {
+      "epoch": 0.186,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.4254557291666665,
+      "learning_rate": 9.171835801843658e-05,
+      "loss": 6.6785,
+      "loss/crossentropy": 1.3895253986120224,
+      "loss/hidden": 3.11328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16994834505021572,
+      "step": 1116
+    },
+    {
+      "epoch": 0.18616666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.5629557291666667,
+      "learning_rate": 9.170392168065857e-05,
+      "loss": 6.7035,
+      "loss/crossentropy": 1.8727948367595673,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2386951968073845,
+      "step": 1117
+    },
+    {
+      "epoch": 0.18633333333333332,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.43515625,
+      "learning_rate": 9.168947390951388e-05,
+      "loss": 6.6923,
+      "loss/crossentropy": 1.233549326658249,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13183335401117802,
+      "step": 1118
+    },
+    {
+      "epoch": 0.1865,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.43515625,
+      "learning_rate": 9.167501470896349e-05,
+      "loss": 6.4367,
+      "loss/crossentropy": 1.3284440636634827,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12750062718987465,
+      "step": 1119
+    },
+    {
+      "epoch": 0.18666666666666668,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.97890625,
+      "learning_rate": 9.166054408297145e-05,
+      "loss": 7.1919,
+      "loss/crossentropy": 2.2246187031269073,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1881609559059143,
+      "step": 1120
+    },
+    {
+      "epoch": 0.18683333333333332,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.939322916666667,
+      "learning_rate": 9.164606203550497e-05,
+      "loss": 6.7847,
+      "loss/crossentropy": 1.483774647116661,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2637944780290127,
+      "step": 1121
+    },
+    {
+      "epoch": 0.187,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.794791666666667,
+      "learning_rate": 9.16315685705344e-05,
+      "loss": 6.9156,
+      "loss/crossentropy": 2.547579824924469,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29180193692445755,
+      "step": 1122
+    },
+    {
+      "epoch": 0.18716666666666668,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.8041666666666667,
+      "learning_rate": 9.161706369203317e-05,
+      "loss": 6.6665,
+      "loss/crossentropy": 1.3284527510404587,
+      "loss/hidden": 3.7578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22660033591091633,
+      "step": 1123
+    },
+    {
+      "epoch": 0.18733333333333332,
+      "grad_norm": 26.875,
+      "grad_norm_var": 2.966666666666667,
+      "learning_rate": 9.160254740397791e-05,
+      "loss": 6.6604,
+      "loss/crossentropy": 1.7407826334238052,
+      "loss/hidden": 3.09375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1453867368400097,
+      "step": 1124
+    },
+    {
+      "epoch": 0.1875,
+      "grad_norm": 26.875,
+      "grad_norm_var": 2.7983723958333333,
+      "learning_rate": 9.158801971034832e-05,
+      "loss": 6.4418,
+      "loss/crossentropy": 1.4215179234743118,
+      "loss/hidden": 3.078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13300453312695026,
+      "step": 1125
+    },
+    {
+      "epoch": 0.18766666666666668,
+      "grad_norm": 27.0,
+      "grad_norm_var": 2.1489583333333333,
+      "learning_rate": 9.157348061512727e-05,
+      "loss": 6.3674,
+      "loss/crossentropy": 1.196425050497055,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12883343547582626,
+      "step": 1126
+    },
+    {
+      "epoch": 0.18783333333333332,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.1083333333333334,
+      "learning_rate": 9.15589301223007e-05,
+      "loss": 6.5302,
+      "loss/crossentropy": 0.9842760860919952,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11342811957001686,
+      "step": 1127
+    },
+    {
+      "epoch": 0.188,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.9559895833333334,
+      "learning_rate": 9.154436823585777e-05,
+      "loss": 6.5862,
+      "loss/crossentropy": 1.5646549314260483,
+      "loss/hidden": 3.65625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.211781557649374,
+      "step": 1128
+    },
+    {
+      "epoch": 0.18816666666666668,
+      "grad_norm": 25.875,
+      "grad_norm_var": 2.3514973958333334,
+      "learning_rate": 9.152979495979063e-05,
+      "loss": 6.7477,
+      "loss/crossentropy": 1.2287218123674393,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24197194911539555,
+      "step": 1129
+    },
+    {
+      "epoch": 0.18833333333333332,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.3583333333333334,
+      "learning_rate": 9.151521029809469e-05,
+      "loss": 6.4685,
+      "loss/crossentropy": 0.9807814359664917,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20455237291753292,
+      "step": 1130
+    },
+    {
+      "epoch": 0.1885,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.1434895833333334,
+      "learning_rate": 9.150061425476838e-05,
+      "loss": 6.5694,
+      "loss/crossentropy": 1.7335754111409187,
+      "loss/hidden": 3.03515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11938268970698118,
+      "step": 1131
+    },
+    {
+      "epoch": 0.18866666666666668,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.2393229166666666,
+      "learning_rate": 9.14860068338133e-05,
+      "loss": 6.7335,
+      "loss/crossentropy": 1.0469383150339127,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19568204693496227,
+      "step": 1132
+    },
+    {
+      "epoch": 0.18883333333333333,
+      "grad_norm": 26.375,
+      "grad_norm_var": 2.0395182291666667,
+      "learning_rate": 9.147138803923416e-05,
+      "loss": 6.4101,
+      "loss/crossentropy": 1.400108441710472,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16938790678977966,
+      "step": 1133
+    },
+    {
+      "epoch": 0.189,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.11015625,
+      "learning_rate": 9.145675787503878e-05,
+      "loss": 6.4386,
+      "loss/crossentropy": 1.6609979271888733,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21109462156891823,
+      "step": 1134
+    },
+    {
+      "epoch": 0.18916666666666668,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.0791015625,
+      "learning_rate": 9.14421163452381e-05,
+      "loss": 6.9789,
+      "loss/crossentropy": 1.4032050222158432,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1367554683238268,
+      "step": 1135
+    },
+    {
+      "epoch": 0.18933333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.62265625,
+      "learning_rate": 9.142746345384619e-05,
+      "loss": 7.2695,
+      "loss/crossentropy": 1.9446362257003784,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2847407441586256,
+      "step": 1136
+    },
+    {
+      "epoch": 0.1895,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.5955729166666666,
+      "learning_rate": 9.141279920488021e-05,
+      "loss": 6.4097,
+      "loss/crossentropy": 1.5586843192577362,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23947792127728462,
+      "step": 1137
+    },
+    {
+      "epoch": 0.18966666666666668,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.5260416666666667,
+      "learning_rate": 9.139812360236046e-05,
+      "loss": 6.8643,
+      "loss/crossentropy": 1.4375022053718567,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14979300647974014,
+      "step": 1138
+    },
+    {
+      "epoch": 0.18983333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.9457682291666667,
+      "learning_rate": 9.138343665031033e-05,
+      "loss": 7.0323,
+      "loss/crossentropy": 2.1014698147773743,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2438003458082676,
+      "step": 1139
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.8893229166666667,
+      "learning_rate": 9.136873835275633e-05,
+      "loss": 6.4694,
+      "loss/crossentropy": 1.7500245869159698,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22757063433527946,
+      "step": 1140
+    },
+    {
+      "epoch": 0.19016666666666668,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.8551432291666667,
+      "learning_rate": 9.135402871372808e-05,
+      "loss": 6.6971,
+      "loss/crossentropy": 1.8896842002868652,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21483681350946426,
+      "step": 1141
+    },
+    {
+      "epoch": 0.19033333333333333,
+      "grad_norm": 27.25,
+      "grad_norm_var": 1.8134765625,
+      "learning_rate": 9.133930773725834e-05,
+      "loss": 6.5363,
+      "loss/crossentropy": 1.075780563056469,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1016013901680708,
+      "step": 1142
+    },
+    {
+      "epoch": 0.1905,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.92265625,
+      "learning_rate": 9.132457542738292e-05,
+      "loss": 6.8569,
+      "loss/crossentropy": 1.4011684656143188,
+      "loss/hidden": 4.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1438417974859476,
+      "step": 1143
+    },
+    {
+      "epoch": 0.19066666666666668,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.0229166666666667,
+      "learning_rate": 9.130983178814077e-05,
+      "loss": 6.7067,
+      "loss/crossentropy": 1.2498696893453598,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17606077156960964,
+      "step": 1144
+    },
+    {
+      "epoch": 0.19083333333333333,
+      "grad_norm": 26.125,
+      "grad_norm_var": 1.9434895833333334,
+      "learning_rate": 9.129507682357394e-05,
+      "loss": 6.6912,
+      "loss/crossentropy": 1.0827951952815056,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1887818519026041,
+      "step": 1145
+    },
+    {
+      "epoch": 0.191,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.9353515625,
+      "learning_rate": 9.128031053772759e-05,
+      "loss": 6.7258,
+      "loss/crossentropy": 1.3786602169275284,
+      "loss/hidden": 3.15625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13905745558440685,
+      "step": 1146
+    },
+    {
+      "epoch": 0.19116666666666668,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.005989583333333,
+      "learning_rate": 9.126553293464998e-05,
+      "loss": 6.7825,
+      "loss/crossentropy": 1.7173727452754974,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1754875760525465,
+      "step": 1147
+    },
+    {
+      "epoch": 0.19133333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.0375,
+      "learning_rate": 9.125074401839249e-05,
+      "loss": 6.7077,
+      "loss/crossentropy": 1.3831033259630203,
+      "loss/hidden": 2.92578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1177002526819706,
+      "step": 1148
+    },
+    {
+      "epoch": 0.1915,
+      "grad_norm": 27.75,
+      "grad_norm_var": 1.7317057291666667,
+      "learning_rate": 9.123594379300955e-05,
+      "loss": 6.2903,
+      "loss/crossentropy": 1.2398637905716896,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11419546976685524,
+      "step": 1149
+    },
+    {
+      "epoch": 0.19166666666666668,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.239518229166667,
+      "learning_rate": 9.122113226255877e-05,
+      "loss": 7.1381,
+      "loss/crossentropy": 1.5385432243347168,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20141547359526157,
+      "step": 1150
+    },
+    {
+      "epoch": 0.19183333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.2181640625,
+      "learning_rate": 9.120630943110077e-05,
+      "loss": 6.9889,
+      "loss/crossentropy": 1.968285232782364,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22922252863645554,
+      "step": 1151
+    },
+    {
+      "epoch": 0.192,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.0504557291666665,
+      "learning_rate": 9.119147530269937e-05,
+      "loss": 6.4616,
+      "loss/crossentropy": 1.6147977262735367,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22360694780945778,
+      "step": 1152
+    },
+    {
+      "epoch": 0.19216666666666668,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.0504557291666665,
+      "learning_rate": 9.117662988142138e-05,
+      "loss": 6.6606,
+      "loss/crossentropy": 1.3119268715381622,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15336495265364647,
+      "step": 1153
+    },
+    {
+      "epoch": 0.19233333333333333,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.145768229166667,
+      "learning_rate": 9.116177317133676e-05,
+      "loss": 6.4666,
+      "loss/crossentropy": 1.5752292722463608,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12842294573783875,
+      "step": 1154
+    },
+    {
+      "epoch": 0.1925,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.1184895833333335,
+      "learning_rate": 9.114690517651859e-05,
+      "loss": 6.7643,
+      "loss/crossentropy": 1.7045340538024902,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16367618553340435,
+      "step": 1155
+    },
+    {
+      "epoch": 0.19266666666666668,
+      "grad_norm": 27.0,
+      "grad_norm_var": 2.314322916666667,
+      "learning_rate": 9.1132025901043e-05,
+      "loss": 6.8107,
+      "loss/crossentropy": 1.4380560964345932,
+      "loss/hidden": 3.6953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20478954166173935,
+      "step": 1156
+    },
+    {
+      "epoch": 0.19283333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.741080729166667,
+      "learning_rate": 9.111713534898922e-05,
+      "loss": 7.2648,
+      "loss/crossentropy": 1.8873392343521118,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2218952253460884,
+      "step": 1157
+    },
+    {
+      "epoch": 0.193,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.767708333333333,
+      "learning_rate": 9.110223352443958e-05,
+      "loss": 6.6598,
+      "loss/crossentropy": 1.3560119718313217,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1457577794790268,
+      "step": 1158
+    },
+    {
+      "epoch": 0.19316666666666665,
+      "grad_norm": 25.5,
+      "grad_norm_var": 3.34765625,
+      "learning_rate": 9.108732043147952e-05,
+      "loss": 6.387,
+      "loss/crossentropy": 1.8567863702774048,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.189379021525383,
+      "step": 1159
+    },
+    {
+      "epoch": 0.19333333333333333,
+      "grad_norm": 28.125,
+      "grad_norm_var": 3.224739583333333,
+      "learning_rate": 9.107239607419753e-05,
+      "loss": 6.9686,
+      "loss/crossentropy": 1.6484321355819702,
+      "loss/hidden": 3.87109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22876212559640408,
+      "step": 1160
+    },
+    {
+      "epoch": 0.1935,
+      "grad_norm": 26.375,
+      "grad_norm_var": 3.146875,
+      "learning_rate": 9.105746045668521e-05,
+      "loss": 6.8195,
+      "loss/crossentropy": 1.8834514319896698,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1649310365319252,
+      "step": 1161
+    },
+    {
+      "epoch": 0.19366666666666665,
+      "grad_norm": 28.5,
+      "grad_norm_var": 3.147330729166667,
+      "learning_rate": 9.104251358303724e-05,
+      "loss": 6.5509,
+      "loss/crossentropy": 1.2517316043376923,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14421220496296883,
+      "step": 1162
+    },
+    {
+      "epoch": 0.19383333333333333,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.851822916666667,
+      "learning_rate": 9.102755545735141e-05,
+      "loss": 6.9415,
+      "loss/crossentropy": 1.566934511065483,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16994167864322662,
+      "step": 1163
+    },
+    {
+      "epoch": 0.194,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.9843098958333334,
+      "learning_rate": 9.101258608372856e-05,
+      "loss": 6.7779,
+      "loss/crossentropy": 1.2031786739826202,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1660716887563467,
+      "step": 1164
+    },
+    {
+      "epoch": 0.19416666666666665,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.4921223958333334,
+      "learning_rate": 9.099760546627261e-05,
+      "loss": 6.8136,
+      "loss/crossentropy": 1.7103138267993927,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14589514583349228,
+      "step": 1165
+    },
+    {
+      "epoch": 0.19433333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.106705729166667,
+      "learning_rate": 9.098261360909064e-05,
+      "loss": 6.9035,
+      "loss/crossentropy": 1.2733410149812698,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13349948450922966,
+      "step": 1166
+    },
+    {
+      "epoch": 0.1945,
+      "grad_norm": 25.5,
+      "grad_norm_var": 3.716666666666667,
+      "learning_rate": 9.096761051629268e-05,
+      "loss": 6.2888,
+      "loss/crossentropy": 1.850477620959282,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18535355478525162,
+      "step": 1167
+    },
+    {
+      "epoch": 0.19466666666666665,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.8208333333333333,
+      "learning_rate": 9.095259619199197e-05,
+      "loss": 7.0673,
+      "loss/crossentropy": 1.661905825138092,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16752731800079346,
+      "step": 1168
+    },
+    {
+      "epoch": 0.19483333333333333,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.815559895833333,
+      "learning_rate": 9.093757064030473e-05,
+      "loss": 7.0351,
+      "loss/crossentropy": 1.4530473500490189,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1607332043349743,
+      "step": 1169
+    },
+    {
+      "epoch": 0.195,
+      "grad_norm": 27.5,
+      "grad_norm_var": 3.815559895833333,
+      "learning_rate": 9.092253386535032e-05,
+      "loss": 6.7947,
+      "loss/crossentropy": 1.9012523591518402,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24264705926179886,
+      "step": 1170
+    },
+    {
+      "epoch": 0.19516666666666665,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.56875,
+      "learning_rate": 9.090748587125118e-05,
+      "loss": 6.4319,
+      "loss/crossentropy": 1.2599057853221893,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13274476863443851,
+      "step": 1171
+    },
+    {
+      "epoch": 0.19533333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 3.4858723958333333,
+      "learning_rate": 9.089242666213276e-05,
+      "loss": 6.709,
+      "loss/crossentropy": 1.3363978862762451,
+      "loss/hidden": 3.60546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22091389633715153,
+      "step": 1172
+    },
+    {
+      "epoch": 0.1955,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.920768229166667,
+      "learning_rate": 9.087735624212365e-05,
+      "loss": 6.4208,
+      "loss/crossentropy": 0.8140209168195724,
+      "loss/hidden": 3.12890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14398249424993992,
+      "step": 1173
+    },
+    {
+      "epoch": 0.19566666666666666,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.8379557291666666,
+      "learning_rate": 9.08622746153555e-05,
+      "loss": 6.5416,
+      "loss/crossentropy": 1.64599147439003,
+      "loss/hidden": 3.14453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13100099749863148,
+      "step": 1174
+    },
+    {
+      "epoch": 0.19583333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.1192057291666666,
+      "learning_rate": 9.084718178596301e-05,
+      "loss": 7.3333,
+      "loss/crossentropy": 1.9611361026763916,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16890819370746613,
+      "step": 1175
+    },
+    {
+      "epoch": 0.196,
+      "grad_norm": 26.25,
+      "grad_norm_var": 3.5010416666666666,
+      "learning_rate": 9.083207775808396e-05,
+      "loss": 6.6961,
+      "loss/crossentropy": 1.419433519244194,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13947676308453083,
+      "step": 1176
+    },
+    {
+      "epoch": 0.19616666666666666,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.133072916666667,
+      "learning_rate": 9.081696253585921e-05,
+      "loss": 6.1697,
+      "loss/crossentropy": 1.317748561501503,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15072812139987946,
+      "step": 1177
+    },
+    {
+      "epoch": 0.19633333333333333,
+      "grad_norm": 28.125,
+      "grad_norm_var": 3.1567057291666667,
+      "learning_rate": 9.080183612343268e-05,
+      "loss": 6.7164,
+      "loss/crossentropy": 1.3343005925416946,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1475490778684616,
+      "step": 1178
+    },
+    {
+      "epoch": 0.1965,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.224739583333333,
+      "learning_rate": 9.078669852495138e-05,
+      "loss": 6.6116,
+      "loss/crossentropy": 1.3958501517772675,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19039911031723022,
+      "step": 1179
+    },
+    {
+      "epoch": 0.19666666666666666,
+      "grad_norm": 26.625,
+      "grad_norm_var": 3.345833333333333,
+      "learning_rate": 9.077154974456534e-05,
+      "loss": 6.5651,
+      "loss/crossentropy": 1.6107353270053864,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17827985808253288,
+      "step": 1180
+    },
+    {
+      "epoch": 0.19683333333333333,
+      "grad_norm": 25.75,
+      "grad_norm_var": 3.2080729166666666,
+      "learning_rate": 9.075638978642771e-05,
+      "loss": 6.1706,
+      "loss/crossentropy": 1.153374306857586,
+      "loss/hidden": 3.12890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09518537670373917,
+      "step": 1181
+    },
+    {
+      "epoch": 0.197,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.0098307291666666,
+      "learning_rate": 9.074121865469467e-05,
+      "loss": 6.6373,
+      "loss/crossentropy": 1.1229224801063538,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.143198374658823,
+      "step": 1182
+    },
+    {
+      "epoch": 0.19716666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.624739583333333,
+      "learning_rate": 9.072603635352548e-05,
+      "loss": 6.803,
+      "loss/crossentropy": 1.1446676403284073,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20634750649333,
+      "step": 1183
+    },
+    {
+      "epoch": 0.19733333333333333,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.593489583333333,
+      "learning_rate": 9.071084288708243e-05,
+      "loss": 6.7214,
+      "loss/crossentropy": 1.3592483848333359,
+      "loss/hidden": 3.12890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12259738147258759,
+      "step": 1184
+    },
+    {
+      "epoch": 0.1975,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.66640625,
+      "learning_rate": 9.069563825953092e-05,
+      "loss": 6.3755,
+      "loss/crossentropy": 0.930594339966774,
+      "loss/hidden": 3.046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09236075170338154,
+      "step": 1185
+    },
+    {
+      "epoch": 0.19766666666666666,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.7134765625,
+      "learning_rate": 9.068042247503936e-05,
+      "loss": 6.6205,
+      "loss/crossentropy": 1.8010877966880798,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2238609902560711,
+      "step": 1186
+    },
+    {
+      "epoch": 0.19783333333333333,
+      "grad_norm": 27.0,
+      "grad_norm_var": 2.809830729166667,
+      "learning_rate": 9.066519553777926e-05,
+      "loss": 6.6321,
+      "loss/crossentropy": 1.4810826033353806,
+      "loss/hidden": 3.71484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1963069550693035,
+      "step": 1187
+    },
+    {
+      "epoch": 0.198,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.7333333333333334,
+      "learning_rate": 9.064995745192518e-05,
+      "loss": 6.9062,
+      "loss/crossentropy": 1.7429955452680588,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20512964576482773,
+      "step": 1188
+    },
+    {
+      "epoch": 0.19816666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.974739583333333,
+      "learning_rate": 9.06347082216547e-05,
+      "loss": 6.9706,
+      "loss/crossentropy": 1.5064394623041153,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13919940404593945,
+      "step": 1189
+    },
+    {
+      "epoch": 0.19833333333333333,
+      "grad_norm": 27.5,
+      "grad_norm_var": 3.003059895833333,
+      "learning_rate": 9.061944785114851e-05,
+      "loss": 6.6419,
+      "loss/crossentropy": 1.9997592568397522,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24072296917438507,
+      "step": 1190
+    },
+    {
+      "epoch": 0.1985,
+      "grad_norm": 26.125,
+      "grad_norm_var": 2.0747395833333333,
+      "learning_rate": 9.060417634459031e-05,
+      "loss": 6.1881,
+      "loss/crossentropy": 1.072800636291504,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11856809444725513,
+      "step": 1191
+    },
+    {
+      "epoch": 0.19866666666666666,
+      "grad_norm": 26.75,
+      "grad_norm_var": 1.9830729166666667,
+      "learning_rate": 9.058889370616689e-05,
+      "loss": 6.0493,
+      "loss/crossentropy": 1.301239162683487,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11753997579216957,
+      "step": 1192
+    },
+    {
+      "epoch": 0.19883333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.8393229166666667,
+      "learning_rate": 9.057359994006806e-05,
+      "loss": 7.1586,
+      "loss/crossentropy": 1.8825919032096863,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.168900016695261,
+      "step": 1193
+    },
+    {
+      "epoch": 0.199,
+      "grad_norm": 26.125,
+      "grad_norm_var": 3.0768229166666665,
+      "learning_rate": 9.055829505048667e-05,
+      "loss": 6.5576,
+      "loss/crossentropy": 1.4147628545761108,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14477220736443996,
+      "step": 1194
+    },
+    {
+      "epoch": 0.19916666666666666,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.0166666666666666,
+      "learning_rate": 9.054297904161868e-05,
+      "loss": 7.0073,
+      "loss/crossentropy": 1.7706734985113144,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19512862339615822,
+      "step": 1195
+    },
+    {
+      "epoch": 0.19933333333333333,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.8955729166666666,
+      "learning_rate": 9.052765191766304e-05,
+      "loss": 6.8255,
+      "loss/crossentropy": 1.1627454906702042,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10302140191197395,
+      "step": 1196
+    },
+    {
+      "epoch": 0.1995,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.598372395833333,
+      "learning_rate": 9.051231368282177e-05,
+      "loss": 6.1381,
+      "loss/crossentropy": 1.7248927056789398,
+      "loss/hidden": 3.11328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14499239809811115,
+      "step": 1197
+    },
+    {
+      "epoch": 0.19966666666666666,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.4593098958333335,
+      "learning_rate": 9.049696434129994e-05,
+      "loss": 6.8588,
+      "loss/crossentropy": 1.5471692085266113,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18744614906609058,
+      "step": 1198
+    },
+    {
+      "epoch": 0.19983333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.328125,
+      "learning_rate": 9.048160389730566e-05,
+      "loss": 6.7448,
+      "loss/crossentropy": 1.419817790389061,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1421039178967476,
+      "step": 1199
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.2994140625,
+      "learning_rate": 9.046623235505007e-05,
+      "loss": 6.6168,
+      "loss/crossentropy": 1.5587895065546036,
+      "loss/hidden": 3.14453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1169568095356226,
+      "step": 1200
+    },
+    {
+      "epoch": 0.20016666666666666,
+      "grad_norm": 27.375,
+      "grad_norm_var": 2.287239583333333,
+      "learning_rate": 9.045084971874738e-05,
+      "loss": 6.9286,
+      "loss/crossentropy": 1.6890553534030914,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13997517712414265,
+      "step": 1201
+    },
+    {
+      "epoch": 0.20033333333333334,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.3421223958333335,
+      "learning_rate": 9.043545599261481e-05,
+      "loss": 6.3552,
+      "loss/crossentropy": 1.7470484524965286,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17869875393807888,
+      "step": 1202
+    },
+    {
+      "epoch": 0.2005,
+      "grad_norm": 26.25,
+      "grad_norm_var": 2.479622395833333,
+      "learning_rate": 9.042005118087267e-05,
+      "loss": 6.5288,
+      "loss/crossentropy": 1.701821357011795,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14979557134211063,
+      "step": 1203
+    },
+    {
+      "epoch": 0.20066666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.606184895833333,
+      "learning_rate": 9.040463528774423e-05,
+      "loss": 6.9567,
+      "loss/crossentropy": 1.4951329976320267,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24193494021892548,
+      "step": 1204
+    },
+    {
+      "epoch": 0.20083333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.8343098958333335,
+      "learning_rate": 9.038920831745587e-05,
+      "loss": 6.8274,
+      "loss/crossentropy": 2.0322689712047577,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15476682782173157,
+      "step": 1205
+    },
+    {
+      "epoch": 0.201,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.8020833333333335,
+      "learning_rate": 9.0373770274237e-05,
+      "loss": 6.6209,
+      "loss/crossentropy": 1.497676894068718,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1388947144150734,
+      "step": 1206
+    },
+    {
+      "epoch": 0.20116666666666666,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.4645182291666665,
+      "learning_rate": 9.035832116232001e-05,
+      "loss": 6.4037,
+      "loss/crossentropy": 1.248465672135353,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16754697728902102,
+      "step": 1207
+    },
+    {
+      "epoch": 0.20133333333333334,
+      "grad_norm": 27.125,
+      "grad_norm_var": 3.388541666666667,
+      "learning_rate": 9.03428609859404e-05,
+      "loss": 6.0206,
+      "loss/crossentropy": 0.5909055694937706,
+      "loss/hidden": 3.0859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.08574395999312401,
+      "step": 1208
+    },
+    {
+      "epoch": 0.2015,
+      "grad_norm": 26.25,
+      "grad_norm_var": 2.9322265625,
+      "learning_rate": 9.032738974933664e-05,
+      "loss": 6.6539,
+      "loss/crossentropy": 1.3955488801002502,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21347402036190033,
+      "step": 1209
+    },
+    {
+      "epoch": 0.20166666666666666,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.65,
+      "learning_rate": 9.031190745675024e-05,
+      "loss": 6.6029,
+      "loss/crossentropy": 1.8016868829727173,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15900519117712975,
+      "step": 1210
+    },
+    {
+      "epoch": 0.20183333333333334,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.5184895833333334,
+      "learning_rate": 9.029641411242579e-05,
+      "loss": 6.8054,
+      "loss/crossentropy": 1.5825860798358917,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20401453599333763,
+      "step": 1211
+    },
+    {
+      "epoch": 0.202,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.005143229166667,
+      "learning_rate": 9.028090972061088e-05,
+      "loss": 6.9967,
+      "loss/crossentropy": 1.3594131767749786,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1763775832951069,
+      "step": 1212
+    },
+    {
+      "epoch": 0.20216666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.9643229166666667,
+      "learning_rate": 9.02653942855561e-05,
+      "loss": 6.8733,
+      "loss/crossentropy": 2.009831130504608,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19624514132738113,
+      "step": 1213
+    },
+    {
+      "epoch": 0.20233333333333334,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.9567057291666665,
+      "learning_rate": 9.024986781151512e-05,
+      "loss": 7.0479,
+      "loss/crossentropy": 1.5291010439395905,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24753132835030556,
+      "step": 1214
+    },
+    {
+      "epoch": 0.2025,
+      "grad_norm": 26.125,
+      "grad_norm_var": 3.2643229166666665,
+      "learning_rate": 9.023433030274459e-05,
+      "loss": 6.971,
+      "loss/crossentropy": 1.1791229546070099,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24390131048858166,
+      "step": 1215
+    },
+    {
+      "epoch": 0.20266666666666666,
+      "grad_norm": 26.125,
+      "grad_norm_var": 3.566666666666667,
+      "learning_rate": 9.021878176350423e-05,
+      "loss": 7.0674,
+      "loss/crossentropy": 1.4992656409740448,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14316371455788612,
+      "step": 1216
+    },
+    {
+      "epoch": 0.20283333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.6186848958333333,
+      "learning_rate": 9.020322219805674e-05,
+      "loss": 6.9972,
+      "loss/crossentropy": 2.2822020053863525,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20745661109685898,
+      "step": 1217
+    },
+    {
+      "epoch": 0.203,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.5530598958333335,
+      "learning_rate": 9.018765161066787e-05,
+      "loss": 7.1042,
+      "loss/crossentropy": 2.0620460212230682,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2415241003036499,
+      "step": 1218
+    },
+    {
+      "epoch": 0.20316666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.49765625,
+      "learning_rate": 9.017207000560639e-05,
+      "loss": 6.8128,
+      "loss/crossentropy": 1.655750960111618,
+      "loss/hidden": 3.796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2676118575036526,
+      "step": 1219
+    },
+    {
+      "epoch": 0.20333333333333334,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.6614583333333335,
+      "learning_rate": 9.015647738714408e-05,
+      "loss": 6.692,
+      "loss/crossentropy": 1.2279730141162872,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16738798841834068,
+      "step": 1220
+    },
+    {
+      "epoch": 0.2035,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.346809895833333,
+      "learning_rate": 9.014087375955573e-05,
+      "loss": 6.615,
+      "loss/crossentropy": 1.3126295655965805,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1448457706719637,
+      "step": 1221
+    },
+    {
+      "epoch": 0.20366666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.3587890625,
+      "learning_rate": 9.012525912711918e-05,
+      "loss": 6.8934,
+      "loss/crossentropy": 1.359935849905014,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.207577807828784,
+      "step": 1222
+    },
+    {
+      "epoch": 0.20383333333333334,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.5291015625,
+      "learning_rate": 9.010963349411529e-05,
+      "loss": 6.6074,
+      "loss/crossentropy": 1.4897202253341675,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1723779635503888,
+      "step": 1223
+    },
+    {
+      "epoch": 0.204,
+      "grad_norm": 26.5,
+      "grad_norm_var": 2.6518229166666667,
+      "learning_rate": 9.009399686482787e-05,
+      "loss": 6.6729,
+      "loss/crossentropy": 2.1506561040878296,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21321503818035126,
+      "step": 1224
+    },
+    {
+      "epoch": 0.20416666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.1176432291666667,
+      "learning_rate": 9.007834924354383e-05,
+      "loss": 6.7883,
+      "loss/crossentropy": 1.533318668603897,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1673147864639759,
+      "step": 1225
+    },
+    {
+      "epoch": 0.20433333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.5518229166666666,
+      "learning_rate": 9.006269063455304e-05,
+      "loss": 6.9214,
+      "loss/crossentropy": 1.954908162355423,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2034966889768839,
+      "step": 1226
+    },
+    {
+      "epoch": 0.2045,
+      "grad_norm": 29.375,
+      "grad_norm_var": 3.55390625,
+      "learning_rate": 9.00470210421484e-05,
+      "loss": 6.8735,
+      "loss/crossentropy": 1.4537492394447327,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15034646075218916,
+      "step": 1227
+    },
+    {
+      "epoch": 0.20466666666666666,
+      "grad_norm": 28.5,
+      "grad_norm_var": 3.24140625,
+      "learning_rate": 9.003134047062579e-05,
+      "loss": 7.0833,
+      "loss/crossentropy": 1.2051249742507935,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21681534126400948,
+      "step": 1228
+    },
+    {
+      "epoch": 0.20483333333333334,
+      "grad_norm": 27.125,
+      "grad_norm_var": 3.3514973958333334,
+      "learning_rate": 9.001564892428415e-05,
+      "loss": 6.7004,
+      "loss/crossentropy": 1.9723464846611023,
+      "loss/hidden": 3.08984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14291589241474867,
+      "step": 1229
+    },
+    {
+      "epoch": 0.205,
+      "grad_norm": 26.625,
+      "grad_norm_var": 3.5843098958333335,
+      "learning_rate": 8.999994640742543e-05,
+      "loss": 6.3032,
+      "loss/crossentropy": 1.516051322221756,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19862085580825806,
+      "step": 1230
+    },
+    {
+      "epoch": 0.20516666666666666,
+      "grad_norm": 28.125,
+      "grad_norm_var": 3.2030598958333334,
+      "learning_rate": 8.998423292435454e-05,
+      "loss": 6.4676,
+      "loss/crossentropy": 1.1445068717002869,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13811507262289524,
+      "step": 1231
+    },
+    {
+      "epoch": 0.20533333333333334,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.8643229166666666,
+      "learning_rate": 8.996850847937941e-05,
+      "loss": 6.4963,
+      "loss/crossentropy": 1.399800255894661,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16321530006825924,
+      "step": 1232
+    },
+    {
+      "epoch": 0.2055,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.0629557291666667,
+      "learning_rate": 8.995277307681099e-05,
+      "loss": 6.7221,
+      "loss/crossentropy": 1.0423889011144638,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17177395336329937,
+      "step": 1233
+    },
+    {
+      "epoch": 0.20566666666666666,
+      "grad_norm": 29.25,
+      "grad_norm_var": 3.0134765625,
+      "learning_rate": 8.993702672096324e-05,
+      "loss": 6.4469,
+      "loss/crossentropy": 1.5280451029539108,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26041166856884956,
+      "step": 1234
+    },
+    {
+      "epoch": 0.20583333333333334,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.73515625,
+      "learning_rate": 8.992126941615313e-05,
+      "loss": 6.9393,
+      "loss/crossentropy": 1.480648934841156,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.227988688275218,
+      "step": 1235
+    },
+    {
+      "epoch": 0.206,
+      "grad_norm": 27.375,
+      "grad_norm_var": 2.802018229166667,
+      "learning_rate": 8.990550116670057e-05,
+      "loss": 6.5075,
+      "loss/crossentropy": 1.583464339375496,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20366531237959862,
+      "step": 1236
+    },
+    {
+      "epoch": 0.20616666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.7535807291666665,
+      "learning_rate": 8.988972197692855e-05,
+      "loss": 6.8314,
+      "loss/crossentropy": 1.8470201790332794,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26271121203899384,
+      "step": 1237
+    },
+    {
+      "epoch": 0.20633333333333334,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.81015625,
+      "learning_rate": 8.987393185116302e-05,
+      "loss": 6.6978,
+      "loss/crossentropy": 1.4702500626444817,
+      "loss/hidden": 3.07421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13370383717119694,
+      "step": 1238
+    },
+    {
+      "epoch": 0.2065,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.6239583333333334,
+      "learning_rate": 8.985813079373292e-05,
+      "loss": 6.6726,
+      "loss/crossentropy": 1.3550107330083847,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.244224788621068,
+      "step": 1239
+    },
+    {
+      "epoch": 0.20666666666666667,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.3358723958333334,
+      "learning_rate": 8.98423188089702e-05,
+      "loss": 6.7607,
+      "loss/crossentropy": 1.8079198896884918,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14915229566395283,
+      "step": 1240
+    },
+    {
+      "epoch": 0.20683333333333334,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.6018229166666667,
+      "learning_rate": 8.982649590120982e-05,
+      "loss": 6.9855,
+      "loss/crossentropy": 1.9233669936656952,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2114737443625927,
+      "step": 1241
+    },
+    {
+      "epoch": 0.207,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.703125,
+      "learning_rate": 8.981066207478971e-05,
+      "loss": 7.1926,
+      "loss/crossentropy": 1.9067326784133911,
+      "loss/hidden": 3.8125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30099838972091675,
+      "step": 1242
+    },
+    {
+      "epoch": 0.20716666666666667,
+      "grad_norm": 27.75,
+      "grad_norm_var": 1.6718098958333334,
+      "learning_rate": 8.97948173340508e-05,
+      "loss": 6.7338,
+      "loss/crossentropy": 1.430239051580429,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17893367540091276,
+      "step": 1243
+    },
+    {
+      "epoch": 0.20733333333333334,
+      "grad_norm": 25.75,
+      "grad_norm_var": 2.0957682291666666,
+      "learning_rate": 8.977896168333702e-05,
+      "loss": 6.5617,
+      "loss/crossentropy": 1.4219672679901123,
+      "loss/hidden": 3.13671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1239619292318821,
+      "step": 1244
+    },
+    {
+      "epoch": 0.2075,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.158072916666667,
+      "learning_rate": 8.976309512699528e-05,
+      "loss": 6.2839,
+      "loss/crossentropy": 1.5226720422506332,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21036674082279205,
+      "step": 1245
+    },
+    {
+      "epoch": 0.20766666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.231705729166667,
+      "learning_rate": 8.97472176693755e-05,
+      "loss": 6.8114,
+      "loss/crossentropy": 1.7841049432754517,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29534048587083817,
+      "step": 1246
+    },
+    {
+      "epoch": 0.20783333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.383072916666667,
+      "learning_rate": 8.973132931483057e-05,
+      "loss": 6.8929,
+      "loss/crossentropy": 1.5661884546279907,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14548476226627827,
+      "step": 1247
+    },
+    {
+      "epoch": 0.208,
+      "grad_norm": 26.125,
+      "grad_norm_var": 2.6874348958333334,
+      "learning_rate": 8.971543006771636e-05,
+      "loss": 6.3195,
+      "loss/crossentropy": 1.2170061022043228,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15822157263755798,
+      "step": 1248
+    },
+    {
+      "epoch": 0.20816666666666667,
+      "grad_norm": 26.125,
+      "grad_norm_var": 2.635872395833333,
+      "learning_rate": 8.969951993239177e-05,
+      "loss": 6.6852,
+      "loss/crossentropy": 2.0130717903375626,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15568870678544044,
+      "step": 1249
+    },
+    {
+      "epoch": 0.20833333333333334,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.8802083333333335,
+      "learning_rate": 8.968359891321862e-05,
+      "loss": 6.6418,
+      "loss/crossentropy": 1.6311540305614471,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22847135923802853,
+      "step": 1250
+    },
+    {
+      "epoch": 0.2085,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.8733723958333335,
+      "learning_rate": 8.966766701456177e-05,
+      "loss": 6.6833,
+      "loss/crossentropy": 1.5942027270793915,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18262358382344246,
+      "step": 1251
+    },
+    {
+      "epoch": 0.20866666666666667,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.8322265625,
+      "learning_rate": 8.965172424078902e-05,
+      "loss": 6.8973,
+      "loss/crossentropy": 1.8061284124851227,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1853354275226593,
+      "step": 1252
+    },
+    {
+      "epoch": 0.20883333333333334,
+      "grad_norm": 24.125,
+      "grad_norm_var": 3.83515625,
+      "learning_rate": 8.963577059627118e-05,
+      "loss": 6.106,
+      "loss/crossentropy": 1.568946361541748,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1581588964909315,
+      "step": 1253
+    },
+    {
+      "epoch": 0.209,
+      "grad_norm": 28.125,
+      "grad_norm_var": 3.8218098958333333,
+      "learning_rate": 8.961980608538203e-05,
+      "loss": 6.487,
+      "loss/crossentropy": 1.403021089732647,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16708368621766567,
+      "step": 1254
+    },
+    {
+      "epoch": 0.20916666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.8577473958333335,
+      "learning_rate": 8.960383071249836e-05,
+      "loss": 6.6891,
+      "loss/crossentropy": 1.3967433869838715,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2843403536826372,
+      "step": 1255
+    },
+    {
+      "epoch": 0.20933333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.9872395833333334,
+      "learning_rate": 8.958784448199987e-05,
+      "loss": 6.8028,
+      "loss/crossentropy": 1.620072603225708,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22326701134443283,
+      "step": 1256
+    },
+    {
+      "epoch": 0.2095,
+      "grad_norm": 26.25,
+      "grad_norm_var": 4.164322916666666,
+      "learning_rate": 8.95718473982693e-05,
+      "loss": 6.5015,
+      "loss/crossentropy": 1.4531026035547256,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17697684653103352,
+      "step": 1257
+    },
+    {
+      "epoch": 0.20966666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.5337890625,
+      "learning_rate": 8.955583946569233e-05,
+      "loss": 6.8476,
+      "loss/crossentropy": 1.7197735011577606,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2148963138461113,
+      "step": 1258
+    },
+    {
+      "epoch": 0.20983333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.8197265625,
+      "learning_rate": 8.95398206886576e-05,
+      "loss": 6.4826,
+      "loss/crossentropy": 1.7892868220806122,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16583251394331455,
+      "step": 1259
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 26.5,
+      "grad_norm_var": 3.6306640625,
+      "learning_rate": 8.95237910715568e-05,
+      "loss": 6.651,
+      "loss/crossentropy": 1.4422977268695831,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1798603367060423,
+      "step": 1260
+    },
+    {
+      "epoch": 0.21016666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 3.5916015625,
+      "learning_rate": 8.950775061878453e-05,
+      "loss": 7.2244,
+      "loss/crossentropy": 1.7096526324748993,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16561134532094002,
+      "step": 1261
+    },
+    {
+      "epoch": 0.21033333333333334,
+      "grad_norm": 28.875,
+      "grad_norm_var": 3.3330729166666666,
+      "learning_rate": 8.949169933473833e-05,
+      "loss": 6.1272,
+      "loss/crossentropy": 1.789107620716095,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23020472191274166,
+      "step": 1262
+    },
+    {
+      "epoch": 0.2105,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.1434895833333334,
+      "learning_rate": 8.94756372238188e-05,
+      "loss": 6.7733,
+      "loss/crossentropy": 1.6946524679660797,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18876988068223,
+      "step": 1263
+    },
+    {
+      "epoch": 0.21066666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.970833333333333,
+      "learning_rate": 8.945956429042943e-05,
+      "loss": 6.9426,
+      "loss/crossentropy": 1.4046456217765808,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13317753560841084,
+      "step": 1264
+    },
+    {
+      "epoch": 0.21083333333333334,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.6754557291666665,
+      "learning_rate": 8.944348053897671e-05,
+      "loss": 6.5402,
+      "loss/crossentropy": 0.9884390383958817,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13939815573394299,
+      "step": 1265
+    },
+    {
+      "epoch": 0.211,
+      "grad_norm": 25.625,
+      "grad_norm_var": 2.8041015625,
+      "learning_rate": 8.94273859738701e-05,
+      "loss": 6.4565,
+      "loss/crossentropy": 1.5082050412893295,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14231885317713022,
+      "step": 1266
+    },
+    {
+      "epoch": 0.21116666666666667,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.871809895833333,
+      "learning_rate": 8.941128059952201e-05,
+      "loss": 6.6774,
+      "loss/crossentropy": 1.5060864835977554,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21303267404437065,
+      "step": 1267
+    },
+    {
+      "epoch": 0.21133333333333335,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.926497395833333,
+      "learning_rate": 8.939516442034781e-05,
+      "loss": 6.7261,
+      "loss/crossentropy": 1.6655118465423584,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2032595369964838,
+      "step": 1268
+    },
+    {
+      "epoch": 0.2115,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.060416666666667,
+      "learning_rate": 8.937903744076587e-05,
+      "loss": 6.9514,
+      "loss/crossentropy": 1.3730367943644524,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1651359535753727,
+      "step": 1269
+    },
+    {
+      "epoch": 0.21166666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.1666015625,
+      "learning_rate": 8.936289966519746e-05,
+      "loss": 6.4229,
+      "loss/crossentropy": 1.5105050131678581,
+      "loss/hidden": 3.0859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12962303683161736,
+      "step": 1270
+    },
+    {
+      "epoch": 0.21183333333333335,
+      "grad_norm": 27.0,
+      "grad_norm_var": 2.3160807291666665,
+      "learning_rate": 8.934675109806688e-05,
+      "loss": 6.6139,
+      "loss/crossentropy": 1.4867517203092575,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14812249317765236,
+      "step": 1271
+    },
+    {
+      "epoch": 0.212,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.2889973958333334,
+      "learning_rate": 8.933059174380131e-05,
+      "loss": 6.6925,
+      "loss/crossentropy": 1.6140912473201752,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15693258680403233,
+      "step": 1272
+    },
+    {
+      "epoch": 0.21216666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.9843098958333334,
+      "learning_rate": 8.931442160683094e-05,
+      "loss": 6.6233,
+      "loss/crossentropy": 1.8490241765975952,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15286258608102798,
+      "step": 1273
+    },
+    {
+      "epoch": 0.21233333333333335,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.8270833333333334,
+      "learning_rate": 8.929824069158894e-05,
+      "loss": 6.8251,
+      "loss/crossentropy": 2.088714152574539,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23050561919808388,
+      "step": 1274
+    },
+    {
+      "epoch": 0.2125,
+      "grad_norm": 27.25,
+      "grad_norm_var": 1.70390625,
+      "learning_rate": 8.928204900251136e-05,
+      "loss": 6.7779,
+      "loss/crossentropy": 1.9452691078186035,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22901257127523422,
+      "step": 1275
+    },
+    {
+      "epoch": 0.21266666666666667,
+      "grad_norm": 27.375,
+      "grad_norm_var": 1.5530598958333333,
+      "learning_rate": 8.926584654403724e-05,
+      "loss": 6.6702,
+      "loss/crossentropy": 1.5190252363681793,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13852976262569427,
+      "step": 1276
+    },
+    {
+      "epoch": 0.21283333333333335,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.7809895833333333,
+      "learning_rate": 8.924963332060863e-05,
+      "loss": 6.9777,
+      "loss/crossentropy": 1.794594019651413,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20917383581399918,
+      "step": 1277
+    },
+    {
+      "epoch": 0.213,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.7624348958333333,
+      "learning_rate": 8.92334093366704e-05,
+      "loss": 6.4988,
+      "loss/crossentropy": 1.4154766499996185,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14726585149765015,
+      "step": 1278
+    },
+    {
+      "epoch": 0.21316666666666667,
+      "grad_norm": 26.25,
+      "grad_norm_var": 1.9801432291666667,
+      "learning_rate": 8.92171745966705e-05,
+      "loss": 6.366,
+      "loss/crossentropy": 1.1493876725435257,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.146550964564085,
+      "step": 1279
+    },
+    {
+      "epoch": 0.21333333333333335,
+      "grad_norm": 24.375,
+      "grad_norm_var": 2.7145182291666665,
+      "learning_rate": 8.920092910505977e-05,
+      "loss": 6.1353,
+      "loss/crossentropy": 1.7373376786708832,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.227845910936594,
+      "step": 1280
+    },
+    {
+      "epoch": 0.2135,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.7348307291666667,
+      "learning_rate": 8.9184672866292e-05,
+      "loss": 6.5413,
+      "loss/crossentropy": 1.9140368103981018,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15690459683537483,
+      "step": 1281
+    },
+    {
+      "epoch": 0.21366666666666667,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.4041015625,
+      "learning_rate": 8.916840588482392e-05,
+      "loss": 6.9562,
+      "loss/crossentropy": 1.494169995188713,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16009102202951908,
+      "step": 1282
+    },
+    {
+      "epoch": 0.21383333333333332,
+      "grad_norm": 33.5,
+      "grad_norm_var": 4.288541666666666,
+      "learning_rate": 8.915212816511522e-05,
+      "loss": 6.643,
+      "loss/crossentropy": 1.3504605740308762,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21957111731171608,
+      "step": 1283
+    },
+    {
+      "epoch": 0.214,
+      "grad_norm": 29.125,
+      "grad_norm_var": 4.288541666666666,
+      "learning_rate": 8.913583971162852e-05,
+      "loss": 6.5445,
+      "loss/crossentropy": 1.2965871542692184,
+      "loss/hidden": 3.0546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12222297862172127,
+      "step": 1284
+    },
+    {
+      "epoch": 0.21416666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 3.872916666666667,
+      "learning_rate": 8.91195405288294e-05,
+      "loss": 6.579,
+      "loss/crossentropy": 1.5752321183681488,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14281318429857492,
+      "step": 1285
+    },
+    {
+      "epoch": 0.21433333333333332,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.8893229166666665,
+      "learning_rate": 8.910323062118639e-05,
+      "loss": 7.0323,
+      "loss/crossentropy": 1.4949658513069153,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16509484313428402,
+      "step": 1286
+    },
+    {
+      "epoch": 0.2145,
+      "grad_norm": 28.125,
+      "grad_norm_var": 3.802018229166667,
+      "learning_rate": 8.908690999317093e-05,
+      "loss": 6.4651,
+      "loss/crossentropy": 1.3085608631372452,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20192678831517696,
+      "step": 1287
+    },
+    {
+      "epoch": 0.21466666666666667,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.7692057291666665,
+      "learning_rate": 8.90705786492574e-05,
+      "loss": 6.6848,
+      "loss/crossentropy": 1.3476620092988014,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1465138727799058,
+      "step": 1288
+    },
+    {
+      "epoch": 0.21483333333333332,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.878125,
+      "learning_rate": 8.905423659392316e-05,
+      "loss": 6.8042,
+      "loss/crossentropy": 2.4582173228263855,
+      "loss/hidden": 3.6796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2926645949482918,
+      "step": 1289
+    },
+    {
+      "epoch": 0.215,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.889518229166667,
+      "learning_rate": 8.903788383164846e-05,
+      "loss": 6.3778,
+      "loss/crossentropy": 1.2794660851359367,
+      "loss/hidden": 4.046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13370629120618105,
+      "step": 1290
+    },
+    {
+      "epoch": 0.21516666666666667,
+      "grad_norm": 28.0,
+      "grad_norm_var": 3.817643229166667,
+      "learning_rate": 8.90215203669165e-05,
+      "loss": 6.4129,
+      "loss/crossentropy": 1.329623505473137,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14849168248474598,
+      "step": 1291
+    },
+    {
+      "epoch": 0.21533333333333332,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.749739583333333,
+      "learning_rate": 8.90051462042134e-05,
+      "loss": 6.7661,
+      "loss/crossentropy": 1.570767730474472,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1540008969604969,
+      "step": 1292
+    },
+    {
+      "epoch": 0.2155,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.4854166666666666,
+      "learning_rate": 8.898876134802826e-05,
+      "loss": 6.6013,
+      "loss/crossentropy": 1.551586925983429,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17839442566037178,
+      "step": 1293
+    },
+    {
+      "epoch": 0.21566666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 3.5035807291666665,
+      "learning_rate": 8.897236580285308e-05,
+      "loss": 6.9122,
+      "loss/crossentropy": 1.8895243406295776,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2788575664162636,
+      "step": 1294
+    },
+    {
+      "epoch": 0.21583333333333332,
+      "grad_norm": 28.875,
+      "grad_norm_var": 3.2041666666666666,
+      "learning_rate": 8.895595957318277e-05,
+      "loss": 6.6273,
+      "loss/crossentropy": 1.9386364817619324,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17969312146306038,
+      "step": 1295
+    },
+    {
+      "epoch": 0.216,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.0205729166666666,
+      "learning_rate": 8.893954266351521e-05,
+      "loss": 6.902,
+      "loss/crossentropy": 2.3904253244400024,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3370712846517563,
+      "step": 1296
+    },
+    {
+      "epoch": 0.21616666666666667,
+      "grad_norm": 28.0,
+      "grad_norm_var": 1.9072916666666666,
+      "learning_rate": 8.892311507835119e-05,
+      "loss": 6.6135,
+      "loss/crossentropy": 1.9212593138217926,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19454152137041092,
+      "step": 1297
+    },
+    {
+      "epoch": 0.21633333333333332,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.966080729166667,
+      "learning_rate": 8.890667682219439e-05,
+      "loss": 6.391,
+      "loss/crossentropy": 1.2680547535419464,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15469966176897287,
+      "step": 1298
+    },
+    {
+      "epoch": 0.2165,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.99765625,
+      "learning_rate": 8.889022789955151e-05,
+      "loss": 6.7694,
+      "loss/crossentropy": 0.947092667222023,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2506480235606432,
+      "step": 1299
+    },
+    {
+      "epoch": 0.21666666666666667,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.0077473958333334,
+      "learning_rate": 8.887376831493205e-05,
+      "loss": 6.2312,
+      "loss/crossentropy": 1.2545424699783325,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1264524646103382,
+      "step": 1300
+    },
+    {
+      "epoch": 0.21683333333333332,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.8614583333333334,
+      "learning_rate": 8.885729807284856e-05,
+      "loss": 7.0381,
+      "loss/crossentropy": 1.5879648625850677,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1830308400094509,
+      "step": 1301
+    },
+    {
+      "epoch": 0.217,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.9056640625,
+      "learning_rate": 8.88408171778164e-05,
+      "loss": 6.9969,
+      "loss/crossentropy": 1.4587052762508392,
+      "loss/hidden": 4.03515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37913890182971954,
+      "step": 1302
+    },
+    {
+      "epoch": 0.21716666666666667,
+      "grad_norm": 27.875,
+      "grad_norm_var": 1.9358723958333333,
+      "learning_rate": 8.882432563435393e-05,
+      "loss": 6.8227,
+      "loss/crossentropy": 1.5047251284122467,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15007029846310616,
+      "step": 1303
+    },
+    {
+      "epoch": 0.21733333333333332,
+      "grad_norm": 28.0,
+      "grad_norm_var": 1.9613932291666667,
+      "learning_rate": 8.88078234469824e-05,
+      "loss": 6.3828,
+      "loss/crossentropy": 1.6883456259965897,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13623754680156708,
+      "step": 1304
+    },
+    {
+      "epoch": 0.2175,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.07890625,
+      "learning_rate": 8.879131062022598e-05,
+      "loss": 6.6566,
+      "loss/crossentropy": 1.5751657485961914,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22802406549453735,
+      "step": 1305
+    },
+    {
+      "epoch": 0.21766666666666667,
+      "grad_norm": 61.0,
+      "grad_norm_var": 67.15,
+      "learning_rate": 8.877478715861173e-05,
+      "loss": 6.668,
+      "loss/crossentropy": 2.0652737617492676,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16132371872663498,
+      "step": 1306
+    },
+    {
+      "epoch": 0.21783333333333332,
+      "grad_norm": 28.25,
+      "grad_norm_var": 67.06223958333334,
+      "learning_rate": 8.875825306666968e-05,
+      "loss": 6.5068,
+      "loss/crossentropy": 1.9428893029689789,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19048425182700157,
+      "step": 1307
+    },
+    {
+      "epoch": 0.218,
+      "grad_norm": 28.0,
+      "grad_norm_var": 67.15,
+      "learning_rate": 8.874170834893272e-05,
+      "loss": 6.529,
+      "loss/crossentropy": 1.4067817330360413,
+      "loss/hidden": 3.07421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13130534626543522,
+      "step": 1308
+    },
+    {
+      "epoch": 0.21816666666666668,
+      "grad_norm": 25.125,
+      "grad_norm_var": 68.90729166666667,
+      "learning_rate": 8.872515300993669e-05,
+      "loss": 6.2013,
+      "loss/crossentropy": 1.3501994907855988,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1739630363881588,
+      "step": 1309
+    },
+    {
+      "epoch": 0.21833333333333332,
+      "grad_norm": 26.5,
+      "grad_norm_var": 69.78430989583333,
+      "learning_rate": 8.870858705422033e-05,
+      "loss": 6.6856,
+      "loss/crossentropy": 2.0368079841136932,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19387321919202805,
+      "step": 1310
+    },
+    {
+      "epoch": 0.2185,
+      "grad_norm": 27.375,
+      "grad_norm_var": 70.22649739583333,
+      "learning_rate": 8.869201048632532e-05,
+      "loss": 6.4329,
+      "loss/crossentropy": 1.5729725360870361,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16083178110420704,
+      "step": 1311
+    },
+    {
+      "epoch": 0.21866666666666668,
+      "grad_norm": 28.125,
+      "grad_norm_var": 70.22649739583333,
+      "learning_rate": 8.867542331079617e-05,
+      "loss": 6.4391,
+      "loss/crossentropy": 1.4303375333547592,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19543035700917244,
+      "step": 1312
+    },
+    {
+      "epoch": 0.21883333333333332,
+      "grad_norm": 29.375,
+      "grad_norm_var": 69.925,
+      "learning_rate": 8.865882553218037e-05,
+      "loss": 6.8722,
+      "loss/crossentropy": 1.5897773951292038,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1667669340968132,
+      "step": 1313
+    },
+    {
+      "epoch": 0.219,
+      "grad_norm": 28.5,
+      "grad_norm_var": 69.615625,
+      "learning_rate": 8.864221715502829e-05,
+      "loss": 6.3114,
+      "loss/crossentropy": 1.2462688088417053,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2083086036145687,
+      "step": 1314
+    },
+    {
+      "epoch": 0.21916666666666668,
+      "grad_norm": 28.25,
+      "grad_norm_var": 69.69212239583334,
+      "learning_rate": 8.862559818389322e-05,
+      "loss": 6.6176,
+      "loss/crossentropy": 1.4975986778736115,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14409909583628178,
+      "step": 1315
+    },
+    {
+      "epoch": 0.21933333333333332,
+      "grad_norm": 30.25,
+      "grad_norm_var": 69.5572265625,
+      "learning_rate": 8.860896862333134e-05,
+      "loss": 6.9355,
+      "loss/crossentropy": 1.6909880936145782,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2077299915254116,
+      "step": 1316
+    },
+    {
+      "epoch": 0.2195,
+      "grad_norm": 28.25,
+      "grad_norm_var": 69.52682291666666,
+      "learning_rate": 8.859232847790175e-05,
+      "loss": 6.2975,
+      "loss/crossentropy": 1.7112976908683777,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21853592619299889,
+      "step": 1317
+    },
+    {
+      "epoch": 0.21966666666666668,
+      "grad_norm": 28.375,
+      "grad_norm_var": 69.46770833333333,
+      "learning_rate": 8.857567775216643e-05,
+      "loss": 6.6071,
+      "loss/crossentropy": 1.5232578814029694,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2179286815226078,
+      "step": 1318
+    },
+    {
+      "epoch": 0.21983333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 69.33958333333334,
+      "learning_rate": 8.855901645069026e-05,
+      "loss": 6.1688,
+      "loss/crossentropy": 1.058205395936966,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10036487132310867,
+      "step": 1319
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 27.625,
+      "grad_norm_var": 69.45149739583333,
+      "learning_rate": 8.854234457804105e-05,
+      "loss": 6.9218,
+      "loss/crossentropy": 1.6375434398651123,
+      "loss/hidden": 3.62890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2968744896352291,
+      "step": 1320
+    },
+    {
+      "epoch": 0.22016666666666668,
+      "grad_norm": 25.875,
+      "grad_norm_var": 70.08098958333333,
+      "learning_rate": 8.852566213878947e-05,
+      "loss": 6.3333,
+      "loss/crossentropy": 1.4782972186803818,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26145337522029877,
+      "step": 1321
+    },
+    {
+      "epoch": 0.22033333333333333,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.5400390625,
+      "learning_rate": 8.850896913750911e-05,
+      "loss": 6.6795,
+      "loss/crossentropy": 1.3669453859329224,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15520895645022392,
+      "step": 1322
+    },
+    {
+      "epoch": 0.2205,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.671875,
+      "learning_rate": 8.849226557877646e-05,
+      "loss": 6.381,
+      "loss/crossentropy": 0.62733593583107,
+      "loss/hidden": 3.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.06343139801174402,
+      "step": 1323
+    },
+    {
+      "epoch": 0.22066666666666668,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.6733723958333333,
+      "learning_rate": 8.84755514671709e-05,
+      "loss": 6.7212,
+      "loss/crossentropy": 1.4413825124502182,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16148651763796806,
+      "step": 1324
+    },
+    {
+      "epoch": 0.22083333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.0978515625,
+      "learning_rate": 8.845882680727469e-05,
+      "loss": 6.3931,
+      "loss/crossentropy": 1.3144546374678612,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1098860977217555,
+      "step": 1325
+    },
+    {
+      "epoch": 0.221,
+      "grad_norm": 29.5,
+      "grad_norm_var": 0.9884765625,
+      "learning_rate": 8.844209160367299e-05,
+      "loss": 6.4334,
+      "loss/crossentropy": 1.5615183413028717,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14797941967844963,
+      "step": 1326
+    },
+    {
+      "epoch": 0.22116666666666668,
+      "grad_norm": 28.625,
+      "grad_norm_var": 0.9207682291666667,
+      "learning_rate": 8.842534586095383e-05,
+      "loss": 7.3761,
+      "loss/crossentropy": 1.8021913468837738,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30935032293200493,
+      "step": 1327
+    },
+    {
+      "epoch": 0.22133333333333333,
+      "grad_norm": 26.5,
+      "grad_norm_var": 1.1552083333333334,
+      "learning_rate": 8.840858958370819e-05,
+      "loss": 6.6065,
+      "loss/crossentropy": 1.3574145883321762,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1361311562359333,
+      "step": 1328
+    },
+    {
+      "epoch": 0.2215,
+      "grad_norm": 26.75,
+      "grad_norm_var": 1.2249348958333333,
+      "learning_rate": 8.839182277652989e-05,
+      "loss": 7.2034,
+      "loss/crossentropy": 2.0402421951293945,
+      "loss/hidden": 3.6484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20759042352437973,
+      "step": 1329
+    },
+    {
+      "epoch": 0.22166666666666668,
+      "grad_norm": 28.0,
+      "grad_norm_var": 1.2192057291666667,
+      "learning_rate": 8.837504544401561e-05,
+      "loss": 6.8273,
+      "loss/crossentropy": 1.333353579044342,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14028004556894302,
+      "step": 1330
+    },
+    {
+      "epoch": 0.22183333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.27890625,
+      "learning_rate": 8.8358257590765e-05,
+      "loss": 6.6867,
+      "loss/crossentropy": 1.4103537499904633,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1657029865309596,
+      "step": 1331
+    },
+    {
+      "epoch": 0.222,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.01015625,
+      "learning_rate": 8.834145922138049e-05,
+      "loss": 6.5112,
+      "loss/crossentropy": 1.5350777506828308,
+      "loss/hidden": 3.0,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13068307749927044,
+      "step": 1332
+    },
+    {
+      "epoch": 0.22216666666666668,
+      "grad_norm": 27.625,
+      "grad_norm_var": 1.0228515625,
+      "learning_rate": 8.832465034046749e-05,
+      "loss": 6.6084,
+      "loss/crossentropy": 1.8636922538280487,
+      "loss/hidden": 3.8203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2078309953212738,
+      "step": 1333
+    },
+    {
+      "epoch": 0.22233333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.99140625,
+      "learning_rate": 8.830783095263425e-05,
+      "loss": 7.034,
+      "loss/crossentropy": 1.598654717206955,
+      "loss/hidden": 3.73046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30865925922989845,
+      "step": 1334
+    },
+    {
+      "epoch": 0.2225,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.99140625,
+      "learning_rate": 8.829100106249189e-05,
+      "loss": 6.3653,
+      "loss/crossentropy": 1.1344598531723022,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1316494233906269,
+      "step": 1335
+    },
+    {
+      "epoch": 0.22266666666666668,
+      "grad_norm": 28.0,
+      "grad_norm_var": 1.9666015625,
+      "learning_rate": 8.827416067465441e-05,
+      "loss": 6.9004,
+      "loss/crossentropy": 1.759160429239273,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21815275400877,
+      "step": 1336
+    },
+    {
+      "epoch": 0.22283333333333333,
+      "grad_norm": 27.25,
+      "grad_norm_var": 1.6364583333333333,
+      "learning_rate": 8.825730979373872e-05,
+      "loss": 6.7587,
+      "loss/crossentropy": 2.0076573491096497,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15654901042580605,
+      "step": 1337
+    },
+    {
+      "epoch": 0.223,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.6364583333333333,
+      "learning_rate": 8.824044842436456e-05,
+      "loss": 6.4088,
+      "loss/crossentropy": 1.6686158776283264,
+      "loss/hidden": 3.0859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13004953786730766,
+      "step": 1338
+    },
+    {
+      "epoch": 0.22316666666666668,
+      "grad_norm": 26.5,
+      "grad_norm_var": 1.7817057291666667,
+      "learning_rate": 8.822357657115459e-05,
+      "loss": 5.8659,
+      "loss/crossentropy": 0.9902602061629295,
+      "loss/hidden": 3.046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09084746427834034,
+      "step": 1339
+    },
+    {
+      "epoch": 0.22333333333333333,
+      "grad_norm": 23.5,
+      "grad_norm_var": 3.18125,
+      "learning_rate": 8.82066942387343e-05,
+      "loss": 6.3085,
+      "loss/crossentropy": 1.19401353597641,
+      "loss/hidden": 3.0703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11941441148519516,
+      "step": 1340
+    },
+    {
+      "epoch": 0.2235,
+      "grad_norm": 26.625,
+      "grad_norm_var": 3.2705729166666666,
+      "learning_rate": 8.818980143173213e-05,
+      "loss": 6.7438,
+      "loss/crossentropy": 2.108260899782181,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.198447797447443,
+      "step": 1341
+    },
+    {
+      "epoch": 0.22366666666666668,
+      "grad_norm": 27.25,
+      "grad_norm_var": 3.0854166666666667,
+      "learning_rate": 8.817289815477928e-05,
+      "loss": 6.7452,
+      "loss/crossentropy": 1.3883316740393639,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15109186992049217,
+      "step": 1342
+    },
+    {
+      "epoch": 0.22383333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.2426432291666667,
+      "learning_rate": 8.815598441250987e-05,
+      "loss": 6.9428,
+      "loss/crossentropy": 1.3261938244104385,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15241075679659843,
+      "step": 1343
+    },
+    {
+      "epoch": 0.224,
+      "grad_norm": 25.75,
+      "grad_norm_var": 3.4020182291666665,
+      "learning_rate": 8.813906020956097e-05,
+      "loss": 6.541,
+      "loss/crossentropy": 1.0693974867463112,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12422368861734867,
+      "step": 1344
+    },
+    {
+      "epoch": 0.22416666666666665,
+      "grad_norm": 25.75,
+      "grad_norm_var": 3.590559895833333,
+      "learning_rate": 8.81221255505724e-05,
+      "loss": 6.11,
+      "loss/crossentropy": 1.3504291325807571,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14698335714638233,
+      "step": 1345
+    },
+    {
+      "epoch": 0.22433333333333333,
+      "grad_norm": 27.75,
+      "grad_norm_var": 3.5822265625,
+      "learning_rate": 8.810518044018689e-05,
+      "loss": 6.881,
+      "loss/crossentropy": 1.6281023919582367,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1453680656850338,
+      "step": 1346
+    },
+    {
+      "epoch": 0.2245,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.497330729166667,
+      "learning_rate": 8.808822488305005e-05,
+      "loss": 6.6608,
+      "loss/crossentropy": 1.4724755585193634,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19075221847742796,
+      "step": 1347
+    },
+    {
+      "epoch": 0.22466666666666665,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.564322916666667,
+      "learning_rate": 8.807125888381035e-05,
+      "loss": 6.6406,
+      "loss/crossentropy": 1.9415532946586609,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19600216299295425,
+      "step": 1348
+    },
+    {
+      "epoch": 0.22483333333333333,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.5900390625,
+      "learning_rate": 8.80542824471191e-05,
+      "loss": 6.5504,
+      "loss/crossentropy": 1.7943297624588013,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14495110884308815,
+      "step": 1349
+    },
+    {
+      "epoch": 0.225,
+      "grad_norm": 26.0,
+      "grad_norm_var": 2.3587890625,
+      "learning_rate": 8.803729557763047e-05,
+      "loss": 6.583,
+      "loss/crossentropy": 1.585892304778099,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15496612340211868,
+      "step": 1350
+    },
+    {
+      "epoch": 0.22516666666666665,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.5041015625,
+      "learning_rate": 8.802029828000156e-05,
+      "loss": 6.8792,
+      "loss/crossentropy": 1.1976583451032639,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12264487985521555,
+      "step": 1351
+    },
+    {
+      "epoch": 0.22533333333333333,
+      "grad_norm": 26.625,
+      "grad_norm_var": 2.49765625,
+      "learning_rate": 8.800329055889223e-05,
+      "loss": 6.5882,
+      "loss/crossentropy": 1.3212222456932068,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10740573238581419,
+      "step": 1352
+    },
+    {
+      "epoch": 0.2255,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.4983723958333335,
+      "learning_rate": 8.798627241896524e-05,
+      "loss": 6.7388,
+      "loss/crossentropy": 2.1766937971115112,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20083113759756088,
+      "step": 1353
+    },
+    {
+      "epoch": 0.22566666666666665,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.484375,
+      "learning_rate": 8.796924386488624e-05,
+      "loss": 6.7047,
+      "loss/crossentropy": 1.630198821425438,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1088574668392539,
+      "step": 1354
+    },
+    {
+      "epoch": 0.22583333333333333,
+      "grad_norm": 27.0,
+      "grad_norm_var": 2.4520833333333334,
+      "learning_rate": 8.795220490132369e-05,
+      "loss": 6.6013,
+      "loss/crossentropy": 1.257098838686943,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1477007195353508,
+      "step": 1355
+    },
+    {
+      "epoch": 0.226,
+      "grad_norm": 26.0,
+      "grad_norm_var": 1.5927083333333334,
+      "learning_rate": 8.793515553294891e-05,
+      "loss": 6.2652,
+      "loss/crossentropy": 1.1517640799283981,
+      "loss/hidden": 3.640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17172394320368767,
+      "step": 1356
+    },
+    {
+      "epoch": 0.22616666666666665,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.5770833333333334,
+      "learning_rate": 8.79180957644361e-05,
+      "loss": 6.3652,
+      "loss/crossentropy": 1.5562289357185364,
+      "loss/hidden": 3.703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31990334391593933,
+      "step": 1357
+    },
+    {
+      "epoch": 0.22633333333333333,
+      "grad_norm": 24.75,
+      "grad_norm_var": 2.051041666666667,
+      "learning_rate": 8.790102560046227e-05,
+      "loss": 6.0902,
+      "loss/crossentropy": 1.7950303554534912,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22284463047981262,
+      "step": 1358
+    },
+    {
+      "epoch": 0.2265,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.255143229166667,
+      "learning_rate": 8.788394504570732e-05,
+      "loss": 6.8729,
+      "loss/crossentropy": 1.338299036026001,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1761884856969118,
+      "step": 1359
+    },
+    {
+      "epoch": 0.22666666666666666,
+      "grad_norm": 27.375,
+      "grad_norm_var": 2.06640625,
+      "learning_rate": 8.786685410485398e-05,
+      "loss": 6.7353,
+      "loss/crossentropy": 1.3051201850175858,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18960629776120186,
+      "step": 1360
+    },
+    {
+      "epoch": 0.22683333333333333,
+      "grad_norm": 28.0,
+      "grad_norm_var": 1.8625,
+      "learning_rate": 8.784975278258783e-05,
+      "loss": 6.4355,
+      "loss/crossentropy": 1.1859900504350662,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.136919679120183,
+      "step": 1361
+    },
+    {
+      "epoch": 0.227,
+      "grad_norm": 27.25,
+      "grad_norm_var": 1.8697916666666667,
+      "learning_rate": 8.783264108359728e-05,
+      "loss": 6.6016,
+      "loss/crossentropy": 1.6561327278614044,
+      "loss/hidden": 3.83984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28435977548360825,
+      "step": 1362
+    },
+    {
+      "epoch": 0.22716666666666666,
+      "grad_norm": 27.0,
+      "grad_norm_var": 1.8113932291666666,
+      "learning_rate": 8.78155190125736e-05,
+      "loss": 6.0377,
+      "loss/crossentropy": 1.2676357477903366,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13790716975927353,
+      "step": 1363
+    },
+    {
+      "epoch": 0.22733333333333333,
+      "grad_norm": 27.5,
+      "grad_norm_var": 1.62265625,
+      "learning_rate": 8.779838657421092e-05,
+      "loss": 6.1669,
+      "loss/crossentropy": 1.3378489539027214,
+      "loss/hidden": 3.09375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1295567862689495,
+      "step": 1364
+    },
+    {
+      "epoch": 0.2275,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.0145833333333334,
+      "learning_rate": 8.778124377320618e-05,
+      "loss": 6.8036,
+      "loss/crossentropy": 2.2961465716362,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18548791483044624,
+      "step": 1365
+    },
+    {
+      "epoch": 0.22766666666666666,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.8718098958333333,
+      "learning_rate": 8.776409061425919e-05,
+      "loss": 6.2881,
+      "loss/crossentropy": 1.2985191643238068,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13449012115597725,
+      "step": 1366
+    },
+    {
+      "epoch": 0.22783333333333333,
+      "grad_norm": 27.625,
+      "grad_norm_var": 1.7139973958333334,
+      "learning_rate": 8.774692710207257e-05,
+      "loss": 6.5953,
+      "loss/crossentropy": 1.0388475880026817,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11377880908548832,
+      "step": 1367
+    },
+    {
+      "epoch": 0.228,
+      "grad_norm": 27.875,
+      "grad_norm_var": 1.6593098958333334,
+      "learning_rate": 8.772975324135179e-05,
+      "loss": 6.3785,
+      "loss/crossentropy": 0.9511722177267075,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09341735392808914,
+      "step": 1368
+    },
+    {
+      "epoch": 0.22816666666666666,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.6749348958333334,
+      "learning_rate": 8.771256903680519e-05,
+      "loss": 6.6922,
+      "loss/crossentropy": 1.429678276181221,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20660696923732758,
+      "step": 1369
+    },
+    {
+      "epoch": 0.22833333333333333,
+      "grad_norm": 26.875,
+      "grad_norm_var": 1.7083333333333333,
+      "learning_rate": 8.769537449314391e-05,
+      "loss": 6.3573,
+      "loss/crossentropy": 1.4579046368598938,
+      "loss/hidden": 3.71875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2587745487689972,
+      "step": 1370
+    },
+    {
+      "epoch": 0.2285,
+      "grad_norm": 26.625,
+      "grad_norm_var": 1.7483723958333333,
+      "learning_rate": 8.76781696150819e-05,
+      "loss": 6.4696,
+      "loss/crossentropy": 1.2540282979607582,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26700376253575087,
+      "step": 1371
+    },
+    {
+      "epoch": 0.22866666666666666,
+      "grad_norm": 27.875,
+      "grad_norm_var": 1.5677083333333333,
+      "learning_rate": 8.766095440733601e-05,
+      "loss": 6.8604,
+      "loss/crossentropy": 1.6994609534740448,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19905509427189827,
+      "step": 1372
+    },
+    {
+      "epoch": 0.22883333333333333,
+      "grad_norm": 27.75,
+      "grad_norm_var": 1.5561848958333333,
+      "learning_rate": 8.764372887462586e-05,
+      "loss": 6.6942,
+      "loss/crossentropy": 1.8636305034160614,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2452586479485035,
+      "step": 1373
+    },
+    {
+      "epoch": 0.229,
+      "grad_norm": 28.75,
+      "grad_norm_var": 0.9853515625,
+      "learning_rate": 8.762649302167395e-05,
+      "loss": 7.1421,
+      "loss/crossentropy": 2.502372086048126,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21404139325022697,
+      "step": 1374
+    },
+    {
+      "epoch": 0.22916666666666666,
+      "grad_norm": 27.625,
+      "grad_norm_var": 0.6494140625,
+      "learning_rate": 8.760924685320557e-05,
+      "loss": 6.3204,
+      "loss/crossentropy": 1.6386250108480453,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2121148481965065,
+      "step": 1375
+    },
+    {
+      "epoch": 0.22933333333333333,
+      "grad_norm": 26.5,
+      "grad_norm_var": 0.7455729166666667,
+      "learning_rate": 8.759199037394887e-05,
+      "loss": 6.2754,
+      "loss/crossentropy": 0.8101483061909676,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13781879469752312,
+      "step": 1376
+    },
+    {
+      "epoch": 0.2295,
+      "grad_norm": 28.75,
+      "grad_norm_var": 0.8072916666666666,
+      "learning_rate": 8.757472358863481e-05,
+      "loss": 6.8081,
+      "loss/crossentropy": 1.4773345291614532,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1636052243411541,
+      "step": 1377
+    },
+    {
+      "epoch": 0.22966666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.1202473958333334,
+      "learning_rate": 8.755744650199716e-05,
+      "loss": 7.2161,
+      "loss/crossentropy": 1.087226115167141,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14092607703059912,
+      "step": 1378
+    },
+    {
+      "epoch": 0.22983333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.2145833333333333,
+      "learning_rate": 8.754015911877255e-05,
+      "loss": 7.0857,
+      "loss/crossentropy": 1.4208160042762756,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15992657095193863,
+      "step": 1379
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 27.375,
+      "grad_norm_var": 1.2259765625,
+      "learning_rate": 8.752286144370041e-05,
+      "loss": 6.3116,
+      "loss/crossentropy": 1.2512414306402206,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1318049468100071,
+      "step": 1380
+    },
+    {
+      "epoch": 0.23016666666666666,
+      "grad_norm": 27.5,
+      "grad_norm_var": 0.9889973958333333,
+      "learning_rate": 8.750555348152298e-05,
+      "loss": 6.6642,
+      "loss/crossentropy": 1.007347248494625,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13616536371409893,
+      "step": 1381
+    },
+    {
+      "epoch": 0.23033333333333333,
+      "grad_norm": 26.0,
+      "grad_norm_var": 1.2247395833333334,
+      "learning_rate": 8.748823523698535e-05,
+      "loss": 6.3399,
+      "loss/crossentropy": 1.5809184312820435,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12614354118704796,
+      "step": 1382
+    },
+    {
+      "epoch": 0.2305,
+      "grad_norm": 26.875,
+      "grad_norm_var": 1.2802083333333334,
+      "learning_rate": 8.747090671483542e-05,
+      "loss": 6.587,
+      "loss/crossentropy": 1.1127055883407593,
+      "loss/hidden": 3.0546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12591952504590154,
+      "step": 1383
+    },
+    {
+      "epoch": 0.23066666666666666,
+      "grad_norm": 25.375,
+      "grad_norm_var": 1.6395833333333334,
+      "learning_rate": 8.745356791982391e-05,
+      "loss": 6.5432,
+      "loss/crossentropy": 0.7250919938087463,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11801298521459103,
+      "step": 1384
+    },
+    {
+      "epoch": 0.23083333333333333,
+      "grad_norm": 26.25,
+      "grad_norm_var": 1.7093098958333333,
+      "learning_rate": 8.74362188567043e-05,
+      "loss": 6.4521,
+      "loss/crossentropy": 1.3951380848884583,
+      "loss/hidden": 3.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12724317610263824,
+      "step": 1385
+    },
+    {
+      "epoch": 0.231,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.8166666666666667,
+      "learning_rate": 8.741885953023301e-05,
+      "loss": 6.6342,
+      "loss/crossentropy": 1.5546142607927322,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15518338978290558,
+      "step": 1386
+    },
+    {
+      "epoch": 0.23116666666666666,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.9791666666666667,
+      "learning_rate": 8.740148994516912e-05,
+      "loss": 7.0334,
+      "loss/crossentropy": 1.6840177774429321,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2906576320528984,
+      "step": 1387
+    },
+    {
+      "epoch": 0.23133333333333334,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.98515625,
+      "learning_rate": 8.738411010627466e-05,
+      "loss": 6.9206,
+      "loss/crossentropy": 1.3421642929315567,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21450426429510117,
+      "step": 1388
+    },
+    {
+      "epoch": 0.2315,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.0125,
+      "learning_rate": 8.736672001831438e-05,
+      "loss": 6.9608,
+      "loss/crossentropy": 1.2448264062404633,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1828536745160818,
+      "step": 1389
+    },
+    {
+      "epoch": 0.23166666666666666,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.9639973958333334,
+      "learning_rate": 8.734931968605589e-05,
+      "loss": 6.4951,
+      "loss/crossentropy": 1.0161764919757843,
+      "loss/hidden": 3.11328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09913336299359798,
+      "step": 1390
+    },
+    {
+      "epoch": 0.23183333333333334,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.2171223958333335,
+      "learning_rate": 8.733190911426958e-05,
+      "loss": 7.5304,
+      "loss/crossentropy": 1.8381727933883667,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22387774847447872,
+      "step": 1391
+    },
+    {
+      "epoch": 0.232,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.0677083333333335,
+      "learning_rate": 8.731448830772864e-05,
+      "loss": 6.5824,
+      "loss/crossentropy": 1.2965693026781082,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18344127014279366,
+      "step": 1392
+    },
+    {
+      "epoch": 0.23216666666666666,
+      "grad_norm": 26.875,
+      "grad_norm_var": 2.1233723958333335,
+      "learning_rate": 8.729705727120911e-05,
+      "loss": 5.9351,
+      "loss/crossentropy": 1.2470285519957542,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17037468776106834,
+      "step": 1393
+    },
+    {
+      "epoch": 0.23233333333333334,
+      "grad_norm": 26.625,
+      "grad_norm_var": 1.8863932291666667,
+      "learning_rate": 8.72796160094898e-05,
+      "loss": 6.486,
+      "loss/crossentropy": 1.7774807810783386,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16827407479286194,
+      "step": 1394
+    },
+    {
+      "epoch": 0.2325,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.69765625,
+      "learning_rate": 8.726216452735232e-05,
+      "loss": 7.0892,
+      "loss/crossentropy": 1.477794349193573,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20362269133329391,
+      "step": 1395
+    },
+    {
+      "epoch": 0.23266666666666666,
+      "grad_norm": 48.0,
+      "grad_norm_var": 27.8666015625,
+      "learning_rate": 8.724470282958111e-05,
+      "loss": 7.069,
+      "loss/crossentropy": 1.7263435125350952,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20600185170769691,
+      "step": 1396
+    },
+    {
+      "epoch": 0.23283333333333334,
+      "grad_norm": 42.25,
+      "grad_norm_var": 38.16087239583333,
+      "learning_rate": 8.722723092096338e-05,
+      "loss": 6.7116,
+      "loss/crossentropy": 0.8946850746870041,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11208548210561275,
+      "step": 1397
+    },
+    {
+      "epoch": 0.233,
+      "grad_norm": 28.0,
+      "grad_norm_var": 37.31712239583333,
+      "learning_rate": 8.720974880628916e-05,
+      "loss": 6.4923,
+      "loss/crossentropy": 2.1065716445446014,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17256246134638786,
+      "step": 1398
+    },
+    {
+      "epoch": 0.23316666666666666,
+      "grad_norm": 26.375,
+      "grad_norm_var": 37.55618489583333,
+      "learning_rate": 8.719225649035126e-05,
+      "loss": 6.3393,
+      "loss/crossentropy": 1.664988398551941,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2271974589675665,
+      "step": 1399
+    },
+    {
+      "epoch": 0.23333333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 35.9853515625,
+      "learning_rate": 8.717475397794531e-05,
+      "loss": 7.2177,
+      "loss/crossentropy": 2.2505394220352173,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22215315327048302,
+      "step": 1400
+    },
+    {
+      "epoch": 0.2335,
+      "grad_norm": 26.125,
+      "grad_norm_var": 36.05625,
+      "learning_rate": 8.715724127386972e-05,
+      "loss": 6.1959,
+      "loss/crossentropy": 1.873531699180603,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16248880326747894,
+      "step": 1401
+    },
+    {
+      "epoch": 0.23366666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 35.97604166666667,
+      "learning_rate": 8.713971838292569e-05,
+      "loss": 6.2601,
+      "loss/crossentropy": 0.9897870719432831,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14599628001451492,
+      "step": 1402
+    },
+    {
+      "epoch": 0.23383333333333334,
+      "grad_norm": 28.625,
+      "grad_norm_var": 36.151041666666664,
+      "learning_rate": 8.712218530991723e-05,
+      "loss": 6.7,
+      "loss/crossentropy": 1.483601376414299,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16519050300121307,
+      "step": 1403
+    },
+    {
+      "epoch": 0.234,
+      "grad_norm": 26.5,
+      "grad_norm_var": 36.8103515625,
+      "learning_rate": 8.710464205965112e-05,
+      "loss": 6.7339,
+      "loss/crossentropy": 2.0020012259483337,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16443980112671852,
+      "step": 1404
+    },
+    {
+      "epoch": 0.23416666666666666,
+      "grad_norm": 26.5,
+      "grad_norm_var": 37.5416015625,
+      "learning_rate": 8.708708863693697e-05,
+      "loss": 6.761,
+      "loss/crossentropy": 1.2835683301091194,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13632741570472717,
+      "step": 1405
+    },
+    {
+      "epoch": 0.23433333333333334,
+      "grad_norm": 27.875,
+      "grad_norm_var": 37.61399739583333,
+      "learning_rate": 8.706952504658712e-05,
+      "loss": 6.4228,
+      "loss/crossentropy": 1.7721076607704163,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1346228215843439,
+      "step": 1406
+    },
+    {
+      "epoch": 0.2345,
+      "grad_norm": 28.875,
+      "grad_norm_var": 37.7150390625,
+      "learning_rate": 8.705195129341672e-05,
+      "loss": 6.7091,
+      "loss/crossentropy": 1.6065224409103394,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13542147167026997,
+      "step": 1407
+    },
+    {
+      "epoch": 0.23466666666666666,
+      "grad_norm": 26.75,
+      "grad_norm_var": 38.25416666666667,
+      "learning_rate": 8.703436738224375e-05,
+      "loss": 6.5667,
+      "loss/crossentropy": 1.2467619329690933,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1695778239518404,
+      "step": 1408
+    },
+    {
+      "epoch": 0.23483333333333334,
+      "grad_norm": 25.75,
+      "grad_norm_var": 38.80201822916667,
+      "learning_rate": 8.701677331788891e-05,
+      "loss": 6.1433,
+      "loss/crossentropy": 1.6390111446380615,
+      "loss/hidden": 3.0703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13177541457116604,
+      "step": 1409
+    },
+    {
+      "epoch": 0.235,
+      "grad_norm": 26.0,
+      "grad_norm_var": 39.10182291666667,
+      "learning_rate": 8.699916910517573e-05,
+      "loss": 6.4737,
+      "loss/crossentropy": 1.3608596995472908,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13710680603981018,
+      "step": 1410
+    },
+    {
+      "epoch": 0.23516666666666666,
+      "grad_norm": 26.875,
+      "grad_norm_var": 39.378580729166664,
+      "learning_rate": 8.69815547489305e-05,
+      "loss": 6.5518,
+      "loss/crossentropy": 1.4065961092710495,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19469621777534485,
+      "step": 1411
+    },
+    {
+      "epoch": 0.23533333333333334,
+      "grad_norm": 28.5,
+      "grad_norm_var": 15.267643229166667,
+      "learning_rate": 8.696393025398229e-05,
+      "loss": 6.8396,
+      "loss/crossentropy": 1.4624530673027039,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17353084683418274,
+      "step": 1412
+    },
+    {
+      "epoch": 0.2355,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.1468098958333335,
+      "learning_rate": 8.694629562516294e-05,
+      "loss": 6.7204,
+      "loss/crossentropy": 0.9706991761922836,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18376579880714417,
+      "step": 1413
+    },
+    {
+      "epoch": 0.23566666666666666,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.1468098958333335,
+      "learning_rate": 8.692865086730713e-05,
+      "loss": 6.4183,
+      "loss/crossentropy": 1.6739065647125244,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19507576152682304,
+      "step": 1414
+    },
+    {
+      "epoch": 0.23583333333333334,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.0561848958333333,
+      "learning_rate": 8.69109959852522e-05,
+      "loss": 6.0193,
+      "loss/crossentropy": 1.0993403047323227,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14124910533428192,
+      "step": 1415
+    },
+    {
+      "epoch": 0.236,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.0119140625,
+      "learning_rate": 8.689333098383842e-05,
+      "loss": 7.2071,
+      "loss/crossentropy": 1.5422320663928986,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22851193323731422,
+      "step": 1416
+    },
+    {
+      "epoch": 0.23616666666666666,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.80390625,
+      "learning_rate": 8.68756558679087e-05,
+      "loss": 7.0285,
+      "loss/crossentropy": 1.507189080119133,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15467006340622902,
+      "step": 1417
+    },
+    {
+      "epoch": 0.23633333333333334,
+      "grad_norm": 28.375,
+      "grad_norm_var": 2.6509765625,
+      "learning_rate": 8.685797064230878e-05,
+      "loss": 6.6319,
+      "loss/crossentropy": 1.0455129817128181,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21261783316731453,
+      "step": 1418
+    },
+    {
+      "epoch": 0.2365,
+      "grad_norm": 25.25,
+      "grad_norm_var": 3.02890625,
+      "learning_rate": 8.684027531188717e-05,
+      "loss": 6.2786,
+      "loss/crossentropy": 1.625629723072052,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16640848852694035,
+      "step": 1419
+    },
+    {
+      "epoch": 0.23666666666666666,
+      "grad_norm": 25.375,
+      "grad_norm_var": 3.2837890625,
+      "learning_rate": 8.682256988149513e-05,
+      "loss": 6.2535,
+      "loss/crossentropy": 0.8649746030569077,
+      "loss/hidden": 3.16015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11157290264964104,
+      "step": 1420
+    },
+    {
+      "epoch": 0.23683333333333334,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.5,
+      "learning_rate": 8.680485435598673e-05,
+      "loss": 6.9396,
+      "loss/crossentropy": 1.6679367423057556,
+      "loss/hidden": 3.7265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2635684050619602,
+      "step": 1421
+    },
+    {
+      "epoch": 0.237,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.6080729166666665,
+      "learning_rate": 8.678712874021874e-05,
+      "loss": 6.7282,
+      "loss/crossentropy": 1.8865498304367065,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15791157633066177,
+      "step": 1422
+    },
+    {
+      "epoch": 0.23716666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 4.169205729166666,
+      "learning_rate": 8.67693930390508e-05,
+      "loss": 6.9616,
+      "loss/crossentropy": 1.1163372993469238,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1212550587952137,
+      "step": 1423
+    },
+    {
+      "epoch": 0.23733333333333334,
+      "grad_norm": 26.5,
+      "grad_norm_var": 4.215559895833334,
+      "learning_rate": 8.67516472573452e-05,
+      "loss": 6.669,
+      "loss/crossentropy": 1.8877412974834442,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2065545618534088,
+      "step": 1424
+    },
+    {
+      "epoch": 0.2375,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.9114583333333335,
+      "learning_rate": 8.673389139996708e-05,
+      "loss": 7.0207,
+      "loss/crossentropy": 1.8748036921024323,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23519740626215935,
+      "step": 1425
+    },
+    {
+      "epoch": 0.23766666666666666,
+      "grad_norm": 28.375,
+      "grad_norm_var": 3.561393229166667,
+      "learning_rate": 8.671612547178428e-05,
+      "loss": 6.6573,
+      "loss/crossentropy": 1.9420278072357178,
+      "loss/hidden": 3.16015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14979233592748642,
+      "step": 1426
+    },
+    {
+      "epoch": 0.23783333333333334,
+      "grad_norm": 27.0,
+      "grad_norm_var": 3.5375,
+      "learning_rate": 8.669834947766746e-05,
+      "loss": 6.5476,
+      "loss/crossentropy": 1.1937883496284485,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12786134146153927,
+      "step": 1427
+    },
+    {
+      "epoch": 0.238,
+      "grad_norm": 25.625,
+      "grad_norm_var": 4.006184895833333,
+      "learning_rate": 8.668056342248998e-05,
+      "loss": 6.5543,
+      "loss/crossentropy": 1.714748054742813,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30871666595339775,
+      "step": 1428
+    },
+    {
+      "epoch": 0.23816666666666667,
+      "grad_norm": 25.625,
+      "grad_norm_var": 3.9934895833333335,
+      "learning_rate": 8.666276731112801e-05,
+      "loss": 6.4467,
+      "loss/crossentropy": 1.2653475105762482,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15050562098622322,
+      "step": 1429
+    },
+    {
+      "epoch": 0.23833333333333334,
+      "grad_norm": 28.5,
+      "grad_norm_var": 4.01640625,
+      "learning_rate": 8.664496114846044e-05,
+      "loss": 6.8582,
+      "loss/crossentropy": 1.4219676703214645,
+      "loss/hidden": 3.77734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30174621753394604,
+      "step": 1430
+    },
+    {
+      "epoch": 0.2385,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.9759765625,
+      "learning_rate": 8.662714493936895e-05,
+      "loss": 6.5823,
+      "loss/crossentropy": 1.614340752363205,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3149548228830099,
+      "step": 1431
+    },
+    {
+      "epoch": 0.23866666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.190559895833333,
+      "learning_rate": 8.660931868873793e-05,
+      "loss": 6.8286,
+      "loss/crossentropy": 1.9330060482025146,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18058581091463566,
+      "step": 1432
+    },
+    {
+      "epoch": 0.23883333333333334,
+      "grad_norm": 27.125,
+      "grad_norm_var": 3.222916666666667,
+      "learning_rate": 8.659148240145456e-05,
+      "loss": 6.7451,
+      "loss/crossentropy": 1.2989310398697853,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14274989068508148,
+      "step": 1433
+    },
+    {
+      "epoch": 0.239,
+      "grad_norm": 26.875,
+      "grad_norm_var": 3.2510416666666666,
+      "learning_rate": 8.657363608240876e-05,
+      "loss": 6.5724,
+      "loss/crossentropy": 1.5316192209720612,
+      "loss/hidden": 3.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2787708677351475,
+      "step": 1434
+    },
+    {
+      "epoch": 0.23916666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.934375,
+      "learning_rate": 8.655577973649321e-05,
+      "loss": 6.5145,
+      "loss/crossentropy": 1.2776052355766296,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13015793077647686,
+      "step": 1435
+    },
+    {
+      "epoch": 0.23933333333333334,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.5056640625,
+      "learning_rate": 8.653791336860331e-05,
+      "loss": 6.4311,
+      "loss/crossentropy": 1.6798433512449265,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17383023910224438,
+      "step": 1436
+    },
+    {
+      "epoch": 0.2395,
+      "grad_norm": 25.75,
+      "grad_norm_var": 2.58515625,
+      "learning_rate": 8.652003698363724e-05,
+      "loss": 6.6174,
+      "loss/crossentropy": 1.4651916921138763,
+      "loss/hidden": 3.83984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1780383065342903,
+      "step": 1437
+    },
+    {
+      "epoch": 0.23966666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.6587890625,
+      "learning_rate": 8.65021505864959e-05,
+      "loss": 6.8334,
+      "loss/crossentropy": 1.892525464296341,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1596686877310276,
+      "step": 1438
+    },
+    {
+      "epoch": 0.23983333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.1697265625,
+      "learning_rate": 8.648425418208294e-05,
+      "loss": 6.8679,
+      "loss/crossentropy": 2.040474846959114,
+      "loss/hidden": 3.609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20210164785385132,
+      "step": 1439
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 25.875,
+      "grad_norm_var": 2.298958333333333,
+      "learning_rate": 8.64663477753048e-05,
+      "loss": 6.6918,
+      "loss/crossentropy": 1.7955405712127686,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1639818660914898,
+      "step": 1440
+    },
+    {
+      "epoch": 0.24016666666666667,
+      "grad_norm": 25.75,
+      "grad_norm_var": 2.378059895833333,
+      "learning_rate": 8.644843137107059e-05,
+      "loss": 6.3101,
+      "loss/crossentropy": 0.5650833696126938,
+      "loss/hidden": 3.0703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.08515303116291761,
+      "step": 1441
+    },
+    {
+      "epoch": 0.24033333333333334,
+      "grad_norm": 27.375,
+      "grad_norm_var": 2.324934895833333,
+      "learning_rate": 8.64305049742922e-05,
+      "loss": 6.7586,
+      "loss/crossentropy": 1.5652510523796082,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1574523150920868,
+      "step": 1442
+    },
+    {
+      "epoch": 0.2405,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.315559895833333,
+      "learning_rate": 8.641256858988424e-05,
+      "loss": 6.4876,
+      "loss/crossentropy": 1.1279072016477585,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11717881634831429,
+      "step": 1443
+    },
+    {
+      "epoch": 0.24066666666666667,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.0837890625,
+      "learning_rate": 8.639462222276409e-05,
+      "loss": 6.6409,
+      "loss/crossentropy": 1.9791463762521744,
+      "loss/hidden": 3.7109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25130971893668175,
+      "step": 1444
+    },
+    {
+      "epoch": 0.24083333333333334,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.9051432291666666,
+      "learning_rate": 8.637666587785184e-05,
+      "loss": 6.8855,
+      "loss/crossentropy": 1.1411218345165253,
+      "loss/hidden": 3.796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2683944068849087,
+      "step": 1445
+    },
+    {
+      "epoch": 0.241,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.8955729166666666,
+      "learning_rate": 8.635869956007034e-05,
+      "loss": 6.665,
+      "loss/crossentropy": 1.7378800809383392,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15992150828242302,
+      "step": 1446
+    },
+    {
+      "epoch": 0.24116666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.01015625,
+      "learning_rate": 8.634072327434515e-05,
+      "loss": 7.1293,
+      "loss/crossentropy": 1.4358764290809631,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15003103949129581,
+      "step": 1447
+    },
+    {
+      "epoch": 0.24133333333333334,
+      "grad_norm": 27.625,
+      "grad_norm_var": 1.740625,
+      "learning_rate": 8.632273702560456e-05,
+      "loss": 6.1288,
+      "loss/crossentropy": 1.080620415508747,
+      "loss/hidden": 2.94140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1031667897477746,
+      "step": 1448
+    },
+    {
+      "epoch": 0.2415,
+      "grad_norm": 26.625,
+      "grad_norm_var": 1.8,
+      "learning_rate": 8.630474081877959e-05,
+      "loss": 6.2774,
+      "loss/crossentropy": 1.6144116818904877,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16698753461241722,
+      "step": 1449
+    },
+    {
+      "epoch": 0.24166666666666667,
+      "grad_norm": 27.375,
+      "grad_norm_var": 1.7572916666666667,
+      "learning_rate": 8.628673465880404e-05,
+      "loss": 7.0422,
+      "loss/crossentropy": 1.8665052950382233,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23085599392652512,
+      "step": 1450
+    },
+    {
+      "epoch": 0.24183333333333334,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.6233723958333333,
+      "learning_rate": 8.626871855061438e-05,
+      "loss": 6.5599,
+      "loss/crossentropy": 1.7175267785787582,
+      "loss/hidden": 3.58984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18754492234438658,
+      "step": 1451
+    },
+    {
+      "epoch": 0.242,
+      "grad_norm": 24.5,
+      "grad_norm_var": 2.2421223958333334,
+      "learning_rate": 8.625069249914983e-05,
+      "loss": 6.3063,
+      "loss/crossentropy": 1.725325345993042,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16669025272130966,
+      "step": 1452
+    },
+    {
+      "epoch": 0.24216666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.133072916666667,
+      "learning_rate": 8.623265650935234e-05,
+      "loss": 6.5514,
+      "loss/crossentropy": 1.4349830448627472,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1410666275769472,
+      "step": 1453
+    },
+    {
+      "epoch": 0.24233333333333335,
+      "grad_norm": 27.25,
+      "grad_norm_var": 1.9010416666666667,
+      "learning_rate": 8.621461058616656e-05,
+      "loss": 6.6931,
+      "loss/crossentropy": 1.1172820925712585,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13138745725154877,
+      "step": 1454
+    },
+    {
+      "epoch": 0.2425,
+      "grad_norm": 26.25,
+      "grad_norm_var": 1.7080729166666666,
+      "learning_rate": 8.61965547345399e-05,
+      "loss": 6.785,
+      "loss/crossentropy": 1.7667811810970306,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21809161081910133,
+      "step": 1455
+    },
+    {
+      "epoch": 0.24266666666666667,
+      "grad_norm": 25.25,
+      "grad_norm_var": 1.8535807291666666,
+      "learning_rate": 8.617848895942247e-05,
+      "loss": 6.5124,
+      "loss/crossentropy": 1.2851376980543137,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13383415527641773,
+      "step": 1456
+    },
+    {
+      "epoch": 0.24283333333333335,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.2889973958333334,
+      "learning_rate": 8.616041326576711e-05,
+      "loss": 7.2151,
+      "loss/crossentropy": 1.7675953954458237,
+      "loss/hidden": 3.62109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21555189229547977,
+      "step": 1457
+    },
+    {
+      "epoch": 0.243,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.511458333333333,
+      "learning_rate": 8.614232765852935e-05,
+      "loss": 6.9415,
+      "loss/crossentropy": 1.381210833787918,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15031128376722336,
+      "step": 1458
+    },
+    {
+      "epoch": 0.24316666666666667,
+      "grad_norm": 26.375,
+      "grad_norm_var": 2.623893229166667,
+      "learning_rate": 8.612423214266749e-05,
+      "loss": 6.7962,
+      "loss/crossentropy": 1.4190790355205536,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13622170314192772,
+      "step": 1459
+    },
+    {
+      "epoch": 0.24333333333333335,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.7520182291666666,
+      "learning_rate": 8.610612672314251e-05,
+      "loss": 6.4547,
+      "loss/crossentropy": 1.1637745425105095,
+      "loss/hidden": 2.93359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10563442297279835,
+      "step": 1460
+    },
+    {
+      "epoch": 0.2435,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.966666666666667,
+      "learning_rate": 8.608801140491811e-05,
+      "loss": 7.0044,
+      "loss/crossentropy": 1.416961282491684,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37132324278354645,
+      "step": 1461
+    },
+    {
+      "epoch": 0.24366666666666667,
+      "grad_norm": 26.625,
+      "grad_norm_var": 3.0122395833333333,
+      "learning_rate": 8.606988619296071e-05,
+      "loss": 6.0629,
+      "loss/crossentropy": 1.3706672191619873,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21816697902977467,
+      "step": 1462
+    },
+    {
+      "epoch": 0.24383333333333335,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.8353515625,
+      "learning_rate": 8.605175109223944e-05,
+      "loss": 6.8633,
+      "loss/crossentropy": 1.5820597410202026,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15302826836705208,
+      "step": 1463
+    },
+    {
+      "epoch": 0.244,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.875,
+      "learning_rate": 8.603360610772612e-05,
+      "loss": 6.6699,
+      "loss/crossentropy": 1.60914845764637,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14544443879276514,
+      "step": 1464
+    },
+    {
+      "epoch": 0.24416666666666667,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.8207682291666667,
+      "learning_rate": 8.601545124439535e-05,
+      "loss": 6.8595,
+      "loss/crossentropy": 1.425622507929802,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16824879124760628,
+      "step": 1465
+    },
+    {
+      "epoch": 0.24433333333333335,
+      "grad_norm": 24.375,
+      "grad_norm_var": 3.4551432291666666,
+      "learning_rate": 8.599728650722434e-05,
+      "loss": 6.6193,
+      "loss/crossentropy": 1.1695657521486282,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1581933069974184,
+      "step": 1466
+    },
+    {
+      "epoch": 0.2445,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.6143229166666666,
+      "learning_rate": 8.597911190119308e-05,
+      "loss": 6.4004,
+      "loss/crossentropy": 1.8788322508335114,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17071765288710594,
+      "step": 1467
+    },
+    {
+      "epoch": 0.24466666666666667,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.036393229166667,
+      "learning_rate": 8.596092743128423e-05,
+      "loss": 6.368,
+      "loss/crossentropy": 1.4089144319295883,
+      "loss/hidden": 3.0703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13847163692116737,
+      "step": 1468
+    },
+    {
+      "epoch": 0.24483333333333332,
+      "grad_norm": 29.375,
+      "grad_norm_var": 3.1275390625,
+      "learning_rate": 8.594273310248318e-05,
+      "loss": 6.7572,
+      "loss/crossentropy": 1.813026711344719,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14645719900727272,
+      "step": 1469
+    },
+    {
+      "epoch": 0.245,
+      "grad_norm": 28.875,
+      "grad_norm_var": 3.1791666666666667,
+      "learning_rate": 8.592452891977798e-05,
+      "loss": 6.9163,
+      "loss/crossentropy": 1.781123161315918,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18610691279172897,
+      "step": 1470
+    },
+    {
+      "epoch": 0.24516666666666667,
+      "grad_norm": 26.625,
+      "grad_norm_var": 3.106705729166667,
+      "learning_rate": 8.590631488815944e-05,
+      "loss": 6.3807,
+      "loss/crossentropy": 1.2462878674268723,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29503070190548897,
+      "step": 1471
+    },
+    {
+      "epoch": 0.24533333333333332,
+      "grad_norm": 25.75,
+      "grad_norm_var": 2.9457682291666667,
+      "learning_rate": 8.588809101262103e-05,
+      "loss": 6.2511,
+      "loss/crossentropy": 1.6031096130609512,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14661357551813126,
+      "step": 1472
+    },
+    {
+      "epoch": 0.2455,
+      "grad_norm": 41.75,
+      "grad_norm_var": 14.711393229166667,
+      "learning_rate": 8.586985729815894e-05,
+      "loss": 6.7183,
+      "loss/crossentropy": 1.0402020961046219,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12379540130496025,
+      "step": 1473
+    },
+    {
+      "epoch": 0.24566666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 14.711393229166667,
+      "learning_rate": 8.585161374977202e-05,
+      "loss": 6.6548,
+      "loss/crossentropy": 1.3159151673316956,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2661251872777939,
+      "step": 1474
+    },
+    {
+      "epoch": 0.24583333333333332,
+      "grad_norm": 30.25,
+      "grad_norm_var": 14.483333333333333,
+      "learning_rate": 8.583336037246186e-05,
+      "loss": 6.6913,
+      "loss/crossentropy": 2.057922273874283,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19244837015867233,
+      "step": 1475
+    },
+    {
+      "epoch": 0.246,
+      "grad_norm": 29.25,
+      "grad_norm_var": 14.4884765625,
+      "learning_rate": 8.581509717123273e-05,
+      "loss": 6.9276,
+      "loss/crossentropy": 1.516150802373886,
+      "loss/hidden": 3.6484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21403126791119576,
+      "step": 1476
+    },
+    {
+      "epoch": 0.24616666666666667,
+      "grad_norm": 27.125,
+      "grad_norm_var": 14.576822916666666,
+      "learning_rate": 8.579682415109156e-05,
+      "loss": 6.662,
+      "loss/crossentropy": 1.8057073950767517,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1675437092781067,
+      "step": 1477
+    },
+    {
+      "epoch": 0.24633333333333332,
+      "grad_norm": 27.5,
+      "grad_norm_var": 14.3822265625,
+      "learning_rate": 8.577854131704805e-05,
+      "loss": 6.5819,
+      "loss/crossentropy": 1.2052525877952576,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13884278573095798,
+      "step": 1478
+    },
+    {
+      "epoch": 0.2465,
+      "grad_norm": 27.0,
+      "grad_norm_var": 14.533072916666667,
+      "learning_rate": 8.576024867411451e-05,
+      "loss": 6.9713,
+      "loss/crossentropy": 1.7933558970689774,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1717188023030758,
+      "step": 1479
+    },
+    {
+      "epoch": 0.24666666666666667,
+      "grad_norm": 27.125,
+      "grad_norm_var": 14.444205729166667,
+      "learning_rate": 8.574194622730599e-05,
+      "loss": 6.6939,
+      "loss/crossentropy": 0.9968490228056908,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2076979111880064,
+      "step": 1480
+    },
+    {
+      "epoch": 0.24683333333333332,
+      "grad_norm": 27.5,
+      "grad_norm_var": 14.444205729166667,
+      "learning_rate": 8.572363398164017e-05,
+      "loss": 6.4582,
+      "loss/crossentropy": 1.9966690838336945,
+      "loss/hidden": 3.83984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20982542261481285,
+      "step": 1481
+    },
+    {
+      "epoch": 0.247,
+      "grad_norm": 24.875,
+      "grad_norm_var": 14.1697265625,
+      "learning_rate": 8.57053119421375e-05,
+      "loss": 6.2283,
+      "loss/crossentropy": 1.6232264339923859,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18004275485873222,
+      "step": 1482
+    },
+    {
+      "epoch": 0.24716666666666667,
+      "grad_norm": 27.875,
+      "grad_norm_var": 14.2125,
+      "learning_rate": 8.568698011382107e-05,
+      "loss": 6.7395,
+      "loss/crossentropy": 1.7975391894578934,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2155589796602726,
+      "step": 1483
+    },
+    {
+      "epoch": 0.24733333333333332,
+      "grad_norm": 28.625,
+      "grad_norm_var": 14.2125,
+      "learning_rate": 8.566863850171663e-05,
+      "loss": 6.397,
+      "loss/crossentropy": 1.1899760961532593,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.08900888357311487,
+      "step": 1484
+    },
+    {
+      "epoch": 0.2475,
+      "grad_norm": 27.75,
+      "grad_norm_var": 14.228580729166667,
+      "learning_rate": 8.565028711085265e-05,
+      "loss": 6.5848,
+      "loss/crossentropy": 1.4124999642372131,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13817033730447292,
+      "step": 1485
+    },
+    {
+      "epoch": 0.24766666666666667,
+      "grad_norm": 25.0,
+      "grad_norm_var": 15.017708333333333,
+      "learning_rate": 8.563192594626027e-05,
+      "loss": 6.339,
+      "loss/crossentropy": 1.0481770262122154,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1259062048047781,
+      "step": 1486
+    },
+    {
+      "epoch": 0.24783333333333332,
+      "grad_norm": 26.25,
+      "grad_norm_var": 15.112434895833333,
+      "learning_rate": 8.56135550129733e-05,
+      "loss": 6.4469,
+      "loss/crossentropy": 1.5840249508619308,
+      "loss/hidden": 3.11328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1716774683445692,
+      "step": 1487
+    },
+    {
+      "epoch": 0.248,
+      "grad_norm": 26.75,
+      "grad_norm_var": 14.8322265625,
+      "learning_rate": 8.559517431602824e-05,
+      "loss": 6.842,
+      "loss/crossentropy": 0.8972706943750381,
+      "loss/hidden": 2.859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.08874005638062954,
+      "step": 1488
+    },
+    {
+      "epoch": 0.24816666666666667,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.1510416666666665,
+      "learning_rate": 8.557678386046428e-05,
+      "loss": 6.7739,
+      "loss/crossentropy": 1.2846410125494003,
+      "loss/hidden": 3.08984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12155664712190628,
+      "step": 1489
+    },
+    {
+      "epoch": 0.24833333333333332,
+      "grad_norm": 26.625,
+      "grad_norm_var": 1.9129557291666666,
+      "learning_rate": 8.555838365132323e-05,
+      "loss": 6.5773,
+      "loss/crossentropy": 1.2601376175880432,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17203301936388016,
+      "step": 1490
+    },
+    {
+      "epoch": 0.2485,
+      "grad_norm": 26.375,
+      "grad_norm_var": 1.35390625,
+      "learning_rate": 8.553997369364963e-05,
+      "loss": 6.124,
+      "loss/crossentropy": 1.2514343559741974,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15664827451109886,
+      "step": 1491
+    },
+    {
+      "epoch": 0.24866666666666667,
+      "grad_norm": 27.875,
+      "grad_norm_var": 1.0796223958333333,
+      "learning_rate": 8.552155399249067e-05,
+      "loss": 6.4607,
+      "loss/crossentropy": 1.7554165720939636,
+      "loss/hidden": 3.69921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.221330925822258,
+      "step": 1492
+    },
+    {
+      "epoch": 0.24883333333333332,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.5458333333333334,
+      "learning_rate": 8.550312455289625e-05,
+      "loss": 6.8281,
+      "loss/crossentropy": 2.2296034395694733,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20713848248124123,
+      "step": 1493
+    },
+    {
+      "epoch": 0.249,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.6718098958333334,
+      "learning_rate": 8.548468537991884e-05,
+      "loss": 6.7593,
+      "loss/crossentropy": 1.497247964143753,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1915741879492998,
+      "step": 1494
+    },
+    {
+      "epoch": 0.24916666666666668,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.6684895833333333,
+      "learning_rate": 8.54662364786137e-05,
+      "loss": 6.6278,
+      "loss/crossentropy": 1.8613238036632538,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21113219112157822,
+      "step": 1495
+    },
+    {
+      "epoch": 0.24933333333333332,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.6684895833333333,
+      "learning_rate": 8.544777785403868e-05,
+      "loss": 6.4625,
+      "loss/crossentropy": 1.6610347032546997,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19034357741475105,
+      "step": 1496
+    },
+    {
+      "epoch": 0.2495,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.7622395833333333,
+      "learning_rate": 8.542930951125432e-05,
+      "loss": 7.0141,
+      "loss/crossentropy": 1.4421517550945282,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2027395162731409,
+      "step": 1497
+    },
+    {
+      "epoch": 0.24966666666666668,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.7275390625,
+      "learning_rate": 8.54108314553238e-05,
+      "loss": 6.5817,
+      "loss/crossentropy": 1.3464898318052292,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15230842307209969,
+      "step": 1498
+    },
+    {
+      "epoch": 0.24983333333333332,
+      "grad_norm": 27.75,
+      "grad_norm_var": 1.7247395833333334,
+      "learning_rate": 8.539234369131301e-05,
+      "loss": 6.9008,
+      "loss/crossentropy": 1.9422021210193634,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26393208652734756,
+      "step": 1499
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 27.625,
+      "grad_norm_var": 1.6559895833333333,
+      "learning_rate": 8.53738462242905e-05,
+      "loss": 6.9373,
+      "loss/crossentropy": 2.0168758928775787,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15413923747837543,
+      "step": 1500
+    },
+    {
+      "epoch": 0.25016666666666665,
+      "grad_norm": 26.625,
+      "grad_norm_var": 1.7093098958333333,
+      "learning_rate": 8.535533905932738e-05,
+      "loss": 6.423,
+      "loss/crossentropy": 1.4030412063002586,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12419278174638748,
+      "step": 1501
+    },
+    {
+      "epoch": 0.25033333333333335,
+      "grad_norm": 24.5,
+      "grad_norm_var": 1.8921223958333333,
+      "learning_rate": 8.533682220149756e-05,
+      "loss": 6.2275,
+      "loss/crossentropy": 0.926719531416893,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09100574813783169,
+      "step": 1502
+    },
+    {
+      "epoch": 0.2505,
+      "grad_norm": 25.125,
+      "grad_norm_var": 2.1552083333333334,
+      "learning_rate": 8.53182956558775e-05,
+      "loss": 6.4744,
+      "loss/crossentropy": 1.3639208525419235,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16878987848758698,
+      "step": 1503
+    },
+    {
+      "epoch": 0.25066666666666665,
+      "grad_norm": 26.875,
+      "grad_norm_var": 2.145247395833333,
+      "learning_rate": 8.52997594275464e-05,
+      "loss": 6.6772,
+      "loss/crossentropy": 1.8602468371391296,
+      "loss/hidden": 3.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13869531266391277,
+      "step": 1504
+    },
+    {
+      "epoch": 0.25083333333333335,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.2514973958333333,
+      "learning_rate": 8.528121352158604e-05,
+      "loss": 6.5704,
+      "loss/crossentropy": 1.6041803359985352,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23281598649919033,
+      "step": 1505
+    },
+    {
+      "epoch": 0.251,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.2113932291666667,
+      "learning_rate": 8.526265794308089e-05,
+      "loss": 6.3723,
+      "loss/crossentropy": 1.3392663225531578,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12616676976904273,
+      "step": 1506
+    },
+    {
+      "epoch": 0.25116666666666665,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.1770833333333335,
+      "learning_rate": 8.524409269711807e-05,
+      "loss": 6.7154,
+      "loss/crossentropy": 1.6676813066005707,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1752212531864643,
+      "step": 1507
+    },
+    {
+      "epoch": 0.25133333333333335,
+      "grad_norm": 25.5,
+      "grad_norm_var": 2.450455729166667,
+      "learning_rate": 8.522551778878736e-05,
+      "loss": 6.6563,
+      "loss/crossentropy": 1.745051771402359,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2105657234787941,
+      "step": 1508
+    },
+    {
+      "epoch": 0.2515,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.084830729166667,
+      "learning_rate": 8.520693322318116e-05,
+      "loss": 6.6398,
+      "loss/crossentropy": 2.4637134671211243,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1597395446151495,
+      "step": 1509
+    },
+    {
+      "epoch": 0.25166666666666665,
+      "grad_norm": 24.375,
+      "grad_norm_var": 2.483268229166667,
+      "learning_rate": 8.518833900539454e-05,
+      "loss": 6.0827,
+      "loss/crossentropy": 1.3924775123596191,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16528458893299103,
+      "step": 1510
+    },
+    {
+      "epoch": 0.25183333333333335,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.4893229166666666,
+      "learning_rate": 8.516973514052519e-05,
+      "loss": 6.2832,
+      "loss/crossentropy": 1.815467745065689,
+      "loss/hidden": 3.58203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2593645825982094,
+      "step": 1511
+    },
+    {
+      "epoch": 0.252,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.4942057291666666,
+      "learning_rate": 8.515112163367351e-05,
+      "loss": 6.5399,
+      "loss/crossentropy": 1.6005598306655884,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17654068022966385,
+      "step": 1512
+    },
+    {
+      "epoch": 0.25216666666666665,
+      "grad_norm": 25.625,
+      "grad_norm_var": 2.4447916666666667,
+      "learning_rate": 8.513249848994246e-05,
+      "loss": 6.2656,
+      "loss/crossentropy": 1.3193487524986267,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14158409647643566,
+      "step": 1513
+    },
+    {
+      "epoch": 0.25233333333333335,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.7514973958333333,
+      "learning_rate": 8.511386571443771e-05,
+      "loss": 6.4974,
+      "loss/crossentropy": 1.1420286148786545,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1334729939699173,
+      "step": 1514
+    },
+    {
+      "epoch": 0.2525,
+      "grad_norm": 26.875,
+      "grad_norm_var": 1.67265625,
+      "learning_rate": 8.50952233122675e-05,
+      "loss": 6.8007,
+      "loss/crossentropy": 1.6864759176969528,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16594232432544231,
+      "step": 1515
+    },
+    {
+      "epoch": 0.25266666666666665,
+      "grad_norm": 31.625,
+      "grad_norm_var": 3.2143229166666667,
+      "learning_rate": 8.50765712885428e-05,
+      "loss": 6.81,
+      "loss/crossentropy": 1.8061297535896301,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15173273161053658,
+      "step": 1516
+    },
+    {
+      "epoch": 0.25283333333333335,
+      "grad_norm": 4362076160.0,
+      "grad_norm_var": 1.189231761948672e+18,
+      "learning_rate": 8.505790964837713e-05,
+      "loss": 6.6166,
+      "loss/crossentropy": 1.3777007535099983,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13933608774095774,
+      "step": 1517
+    },
+    {
+      "epoch": 0.253,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.189231761657867e+18,
+      "learning_rate": 8.503923839688667e-05,
+      "loss": 6.5272,
+      "loss/crossentropy": 1.1747866570949554,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1295059323310852,
+      "step": 1518
+    },
+    {
+      "epoch": 0.25316666666666665,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.1892317614261317e+18,
+      "learning_rate": 8.502055753919032e-05,
+      "loss": 6.4421,
+      "loss/crossentropy": 1.1869618147611618,
+      "loss/hidden": 3.13671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1322761494666338,
+      "step": 1519
+    },
+    {
+      "epoch": 0.25333333333333335,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.1892317613488865e+18,
+      "learning_rate": 8.500186708040949e-05,
+      "loss": 6.7006,
+      "loss/crossentropy": 1.3832582533359528,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17695295624434948,
+      "step": 1520
+    },
+    {
+      "epoch": 0.2535,
+      "grad_norm": 27.375,
+      "grad_norm_var": 1.1892317614034125e+18,
+      "learning_rate": 8.498316702566828e-05,
+      "loss": 7.0269,
+      "loss/crossentropy": 1.9856608510017395,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19019784405827522,
+      "step": 1521
+    },
+    {
+      "epoch": 0.25366666666666665,
+      "grad_norm": 26.125,
+      "grad_norm_var": 1.1892317614397632e+18,
+      "learning_rate": 8.496445738009342e-05,
+      "loss": 6.4003,
+      "loss/crossentropy": 1.2322729229927063,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14161593653261662,
+      "step": 1522
+    },
+    {
+      "epoch": 0.25383333333333336,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.1892317614397632e+18,
+      "learning_rate": 8.494573814881426e-05,
+      "loss": 6.7585,
+      "loss/crossentropy": 1.3647176027297974,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23197595309466124,
+      "step": 1523
+    },
+    {
+      "epoch": 0.254,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.1892317613443428e+18,
+      "learning_rate": 8.49270093369628e-05,
+      "loss": 6.4966,
+      "loss/crossentropy": 1.4593277126550674,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12590165250003338,
+      "step": 1524
+    },
+    {
+      "epoch": 0.25416666666666665,
+      "grad_norm": 54.5,
+      "grad_norm_var": 1.1892317603628756e+18,
+      "learning_rate": 8.490827094967363e-05,
+      "loss": 6.9594,
+      "loss/crossentropy": 1.4991252273321152,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15912701189517975,
+      "step": 1525
+    },
+    {
+      "epoch": 0.25433333333333336,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.1892317601584033e+18,
+      "learning_rate": 8.488952299208401e-05,
+      "loss": 6.8981,
+      "loss/crossentropy": 1.668123185634613,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17216939851641655,
+      "step": 1526
+    },
+    {
+      "epoch": 0.2545,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.189231760090246e+18,
+      "learning_rate": 8.487076546933378e-05,
+      "loss": 6.4686,
+      "loss/crossentropy": 1.8503113985061646,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17350387945771217,
+      "step": 1527
+    },
+    {
+      "epoch": 0.25466666666666665,
+      "grad_norm": 41.5,
+      "grad_norm_var": 1.189231759554074e+18,
+      "learning_rate": 8.485199838656543e-05,
+      "loss": 6.1896,
+      "loss/crossentropy": 1.2597735971212387,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20131279900670052,
+      "step": 1528
+    },
+    {
+      "epoch": 0.25483333333333336,
+      "grad_norm": 26.75,
+      "grad_norm_var": 1.1892317595131796e+18,
+      "learning_rate": 8.483322174892404e-05,
+      "loss": 6.4049,
+      "loss/crossentropy": 1.3848628252744675,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15812037885189056,
+      "step": 1529
+    },
+    {
+      "epoch": 0.255,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.1892317594404782e+18,
+      "learning_rate": 8.481443556155735e-05,
+      "loss": 6.8139,
+      "loss/crossentropy": 1.1715636402368546,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18708932399749756,
+      "step": 1530
+    },
+    {
+      "epoch": 0.25516666666666665,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.189231759367777e+18,
+      "learning_rate": 8.479563982961571e-05,
+      "loss": 6.8292,
+      "loss/crossentropy": 1.2184790670871735,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09692073054611683,
+      "step": 1531
+    },
+    {
+      "epoch": 0.25533333333333336,
+      "grad_norm": 33.75,
+      "grad_norm_var": 1.1892317592905318e+18,
+      "learning_rate": 8.477683455825207e-05,
+      "loss": 6.6891,
+      "loss/crossentropy": 1.6362611651420593,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13843068294227123,
+      "step": 1532
+    },
+    {
+      "epoch": 0.2555,
+      "grad_norm": 28.125,
+      "grad_norm_var": 50.97337239583333,
+      "learning_rate": 8.4758019752622e-05,
+      "loss": 6.7058,
+      "loss/crossentropy": 1.6494111716747284,
+      "loss/hidden": 3.09765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1929592378437519,
+      "step": 1533
+    },
+    {
+      "epoch": 0.25566666666666665,
+      "grad_norm": 26.75,
+      "grad_norm_var": 52.291080729166666,
+      "learning_rate": 8.473919541788366e-05,
+      "loss": 6.274,
+      "loss/crossentropy": 1.3822196274995804,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18195270374417305,
+      "step": 1534
+    },
+    {
+      "epoch": 0.25583333333333336,
+      "grad_norm": 27.5,
+      "grad_norm_var": 53.1119140625,
+      "learning_rate": 8.472036155919791e-05,
+      "loss": 6.731,
+      "loss/crossentropy": 1.2519934549927711,
+      "loss/hidden": 2.98828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10057607851922512,
+      "step": 1535
+    },
+    {
+      "epoch": 0.256,
+      "grad_norm": 38.5,
+      "grad_norm_var": 56.328059895833334,
+      "learning_rate": 8.470151818172809e-05,
+      "loss": 6.7967,
+      "loss/crossentropy": 1.737927407026291,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22391091287136078,
+      "step": 1536
+    },
+    {
+      "epoch": 0.25616666666666665,
+      "grad_norm": 24.875,
+      "grad_norm_var": 58.0962890625,
+      "learning_rate": 8.468266529064025e-05,
+      "loss": 6.1896,
+      "loss/crossentropy": 1.3031770437955856,
+      "loss/hidden": 3.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12873683124780655,
+      "step": 1537
+    },
+    {
+      "epoch": 0.25633333333333336,
+      "grad_norm": 23.875,
+      "grad_norm_var": 59.9806640625,
+      "learning_rate": 8.466380289110303e-05,
+      "loss": 5.8714,
+      "loss/crossentropy": 1.1562596336007118,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15082458686083555,
+      "step": 1538
+    },
+    {
+      "epoch": 0.2565,
+      "grad_norm": 26.75,
+      "grad_norm_var": 60.804622395833334,
+      "learning_rate": 8.464493098828763e-05,
+      "loss": 6.41,
+      "loss/crossentropy": 1.9813059270381927,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17487092316150665,
+      "step": 1539
+    },
+    {
+      "epoch": 0.25666666666666665,
+      "grad_norm": 25.25,
+      "grad_norm_var": 62.462239583333336,
+      "learning_rate": 8.462604958736793e-05,
+      "loss": 6.0257,
+      "loss/crossentropy": 1.2350340336561203,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13172156736254692,
+      "step": 1540
+    },
+    {
+      "epoch": 0.25683333333333336,
+      "grad_norm": 27.75,
+      "grad_norm_var": 23.089583333333334,
+      "learning_rate": 8.460715869352035e-05,
+      "loss": 6.6258,
+      "loss/crossentropy": 1.2772248834371567,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18044393695890903,
+      "step": 1541
+    },
+    {
+      "epoch": 0.257,
+      "grad_norm": 39.25,
+      "grad_norm_var": 29.362239583333334,
+      "learning_rate": 8.458825831192392e-05,
+      "loss": 5.978,
+      "loss/crossentropy": 1.04556904733181,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1417969260364771,
+      "step": 1542
+    },
+    {
+      "epoch": 0.25716666666666665,
+      "grad_norm": 30.625,
+      "grad_norm_var": 29.29140625,
+      "learning_rate": 8.456934844776032e-05,
+      "loss": 7.0604,
+      "loss/crossentropy": 1.921929270029068,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22601721063256264,
+      "step": 1543
+    },
+    {
+      "epoch": 0.25733333333333336,
+      "grad_norm": 27.5,
+      "grad_norm_var": 19.987239583333334,
+      "learning_rate": 8.455042910621379e-05,
+      "loss": 6.2624,
+      "loss/crossentropy": 0.8490016311407089,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.08929689973592758,
+      "step": 1544
+    },
+    {
+      "epoch": 0.2575,
+      "grad_norm": 26.875,
+      "grad_norm_var": 19.9494140625,
+      "learning_rate": 8.453150029247114e-05,
+      "loss": 6.6527,
+      "loss/crossentropy": 1.2423966974020004,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20903770253062248,
+      "step": 1545
+    },
+    {
+      "epoch": 0.25766666666666665,
+      "grad_norm": 24.75,
+      "grad_norm_var": 21.122916666666665,
+      "learning_rate": 8.451256201172186e-05,
+      "loss": 6.319,
+      "loss/crossentropy": 1.611857533454895,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19947092235088348,
+      "step": 1546
+    },
+    {
+      "epoch": 0.25783333333333336,
+      "grad_norm": 25.0,
+      "grad_norm_var": 22.0291015625,
+      "learning_rate": 8.449361426915797e-05,
+      "loss": 6.5528,
+      "loss/crossentropy": 1.5101568177342415,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1427165437489748,
+      "step": 1547
+    },
+    {
+      "epoch": 0.258,
+      "grad_norm": 28.125,
+      "grad_norm_var": 20.121875,
+      "learning_rate": 8.447465706997408e-05,
+      "loss": 6.5579,
+      "loss/crossentropy": 1.8341155499219894,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1619478166103363,
+      "step": 1548
+    },
+    {
+      "epoch": 0.25816666666666666,
+      "grad_norm": 27.25,
+      "grad_norm_var": 20.1806640625,
+      "learning_rate": 8.445569041936743e-05,
+      "loss": 6.5625,
+      "loss/crossentropy": 1.8451082110404968,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2067696861922741,
+      "step": 1549
+    },
+    {
+      "epoch": 0.25833333333333336,
+      "grad_norm": 27.125,
+      "grad_norm_var": 20.11875,
+      "learning_rate": 8.443671432253784e-05,
+      "loss": 6.5914,
+      "loss/crossentropy": 1.3312060832977295,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1642481330782175,
+      "step": 1550
+    },
+    {
+      "epoch": 0.2585,
+      "grad_norm": 30.625,
+      "grad_norm_var": 20.442643229166666,
+      "learning_rate": 8.44177287846877e-05,
+      "loss": 6.6404,
+      "loss/crossentropy": 1.4000188559293747,
+      "loss/hidden": 3.66796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24680931121110916,
+      "step": 1551
+    },
+    {
+      "epoch": 0.25866666666666666,
+      "grad_norm": 24.625,
+      "grad_norm_var": 13.758072916666666,
+      "learning_rate": 8.439873381102203e-05,
+      "loss": 6.1384,
+      "loss/crossentropy": 1.4691539406776428,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13326716050505638,
+      "step": 1552
+    },
+    {
+      "epoch": 0.25883333333333336,
+      "grad_norm": 28.375,
+      "grad_norm_var": 13.29140625,
+      "learning_rate": 8.437972940674838e-05,
+      "loss": 6.8103,
+      "loss/crossentropy": 1.243903011083603,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1517253126949072,
+      "step": 1553
+    },
+    {
+      "epoch": 0.259,
+      "grad_norm": 23.875,
+      "grad_norm_var": 13.29140625,
+      "learning_rate": 8.436071557707692e-05,
+      "loss": 6.3399,
+      "loss/crossentropy": 1.2234562933444977,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16726651415228844,
+      "step": 1554
+    },
+    {
+      "epoch": 0.25916666666666666,
+      "grad_norm": 26.25,
+      "grad_norm_var": 13.37265625,
+      "learning_rate": 8.434169232722043e-05,
+      "loss": 6.2494,
+      "loss/crossentropy": 1.3654979467391968,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16484334878623486,
+      "step": 1555
+    },
+    {
+      "epoch": 0.25933333333333336,
+      "grad_norm": 26.75,
+      "grad_norm_var": 13.02265625,
+      "learning_rate": 8.432265966239419e-05,
+      "loss": 7.0852,
+      "loss/crossentropy": 2.001293331384659,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19069082662463188,
+      "step": 1556
+    },
+    {
+      "epoch": 0.2595,
+      "grad_norm": 29.5,
+      "grad_norm_var": 13.203125,
+      "learning_rate": 8.430361758781616e-05,
+      "loss": 6.7866,
+      "loss/crossentropy": 2.1050551533699036,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24255291372537613,
+      "step": 1557
+    },
+    {
+      "epoch": 0.25966666666666666,
+      "grad_norm": 27.375,
+      "grad_norm_var": 4.0556640625,
+      "learning_rate": 8.42845661087068e-05,
+      "loss": 6.5634,
+      "loss/crossentropy": 1.4958145320415497,
+      "loss/hidden": 3.7890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2499004602432251,
+      "step": 1558
+    },
+    {
+      "epoch": 0.25983333333333336,
+      "grad_norm": 24.625,
+      "grad_norm_var": 3.5369140625,
+      "learning_rate": 8.42655052302892e-05,
+      "loss": 5.9745,
+      "loss/crossentropy": 1.2098843902349472,
+      "loss/hidden": 3.10546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11175981163978577,
+      "step": 1559
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.753059895833333,
+      "learning_rate": 8.424643495778902e-05,
+      "loss": 6.377,
+      "loss/crossentropy": 1.6902850717306137,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17755519971251488,
+      "step": 1560
+    },
+    {
+      "epoch": 0.26016666666666666,
+      "grad_norm": 26.125,
+      "grad_norm_var": 3.7874348958333335,
+      "learning_rate": 8.422735529643444e-05,
+      "loss": 6.1022,
+      "loss/crossentropy": 1.4672711193561554,
+      "loss/hidden": 3.01953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11636890098452568,
+      "step": 1561
+    },
+    {
+      "epoch": 0.26033333333333336,
+      "grad_norm": 27.125,
+      "grad_norm_var": 3.484375,
+      "learning_rate": 8.42082662514563e-05,
+      "loss": 6.988,
+      "loss/crossentropy": 1.8623914122581482,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18818616680800915,
+      "step": 1562
+    },
+    {
+      "epoch": 0.2605,
+      "grad_norm": 27.0,
+      "grad_norm_var": 3.209375,
+      "learning_rate": 8.418916782808795e-05,
+      "loss": 6.2558,
+      "loss/crossentropy": 1.8625478744506836,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19597026705741882,
+      "step": 1563
+    },
+    {
+      "epoch": 0.26066666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.734375,
+      "learning_rate": 8.417006003156532e-05,
+      "loss": 6.8321,
+      "loss/crossentropy": 1.1682617217302322,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2345657404512167,
+      "step": 1564
+    },
+    {
+      "epoch": 0.2608333333333333,
+      "grad_norm": 25.625,
+      "grad_norm_var": 3.8926432291666666,
+      "learning_rate": 8.415094286712694e-05,
+      "loss": 6.4499,
+      "loss/crossentropy": 1.6742401123046875,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13021016493439674,
+      "step": 1565
+    },
+    {
+      "epoch": 0.261,
+      "grad_norm": 28.875,
+      "grad_norm_var": 4.085872395833333,
+      "learning_rate": 8.413181634001391e-05,
+      "loss": 6.7084,
+      "loss/crossentropy": 1.5285805314779282,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1491610729135573,
+      "step": 1566
+    },
+    {
+      "epoch": 0.26116666666666666,
+      "grad_norm": 31.75,
+      "grad_norm_var": 4.674739583333333,
+      "learning_rate": 8.411268045546983e-05,
+      "loss": 6.8492,
+      "loss/crossentropy": 1.7009564563632011,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1385979875922203,
+      "step": 1567
+    },
+    {
+      "epoch": 0.2613333333333333,
+      "grad_norm": 28.125,
+      "grad_norm_var": 4.193489583333333,
+      "learning_rate": 8.409353521874093e-05,
+      "loss": 6.4426,
+      "loss/crossentropy": 1.229106530547142,
+      "loss/hidden": 3.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11261659115552902,
+      "step": 1568
+    },
+    {
+      "epoch": 0.2615,
+      "grad_norm": 25.0,
+      "grad_norm_var": 4.518684895833333,
+      "learning_rate": 8.4074380635076e-05,
+      "loss": 6.1533,
+      "loss/crossentropy": 1.7832941710948944,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17035359889268875,
+      "step": 1569
+    },
+    {
+      "epoch": 0.26166666666666666,
+      "grad_norm": 25.25,
+      "grad_norm_var": 4.008072916666666,
+      "learning_rate": 8.405521670972634e-05,
+      "loss": 6.6411,
+      "loss/crossentropy": 1.351745367050171,
+      "loss/hidden": 3.0859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11264829337596893,
+      "step": 1570
+    },
+    {
+      "epoch": 0.2618333333333333,
+      "grad_norm": 27.25,
+      "grad_norm_var": 3.9184895833333333,
+      "learning_rate": 8.40360434479459e-05,
+      "loss": 6.5477,
+      "loss/crossentropy": 1.6367198377847672,
+      "loss/hidden": 3.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14650331810116768,
+      "step": 1571
+    },
+    {
+      "epoch": 0.262,
+      "grad_norm": 24.875,
+      "grad_norm_var": 4.313997395833334,
+      "learning_rate": 8.40168608549911e-05,
+      "loss": 6.2995,
+      "loss/crossentropy": 0.9696106985211372,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12367397360503674,
+      "step": 1572
+    },
+    {
+      "epoch": 0.26216666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 4.1853515625,
+      "learning_rate": 8.399766893612096e-05,
+      "loss": 6.4659,
+      "loss/crossentropy": 1.5791050642728806,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18380821123719215,
+      "step": 1573
+    },
+    {
+      "epoch": 0.2623333333333333,
+      "grad_norm": 26.25,
+      "grad_norm_var": 4.25390625,
+      "learning_rate": 8.397846769659707e-05,
+      "loss": 6.3502,
+      "loss/crossentropy": 1.3836457431316376,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17323687486350536,
+      "step": 1574
+    },
+    {
+      "epoch": 0.2625,
+      "grad_norm": 25.25,
+      "grad_norm_var": 4.060872395833333,
+      "learning_rate": 8.395925714168356e-05,
+      "loss": 6.447,
+      "loss/crossentropy": 1.28899946808815,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12180336937308311,
+      "step": 1575
+    },
+    {
+      "epoch": 0.26266666666666666,
+      "grad_norm": 27.625,
+      "grad_norm_var": 3.9184895833333333,
+      "learning_rate": 8.39400372766471e-05,
+      "loss": 6.2646,
+      "loss/crossentropy": 0.9761116281151772,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20371597120538354,
+      "step": 1576
+    },
+    {
+      "epoch": 0.2628333333333333,
+      "grad_norm": 38.75,
+      "grad_norm_var": 12.065559895833333,
+      "learning_rate": 8.392080810675691e-05,
+      "loss": 6.7719,
+      "loss/crossentropy": 1.8322997987270355,
+      "loss/hidden": 3.59375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23163571022450924,
+      "step": 1577
+    },
+    {
+      "epoch": 0.263,
+      "grad_norm": 26.25,
+      "grad_norm_var": 12.214583333333334,
+      "learning_rate": 8.390156963728482e-05,
+      "loss": 6.1365,
+      "loss/crossentropy": 1.194678246974945,
+      "loss/hidden": 3.56640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2539881942793727,
+      "step": 1578
+    },
+    {
+      "epoch": 0.26316666666666666,
+      "grad_norm": 23.875,
+      "grad_norm_var": 13.215559895833334,
+      "learning_rate": 8.388232187350512e-05,
+      "loss": 6.0593,
+      "loss/crossentropy": 1.3664326220750809,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10707986168563366,
+      "step": 1579
+    },
+    {
+      "epoch": 0.2633333333333333,
+      "grad_norm": 28.5,
+      "grad_norm_var": 12.864322916666667,
+      "learning_rate": 8.386306482069473e-05,
+      "loss": 6.8437,
+      "loss/crossentropy": 1.850177824497223,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17907489091157913,
+      "step": 1580
+    },
+    {
+      "epoch": 0.2635,
+      "grad_norm": 27.75,
+      "grad_norm_var": 12.575455729166666,
+      "learning_rate": 8.384379848413304e-05,
+      "loss": 6.6427,
+      "loss/crossentropy": 1.733255460858345,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1370087955147028,
+      "step": 1581
+    },
+    {
+      "epoch": 0.26366666666666666,
+      "grad_norm": 26.75,
+      "grad_norm_var": 12.545572916666666,
+      "learning_rate": 8.382452286910206e-05,
+      "loss": 6.2584,
+      "loss/crossentropy": 1.3558062016963959,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13898317702114582,
+      "step": 1582
+    },
+    {
+      "epoch": 0.2638333333333333,
+      "grad_norm": 27.75,
+      "grad_norm_var": 11.35390625,
+      "learning_rate": 8.380523798088631e-05,
+      "loss": 6.7305,
+      "loss/crossentropy": 1.2056031823158264,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10830382723361254,
+      "step": 1583
+    },
+    {
+      "epoch": 0.264,
+      "grad_norm": 26.75,
+      "grad_norm_var": 11.337434895833333,
+      "learning_rate": 8.378594382477282e-05,
+      "loss": 6.6729,
+      "loss/crossentropy": 2.236360639333725,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25125929713249207,
+      "step": 1584
+    },
+    {
+      "epoch": 0.26416666666666666,
+      "grad_norm": 24.75,
+      "grad_norm_var": 11.4181640625,
+      "learning_rate": 8.376664040605122e-05,
+      "loss": 6.386,
+      "loss/crossentropy": 0.9779779016971588,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11395659483969212,
+      "step": 1585
+    },
+    {
+      "epoch": 0.2643333333333333,
+      "grad_norm": 26.625,
+      "grad_norm_var": 11.1625,
+      "learning_rate": 8.374732773001366e-05,
+      "loss": 6.3123,
+      "loss/crossentropy": 1.4567594081163406,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17596044018864632,
+      "step": 1586
+    },
+    {
+      "epoch": 0.2645,
+      "grad_norm": 35.5,
+      "grad_norm_var": 15.27890625,
+      "learning_rate": 8.372800580195479e-05,
+      "loss": 6.7364,
+      "loss/crossentropy": 1.9195697903633118,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16548319160938263,
+      "step": 1587
+    },
+    {
+      "epoch": 0.26466666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 14.9947265625,
+      "learning_rate": 8.370867462717183e-05,
+      "loss": 6.7132,
+      "loss/crossentropy": 1.2127940505743027,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09943809919059277,
+      "step": 1588
+    },
+    {
+      "epoch": 0.2648333333333333,
+      "grad_norm": 26.875,
+      "grad_norm_var": 15.062239583333334,
+      "learning_rate": 8.368933421096454e-05,
+      "loss": 6.7852,
+      "loss/crossentropy": 0.9601086676120758,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11084135621786118,
+      "step": 1589
+    },
+    {
+      "epoch": 0.265,
+      "grad_norm": 24.625,
+      "grad_norm_var": 15.630143229166666,
+      "learning_rate": 8.366998455863522e-05,
+      "loss": 6.333,
+      "loss/crossentropy": 1.4967469424009323,
+      "loss/hidden": 3.2578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1625440139323473,
+      "step": 1590
+    },
+    {
+      "epoch": 0.26516666666666666,
+      "grad_norm": 26.0,
+      "grad_norm_var": 15.389518229166667,
+      "learning_rate": 8.365062567548867e-05,
+      "loss": 6.2864,
+      "loss/crossentropy": 1.6240037083625793,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15611721575260162,
+      "step": 1591
+    },
+    {
+      "epoch": 0.2653333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 15.605989583333333,
+      "learning_rate": 8.363125756683223e-05,
+      "loss": 6.6876,
+      "loss/crossentropy": 1.5841274857521057,
+      "loss/hidden": 3.83203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2983058914542198,
+      "step": 1592
+    },
+    {
+      "epoch": 0.2655,
+      "grad_norm": 24.875,
+      "grad_norm_var": 8.126497395833333,
+      "learning_rate": 8.361188023797582e-05,
+      "loss": 6.6541,
+      "loss/crossentropy": 1.7854099571704865,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14649183489382267,
+      "step": 1593
+    },
+    {
+      "epoch": 0.26566666666666666,
+      "grad_norm": 28.25,
+      "grad_norm_var": 8.0869140625,
+      "learning_rate": 8.359249369423177e-05,
+      "loss": 6.2296,
+      "loss/crossentropy": 1.119899682700634,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12407440692186356,
+      "step": 1594
+    },
+    {
+      "epoch": 0.2658333333333333,
+      "grad_norm": 27.125,
+      "grad_norm_var": 7.1931640625,
+      "learning_rate": 8.357309794091507e-05,
+      "loss": 6.3103,
+      "loss/crossentropy": 1.140097290277481,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11549050640314817,
+      "step": 1595
+    },
+    {
+      "epoch": 0.266,
+      "grad_norm": 28.125,
+      "grad_norm_var": 7.16015625,
+      "learning_rate": 8.355369298334316e-05,
+      "loss": 6.7788,
+      "loss/crossentropy": 1.529033437371254,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21456463262438774,
+      "step": 1596
+    },
+    {
+      "epoch": 0.26616666666666666,
+      "grad_norm": 26.625,
+      "grad_norm_var": 7.2228515625,
+      "learning_rate": 8.3534278826836e-05,
+      "loss": 6.7892,
+      "loss/crossentropy": 1.5850271880626678,
+      "loss/hidden": 3.20703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.130172586068511,
+      "step": 1597
+    },
+    {
+      "epoch": 0.2663333333333333,
+      "grad_norm": 25.25,
+      "grad_norm_var": 7.5275390625,
+      "learning_rate": 8.351485547671613e-05,
+      "loss": 6.3052,
+      "loss/crossentropy": 1.7502561509609222,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15570508502423763,
+      "step": 1598
+    },
+    {
+      "epoch": 0.2665,
+      "grad_norm": 25.125,
+      "grad_norm_var": 7.8625,
+      "learning_rate": 8.349542293830855e-05,
+      "loss": 6.2877,
+      "loss/crossentropy": 1.6792922914028168,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23954294063150883,
+      "step": 1599
+    },
+    {
+      "epoch": 0.26666666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 8.01015625,
+      "learning_rate": 8.347598121694078e-05,
+      "loss": 6.6913,
+      "loss/crossentropy": 1.8723644018173218,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20333952084183693,
+      "step": 1600
+    },
+    {
+      "epoch": 0.2668333333333333,
+      "grad_norm": 28.5,
+      "grad_norm_var": 7.5375,
+      "learning_rate": 8.345653031794292e-05,
+      "loss": 7.1074,
+      "loss/crossentropy": 1.974313199520111,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.148883655667305,
+      "step": 1601
+    },
+    {
+      "epoch": 0.267,
+      "grad_norm": 26.375,
+      "grad_norm_var": 7.576822916666667,
+      "learning_rate": 8.343707024664751e-05,
+      "loss": 6.6407,
+      "loss/crossentropy": 1.3841271996498108,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16643064469099045,
+      "step": 1602
+    },
+    {
+      "epoch": 0.26716666666666666,
+      "grad_norm": 27.5,
+      "grad_norm_var": 3.226822916666667,
+      "learning_rate": 8.341760100838965e-05,
+      "loss": 6.6656,
+      "loss/crossentropy": 1.0456174314022064,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13179867714643478,
+      "step": 1603
+    },
+    {
+      "epoch": 0.2673333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.670768229166667,
+      "learning_rate": 8.339812260850696e-05,
+      "loss": 7.5224,
+      "loss/crossentropy": 2.02323842048645,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26648833230137825,
+      "step": 1604
+    },
+    {
+      "epoch": 0.2675,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.6681640625,
+      "learning_rate": 8.337863505233953e-05,
+      "loss": 6.3042,
+      "loss/crossentropy": 1.5908322036266327,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1915053054690361,
+      "step": 1605
+    },
+    {
+      "epoch": 0.26766666666666666,
+      "grad_norm": 26.25,
+      "grad_norm_var": 2.3,
+      "learning_rate": 8.335913834522999e-05,
+      "loss": 6.3944,
+      "loss/crossentropy": 1.5703306943178177,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13484930247068405,
+      "step": 1606
+    },
+    {
+      "epoch": 0.2678333333333333,
+      "grad_norm": 26.25,
+      "grad_norm_var": 2.2643229166666665,
+      "learning_rate": 8.333963249252348e-05,
+      "loss": 6.2017,
+      "loss/crossentropy": 1.5306211411952972,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24524124339222908,
+      "step": 1607
+    },
+    {
+      "epoch": 0.268,
+      "grad_norm": 26.5,
+      "grad_norm_var": 1.7247395833333334,
+      "learning_rate": 8.332011749956763e-05,
+      "loss": 6.6625,
+      "loss/crossentropy": 2.0105501264333725,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19223702698946,
+      "step": 1608
+    },
+    {
+      "epoch": 0.26816666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.5997395833333334,
+      "learning_rate": 8.330059337171258e-05,
+      "loss": 6.4999,
+      "loss/crossentropy": 1.5219484716653824,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13616113737225533,
+      "step": 1609
+    },
+    {
+      "epoch": 0.2683333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.79765625,
+      "learning_rate": 8.328106011431101e-05,
+      "loss": 6.7819,
+      "loss/crossentropy": 1.3084082007408142,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22154392953962088,
+      "step": 1610
+    },
+    {
+      "epoch": 0.2685,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.29140625,
+      "learning_rate": 8.326151773271804e-05,
+      "loss": 6.8103,
+      "loss/crossentropy": 1.229638785123825,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20553592592477798,
+      "step": 1611
+    },
+    {
+      "epoch": 0.26866666666666666,
+      "grad_norm": 26.125,
+      "grad_norm_var": 2.3705729166666667,
+      "learning_rate": 8.324196623229135e-05,
+      "loss": 6.2954,
+      "loss/crossentropy": 1.427354633808136,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14054181426763535,
+      "step": 1612
+    },
+    {
+      "epoch": 0.2688333333333333,
+      "grad_norm": 25.5,
+      "grad_norm_var": 2.559830729166667,
+      "learning_rate": 8.322240561839109e-05,
+      "loss": 6.4815,
+      "loss/crossentropy": 1.7224322333931923,
+      "loss/hidden": 3.1171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16631003841757774,
+      "step": 1613
+    },
+    {
+      "epoch": 0.269,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.2660807291666667,
+      "learning_rate": 8.32028358963799e-05,
+      "loss": 6.5355,
+      "loss/crossentropy": 1.6903588771820068,
+      "loss/hidden": 3.0703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13924129493534565,
+      "step": 1614
+    },
+    {
+      "epoch": 0.26916666666666667,
+      "grad_norm": 27.5,
+      "grad_norm_var": 1.89375,
+      "learning_rate": 8.318325707162293e-05,
+      "loss": 6.9817,
+      "loss/crossentropy": 1.8187223076820374,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17342381179332733,
+      "step": 1615
+    },
+    {
+      "epoch": 0.2693333333333333,
+      "grad_norm": 25.75,
+      "grad_norm_var": 1.9309895833333333,
+      "learning_rate": 8.316366914948783e-05,
+      "loss": 6.5086,
+      "loss/crossentropy": 1.3641190081834793,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18485832773149014,
+      "step": 1616
+    },
+    {
+      "epoch": 0.2695,
+      "grad_norm": 25.625,
+      "grad_norm_var": 2.0103515625,
+      "learning_rate": 8.314407213534476e-05,
+      "loss": 6.2474,
+      "loss/crossentropy": 1.3395388424396515,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13723072689026594,
+      "step": 1617
+    },
+    {
+      "epoch": 0.26966666666666667,
+      "grad_norm": 25.75,
+      "grad_norm_var": 2.101822916666667,
+      "learning_rate": 8.312446603456632e-05,
+      "loss": 6.5276,
+      "loss/crossentropy": 1.1821232289075851,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1599526572972536,
+      "step": 1618
+    },
+    {
+      "epoch": 0.2698333333333333,
+      "grad_norm": 25.875,
+      "grad_norm_var": 2.1889973958333333,
+      "learning_rate": 8.310485085252767e-05,
+      "loss": 6.6453,
+      "loss/crossentropy": 1.1522404998540878,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11603707447648048,
+      "step": 1619
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 27.0,
+      "grad_norm_var": 1.9497395833333333,
+      "learning_rate": 8.308522659460641e-05,
+      "loss": 6.1391,
+      "loss/crossentropy": 1.7048717737197876,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26493560522794724,
+      "step": 1620
+    },
+    {
+      "epoch": 0.27016666666666667,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.9497395833333333,
+      "learning_rate": 8.306559326618259e-05,
+      "loss": 6.9394,
+      "loss/crossentropy": 1.5971474349498749,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23606881871819496,
+      "step": 1621
+    },
+    {
+      "epoch": 0.2703333333333333,
+      "grad_norm": 24.75,
+      "grad_norm_var": 2.224739583333333,
+      "learning_rate": 8.304595087263889e-05,
+      "loss": 6.1608,
+      "loss/crossentropy": 1.8884839713573456,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1541367694735527,
+      "step": 1622
+    },
+    {
+      "epoch": 0.2705,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.5559895833333335,
+      "learning_rate": 8.30262994193603e-05,
+      "loss": 6.8455,
+      "loss/crossentropy": 1.5760976076126099,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1386857032775879,
+      "step": 1623
+    },
+    {
+      "epoch": 0.27066666666666667,
+      "grad_norm": 24.75,
+      "grad_norm_var": 2.8677083333333333,
+      "learning_rate": 8.300663891173443e-05,
+      "loss": 6.61,
+      "loss/crossentropy": 1.636210411787033,
+      "loss/hidden": 3.01171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14123854786157608,
+      "step": 1624
+    },
+    {
+      "epoch": 0.2708333333333333,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.5921223958333335,
+      "learning_rate": 8.298696935515132e-05,
+      "loss": 6.5315,
+      "loss/crossentropy": 1.2037115097045898,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13143595308065414,
+      "step": 1625
+    },
+    {
+      "epoch": 0.271,
+      "grad_norm": 26.0,
+      "grad_norm_var": 2.1791015625,
+      "learning_rate": 8.296729075500344e-05,
+      "loss": 6.4103,
+      "loss/crossentropy": 1.3744913935661316,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16979750245809555,
+      "step": 1626
+    },
+    {
+      "epoch": 0.27116666666666667,
+      "grad_norm": 27.5,
+      "grad_norm_var": 1.365625,
+      "learning_rate": 8.294760311668586e-05,
+      "loss": 6.4506,
+      "loss/crossentropy": 1.196798488497734,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1543511599302292,
+      "step": 1627
+    },
+    {
+      "epoch": 0.2713333333333333,
+      "grad_norm": 25.75,
+      "grad_norm_var": 1.3884765625,
+      "learning_rate": 8.2927906445596e-05,
+      "loss": 6.4978,
+      "loss/crossentropy": 1.481429547071457,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17183401808142662,
+      "step": 1628
+    },
+    {
+      "epoch": 0.2715,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.703125,
+      "learning_rate": 8.290820074713384e-05,
+      "loss": 6.4526,
+      "loss/crossentropy": 1.618128389120102,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21382078900933266,
+      "step": 1629
+    },
+    {
+      "epoch": 0.27166666666666667,
+      "grad_norm": 28.5,
+      "grad_norm_var": 1.91015625,
+      "learning_rate": 8.28884860267018e-05,
+      "loss": 6.5838,
+      "loss/crossentropy": 2.148876816034317,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19500019028782845,
+      "step": 1630
+    },
+    {
+      "epoch": 0.2718333333333333,
+      "grad_norm": 26.5,
+      "grad_norm_var": 1.8622395833333334,
+      "learning_rate": 8.28687622897048e-05,
+      "loss": 5.9952,
+      "loss/crossentropy": 1.6788550913333893,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1563299372792244,
+      "step": 1631
+    },
+    {
+      "epoch": 0.272,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.114518229166667,
+      "learning_rate": 8.284902954155019e-05,
+      "loss": 6.8839,
+      "loss/crossentropy": 1.2067348659038544,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16382478922605515,
+      "step": 1632
+    },
+    {
+      "epoch": 0.27216666666666667,
+      "grad_norm": 24.75,
+      "grad_norm_var": 2.3,
+      "learning_rate": 8.282928778764783e-05,
+      "loss": 6.1469,
+      "loss/crossentropy": 1.2932213842868805,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14870959147810936,
+      "step": 1633
+    },
+    {
+      "epoch": 0.2723333333333333,
+      "grad_norm": 26.375,
+      "grad_norm_var": 2.241080729166667,
+      "learning_rate": 8.280953703341004e-05,
+      "loss": 6.2526,
+      "loss/crossentropy": 0.8585210591554642,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14180009625852108,
+      "step": 1634
+    },
+    {
+      "epoch": 0.2725,
+      "grad_norm": 27.0,
+      "grad_norm_var": 2.1830729166666667,
+      "learning_rate": 8.278977728425157e-05,
+      "loss": 6.8247,
+      "loss/crossentropy": 2.1694380342960358,
+      "loss/hidden": 3.515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2599971890449524,
+      "step": 1635
+    },
+    {
+      "epoch": 0.27266666666666667,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.1822916666666665,
+      "learning_rate": 8.27700085455897e-05,
+      "loss": 6.3411,
+      "loss/crossentropy": 1.5276328772306442,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20902734249830246,
+      "step": 1636
+    },
+    {
+      "epoch": 0.2728333333333333,
+      "grad_norm": 25.75,
+      "grad_norm_var": 2.248893229166667,
+      "learning_rate": 8.275023082284413e-05,
+      "loss": 6.6245,
+      "loss/crossentropy": 2.339494287967682,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16161833330988884,
+      "step": 1637
+    },
+    {
+      "epoch": 0.273,
+      "grad_norm": 25.875,
+      "grad_norm_var": 2.0268229166666667,
+      "learning_rate": 8.273044412143704e-05,
+      "loss": 6.8519,
+      "loss/crossentropy": 1.2134160548448563,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12258456461131573,
+      "step": 1638
+    },
+    {
+      "epoch": 0.27316666666666667,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.7426432291666667,
+      "learning_rate": 8.271064844679306e-05,
+      "loss": 6.5726,
+      "loss/crossentropy": 1.2051786482334137,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2356526292860508,
+      "step": 1639
+    },
+    {
+      "epoch": 0.2733333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.5934895833333333,
+      "learning_rate": 8.269084380433929e-05,
+      "loss": 6.3446,
+      "loss/crossentropy": 1.6763530224561691,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22781694680452347,
+      "step": 1640
+    },
+    {
+      "epoch": 0.2735,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.7809895833333333,
+      "learning_rate": 8.267103019950529e-05,
+      "loss": 7.0837,
+      "loss/crossentropy": 1.754968911409378,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14769989624619484,
+      "step": 1641
+    },
+    {
+      "epoch": 0.27366666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.846875,
+      "learning_rate": 8.265120763772303e-05,
+      "loss": 6.7462,
+      "loss/crossentropy": 1.3203455954790115,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17968839779496193,
+      "step": 1642
+    },
+    {
+      "epoch": 0.2738333333333333,
+      "grad_norm": 25.25,
+      "grad_norm_var": 2.09765625,
+      "learning_rate": 8.263137612442706e-05,
+      "loss": 6.4217,
+      "loss/crossentropy": 1.823981136083603,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16684725135564804,
+      "step": 1643
+    },
+    {
+      "epoch": 0.274,
+      "grad_norm": 25.875,
+      "grad_norm_var": 2.075455729166667,
+      "learning_rate": 8.261153566505424e-05,
+      "loss": 6.4205,
+      "loss/crossentropy": 1.4490568339824677,
+      "loss/hidden": 3.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14278194680809975,
+      "step": 1644
+    },
+    {
+      "epoch": 0.27416666666666667,
+      "grad_norm": 26.75,
+      "grad_norm_var": 1.8684895833333333,
+      "learning_rate": 8.259168626504395e-05,
+      "loss": 6.6623,
+      "loss/crossentropy": 1.949517160654068,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16612081229686737,
+      "step": 1645
+    },
+    {
+      "epoch": 0.2743333333333333,
+      "grad_norm": 26.75,
+      "grad_norm_var": 1.7135416666666667,
+      "learning_rate": 8.257182792983802e-05,
+      "loss": 6.8077,
+      "loss/crossentropy": 1.3662299513816833,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13424787670373917,
+      "step": 1646
+    },
+    {
+      "epoch": 0.2745,
+      "grad_norm": 25.375,
+      "grad_norm_var": 1.8535807291666666,
+      "learning_rate": 8.255196066488075e-05,
+      "loss": 6.3715,
+      "loss/crossentropy": 1.7718872725963593,
+      "loss/hidden": 3.12890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11640601232647896,
+      "step": 1647
+    },
+    {
+      "epoch": 0.27466666666666667,
+      "grad_norm": 24.625,
+      "grad_norm_var": 1.8270182291666666,
+      "learning_rate": 8.253208447561882e-05,
+      "loss": 6.1833,
+      "loss/crossentropy": 1.5564922764897346,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1758194393478334,
+      "step": 1648
+    },
+    {
+      "epoch": 0.2748333333333333,
+      "grad_norm": 25.625,
+      "grad_norm_var": 1.6625,
+      "learning_rate": 8.251219936750144e-05,
+      "loss": 6.1891,
+      "loss/crossentropy": 1.59621924161911,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20559540763497353,
+      "step": 1649
+    },
+    {
+      "epoch": 0.275,
+      "grad_norm": 26.875,
+      "grad_norm_var": 1.6614583333333333,
+      "learning_rate": 8.249230534598021e-05,
+      "loss": 6.5536,
+      "loss/crossentropy": 1.4750879406929016,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15503276512026787,
+      "step": 1650
+    },
+    {
+      "epoch": 0.27516666666666667,
+      "grad_norm": 25.0,
+      "grad_norm_var": 1.8197916666666667,
+      "learning_rate": 8.247240241650918e-05,
+      "loss": 6.18,
+      "loss/crossentropy": 1.089761197566986,
+      "loss/hidden": 3.08984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12325790617614985,
+      "step": 1651
+    },
+    {
+      "epoch": 0.2753333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.283333333333333,
+      "learning_rate": 8.245249058454487e-05,
+      "loss": 6.6376,
+      "loss/crossentropy": 1.4199354499578476,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1404520981013775,
+      "step": 1652
+    },
+    {
+      "epoch": 0.2755,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.412239583333333,
+      "learning_rate": 8.243256985554621e-05,
+      "loss": 6.5786,
+      "loss/crossentropy": 1.7468246966600418,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18941262364387512,
+      "step": 1653
+    },
+    {
+      "epoch": 0.27566666666666667,
+      "grad_norm": 26.875,
+      "grad_norm_var": 2.343489583333333,
+      "learning_rate": 8.241264023497457e-05,
+      "loss": 6.3791,
+      "loss/crossentropy": 1.655304729938507,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2107965499162674,
+      "step": 1654
+    },
+    {
+      "epoch": 0.2758333333333333,
+      "grad_norm": 25.875,
+      "grad_norm_var": 2.298958333333333,
+      "learning_rate": 8.239270172829379e-05,
+      "loss": 6.2451,
+      "loss/crossentropy": 2.322205126285553,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20676938816905022,
+      "step": 1655
+    },
+    {
+      "epoch": 0.276,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.190559895833333,
+      "learning_rate": 8.237275434097012e-05,
+      "loss": 6.6079,
+      "loss/crossentropy": 1.137220598757267,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13325241953134537,
+      "step": 1656
+    },
+    {
+      "epoch": 0.27616666666666667,
+      "grad_norm": 26.75,
+      "grad_norm_var": 1.9051432291666666,
+      "learning_rate": 8.235279807847223e-05,
+      "loss": 6.2816,
+      "loss/crossentropy": 1.6288244128227234,
+      "loss/hidden": 3.09765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15651296451687813,
+      "step": 1657
+    },
+    {
+      "epoch": 0.2763333333333333,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.6080729166666667,
+      "learning_rate": 8.233283294627125e-05,
+      "loss": 6.381,
+      "loss/crossentropy": 1.3976692855358124,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1365256905555725,
+      "step": 1658
+    },
+    {
+      "epoch": 0.2765,
+      "grad_norm": 37.75,
+      "grad_norm_var": 9.264322916666666,
+      "learning_rate": 8.231285894984076e-05,
+      "loss": 6.7407,
+      "loss/crossentropy": 1.2789996713399887,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23060336709022522,
+      "step": 1659
+    },
+    {
+      "epoch": 0.27666666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 9.4681640625,
+      "learning_rate": 8.22928760946567e-05,
+      "loss": 6.846,
+      "loss/crossentropy": 1.5556926429271698,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17659665271639824,
+      "step": 1660
+    },
+    {
+      "epoch": 0.2768333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 10.047330729166667,
+      "learning_rate": 8.227288438619754e-05,
+      "loss": 6.8995,
+      "loss/crossentropy": 2.1362548172473907,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25567465275526047,
+      "step": 1661
+    },
+    {
+      "epoch": 0.277,
+      "grad_norm": 27.375,
+      "grad_norm_var": 9.98515625,
+      "learning_rate": 8.225288382994407e-05,
+      "loss": 6.3478,
+      "loss/crossentropy": 1.5700103342533112,
+      "loss/hidden": 3.078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13714691624045372,
+      "step": 1662
+    },
+    {
+      "epoch": 0.2771666666666667,
+      "grad_norm": 28.0,
+      "grad_norm_var": 9.5572265625,
+      "learning_rate": 8.223287443137957e-05,
+      "loss": 6.7144,
+      "loss/crossentropy": 1.6021295487880707,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16361736506223679,
+      "step": 1663
+    },
+    {
+      "epoch": 0.2773333333333333,
+      "grad_norm": 27.125,
+      "grad_norm_var": 8.825455729166666,
+      "learning_rate": 8.221285619598975e-05,
+      "loss": 6.5565,
+      "loss/crossentropy": 1.6454369276762009,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17043303325772285,
+      "step": 1664
+    },
+    {
+      "epoch": 0.2775,
+      "grad_norm": 28.25,
+      "grad_norm_var": 8.372916666666667,
+      "learning_rate": 8.21928291292627e-05,
+      "loss": 6.7848,
+      "loss/crossentropy": 1.7023286670446396,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20277707651257515,
+      "step": 1665
+    },
+    {
+      "epoch": 0.2776666666666667,
+      "grad_norm": 27.375,
+      "grad_norm_var": 8.292708333333334,
+      "learning_rate": 8.217279323668895e-05,
+      "loss": 6.8649,
+      "loss/crossentropy": 1.3222306594252586,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17969263438135386,
+      "step": 1666
+    },
+    {
+      "epoch": 0.2778333333333333,
+      "grad_norm": 25.625,
+      "grad_norm_var": 8.0384765625,
+      "learning_rate": 8.215274852376147e-05,
+      "loss": 6.1491,
+      "loss/crossentropy": 1.8724027872085571,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12196112424135208,
+      "step": 1667
+    },
+    {
+      "epoch": 0.278,
+      "grad_norm": 23.75,
+      "grad_norm_var": 9.2931640625,
+      "learning_rate": 8.213269499597565e-05,
+      "loss": 5.8931,
+      "loss/crossentropy": 1.1713724061846733,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12717099487781525,
+      "step": 1668
+    },
+    {
+      "epoch": 0.2781666666666667,
+      "grad_norm": 28.0,
+      "grad_norm_var": 9.278059895833334,
+      "learning_rate": 8.211263265882923e-05,
+      "loss": 6.9227,
+      "loss/crossentropy": 1.6045489609241486,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17086264491081238,
+      "step": 1669
+    },
+    {
+      "epoch": 0.2783333333333333,
+      "grad_norm": 28.75,
+      "grad_norm_var": 9.214583333333334,
+      "learning_rate": 8.209256151782243e-05,
+      "loss": 6.7527,
+      "loss/crossentropy": 1.3341772854328156,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1483191940933466,
+      "step": 1670
+    },
+    {
+      "epoch": 0.2785,
+      "grad_norm": 27.375,
+      "grad_norm_var": 8.905208333333333,
+      "learning_rate": 8.207248157845791e-05,
+      "loss": 6.2651,
+      "loss/crossentropy": 1.860980048775673,
+      "loss/hidden": 3.14453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11668968573212624,
+      "step": 1671
+    },
+    {
+      "epoch": 0.2786666666666667,
+      "grad_norm": 26.625,
+      "grad_norm_var": 9.054622395833333,
+      "learning_rate": 8.205239284624062e-05,
+      "loss": 6.6759,
+      "loss/crossentropy": 1.6459452509880066,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22845373675227165,
+      "step": 1672
+    },
+    {
+      "epoch": 0.2788333333333333,
+      "grad_norm": 24.75,
+      "grad_norm_var": 9.6775390625,
+      "learning_rate": 8.203229532667807e-05,
+      "loss": 6.4217,
+      "loss/crossentropy": 1.53298020362854,
+      "loss/hidden": 3.0234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1260205153375864,
+      "step": 1673
+    },
+    {
+      "epoch": 0.279,
+      "grad_norm": 27.875,
+      "grad_norm_var": 9.6228515625,
+      "learning_rate": 8.201218902528009e-05,
+      "loss": 6.4527,
+      "loss/crossentropy": 1.869560956954956,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19912932440638542,
+      "step": 1674
+    },
+    {
+      "epoch": 0.2791666666666667,
+      "grad_norm": 25.25,
+      "grad_norm_var": 3.2556640625,
+      "learning_rate": 8.199207394755893e-05,
+      "loss": 6.0211,
+      "loss/crossentropy": 1.2897258251905441,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20842219702899456,
+      "step": 1675
+    },
+    {
+      "epoch": 0.2793333333333333,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.840625,
+      "learning_rate": 8.197195009902924e-05,
+      "loss": 6.7164,
+      "loss/crossentropy": 1.6986341327428818,
+      "loss/hidden": 3.59765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21361364796757698,
+      "step": 1676
+    },
+    {
+      "epoch": 0.2795,
+      "grad_norm": 25.125,
+      "grad_norm_var": 2.1228515625,
+      "learning_rate": 8.195181748520811e-05,
+      "loss": 6.538,
+      "loss/crossentropy": 1.6305998861789703,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2566416822373867,
+      "step": 1677
+    },
+    {
+      "epoch": 0.2796666666666667,
+      "grad_norm": 24.375,
+      "grad_norm_var": 2.4572265625,
+      "learning_rate": 8.193167611161499e-05,
+      "loss": 5.9843,
+      "loss/crossentropy": 1.6780350655317307,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17056487873196602,
+      "step": 1678
+    },
+    {
+      "epoch": 0.2798333333333333,
+      "grad_norm": 27.375,
+      "grad_norm_var": 2.36640625,
+      "learning_rate": 8.191152598377178e-05,
+      "loss": 7.0966,
+      "loss/crossentropy": 1.5251629650592804,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20536588318645954,
+      "step": 1679
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 25.375,
+      "grad_norm_var": 2.4302083333333333,
+      "learning_rate": 8.189136710720272e-05,
+      "loss": 6.4033,
+      "loss/crossentropy": 1.704188033938408,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11363850720226765,
+      "step": 1680
+    },
+    {
+      "epoch": 0.2801666666666667,
+      "grad_norm": 25.5,
+      "grad_norm_var": 2.249739583333333,
+      "learning_rate": 8.18711994874345e-05,
+      "loss": 6.3669,
+      "loss/crossentropy": 1.5498753786087036,
+      "loss/hidden": 3.11328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13964795740321279,
+      "step": 1681
+    },
+    {
+      "epoch": 0.2803333333333333,
+      "grad_norm": 24.5,
+      "grad_norm_var": 2.3530598958333333,
+      "learning_rate": 8.185102312999617e-05,
+      "loss": 6.105,
+      "loss/crossentropy": 1.3350743502378464,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15579680539667606,
+      "step": 1682
+    },
+    {
+      "epoch": 0.2805,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.9364583333333334,
+      "learning_rate": 8.183083804041921e-05,
+      "loss": 6.9201,
+      "loss/crossentropy": 1.4282542169094086,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19400536641478539,
+      "step": 1683
+    },
+    {
+      "epoch": 0.2806666666666667,
+      "grad_norm": 25.375,
+      "grad_norm_var": 2.5395182291666667,
+      "learning_rate": 8.181064422423748e-05,
+      "loss": 6.295,
+      "loss/crossentropy": 1.6156925559043884,
+      "loss/hidden": 3.01953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12244082801043987,
+      "step": 1684
+    },
+    {
+      "epoch": 0.2808333333333333,
+      "grad_norm": 26.875,
+      "grad_norm_var": 2.3854166666666665,
+      "learning_rate": 8.179044168698721e-05,
+      "loss": 6.7027,
+      "loss/crossentropy": 2.2172724902629852,
+      "loss/hidden": 3.15234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16203631088137627,
+      "step": 1685
+    },
+    {
+      "epoch": 0.281,
+      "grad_norm": 23.875,
+      "grad_norm_var": 2.327018229166667,
+      "learning_rate": 8.177023043420705e-05,
+      "loss": 6.33,
+      "loss/crossentropy": 1.400356575846672,
+      "loss/hidden": 3.08984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11868741922080517,
+      "step": 1686
+    },
+    {
+      "epoch": 0.2811666666666667,
+      "grad_norm": 35.0,
+      "grad_norm_var": 7.287239583333333,
+      "learning_rate": 8.175001047143804e-05,
+      "loss": 6.6259,
+      "loss/crossentropy": 2.0223781168460846,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17344199120998383,
+      "step": 1687
+    },
+    {
+      "epoch": 0.2813333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 7.745247395833333,
+      "learning_rate": 8.172978180422358e-05,
+      "loss": 6.7435,
+      "loss/crossentropy": 1.1300409510731697,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15298896189779043,
+      "step": 1688
+    },
+    {
+      "epoch": 0.2815,
+      "grad_norm": 26.0,
+      "grad_norm_var": 7.516080729166666,
+      "learning_rate": 8.170954443810948e-05,
+      "loss": 6.4089,
+      "loss/crossentropy": 1.867250233888626,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18279875442385674,
+      "step": 1689
+    },
+    {
+      "epoch": 0.2816666666666667,
+      "grad_norm": 25.25,
+      "grad_norm_var": 7.566666666666666,
+      "learning_rate": 8.168929837864395e-05,
+      "loss": 6.4647,
+      "loss/crossentropy": 1.1209526397287846,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09244260424748063,
+      "step": 1690
+    },
+    {
+      "epoch": 0.2818333333333333,
+      "grad_norm": 28.25,
+      "grad_norm_var": 7.579166666666667,
+      "learning_rate": 8.16690436313775e-05,
+      "loss": 6.7581,
+      "loss/crossentropy": 1.2715392410755157,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13625416904687881,
+      "step": 1691
+    },
+    {
+      "epoch": 0.282,
+      "grad_norm": 24.625,
+      "grad_norm_var": 7.816666666666666,
+      "learning_rate": 8.164878020186317e-05,
+      "loss": 6.1189,
+      "loss/crossentropy": 1.3640960454940796,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11034796759486198,
+      "step": 1692
+    },
+    {
+      "epoch": 0.2821666666666667,
+      "grad_norm": 25.25,
+      "grad_norm_var": 7.792643229166667,
+      "learning_rate": 8.162850809565623e-05,
+      "loss": 6.5202,
+      "loss/crossentropy": 1.6748483180999756,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27111995220184326,
+      "step": 1693
+    },
+    {
+      "epoch": 0.2823333333333333,
+      "grad_norm": 26.125,
+      "grad_norm_var": 7.4572265625,
+      "learning_rate": 8.160822731831441e-05,
+      "loss": 6.4408,
+      "loss/crossentropy": 2.241428554058075,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1700704414397478,
+      "step": 1694
+    },
+    {
+      "epoch": 0.2825,
+      "grad_norm": 25.375,
+      "grad_norm_var": 7.5384765625,
+      "learning_rate": 8.158793787539782e-05,
+      "loss": 6.2016,
+      "loss/crossentropy": 1.798877015709877,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1579480841755867,
+      "step": 1695
+    },
+    {
+      "epoch": 0.2826666666666667,
+      "grad_norm": 25.375,
+      "grad_norm_var": 7.5384765625,
+      "learning_rate": 8.156763977246889e-05,
+      "loss": 6.2518,
+      "loss/crossentropy": 1.4084021598100662,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14755032770335674,
+      "step": 1696
+    },
+    {
+      "epoch": 0.2828333333333333,
+      "grad_norm": 26.5,
+      "grad_norm_var": 7.452018229166667,
+      "learning_rate": 8.154733301509248e-05,
+      "loss": 6.7187,
+      "loss/crossentropy": 1.2295970767736435,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15269313007593155,
+      "step": 1697
+    },
+    {
+      "epoch": 0.283,
+      "grad_norm": 28.375,
+      "grad_norm_var": 7.264322916666667,
+      "learning_rate": 8.152701760883581e-05,
+      "loss": 6.7106,
+      "loss/crossentropy": 1.2212089449167252,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10710981115698814,
+      "step": 1698
+    },
+    {
+      "epoch": 0.2831666666666667,
+      "grad_norm": 27.0,
+      "grad_norm_var": 6.882291666666666,
+      "learning_rate": 8.150669355926846e-05,
+      "loss": 6.5067,
+      "loss/crossentropy": 1.7339929342269897,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17183758690953255,
+      "step": 1699
+    },
+    {
+      "epoch": 0.2833333333333333,
+      "grad_norm": 28.625,
+      "grad_norm_var": 6.933072916666666,
+      "learning_rate": 8.148636087196237e-05,
+      "loss": 6.8658,
+      "loss/crossentropy": 2.0025227665901184,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1932712383568287,
+      "step": 1700
+    },
+    {
+      "epoch": 0.2835,
+      "grad_norm": 26.375,
+      "grad_norm_var": 6.955989583333333,
+      "learning_rate": 8.146601955249188e-05,
+      "loss": 6.5164,
+      "loss/crossentropy": 1.7747209072113037,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17721500620245934,
+      "step": 1701
+    },
+    {
+      "epoch": 0.2836666666666667,
+      "grad_norm": 26.0,
+      "grad_norm_var": 6.366080729166667,
+      "learning_rate": 8.144566960643367e-05,
+      "loss": 6.445,
+      "loss/crossentropy": 0.9415745139122009,
+      "loss/hidden": 3.0859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.07015350181609392,
+      "step": 1702
+    },
+    {
+      "epoch": 0.2838333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.4530598958333334,
+      "learning_rate": 8.142531103936678e-05,
+      "loss": 6.6023,
+      "loss/crossentropy": 1.7318656668066978,
+      "loss/hidden": 3.08984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1300593465566635,
+      "step": 1703
+    },
+    {
+      "epoch": 0.284,
+      "grad_norm": 22.875,
+      "grad_norm_var": 2.861458333333333,
+      "learning_rate": 8.140494385687265e-05,
+      "loss": 5.9511,
+      "loss/crossentropy": 1.8381174504756927,
+      "loss/hidden": 3.16015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.162352429702878,
+      "step": 1704
+    },
+    {
+      "epoch": 0.2841666666666667,
+      "grad_norm": 26.875,
+      "grad_norm_var": 2.8692057291666666,
+      "learning_rate": 8.138456806453503e-05,
+      "loss": 6.4049,
+      "loss/crossentropy": 1.34805366396904,
+      "loss/hidden": 3.12890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13075863756239414,
+      "step": 1705
+    },
+    {
+      "epoch": 0.2843333333333333,
+      "grad_norm": 24.25,
+      "grad_norm_var": 3.084830729166667,
+      "learning_rate": 8.136418366794008e-05,
+      "loss": 6.319,
+      "loss/crossentropy": 1.459059625864029,
+      "loss/hidden": 3.12109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11154993809759617,
+      "step": 1706
+    },
+    {
+      "epoch": 0.2845,
+      "grad_norm": 24.5,
+      "grad_norm_var": 3.006705729166667,
+      "learning_rate": 8.13437906726763e-05,
+      "loss": 6.4387,
+      "loss/crossentropy": 1.59040267765522,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1609763242304325,
+      "step": 1707
+    },
+    {
+      "epoch": 0.2846666666666667,
+      "grad_norm": 26.25,
+      "grad_norm_var": 2.851822916666667,
+      "learning_rate": 8.132338908433454e-05,
+      "loss": 6.3896,
+      "loss/crossentropy": 0.9835031479597092,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1490474995225668,
+      "step": 1708
+    },
+    {
+      "epoch": 0.2848333333333333,
+      "grad_norm": 28.125,
+      "grad_norm_var": 3.003059895833333,
+      "learning_rate": 8.130297890850802e-05,
+      "loss": 6.7328,
+      "loss/crossentropy": 1.6215149462223053,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19060637801885605,
+      "step": 1709
+    },
+    {
+      "epoch": 0.285,
+      "grad_norm": 26.875,
+      "grad_norm_var": 3.012434895833333,
+      "learning_rate": 8.128256015079229e-05,
+      "loss": 6.3302,
+      "loss/crossentropy": 1.2029203176498413,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1370858997106552,
+      "step": 1710
+    },
+    {
+      "epoch": 0.2851666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.968489583333333,
+      "learning_rate": 8.126213281678528e-05,
+      "loss": 6.8607,
+      "loss/crossentropy": 1.386208288371563,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21757697314023972,
+      "step": 1711
+    },
+    {
+      "epoch": 0.2853333333333333,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.968489583333333,
+      "learning_rate": 8.124169691208723e-05,
+      "loss": 6.4853,
+      "loss/crossentropy": 2.022136479616165,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23764223605394363,
+      "step": 1712
+    },
+    {
+      "epoch": 0.2855,
+      "grad_norm": 24.375,
+      "grad_norm_var": 3.3082682291666665,
+      "learning_rate": 8.122125244230079e-05,
+      "loss": 6.0407,
+      "loss/crossentropy": 1.2282831519842148,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16151340305805206,
+      "step": 1713
+    },
+    {
+      "epoch": 0.2856666666666667,
+      "grad_norm": 26.25,
+      "grad_norm_var": 3.0791666666666666,
+      "learning_rate": 8.120079941303094e-05,
+      "loss": 6.4671,
+      "loss/crossentropy": 1.2770699262619019,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20052243396639824,
+      "step": 1714
+    },
+    {
+      "epoch": 0.28583333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.811393229166667,
+      "learning_rate": 8.118033782988496e-05,
+      "loss": 6.6145,
+      "loss/crossentropy": 0.705689087510109,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.07864415924996138,
+      "step": 1715
+    },
+    {
+      "epoch": 0.286,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.047330729166666,
+      "learning_rate": 8.115986769847252e-05,
+      "loss": 6.6003,
+      "loss/crossentropy": 1.7134449481964111,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19462409429252148,
+      "step": 1716
+    },
+    {
+      "epoch": 0.2861666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 4.061458333333333,
+      "learning_rate": 8.113938902440564e-05,
+      "loss": 6.7364,
+      "loss/crossentropy": 1.8991410434246063,
+      "loss/hidden": 3.046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14321943186223507,
+      "step": 1717
+    },
+    {
+      "epoch": 0.28633333333333333,
+      "grad_norm": 26.375,
+      "grad_norm_var": 4.034309895833333,
+      "learning_rate": 8.111890181329863e-05,
+      "loss": 6.6163,
+      "loss/crossentropy": 1.3636438250541687,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16221468150615692,
+      "step": 1718
+    },
+    {
+      "epoch": 0.2865,
+      "grad_norm": 25.875,
+      "grad_norm_var": 3.52265625,
+      "learning_rate": 8.109840607076821e-05,
+      "loss": 6.3513,
+      "loss/crossentropy": 1.256602942943573,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20821137353777885,
+      "step": 1719
+    },
+    {
+      "epoch": 0.2866666666666667,
+      "grad_norm": 25.875,
+      "grad_norm_var": 2.62890625,
+      "learning_rate": 8.107790180243338e-05,
+      "loss": 6.6699,
+      "loss/crossentropy": 1.7374268770217896,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16650890931487083,
+      "step": 1720
+    },
+    {
+      "epoch": 0.28683333333333333,
+      "grad_norm": 25.25,
+      "grad_norm_var": 2.756705729166667,
+      "learning_rate": 8.105738901391552e-05,
+      "loss": 6.488,
+      "loss/crossentropy": 1.6735858917236328,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17187855020165443,
+      "step": 1721
+    },
+    {
+      "epoch": 0.287,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.3978515625,
+      "learning_rate": 8.103686771083831e-05,
+      "loss": 6.5816,
+      "loss/crossentropy": 1.6326270401477814,
+      "loss/hidden": 3.03515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13067574799060822,
+      "step": 1722
+    },
+    {
+      "epoch": 0.2871666666666667,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.0479166666666666,
+      "learning_rate": 8.101633789882781e-05,
+      "loss": 6.2152,
+      "loss/crossentropy": 1.4857535660266876,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15544962510466576,
+      "step": 1723
+    },
+    {
+      "epoch": 0.28733333333333333,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.064322916666667,
+      "learning_rate": 8.099579958351235e-05,
+      "loss": 6.4505,
+      "loss/crossentropy": 1.534998044371605,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14404334872961044,
+      "step": 1724
+    },
+    {
+      "epoch": 0.2875,
+      "grad_norm": 26.0,
+      "grad_norm_var": 2.0587890625,
+      "learning_rate": 8.097525277052264e-05,
+      "loss": 6.1291,
+      "loss/crossentropy": 1.4534046351909637,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1526658609509468,
+      "step": 1725
+    },
+    {
+      "epoch": 0.2876666666666667,
+      "grad_norm": 27.625,
+      "grad_norm_var": 2.0837890625,
+      "learning_rate": 8.095469746549172e-05,
+      "loss": 6.5274,
+      "loss/crossentropy": 1.9695439636707306,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18964740447700024,
+      "step": 1726
+    },
+    {
+      "epoch": 0.28783333333333333,
+      "grad_norm": 25.5,
+      "grad_norm_var": 2.2223307291666665,
+      "learning_rate": 8.093413367405489e-05,
+      "loss": 6.5136,
+      "loss/crossentropy": 1.164928138256073,
+      "loss/hidden": 3.1171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13709484552964568,
+      "step": 1727
+    },
+    {
+      "epoch": 0.288,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.1666666666666665,
+      "learning_rate": 8.091356140184991e-05,
+      "loss": 6.5013,
+      "loss/crossentropy": 0.9563533067703247,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1406487450003624,
+      "step": 1728
+    },
+    {
+      "epoch": 0.2881666666666667,
+      "grad_norm": 26.25,
+      "grad_norm_var": 1.7613932291666667,
+      "learning_rate": 8.089298065451672e-05,
+      "loss": 6.4944,
+      "loss/crossentropy": 1.8267393708229065,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24176055938005447,
+      "step": 1729
+    },
+    {
+      "epoch": 0.28833333333333333,
+      "grad_norm": 27.0,
+      "grad_norm_var": 1.7223307291666667,
+      "learning_rate": 8.087239143769768e-05,
+      "loss": 6.2706,
+      "loss/crossentropy": 1.5068450272083282,
+      "loss/hidden": 3.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17829317040741444,
+      "step": 1730
+    },
+    {
+      "epoch": 0.2885,
+      "grad_norm": 26.875,
+      "grad_norm_var": 1.1504557291666666,
+      "learning_rate": 8.085179375703744e-05,
+      "loss": 6.3447,
+      "loss/crossentropy": 0.6840021386742592,
+      "loss/hidden": 3.12890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09808410704135895,
+      "step": 1731
+    },
+    {
+      "epoch": 0.2886666666666667,
+      "grad_norm": 26.0,
+      "grad_norm_var": 0.7268229166666667,
+      "learning_rate": 8.083118761818295e-05,
+      "loss": 6.2977,
+      "loss/crossentropy": 1.4680070132017136,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2003943156450987,
+      "step": 1732
+    },
+    {
+      "epoch": 0.28883333333333333,
+      "grad_norm": 27.0,
+      "grad_norm_var": 0.7104166666666667,
+      "learning_rate": 8.081057302678352e-05,
+      "loss": 6.4703,
+      "loss/crossentropy": 1.6793077364563942,
+      "loss/hidden": 3.06640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12762292474508286,
+      "step": 1733
+    },
+    {
+      "epoch": 0.289,
+      "grad_norm": 27.0,
+      "grad_norm_var": 0.7139973958333333,
+      "learning_rate": 8.078994998849076e-05,
+      "loss": 6.0544,
+      "loss/crossentropy": 1.674924522638321,
+      "loss/hidden": 2.99609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12003491073846817,
+      "step": 1734
+    },
+    {
+      "epoch": 0.2891666666666667,
+      "grad_norm": 26.5,
+      "grad_norm_var": 0.67265625,
+      "learning_rate": 8.076931850895859e-05,
+      "loss": 6.761,
+      "loss/crossentropy": 1.4618712961673737,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13911530375480652,
+      "step": 1735
+    },
+    {
+      "epoch": 0.28933333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.1375,
+      "learning_rate": 8.074867859384322e-05,
+      "loss": 6.864,
+      "loss/crossentropy": 1.2650126516819,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12331366911530495,
+      "step": 1736
+    },
+    {
+      "epoch": 0.2895,
+      "grad_norm": 26.25,
+      "grad_norm_var": 0.975,
+      "learning_rate": 8.072803024880322e-05,
+      "loss": 6.4534,
+      "loss/crossentropy": 1.5010119080543518,
+      "loss/hidden": 3.03125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12273932807147503,
+      "step": 1737
+    },
+    {
+      "epoch": 0.2896666666666667,
+      "grad_norm": 24.75,
+      "grad_norm_var": 1.2643229166666667,
+      "learning_rate": 8.070737347949947e-05,
+      "loss": 6.5734,
+      "loss/crossentropy": 1.2331020385026932,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14210047386586666,
+      "step": 1738
+    },
+    {
+      "epoch": 0.28983333333333333,
+      "grad_norm": 27.5,
+      "grad_norm_var": 1.2520182291666666,
+      "learning_rate": 8.068670829159511e-05,
+      "loss": 6.5426,
+      "loss/crossentropy": 2.061487466096878,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18562847934663296,
+      "step": 1739
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 26.625,
+      "grad_norm_var": 1.15390625,
+      "learning_rate": 8.066603469075564e-05,
+      "loss": 6.9365,
+      "loss/crossentropy": 1.7315274775028229,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18126662634313107,
+      "step": 1740
+    },
+    {
+      "epoch": 0.2901666666666667,
+      "grad_norm": 27.5,
+      "grad_norm_var": 1.14765625,
+      "learning_rate": 8.064535268264883e-05,
+      "loss": 6.6233,
+      "loss/crossentropy": 1.6371970251202583,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23528031632304192,
+      "step": 1741
+    },
+    {
+      "epoch": 0.29033333333333333,
+      "grad_norm": 27.625,
+      "grad_norm_var": 1.14765625,
+      "learning_rate": 8.062466227294477e-05,
+      "loss": 6.7399,
+      "loss/crossentropy": 1.3356368094682693,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23707100003957748,
+      "step": 1742
+    },
+    {
+      "epoch": 0.2905,
+      "grad_norm": 27.25,
+      "grad_norm_var": 1.0291666666666666,
+      "learning_rate": 8.060396346731587e-05,
+      "loss": 6.8194,
+      "loss/crossentropy": 1.7589203417301178,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16239910200238228,
+      "step": 1743
+    },
+    {
+      "epoch": 0.2906666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.2619140625,
+      "learning_rate": 8.058325627143681e-05,
+      "loss": 6.4502,
+      "loss/crossentropy": 1.591295599937439,
+      "loss/hidden": 3.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19966844655573368,
+      "step": 1744
+    },
+    {
+      "epoch": 0.29083333333333333,
+      "grad_norm": 26.25,
+      "grad_norm_var": 1.2619140625,
+      "learning_rate": 8.056254069098459e-05,
+      "loss": 6.463,
+      "loss/crossentropy": 1.9710794389247894,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16523264348506927,
+      "step": 1745
+    },
+    {
+      "epoch": 0.291,
+      "grad_norm": 26.5,
+      "grad_norm_var": 1.2801432291666666,
+      "learning_rate": 8.05418167316385e-05,
+      "loss": 6.6769,
+      "loss/crossentropy": 1.5565124452114105,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19607799127697945,
+      "step": 1746
+    },
+    {
+      "epoch": 0.2911666666666667,
+      "grad_norm": 28.125,
+      "grad_norm_var": 1.3556640625,
+      "learning_rate": 8.052108439908013e-05,
+      "loss": 6.7603,
+      "loss/crossentropy": 1.780187577009201,
+      "loss/hidden": 3.11328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15015039592981339,
+      "step": 1747
+    },
+    {
+      "epoch": 0.29133333333333333,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.271875,
+      "learning_rate": 8.050034369899337e-05,
+      "loss": 6.5149,
+      "loss/crossentropy": 1.2236396819353104,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11750190891325474,
+      "step": 1748
+    },
+    {
+      "epoch": 0.2915,
+      "grad_norm": 27.5,
+      "grad_norm_var": 1.2770833333333333,
+      "learning_rate": 8.04795946370644e-05,
+      "loss": 6.6239,
+      "loss/crossentropy": 1.2788297533988953,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13053389266133308,
+      "step": 1749
+    },
+    {
+      "epoch": 0.2916666666666667,
+      "grad_norm": 26.625,
+      "grad_norm_var": 1.2952473958333333,
+      "learning_rate": 8.04588372189817e-05,
+      "loss": 5.9622,
+      "loss/crossentropy": 1.1857105791568756,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16278855875134468,
+      "step": 1750
+    },
+    {
+      "epoch": 0.29183333333333333,
+      "grad_norm": 27.75,
+      "grad_norm_var": 1.2822265625,
+      "learning_rate": 8.043807145043604e-05,
+      "loss": 6.3428,
+      "loss/crossentropy": 1.0286138355731964,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.162134924903512,
+      "step": 1751
+    },
+    {
+      "epoch": 0.292,
+      "grad_norm": 24.625,
+      "grad_norm_var": 1.2561848958333333,
+      "learning_rate": 8.041729733712045e-05,
+      "loss": 6.3891,
+      "loss/crossentropy": 1.474233716726303,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18093383312225342,
+      "step": 1752
+    },
+    {
+      "epoch": 0.2921666666666667,
+      "grad_norm": 26.125,
+      "grad_norm_var": 1.2684895833333334,
+      "learning_rate": 8.039651488473028e-05,
+      "loss": 6.6086,
+      "loss/crossentropy": 1.9393830597400665,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16801822558045387,
+      "step": 1753
+    },
+    {
+      "epoch": 0.29233333333333333,
+      "grad_norm": 24.75,
+      "grad_norm_var": 1.2684895833333334,
+      "learning_rate": 8.037572409896315e-05,
+      "loss": 6.0165,
+      "loss/crossentropy": 1.1125195175409317,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14247208461165428,
+      "step": 1754
+    },
+    {
+      "epoch": 0.2925,
+      "grad_norm": 24.0,
+      "grad_norm_var": 1.7643229166666667,
+      "learning_rate": 8.0354924985519e-05,
+      "loss": 6.1801,
+      "loss/crossentropy": 1.8784036189317703,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13121195137500763,
+      "step": 1755
+    },
+    {
+      "epoch": 0.2926666666666667,
+      "grad_norm": 25.375,
+      "grad_norm_var": 1.875,
+      "learning_rate": 8.033411755009999e-05,
+      "loss": 6.0551,
+      "loss/crossentropy": 1.4668779373168945,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12251794151961803,
+      "step": 1756
+    },
+    {
+      "epoch": 0.29283333333333333,
+      "grad_norm": 26.75,
+      "grad_norm_var": 1.82265625,
+      "learning_rate": 8.031330179841062e-05,
+      "loss": 6.2527,
+      "loss/crossentropy": 0.960818275809288,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09816616401076317,
+      "step": 1757
+    },
+    {
+      "epoch": 0.293,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.351822916666667,
+      "learning_rate": 8.029247773615764e-05,
+      "loss": 6.7504,
+      "loss/crossentropy": 1.881619244813919,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1748347170650959,
+      "step": 1758
+    },
+    {
+      "epoch": 0.2931666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.601822916666667,
+      "learning_rate": 8.027164536905008e-05,
+      "loss": 6.5377,
+      "loss/crossentropy": 1.3765027970075607,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13398028910160065,
+      "step": 1759
+    },
+    {
+      "epoch": 0.29333333333333333,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.3166015625,
+      "learning_rate": 8.025080470279924e-05,
+      "loss": 6.656,
+      "loss/crossentropy": 2.0068330615758896,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21021978929638863,
+      "step": 1760
+    },
+    {
+      "epoch": 0.2935,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.3400390625,
+      "learning_rate": 8.022995574311876e-05,
+      "loss": 6.8367,
+      "loss/crossentropy": 1.4760197699069977,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23727067932486534,
+      "step": 1761
+    },
+    {
+      "epoch": 0.2936666666666667,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.366080729166667,
+      "learning_rate": 8.020909849572444e-05,
+      "loss": 6.2134,
+      "loss/crossentropy": 2.564862221479416,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1808963194489479,
+      "step": 1762
+    },
+    {
+      "epoch": 0.29383333333333334,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.467643229166667,
+      "learning_rate": 8.018823296633441e-05,
+      "loss": 6.9278,
+      "loss/crossentropy": 1.4544070810079575,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2306374479085207,
+      "step": 1763
+    },
+    {
+      "epoch": 0.294,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.751822916666667,
+      "learning_rate": 8.016735916066913e-05,
+      "loss": 6.6822,
+      "loss/crossentropy": 1.4069459438323975,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18433349765837193,
+      "step": 1764
+    },
+    {
+      "epoch": 0.2941666666666667,
+      "grad_norm": 22.75,
+      "grad_norm_var": 3.8354166666666667,
+      "learning_rate": 8.014647708445124e-05,
+      "loss": 5.9026,
+      "loss/crossentropy": 1.4110225439071655,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14669842272996902,
+      "step": 1765
+    },
+    {
+      "epoch": 0.29433333333333334,
+      "grad_norm": 27.375,
+      "grad_norm_var": 3.8643229166666666,
+      "learning_rate": 8.012558674340566e-05,
+      "loss": 6.2119,
+      "loss/crossentropy": 1.3153589367866516,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14798353984951973,
+      "step": 1766
+    },
+    {
+      "epoch": 0.2945,
+      "grad_norm": 27.625,
+      "grad_norm_var": 3.848372395833333,
+      "learning_rate": 8.010468814325964e-05,
+      "loss": 6.7078,
+      "loss/crossentropy": 1.047652706503868,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14315370097756386,
+      "step": 1767
+    },
+    {
+      "epoch": 0.2946666666666667,
+      "grad_norm": 27.875,
+      "grad_norm_var": 3.5978515625,
+      "learning_rate": 8.008378128974262e-05,
+      "loss": 6.7981,
+      "loss/crossentropy": 1.613780677318573,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25304109044373035,
+      "step": 1768
+    },
+    {
+      "epoch": 0.29483333333333334,
+      "grad_norm": 24.375,
+      "grad_norm_var": 3.9770182291666667,
+      "learning_rate": 8.006286618858635e-05,
+      "loss": 6.1878,
+      "loss/crossentropy": 1.6416444182395935,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13193977810442448,
+      "step": 1769
+    },
+    {
+      "epoch": 0.295,
+      "grad_norm": 23.75,
+      "grad_norm_var": 4.315559895833333,
+      "learning_rate": 8.004194284552477e-05,
+      "loss": 6.2519,
+      "loss/crossentropy": 1.581135168671608,
+      "loss/hidden": 2.94140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11294459365308285,
+      "step": 1770
+    },
+    {
+      "epoch": 0.2951666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.7056640625,
+      "learning_rate": 8.002101126629421e-05,
+      "loss": 7.1182,
+      "loss/crossentropy": 2.08039990067482,
+      "loss/hidden": 3.984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.257803812623024,
+      "step": 1771
+    },
+    {
+      "epoch": 0.29533333333333334,
+      "grad_norm": 26.5,
+      "grad_norm_var": 5.49765625,
+      "learning_rate": 8.000007145663312e-05,
+      "loss": 6.5124,
+      "loss/crossentropy": 1.7909472584724426,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21027473732829094,
+      "step": 1772
+    },
+    {
+      "epoch": 0.2955,
+      "grad_norm": 27.0,
+      "grad_norm_var": 5.48125,
+      "learning_rate": 7.997912342228232e-05,
+      "loss": 6.5355,
+      "loss/crossentropy": 1.297078013420105,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1581488400697708,
+      "step": 1773
+    },
+    {
+      "epoch": 0.2956666666666667,
+      "grad_norm": 27.75,
+      "grad_norm_var": 5.1384765625,
+      "learning_rate": 7.99581671689848e-05,
+      "loss": 6.2334,
+      "loss/crossentropy": 1.3412351608276367,
+      "loss/hidden": 3.61328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2025449462234974,
+      "step": 1774
+    },
+    {
+      "epoch": 0.29583333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 6.2978515625,
+      "learning_rate": 7.993720270248584e-05,
+      "loss": 6.4776,
+      "loss/crossentropy": 1.4673235714435577,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1532964836806059,
+      "step": 1775
+    },
+    {
+      "epoch": 0.296,
+      "grad_norm": 29.0,
+      "grad_norm_var": 6.443684895833333,
+      "learning_rate": 7.991623002853296e-05,
+      "loss": 6.469,
+      "loss/crossentropy": 1.1405130997300148,
+      "loss/hidden": 3.03515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1286987140774727,
+      "step": 1776
+    },
+    {
+      "epoch": 0.2961666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 6.573372395833333,
+      "learning_rate": 7.989524915287595e-05,
+      "loss": 7.0811,
+      "loss/crossentropy": 1.386097103357315,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16061494871973991,
+      "step": 1777
+    },
+    {
+      "epoch": 0.29633333333333334,
+      "grad_norm": 26.125,
+      "grad_norm_var": 6.71875,
+      "learning_rate": 7.987426008126683e-05,
+      "loss": 6.0542,
+      "loss/crossentropy": 1.760521948337555,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18958137556910515,
+      "step": 1778
+    },
+    {
+      "epoch": 0.2965,
+      "grad_norm": 27.875,
+      "grad_norm_var": 6.64765625,
+      "learning_rate": 7.985326281945989e-05,
+      "loss": 6.9317,
+      "loss/crossentropy": 1.645604282617569,
+      "loss/hidden": 3.80078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32877641171216965,
+      "step": 1779
+    },
+    {
+      "epoch": 0.2966666666666667,
+      "grad_norm": 24.375,
+      "grad_norm_var": 7.069205729166667,
+      "learning_rate": 7.983225737321163e-05,
+      "loss": 6.0512,
+      "loss/crossentropy": 1.0108902752399445,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14840376935899258,
+      "step": 1780
+    },
+    {
+      "epoch": 0.29683333333333334,
+      "grad_norm": 26.625,
+      "grad_norm_var": 5.694791666666666,
+      "learning_rate": 7.98112437482808e-05,
+      "loss": 6.6746,
+      "loss/crossentropy": 1.5815895348787308,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19548891857266426,
+      "step": 1781
+    },
+    {
+      "epoch": 0.297,
+      "grad_norm": 28.625,
+      "grad_norm_var": 5.776822916666666,
+      "learning_rate": 7.979022195042842e-05,
+      "loss": 6.7206,
+      "loss/crossentropy": 1.7035427242517471,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1873407606035471,
+      "step": 1782
+    },
+    {
+      "epoch": 0.2971666666666667,
+      "grad_norm": 28.625,
+      "grad_norm_var": 5.849739583333333,
+      "learning_rate": 7.976919198541776e-05,
+      "loss": 6.5586,
+      "loss/crossentropy": 1.5671730041503906,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16458377614617348,
+      "step": 1783
+    },
+    {
+      "epoch": 0.29733333333333334,
+      "grad_norm": 28.125,
+      "grad_norm_var": 5.8625,
+      "learning_rate": 7.974815385901426e-05,
+      "loss": 6.5158,
+      "loss/crossentropy": 1.3500420153141022,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13635024055838585,
+      "step": 1784
+    },
+    {
+      "epoch": 0.2975,
+      "grad_norm": 24.125,
+      "grad_norm_var": 5.974739583333333,
+      "learning_rate": 7.972710757698567e-05,
+      "loss": 6.0873,
+      "loss/crossentropy": 1.2957044690847397,
+      "loss/hidden": 3.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09067535400390625,
+      "step": 1785
+    },
+    {
+      "epoch": 0.2976666666666667,
+      "grad_norm": 24.5,
+      "grad_norm_var": 5.623958333333333,
+      "learning_rate": 7.970605314510194e-05,
+      "loss": 6.2344,
+      "loss/crossentropy": 1.358672022819519,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.239708811044693,
+      "step": 1786
+    },
+    {
+      "epoch": 0.29783333333333334,
+      "grad_norm": 22.75,
+      "grad_norm_var": 5.268489583333333,
+      "learning_rate": 7.968499056913524e-05,
+      "loss": 6.0454,
+      "loss/crossentropy": 1.1282762587070465,
+      "loss/hidden": 2.98828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09060000814497471,
+      "step": 1787
+    },
+    {
+      "epoch": 0.298,
+      "grad_norm": 25.125,
+      "grad_norm_var": 5.4869140625,
+      "learning_rate": 7.966391985486003e-05,
+      "loss": 6.6858,
+      "loss/crossentropy": 1.781815230846405,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18372655287384987,
+      "step": 1788
+    },
+    {
+      "epoch": 0.2981666666666667,
+      "grad_norm": 26.0,
+      "grad_norm_var": 5.544205729166666,
+      "learning_rate": 7.964284100805297e-05,
+      "loss": 6.735,
+      "loss/crossentropy": 1.3905943930149078,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23225440457463264,
+      "step": 1789
+    },
+    {
+      "epoch": 0.29833333333333334,
+      "grad_norm": 24.625,
+      "grad_norm_var": 5.799739583333333,
+      "learning_rate": 7.96217540344929e-05,
+      "loss": 6.2843,
+      "loss/crossentropy": 1.9590551406145096,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18367251940071583,
+      "step": 1790
+    },
+    {
+      "epoch": 0.2985,
+      "grad_norm": 27.875,
+      "grad_norm_var": 4.1306640625,
+      "learning_rate": 7.960065893996098e-05,
+      "loss": 6.3292,
+      "loss/crossentropy": 1.605180263519287,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18637590110301971,
+      "step": 1791
+    },
+    {
+      "epoch": 0.2986666666666667,
+      "grad_norm": 27.125,
+      "grad_norm_var": 3.715625,
+      "learning_rate": 7.957955573024052e-05,
+      "loss": 6.2751,
+      "loss/crossentropy": 1.7973419725894928,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14722765982151031,
+      "step": 1792
+    },
+    {
+      "epoch": 0.29883333333333334,
+      "grad_norm": 26.0,
+      "grad_norm_var": 3.215625,
+      "learning_rate": 7.95584444111171e-05,
+      "loss": 6.6446,
+      "loss/crossentropy": 1.5896763503551483,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1403478067368269,
+      "step": 1793
+    },
+    {
+      "epoch": 0.299,
+      "grad_norm": 27.5,
+      "grad_norm_var": 3.3280598958333334,
+      "learning_rate": 7.95373249883785e-05,
+      "loss": 6.5739,
+      "loss/crossentropy": 1.9001854956150055,
+      "loss/hidden": 3.640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22470081970095634,
+      "step": 1794
+    },
+    {
+      "epoch": 0.2991666666666667,
+      "grad_norm": 6408896512.0,
+      "grad_norm_var": 2.5671221354096317e+18,
+      "learning_rate": 7.951619746781474e-05,
+      "loss": 7.1923,
+      "loss/crossentropy": 2.1498090028762817,
+      "loss/hidden": 6.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32943011447787285,
+      "step": 1795
+    },
+    {
+      "epoch": 0.29933333333333334,
+      "grad_norm": 34.25,
+      "grad_norm_var": 2.567122134882233e+18,
+      "learning_rate": 7.949506185521802e-05,
+      "loss": 6.7547,
+      "loss/crossentropy": 1.0805221498012543,
+      "loss/hidden": 3.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1237840848043561,
+      "step": 1796
+    },
+    {
+      "epoch": 0.2995,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.5671221346485755e+18,
+      "learning_rate": 7.947391815638284e-05,
+      "loss": 6.6611,
+      "loss/crossentropy": 2.0383740961551666,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17835279554128647,
+      "step": 1797
+    },
+    {
+      "epoch": 0.2996666666666667,
+      "grad_norm": 26.25,
+      "grad_norm_var": 2.567122134775418e+18,
+      "learning_rate": 7.945276637710582e-05,
+      "loss": 6.4153,
+      "loss/crossentropy": 1.6993253529071808,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3014073744416237,
+      "step": 1798
+    },
+    {
+      "epoch": 0.29983333333333334,
+      "grad_norm": 25.75,
+      "grad_norm_var": 2.5671221349289646e+18,
+      "learning_rate": 7.943160652318585e-05,
+      "loss": 6.3668,
+      "loss/crossentropy": 1.5742650628089905,
+      "loss/hidden": 3.2421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16767333820462227,
+      "step": 1799
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.567122134975696e+18,
+      "learning_rate": 7.941043860042403e-05,
+      "loss": 6.331,
+      "loss/crossentropy": 1.8069534599781036,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24792053177952766,
+      "step": 1800
+    },
+    {
+      "epoch": 0.3001666666666667,
+      "grad_norm": 23.625,
+      "grad_norm_var": 2.5671221350024e+18,
+      "learning_rate": 7.938926261462366e-05,
+      "loss": 6.2342,
+      "loss/crossentropy": 1.0832974091172218,
+      "loss/hidden": 3.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10927399434149265,
+      "step": 1801
+    },
+    {
+      "epoch": 0.30033333333333334,
+      "grad_norm": 26.375,
+      "grad_norm_var": 2.5671221349022607e+18,
+      "learning_rate": 7.936807857159026e-05,
+      "loss": 6.7035,
+      "loss/crossentropy": 1.6718868017196655,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16467371210455894,
+      "step": 1802
+    },
+    {
+      "epoch": 0.3005,
+      "grad_norm": 27.5,
+      "grad_norm_var": 2.5671221346485755e+18,
+      "learning_rate": 7.934688647713158e-05,
+      "loss": 6.8929,
+      "loss/crossentropy": 2.2081165313720703,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17353184521198273,
+      "step": 1803
+    },
+    {
+      "epoch": 0.3006666666666667,
+      "grad_norm": 24.75,
+      "grad_norm_var": 2.567122134668603e+18,
+      "learning_rate": 7.932568633705752e-05,
+      "loss": 6.3388,
+      "loss/crossentropy": 0.8409457728266716,
+      "loss/hidden": 2.9609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.08156785555183887,
+      "step": 1804
+    },
+    {
+      "epoch": 0.30083333333333334,
+      "grad_norm": 22.375,
+      "grad_norm_var": 2.5671221348622054e+18,
+      "learning_rate": 7.930447815718022e-05,
+      "loss": 6.1252,
+      "loss/crossentropy": 1.269652098417282,
+      "loss/hidden": 3.03125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12972666136920452,
+      "step": 1805
+    },
+    {
+      "epoch": 0.301,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.567122134681955e+18,
+      "learning_rate": 7.928326194331404e-05,
+      "loss": 6.2914,
+      "loss/crossentropy": 0.919536791741848,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1664884351193905,
+      "step": 1806
+    },
+    {
+      "epoch": 0.3011666666666667,
+      "grad_norm": 24.5,
+      "grad_norm_var": 2.5671221348622054e+18,
+      "learning_rate": 7.926203770127552e-05,
+      "loss": 6.57,
+      "loss/crossentropy": 1.2009000331163406,
+      "loss/hidden": 3.00390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12875444488599896,
+      "step": 1807
+    },
+    {
+      "epoch": 0.30133333333333334,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.567122134782094e+18,
+      "learning_rate": 7.924080543688337e-05,
+      "loss": 6.5496,
+      "loss/crossentropy": 1.196403868496418,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14112555235624313,
+      "step": 1808
+    },
+    {
+      "epoch": 0.3015,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.567122134681955e+18,
+      "learning_rate": 7.921956515595861e-05,
+      "loss": 6.6477,
+      "loss/crossentropy": 1.5183096826076508,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25990092381834984,
+      "step": 1809
+    },
+    {
+      "epoch": 0.3016666666666667,
+      "grad_norm": 26.875,
+      "grad_norm_var": 2.5671221347153347e+18,
+      "learning_rate": 7.919831686432433e-05,
+      "loss": 6.8476,
+      "loss/crossentropy": 1.5886277109384537,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18553652241826057,
+      "step": 1810
+    },
+    {
+      "epoch": 0.30183333333333334,
+      "grad_norm": 24.75,
+      "grad_norm_var": 8.31015625,
+      "learning_rate": 7.917706056780587e-05,
+      "loss": 6.238,
+      "loss/crossentropy": 1.0148934796452522,
+      "loss/hidden": 3.0703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09379838779568672,
+      "step": 1811
+    },
+    {
+      "epoch": 0.302,
+      "grad_norm": 24.75,
+      "grad_norm_var": 4.589322916666666,
+      "learning_rate": 7.915579627223079e-05,
+      "loss": 6.6423,
+      "loss/crossentropy": 1.0143003314733505,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10569844581186771,
+      "step": 1812
+    },
+    {
+      "epoch": 0.30216666666666664,
+      "grad_norm": 24.25,
+      "grad_norm_var": 3.176041666666667,
+      "learning_rate": 7.913452398342881e-05,
+      "loss": 6.4593,
+      "loss/crossentropy": 1.60957869887352,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17155874334275723,
+      "step": 1813
+    },
+    {
+      "epoch": 0.30233333333333334,
+      "grad_norm": 25.625,
+      "grad_norm_var": 3.1666015625,
+      "learning_rate": 7.911324370723183e-05,
+      "loss": 6.1152,
+      "loss/crossentropy": 1.5737018585205078,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17441192269325256,
+      "step": 1814
+    },
+    {
+      "epoch": 0.3025,
+      "grad_norm": 27.875,
+      "grad_norm_var": 3.433333333333333,
+      "learning_rate": 7.909195544947398e-05,
+      "loss": 6.6598,
+      "loss/crossentropy": 1.7213964462280273,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2511589825153351,
+      "step": 1815
+    },
+    {
+      "epoch": 0.30266666666666664,
+      "grad_norm": 26.25,
+      "grad_norm_var": 3.3208333333333333,
+      "learning_rate": 7.907065921599154e-05,
+      "loss": 6.5811,
+      "loss/crossentropy": 1.8362173736095428,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19294562935829163,
+      "step": 1816
+    },
+    {
+      "epoch": 0.30283333333333334,
+      "grad_norm": 27.625,
+      "grad_norm_var": 3.120833333333333,
+      "learning_rate": 7.9049355012623e-05,
+      "loss": 6.435,
+      "loss/crossentropy": 1.4664649963378906,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16518542543053627,
+      "step": 1817
+    },
+    {
+      "epoch": 0.303,
+      "grad_norm": 24.125,
+      "grad_norm_var": 3.3622395833333334,
+      "learning_rate": 7.902804284520903e-05,
+      "loss": 6.2792,
+      "loss/crossentropy": 2.1789761781692505,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1664630975574255,
+      "step": 1818
+    },
+    {
+      "epoch": 0.30316666666666664,
+      "grad_norm": 28.125,
+      "grad_norm_var": 3.512955729166667,
+      "learning_rate": 7.900672271959247e-05,
+      "loss": 6.821,
+      "loss/crossentropy": 2.0341483652591705,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2923608161509037,
+      "step": 1819
+    },
+    {
+      "epoch": 0.30333333333333334,
+      "grad_norm": 27.0,
+      "grad_norm_var": 3.4473307291666666,
+      "learning_rate": 7.898539464161838e-05,
+      "loss": 6.6761,
+      "loss/crossentropy": 1.9393551647663116,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25565091148018837,
+      "step": 1820
+    },
+    {
+      "epoch": 0.3035,
+      "grad_norm": 26.875,
+      "grad_norm_var": 2.4395182291666666,
+      "learning_rate": 7.896405861713394e-05,
+      "loss": 6.3848,
+      "loss/crossentropy": 1.046747848391533,
+      "loss/hidden": 2.9921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12378543987870216,
+      "step": 1821
+    },
+    {
+      "epoch": 0.30366666666666664,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.3059895833333335,
+      "learning_rate": 7.894271465198857e-05,
+      "loss": 6.6549,
+      "loss/crossentropy": 1.3154726698994637,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12605149671435356,
+      "step": 1822
+    },
+    {
+      "epoch": 0.30383333333333334,
+      "grad_norm": 25.125,
+      "grad_norm_var": 2.1728515625,
+      "learning_rate": 7.892136275203383e-05,
+      "loss": 6.224,
+      "loss/crossentropy": 1.3829431384801865,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11828980594873428,
+      "step": 1823
+    },
+    {
+      "epoch": 0.304,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.0143229166666665,
+      "learning_rate": 7.890000292312346e-05,
+      "loss": 6.3587,
+      "loss/crossentropy": 0.8525128811597824,
+      "loss/hidden": 3.05078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.08960277959704399,
+      "step": 1824
+    },
+    {
+      "epoch": 0.30416666666666664,
+      "grad_norm": 26.25,
+      "grad_norm_var": 1.8577473958333333,
+      "learning_rate": 7.887863517111338e-05,
+      "loss": 6.5897,
+      "loss/crossentropy": 1.4181907325983047,
+      "loss/hidden": 3.52734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17713713832199574,
+      "step": 1825
+    },
+    {
+      "epoch": 0.30433333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.4436848958333335,
+      "learning_rate": 7.88572595018617e-05,
+      "loss": 6.8242,
+      "loss/crossentropy": 1.2928383871912956,
+      "loss/hidden": 3.06640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12669945321977139,
+      "step": 1826
+    },
+    {
+      "epoch": 0.3045,
+      "grad_norm": 25.875,
+      "grad_norm_var": 2.2684895833333334,
+      "learning_rate": 7.883587592122863e-05,
+      "loss": 6.2948,
+      "loss/crossentropy": 1.237380012869835,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1380482241511345,
+      "step": 1827
+    },
+    {
+      "epoch": 0.30466666666666664,
+      "grad_norm": 25.625,
+      "grad_norm_var": 2.1103515625,
+      "learning_rate": 7.881448443507664e-05,
+      "loss": 6.3056,
+      "loss/crossentropy": 1.4298370033502579,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19418893568217754,
+      "step": 1828
+    },
+    {
+      "epoch": 0.30483333333333335,
+      "grad_norm": 25.75,
+      "grad_norm_var": 1.7869140625,
+      "learning_rate": 7.879308504927035e-05,
+      "loss": 6.4656,
+      "loss/crossentropy": 2.142209440469742,
+      "loss/hidden": 3.68359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29572315886616707,
+      "step": 1829
+    },
+    {
+      "epoch": 0.305,
+      "grad_norm": 26.125,
+      "grad_norm_var": 1.7332682291666666,
+      "learning_rate": 7.877167776967645e-05,
+      "loss": 6.7122,
+      "loss/crossentropy": 1.8007521629333496,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20245215110480785,
+      "step": 1830
+    },
+    {
+      "epoch": 0.30516666666666664,
+      "grad_norm": 27.0,
+      "grad_norm_var": 1.6434895833333334,
+      "learning_rate": 7.875026260216393e-05,
+      "loss": 6.3127,
+      "loss/crossentropy": 1.238753080368042,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15646503865718842,
+      "step": 1831
+    },
+    {
+      "epoch": 0.30533333333333335,
+      "grad_norm": 25.375,
+      "grad_norm_var": 1.7369140625,
+      "learning_rate": 7.872883955260387e-05,
+      "loss": 6.3482,
+      "loss/crossentropy": 2.4605717062950134,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17422811686992645,
+      "step": 1832
+    },
+    {
+      "epoch": 0.3055,
+      "grad_norm": 27.0,
+      "grad_norm_var": 1.6747395833333334,
+      "learning_rate": 7.87074086268695e-05,
+      "loss": 6.7203,
+      "loss/crossentropy": 1.5636233538389206,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1645506266504526,
+      "step": 1833
+    },
+    {
+      "epoch": 0.30566666666666664,
+      "grad_norm": 25.25,
+      "grad_norm_var": 1.3905598958333334,
+      "learning_rate": 7.868596983083623e-05,
+      "loss": 6.2717,
+      "loss/crossentropy": 1.3363309055566788,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16017024964094162,
+      "step": 1834
+    },
+    {
+      "epoch": 0.30583333333333335,
+      "grad_norm": 25.75,
+      "grad_norm_var": 1.265625,
+      "learning_rate": 7.866452317038164e-05,
+      "loss": 6.4717,
+      "loss/crossentropy": 1.3597393333911896,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24192249029874802,
+      "step": 1835
+    },
+    {
+      "epoch": 0.306,
+      "grad_norm": 26.0,
+      "grad_norm_var": 1.2572916666666667,
+      "learning_rate": 7.864306865138545e-05,
+      "loss": 6.3713,
+      "loss/crossentropy": 2.0953482687473297,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17437477968633175,
+      "step": 1836
+    },
+    {
+      "epoch": 0.30616666666666664,
+      "grad_norm": 23.5,
+      "grad_norm_var": 1.7582682291666667,
+      "learning_rate": 7.862160627972955e-05,
+      "loss": 6.0975,
+      "loss/crossentropy": 1.5605383813381195,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1994226761162281,
+      "step": 1837
+    },
+    {
+      "epoch": 0.30633333333333335,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.046875,
+      "learning_rate": 7.860013606129796e-05,
+      "loss": 6.5668,
+      "loss/crossentropy": 1.7478504478931427,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14831626787781715,
+      "step": 1838
+    },
+    {
+      "epoch": 0.3065,
+      "grad_norm": 25.0,
+      "grad_norm_var": 2.067122395833333,
+      "learning_rate": 7.857865800197684e-05,
+      "loss": 6.4458,
+      "loss/crossentropy": 1.0507306307554245,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13380114827305079,
+      "step": 1839
+    },
+    {
+      "epoch": 0.30666666666666664,
+      "grad_norm": 25.5,
+      "grad_norm_var": 1.8822265625,
+      "learning_rate": 7.855717210765456e-05,
+      "loss": 6.455,
+      "loss/crossentropy": 1.7240546941757202,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20109525695443153,
+      "step": 1840
+    },
+    {
+      "epoch": 0.30683333333333335,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.0494140625,
+      "learning_rate": 7.85356783842216e-05,
+      "loss": 6.6225,
+      "loss/crossentropy": 1.9693889617919922,
+      "loss/hidden": 3.5546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21837937459349632,
+      "step": 1841
+    },
+    {
+      "epoch": 0.307,
+      "grad_norm": 25.625,
+      "grad_norm_var": 1.3462890625,
+      "learning_rate": 7.851417683757053e-05,
+      "loss": 6.4929,
+      "loss/crossentropy": 1.919701337814331,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15016796812415123,
+      "step": 1842
+    },
+    {
+      "epoch": 0.30716666666666664,
+      "grad_norm": 26.5,
+      "grad_norm_var": 1.3622395833333334,
+      "learning_rate": 7.849266747359619e-05,
+      "loss": 6.8308,
+      "loss/crossentropy": 1.6118244975805283,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20529415644705296,
+      "step": 1843
+    },
+    {
+      "epoch": 0.30733333333333335,
+      "grad_norm": 25.0,
+      "grad_norm_var": 1.4192057291666667,
+      "learning_rate": 7.847115029819547e-05,
+      "loss": 6.4478,
+      "loss/crossentropy": 1.3020363003015518,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12500270456075668,
+      "step": 1844
+    },
+    {
+      "epoch": 0.3075,
+      "grad_norm": 26.5,
+      "grad_norm_var": 1.4317057291666666,
+      "learning_rate": 7.84496253172674e-05,
+      "loss": 6.7033,
+      "loss/crossentropy": 1.2071290761232376,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16696172766387463,
+      "step": 1845
+    },
+    {
+      "epoch": 0.30766666666666664,
+      "grad_norm": 25.25,
+      "grad_norm_var": 1.4677083333333334,
+      "learning_rate": 7.84280925367132e-05,
+      "loss": 6.5151,
+      "loss/crossentropy": 1.4631052911281586,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18882737681269646,
+      "step": 1846
+    },
+    {
+      "epoch": 0.30783333333333335,
+      "grad_norm": 28.0,
+      "grad_norm_var": 1.6677083333333333,
+      "learning_rate": 7.84065519624362e-05,
+      "loss": 7.026,
+      "loss/crossentropy": 2.259757697582245,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.192843746393919,
+      "step": 1847
+    },
+    {
+      "epoch": 0.308,
+      "grad_norm": 26.625,
+      "grad_norm_var": 1.6559895833333333,
+      "learning_rate": 7.838500360034188e-05,
+      "loss": 6.1788,
+      "loss/crossentropy": 1.2726633250713348,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1777363307774067,
+      "step": 1848
+    },
+    {
+      "epoch": 0.30816666666666664,
+      "grad_norm": 23.625,
+      "grad_norm_var": 1.9671223958333333,
+      "learning_rate": 7.836344745633783e-05,
+      "loss": 6.3957,
+      "loss/crossentropy": 1.7389848679304123,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16802443750202656,
+      "step": 1849
+    },
+    {
+      "epoch": 0.30833333333333335,
+      "grad_norm": 26.0,
+      "grad_norm_var": 1.9374348958333334,
+      "learning_rate": 7.83418835363338e-05,
+      "loss": 6.6081,
+      "loss/crossentropy": 1.580287128686905,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17671324871480465,
+      "step": 1850
+    },
+    {
+      "epoch": 0.3085,
+      "grad_norm": 24.25,
+      "grad_norm_var": 2.1171223958333334,
+      "learning_rate": 7.832031184624164e-05,
+      "loss": 6.0687,
+      "loss/crossentropy": 1.1475929617881775,
+      "loss/hidden": 3.16015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.126324862241745,
+      "step": 1851
+    },
+    {
+      "epoch": 0.30866666666666664,
+      "grad_norm": 26.625,
+      "grad_norm_var": 2.15390625,
+      "learning_rate": 7.829873239197538e-05,
+      "loss": 6.6017,
+      "loss/crossentropy": 2.1429725289344788,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23913269490003586,
+      "step": 1852
+    },
+    {
+      "epoch": 0.30883333333333335,
+      "grad_norm": 26.0,
+      "grad_norm_var": 1.74765625,
+      "learning_rate": 7.827714517945115e-05,
+      "loss": 6.771,
+      "loss/crossentropy": 1.3311758264899254,
+      "loss/hidden": 3.19140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12895291671156883,
+      "step": 1853
+    },
+    {
+      "epoch": 0.309,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.4160807291666666,
+      "learning_rate": 7.825555021458716e-05,
+      "loss": 6.7071,
+      "loss/crossentropy": 1.272746354341507,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21288197487592697,
+      "step": 1854
+    },
+    {
+      "epoch": 0.30916666666666665,
+      "grad_norm": 25.625,
+      "grad_norm_var": 1.3604166666666666,
+      "learning_rate": 7.823394750330387e-05,
+      "loss": 6.6636,
+      "loss/crossentropy": 2.123755916953087,
+      "loss/hidden": 3.1640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14926998503506184,
+      "step": 1855
+    },
+    {
+      "epoch": 0.30933333333333335,
+      "grad_norm": 26.875,
+      "grad_norm_var": 1.3869140625,
+      "learning_rate": 7.821233705152371e-05,
+      "loss": 6.3081,
+      "loss/crossentropy": 1.1460856795310974,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11831165105104446,
+      "step": 1856
+    },
+    {
+      "epoch": 0.3095,
+      "grad_norm": 27.75,
+      "grad_norm_var": 1.3869140625,
+      "learning_rate": 7.819071886517134e-05,
+      "loss": 6.8787,
+      "loss/crossentropy": 1.1290220320224762,
+      "loss/hidden": 3.65234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2073416095227003,
+      "step": 1857
+    },
+    {
+      "epoch": 0.30966666666666665,
+      "grad_norm": 23.625,
+      "grad_norm_var": 1.7598307291666666,
+      "learning_rate": 7.816909295017352e-05,
+      "loss": 6.0213,
+      "loss/crossentropy": 1.3497294038534164,
+      "loss/hidden": 3.1171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12365151569247246,
+      "step": 1858
+    },
+    {
+      "epoch": 0.30983333333333335,
+      "grad_norm": 26.625,
+      "grad_norm_var": 1.7697916666666667,
+      "learning_rate": 7.81474593124591e-05,
+      "loss": 6.2203,
+      "loss/crossentropy": 1.2937806993722916,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13160284096375108,
+      "step": 1859
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 26.25,
+      "grad_norm_var": 1.7059895833333334,
+      "learning_rate": 7.812581795795907e-05,
+      "loss": 6.2279,
+      "loss/crossentropy": 1.8884514570236206,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20325405150651932,
+      "step": 1860
+    },
+    {
+      "epoch": 0.31016666666666665,
+      "grad_norm": 25.375,
+      "grad_norm_var": 1.7171223958333333,
+      "learning_rate": 7.810416889260653e-05,
+      "loss": 6.2427,
+      "loss/crossentropy": 1.5505582392215729,
+      "loss/hidden": 3.13671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13619203492999077,
+      "step": 1861
+    },
+    {
+      "epoch": 0.31033333333333335,
+      "grad_norm": 27.5,
+      "grad_norm_var": 1.8155598958333334,
+      "learning_rate": 7.80825121223367e-05,
+      "loss": 7.0282,
+      "loss/crossentropy": 1.8412186950445175,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1582876518368721,
+      "step": 1862
+    },
+    {
+      "epoch": 0.3105,
+      "grad_norm": 27.25,
+      "grad_norm_var": 1.6624348958333333,
+      "learning_rate": 7.80608476530869e-05,
+      "loss": 6.4361,
+      "loss/crossentropy": 1.3256421834230423,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1557693425565958,
+      "step": 1863
+    },
+    {
+      "epoch": 0.31066666666666665,
+      "grad_norm": 25.125,
+      "grad_norm_var": 1.6921223958333333,
+      "learning_rate": 7.803917549079655e-05,
+      "loss": 6.2038,
+      "loss/crossentropy": 1.6339230090379715,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1802824381738901,
+      "step": 1864
+    },
+    {
+      "epoch": 0.31083333333333335,
+      "grad_norm": 25.125,
+      "grad_norm_var": 1.3624348958333334,
+      "learning_rate": 7.801749564140724e-05,
+      "loss": 6.6307,
+      "loss/crossentropy": 1.5514553636312485,
+      "loss/hidden": 2.98828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11378040164709091,
+      "step": 1865
+    },
+    {
+      "epoch": 0.311,
+      "grad_norm": 26.5,
+      "grad_norm_var": 1.3733723958333333,
+      "learning_rate": 7.799580811086258e-05,
+      "loss": 6.7113,
+      "loss/crossentropy": 1.5972132235765457,
+      "loss/hidden": 3.81640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30467648059129715,
+      "step": 1866
+    },
+    {
+      "epoch": 0.31116666666666665,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.4184895833333333,
+      "learning_rate": 7.797411290510835e-05,
+      "loss": 6.3771,
+      "loss/crossentropy": 1.295072615146637,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14648963510990143,
+      "step": 1867
+    },
+    {
+      "epoch": 0.31133333333333335,
+      "grad_norm": 25.5,
+      "grad_norm_var": 1.4577473958333333,
+      "learning_rate": 7.795241003009241e-05,
+      "loss": 6.3109,
+      "loss/crossentropy": 1.1807305663824081,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1505681350827217,
+      "step": 1868
+    },
+    {
+      "epoch": 0.3115,
+      "grad_norm": 27.375,
+      "grad_norm_var": 1.5229166666666667,
+      "learning_rate": 7.793069949176473e-05,
+      "loss": 7.0231,
+      "loss/crossentropy": 1.5879820436239243,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2128017395734787,
+      "step": 1869
+    },
+    {
+      "epoch": 0.31166666666666665,
+      "grad_norm": 25.0,
+      "grad_norm_var": 1.5926432291666666,
+      "learning_rate": 7.790898129607738e-05,
+      "loss": 6.1057,
+      "loss/crossentropy": 1.324481725692749,
+      "loss/hidden": 3.2265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13999750092625618,
+      "step": 1870
+    },
+    {
+      "epoch": 0.31183333333333335,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.8072916666666667,
+      "learning_rate": 7.788725544898452e-05,
+      "loss": 6.4336,
+      "loss/crossentropy": 1.0379912182688713,
+      "loss/hidden": 2.90625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09183950256556273,
+      "step": 1871
+    },
+    {
+      "epoch": 0.312,
+      "grad_norm": 26.625,
+      "grad_norm_var": 1.7955729166666667,
+      "learning_rate": 7.78655219564424e-05,
+      "loss": 6.4524,
+      "loss/crossentropy": 1.538333684206009,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12575656548142433,
+      "step": 1872
+    },
+    {
+      "epoch": 0.31216666666666665,
+      "grad_norm": 26.5,
+      "grad_norm_var": 1.6666666666666667,
+      "learning_rate": 7.784378082440941e-05,
+      "loss": 6.4221,
+      "loss/crossentropy": 1.8402161598205566,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18546854704618454,
+      "step": 1873
+    },
+    {
+      "epoch": 0.31233333333333335,
+      "grad_norm": 26.0,
+      "grad_norm_var": 1.1681640625,
+      "learning_rate": 7.782203205884598e-05,
+      "loss": 6.6287,
+      "loss/crossentropy": 1.447823904454708,
+      "loss/hidden": 3.1171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11326228268444538,
+      "step": 1874
+    },
+    {
+      "epoch": 0.3125,
+      "grad_norm": 24.5,
+      "grad_norm_var": 1.40390625,
+      "learning_rate": 7.780027566571465e-05,
+      "loss": 6.2397,
+      "loss/crossentropy": 1.87068110704422,
+      "loss/hidden": 3.14453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1607346385717392,
+      "step": 1875
+    },
+    {
+      "epoch": 0.31266666666666665,
+      "grad_norm": 28.0,
+      "grad_norm_var": 1.5770833333333334,
+      "learning_rate": 7.777851165098012e-05,
+      "loss": 6.7472,
+      "loss/crossentropy": 1.9179079830646515,
+      "loss/hidden": 3.671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19811002910137177,
+      "step": 1876
+    },
+    {
+      "epoch": 0.31283333333333335,
+      "grad_norm": 24.875,
+      "grad_norm_var": 1.6635416666666667,
+      "learning_rate": 7.775674002060905e-05,
+      "loss": 6.3514,
+      "loss/crossentropy": 1.3205182254314423,
+      "loss/hidden": 2.93359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13965397514402866,
+      "step": 1877
+    },
+    {
+      "epoch": 0.313,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.8080729166666667,
+      "learning_rate": 7.773496078057028e-05,
+      "loss": 6.3118,
+      "loss/crossentropy": 1.3714709877967834,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1656615175306797,
+      "step": 1878
+    },
+    {
+      "epoch": 0.31316666666666665,
+      "grad_norm": 25.0,
+      "grad_norm_var": 1.8854166666666667,
+      "learning_rate": 7.771317393683471e-05,
+      "loss": 6.5127,
+      "loss/crossentropy": 1.2658039927482605,
+      "loss/hidden": 3.02734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09252388961613178,
+      "step": 1879
+    },
+    {
+      "epoch": 0.31333333333333335,
+      "grad_norm": 24.0,
+      "grad_norm_var": 2.1426432291666666,
+      "learning_rate": 7.769137949537532e-05,
+      "loss": 6.4282,
+      "loss/crossentropy": 1.9153670966625214,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21619888022542,
+      "step": 1880
+    },
+    {
+      "epoch": 0.3135,
+      "grad_norm": 27.25,
+      "grad_norm_var": 2.1083333333333334,
+      "learning_rate": 7.766957746216721e-05,
+      "loss": 6.6214,
+      "loss/crossentropy": 1.9913443624973297,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1960178092122078,
+      "step": 1881
+    },
+    {
+      "epoch": 0.31366666666666665,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.4259765625,
+      "learning_rate": 7.764776784318751e-05,
+      "loss": 6.7529,
+      "loss/crossentropy": 2.0023012459278107,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23548029735684395,
+      "step": 1882
+    },
+    {
+      "epoch": 0.31383333333333335,
+      "grad_norm": 27.375,
+      "grad_norm_var": 2.239518229166667,
+      "learning_rate": 7.762595064441542e-05,
+      "loss": 6.6397,
+      "loss/crossentropy": 2.1164674758911133,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21258943900465965,
+      "step": 1883
+    },
+    {
+      "epoch": 0.314,
+      "grad_norm": 24.75,
+      "grad_norm_var": 2.3692057291666666,
+      "learning_rate": 7.76041258718323e-05,
+      "loss": 6.1556,
+      "loss/crossentropy": 1.4157161712646484,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1337559726089239,
+      "step": 1884
+    },
+    {
+      "epoch": 0.31416666666666665,
+      "grad_norm": 27.125,
+      "grad_norm_var": 2.340559895833333,
+      "learning_rate": 7.758229353142152e-05,
+      "loss": 6.6223,
+      "loss/crossentropy": 1.4183429181575775,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12671282701194286,
+      "step": 1885
+    },
+    {
+      "epoch": 0.31433333333333335,
+      "grad_norm": 24.875,
+      "grad_norm_var": 2.3645833333333335,
+      "learning_rate": 7.756045362916853e-05,
+      "loss": 5.9262,
+      "loss/crossentropy": 1.421977013349533,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12829083390533924,
+      "step": 1886
+    },
+    {
+      "epoch": 0.3145,
+      "grad_norm": 24.125,
+      "grad_norm_var": 2.3968098958333335,
+      "learning_rate": 7.753860617106086e-05,
+      "loss": 6.7611,
+      "loss/crossentropy": 1.9478589296340942,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19985918700695038,
+      "step": 1887
+    },
+    {
+      "epoch": 0.31466666666666665,
+      "grad_norm": 24.75,
+      "grad_norm_var": 2.4895833333333335,
+      "learning_rate": 7.751675116308812e-05,
+      "loss": 6.2293,
+      "loss/crossentropy": 0.9526519030332565,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12802209705114365,
+      "step": 1888
+    },
+    {
+      "epoch": 0.31483333333333335,
+      "grad_norm": 25.0,
+      "grad_norm_var": 2.5302083333333334,
+      "learning_rate": 7.7494888611242e-05,
+      "loss": 6.3645,
+      "loss/crossentropy": 1.7581877708435059,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19176558405160904,
+      "step": 1889
+    },
+    {
+      "epoch": 0.315,
+      "grad_norm": 25.25,
+      "grad_norm_var": 2.5559895833333335,
+      "learning_rate": 7.747301852151621e-05,
+      "loss": 6.2565,
+      "loss/crossentropy": 1.375806987285614,
+      "loss/hidden": 3.22265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12717818468809128,
+      "step": 1890
+    },
+    {
+      "epoch": 0.31516666666666665,
+      "grad_norm": 26.0,
+      "grad_norm_var": 2.4247395833333334,
+      "learning_rate": 7.74511408999066e-05,
+      "loss": 6.37,
+      "loss/crossentropy": 1.313630610704422,
+      "loss/hidden": 3.66015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10268084704875946,
+      "step": 1891
+    },
+    {
+      "epoch": 0.31533333333333335,
+      "grad_norm": 26.5,
+      "grad_norm_var": 2.155989583333333,
+      "learning_rate": 7.7429255752411e-05,
+      "loss": 6.3582,
+      "loss/crossentropy": 1.595801293849945,
+      "loss/hidden": 3.21484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1375584239140153,
+      "step": 1892
+    },
+    {
+      "epoch": 0.3155,
+      "grad_norm": 27.0,
+      "grad_norm_var": 2.159309895833333,
+      "learning_rate": 7.740736308502938e-05,
+      "loss": 6.1422,
+      "loss/crossentropy": 1.9210147261619568,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1734856776893139,
+      "step": 1893
+    },
+    {
+      "epoch": 0.31566666666666665,
+      "grad_norm": 24.375,
+      "grad_norm_var": 1.93125,
+      "learning_rate": 7.738546290376373e-05,
+      "loss": 6.1822,
+      "loss/crossentropy": 1.2764112129807472,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10514301294460893,
+      "step": 1894
+    },
+    {
+      "epoch": 0.31583333333333335,
+      "grad_norm": 26.875,
+      "grad_norm_var": 1.9634765625,
+      "learning_rate": 7.736355521461811e-05,
+      "loss": 6.73,
+      "loss/crossentropy": 1.3645439743995667,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17711367085576057,
+      "step": 1895
+    },
+    {
+      "epoch": 0.316,
+      "grad_norm": 26.875,
+      "grad_norm_var": 1.7643229166666667,
+      "learning_rate": 7.734164002359863e-05,
+      "loss": 6.2444,
+      "loss/crossentropy": 1.286956861615181,
+      "loss/hidden": 3.09765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1195491487160325,
+      "step": 1896
+    },
+    {
+      "epoch": 0.31616666666666665,
+      "grad_norm": 25.625,
+      "grad_norm_var": 1.6686848958333333,
+      "learning_rate": 7.731971733671346e-05,
+      "loss": 6.1124,
+      "loss/crossentropy": 1.4985006153583527,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2696618605405092,
+      "step": 1897
+    },
+    {
+      "epoch": 0.31633333333333336,
+      "grad_norm": 26.875,
+      "grad_norm_var": 1.2348307291666667,
+      "learning_rate": 7.729778715997284e-05,
+      "loss": 6.2936,
+      "loss/crossentropy": 0.9917602837085724,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13795838318765163,
+      "step": 1898
+    },
+    {
+      "epoch": 0.3165,
+      "grad_norm": 25.5,
+      "grad_norm_var": 1.0697916666666667,
+      "learning_rate": 7.727584949938907e-05,
+      "loss": 6.5076,
+      "loss/crossentropy": 1.3462112173438072,
+      "loss/hidden": 3.1328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13541871309280396,
+      "step": 1899
+    },
+    {
+      "epoch": 0.31666666666666665,
+      "grad_norm": 27.375,
+      "grad_norm_var": 1.1613932291666667,
+      "learning_rate": 7.725390436097643e-05,
+      "loss": 6.207,
+      "loss/crossentropy": 1.4423311352729797,
+      "loss/hidden": 3.5625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21663590148091316,
+      "step": 1900
+    },
+    {
+      "epoch": 0.31683333333333336,
+      "grad_norm": 25.75,
+      "grad_norm_var": 1.0518229166666666,
+      "learning_rate": 7.723195175075136e-05,
+      "loss": 6.069,
+      "loss/crossentropy": 1.1143035739660263,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13193629682064056,
+      "step": 1901
+    },
+    {
+      "epoch": 0.317,
+      "grad_norm": 25.75,
+      "grad_norm_var": 0.9921223958333333,
+      "learning_rate": 7.720999167473227e-05,
+      "loss": 6.2683,
+      "loss/crossentropy": 1.0505619198083878,
+      "loss/hidden": 3.109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11662804149091244,
+      "step": 1902
+    },
+    {
+      "epoch": 0.31716666666666665,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.871875,
+      "learning_rate": 7.718802413893963e-05,
+      "loss": 6.9684,
+      "loss/crossentropy": 1.5587396919727325,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20114586129784584,
+      "step": 1903
+    },
+    {
+      "epoch": 0.31733333333333336,
+      "grad_norm": 26.125,
+      "grad_norm_var": 3.686393229166667,
+      "learning_rate": 7.716604914939598e-05,
+      "loss": 6.2411,
+      "loss/crossentropy": 1.4164672195911407,
+      "loss/hidden": 3.140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12767393980175257,
+      "step": 1904
+    },
+    {
+      "epoch": 0.3175,
+      "grad_norm": 26.125,
+      "grad_norm_var": 3.5416666666666665,
+      "learning_rate": 7.714406671212589e-05,
+      "loss": 6.9527,
+      "loss/crossentropy": 1.6486473083496094,
+      "loss/hidden": 3.53125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1878268513828516,
+      "step": 1905
+    },
+    {
+      "epoch": 0.31766666666666665,
+      "grad_norm": 24.75,
+      "grad_norm_var": 3.644791666666667,
+      "learning_rate": 7.712207683315594e-05,
+      "loss": 6.4158,
+      "loss/crossentropy": 1.164580836892128,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1938272463157773,
+      "step": 1906
+    },
+    {
+      "epoch": 0.31783333333333336,
+      "grad_norm": 23.0,
+      "grad_norm_var": 4.419791666666667,
+      "learning_rate": 7.710007951851481e-05,
+      "loss": 6.2314,
+      "loss/crossentropy": 1.6975385248661041,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12435791827738285,
+      "step": 1907
+    },
+    {
+      "epoch": 0.318,
+      "grad_norm": 26.25,
+      "grad_norm_var": 4.418489583333334,
+      "learning_rate": 7.707807477423319e-05,
+      "loss": 6.7084,
+      "loss/crossentropy": 2.0439485013484955,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13873123191297054,
+      "step": 1908
+    },
+    {
+      "epoch": 0.31816666666666665,
+      "grad_norm": 26.0,
+      "grad_norm_var": 4.39140625,
+      "learning_rate": 7.705606260634379e-05,
+      "loss": 6.3084,
+      "loss/crossentropy": 1.9195210337638855,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16824555397033691,
+      "step": 1909
+    },
+    {
+      "epoch": 0.31833333333333336,
+      "grad_norm": 25.0,
+      "grad_norm_var": 4.258268229166666,
+      "learning_rate": 7.703404302088138e-05,
+      "loss": 5.9442,
+      "loss/crossentropy": 1.4394736289978027,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.148159084841609,
+      "step": 1910
+    },
+    {
+      "epoch": 0.3185,
+      "grad_norm": 25.75,
+      "grad_norm_var": 4.251822916666667,
+      "learning_rate": 7.701201602388276e-05,
+      "loss": 6.2713,
+      "loss/crossentropy": 1.2391781955957413,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15659042447805405,
+      "step": 1911
+    },
+    {
+      "epoch": 0.31866666666666665,
+      "grad_norm": 28.125,
+      "grad_norm_var": 4.45625,
+      "learning_rate": 7.698998162138673e-05,
+      "loss": 6.354,
+      "loss/crossentropy": 1.7924777567386627,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15180748887360096,
+      "step": 1912
+    },
+    {
+      "epoch": 0.31883333333333336,
+      "grad_norm": 26.625,
+      "grad_norm_var": 4.427083333333333,
+      "learning_rate": 7.696793981943417e-05,
+      "loss": 6.7418,
+      "loss/crossentropy": 1.4420042484998703,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1826731665059924,
+      "step": 1913
+    },
+    {
+      "epoch": 0.319,
+      "grad_norm": 24.375,
+      "grad_norm_var": 4.651041666666667,
+      "learning_rate": 7.694589062406796e-05,
+      "loss": 6.2945,
+      "loss/crossentropy": 1.643314927816391,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1626468040049076,
+      "step": 1914
+    },
+    {
+      "epoch": 0.31916666666666665,
+      "grad_norm": 26.75,
+      "grad_norm_var": 4.62890625,
+      "learning_rate": 7.692383404133301e-05,
+      "loss": 6.3934,
+      "loss/crossentropy": 1.2791866958141327,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16720141470432281,
+      "step": 1915
+    },
+    {
+      "epoch": 0.31933333333333336,
+      "grad_norm": 24.5,
+      "grad_norm_var": 4.7322265625,
+      "learning_rate": 7.690177007727625e-05,
+      "loss": 6.2058,
+      "loss/crossentropy": 1.4310691952705383,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16014227829873562,
+      "step": 1916
+    },
+    {
+      "epoch": 0.3195,
+      "grad_norm": 27.25,
+      "grad_norm_var": 4.7994140625,
+      "learning_rate": 7.687969873794667e-05,
+      "loss": 6.695,
+      "loss/crossentropy": 1.2179763466119766,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16675520688295364,
+      "step": 1917
+    },
+    {
+      "epoch": 0.31966666666666665,
+      "grad_norm": 29.75,
+      "grad_norm_var": 5.553580729166667,
+      "learning_rate": 7.685762002939523e-05,
+      "loss": 6.6029,
+      "loss/crossentropy": 1.2701485753059387,
+      "loss/hidden": 3.23046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11678503267467022,
+      "step": 1918
+    },
+    {
+      "epoch": 0.31983333333333336,
+      "grad_norm": 27.875,
+      "grad_norm_var": 2.726822916666667,
+      "learning_rate": 7.683553395767492e-05,
+      "loss": 6.9142,
+      "loss/crossentropy": 1.4391902685165405,
+      "loss/hidden": 3.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.141116825863719,
+      "step": 1919
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.004622395833333,
+      "learning_rate": 7.681344052884077e-05,
+      "loss": 6.7391,
+      "loss/crossentropy": 1.7268415987491608,
+      "loss/hidden": 3.54296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19017089530825615,
+      "step": 1920
+    },
+    {
+      "epoch": 0.32016666666666665,
+      "grad_norm": 27.375,
+      "grad_norm_var": 3.0775390625,
+      "learning_rate": 7.679133974894983e-05,
+      "loss": 6.8914,
+      "loss/crossentropy": 2.178110897541046,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19218476489186287,
+      "step": 1921
+    },
+    {
+      "epoch": 0.32033333333333336,
+      "grad_norm": 25.125,
+      "grad_norm_var": 3.00625,
+      "learning_rate": 7.676923162406115e-05,
+      "loss": 6.3706,
+      "loss/crossentropy": 1.1894247233867645,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18001854792237282,
+      "step": 1922
+    },
+    {
+      "epoch": 0.3205,
+      "grad_norm": 26.625,
+      "grad_norm_var": 2.1962890625,
+      "learning_rate": 7.674711616023581e-05,
+      "loss": 6.4039,
+      "loss/crossentropy": 1.5494977235794067,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15634138695895672,
+      "step": 1923
+    },
+    {
+      "epoch": 0.32066666666666666,
+      "grad_norm": 26.75,
+      "grad_norm_var": 2.1884765625,
+      "learning_rate": 7.672499336353687e-05,
+      "loss": 6.5562,
+      "loss/crossentropy": 1.5956921577453613,
+      "loss/hidden": 3.57421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21293289214372635,
+      "step": 1924
+    },
+    {
+      "epoch": 0.32083333333333336,
+      "grad_norm": 24.5,
+      "grad_norm_var": 2.4556640625,
+      "learning_rate": 7.670286324002944e-05,
+      "loss": 6.3067,
+      "loss/crossentropy": 1.5231582522392273,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15446319803595543,
+      "step": 1925
+    },
+    {
+      "epoch": 0.321,
+      "grad_norm": 41.0,
+      "grad_norm_var": 15.172330729166667,
+      "learning_rate": 7.668072579578058e-05,
+      "loss": 6.78,
+      "loss/crossentropy": 1.3537965267896652,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1880318894982338,
+      "step": 1926
+    },
+    {
+      "epoch": 0.32116666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 15.037239583333333,
+      "learning_rate": 7.665858103685944e-05,
+      "loss": 6.1524,
+      "loss/crossentropy": 1.3253830820322037,
+      "loss/hidden": 3.02734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.09621228277683258,
+      "step": 1927
+    },
+    {
+      "epoch": 0.32133333333333336,
+      "grad_norm": 27.875,
+      "grad_norm_var": 15.028125,
+      "learning_rate": 7.663642896933712e-05,
+      "loss": 6.4819,
+      "loss/crossentropy": 1.6501772999763489,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.231500755995512,
+      "step": 1928
+    },
+    {
+      "epoch": 0.3215,
+      "grad_norm": 27.5,
+      "grad_norm_var": 14.948372395833333,
+      "learning_rate": 7.66142695992867e-05,
+      "loss": 6.596,
+      "loss/crossentropy": 1.375167578458786,
+      "loss/hidden": 3.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11396941542625427,
+      "step": 1929
+    },
+    {
+      "epoch": 0.32166666666666666,
+      "grad_norm": 27.25,
+      "grad_norm_var": 14.162239583333333,
+      "learning_rate": 7.659210293278334e-05,
+      "loss": 6.3975,
+      "loss/crossentropy": 1.6278154253959656,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2192364428192377,
+      "step": 1930
+    },
+    {
+      "epoch": 0.32183333333333336,
+      "grad_norm": 26.375,
+      "grad_norm_var": 14.231184895833334,
+      "learning_rate": 7.656992897590414e-05,
+      "loss": 6.595,
+      "loss/crossentropy": 1.3742298930883408,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12720617931336164,
+      "step": 1931
+    },
+    {
+      "epoch": 0.322,
+      "grad_norm": 26.375,
+      "grad_norm_var": 13.593489583333334,
+      "learning_rate": 7.654774773472823e-05,
+      "loss": 6.2011,
+      "loss/crossentropy": 1.7735658288002014,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17184998467564583,
+      "step": 1932
+    },
+    {
+      "epoch": 0.32216666666666666,
+      "grad_norm": 26.125,
+      "grad_norm_var": 13.792122395833333,
+      "learning_rate": 7.65255592153367e-05,
+      "loss": 6.2184,
+      "loss/crossentropy": 1.8436724245548248,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17776019871234894,
+      "step": 1933
+    },
+    {
+      "epoch": 0.32233333333333336,
+      "grad_norm": 25.875,
+      "grad_norm_var": 13.814322916666667,
+      "learning_rate": 7.650336342381269e-05,
+      "loss": 6.1072,
+      "loss/crossentropy": 1.1867903620004654,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1665213666856289,
+      "step": 1934
+    },
+    {
+      "epoch": 0.3225,
+      "grad_norm": 23.375,
+      "grad_norm_var": 14.995572916666667,
+      "learning_rate": 7.648116036624126e-05,
+      "loss": 6.5229,
+      "loss/crossentropy": 1.2136850953102112,
+      "loss/hidden": 3.05859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11723973229527473,
+      "step": 1935
+    },
+    {
+      "epoch": 0.32266666666666666,
+      "grad_norm": 24.125,
+      "grad_norm_var": 15.620768229166666,
+      "learning_rate": 7.645895004870954e-05,
+      "loss": 6.3716,
+      "loss/crossentropy": 1.7954025566577911,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14210795238614082,
+      "step": 1936
+    },
+    {
+      "epoch": 0.32283333333333336,
+      "grad_norm": 26.5,
+      "grad_norm_var": 15.64765625,
+      "learning_rate": 7.643673247730658e-05,
+      "loss": 6.3718,
+      "loss/crossentropy": 1.3012201935052872,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18409869819879532,
+      "step": 1937
+    },
+    {
+      "epoch": 0.323,
+      "grad_norm": 25.75,
+      "grad_norm_var": 15.5041015625,
+      "learning_rate": 7.64145076581235e-05,
+      "loss": 6.4638,
+      "loss/crossentropy": 1.5930643379688263,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1491758320480585,
+      "step": 1938
+    },
+    {
+      "epoch": 0.32316666666666666,
+      "grad_norm": 24.375,
+      "grad_norm_var": 15.9869140625,
+      "learning_rate": 7.639227559725332e-05,
+      "loss": 6.3902,
+      "loss/crossentropy": 1.3990869522094727,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15044845081865788,
+      "step": 1939
+    },
+    {
+      "epoch": 0.3233333333333333,
+      "grad_norm": 25.375,
+      "grad_norm_var": 16.158072916666665,
+      "learning_rate": 7.637003630079111e-05,
+      "loss": 5.9661,
+      "loss/crossentropy": 1.3771826028823853,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1573531161993742,
+      "step": 1940
+    },
+    {
+      "epoch": 0.3235,
+      "grad_norm": 23.5,
+      "grad_norm_var": 16.54765625,
+      "learning_rate": 7.634778977483389e-05,
+      "loss": 5.9605,
+      "loss/crossentropy": 1.3241405487060547,
+      "loss/hidden": 3.125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13419854640960693,
+      "step": 1941
+    },
+    {
+      "epoch": 0.32366666666666666,
+      "grad_norm": 23.5,
+      "grad_norm_var": 2.76640625,
+      "learning_rate": 7.632553602548065e-05,
+      "loss": 5.8775,
+      "loss/crossentropy": 1.3533181101083755,
+      "loss/hidden": 3.12109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12711031548678875,
+      "step": 1942
+    },
+    {
+      "epoch": 0.3238333333333333,
+      "grad_norm": 24.875,
+      "grad_norm_var": 2.124739583333333,
+      "learning_rate": 7.630327505883242e-05,
+      "loss": 6.5926,
+      "loss/crossentropy": 1.1828708052635193,
+      "loss/hidden": 3.75390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25145961344242096,
+      "step": 1943
+    },
+    {
+      "epoch": 0.324,
+      "grad_norm": 25.625,
+      "grad_norm_var": 1.7427083333333333,
+      "learning_rate": 7.628100688099215e-05,
+      "loss": 6.8952,
+      "loss/crossentropy": 1.120570808649063,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27618863247334957,
+      "step": 1944
+    },
+    {
+      "epoch": 0.32416666666666666,
+      "grad_norm": 25.875,
+      "grad_norm_var": 1.4541015625,
+      "learning_rate": 7.62587314980648e-05,
+      "loss": 6.5664,
+      "loss/crossentropy": 1.1782270669937134,
+      "loss/hidden": 3.0390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11352577339857817,
+      "step": 1945
+    },
+    {
+      "epoch": 0.3243333333333333,
+      "grad_norm": 27.75,
+      "grad_norm_var": 1.5994140625,
+      "learning_rate": 7.623644891615727e-05,
+      "loss": 6.3751,
+      "loss/crossentropy": 1.2841320484876633,
+      "loss/hidden": 3.70703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1850954368710518,
+      "step": 1946
+    },
+    {
+      "epoch": 0.3245,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.854622395833333,
+      "learning_rate": 7.621415914137846e-05,
+      "loss": 7.1795,
+      "loss/crossentropy": 1.727049320936203,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16083285585045815,
+      "step": 1947
+    },
+    {
+      "epoch": 0.32466666666666666,
+      "grad_norm": 27.375,
+      "grad_norm_var": 4.013997395833333,
+      "learning_rate": 7.619186217983924e-05,
+      "loss": 6.5735,
+      "loss/crossentropy": 1.2679304331541061,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21421366184949875,
+      "step": 1948
+    },
+    {
+      "epoch": 0.3248333333333333,
+      "grad_norm": 24.0,
+      "grad_norm_var": 4.17890625,
+      "learning_rate": 7.616955803765249e-05,
+      "loss": 6.6339,
+      "loss/crossentropy": 1.2826143652200699,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1158591378480196,
+      "step": 1949
+    },
+    {
+      "epoch": 0.325,
+      "grad_norm": 25.25,
+      "grad_norm_var": 4.178580729166667,
+      "learning_rate": 7.614724672093296e-05,
+      "loss": 6.329,
+      "loss/crossentropy": 1.2010756433010101,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.162882711738348,
+      "step": 1950
+    },
+    {
+      "epoch": 0.32516666666666666,
+      "grad_norm": 24.375,
+      "grad_norm_var": 3.9525390625,
+      "learning_rate": 7.612492823579745e-05,
+      "loss": 6.4501,
+      "loss/crossentropy": 1.4341825097799301,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10451747477054596,
+      "step": 1951
+    },
+    {
+      "epoch": 0.3253333333333333,
+      "grad_norm": 26.0,
+      "grad_norm_var": 3.803125,
+      "learning_rate": 7.61026025883647e-05,
+      "loss": 6.2596,
+      "loss/crossentropy": 1.3979451805353165,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2629561945796013,
+      "step": 1952
+    },
+    {
+      "epoch": 0.3255,
+      "grad_norm": 27.625,
+      "grad_norm_var": 3.9994140625,
+      "learning_rate": 7.60802697847554e-05,
+      "loss": 6.7116,
+      "loss/crossentropy": 1.3776356428861618,
+      "loss/hidden": 2.96484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10853162780404091,
+      "step": 1953
+    },
+    {
+      "epoch": 0.32566666666666666,
+      "grad_norm": 25.875,
+      "grad_norm_var": 3.999739583333333,
+      "learning_rate": 7.605792983109222e-05,
+      "loss": 6.3798,
+      "loss/crossentropy": 1.2861874103546143,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14763324335217476,
+      "step": 1954
+    },
+    {
+      "epoch": 0.3258333333333333,
+      "grad_norm": 27.125,
+      "grad_norm_var": 3.951041666666667,
+      "learning_rate": 7.60355827334998e-05,
+      "loss": 6.2447,
+      "loss/crossentropy": 1.6014887690544128,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1440642587840557,
+      "step": 1955
+    },
+    {
+      "epoch": 0.326,
+      "grad_norm": 24.875,
+      "grad_norm_var": 4.00625,
+      "learning_rate": 7.60132284981047e-05,
+      "loss": 5.9239,
+      "loss/crossentropy": 1.4988811761140823,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1265989039093256,
+      "step": 1956
+    },
+    {
+      "epoch": 0.32616666666666666,
+      "grad_norm": 24.125,
+      "grad_norm_var": 3.8275390625,
+      "learning_rate": 7.599086713103547e-05,
+      "loss": 6.2312,
+      "loss/crossentropy": 1.4320566207170486,
+      "loss/hidden": 2.953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10050812736153603,
+      "step": 1957
+    },
+    {
+      "epoch": 0.3263333333333333,
+      "grad_norm": 24.25,
+      "grad_norm_var": 3.6150390625,
+      "learning_rate": 7.596849863842263e-05,
+      "loss": 6.1595,
+      "loss/crossentropy": 1.0710847079753876,
+      "loss/hidden": 3.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15653483383357525,
+      "step": 1958
+    },
+    {
+      "epoch": 0.3265,
+      "grad_norm": 24.75,
+      "grad_norm_var": 3.63515625,
+      "learning_rate": 7.594612302639859e-05,
+      "loss": 6.3477,
+      "loss/crossentropy": 1.452481895685196,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1926157921552658,
+      "step": 1959
+    },
+    {
+      "epoch": 0.32666666666666666,
+      "grad_norm": 23.25,
+      "grad_norm_var": 4.111393229166667,
+      "learning_rate": 7.592374030109777e-05,
+      "loss": 6.2755,
+      "loss/crossentropy": 1.5255981087684631,
+      "loss/hidden": 3.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.11009975895285606,
+      "step": 1960
+    },
+    {
+      "epoch": 0.3268333333333333,
+      "grad_norm": 25.125,
+      "grad_norm_var": 4.145768229166666,
+      "learning_rate": 7.590135046865651e-05,
+      "loss": 6.3481,
+      "loss/crossentropy": 1.5504071563482285,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13940811716020107,
+      "step": 1961
+    },
+    {
+      "epoch": 0.327,
+      "grad_norm": 26.75,
+      "grad_norm_var": 3.9509765625,
+      "learning_rate": 7.587895353521314e-05,
+      "loss": 6.4289,
+      "loss/crossentropy": 1.4041942059993744,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1428377777338028,
+      "step": 1962
+    },
+    {
+      "epoch": 0.32716666666666666,
+      "grad_norm": 24.875,
+      "grad_norm_var": 1.7233723958333333,
+      "learning_rate": 7.585654950690786e-05,
+      "loss": 6.467,
+      "loss/crossentropy": 1.6645765453577042,
+      "loss/hidden": 3.1484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1239413321018219,
+      "step": 1963
+    },
+    {
+      "epoch": 0.3273333333333333,
+      "grad_norm": 25.125,
+      "grad_norm_var": 1.4327473958333334,
+      "learning_rate": 7.58341383898829e-05,
+      "loss": 6.3363,
+      "loss/crossentropy": 1.437987893819809,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15620095655322075,
+      "step": 1964
+    },
+    {
+      "epoch": 0.3275,
+      "grad_norm": 25.125,
+      "grad_norm_var": 1.3302083333333334,
+      "learning_rate": 7.581172019028238e-05,
+      "loss": 5.9472,
+      "loss/crossentropy": 1.605808474123478,
+      "loss/hidden": 3.00390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1253122929483652,
+      "step": 1965
+    },
+    {
+      "epoch": 0.32766666666666666,
+      "grad_norm": 27.5,
+      "grad_norm_var": 1.6372395833333333,
+      "learning_rate": 7.578929491425238e-05,
+      "loss": 6.6689,
+      "loss/crossentropy": 1.294993743300438,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13339629210531712,
+      "step": 1966
+    },
+    {
+      "epoch": 0.3278333333333333,
+      "grad_norm": 25.5,
+      "grad_norm_var": 1.5593098958333333,
+      "learning_rate": 7.576686256794091e-05,
+      "loss": 6.2132,
+      "loss/crossentropy": 2.2327435314655304,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17829161509871483,
+      "step": 1967
+    },
+    {
+      "epoch": 0.328,
+      "grad_norm": 26.375,
+      "grad_norm_var": 1.5934895833333333,
+      "learning_rate": 7.574442315749793e-05,
+      "loss": 6.4249,
+      "loss/crossentropy": 1.466038852930069,
+      "loss/hidden": 3.1796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19742033816874027,
+      "step": 1968
+    },
+    {
+      "epoch": 0.32816666666666666,
+      "grad_norm": 23.625,
+      "grad_norm_var": 1.4684895833333333,
+      "learning_rate": 7.572197668907532e-05,
+      "loss": 6.4226,
+      "loss/crossentropy": 1.8144182711839676,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20674077793955803,
+      "step": 1969
+    },
+    {
+      "epoch": 0.3283333333333333,
+      "grad_norm": 26.125,
+      "grad_norm_var": 1.4927083333333333,
+      "learning_rate": 7.569952316882694e-05,
+      "loss": 6.2174,
+      "loss/crossentropy": 1.3762106895446777,
+      "loss/hidden": 2.96484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1315699713304639,
+      "step": 1970
+    },
+    {
+      "epoch": 0.3285,
+      "grad_norm": 26.375,
+      "grad_norm_var": 1.3434895833333333,
+      "learning_rate": 7.567706260290851e-05,
+      "loss": 6.4584,
+      "loss/crossentropy": 1.0963778346776962,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18599781021475792,
+      "step": 1971
+    },
+    {
+      "epoch": 0.32866666666666666,
+      "grad_norm": 24.625,
+      "grad_norm_var": 1.359375,
+      "learning_rate": 7.565459499747775e-05,
+      "loss": 5.9722,
+      "loss/crossentropy": 1.5091693997383118,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2419084869325161,
+      "step": 1972
+    },
+    {
+      "epoch": 0.3288333333333333,
+      "grad_norm": 24.375,
+      "grad_norm_var": 1.3268229166666667,
+      "learning_rate": 7.563212035869425e-05,
+      "loss": 6.2622,
+      "loss/crossentropy": 1.4423122555017471,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15351855754852295,
+      "step": 1973
+    },
+    {
+      "epoch": 0.329,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.4660807291666667,
+      "learning_rate": 7.56096386927196e-05,
+      "loss": 6.3567,
+      "loss/crossentropy": 2.25230872631073,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19635501876473427,
+      "step": 1974
+    },
+    {
+      "epoch": 0.32916666666666666,
+      "grad_norm": 26.625,
+      "grad_norm_var": 1.5197916666666667,
+      "learning_rate": 7.558715000571726e-05,
+      "loss": 6.6388,
+      "loss/crossentropy": 1.3071192353963852,
+      "loss/hidden": 3.203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14108655788004398,
+      "step": 1975
+    },
+    {
+      "epoch": 0.3293333333333333,
+      "grad_norm": 27.875,
+      "grad_norm_var": 1.4499348958333333,
+      "learning_rate": 7.55646543038526e-05,
+      "loss": 6.7195,
+      "loss/crossentropy": 1.2048006281256676,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12995907431468368,
+      "step": 1976
+    },
+    {
+      "epoch": 0.3295,
+      "grad_norm": 26.0,
+      "grad_norm_var": 1.4166666666666667,
+      "learning_rate": 7.5542151593293e-05,
+      "loss": 6.4977,
+      "loss/crossentropy": 1.1852080002427101,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10718733165413141,
+      "step": 1977
+    },
+    {
+      "epoch": 0.32966666666666666,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.153580729166667,
+      "learning_rate": 7.551964188020766e-05,
+      "loss": 6.5545,
+      "loss/crossentropy": 1.4078846126794815,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1846070997416973,
+      "step": 1978
+    },
+    {
+      "epoch": 0.3298333333333333,
+      "grad_norm": 25.5,
+      "grad_norm_var": 2.0809895833333334,
+      "learning_rate": 7.549712517076777e-05,
+      "loss": 6.3851,
+      "loss/crossentropy": 2.0256086885929108,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1983210016041994,
+      "step": 1979
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 25.0,
+      "grad_norm_var": 2.0978515625,
+      "learning_rate": 7.547460147114641e-05,
+      "loss": 6.0587,
+      "loss/crossentropy": 1.5401402115821838,
+      "loss/hidden": 3.03125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.12004663981497288,
+      "step": 1980
+    },
+    {
+      "epoch": 0.33016666666666666,
+      "grad_norm": 26.5,
+      "grad_norm_var": 2.042708333333333,
+      "learning_rate": 7.545207078751857e-05,
+      "loss": 6.3717,
+      "loss/crossentropy": 1.78058473020792,
+      "loss/hidden": 3.12109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18446800578385592,
+      "step": 1981
+    },
+    {
+      "epoch": 0.3303333333333333,
+      "grad_norm": 26.125,
+      "grad_norm_var": 1.9145182291666667,
+      "learning_rate": 7.542953312606117e-05,
+      "loss": 6.5679,
+      "loss/crossentropy": 1.3367803692817688,
+      "loss/hidden": 3.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15712000243365765,
+      "step": 1982
+    },
+    {
+      "epoch": 0.3305,
+      "grad_norm": 26.125,
+      "grad_norm_var": 1.89140625,
+      "learning_rate": 7.540698849295305e-05,
+      "loss": 6.6764,
+      "loss/crossentropy": 1.5010664910078049,
+      "loss/hidden": 3.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.13286457024514675,
+      "step": 1983
+    },
+    {
+      "epoch": 0.33066666666666666,
+      "grad_norm": 27.125,
+      "grad_norm_var": 1.953125,
+      "learning_rate": 7.538443689437492e-05,
+      "loss": 6.6207,
+      "loss/crossentropy": 1.2388575226068497,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16265138238668442,
+      "step": 1984
+    },
+    {
+      "epoch": 0.3308333333333333,
+      "grad_norm": 26.0,
+      "grad_norm_var": 1.5041015625,
+      "learning_rate": 7.536187833650947e-05,
+      "loss": 6.6566,
+      "loss/crossentropy": 1.2816483676433563,
+      "loss/hidden": 3.12890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.10099411942064762,
+      "step": 1985
+    },
+    {
+      "epoch": 0.331,
+      "grad_norm": 26.375,
+      "grad_norm_var": 1.5020182291666666,
+      "learning_rate": 7.53393128255412e-05,
+      "loss": 6.8827,
+      "loss/crossentropy": 0.948730118572712,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3065859228372574,
+      "step": 1986
+    },
+    {
+      "epoch": 0.33116666666666666,
+      "grad_norm": 26.5,
+      "grad_norm_var": 1.50390625,
+      "learning_rate": 7.531674036765662e-05,
+      "loss": 6.8775,
+      "loss/crossentropy": 1.9316487163305283,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15175139717757702,
+      "step": 1987
+    },
+    {
+      "epoch": 0.3313333333333333,
+      "grad_norm": 26.125,
+      "grad_norm_var": 1.30390625,
+      "learning_rate": 7.52941609690441e-05,
+      "loss": 6.6747,
+      "loss/crossentropy": 1.6824732720851898,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19663692638278008,
+      "step": 1988
+    },
+    {
+      "epoch": 0.3315,
+      "grad_norm": 28.25,
+      "grad_norm_var": 1.1848307291666667,
+      "learning_rate": 7.52715746358939e-05,
+      "loss": 6.503,
+      "loss/crossentropy": 1.5091774463653564,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29703205451369286,
+      "step": 1989
+    },
+    {
+      "epoch": 0.33166666666666667,
+      "grad_norm": 26.0,
+      "grad_norm_var": 1.1947916666666667,
+      "learning_rate": 7.524898137439814e-05,
+      "loss": 6.4472,
+      "loss/crossentropy": 1.7559304237365723,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24768796935677528,
+      "step": 1990
+    },
+    {
+      "epoch": 0.3318333333333333,
+      "grad_norm": 26.5,
+      "grad_norm_var": 1.1952473958333334,
+      "learning_rate": 7.522638119075096e-05,
+      "loss": 6.5614,
+      "loss/crossentropy": 0.6225664392113686,
+      "loss/hidden": 2.796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.06071426905691624,
+      "step": 1991
+    },
+    {
+      "epoch": 0.332,
+      "grad_norm": 25.625,
+      "grad_norm_var": 1.1249348958333334,
+      "learning_rate": 7.520377409114831e-05,
+      "loss": 6.2228,
+      "loss/crossentropy": 1.62913578748703,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1353142112493515,
+      "step": 1992
+    },
+    {
+      "epoch": 0.33216666666666667,
+      "grad_norm": 27.25,
+      "grad_norm_var": 1.1483723958333334,
+      "learning_rate": 7.518116008178805e-05,
+      "loss": 6.7077,
+      "loss/crossentropy": 1.8163381218910217,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.14655482582747936,
+      "step": 1993
+    },
+    {
+      "epoch": 0.3323333333333333,
+      "grad_norm": 23.25,
+      "grad_norm_var": 1.1643229166666667,
+      "learning_rate": 7.515853916886993e-05,
+      "loss": 6.268,
+      "loss/crossentropy": 1.3509615659713745,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1473417840898037,
+      "step": 1994
+    },
+    {
+      "epoch": 0.3325,
+      "grad_norm": 26.5,
+      "grad_norm_var": 1.14140625,
+      "learning_rate": 7.513591135859561e-05,
+      "loss": 6.6178,
+      "loss/crossentropy": 1.7216495871543884,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15290702134370804,
+      "step": 1995
+    },
+    {
+      "epoch": 0.33266666666666667,
+      "grad_norm": 25.875,
+      "grad_norm_var": 1.0488932291666666,
+      "learning_rate": 7.511327665716863e-05,
+      "loss": 6.3424,
+      "loss/crossentropy": 1.327872171998024,
+      "loss/hidden": 3.171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15156515128910542,
+      "step": 1996
+    },
+    {
+      "epoch": 0.3328333333333333,
+      "grad_norm": 22.375,
+      "grad_norm_var": 1.9791666666666667,
+      "learning_rate": 7.509063507079443e-05,
+      "loss": 6.1078,
+      "loss/crossentropy": 1.865316480398178,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16528227180242538,
+      "step": 1997
+    },
+    {
+      "epoch": 0.333,
+      "grad_norm": 25.5,
+      "grad_norm_var": 1.9931640625,
+      "learning_rate": 7.506798660568031e-05,
+      "loss": 6.4771,
+      "loss/crossentropy": 1.3237116634845734,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15546652302145958,
+      "step": 1998
+    },
+    {
+      "epoch": 0.33316666666666667,
+      "grad_norm": 24.5,
+      "grad_norm_var": 2.12265625,
+      "learning_rate": 7.50453312680355e-05,
+      "loss": 6.2661,
+      "loss/crossentropy": 1.6104268431663513,
+      "loss/hidden": 3.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18106029368937016,
+      "step": 1999
+    },
+    {
+      "epoch": 0.3333333333333333,
+      "grad_norm": 25.75,
+      "grad_norm_var": 2.0087890625,
+      "learning_rate": 7.502266906407107e-05,
+      "loss": 6.5924,
+      "loss/crossentropy": 1.7930181920528412,
+      "loss/hidden": 3.1875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1696604359894991,
+      "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
+}