diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,24018 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.25,
+  "eval_steps": 250,
+  "global_step": 2000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.000125,
+      "grad_norm": 2.537714958190918,
+      "learning_rate": 1.0000000000000002e-06,
+      "loss": 0.5468,
+      "loss/crossentropy": 2.2066214084625244,
+      "loss/hidden": 0.248046875,
+      "loss/logits": 0.03443578630685806,
+      "loss/reg": 0.026429571211338043,
+      "step": 1
+    },
+    {
+      "epoch": 0.00025,
+      "grad_norm": 2.4728448390960693,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 0.6642,
+      "loss/crossentropy": 2.132329225540161,
+      "loss/hidden": 0.345703125,
+      "loss/logits": 0.05424630641937256,
+      "loss/reg": 0.026429571211338043,
+      "step": 2
+    },
+    {
+      "epoch": 0.000375,
+      "grad_norm": 2.773984670639038,
+      "learning_rate": 3e-06,
+      "loss": 0.5822,
+      "loss/crossentropy": 2.3457791805267334,
+      "loss/hidden": 0.2734375,
+      "loss/logits": 0.044443465769290924,
+      "loss/reg": 0.02642953023314476,
+      "step": 3
+    },
+    {
+      "epoch": 0.0005,
+      "grad_norm": 4.14040470123291,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 0.7192,
+      "loss/crossentropy": 2.7209200859069824,
+      "loss/hidden": 0.35546875,
+      "loss/logits": 0.09940779209136963,
+      "loss/reg": 0.026429466903209686,
+      "step": 4
+    },
+    {
+      "epoch": 0.000625,
+      "grad_norm": 1.9164764881134033,
+      "learning_rate": 5e-06,
+      "loss": 0.5467,
+      "loss/crossentropy": 2.4304752349853516,
+      "loss/hidden": 0.244140625,
+      "loss/logits": 0.03826362267136574,
+      "loss/reg": 0.02642936445772648,
+      "step": 5
+    },
+    {
+      "epoch": 0.00075,
+      "grad_norm": 1.9878246784210205,
+      "learning_rate": 6e-06,
+      "loss": 0.517,
+      "loss/crossentropy": 2.472181797027588,
+      "loss/hidden": 0.2255859375,
+      "loss/logits": 0.027161670848727226,
+      "loss/reg": 0.02642924338579178,
+      "step": 6
+    },
+    {
+      "epoch": 0.000875,
+      "grad_norm": 2.1939733028411865,
+      "learning_rate": 7.000000000000001e-06,
+      "loss": 0.6043,
+      "loss/crossentropy": 2.241501808166504,
+      "loss/hidden": 0.298828125,
+      "loss/logits": 0.04118040204048157,
+      "loss/reg": 0.02642909064888954,
+      "step": 7
+    },
+    {
+      "epoch": 0.001,
+      "grad_norm": 3.516223907470703,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 0.5199,
+      "loss/crossentropy": 2.409766912460327,
+      "loss/hidden": 0.2236328125,
+      "loss/logits": 0.032000549137592316,
+      "loss/reg": 0.02642889879643917,
+      "step": 8
+    },
+    {
+      "epoch": 0.001125,
+      "grad_norm": 1.9335486888885498,
+      "learning_rate": 9e-06,
+      "loss": 0.5575,
+      "loss/crossentropy": 2.6256861686706543,
+      "loss/hidden": 0.255859375,
+      "loss/logits": 0.037392452359199524,
+      "loss/reg": 0.02642873302102089,
+      "step": 9
+    },
+    {
+      "epoch": 0.00125,
+      "grad_norm": 1.6782876253128052,
+      "learning_rate": 1e-05,
+      "loss": 0.5162,
+      "loss/crossentropy": 2.1947107315063477,
+      "loss/hidden": 0.2255859375,
+      "loss/logits": 0.026354767382144928,
+      "loss/reg": 0.026428483426570892,
+      "step": 10
+    },
+    {
+      "epoch": 0.001375,
+      "grad_norm": 10.848552703857422,
+      "learning_rate": 1.1000000000000001e-05,
+      "loss": 1.0046,
+      "loss/crossentropy": 3.0539069175720215,
+      "loss/hidden": 0.640625,
+      "loss/logits": 0.09970991313457489,
+      "loss/reg": 0.026428230106830597,
+      "step": 11
+    },
+    {
+      "epoch": 0.0015,
+      "grad_norm": 2.237061023712158,
+      "learning_rate": 1.2e-05,
+      "loss": 0.563,
+      "loss/crossentropy": 2.5601325035095215,
+      "loss/hidden": 0.248046875,
+      "loss/logits": 0.050660137087106705,
+      "loss/reg": 0.02642793208360672,
+      "step": 12
+    },
+    {
+      "epoch": 0.001625,
+      "grad_norm": 1.4406346082687378,
+      "learning_rate": 1.3000000000000001e-05,
+      "loss": 0.507,
+      "loss/crossentropy": 1.965380311012268,
+      "loss/hidden": 0.2197265625,
+      "loss/logits": 0.02295786701142788,
+      "loss/reg": 0.02642756886780262,
+      "step": 13
+    },
+    {
+      "epoch": 0.00175,
+      "grad_norm": 3.0757036209106445,
+      "learning_rate": 1.4000000000000001e-05,
+      "loss": 0.7761,
+      "loss/crossentropy": 2.15138840675354,
+      "loss/hidden": 0.4375,
+      "loss/logits": 0.07431840896606445,
+      "loss/reg": 0.026427194476127625,
+      "step": 14
+    },
+    {
+      "epoch": 0.001875,
+      "grad_norm": 2.8731143474578857,
+      "learning_rate": 1.5e-05,
+      "loss": 0.4684,
+      "loss/crossentropy": 2.5530812740325928,
+      "loss/hidden": 0.1845703125,
+      "loss/logits": 0.019558344036340714,
+      "loss/reg": 0.026426764205098152,
+      "step": 15
+    },
+    {
+      "epoch": 0.002,
+      "grad_norm": 2.5288755893707275,
+      "grad_norm_var": 4.846526347105633,
+      "learning_rate": 1.6000000000000003e-05,
+      "loss": 0.5781,
+      "loss/crossentropy": 2.5096747875213623,
+      "loss/hidden": 0.275390625,
+      "loss/logits": 0.0384209081530571,
+      "loss/reg": 0.026426298543810844,
+      "step": 16
+    },
+    {
+      "epoch": 0.002125,
+      "grad_norm": 2.0281474590301514,
+      "grad_norm_var": 4.89482291121508,
+      "learning_rate": 1.7000000000000003e-05,
+      "loss": 0.5318,
+      "loss/crossentropy": 2.396097421646118,
+      "loss/hidden": 0.232421875,
+      "loss/logits": 0.03516196087002754,
+      "loss/reg": 0.026425909250974655,
+      "step": 17
+    },
+    {
+      "epoch": 0.00225,
+      "grad_norm": 2.4487411975860596,
+      "grad_norm_var": 4.896482229626747,
+      "learning_rate": 1.8e-05,
+      "loss": 0.5984,
+      "loss/crossentropy": 2.3916616439819336,
+      "loss/hidden": 0.28125,
+      "loss/logits": 0.052923329174518585,
+      "loss/reg": 0.026425503194332123,
+      "step": 18
+    },
+    {
+      "epoch": 0.002375,
+      "grad_norm": 1.986022710800171,
+      "grad_norm_var": 4.956548008938709,
+      "learning_rate": 1.9e-05,
+      "loss": 0.5504,
+      "loss/crossentropy": 2.4791200160980225,
+      "loss/hidden": 0.2421875,
+      "loss/logits": 0.04391499236226082,
+      "loss/reg": 0.026425078511238098,
+      "step": 19
+    },
+    {
+      "epoch": 0.0025,
+      "grad_norm": 2.0934784412384033,
+      "grad_norm_var": 4.887277710992484,
+      "learning_rate": 2e-05,
+      "loss": 0.5369,
+      "loss/crossentropy": 2.1741297245025635,
+      "loss/hidden": 0.2392578125,
+      "loss/logits": 0.03335873782634735,
+      "loss/reg": 0.026424556970596313,
+      "step": 20
+    },
+    {
+      "epoch": 0.002625,
+      "grad_norm": 1.9445254802703857,
+      "grad_norm_var": 4.884025740026245,
+      "learning_rate": 2.1e-05,
+      "loss": 0.4865,
+      "loss/crossentropy": 2.45112943649292,
+      "loss/hidden": 0.1962890625,
+      "loss/logits": 0.025960583239793777,
+      "loss/reg": 0.02642405778169632,
+      "step": 21
+    },
+    {
+      "epoch": 0.00275,
+      "grad_norm": 3.070704221725464,
+      "grad_norm_var": 4.8399171328504895,
+      "learning_rate": 2.2000000000000003e-05,
+      "loss": 0.5887,
+      "loss/crossentropy": 2.1550512313842773,
+      "loss/hidden": 0.275390625,
+      "loss/logits": 0.049097511917352676,
+      "loss/reg": 0.02642347477376461,
+      "step": 22
+    },
+    {
+      "epoch": 0.002875,
+      "grad_norm": 2.2452821731567383,
+      "grad_norm_var": 4.835466428034186,
+      "learning_rate": 2.3000000000000003e-05,
+      "loss": 0.5482,
+      "loss/crossentropy": 2.1640255451202393,
+      "loss/hidden": 0.2470703125,
+      "loss/logits": 0.036868080496788025,
+      "loss/reg": 0.026422815397381783,
+      "step": 23
+    },
+    {
+      "epoch": 0.003,
+      "grad_norm": 2.032148838043213,
+      "grad_norm_var": 4.84560617678243,
+      "learning_rate": 2.4e-05,
+      "loss": 0.5756,
+      "loss/crossentropy": 2.323482036590576,
+      "loss/hidden": 0.271484375,
+      "loss/logits": 0.03993295133113861,
+      "loss/reg": 0.02642211876809597,
+      "step": 24
+    },
+    {
+      "epoch": 0.003125,
+      "grad_norm": 1.763465404510498,
+      "grad_norm_var": 4.8665883230545335,
+      "learning_rate": 2.5e-05,
+      "loss": 0.5139,
+      "loss/crossentropy": 2.33661150932312,
+      "loss/hidden": 0.22265625,
+      "loss/logits": 0.02701444923877716,
+      "loss/reg": 0.026421383023262024,
+      "step": 25
+    },
+    {
+      "epoch": 0.00325,
+      "grad_norm": 1.7001625299453735,
+      "grad_norm_var": 4.863438686484135,
+      "learning_rate": 2.6000000000000002e-05,
+      "loss": 0.5696,
+      "loss/crossentropy": 2.2305383682250977,
+      "loss/hidden": 0.275390625,
+      "loss/logits": 0.02997000887989998,
+      "loss/reg": 0.026420695707201958,
+      "step": 26
+    },
+    {
+      "epoch": 0.003375,
+      "grad_norm": 3.474130392074585,
+      "grad_norm_var": 0.318824614624526,
+      "learning_rate": 2.7000000000000002e-05,
+      "loss": 0.5456,
+      "loss/crossentropy": 2.1680750846862793,
+      "loss/hidden": 0.2451171875,
+      "loss/logits": 0.036261945962905884,
+      "loss/reg": 0.0264199897646904,
+      "step": 27
+    },
+    {
+      "epoch": 0.0035,
+      "grad_norm": 3.8201987743377686,
+      "grad_norm_var": 0.46030846745106163,
+      "learning_rate": 2.8000000000000003e-05,
+      "loss": 0.5181,
+      "loss/crossentropy": 2.418672800064087,
+      "loss/hidden": 0.224609375,
+      "loss/logits": 0.029263213276863098,
+      "loss/reg": 0.02641921117901802,
+      "step": 28
+    },
+    {
+      "epoch": 0.003625,
+      "grad_norm": 1.9781090021133423,
+      "grad_norm_var": 0.40905077024170067,
+      "learning_rate": 2.9e-05,
+      "loss": 0.5379,
+      "loss/crossentropy": 2.390868663787842,
+      "loss/hidden": 0.2392578125,
+      "loss/logits": 0.034465983510017395,
+      "loss/reg": 0.026418352499604225,
+      "step": 29
+    },
+    {
+      "epoch": 0.00375,
+      "grad_norm": 1.6551319360733032,
+      "grad_norm_var": 0.41503895204729413,
+      "learning_rate": 3e-05,
+      "loss": 0.496,
+      "loss/crossentropy": 2.5960400104522705,
+      "loss/hidden": 0.205078125,
+      "loss/logits": 0.02678578905761242,
+      "loss/reg": 0.026417305693030357,
+      "step": 30
+    },
+    {
+      "epoch": 0.003875,
+      "grad_norm": 1.7136921882629395,
+      "grad_norm_var": 0.4185952392532807,
+      "learning_rate": 3.1e-05,
+      "loss": 0.5235,
+      "loss/crossentropy": 2.349839687347412,
+      "loss/hidden": 0.2275390625,
+      "loss/logits": 0.031792763620615005,
+      "loss/reg": 0.026416433975100517,
+      "step": 31
+    },
+    {
+      "epoch": 0.004,
+      "grad_norm": 1.9992157220840454,
+      "grad_norm_var": 0.41856547198587274,
+      "learning_rate": 3.2000000000000005e-05,
+      "loss": 0.4928,
+      "loss/crossentropy": 2.3164803981781006,
+      "loss/hidden": 0.2041015625,
+      "loss/logits": 0.024579893797636032,
+      "loss/reg": 0.02641524001955986,
+      "step": 32
+    },
+    {
+      "epoch": 0.004125,
+      "grad_norm": 2.705052614212036,
+      "grad_norm_var": 0.42744416353313663,
+      "learning_rate": 3.3e-05,
+      "loss": 0.5732,
+      "loss/crossentropy": 2.42107892036438,
+      "loss/hidden": 0.275390625,
+      "loss/logits": 0.03370767831802368,
+      "loss/reg": 0.02641397900879383,
+      "step": 33
+    },
+    {
+      "epoch": 0.00425,
+      "grad_norm": 1.8898464441299438,
+      "grad_norm_var": 0.4350913020843951,
+      "learning_rate": 3.4000000000000007e-05,
+      "loss": 0.5531,
+      "loss/crossentropy": 2.4147770404815674,
+      "loss/hidden": 0.25390625,
+      "loss/logits": 0.03504405915737152,
+      "loss/reg": 0.026412710547447205,
+      "step": 34
+    },
+    {
+      "epoch": 0.004375,
+      "grad_norm": 4.9570159912109375,
+      "grad_norm_var": 0.8804344329355491,
+      "learning_rate": 3.5e-05,
+      "loss": 0.6763,
+      "loss/crossentropy": 1.6753497123718262,
+      "loss/hidden": 0.376953125,
+      "loss/logits": 0.03519564867019653,
+      "loss/reg": 0.0264116358011961,
+      "step": 35
+    },
+    {
+      "epoch": 0.0045,
+      "grad_norm": 4.928956508636475,
+      "grad_norm_var": 1.2518721453244992,
+      "learning_rate": 3.6e-05,
+      "loss": 0.7329,
+      "loss/crossentropy": 2.6104867458343506,
+      "loss/hidden": 0.400390625,
+      "loss/logits": 0.06845290958881378,
+      "loss/reg": 0.02641039527952671,
+      "step": 36
+    },
+    {
+      "epoch": 0.004625,
+      "grad_norm": 7.503647327423096,
+      "grad_norm_var": 2.684651641752033,
+      "learning_rate": 3.7e-05,
+      "loss": 0.6258,
+      "loss/crossentropy": 2.2158656120300293,
+      "loss/hidden": 0.318359375,
+      "loss/logits": 0.043342188000679016,
+      "loss/reg": 0.026409219950437546,
+      "step": 37
+    },
+    {
+      "epoch": 0.00475,
+      "grad_norm": 2.6838622093200684,
+      "grad_norm_var": 2.6885420074665602,
+      "learning_rate": 3.8e-05,
+      "loss": 0.5939,
+      "loss/crossentropy": 2.344879627227783,
+      "loss/hidden": 0.28515625,
+      "loss/logits": 0.04461552947759628,
+      "loss/reg": 0.026408080011606216,
+      "step": 38
+    },
+    {
+      "epoch": 0.004875,
+      "grad_norm": 3.357893705368042,
+      "grad_norm_var": 2.662758933855309,
+      "learning_rate": 3.9000000000000006e-05,
+      "loss": 0.5729,
+      "loss/crossentropy": 2.6759543418884277,
+      "loss/hidden": 0.275390625,
+      "loss/logits": 0.033413954079151154,
+      "loss/reg": 0.026406895369291306,
+      "step": 39
+    },
+    {
+      "epoch": 0.005,
+      "grad_norm": 3.0177316665649414,
+      "grad_norm_var": 2.5949485604856193,
+      "learning_rate": 4e-05,
+      "loss": 0.7498,
+      "loss/crossentropy": 2.2261273860931396,
+      "loss/hidden": 0.408203125,
+      "loss/logits": 0.07758316397666931,
+      "loss/reg": 0.026405224576592445,
+      "step": 40
+    },
+    {
+      "epoch": 0.005125,
+      "grad_norm": 2.1196699142456055,
+      "grad_norm_var": 2.54074274703229,
+      "learning_rate": 4.1e-05,
+      "loss": 0.6396,
+      "loss/crossentropy": 2.193378448486328,
+      "loss/hidden": 0.30859375,
+      "loss/logits": 0.06692355871200562,
+      "loss/reg": 0.026403924450278282,
+      "step": 41
+    },
+    {
+      "epoch": 0.00525,
+      "grad_norm": 2.456051826477051,
+      "grad_norm_var": 2.435973046683167,
+      "learning_rate": 4.2e-05,
+      "loss": 0.5571,
+      "loss/crossentropy": 1.9526888132095337,
+      "loss/hidden": 0.26171875,
+      "loss/logits": 0.03133418411016464,
+      "loss/reg": 0.026402529329061508,
+      "step": 42
+    },
+    {
+      "epoch": 0.005375,
+      "grad_norm": 2.257375717163086,
+      "grad_norm_var": 2.474501380785125,
+      "learning_rate": 4.3e-05,
+      "loss": 0.5544,
+      "loss/crossentropy": 2.3284847736358643,
+      "loss/hidden": 0.25390625,
+      "loss/logits": 0.03650724142789841,
+      "loss/reg": 0.026400938630104065,
+      "step": 43
+    },
+    {
+      "epoch": 0.0055,
+      "grad_norm": 2.9145264625549316,
+      "grad_norm_var": 2.4345975605903694,
+      "learning_rate": 4.4000000000000006e-05,
+      "loss": 0.5175,
+      "loss/crossentropy": 2.295241594314575,
+      "loss/hidden": 0.220703125,
+      "loss/logits": 0.03284794092178345,
+      "loss/reg": 0.026399515569210052,
+      "step": 44
+    },
+    {
+      "epoch": 0.005625,
+      "grad_norm": 3.1294264793395996,
+      "grad_norm_var": 2.3592519473156615,
+      "learning_rate": 4.5e-05,
+      "loss": 0.5567,
+      "loss/crossentropy": 2.660597085952759,
+      "loss/hidden": 0.255859375,
+      "loss/logits": 0.03686758130788803,
+      "loss/reg": 0.026398126035928726,
+      "step": 45
+    },
+    {
+      "epoch": 0.00575,
+      "grad_norm": 2.197265863418579,
+      "grad_norm_var": 2.2745842657817748,
+      "learning_rate": 4.600000000000001e-05,
+      "loss": 0.5512,
+      "loss/crossentropy": 2.3832643032073975,
+      "loss/hidden": 0.2490234375,
+      "loss/logits": 0.038256023079156876,
+      "loss/reg": 0.02639671601355076,
+      "step": 46
+    },
+    {
+      "epoch": 0.005875,
+      "grad_norm": 2.883378744125366,
+      "grad_norm_var": 2.141634704665381,
+      "learning_rate": 4.7e-05,
+      "loss": 0.5298,
+      "loss/crossentropy": 2.6035244464874268,
+      "loss/hidden": 0.2333984375,
+      "loss/logits": 0.03240815922617912,
+      "loss/reg": 0.026395246386528015,
+      "step": 47
+    },
+    {
+      "epoch": 0.006,
+      "grad_norm": 3.1519744396209717,
+      "grad_norm_var": 2.0420385103816727,
+      "learning_rate": 4.8e-05,
+      "loss": 0.5385,
+      "loss/crossentropy": 2.250037908554077,
+      "loss/hidden": 0.244140625,
+      "loss/logits": 0.03043752908706665,
+      "loss/reg": 0.026393571868538857,
+      "step": 48
+    },
+    {
+      "epoch": 0.006125,
+      "grad_norm": 3.187680244445801,
+      "grad_norm_var": 2.0209109756516943,
+      "learning_rate": 4.9e-05,
+      "loss": 0.5614,
+      "loss/crossentropy": 2.366483688354492,
+      "loss/hidden": 0.263671875,
+      "loss/logits": 0.033859170973300934,
+      "loss/reg": 0.02639181725680828,
+      "step": 49
+    },
+    {
+      "epoch": 0.00625,
+      "grad_norm": 2.3717658519744873,
+      "grad_norm_var": 1.9454730589909708,
+      "learning_rate": 5e-05,
+      "loss": 0.5865,
+      "loss/crossentropy": 2.007732391357422,
+      "loss/hidden": 0.2890625,
+      "loss/logits": 0.0335388109087944,
+      "loss/reg": 0.02638987824320793,
+      "step": 50
+    },
+    {
+      "epoch": 0.006375,
+      "grad_norm": 3.658735990524292,
+      "grad_norm_var": 1.767425501826429,
+      "learning_rate": 5.1000000000000006e-05,
+      "loss": 0.5028,
+      "loss/crossentropy": 2.511072874069214,
+      "loss/hidden": 0.2099609375,
+      "loss/logits": 0.02893088385462761,
+      "loss/reg": 0.026387827470898628,
+      "step": 51
+    },
+    {
+      "epoch": 0.0065,
+      "grad_norm": 2.5912654399871826,
+      "grad_norm_var": 1.582150273328572,
+      "learning_rate": 5.2000000000000004e-05,
+      "loss": 0.5619,
+      "loss/crossentropy": 2.3280093669891357,
+      "loss/hidden": 0.263671875,
+      "loss/logits": 0.03436018154025078,
+      "loss/reg": 0.026385735720396042,
+      "step": 52
+    },
+    {
+      "epoch": 0.006625,
+      "grad_norm": 2.0419421195983887,
+      "grad_norm_var": 0.23432357463535497,
+      "learning_rate": 5.300000000000001e-05,
+      "loss": 0.5674,
+      "loss/crossentropy": 2.3851194381713867,
+      "loss/hidden": 0.263671875,
+      "loss/logits": 0.039869021624326706,
+      "loss/reg": 0.026383817195892334,
+      "step": 53
+    },
+    {
+      "epoch": 0.00675,
+      "grad_norm": 2.4164810180664062,
+      "grad_norm_var": 0.24119551692934707,
+      "learning_rate": 5.4000000000000005e-05,
+      "loss": 0.6087,
+      "loss/crossentropy": 2.6006996631622314,
+      "loss/hidden": 0.296875,
+      "loss/logits": 0.04797635227441788,
+      "loss/reg": 0.026381801813840866,
+      "step": 54
+    },
+    {
+      "epoch": 0.006875,
+      "grad_norm": 2.697831153869629,
+      "grad_norm_var": 0.2135682431387058,
+      "learning_rate": 5.500000000000001e-05,
+      "loss": 0.523,
+      "loss/crossentropy": 2.472208261489868,
+      "loss/hidden": 0.2275390625,
+      "loss/logits": 0.031705208122730255,
+      "loss/reg": 0.026379752904176712,
+      "step": 55
+    },
+    {
+      "epoch": 0.007,
+      "grad_norm": 4.182509422302246,
+      "grad_norm_var": 0.34874494246430365,
+      "learning_rate": 5.6000000000000006e-05,
+      "loss": 0.6766,
+      "loss/crossentropy": 2.693652868270874,
+      "loss/hidden": 0.3671875,
+      "loss/logits": 0.04566050320863724,
+      "loss/reg": 0.02637762948870659,
+      "step": 56
+    },
+    {
+      "epoch": 0.007125,
+      "grad_norm": 2.231238842010498,
+      "grad_norm_var": 0.33990645656106155,
+      "learning_rate": 5.6999999999999996e-05,
+      "loss": 0.5811,
+      "loss/crossentropy": 2.4935543537139893,
+      "loss/hidden": 0.28125,
+      "loss/logits": 0.03613065183162689,
+      "loss/reg": 0.026375366374850273,
+      "step": 57
+    },
+    {
+      "epoch": 0.00725,
+      "grad_norm": 2.0192184448242188,
+      "grad_norm_var": 0.37029866859904437,
+      "learning_rate": 5.8e-05,
+      "loss": 0.5285,
+      "loss/crossentropy": 2.192227840423584,
+      "loss/hidden": 0.2294921875,
+      "loss/logits": 0.03531934320926666,
+      "loss/reg": 0.02637314423918724,
+      "step": 58
+    },
+    {
+      "epoch": 0.007375,
+      "grad_norm": 2.3108532428741455,
+      "grad_norm_var": 0.36699486123436575,
+      "learning_rate": 5.9e-05,
+      "loss": 0.507,
+      "loss/crossentropy": 2.39101243019104,
+      "loss/hidden": 0.2177734375,
+      "loss/logits": 0.025539016351103783,
+      "loss/reg": 0.02637065388262272,
+      "step": 59
+    },
+    {
+      "epoch": 0.0075,
+      "grad_norm": 2.049551486968994,
+      "grad_norm_var": 0.3946811437011318,
+      "learning_rate": 6e-05,
+      "loss": 0.5351,
+      "loss/crossentropy": 2.7062017917633057,
+      "loss/hidden": 0.2353515625,
+      "loss/logits": 0.03605186939239502,
+      "loss/reg": 0.026368385180830956,
+      "step": 60
+    },
+    {
+      "epoch": 0.007625,
+      "grad_norm": 2.6327223777770996,
+      "grad_norm_var": 0.38133460463904284,
+      "learning_rate": 6.1e-05,
+      "loss": 0.5344,
+      "loss/crossentropy": 2.0810956954956055,
+      "loss/hidden": 0.23828125,
+      "loss/logits": 0.03246723860502243,
+      "loss/reg": 0.02636607363820076,
+      "step": 61
+    },
+    {
+      "epoch": 0.00775,
+      "grad_norm": 2.06585955619812,
+      "grad_norm_var": 0.39059185941555535,
+      "learning_rate": 6.2e-05,
+      "loss": 0.5202,
+      "loss/crossentropy": 2.6240835189819336,
+      "loss/hidden": 0.224609375,
+      "loss/logits": 0.03200242295861244,
+      "loss/reg": 0.026363445445895195,
+      "step": 62
+    },
+    {
+      "epoch": 0.007875,
+      "grad_norm": 2.109790563583374,
+      "grad_norm_var": 0.40452198957435875,
+      "learning_rate": 6.3e-05,
+      "loss": 0.5249,
+      "loss/crossentropy": 2.587536573410034,
+      "loss/hidden": 0.2294921875,
+      "loss/logits": 0.03178905323147774,
+      "loss/reg": 0.02636083774268627,
+      "step": 63
+    },
+    {
+      "epoch": 0.008,
+      "grad_norm": 3.818783760070801,
+      "grad_norm_var": 0.48072296241430573,
+      "learning_rate": 6.400000000000001e-05,
+      "loss": 0.6632,
+      "loss/crossentropy": 2.011171817779541,
+      "loss/hidden": 0.353515625,
+      "loss/logits": 0.04613731801509857,
+      "loss/reg": 0.02635800838470459,
+      "step": 64
+    },
+    {
+      "epoch": 0.008125,
+      "grad_norm": 2.4136369228363037,
+      "grad_norm_var": 0.46258887231494605,
+      "learning_rate": 6.500000000000001e-05,
+      "loss": 0.522,
+      "loss/crossentropy": 2.600787401199341,
+      "loss/hidden": 0.224609375,
+      "loss/logits": 0.03384025767445564,
+      "loss/reg": 0.026355121284723282,
+      "step": 65
+    },
+    {
+      "epoch": 0.00825,
+      "grad_norm": 2.3908252716064453,
+      "grad_norm_var": 0.46202963925557394,
+      "learning_rate": 6.6e-05,
+      "loss": 0.5859,
+      "loss/crossentropy": 2.1056201457977295,
+      "loss/hidden": 0.279296875,
+      "loss/logits": 0.04309317469596863,
+      "loss/reg": 0.02635251171886921,
+      "step": 66
+    },
+    {
+      "epoch": 0.008375,
+      "grad_norm": 2.5190653800964355,
+      "grad_norm_var": 0.38262308323354144,
+      "learning_rate": 6.7e-05,
+      "loss": 0.5391,
+      "loss/crossentropy": 2.5527184009552,
+      "loss/hidden": 0.2421875,
+      "loss/logits": 0.0334152951836586,
+      "loss/reg": 0.02634957991540432,
+      "step": 67
+    },
+    {
+      "epoch": 0.0085,
+      "grad_norm": 2.500368595123291,
+      "grad_norm_var": 0.3824057294099087,
+      "learning_rate": 6.800000000000001e-05,
+      "loss": 0.5958,
+      "loss/crossentropy": 2.30499267578125,
+      "loss/hidden": 0.28125,
+      "loss/logits": 0.05108712613582611,
+      "loss/reg": 0.026346800848841667,
+      "step": 68
+    },
+    {
+      "epoch": 0.008625,
+      "grad_norm": 2.7905988693237305,
+      "grad_norm_var": 0.3692126592154902,
+      "learning_rate": 6.9e-05,
+      "loss": 0.6049,
+      "loss/crossentropy": 2.3807146549224854,
+      "loss/hidden": 0.294921875,
+      "loss/logits": 0.04648623988032341,
+      "loss/reg": 0.026344334706664085,
+      "step": 69
+    },
+    {
+      "epoch": 0.00875,
+      "grad_norm": 2.147470235824585,
+      "grad_norm_var": 0.3793077808516782,
+      "learning_rate": 7e-05,
+      "loss": 0.5345,
+      "loss/crossentropy": 2.627505302429199,
+      "loss/hidden": 0.236328125,
+      "loss/logits": 0.034769318997859955,
+      "loss/reg": 0.026341637596488,
+      "step": 70
+    },
+    {
+      "epoch": 0.008875,
+      "grad_norm": 2.6987268924713135,
+      "grad_norm_var": 0.3793248871627156,
+      "learning_rate": 7.1e-05,
+      "loss": 0.5722,
+      "loss/crossentropy": 2.382685899734497,
+      "loss/hidden": 0.271484375,
+      "loss/logits": 0.03735022246837616,
+      "loss/reg": 0.02633870206773281,
+      "step": 71
+    },
+    {
+      "epoch": 0.009,
+      "grad_norm": 3.085085153579712,
+      "grad_norm_var": 0.2164648496489896,
+      "learning_rate": 7.2e-05,
+      "loss": 0.5882,
+      "loss/crossentropy": 2.371429681777954,
+      "loss/hidden": 0.27734375,
+      "loss/logits": 0.04748620092868805,
+      "loss/reg": 0.02633603662252426,
+      "step": 72
+    },
+    {
+      "epoch": 0.009125,
+      "grad_norm": 4.158353328704834,
+      "grad_norm_var": 0.3829897758196862,
+      "learning_rate": 7.3e-05,
+      "loss": 0.8663,
+      "loss/crossentropy": 2.29622745513916,
+      "loss/hidden": 0.5234375,
+      "loss/logits": 0.07955377548933029,
+      "loss/reg": 0.026333071291446686,
+      "step": 73
+    },
+    {
+      "epoch": 0.00925,
+      "grad_norm": 2.111111879348755,
+      "grad_norm_var": 0.37631661688178514,
+      "learning_rate": 7.4e-05,
+      "loss": 0.5468,
+      "loss/crossentropy": 2.29744815826416,
+      "loss/hidden": 0.24609375,
+      "loss/logits": 0.037446070462465286,
+      "loss/reg": 0.026330096647143364,
+      "step": 74
+    },
+    {
+      "epoch": 0.009375,
+      "grad_norm": 2.545919179916382,
+      "grad_norm_var": 0.37031037444480336,
+      "learning_rate": 7.500000000000001e-05,
+      "loss": 0.625,
+      "loss/crossentropy": 2.4376375675201416,
+      "loss/hidden": 0.306640625,
+      "loss/logits": 0.05505819618701935,
+      "loss/reg": 0.026326792314648628,
+      "step": 75
+    },
+    {
+      "epoch": 0.0095,
+      "grad_norm": 2.362215042114258,
+      "grad_norm_var": 0.35233204024674003,
+      "learning_rate": 7.6e-05,
+      "loss": 0.5747,
+      "loss/crossentropy": 2.677924156188965,
+      "loss/hidden": 0.267578125,
+      "loss/logits": 0.04392882436513901,
+      "loss/reg": 0.026323769241571426,
+      "step": 76
+    },
+    {
+      "epoch": 0.009625,
+      "grad_norm": 3.135709762573242,
+      "grad_norm_var": 0.3671929300455114,
+      "learning_rate": 7.7e-05,
+      "loss": 0.7113,
+      "loss/crossentropy": 1.972798466682434,
+      "loss/hidden": 0.384765625,
+      "loss/logits": 0.06329117715358734,
+      "loss/reg": 0.026320943608880043,
+      "step": 77
+    },
+    {
+      "epoch": 0.00975,
+      "grad_norm": 4.418634414672852,
+      "grad_norm_var": 0.5210260544686395,
+      "learning_rate": 7.800000000000001e-05,
+      "loss": 0.6851,
+      "loss/crossentropy": 2.558809518814087,
+      "loss/hidden": 0.357421875,
+      "loss/logits": 0.06447892636060715,
+      "loss/reg": 0.02631756290793419,
+      "step": 78
+    },
+    {
+      "epoch": 0.009875,
+      "grad_norm": 3.9261293411254883,
+      "grad_norm_var": 0.55391532710314,
+      "learning_rate": 7.900000000000001e-05,
+      "loss": 0.5968,
+      "loss/crossentropy": 2.6102137565612793,
+      "loss/hidden": 0.28515625,
+      "loss/logits": 0.04846350848674774,
+      "loss/reg": 0.026314500719308853,
+      "step": 79
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.1532020568847656,
+      "grad_norm_var": 0.5035194586586452,
+      "learning_rate": 8e-05,
+      "loss": 0.7066,
+      "loss/crossentropy": 2.36220121383667,
+      "loss/hidden": 0.38671875,
+      "loss/logits": 0.05672474205493927,
+      "loss/reg": 0.026311254128813744,
+      "step": 80
+    },
+    {
+      "epoch": 0.010125,
+      "grad_norm": 3.557161808013916,
+      "grad_norm_var": 0.5115010248662256,
+      "learning_rate": 8.1e-05,
+      "loss": 0.6647,
+      "loss/crossentropy": 2.412325859069824,
+      "loss/hidden": 0.345703125,
+      "loss/logits": 0.05596970394253731,
+      "loss/reg": 0.026307715103030205,
+      "step": 81
+    },
+    {
+      "epoch": 0.01025,
+      "grad_norm": 2.2158408164978027,
+      "grad_norm_var": 0.5268993015208875,
+      "learning_rate": 8.2e-05,
+      "loss": 0.5575,
+      "loss/crossentropy": 2.3789050579071045,
+      "loss/hidden": 0.255859375,
+      "loss/logits": 0.03855578228831291,
+      "loss/reg": 0.0263040903955698,
+      "step": 82
+    },
+    {
+      "epoch": 0.010375,
+      "grad_norm": 3.2140979766845703,
+      "grad_norm_var": 0.5164286227091848,
+      "learning_rate": 8.3e-05,
+      "loss": 0.5548,
+      "loss/crossentropy": 2.401925802230835,
+      "loss/hidden": 0.2578125,
+      "loss/logits": 0.03401009738445282,
+      "loss/reg": 0.026300577446818352,
+      "step": 83
+    },
+    {
+      "epoch": 0.0105,
+      "grad_norm": 2.4155867099761963,
+      "grad_norm_var": 0.5225404018326155,
+      "learning_rate": 8.4e-05,
+      "loss": 0.5432,
+      "loss/crossentropy": 2.6546974182128906,
+      "loss/hidden": 0.244140625,
+      "loss/logits": 0.03612750768661499,
+      "loss/reg": 0.02629682794213295,
+      "step": 84
+    },
+    {
+      "epoch": 0.010625,
+      "grad_norm": 4.232295036315918,
+      "grad_norm_var": 0.6129643025970267,
+      "learning_rate": 8.5e-05,
+      "loss": 0.7199,
+      "loss/crossentropy": 2.2300524711608887,
+      "loss/hidden": 0.412109375,
+      "loss/logits": 0.044823646545410156,
+      "loss/reg": 0.0262930728495121,
+      "step": 85
+    },
+    {
+      "epoch": 0.01075,
+      "grad_norm": 2.7160282135009766,
+      "grad_norm_var": 0.5620128324129702,
+      "learning_rate": 8.6e-05,
+      "loss": 0.6973,
+      "loss/crossentropy": 2.2953288555145264,
+      "loss/hidden": 0.37109375,
+      "loss/logits": 0.06333646178245544,
+      "loss/reg": 0.02628917805850506,
+      "step": 86
+    },
+    {
+      "epoch": 0.010875,
+      "grad_norm": 3.0872819423675537,
+      "grad_norm_var": 0.5495392294868544,
+      "learning_rate": 8.7e-05,
+      "loss": 0.636,
+      "loss/crossentropy": 2.449223756790161,
+      "loss/hidden": 0.328125,
+      "loss/logits": 0.045011188834905624,
+      "loss/reg": 0.02628495544195175,
+      "step": 87
+    },
+    {
+      "epoch": 0.011,
+      "grad_norm": 2.6966607570648193,
+      "grad_norm_var": 0.5621192378124493,
+      "learning_rate": 8.800000000000001e-05,
+      "loss": 0.6272,
+      "loss/crossentropy": 2.5449047088623047,
+      "loss/hidden": 0.31640625,
+      "loss/logits": 0.04802623763680458,
+      "loss/reg": 0.026280568912625313,
+      "step": 88
+    },
+    {
+      "epoch": 0.011125,
+      "grad_norm": 2.9160921573638916,
+      "grad_norm_var": 0.48685408890983506,
+      "learning_rate": 8.900000000000001e-05,
+      "loss": 0.6278,
+      "loss/crossentropy": 2.1442108154296875,
+      "loss/hidden": 0.310546875,
+      "loss/logits": 0.054462507367134094,
+      "loss/reg": 0.02627684734761715,
+      "step": 89
+    },
+    {
+      "epoch": 0.01125,
+      "grad_norm": 3.3378944396972656,
+      "grad_norm_var": 0.4283231906687052,
+      "learning_rate": 9e-05,
+      "loss": 0.6536,
+      "loss/crossentropy": 2.4361062049865723,
+      "loss/hidden": 0.349609375,
+      "loss/logits": 0.04130454361438751,
+      "loss/reg": 0.026273205876350403,
+      "step": 90
+    },
+    {
+      "epoch": 0.011375,
+      "grad_norm": 2.597607374191284,
+      "grad_norm_var": 0.42452911296148627,
+      "learning_rate": 9.1e-05,
+      "loss": 0.6391,
+      "loss/crossentropy": 2.0079762935638428,
+      "loss/hidden": 0.3203125,
+      "loss/logits": 0.056107863783836365,
+      "loss/reg": 0.02626909501850605,
+      "step": 91
+    },
+    {
+      "epoch": 0.0115,
+      "grad_norm": 4.960971355438232,
+      "grad_norm_var": 0.5826997127177641,
+      "learning_rate": 9.200000000000001e-05,
+      "loss": 0.6735,
+      "loss/crossentropy": 2.708275079727173,
+      "loss/hidden": 0.349609375,
+      "loss/logits": 0.06119866296648979,
+      "loss/reg": 0.026265164837241173,
+      "step": 92
+    },
+    {
+      "epoch": 0.011625,
+      "grad_norm": 3.8193323612213135,
+      "grad_norm_var": 0.5981799563928756,
+      "learning_rate": 9.300000000000001e-05,
+      "loss": 0.8429,
+      "loss/crossentropy": 2.4117016792297363,
+      "loss/hidden": 0.498046875,
+      "loss/logits": 0.08221981674432755,
+      "loss/reg": 0.026260720565915108,
+      "step": 93
+    },
+    {
+      "epoch": 0.01175,
+      "grad_norm": 3.434213638305664,
+      "grad_norm_var": 0.5157332557296352,
+      "learning_rate": 9.4e-05,
+      "loss": 0.6738,
+      "loss/crossentropy": 2.62178111076355,
+      "loss/hidden": 0.36328125,
+      "loss/logits": 0.04794853553175926,
+      "loss/reg": 0.02625615894794464,
+      "step": 94
+    },
+    {
+      "epoch": 0.011875,
+      "grad_norm": 3.0944480895996094,
+      "grad_norm_var": 0.4859309000509171,
+      "learning_rate": 9.5e-05,
+      "loss": 0.6829,
+      "loss/crossentropy": 2.582462787628174,
+      "loss/hidden": 0.353515625,
+      "loss/logits": 0.0669020414352417,
+      "loss/reg": 0.026251958683133125,
+      "step": 95
+    },
+    {
+      "epoch": 0.012,
+      "grad_norm": 3.9548256397247314,
+      "grad_norm_var": 0.5194300484800329,
+      "learning_rate": 9.6e-05,
+      "loss": 0.8508,
+      "loss/crossentropy": 2.3243165016174316,
+      "loss/hidden": 0.49609375,
+      "loss/logits": 0.09221720695495605,
+      "loss/reg": 0.026247689500451088,
+      "step": 96
+    },
+    {
+      "epoch": 0.012125,
+      "grad_norm": 8.949115753173828,
+      "grad_norm_var": 2.5460815450669125,
+      "learning_rate": 9.7e-05,
+      "loss": 0.9668,
+      "loss/crossentropy": 2.3593697547912598,
+      "loss/hidden": 0.62109375,
+      "loss/logits": 0.08330727368593216,
+      "loss/reg": 0.026243869215250015,
+      "step": 97
+    },
+    {
+      "epoch": 0.01225,
+      "grad_norm": 3.874511957168579,
+      "grad_norm_var": 2.411331023611861,
+      "learning_rate": 9.8e-05,
+      "loss": 0.7272,
+      "loss/crossentropy": 1.9933784008026123,
+      "loss/hidden": 0.41015625,
+      "loss/logits": 0.05464401841163635,
+      "loss/reg": 0.026239972561597824,
+      "step": 98
+    },
+    {
+      "epoch": 0.012375,
+      "grad_norm": 5.088143825531006,
+      "grad_norm_var": 2.507843574169987,
+      "learning_rate": 9.900000000000001e-05,
+      "loss": 0.6761,
+      "loss/crossentropy": 2.578767776489258,
+      "loss/hidden": 0.3515625,
+      "loss/logits": 0.0621890164911747,
+      "loss/reg": 0.026235179975628853,
+      "step": 99
+    },
+    {
+      "epoch": 0.0125,
+      "grad_norm": 3.9010627269744873,
+      "grad_norm_var": 2.366914585761602,
+      "learning_rate": 0.0001,
+      "loss": 0.7051,
+      "loss/crossentropy": 2.4717133045196533,
+      "loss/hidden": 0.375,
+      "loss/logits": 0.0677795261144638,
+      "loss/reg": 0.026230769231915474,
+      "step": 100
+    },
+    {
+      "epoch": 0.012625,
+      "grad_norm": 5.50706148147583,
+      "grad_norm_var": 2.522191588171327,
+      "learning_rate": 0.0001,
+      "loss": 0.7765,
+      "loss/crossentropy": 2.3764612674713135,
+      "loss/hidden": 0.44921875,
+      "loss/logits": 0.06503438949584961,
+      "loss/reg": 0.02622627653181553,
+      "step": 101
+    },
+    {
+      "epoch": 0.01275,
+      "grad_norm": 5.103200435638428,
+      "grad_norm_var": 2.470966679212188,
+      "learning_rate": 0.0001,
+      "loss": 0.7008,
+      "loss/crossentropy": 2.5796542167663574,
+      "loss/hidden": 0.38671875,
+      "loss/logits": 0.05191829800605774,
+      "loss/reg": 0.026221245527267456,
+      "step": 102
+    },
+    {
+      "epoch": 0.012875,
+      "grad_norm": 18.05303192138672,
+      "grad_norm_var": 14.358413039824521,
+      "learning_rate": 0.0001,
+      "loss": 1.0008,
+      "loss/crossentropy": 1.927337646484375,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.05896752327680588,
+      "loss/reg": 0.02621658518910408,
+      "step": 103
+    },
+    {
+      "epoch": 0.013,
+      "grad_norm": 3.410438299179077,
+      "grad_norm_var": 14.16338361533652,
+      "learning_rate": 0.0001,
+      "loss": 0.735,
+      "loss/crossentropy": 2.290928363800049,
+      "loss/hidden": 0.40625,
+      "loss/logits": 0.06666909158229828,
+      "loss/reg": 0.026211561635136604,
+      "step": 104
+    },
+    {
+      "epoch": 0.013125,
+      "grad_norm": 3.117622137069702,
+      "grad_norm_var": 14.10656391346422,
+      "learning_rate": 0.0001,
+      "loss": 0.6665,
+      "loss/crossentropy": 2.6549246311187744,
+      "loss/hidden": 0.353515625,
+      "loss/logits": 0.05095440149307251,
+      "loss/reg": 0.026206739246845245,
+      "step": 105
+    },
+    {
+      "epoch": 0.01325,
+      "grad_norm": 3.9999241828918457,
+      "grad_norm_var": 13.97508509706009,
+      "learning_rate": 0.0001,
+      "loss": 0.8082,
+      "loss/crossentropy": 2.460174798965454,
+      "loss/hidden": 0.46484375,
+      "loss/logits": 0.08136071264743805,
+      "loss/reg": 0.02620157040655613,
+      "step": 106
+    },
+    {
+      "epoch": 0.013375,
+      "grad_norm": 3.405712842941284,
+      "grad_norm_var": 13.73775124044489,
+      "learning_rate": 0.0001,
+      "loss": 0.6518,
+      "loss/crossentropy": 2.521803855895996,
+      "loss/hidden": 0.3359375,
+      "loss/logits": 0.053909383714199066,
+      "loss/reg": 0.02619684301316738,
+      "step": 107
+    },
+    {
+      "epoch": 0.0135,
+      "grad_norm": 3.615098237991333,
+      "grad_norm_var": 13.899167673014775,
+      "learning_rate": 0.0001,
+      "loss": 0.7068,
+      "loss/crossentropy": 2.510159969329834,
+      "loss/hidden": 0.37890625,
+      "loss/logits": 0.06601101160049438,
+      "loss/reg": 0.02619197592139244,
+      "step": 108
+    },
+    {
+      "epoch": 0.013625,
+      "grad_norm": 4.2520599365234375,
+      "grad_norm_var": 13.834356012442765,
+      "learning_rate": 0.0001,
+      "loss": 0.735,
+      "loss/crossentropy": 2.508683681488037,
+      "loss/hidden": 0.41015625,
+      "loss/logits": 0.0629870742559433,
+      "loss/reg": 0.0261868704110384,
+      "step": 109
+    },
+    {
+      "epoch": 0.01375,
+      "grad_norm": 3.215749979019165,
+      "grad_norm_var": 13.887973421518442,
+      "learning_rate": 0.0001,
+      "loss": 0.8268,
+      "loss/crossentropy": 2.3564252853393555,
+      "loss/hidden": 0.48046875,
+      "loss/logits": 0.08449074625968933,
+      "loss/reg": 0.026181429624557495,
+      "step": 110
+    },
+    {
+      "epoch": 0.013875,
+      "grad_norm": 4.598328590393066,
+      "grad_norm_var": 13.615373346458785,
+      "learning_rate": 0.0001,
+      "loss": 0.7398,
+      "loss/crossentropy": 2.366943359375,
+      "loss/hidden": 0.41796875,
+      "loss/logits": 0.060083672404289246,
+      "loss/reg": 0.026176555082201958,
+      "step": 111
+    },
+    {
+      "epoch": 0.014,
+      "grad_norm": 2.758070707321167,
+      "grad_norm_var": 13.912012390229316,
+      "learning_rate": 0.0001,
+      "loss": 0.6839,
+      "loss/crossentropy": 2.3351521492004395,
+      "loss/hidden": 0.365234375,
+      "loss/logits": 0.056987129151821136,
+      "loss/reg": 0.0261719711124897,
+      "step": 112
+    },
+    {
+      "epoch": 0.014125,
+      "grad_norm": 2.9389584064483643,
+      "grad_norm_var": 13.147693721903025,
+      "learning_rate": 0.0001,
+      "loss": 0.8964,
+      "loss/crossentropy": 2.188626766204834,
+      "loss/hidden": 0.54296875,
+      "loss/logits": 0.09171397984027863,
+      "loss/reg": 0.026167072355747223,
+      "step": 113
+    },
+    {
+      "epoch": 0.01425,
+      "grad_norm": 2.8545026779174805,
+      "grad_norm_var": 13.338918720074266,
+      "learning_rate": 0.0001,
+      "loss": 0.6652,
+      "loss/crossentropy": 2.488462448120117,
+      "loss/hidden": 0.341796875,
+      "loss/logits": 0.06177069991827011,
+      "loss/reg": 0.026162149384617805,
+      "step": 114
+    },
+    {
+      "epoch": 0.014375,
+      "grad_norm": 3.343590497970581,
+      "grad_norm_var": 13.447848849906688,
+      "learning_rate": 0.0001,
+      "loss": 0.7317,
+      "loss/crossentropy": 2.4826672077178955,
+      "loss/hidden": 0.396484375,
+      "loss/logits": 0.07369040697813034,
+      "loss/reg": 0.026156950742006302,
+      "step": 115
+    },
+    {
+      "epoch": 0.0145,
+      "grad_norm": 5.309541702270508,
+      "grad_norm_var": 13.435010363164546,
+      "learning_rate": 0.0001,
+      "loss": 0.6918,
+      "loss/crossentropy": 2.715517282485962,
+      "loss/hidden": 0.376953125,
+      "loss/logits": 0.05337735265493393,
+      "loss/reg": 0.026151426136493683,
+      "step": 116
+    },
+    {
+      "epoch": 0.014625,
+      "grad_norm": 3.413027763366699,
+      "grad_norm_var": 13.488672790501717,
+      "learning_rate": 0.0001,
+      "loss": 0.7942,
+      "loss/crossentropy": 2.3932089805603027,
+      "loss/hidden": 0.451171875,
+      "loss/logits": 0.08154396712779999,
+      "loss/reg": 0.026146216318011284,
+      "step": 117
+    },
+    {
+      "epoch": 0.01475,
+      "grad_norm": 2.735275983810425,
+      "grad_norm_var": 13.676075950246783,
+      "learning_rate": 0.0001,
+      "loss": 0.7606,
+      "loss/crossentropy": 2.2933082580566406,
+      "loss/hidden": 0.4296875,
+      "loss/logits": 0.06949938833713531,
+      "loss/reg": 0.02614082768559456,
+      "step": 118
+    },
+    {
+      "epoch": 0.014875,
+      "grad_norm": 3.1346964836120605,
+      "grad_norm_var": 0.5056645529599865,
+      "learning_rate": 0.0001,
+      "loss": 0.7759,
+      "loss/crossentropy": 2.331713914871216,
+      "loss/hidden": 0.4375,
+      "loss/logits": 0.07704727351665497,
+      "loss/reg": 0.026135168969631195,
+      "step": 119
+    },
+    {
+      "epoch": 0.015,
+      "grad_norm": 3.8077635765075684,
+      "grad_norm_var": 0.5104468723684629,
+      "learning_rate": 0.0001,
+      "loss": 0.6882,
+      "loss/crossentropy": 2.331220865249634,
+      "loss/hidden": 0.373046875,
+      "loss/logits": 0.05386776477098465,
+      "loss/reg": 0.02612963318824768,
+      "step": 120
+    },
+    {
+      "epoch": 0.015125,
+      "grad_norm": 3.209914445877075,
+      "grad_norm_var": 0.5058893418769751,
+      "learning_rate": 0.0001,
+      "loss": 0.7536,
+      "loss/crossentropy": 2.352771759033203,
+      "loss/hidden": 0.4140625,
+      "loss/logits": 0.07825946807861328,
+      "loss/reg": 0.026124266907572746,
+      "step": 121
+    },
+    {
+      "epoch": 0.01525,
+      "grad_norm": 3.3548500537872314,
+      "grad_norm_var": 0.49208198737674874,
+      "learning_rate": 0.0001,
+      "loss": 0.7088,
+      "loss/crossentropy": 2.483644723892212,
+      "loss/hidden": 0.384765625,
+      "loss/logits": 0.06287863850593567,
+      "loss/reg": 0.0261182002723217,
+      "step": 122
+    },
+    {
+      "epoch": 0.015375,
+      "grad_norm": 3.9953765869140625,
+      "grad_norm_var": 0.5066601541023895,
+      "learning_rate": 0.0001,
+      "loss": 0.7792,
+      "loss/crossentropy": 2.6117637157440186,
+      "loss/hidden": 0.435546875,
+      "loss/logits": 0.08250629901885986,
+      "loss/reg": 0.026112213730812073,
+      "step": 123
+    },
+    {
+      "epoch": 0.0155,
+      "grad_norm": 3.1783852577209473,
+      "grad_norm_var": 0.5138316405800327,
+      "learning_rate": 0.0001,
+      "loss": 0.7559,
+      "loss/crossentropy": 2.401679754257202,
+      "loss/hidden": 0.423828125,
+      "loss/logits": 0.07104581594467163,
+      "loss/reg": 0.02610679157078266,
+      "step": 124
+    },
+    {
+      "epoch": 0.015625,
+      "grad_norm": 3.2759885787963867,
+      "grad_norm_var": 0.47631527116517774,
+      "learning_rate": 0.0001,
+      "loss": 0.8262,
+      "loss/crossentropy": 2.3979361057281494,
+      "loss/hidden": 0.486328125,
+      "loss/logits": 0.0788530558347702,
+      "loss/reg": 0.026101654395461082,
+      "step": 125
+    },
+    {
+      "epoch": 0.01575,
+      "grad_norm": 4.0768632888793945,
+      "grad_norm_var": 0.4963098069624029,
+      "learning_rate": 0.0001,
+      "loss": 0.6983,
+      "loss/crossentropy": 2.5287113189697266,
+      "loss/hidden": 0.375,
+      "loss/logits": 0.062308911234140396,
+      "loss/reg": 0.026096193119883537,
+      "step": 126
+    },
+    {
+      "epoch": 0.015875,
+      "grad_norm": 4.300101280212402,
+      "grad_norm_var": 0.45815803943682926,
+      "learning_rate": 0.0001,
+      "loss": 0.858,
+      "loss/crossentropy": 2.255234956741333,
+      "loss/hidden": 0.50390625,
+      "loss/logits": 0.0932290330529213,
+      "loss/reg": 0.026090849190950394,
+      "step": 127
+    },
+    {
+      "epoch": 0.016,
+      "grad_norm": 3.303663492202759,
+      "grad_norm_var": 0.42421384752866137,
+      "learning_rate": 0.0001,
+      "loss": 0.7284,
+      "loss/crossentropy": 2.528862476348877,
+      "loss/hidden": 0.40234375,
+      "loss/logits": 0.06523742526769638,
+      "loss/reg": 0.026085302233695984,
+      "step": 128
+    },
+    {
+      "epoch": 0.016125,
+      "grad_norm": 6.868241310119629,
+      "grad_norm_var": 1.0876227157335427,
+      "learning_rate": 0.0001,
+      "loss": 0.8999,
+      "loss/crossentropy": 2.6554996967315674,
+      "loss/hidden": 0.5390625,
+      "loss/logits": 0.1000661626458168,
+      "loss/reg": 0.02607985958456993,
+      "step": 129
+    },
+    {
+      "epoch": 0.01625,
+      "grad_norm": 3.3035075664520264,
+      "grad_norm_var": 1.046006684658701,
+      "learning_rate": 0.0001,
+      "loss": 0.7367,
+      "loss/crossentropy": 2.293642282485962,
+      "loss/hidden": 0.40625,
+      "loss/logits": 0.06973426043987274,
+      "loss/reg": 0.026074659079313278,
+      "step": 130
+    },
+    {
+      "epoch": 0.016375,
+      "grad_norm": 4.2563276290893555,
+      "grad_norm_var": 1.0439696727840135,
+      "learning_rate": 0.0001,
+      "loss": 0.7754,
+      "loss/crossentropy": 2.2192564010620117,
+      "loss/hidden": 0.447265625,
+      "loss/logits": 0.06743350625038147,
+      "loss/reg": 0.026069074869155884,
+      "step": 131
+    },
+    {
+      "epoch": 0.0165,
+      "grad_norm": 4.646778583526611,
+      "grad_norm_var": 0.9420233457711596,
+      "learning_rate": 0.0001,
+      "loss": 0.7496,
+      "loss/crossentropy": 2.430368423461914,
+      "loss/hidden": 0.4140625,
+      "loss/logits": 0.07494455575942993,
+      "loss/reg": 0.026063458994030952,
+      "step": 132
+    },
+    {
+      "epoch": 0.016625,
+      "grad_norm": 7.465832233428955,
+      "grad_norm_var": 1.7574380087301391,
+      "learning_rate": 0.0001,
+      "loss": 0.9607,
+      "loss/crossentropy": 2.6182897090911865,
+      "loss/hidden": 0.57421875,
+      "loss/logits": 0.1258610337972641,
+      "loss/reg": 0.026057813316583633,
+      "step": 133
+    },
+    {
+      "epoch": 0.01675,
+      "grad_norm": 4.5479936599731445,
+      "grad_norm_var": 1.6433309350178509,
+      "learning_rate": 0.0001,
+      "loss": 0.9529,
+      "loss/crossentropy": 2.238551378250122,
+      "loss/hidden": 0.58984375,
+      "loss/logits": 0.10252824425697327,
+      "loss/reg": 0.02605200558900833,
+      "step": 134
+    },
+    {
+      "epoch": 0.016875,
+      "grad_norm": 3.4055774211883545,
+      "grad_norm_var": 1.6105102483452751,
+      "learning_rate": 0.0001,
+      "loss": 0.8407,
+      "loss/crossentropy": 2.4216599464416504,
+      "loss/hidden": 0.498046875,
+      "loss/logits": 0.08216647803783417,
+      "loss/reg": 0.026046328246593475,
+      "step": 135
+    },
+    {
+      "epoch": 0.017,
+      "grad_norm": 8.839641571044922,
+      "grad_norm_var": 2.938344740358995,
+      "learning_rate": 0.0001,
+      "loss": 1.1023,
+      "loss/crossentropy": 2.534442901611328,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.1310025304555893,
+      "loss/reg": 0.026040658354759216,
+      "step": 136
+    },
+    {
+      "epoch": 0.017125,
+      "grad_norm": 4.6421589851379395,
+      "grad_norm_var": 2.819843479450094,
+      "learning_rate": 0.0001,
+      "loss": 0.7418,
+      "loss/crossentropy": 2.605559825897217,
+      "loss/hidden": 0.416015625,
+      "loss/logits": 0.06548085808753967,
+      "loss/reg": 0.02603481523692608,
+      "step": 137
+    },
+    {
+      "epoch": 0.01725,
+      "grad_norm": 3.1547701358795166,
+      "grad_norm_var": 2.8553314644504555,
+      "learning_rate": 0.0001,
+      "loss": 0.731,
+      "loss/crossentropy": 2.5905325412750244,
+      "loss/hidden": 0.40234375,
+      "loss/logits": 0.06835847347974777,
+      "loss/reg": 0.02602926455438137,
+      "step": 138
+    },
+    {
+      "epoch": 0.017375,
+      "grad_norm": 4.074351787567139,
+      "grad_norm_var": 2.849577549707145,
+      "learning_rate": 0.0001,
+      "loss": 0.9582,
+      "loss/crossentropy": 2.1483733654022217,
+      "loss/hidden": 0.60546875,
+      "loss/logits": 0.0924658477306366,
+      "loss/reg": 0.026023706421256065,
+      "step": 139
+    },
+    {
+      "epoch": 0.0175,
+      "grad_norm": 3.758636713027954,
+      "grad_norm_var": 2.7618912420839155,
+      "learning_rate": 0.0001,
+      "loss": 0.9996,
+      "loss/crossentropy": 2.335742473602295,
+      "loss/hidden": 0.6328125,
+      "loss/logits": 0.10663188993930817,
+      "loss/reg": 0.026017924770712852,
+      "step": 140
+    },
+    {
+      "epoch": 0.017625,
+      "grad_norm": 4.186927795410156,
+      "grad_norm_var": 2.6505093919275544,
+      "learning_rate": 0.0001,
+      "loss": 0.8414,
+      "loss/crossentropy": 2.281843423843384,
+      "loss/hidden": 0.498046875,
+      "loss/logits": 0.08321215212345123,
+      "loss/reg": 0.026011699810624123,
+      "step": 141
+    },
+    {
+      "epoch": 0.01775,
+      "grad_norm": 3.1666276454925537,
+      "grad_norm_var": 2.775123140671519,
+      "learning_rate": 0.0001,
+      "loss": 0.7768,
+      "loss/crossentropy": 2.4267935752868652,
+      "loss/hidden": 0.44140625,
+      "loss/logits": 0.07538889348506927,
+      "loss/reg": 0.02600528486073017,
+      "step": 142
+    },
+    {
+      "epoch": 0.017875,
+      "grad_norm": 6.386529445648193,
+      "grad_norm_var": 2.9581845034072245,
+      "learning_rate": 0.0001,
+      "loss": 0.8401,
+      "loss/crossentropy": 2.7848174571990967,
+      "loss/hidden": 0.50390625,
+      "loss/logits": 0.07621172070503235,
+      "loss/reg": 0.025999369099736214,
+      "step": 143
+    },
+    {
+      "epoch": 0.018,
+      "grad_norm": 3.8083512783050537,
+      "grad_norm_var": 2.876745593692829,
+      "learning_rate": 0.0001,
+      "loss": 0.9046,
+      "loss/crossentropy": 2.341048002243042,
+      "loss/hidden": 0.56640625,
+      "loss/logits": 0.0782276839017868,
+      "loss/reg": 0.0259928647428751,
+      "step": 144
+    },
+    {
+      "epoch": 0.018125,
+      "grad_norm": 4.083465576171875,
+      "grad_norm_var": 2.5868089188751657,
+      "learning_rate": 0.0001,
+      "loss": 0.9449,
+      "loss/crossentropy": 2.3942995071411133,
+      "loss/hidden": 0.59375,
+      "loss/logits": 0.09130540490150452,
+      "loss/reg": 0.025986921042203903,
+      "step": 145
+    },
+    {
+      "epoch": 0.01825,
+      "grad_norm": 3.654815673828125,
+      "grad_norm_var": 2.533420197907486,
+      "learning_rate": 0.0001,
+      "loss": 0.9459,
+      "loss/crossentropy": 2.2414467334747314,
+      "loss/hidden": 0.5859375,
+      "loss/logits": 0.10016702860593796,
+      "loss/reg": 0.02598092146217823,
+      "step": 146
+    },
+    {
+      "epoch": 0.018375,
+      "grad_norm": 5.4976935386657715,
+      "grad_norm_var": 2.567896035243579,
+      "learning_rate": 0.0001,
+      "loss": 0.9079,
+      "loss/crossentropy": 2.1185100078582764,
+      "loss/hidden": 0.58203125,
+      "loss/logits": 0.06617112457752228,
+      "loss/reg": 0.025974513962864876,
+      "step": 147
+    },
+    {
+      "epoch": 0.0185,
+      "grad_norm": 3.4107933044433594,
+      "grad_norm_var": 2.673383097164577,
+      "learning_rate": 0.0001,
+      "loss": 0.7982,
+      "loss/crossentropy": 2.417313575744629,
+      "loss/hidden": 0.462890625,
+      "loss/logits": 0.07567355036735535,
+      "loss/reg": 0.025968506932258606,
+      "step": 148
+    },
+    {
+      "epoch": 0.018625,
+      "grad_norm": 3.589749574661255,
+      "grad_norm_var": 2.1469293827286418,
+      "learning_rate": 0.0001,
+      "loss": 0.8088,
+      "loss/crossentropy": 2.2554194927215576,
+      "loss/hidden": 0.478515625,
+      "loss/logits": 0.07071521133184433,
+      "loss/reg": 0.025961775332689285,
+      "step": 149
+    },
+    {
+      "epoch": 0.01875,
+      "grad_norm": 4.003805160522461,
+      "grad_norm_var": 2.1538296896943887,
+      "learning_rate": 0.0001,
+      "loss": 0.7856,
+      "loss/crossentropy": 2.7128918170928955,
+      "loss/hidden": 0.453125,
+      "loss/logits": 0.07290500402450562,
+      "loss/reg": 0.025955306366086006,
+      "step": 150
+    },
+    {
+      "epoch": 0.018875,
+      "grad_norm": 3.536449432373047,
+      "grad_norm_var": 2.1383506752067736,
+      "learning_rate": 0.0001,
+      "loss": 0.7436,
+      "loss/crossentropy": 2.509995460510254,
+      "loss/hidden": 0.421875,
+      "loss/logits": 0.062191903591156006,
+      "loss/reg": 0.02594931609928608,
+      "step": 151
+    },
+    {
+      "epoch": 0.019,
+      "grad_norm": 3.434654951095581,
+      "grad_norm_var": 0.7374638182579057,
+      "learning_rate": 0.0001,
+      "loss": 0.9118,
+      "loss/crossentropy": 2.3447437286376953,
+      "loss/hidden": 0.55859375,
+      "loss/logits": 0.0937500149011612,
+      "loss/reg": 0.025943227112293243,
+      "step": 152
+    },
+    {
+      "epoch": 0.019125,
+      "grad_norm": 8.066261291503906,
+      "grad_norm_var": 1.7522972641868202,
+      "learning_rate": 0.0001,
+      "loss": 0.8648,
+      "loss/crossentropy": 2.4481232166290283,
+      "loss/hidden": 0.52734375,
+      "loss/logits": 0.07811163365840912,
+      "loss/reg": 0.025936946272850037,
+      "step": 153
+    },
+    {
+      "epoch": 0.01925,
+      "grad_norm": 3.2214348316192627,
+      "grad_norm_var": 1.7429433318934864,
+      "learning_rate": 0.0001,
+      "loss": 0.7673,
+      "loss/crossentropy": 2.3142549991607666,
+      "loss/hidden": 0.43359375,
+      "loss/logits": 0.07435894012451172,
+      "loss/reg": 0.02593095973134041,
+      "step": 154
+    },
+    {
+      "epoch": 0.019375,
+      "grad_norm": 2.854038715362549,
+      "grad_norm_var": 1.8633807825236384,
+      "learning_rate": 0.0001,
+      "loss": 0.7601,
+      "loss/crossentropy": 2.7022647857666016,
+      "loss/hidden": 0.43359375,
+      "loss/logits": 0.06724615395069122,
+      "loss/reg": 0.025924943387508392,
+      "step": 155
+    },
+    {
+      "epoch": 0.0195,
+      "grad_norm": 3.1763484477996826,
+      "grad_norm_var": 1.9162196068123232,
+      "learning_rate": 0.0001,
+      "loss": 0.9,
+      "loss/crossentropy": 2.574676036834717,
+      "loss/hidden": 0.54296875,
+      "loss/logits": 0.09785018861293793,
+      "loss/reg": 0.025919148698449135,
+      "step": 156
+    },
+    {
+      "epoch": 0.019625,
+      "grad_norm": 3.999523162841797,
+      "grad_norm_var": 1.9169889601133074,
+      "learning_rate": 0.0001,
+      "loss": 0.858,
+      "loss/crossentropy": 2.480532646179199,
+      "loss/hidden": 0.515625,
+      "loss/logits": 0.08323468267917633,
+      "loss/reg": 0.025913061574101448,
+      "step": 157
+    },
+    {
+      "epoch": 0.01975,
+      "grad_norm": 2.9662230014801025,
+      "grad_norm_var": 1.944924590139983,
+      "learning_rate": 0.0001,
+      "loss": 0.818,
+      "loss/crossentropy": 2.459967613220215,
+      "loss/hidden": 0.48046875,
+      "loss/logits": 0.07847169041633606,
+      "loss/reg": 0.025906959548592567,
+      "step": 158
+    },
+    {
+      "epoch": 0.019875,
+      "grad_norm": 7.074191093444824,
+      "grad_norm_var": 2.1836107796529065,
+      "learning_rate": 0.0001,
+      "loss": 0.9801,
+      "loss/crossentropy": 2.2858352661132812,
+      "loss/hidden": 0.6484375,
+      "loss/logits": 0.07268328964710236,
+      "loss/reg": 0.025901462882757187,
+      "step": 159
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.6333370208740234,
+      "grad_norm_var": 2.193465227977892,
+      "learning_rate": 0.0001,
+      "loss": 0.9048,
+      "loss/crossentropy": 2.4626388549804688,
+      "loss/hidden": 0.5546875,
+      "loss/logits": 0.09115847945213318,
+      "loss/reg": 0.02589540183544159,
+      "step": 160
+    },
+    {
+      "epoch": 0.020125,
+      "grad_norm": 4.283749103546143,
+      "grad_norm_var": 2.1945247126451437,
+      "learning_rate": 0.0001,
+      "loss": 0.9251,
+      "loss/crossentropy": 2.5746121406555176,
+      "loss/hidden": 0.58203125,
+      "loss/logits": 0.0841851755976677,
+      "loss/reg": 0.025888830423355103,
+      "step": 161
+    },
+    {
+      "epoch": 0.02025,
+      "grad_norm": 3.628138542175293,
+      "grad_norm_var": 2.196331220420876,
+      "learning_rate": 0.0001,
+      "loss": 0.9408,
+      "loss/crossentropy": 2.398010730743408,
+      "loss/hidden": 0.58203125,
+      "loss/logits": 0.09991887211799622,
+      "loss/reg": 0.025882074609398842,
+      "step": 162
+    },
+    {
+      "epoch": 0.020375,
+      "grad_norm": 3.5412559509277344,
+      "grad_norm_var": 2.0836172065033507,
+      "learning_rate": 0.0001,
+      "loss": 0.857,
+      "loss/crossentropy": 2.5064220428466797,
+      "loss/hidden": 0.50390625,
+      "loss/logits": 0.0943569540977478,
+      "loss/reg": 0.025876009836792946,
+      "step": 163
+    },
+    {
+      "epoch": 0.0205,
+      "grad_norm": 3.498668670654297,
+      "grad_norm_var": 2.0768887394910185,
+      "learning_rate": 0.0001,
+      "loss": 0.8917,
+      "loss/crossentropy": 2.535529375076294,
+      "loss/hidden": 0.52734375,
+      "loss/logits": 0.10565976053476334,
+      "loss/reg": 0.02587023191154003,
+      "step": 164
+    },
+    {
+      "epoch": 0.020625,
+      "grad_norm": 3.14025616645813,
+      "grad_norm_var": 2.116006039378421,
+      "learning_rate": 0.0001,
+      "loss": 0.7816,
+      "loss/crossentropy": 2.7250640392303467,
+      "loss/hidden": 0.453125,
+      "loss/logits": 0.06980661302804947,
+      "loss/reg": 0.025863803923130035,
+      "step": 165
+    },
+    {
+      "epoch": 0.02075,
+      "grad_norm": 2.6821706295013428,
+      "grad_norm_var": 2.2251478520018773,
+      "learning_rate": 0.0001,
+      "loss": 0.7558,
+      "loss/crossentropy": 2.230567693710327,
+      "loss/hidden": 0.43359375,
+      "loss/logits": 0.06358660757541656,
+      "loss/reg": 0.02585742622613907,
+      "step": 166
+    },
+    {
+      "epoch": 0.020875,
+      "grad_norm": 3.8048856258392334,
+      "grad_norm_var": 2.2158862694911607,
+      "learning_rate": 0.0001,
+      "loss": 0.7965,
+      "loss/crossentropy": 2.69014310836792,
+      "loss/hidden": 0.453125,
+      "loss/logits": 0.08488957583904266,
+      "loss/reg": 0.025850988924503326,
+      "step": 167
+    },
+    {
+      "epoch": 0.021,
+      "grad_norm": 4.305826187133789,
+      "grad_norm_var": 2.2048741298976515,
+      "learning_rate": 0.0001,
+      "loss": 1.1009,
+      "loss/crossentropy": 2.2925851345062256,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.10412566363811493,
+      "loss/reg": 0.025845136493444443,
+      "step": 168
+    },
+    {
+      "epoch": 0.021125,
+      "grad_norm": 4.051706790924072,
+      "grad_norm_var": 1.0314628897999674,
+      "learning_rate": 0.0001,
+      "loss": 0.7756,
+      "loss/crossentropy": 2.4515578746795654,
+      "loss/hidden": 0.4453125,
+      "loss/logits": 0.07186460494995117,
+      "loss/reg": 0.02583896555006504,
+      "step": 169
+    },
+    {
+      "epoch": 0.02125,
+      "grad_norm": 3.3800575733184814,
+      "grad_norm_var": 1.022039210437893,
+      "learning_rate": 0.0001,
+      "loss": 0.798,
+      "loss/crossentropy": 2.3387436866760254,
+      "loss/hidden": 0.455078125,
+      "loss/logits": 0.08458675444126129,
+      "loss/reg": 0.025832952931523323,
+      "step": 170
+    },
+    {
+      "epoch": 0.021375,
+      "grad_norm": 3.069735527038574,
+      "grad_norm_var": 0.9991429378891439,
+      "learning_rate": 0.0001,
+      "loss": 0.7416,
+      "loss/crossentropy": 2.660727024078369,
+      "loss/hidden": 0.412109375,
+      "loss/logits": 0.07125158607959747,
+      "loss/reg": 0.025827286764979362,
+      "step": 171
+    },
+    {
+      "epoch": 0.0215,
+      "grad_norm": 3.5827748775482178,
+      "grad_norm_var": 0.9775809993657352,
+      "learning_rate": 0.0001,
+      "loss": 0.9301,
+      "loss/crossentropy": 2.160230875015259,
+      "loss/hidden": 0.5703125,
+      "loss/logits": 0.10158500075340271,
+      "loss/reg": 0.025821613147854805,
+      "step": 172
+    },
+    {
+      "epoch": 0.021625,
+      "grad_norm": 3.0348143577575684,
+      "grad_norm_var": 1.008817027257092,
+      "learning_rate": 0.0001,
+      "loss": 0.9059,
+      "loss/crossentropy": 2.5519795417785645,
+      "loss/hidden": 0.5390625,
+      "loss/logits": 0.10867513716220856,
+      "loss/reg": 0.025815250352025032,
+      "step": 173
+    },
+    {
+      "epoch": 0.02175,
+      "grad_norm": 3.325514316558838,
+      "grad_norm_var": 0.980302655794393,
+      "learning_rate": 0.0001,
+      "loss": 0.7651,
+      "loss/crossentropy": 2.3060855865478516,
+      "loss/hidden": 0.43359375,
+      "loss/logits": 0.0733788013458252,
+      "loss/reg": 0.02580902725458145,
+      "step": 174
+    },
+    {
+      "epoch": 0.021875,
+      "grad_norm": 2.9402523040771484,
+      "grad_norm_var": 0.21740374576502078,
+      "learning_rate": 0.0001,
+      "loss": 0.859,
+      "loss/crossentropy": 2.4109151363372803,
+      "loss/hidden": 0.515625,
+      "loss/logits": 0.0853327289223671,
+      "loss/reg": 0.025802936404943466,
+      "step": 175
+    },
+    {
+      "epoch": 0.022,
+      "grad_norm": 4.970669746398926,
+      "grad_norm_var": 0.354037293260262,
+      "learning_rate": 0.0001,
+      "loss": 0.9835,
+      "loss/crossentropy": 2.9366648197174072,
+      "loss/hidden": 0.6171875,
+      "loss/logits": 0.10840301960706711,
+      "loss/reg": 0.02579565905034542,
+      "step": 176
+    },
+    {
+      "epoch": 0.022125,
+      "grad_norm": 4.438536643981934,
+      "grad_norm_var": 0.37010993593279384,
+      "learning_rate": 0.0001,
+      "loss": 0.8574,
+      "loss/crossentropy": 2.4177615642547607,
+      "loss/hidden": 0.51171875,
+      "loss/logits": 0.08775197714567184,
+      "loss/reg": 0.025788920000195503,
+      "step": 177
+    },
+    {
+      "epoch": 0.02225,
+      "grad_norm": 4.3905863761901855,
+      "grad_norm_var": 0.41060424896311526,
+      "learning_rate": 0.0001,
+      "loss": 0.8796,
+      "loss/crossentropy": 2.320418119430542,
+      "loss/hidden": 0.53125,
+      "loss/logits": 0.09051363915205002,
+      "loss/reg": 0.02578234300017357,
+      "step": 178
+    },
+    {
+      "epoch": 0.022375,
+      "grad_norm": 3.2044453620910645,
+      "grad_norm_var": 0.42189777730298467,
+      "learning_rate": 0.0001,
+      "loss": 0.785,
+      "loss/crossentropy": 2.327108383178711,
+      "loss/hidden": 0.447265625,
+      "loss/logits": 0.08003242313861847,
+      "loss/reg": 0.025775177404284477,
+      "step": 179
+    },
+    {
+      "epoch": 0.0225,
+      "grad_norm": 3.2016260623931885,
+      "grad_norm_var": 0.4319725268587102,
+      "learning_rate": 0.0001,
+      "loss": 0.8076,
+      "loss/crossentropy": 2.6913022994995117,
+      "loss/hidden": 0.466796875,
+      "loss/logits": 0.08316424489021301,
+      "loss/reg": 0.02576799876987934,
+      "step": 180
+    },
+    {
+      "epoch": 0.022625,
+      "grad_norm": 3.216141939163208,
+      "grad_norm_var": 0.42772885748243633,
+      "learning_rate": 0.0001,
+      "loss": 0.7782,
+      "loss/crossentropy": 2.4444668292999268,
+      "loss/hidden": 0.447265625,
+      "loss/logits": 0.07332297414541245,
+      "loss/reg": 0.025760415941476822,
+      "step": 181
+    },
+    {
+      "epoch": 0.02275,
+      "grad_norm": 3.6005637645721436,
+      "grad_norm_var": 0.3680557604441513,
+      "learning_rate": 0.0001,
+      "loss": 0.8578,
+      "loss/crossentropy": 2.2084226608276367,
+      "loss/hidden": 0.51171875,
+      "loss/logits": 0.0885147675871849,
+      "loss/reg": 0.025753989815711975,
+      "step": 182
+    },
+    {
+      "epoch": 0.022875,
+      "grad_norm": 4.19577693939209,
+      "grad_norm_var": 0.3852931468556484,
+      "learning_rate": 0.0001,
+      "loss": 0.9394,
+      "loss/crossentropy": 2.557783842086792,
+      "loss/hidden": 0.56640625,
+      "loss/logits": 0.11552520841360092,
+      "loss/reg": 0.025747526437044144,
+      "step": 183
+    },
+    {
+      "epoch": 0.023,
+      "grad_norm": 3.024552822113037,
+      "grad_norm_var": 0.38129301153876227,
+      "learning_rate": 0.0001,
+      "loss": 0.8858,
+      "loss/crossentropy": 2.994615316390991,
+      "loss/hidden": 0.52734375,
+      "loss/logits": 0.10108112543821335,
+      "loss/reg": 0.02574075385928154,
+      "step": 184
+    },
+    {
+      "epoch": 0.023125,
+      "grad_norm": 3.3255319595336914,
+      "grad_norm_var": 0.3706833429951111,
+      "learning_rate": 0.0001,
+      "loss": 0.8407,
+      "loss/crossentropy": 2.677245855331421,
+      "loss/hidden": 0.4765625,
+      "loss/logits": 0.10678394883871078,
+      "loss/reg": 0.02573317475616932,
+      "step": 185
+    },
+    {
+      "epoch": 0.02325,
+      "grad_norm": 3.341599464416504,
+      "grad_norm_var": 0.3716797590150757,
+      "learning_rate": 0.0001,
+      "loss": 0.9127,
+      "loss/crossentropy": 2.156444549560547,
+      "loss/hidden": 0.56640625,
+      "loss/logits": 0.0890708938241005,
+      "loss/reg": 0.02572541870176792,
+      "step": 186
+    },
+    {
+      "epoch": 0.023375,
+      "grad_norm": 2.925915241241455,
+      "grad_norm_var": 0.3822577484351124,
+      "learning_rate": 0.0001,
+      "loss": 0.8815,
+      "loss/crossentropy": 2.2716755867004395,
+      "loss/hidden": 0.53125,
+      "loss/logits": 0.09304732084274292,
+      "loss/reg": 0.02571748197078705,
+      "step": 187
+    },
+    {
+      "epoch": 0.0235,
+      "grad_norm": 4.192226886749268,
+      "grad_norm_var": 0.40854537365233884,
+      "learning_rate": 0.0001,
+      "loss": 0.8547,
+      "loss/crossentropy": 2.378901720046997,
+      "loss/hidden": 0.5078125,
+      "loss/logits": 0.08977752178907394,
+      "loss/reg": 0.025709524750709534,
+      "step": 188
+    },
+    {
+      "epoch": 0.023625,
+      "grad_norm": 5.648179054260254,
+      "grad_norm_var": 0.6443691048121629,
+      "learning_rate": 0.0001,
+      "loss": 1.0032,
+      "loss/crossentropy": 2.4307687282562256,
+      "loss/hidden": 0.63671875,
+      "loss/logits": 0.10948194563388824,
+      "loss/reg": 0.025701580569148064,
+      "step": 189
+    },
+    {
+      "epoch": 0.02375,
+      "grad_norm": 6.345841884613037,
+      "grad_norm_var": 1.045029826307897,
+      "learning_rate": 0.0001,
+      "loss": 1.0995,
+      "loss/crossentropy": 2.279203414916992,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.09642117470502853,
+      "loss/reg": 0.025694590061903,
+      "step": 190
+    },
+    {
+      "epoch": 0.023875,
+      "grad_norm": 4.242865085601807,
+      "grad_norm_var": 0.9782837984014707,
+      "learning_rate": 0.0001,
+      "loss": 1.155,
+      "loss/crossentropy": 2.235325574874878,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.10906486213207245,
+      "loss/reg": 0.025686509907245636,
+      "step": 191
+    },
+    {
+      "epoch": 0.024,
+      "grad_norm": 6.347895622253418,
+      "grad_norm_var": 1.272032888242258,
+      "learning_rate": 0.0001,
+      "loss": 1.0789,
+      "loss/crossentropy": 2.5386898517608643,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.1463102549314499,
+      "loss/reg": 0.025679145008325577,
+      "step": 192
+    },
+    {
+      "epoch": 0.024125,
+      "grad_norm": 3.077846050262451,
+      "grad_norm_var": 1.3268106432816444,
+      "learning_rate": 0.0001,
+      "loss": 0.855,
+      "loss/crossentropy": 2.5889694690704346,
+      "loss/hidden": 0.515625,
+      "loss/logits": 0.08266487717628479,
+      "loss/reg": 0.025671878829598427,
+      "step": 193
+    },
+    {
+      "epoch": 0.02425,
+      "grad_norm": 3.672849416732788,
+      "grad_norm_var": 1.323313109234428,
+      "learning_rate": 0.0001,
+      "loss": 0.8213,
+      "loss/crossentropy": 2.269009590148926,
+      "loss/hidden": 0.4921875,
+      "loss/logits": 0.07244250178337097,
+      "loss/reg": 0.02566472254693508,
+      "step": 194
+    },
+    {
+      "epoch": 0.024375,
+      "grad_norm": 3.13712215423584,
+      "grad_norm_var": 1.330492936258482,
+      "learning_rate": 0.0001,
+      "loss": 0.9044,
+      "loss/crossentropy": 2.3420536518096924,
+      "loss/hidden": 0.5390625,
+      "loss/logits": 0.10874692350625992,
+      "loss/reg": 0.02565707452595234,
+      "step": 195
+    },
+    {
+      "epoch": 0.0245,
+      "grad_norm": 5.941372871398926,
+      "grad_norm_var": 1.5194802994125645,
+      "learning_rate": 0.0001,
+      "loss": 1.0268,
+      "loss/crossentropy": 2.245668649673462,
+      "loss/hidden": 0.63671875,
+      "loss/logits": 0.13362175226211548,
+      "loss/reg": 0.02564912661910057,
+      "step": 196
+    },
+    {
+      "epoch": 0.024625,
+      "grad_norm": 2.8778631687164307,
+      "grad_norm_var": 1.5682913914576866,
+      "learning_rate": 0.0001,
+      "loss": 0.933,
+      "loss/crossentropy": 2.4744086265563965,
+      "loss/hidden": 0.57421875,
+      "loss/logits": 0.10236240178346634,
+      "loss/reg": 0.025642510503530502,
+      "step": 197
+    },
+    {
+      "epoch": 0.02475,
+      "grad_norm": 5.235295295715332,
+      "grad_norm_var": 1.6223942527523605,
+      "learning_rate": 0.0001,
+      "loss": 0.993,
+      "loss/crossentropy": 2.3120462894439697,
+      "loss/hidden": 0.61328125,
+      "loss/logits": 0.12334179133176804,
+      "loss/reg": 0.025634463876485825,
+      "step": 198
+    },
+    {
+      "epoch": 0.024875,
+      "grad_norm": 3.2772397994995117,
+      "grad_norm_var": 1.6781902664948347,
+      "learning_rate": 0.0001,
+      "loss": 0.9134,
+      "loss/crossentropy": 2.2896904945373535,
+      "loss/hidden": 0.5625,
+      "loss/logits": 0.0946369469165802,
+      "loss/reg": 0.025627706199884415,
+      "step": 199
+    },
+    {
+      "epoch": 0.025,
+      "grad_norm": 4.021130084991455,
+      "grad_norm_var": 1.5889382838258257,
+      "learning_rate": 0.0001,
+      "loss": 0.9467,
+      "loss/crossentropy": 2.3281702995300293,
+      "loss/hidden": 0.5859375,
+      "loss/logits": 0.10459813475608826,
+      "loss/reg": 0.025620225816965103,
+      "step": 200
+    },
+    {
+      "epoch": 0.025125,
+      "grad_norm": 3.3705508708953857,
+      "grad_norm_var": 1.5836618344967157,
+      "learning_rate": 0.0001,
+      "loss": 1.0247,
+      "loss/crossentropy": 2.3704278469085693,
+      "loss/hidden": 0.640625,
+      "loss/logits": 0.12795141339302063,
+      "loss/reg": 0.025613589212298393,
+      "step": 201
+    },
+    {
+      "epoch": 0.02525,
+      "grad_norm": 5.586423397064209,
+      "grad_norm_var": 1.6331597901730046,
+      "learning_rate": 0.0001,
+      "loss": 1.2703,
+      "loss/crossentropy": 2.3346736431121826,
+      "loss/hidden": 0.83984375,
+      "loss/logits": 0.17438159883022308,
+      "loss/reg": 0.025606893002986908,
+      "step": 202
+    },
+    {
+      "epoch": 0.025375,
+      "grad_norm": 6.558523654937744,
+      "grad_norm_var": 1.7590475344041898,
+      "learning_rate": 0.0001,
+      "loss": 1.0711,
+      "loss/crossentropy": 2.264883518218994,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.10028564184904099,
+      "loss/reg": 0.025599893182516098,
+      "step": 203
+    },
+    {
+      "epoch": 0.0255,
+      "grad_norm": 3.024080991744995,
+      "grad_norm_var": 1.9071946132324447,
+      "learning_rate": 0.0001,
+      "loss": 0.9349,
+      "loss/crossentropy": 2.505457878112793,
+      "loss/hidden": 0.58203125,
+      "loss/logits": 0.09690214693546295,
+      "loss/reg": 0.025592036545276642,
+      "step": 204
+    },
+    {
+      "epoch": 0.025625,
+      "grad_norm": 3.268216133117676,
+      "grad_norm_var": 1.9040994009139534,
+      "learning_rate": 0.0001,
+      "loss": 0.8433,
+      "loss/crossentropy": 2.629786968231201,
+      "loss/hidden": 0.5,
+      "loss/logits": 0.08743932843208313,
+      "loss/reg": 0.025583887472748756,
+      "step": 205
+    },
+    {
+      "epoch": 0.02575,
+      "grad_norm": 5.203437328338623,
+      "grad_norm_var": 1.6853258867355625,
+      "learning_rate": 0.0001,
+      "loss": 0.9278,
+      "loss/crossentropy": 2.368603229522705,
+      "loss/hidden": 0.5625,
+      "loss/logits": 0.10955986380577087,
+      "loss/reg": 0.025574835017323494,
+      "step": 206
+    },
+    {
+      "epoch": 0.025875,
+      "grad_norm": 5.106112480163574,
+      "grad_norm_var": 1.725017173963382,
+      "learning_rate": 0.0001,
+      "loss": 0.9881,
+      "loss/crossentropy": 2.554746389389038,
+      "loss/hidden": 0.6171875,
+      "loss/logits": 0.11524944007396698,
+      "loss/reg": 0.02556804195046425,
+      "step": 207
+    },
+    {
+      "epoch": 0.026,
+      "grad_norm": 8.258187294006348,
+      "grad_norm_var": 2.4602814049521387,
+      "learning_rate": 0.0001,
+      "loss": 1.3675,
+      "loss/crossentropy": 2.4391205310821533,
+      "loss/hidden": 0.94140625,
+      "loss/logits": 0.17047560214996338,
+      "loss/reg": 0.025560656562447548,
+      "step": 208
+    },
+    {
+      "epoch": 0.026125,
+      "grad_norm": 3.8223764896392822,
+      "grad_norm_var": 2.3561294395388566,
+      "learning_rate": 0.0001,
+      "loss": 1.02,
+      "loss/crossentropy": 2.5300426483154297,
+      "loss/hidden": 0.62890625,
+      "loss/logits": 0.13556598126888275,
+      "loss/reg": 0.025551345199346542,
+      "step": 209
+    },
+    {
+      "epoch": 0.02625,
+      "grad_norm": 3.6237666606903076,
+      "grad_norm_var": 2.36184075461094,
+      "learning_rate": 0.0001,
+      "loss": 1.0218,
+      "loss/crossentropy": 2.3368213176727295,
+      "loss/hidden": 0.6640625,
+      "loss/logits": 0.10230866074562073,
+      "loss/reg": 0.025544527918100357,
+      "step": 210
+    },
+    {
+      "epoch": 0.026375,
+      "grad_norm": 3.1394050121307373,
+      "grad_norm_var": 2.3614203164344407,
+      "learning_rate": 0.0001,
+      "loss": 0.964,
+      "loss/crossentropy": 2.338050603866577,
+      "loss/hidden": 0.609375,
+      "loss/logits": 0.09931059181690216,
+      "loss/reg": 0.025535617023706436,
+      "step": 211
+    },
+    {
+      "epoch": 0.0265,
+      "grad_norm": 3.1498186588287354,
+      "grad_norm_var": 2.319283484830568,
+      "learning_rate": 0.0001,
+      "loss": 0.8039,
+      "loss/crossentropy": 2.5108940601348877,
+      "loss/hidden": 0.46875,
+      "loss/logits": 0.07982419431209564,
+      "loss/reg": 0.025528721511363983,
+      "step": 212
+    },
+    {
+      "epoch": 0.026625,
+      "grad_norm": 3.334510326385498,
+      "grad_norm_var": 2.2429786468962374,
+      "learning_rate": 0.0001,
+      "loss": 1.1105,
+      "loss/crossentropy": 2.458519697189331,
+      "loss/hidden": 0.71875,
+      "loss/logits": 0.1365831047296524,
+      "loss/reg": 0.02552017569541931,
+      "step": 213
+    },
+    {
+      "epoch": 0.02675,
+      "grad_norm": 3.3243789672851562,
+      "grad_norm_var": 2.2516768547287977,
+      "learning_rate": 0.0001,
+      "loss": 1.0016,
+      "loss/crossentropy": 2.2530109882354736,
+      "loss/hidden": 0.62890625,
+      "loss/logits": 0.11759582161903381,
+      "loss/reg": 0.025511734187602997,
+      "step": 214
+    },
+    {
+      "epoch": 0.026875,
+      "grad_norm": 3.3768937587738037,
+      "grad_norm_var": 2.2393156807375245,
+      "learning_rate": 0.0001,
+      "loss": 1.0452,
+      "loss/crossentropy": 2.3267643451690674,
+      "loss/hidden": 0.65625,
+      "loss/logits": 0.13390058279037476,
+      "loss/reg": 0.0255054272711277,
+      "step": 215
+    },
+    {
+      "epoch": 0.027,
+      "grad_norm": 7.391561031341553,
+      "grad_norm_var": 2.841738119885866,
+      "learning_rate": 0.0001,
+      "loss": 1.2263,
+      "loss/crossentropy": 2.285508394241333,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.1510239541530609,
+      "loss/reg": 0.025499247014522552,
+      "step": 216
+    },
+    {
+      "epoch": 0.027125,
+      "grad_norm": 3.143969774246216,
+      "grad_norm_var": 2.8781965049841705,
+      "learning_rate": 0.0001,
+      "loss": 0.9421,
+      "loss/crossentropy": 2.6111316680908203,
+      "loss/hidden": 0.5703125,
+      "loss/logits": 0.11691074818372726,
+      "loss/reg": 0.025491848587989807,
+      "step": 217
+    },
+    {
+      "epoch": 0.02725,
+      "grad_norm": 4.989267826080322,
+      "grad_norm_var": 2.8105564664802234,
+      "learning_rate": 0.0001,
+      "loss": 0.9104,
+      "loss/crossentropy": 2.7856109142303467,
+      "loss/hidden": 0.546875,
+      "loss/logits": 0.10870229452848434,
+      "loss/reg": 0.025485411286354065,
+      "step": 218
+    },
+    {
+      "epoch": 0.027375,
+      "grad_norm": 8.867380142211914,
+      "grad_norm_var": 3.802177537112387,
+      "learning_rate": 0.0001,
+      "loss": 1.2976,
+      "loss/crossentropy": 2.414778470993042,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.199102982878685,
+      "loss/reg": 0.025478005409240723,
+      "step": 219
+    },
+    {
+      "epoch": 0.0275,
+      "grad_norm": 3.949193239212036,
+      "grad_norm_var": 3.665725599495321,
+      "learning_rate": 0.0001,
+      "loss": 1.0781,
+      "loss/crossentropy": 2.3898277282714844,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.14368270337581635,
+      "loss/reg": 0.02547168917953968,
+      "step": 220
+    },
+    {
+      "epoch": 0.027625,
+      "grad_norm": 7.980153560638428,
+      "grad_norm_var": 4.202985170085262,
+      "learning_rate": 0.0001,
+      "loss": 1.4005,
+      "loss/crossentropy": 2.0598959922790527,
+      "loss/hidden": 0.9921875,
+      "loss/logits": 0.15361803770065308,
+      "loss/reg": 0.02546495571732521,
+      "step": 221
+    },
+    {
+      "epoch": 0.02775,
+      "grad_norm": 4.118736743927002,
+      "grad_norm_var": 4.234989890674561,
+      "learning_rate": 0.0001,
+      "loss": 0.9084,
+      "loss/crossentropy": 2.2568750381469727,
+      "loss/hidden": 0.55859375,
+      "loss/logits": 0.09525588899850845,
+      "loss/reg": 0.02545757219195366,
+      "step": 222
+    },
+    {
+      "epoch": 0.027875,
+      "grad_norm": 3.730299711227417,
+      "grad_norm_var": 4.306033514824207,
+      "learning_rate": 0.0001,
+      "loss": 0.9394,
+      "loss/crossentropy": 2.7180914878845215,
+      "loss/hidden": 0.57421875,
+      "loss/logits": 0.11064038425683975,
+      "loss/reg": 0.025450890883803368,
+      "step": 223
+    },
+    {
+      "epoch": 0.028,
+      "grad_norm": 3.138925552368164,
+      "grad_norm_var": 3.557911666554559,
+      "learning_rate": 0.0001,
+      "loss": 0.8406,
+      "loss/crossentropy": 2.3719072341918945,
+      "loss/hidden": 0.5078125,
+      "loss/logits": 0.07834647595882416,
+      "loss/reg": 0.02544352412223816,
+      "step": 224
+    },
+    {
+      "epoch": 0.028125,
+      "grad_norm": 3.478994846343994,
+      "grad_norm_var": 3.593674795871404,
+      "learning_rate": 0.0001,
+      "loss": 0.9072,
+      "loss/crossentropy": 2.4101219177246094,
+      "loss/hidden": 0.56640625,
+      "loss/logits": 0.0863800048828125,
+      "loss/reg": 0.025437019765377045,
+      "step": 225
+    },
+    {
+      "epoch": 0.02825,
+      "grad_norm": 5.091615676879883,
+      "grad_norm_var": 3.572291640880083,
+      "learning_rate": 0.0001,
+      "loss": 0.9864,
+      "loss/crossentropy": 2.2258856296539307,
+      "loss/hidden": 0.640625,
+      "loss/logits": 0.09143185615539551,
+      "loss/reg": 0.025430168956518173,
+      "step": 226
+    },
+    {
+      "epoch": 0.028375,
+      "grad_norm": 3.616190195083618,
+      "grad_norm_var": 3.4991896025782796,
+      "learning_rate": 0.0001,
+      "loss": 0.9246,
+      "loss/crossentropy": 2.765105724334717,
+      "loss/hidden": 0.5703125,
+      "loss/logits": 0.10003923624753952,
+      "loss/reg": 0.025423482060432434,
+      "step": 227
+    },
+    {
+      "epoch": 0.0285,
+      "grad_norm": 4.406581878662109,
+      "grad_norm_var": 3.364516245493509,
+      "learning_rate": 0.0001,
+      "loss": 0.98,
+      "loss/crossentropy": 2.41001033782959,
+      "loss/hidden": 0.6171875,
+      "loss/logits": 0.10860306769609451,
+      "loss/reg": 0.025416266173124313,
+      "step": 228
+    },
+    {
+      "epoch": 0.028625,
+      "grad_norm": 3.003995418548584,
+      "grad_norm_var": 3.4280449285692023,
+      "learning_rate": 0.0001,
+      "loss": 0.9759,
+      "loss/crossentropy": 2.518749952316284,
+      "loss/hidden": 0.62109375,
+      "loss/logits": 0.10075733810663223,
+      "loss/reg": 0.025409165769815445,
+      "step": 229
+    },
+    {
+      "epoch": 0.02875,
+      "grad_norm": 4.073727130889893,
+      "grad_norm_var": 3.3356380380428576,
+      "learning_rate": 0.0001,
+      "loss": 0.8881,
+      "loss/crossentropy": 2.6824846267700195,
+      "loss/hidden": 0.5390625,
+      "loss/logits": 0.09498724341392517,
+      "loss/reg": 0.025402268394827843,
+      "step": 230
+    },
+    {
+      "epoch": 0.028875,
+      "grad_norm": 3.8958635330200195,
+      "grad_norm_var": 3.2645611787952435,
+      "learning_rate": 0.0001,
+      "loss": 0.8794,
+      "loss/crossentropy": 2.684971332550049,
+      "loss/hidden": 0.53515625,
+      "loss/logits": 0.09024453163146973,
+      "loss/reg": 0.025395380333065987,
+      "step": 231
+    },
+    {
+      "epoch": 0.029,
+      "grad_norm": 3.5619406700134277,
+      "grad_norm_var": 2.796506014438906,
+      "learning_rate": 0.0001,
+      "loss": 0.957,
+      "loss/crossentropy": 2.7883284091949463,
+      "loss/hidden": 0.6015625,
+      "loss/logits": 0.10159540176391602,
+      "loss/reg": 0.02538810484111309,
+      "step": 232
+    },
+    {
+      "epoch": 0.029125,
+      "grad_norm": 12.658771514892578,
+      "grad_norm_var": 6.809983669727001,
+      "learning_rate": 0.0001,
+      "loss": 1.1843,
+      "loss/crossentropy": 2.537827253341675,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.14144758880138397,
+      "loss/reg": 0.02538110502064228,
+      "step": 233
+    },
+    {
+      "epoch": 0.02925,
+      "grad_norm": 8.465475082397461,
+      "grad_norm_var": 7.543990683504188,
+      "learning_rate": 0.0001,
+      "loss": 1.2958,
+      "loss/crossentropy": 2.6715972423553467,
+      "loss/hidden": 0.875,
+      "loss/logits": 0.16701380908489227,
+      "loss/reg": 0.02537420578300953,
+      "step": 234
+    },
+    {
+      "epoch": 0.029375,
+      "grad_norm": 6.49767541885376,
+      "grad_norm_var": 6.75275709893707,
+      "learning_rate": 0.0001,
+      "loss": 1.0826,
+      "loss/crossentropy": 2.316210985183716,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.11794352531433105,
+      "loss/reg": 0.025367144495248795,
+      "step": 235
+    },
+    {
+      "epoch": 0.0295,
+      "grad_norm": 4.668674945831299,
+      "grad_norm_var": 6.674304000957216,
+      "learning_rate": 0.0001,
+      "loss": 1.0642,
+      "loss/crossentropy": 1.956210732460022,
+      "loss/hidden": 0.69921875,
+      "loss/logits": 0.11135473847389221,
+      "loss/reg": 0.02535996399819851,
+      "step": 236
+    },
+    {
+      "epoch": 0.029625,
+      "grad_norm": 6.132915019989014,
+      "grad_norm_var": 6.19031909782113,
+      "learning_rate": 0.0001,
+      "loss": 1.0462,
+      "loss/crossentropy": 2.6552460193634033,
+      "loss/hidden": 0.6640625,
+      "loss/logits": 0.12862679362297058,
+      "loss/reg": 0.025352442637085915,
+      "step": 237
+    },
+    {
+      "epoch": 0.02975,
+      "grad_norm": 4.94125509262085,
+      "grad_norm_var": 6.132251305092321,
+      "learning_rate": 0.0001,
+      "loss": 1.2795,
+      "loss/crossentropy": 2.204523801803589,
+      "loss/hidden": 0.84765625,
+      "loss/logits": 0.17835499346256256,
+      "loss/reg": 0.025344664230942726,
+      "step": 238
+    },
+    {
+      "epoch": 0.029875,
+      "grad_norm": 3.7683677673339844,
+      "grad_norm_var": 6.125464850588167,
+      "learning_rate": 0.0001,
+      "loss": 0.985,
+      "loss/crossentropy": 2.652397632598877,
+      "loss/hidden": 0.625,
+      "loss/logits": 0.10665792226791382,
+      "loss/reg": 0.0253366157412529,
+      "step": 239
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.732015609741211,
+      "grad_norm_var": 5.870172361717241,
+      "learning_rate": 0.0001,
+      "loss": 1.1018,
+      "loss/crossentropy": 2.544055938720703,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.13754525780677795,
+      "loss/reg": 0.02532930299639702,
+      "step": 240
+    },
+    {
+      "epoch": 0.030125,
+      "grad_norm": 4.046056270599365,
+      "grad_norm_var": 5.761120866273571,
+      "learning_rate": 0.0001,
+      "loss": 0.9396,
+      "loss/crossentropy": 2.6015427112579346,
+      "loss/hidden": 0.5859375,
+      "loss/logits": 0.10041546076536179,
+      "loss/reg": 0.025322062894701958,
+      "step": 241
+    },
+    {
+      "epoch": 0.03025,
+      "grad_norm": 3.066027879714966,
+      "grad_norm_var": 6.052926687728684,
+      "learning_rate": 0.0001,
+      "loss": 1.0668,
+      "loss/crossentropy": 2.4402151107788086,
+      "loss/hidden": 0.6875,
+      "loss/logits": 0.12614867091178894,
+      "loss/reg": 0.0253145694732666,
+      "step": 242
+    },
+    {
+      "epoch": 0.030375,
+      "grad_norm": 3.168888807296753,
+      "grad_norm_var": 6.1536859873832555,
+      "learning_rate": 0.0001,
+      "loss": 0.9413,
+      "loss/crossentropy": 2.5689940452575684,
+      "loss/hidden": 0.58984375,
+      "loss/logits": 0.09841729700565338,
+      "loss/reg": 0.02530776336789131,
+      "step": 243
+    },
+    {
+      "epoch": 0.0305,
+      "grad_norm": 3.806450366973877,
+      "grad_norm_var": 6.229122059899357,
+      "learning_rate": 0.0001,
+      "loss": 1.021,
+      "loss/crossentropy": 2.2438809871673584,
+      "loss/hidden": 0.6328125,
+      "loss/logits": 0.13515594601631165,
+      "loss/reg": 0.02530042454600334,
+      "step": 244
+    },
+    {
+      "epoch": 0.030625,
+      "grad_norm": 4.313736915588379,
+      "grad_norm_var": 5.982441934425083,
+      "learning_rate": 0.0001,
+      "loss": 1.0307,
+      "loss/crossentropy": 2.5067784786224365,
+      "loss/hidden": 0.65234375,
+      "loss/logits": 0.1253766119480133,
+      "loss/reg": 0.025293124839663506,
+      "step": 245
+    },
+    {
+      "epoch": 0.03075,
+      "grad_norm": 4.419394016265869,
+      "grad_norm_var": 5.942040082684442,
+      "learning_rate": 0.0001,
+      "loss": 0.8957,
+      "loss/crossentropy": 2.4725239276885986,
+      "loss/hidden": 0.55078125,
+      "loss/logits": 0.09201550483703613,
+      "loss/reg": 0.02528616413474083,
+      "step": 246
+    },
+    {
+      "epoch": 0.030875,
+      "grad_norm": 3.709151268005371,
+      "grad_norm_var": 5.975041529003943,
+      "learning_rate": 0.0001,
+      "loss": 0.9152,
+      "loss/crossentropy": 2.413250207901001,
+      "loss/hidden": 0.56640625,
+      "loss/logits": 0.09596529603004456,
+      "loss/reg": 0.025278838351368904,
+      "step": 247
+    },
+    {
+      "epoch": 0.031,
+      "grad_norm": 3.4139935970306396,
+      "grad_norm_var": 6.007189625317282,
+      "learning_rate": 0.0001,
+      "loss": 1.0268,
+      "loss/crossentropy": 2.1580560207366943,
+      "loss/hidden": 0.65625,
+      "loss/logits": 0.11786328256130219,
+      "loss/reg": 0.025271562859416008,
+      "step": 248
+    },
+    {
+      "epoch": 0.031125,
+      "grad_norm": 6.089378833770752,
+      "grad_norm_var": 2.0950588257899443,
+      "learning_rate": 0.0001,
+      "loss": 1.0999,
+      "loss/crossentropy": 2.5466785430908203,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.1441519856452942,
+      "loss/reg": 0.025264522060751915,
+      "step": 249
+    },
+    {
+      "epoch": 0.03125,
+      "grad_norm": 4.001245498657227,
+      "grad_norm_var": 1.1007847740596406,
+      "learning_rate": 0.0001,
+      "loss": 0.9263,
+      "loss/crossentropy": 2.433518171310425,
+      "loss/hidden": 0.578125,
+      "loss/logits": 0.09559185057878494,
+      "loss/reg": 0.025257611647248268,
+      "step": 250
+    },
+    {
+      "epoch": 0.031375,
+      "grad_norm": 4.982790946960449,
+      "grad_norm_var": 0.8252532202355399,
+      "learning_rate": 0.0001,
+      "loss": 1.1439,
+      "loss/crossentropy": 2.3542721271514893,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.16091328859329224,
+      "loss/reg": 0.025250321254134178,
+      "step": 251
+    },
+    {
+      "epoch": 0.0315,
+      "grad_norm": 3.6227378845214844,
+      "grad_norm_var": 0.8462248829388517,
+      "learning_rate": 0.0001,
+      "loss": 1.1789,
+      "loss/crossentropy": 2.1333043575286865,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.1374010145664215,
+      "loss/reg": 0.025242896750569344,
+      "step": 252
+    },
+    {
+      "epoch": 0.031625,
+      "grad_norm": 3.415194511413574,
+      "grad_norm_var": 0.6304077366129337,
+      "learning_rate": 0.0001,
+      "loss": 1.0916,
+      "loss/crossentropy": 2.23525071144104,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.1243971735239029,
+      "loss/reg": 0.025235962122678757,
+      "step": 253
+    },
+    {
+      "epoch": 0.03175,
+      "grad_norm": 3.7671289443969727,
+      "grad_norm_var": 0.5838590152321442,
+      "learning_rate": 0.0001,
+      "loss": 0.908,
+      "loss/crossentropy": 2.254054546356201,
+      "loss/hidden": 0.56640625,
+      "loss/logits": 0.0893106684088707,
+      "loss/reg": 0.025228681042790413,
+      "step": 254
+    },
+    {
+      "epoch": 0.031875,
+      "grad_norm": 4.164376735687256,
+      "grad_norm_var": 0.5803655311074387,
+      "learning_rate": 0.0001,
+      "loss": 0.9548,
+      "loss/crossentropy": 2.4128520488739014,
+      "loss/hidden": 0.5859375,
+      "loss/logits": 0.11668873578310013,
+      "loss/reg": 0.025221774354577065,
+      "step": 255
+    },
+    {
+      "epoch": 0.032,
+      "grad_norm": 3.6412672996520996,
+      "grad_norm_var": 0.5547959425019464,
+      "learning_rate": 0.0001,
+      "loss": 1.0098,
+      "loss/crossentropy": 2.7745320796966553,
+      "loss/hidden": 0.62890625,
+      "loss/logits": 0.12875014543533325,
+      "loss/reg": 0.025214577093720436,
+      "step": 256
+    },
+    {
+      "epoch": 0.032125,
+      "grad_norm": 2.900871515274048,
+      "grad_norm_var": 0.6261772657264061,
+      "learning_rate": 0.0001,
+      "loss": 0.9348,
+      "loss/crossentropy": 2.286968469619751,
+      "loss/hidden": 0.578125,
+      "loss/logits": 0.10455667227506638,
+      "loss/reg": 0.02520710788667202,
+      "step": 257
+    },
+    {
+      "epoch": 0.03225,
+      "grad_norm": 3.6853647232055664,
+      "grad_norm_var": 0.5808564529014478,
+      "learning_rate": 0.0001,
+      "loss": 0.9013,
+      "loss/crossentropy": 2.4515562057495117,
+      "loss/hidden": 0.55078125,
+      "loss/logits": 0.09847953915596008,
+      "loss/reg": 0.025199349969625473,
+      "step": 258
+    },
+    {
+      "epoch": 0.032375,
+      "grad_norm": 2.9091956615448,
+      "grad_norm_var": 0.6119059054418109,
+      "learning_rate": 0.0001,
+      "loss": 0.8784,
+      "loss/crossentropy": 2.163628339767456,
+      "loss/hidden": 0.5390625,
+      "loss/logits": 0.08743810653686523,
+      "loss/reg": 0.025191258639097214,
+      "step": 259
+    },
+    {
+      "epoch": 0.0325,
+      "grad_norm": 2.9488673210144043,
+      "grad_norm_var": 0.6717290813098125,
+      "learning_rate": 0.0001,
+      "loss": 0.9195,
+      "loss/crossentropy": 2.3084585666656494,
+      "loss/hidden": 0.5546875,
+      "loss/logits": 0.11298206448554993,
+      "loss/reg": 0.025183765217661858,
+      "step": 260
+    },
+    {
+      "epoch": 0.032625,
+      "grad_norm": 3.4345953464508057,
+      "grad_norm_var": 0.6684943296663647,
+      "learning_rate": 0.0001,
+      "loss": 0.8851,
+      "loss/crossentropy": 2.19558048248291,
+      "loss/hidden": 0.546875,
+      "loss/logits": 0.08649411797523499,
+      "loss/reg": 0.02517561800777912,
+      "step": 261
+    },
+    {
+      "epoch": 0.03275,
+      "grad_norm": 28.159074783325195,
+      "grad_norm_var": 37.79188620028727,
+      "learning_rate": 0.0001,
+      "loss": 1.1781,
+      "loss/crossentropy": 2.839057207107544,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.13732999563217163,
+      "loss/reg": 0.025168145075440407,
+      "step": 262
+    },
+    {
+      "epoch": 0.032875,
+      "grad_norm": 2.734104871749878,
+      "grad_norm_var": 38.05849364469687,
+      "learning_rate": 0.0001,
+      "loss": 0.8892,
+      "loss/crossentropy": 2.4754340648651123,
+      "loss/hidden": 0.54296875,
+      "loss/logits": 0.09462890028953552,
+      "loss/reg": 0.025160137563943863,
+      "step": 263
+    },
+    {
+      "epoch": 0.033,
+      "grad_norm": 4.089654922485352,
+      "grad_norm_var": 37.922354469790704,
+      "learning_rate": 0.0001,
+      "loss": 0.9326,
+      "loss/crossentropy": 2.476450204849243,
+      "loss/hidden": 0.57421875,
+      "loss/logits": 0.1068512499332428,
+      "loss/reg": 0.025151856243610382,
+      "step": 264
+    },
+    {
+      "epoch": 0.033125,
+      "grad_norm": 3.838066577911377,
+      "grad_norm_var": 37.99741003814723,
+      "learning_rate": 0.0001,
+      "loss": 1.0909,
+      "loss/crossentropy": 2.114243268966675,
+      "loss/hidden": 0.734375,
+      "loss/logits": 0.10510236769914627,
+      "loss/reg": 0.025143325328826904,
+      "step": 265
+    },
+    {
+      "epoch": 0.03325,
+      "grad_norm": 3.0674874782562256,
+      "grad_norm_var": 38.19410456778619,
+      "learning_rate": 0.0001,
+      "loss": 0.8898,
+      "loss/crossentropy": 2.6426563262939453,
+      "loss/hidden": 0.54296875,
+      "loss/logits": 0.09550425410270691,
+      "loss/reg": 0.025135278701782227,
+      "step": 266
+    },
+    {
+      "epoch": 0.033375,
+      "grad_norm": 7.777696132659912,
+      "grad_norm_var": 38.64421623432597,
+      "learning_rate": 0.0001,
+      "loss": 1.0482,
+      "loss/crossentropy": 2.640554666519165,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.11728060245513916,
+      "loss/reg": 0.025126684457063675,
+      "step": 267
+    },
+    {
+      "epoch": 0.0335,
+      "grad_norm": 3.6375653743743896,
+      "grad_norm_var": 38.64099364344996,
+      "learning_rate": 0.0001,
+      "loss": 0.9888,
+      "loss/crossentropy": 2.227538824081421,
+      "loss/hidden": 0.63671875,
+      "loss/logits": 0.1008879542350769,
+      "loss/reg": 0.0251180287450552,
+      "step": 268
+    },
+    {
+      "epoch": 0.033625,
+      "grad_norm": 4.305724620819092,
+      "grad_norm_var": 38.4714335626231,
+      "learning_rate": 0.0001,
+      "loss": 1.1268,
+      "loss/crossentropy": 2.556248903274536,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.1374487727880478,
+      "loss/reg": 0.02511041797697544,
+      "step": 269
+    },
+    {
+      "epoch": 0.03375,
+      "grad_norm": 3.658052921295166,
+      "grad_norm_var": 38.4947077039297,
+      "learning_rate": 0.0001,
+      "loss": 0.9616,
+      "loss/crossentropy": 2.437307596206665,
+      "loss/hidden": 0.60546875,
+      "loss/logits": 0.1050904244184494,
+      "loss/reg": 0.025102900341153145,
+      "step": 270
+    },
+    {
+      "epoch": 0.033875,
+      "grad_norm": 6.534849643707275,
+      "grad_norm_var": 38.483973576312195,
+      "learning_rate": 0.0001,
+      "loss": 1.2119,
+      "loss/crossentropy": 2.36796236038208,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.1445278525352478,
+      "loss/reg": 0.025095317512750626,
+      "step": 271
+    },
+    {
+      "epoch": 0.034,
+      "grad_norm": 5.486751079559326,
+      "grad_norm_var": 38.24988881420663,
+      "learning_rate": 0.0001,
+      "loss": 1.0337,
+      "loss/crossentropy": 2.517333984375,
+      "loss/hidden": 0.66015625,
+      "loss/logits": 0.12263435125350952,
+      "loss/reg": 0.025087665766477585,
+      "step": 272
+    },
+    {
+      "epoch": 0.034125,
+      "grad_norm": 3.231940746307373,
+      "grad_norm_var": 38.13878485092763,
+      "learning_rate": 0.0001,
+      "loss": 1.0586,
+      "loss/crossentropy": 2.6812305450439453,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.13204006850719452,
+      "loss/reg": 0.025080092251300812,
+      "step": 273
+    },
+    {
+      "epoch": 0.03425,
+      "grad_norm": 4.71685791015625,
+      "grad_norm_var": 37.942827296069424,
+      "learning_rate": 0.0001,
+      "loss": 1.0409,
+      "loss/crossentropy": 2.093151807785034,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.11049959808588028,
+      "loss/reg": 0.02507280558347702,
+      "step": 274
+    },
+    {
+      "epoch": 0.034375,
+      "grad_norm": 3.7086901664733887,
+      "grad_norm_var": 37.68973967522894,
+      "learning_rate": 0.0001,
+      "loss": 0.9144,
+      "loss/crossentropy": 2.2392518520355225,
+      "loss/hidden": 0.57421875,
+      "loss/logits": 0.08954055607318878,
+      "loss/reg": 0.02506582997739315,
+      "step": 275
+    },
+    {
+      "epoch": 0.0345,
+      "grad_norm": 5.537543773651123,
+      "grad_norm_var": 37.1561912525544,
+      "learning_rate": 0.0001,
+      "loss": 1.3586,
+      "loss/crossentropy": 1.8622019290924072,
+      "loss/hidden": 0.96484375,
+      "loss/logits": 0.14318415522575378,
+      "loss/reg": 0.025058824568986893,
+      "step": 276
+    },
+    {
+      "epoch": 0.034625,
+      "grad_norm": 2.898383617401123,
+      "grad_norm_var": 37.34827444255352,
+      "learning_rate": 0.0001,
+      "loss": 0.9413,
+      "loss/crossentropy": 2.4155869483947754,
+      "loss/hidden": 0.578125,
+      "loss/logits": 0.11263684928417206,
+      "loss/reg": 0.025052132084965706,
+      "step": 277
+    },
+    {
+      "epoch": 0.03475,
+      "grad_norm": 2.803711175918579,
+      "grad_norm_var": 2.0625830733920933,
+      "learning_rate": 0.0001,
+      "loss": 0.9404,
+      "loss/crossentropy": 2.2058048248291016,
+      "loss/hidden": 0.59765625,
+      "loss/logits": 0.09232598543167114,
+      "loss/reg": 0.02504453808069229,
+      "step": 278
+    },
+    {
+      "epoch": 0.034875,
+      "grad_norm": 3.180114507675171,
+      "grad_norm_var": 1.9847680294286107,
+      "learning_rate": 0.0001,
+      "loss": 0.9311,
+      "loss/crossentropy": 2.6069722175598145,
+      "loss/hidden": 0.57421875,
+      "loss/logits": 0.10655493289232254,
+      "loss/reg": 0.02503693662583828,
+      "step": 279
+    },
+    {
+      "epoch": 0.035,
+      "grad_norm": 5.609209060668945,
+      "grad_norm_var": 2.0906055341906256,
+      "learning_rate": 0.0001,
+      "loss": 1.0675,
+      "loss/crossentropy": 2.5023670196533203,
+      "loss/hidden": 0.68359375,
+      "loss/logits": 0.13361594080924988,
+      "loss/reg": 0.025029515847563744,
+      "step": 280
+    },
+    {
+      "epoch": 0.035125,
+      "grad_norm": 5.165370464324951,
+      "grad_norm_var": 2.105772188928517,
+      "learning_rate": 0.0001,
+      "loss": 1.1103,
+      "loss/crossentropy": 2.5107007026672363,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.12183347344398499,
+      "loss/reg": 0.025022249668836594,
+      "step": 281
+    },
+    {
+      "epoch": 0.03525,
+      "grad_norm": 4.322115421295166,
+      "grad_norm_var": 1.9716269568170388,
+      "learning_rate": 0.0001,
+      "loss": 0.95,
+      "loss/crossentropy": 2.483142137527466,
+      "loss/hidden": 0.5859375,
+      "loss/logits": 0.11390361189842224,
+      "loss/reg": 0.025014575570821762,
+      "step": 282
+    },
+    {
+      "epoch": 0.035375,
+      "grad_norm": 4.432065486907959,
+      "grad_norm_var": 1.2250959918754403,
+      "learning_rate": 0.0001,
+      "loss": 0.9886,
+      "loss/crossentropy": 2.1206424236297607,
+      "loss/hidden": 0.6328125,
+      "loss/logits": 0.10568062961101532,
+      "loss/reg": 0.025007015094161034,
+      "step": 283
+    },
+    {
+      "epoch": 0.0355,
+      "grad_norm": 4.0429558753967285,
+      "grad_norm_var": 1.198112283867575,
+      "learning_rate": 0.0001,
+      "loss": 1.0004,
+      "loss/crossentropy": 2.6465060710906982,
+      "loss/hidden": 0.6328125,
+      "loss/logits": 0.11765305697917938,
+      "loss/reg": 0.024997249245643616,
+      "step": 284
+    },
+    {
+      "epoch": 0.035625,
+      "grad_norm": 3.551006555557251,
+      "grad_norm_var": 1.23838358717468,
+      "learning_rate": 0.0001,
+      "loss": 0.9764,
+      "loss/crossentropy": 2.4274182319641113,
+      "loss/hidden": 0.61328125,
+      "loss/logits": 0.11320526152849197,
+      "loss/reg": 0.02498740889132023,
+      "step": 285
+    },
+    {
+      "epoch": 0.03575,
+      "grad_norm": 2.874697685241699,
+      "grad_norm_var": 1.344305852802292,
+      "learning_rate": 0.0001,
+      "loss": 1.0133,
+      "loss/crossentropy": 2.512948751449585,
+      "loss/hidden": 0.6484375,
+      "loss/logits": 0.11510799080133438,
+      "loss/reg": 0.02497740648686886,
+      "step": 286
+    },
+    {
+      "epoch": 0.035875,
+      "grad_norm": 3.4613194465637207,
+      "grad_norm_var": 1.0008425760722102,
+      "learning_rate": 0.0001,
+      "loss": 0.9274,
+      "loss/crossentropy": 2.4450602531433105,
+      "loss/hidden": 0.5703125,
+      "loss/logits": 0.10743874311447144,
+      "loss/reg": 0.024968616664409637,
+      "step": 287
+    },
+    {
+      "epoch": 0.036,
+      "grad_norm": 3.452150344848633,
+      "grad_norm_var": 0.8735820507410946,
+      "learning_rate": 0.0001,
+      "loss": 0.952,
+      "loss/crossentropy": 2.304729700088501,
+      "loss/hidden": 0.578125,
+      "loss/logits": 0.1242954432964325,
+      "loss/reg": 0.024959923699498177,
+      "step": 288
+    },
+    {
+      "epoch": 0.036125,
+      "grad_norm": 3.3603436946868896,
+      "grad_norm_var": 0.8625457550688983,
+      "learning_rate": 0.0001,
+      "loss": 0.9447,
+      "loss/crossentropy": 2.425712823867798,
+      "loss/hidden": 0.58984375,
+      "loss/logits": 0.1053181067109108,
+      "loss/reg": 0.024950530380010605,
+      "step": 289
+    },
+    {
+      "epoch": 0.03625,
+      "grad_norm": 3.3667662143707275,
+      "grad_norm_var": 0.8374846368179986,
+      "learning_rate": 0.0001,
+      "loss": 0.9067,
+      "loss/crossentropy": 2.2623162269592285,
+      "loss/hidden": 0.55859375,
+      "loss/logits": 0.098650723695755,
+      "loss/reg": 0.02494119666516781,
+      "step": 290
+    },
+    {
+      "epoch": 0.036375,
+      "grad_norm": 2.7996857166290283,
+      "grad_norm_var": 0.9075153562133816,
+      "learning_rate": 0.0001,
+      "loss": 0.9594,
+      "loss/crossentropy": 2.370417356491089,
+      "loss/hidden": 0.59765625,
+      "loss/logits": 0.11243726313114166,
+      "loss/reg": 0.024933209642767906,
+      "step": 291
+    },
+    {
+      "epoch": 0.0365,
+      "grad_norm": 2.9999380111694336,
+      "grad_norm_var": 0.7233017120786666,
+      "learning_rate": 0.0001,
+      "loss": 1.0396,
+      "loss/crossentropy": 2.286776542663574,
+      "loss/hidden": 0.6640625,
+      "loss/logits": 0.12626898288726807,
+      "loss/reg": 0.02492516115307808,
+      "step": 292
+    },
+    {
+      "epoch": 0.036625,
+      "grad_norm": 3.880777597427368,
+      "grad_norm_var": 0.685825505757979,
+      "learning_rate": 0.0001,
+      "loss": 0.9764,
+      "loss/crossentropy": 2.613823652267456,
+      "loss/hidden": 0.61328125,
+      "loss/logits": 0.11395551264286041,
+      "loss/reg": 0.02491726726293564,
+      "step": 293
+    },
+    {
+      "epoch": 0.03675,
+      "grad_norm": 4.2950568199157715,
+      "grad_norm_var": 0.6453385025094112,
+      "learning_rate": 0.0001,
+      "loss": 1.0535,
+      "loss/crossentropy": 2.6882171630859375,
+      "loss/hidden": 0.66796875,
+      "loss/logits": 0.13644324243068695,
+      "loss/reg": 0.02490835078060627,
+      "step": 294
+    },
+    {
+      "epoch": 0.036875,
+      "grad_norm": 3.61423659324646,
+      "grad_norm_var": 0.6212598300915251,
+      "learning_rate": 0.0001,
+      "loss": 1.155,
+      "loss/crossentropy": 2.0867068767547607,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.13649392127990723,
+      "loss/reg": 0.024899456650018692,
+      "step": 295
+    },
+    {
+      "epoch": 0.037,
+      "grad_norm": 3.9245386123657227,
+      "grad_norm_var": 0.3982568915415859,
+      "learning_rate": 0.0001,
+      "loss": 0.951,
+      "loss/crossentropy": 2.703791618347168,
+      "loss/hidden": 0.5859375,
+      "loss/logits": 0.11614765971899033,
+      "loss/reg": 0.02489159069955349,
+      "step": 296
+    },
+    {
+      "epoch": 0.037125,
+      "grad_norm": 4.020047187805176,
+      "grad_norm_var": 0.2597397925732442,
+      "learning_rate": 0.0001,
+      "loss": 0.9707,
+      "loss/crossentropy": 2.429636001586914,
+      "loss/hidden": 0.61328125,
+      "loss/logits": 0.1086028665304184,
+      "loss/reg": 0.024883201345801353,
+      "step": 297
+    },
+    {
+      "epoch": 0.03725,
+      "grad_norm": 4.086516857147217,
+      "grad_norm_var": 0.24209119003572066,
+      "learning_rate": 0.0001,
+      "loss": 1.0769,
+      "loss/crossentropy": 2.8581056594848633,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.14844708144664764,
+      "loss/reg": 0.024874594062566757,
+      "step": 298
+    },
+    {
+      "epoch": 0.037375,
+      "grad_norm": 2.956085443496704,
+      "grad_norm_var": 0.22141400399237127,
+      "learning_rate": 0.0001,
+      "loss": 0.9903,
+      "loss/crossentropy": 2.4802041053771973,
+      "loss/hidden": 0.609375,
+      "loss/logits": 0.13230839371681213,
+      "loss/reg": 0.024866636842489243,
+      "step": 299
+    },
+    {
+      "epoch": 0.0375,
+      "grad_norm": 3.165804386138916,
+      "grad_norm_var": 0.2110158468875736,
+      "learning_rate": 0.0001,
+      "loss": 1.0347,
+      "loss/crossentropy": 2.4370639324188232,
+      "loss/hidden": 0.66015625,
+      "loss/logits": 0.1259341686964035,
+      "loss/reg": 0.024857714772224426,
+      "step": 300
+    },
+    {
+      "epoch": 0.037625,
+      "grad_norm": 3.0942695140838623,
+      "grad_norm_var": 0.22022059823099174,
+      "learning_rate": 0.0001,
+      "loss": 0.9747,
+      "loss/crossentropy": 2.674190044403076,
+      "loss/hidden": 0.6015625,
+      "loss/logits": 0.12467385828495026,
+      "loss/reg": 0.024848785251379013,
+      "step": 301
+    },
+    {
+      "epoch": 0.03775,
+      "grad_norm": 2.949205160140991,
+      "grad_norm_var": 0.21475779393049,
+      "learning_rate": 0.0001,
+      "loss": 0.8898,
+      "loss/crossentropy": 2.491746664047241,
+      "loss/hidden": 0.55078125,
+      "loss/logits": 0.09065801650285721,
+      "loss/reg": 0.02483983524143696,
+      "step": 302
+    },
+    {
+      "epoch": 0.037875,
+      "grad_norm": 3.549431324005127,
+      "grad_norm_var": 0.21520952048913009,
+      "learning_rate": 0.0001,
+      "loss": 0.9197,
+      "loss/crossentropy": 2.7578208446502686,
+      "loss/hidden": 0.57421875,
+      "loss/logits": 0.09712585806846619,
+      "loss/reg": 0.024831857532262802,
+      "step": 303
+    },
+    {
+      "epoch": 0.038,
+      "grad_norm": 3.0621931552886963,
+      "grad_norm_var": 0.22562503941355938,
+      "learning_rate": 0.0001,
+      "loss": 0.894,
+      "loss/crossentropy": 2.440351724624634,
+      "loss/hidden": 0.546875,
+      "loss/logits": 0.09888219833374023,
+      "loss/reg": 0.02482294850051403,
+      "step": 304
+    },
+    {
+      "epoch": 0.038125,
+      "grad_norm": 3.334942579269409,
+      "grad_norm_var": 0.22595311715915212,
+      "learning_rate": 0.0001,
+      "loss": 1.1509,
+      "loss/crossentropy": 2.378168821334839,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.15670377016067505,
+      "loss/reg": 0.024814244359731674,
+      "step": 305
+    },
+    {
+      "epoch": 0.03825,
+      "grad_norm": 3.3253893852233887,
+      "grad_norm_var": 0.22648465837488155,
+      "learning_rate": 0.0001,
+      "loss": 0.9744,
+      "loss/crossentropy": 2.4409470558166504,
+      "loss/hidden": 0.6015625,
+      "loss/logits": 0.12483348697423935,
+      "loss/reg": 0.024805361405014992,
+      "step": 306
+    },
+    {
+      "epoch": 0.038375,
+      "grad_norm": 3.1687541007995605,
+      "grad_norm_var": 0.20343285009947346,
+      "learning_rate": 0.0001,
+      "loss": 0.8998,
+      "loss/crossentropy": 2.3339829444885254,
+      "loss/hidden": 0.55078125,
+      "loss/logits": 0.10110392421483994,
+      "loss/reg": 0.02479635737836361,
+      "step": 307
+    },
+    {
+      "epoch": 0.0385,
+      "grad_norm": 4.096661567687988,
+      "grad_norm_var": 0.21071919009248533,
+      "learning_rate": 0.0001,
+      "loss": 1.0154,
+      "loss/crossentropy": 2.468813180923462,
+      "loss/hidden": 0.64453125,
+      "loss/logits": 0.12295258045196533,
+      "loss/reg": 0.024787236005067825,
+      "step": 308
+    },
+    {
+      "epoch": 0.038625,
+      "grad_norm": 3.097642660140991,
+      "grad_norm_var": 0.2127095324618587,
+      "learning_rate": 0.0001,
+      "loss": 0.8772,
+      "loss/crossentropy": 2.5380513668060303,
+      "loss/hidden": 0.53125,
+      "loss/logits": 0.09821398556232452,
+      "loss/reg": 0.024777989834547043,
+      "step": 309
+    },
+    {
+      "epoch": 0.03875,
+      "grad_norm": 4.647727012634277,
+      "grad_norm_var": 0.25863060133758775,
+      "learning_rate": 0.0001,
+      "loss": 0.9162,
+      "loss/crossentropy": 2.0996336936950684,
+      "loss/hidden": 0.5625,
+      "loss/logits": 0.10599061101675034,
+      "loss/reg": 0.024768849834799767,
+      "step": 310
+    },
+    {
+      "epoch": 0.038875,
+      "grad_norm": 3.627246141433716,
+      "grad_norm_var": 0.25882920418562944,
+      "learning_rate": 0.0001,
+      "loss": 1.0161,
+      "loss/crossentropy": 2.2958993911743164,
+      "loss/hidden": 0.65234375,
+      "loss/logits": 0.11611323803663254,
+      "loss/reg": 0.024759870022535324,
+      "step": 311
+    },
+    {
+      "epoch": 0.039,
+      "grad_norm": 3.7236764430999756,
+      "grad_norm_var": 0.2501591619921593,
+      "learning_rate": 0.0001,
+      "loss": 1.1308,
+      "loss/crossentropy": 2.4322752952575684,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.1410805881023407,
+      "loss/reg": 0.024751078337430954,
+      "step": 312
+    },
+    {
+      "epoch": 0.039125,
+      "grad_norm": 3.998807430267334,
+      "grad_norm_var": 0.24869789076210413,
+      "learning_rate": 0.0001,
+      "loss": 1.0749,
+      "loss/crossentropy": 2.374758720397949,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.1243831142783165,
+      "loss/reg": 0.024742012843489647,
+      "step": 313
+    },
+    {
+      "epoch": 0.03925,
+      "grad_norm": 4.315618515014648,
+      "grad_norm_var": 0.2701154191311919,
+      "learning_rate": 0.0001,
+      "loss": 0.9542,
+      "loss/crossentropy": 2.290766477584839,
+      "loss/hidden": 0.60546875,
+      "loss/logits": 0.10138154029846191,
+      "loss/reg": 0.02473386563360691,
+      "step": 314
+    },
+    {
+      "epoch": 0.039375,
+      "grad_norm": 11.789870262145996,
+      "grad_norm_var": 4.498354875640615,
+      "learning_rate": 0.0001,
+      "loss": 1.6276,
+      "loss/crossentropy": 2.391585350036621,
+      "loss/hidden": 1.25,
+      "loss/logits": 0.13034726679325104,
+      "loss/reg": 0.02472575195133686,
+      "step": 315
+    },
+    {
+      "epoch": 0.0395,
+      "grad_norm": 3.499288320541382,
+      "grad_norm_var": 4.465581075155145,
+      "learning_rate": 0.0001,
+      "loss": 0.9478,
+      "loss/crossentropy": 2.4527926445007324,
+      "loss/hidden": 0.5703125,
+      "loss/logits": 0.13031822443008423,
+      "loss/reg": 0.024717645719647408,
+      "step": 316
+    },
+    {
+      "epoch": 0.039625,
+      "grad_norm": 3.0139851570129395,
+      "grad_norm_var": 4.476536239649594,
+      "learning_rate": 0.0001,
+      "loss": 0.9555,
+      "loss/crossentropy": 2.506457805633545,
+      "loss/hidden": 0.58203125,
+      "loss/logits": 0.12640972435474396,
+      "loss/reg": 0.02470862865447998,
+      "step": 317
+    },
+    {
+      "epoch": 0.03975,
+      "grad_norm": 3.671523094177246,
+      "grad_norm_var": 4.4007183053584145,
+      "learning_rate": 0.0001,
+      "loss": 1.0902,
+      "loss/crossentropy": 2.1730313301086426,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.13228294253349304,
+      "loss/reg": 0.024700626730918884,
+      "step": 318
+    },
+    {
+      "epoch": 0.039875,
+      "grad_norm": 3.453159809112549,
+      "grad_norm_var": 4.408623714873802,
+      "learning_rate": 0.0001,
+      "loss": 0.9518,
+      "loss/crossentropy": 2.7452075481414795,
+      "loss/hidden": 0.58984375,
+      "loss/logits": 0.11500866711139679,
+      "loss/reg": 0.02469259686768055,
+      "step": 319
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 8.18558406829834,
+      "grad_norm_var": 5.330579476502794,
+      "learning_rate": 0.0001,
+      "loss": 1.2593,
+      "loss/crossentropy": 2.295231819152832,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.2312176525592804,
+      "loss/reg": 0.024684064090251923,
+      "step": 320
+    },
+    {
+      "epoch": 0.040125,
+      "grad_norm": 4.58513879776001,
+      "grad_norm_var": 5.245000173569268,
+      "learning_rate": 0.0001,
+      "loss": 1.0428,
+      "loss/crossentropy": 2.3456668853759766,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.12027865648269653,
+      "loss/reg": 0.024676108732819557,
+      "step": 321
+    },
+    {
+      "epoch": 0.04025,
+      "grad_norm": 3.6148953437805176,
+      "grad_norm_var": 5.20441494141252,
+      "learning_rate": 0.0001,
+      "loss": 0.885,
+      "loss/crossentropy": 2.3615198135375977,
+      "loss/hidden": 0.53515625,
+      "loss/logits": 0.10311760008335114,
+      "loss/reg": 0.02466769702732563,
+      "step": 322
+    },
+    {
+      "epoch": 0.040375,
+      "grad_norm": 4.596187114715576,
+      "grad_norm_var": 5.072570501388933,
+      "learning_rate": 0.0001,
+      "loss": 1.3642,
+      "loss/crossentropy": 1.9407044649124146,
+      "loss/hidden": 0.9140625,
+      "loss/logits": 0.20351722836494446,
+      "loss/reg": 0.02465960383415222,
+      "step": 323
+    },
+    {
+      "epoch": 0.0405,
+      "grad_norm": 4.305622577667236,
+      "grad_norm_var": 5.060723771971758,
+      "learning_rate": 0.0001,
+      "loss": 1.2854,
+      "loss/crossentropy": 2.1887283325195312,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.20682212710380554,
+      "loss/reg": 0.024651024490594864,
+      "step": 324
+    },
+    {
+      "epoch": 0.040625,
+      "grad_norm": 3.1052956581115723,
+      "grad_norm_var": 5.059160883569213,
+      "learning_rate": 0.0001,
+      "loss": 0.9277,
+      "loss/crossentropy": 2.7552340030670166,
+      "loss/hidden": 0.56640625,
+      "loss/logits": 0.11491444706916809,
+      "loss/reg": 0.024642454460263252,
+      "step": 325
+    },
+    {
+      "epoch": 0.04075,
+      "grad_norm": 3.902174234390259,
+      "grad_norm_var": 5.092472426369553,
+      "learning_rate": 0.0001,
+      "loss": 1.0922,
+      "loss/crossentropy": 2.328878402709961,
+      "loss/hidden": 0.70703125,
+      "loss/logits": 0.13887017965316772,
+      "loss/reg": 0.024634363129734993,
+      "step": 326
+    },
+    {
+      "epoch": 0.040875,
+      "grad_norm": 3.144326686859131,
+      "grad_norm_var": 5.168830163995767,
+      "learning_rate": 0.0001,
+      "loss": 0.9028,
+      "loss/crossentropy": 2.765711545944214,
+      "loss/hidden": 0.55078125,
+      "loss/logits": 0.10574661940336227,
+      "loss/reg": 0.02462565153837204,
+      "step": 327
+    },
+    {
+      "epoch": 0.041,
+      "grad_norm": 4.085259914398193,
+      "grad_norm_var": 5.136846736797656,
+      "learning_rate": 0.0001,
+      "loss": 1.1366,
+      "loss/crossentropy": 2.2278430461883545,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.13655412197113037,
+      "loss/reg": 0.0246175117790699,
+      "step": 328
+    },
+    {
+      "epoch": 0.041125,
+      "grad_norm": 3.578554391860962,
+      "grad_norm_var": 5.180404969237465,
+      "learning_rate": 0.0001,
+      "loss": 1.0227,
+      "loss/crossentropy": 2.7795305252075195,
+      "loss/hidden": 0.640625,
+      "loss/logits": 0.13600832223892212,
+      "loss/reg": 0.024609515443444252,
+      "step": 329
+    },
+    {
+      "epoch": 0.04125,
+      "grad_norm": 3.7418434619903564,
+      "grad_norm_var": 5.219134310055354,
+      "learning_rate": 0.0001,
+      "loss": 1.0244,
+      "loss/crossentropy": 2.636258840560913,
+      "loss/hidden": 0.63671875,
+      "loss/logits": 0.1417178213596344,
+      "loss/reg": 0.024600951001048088,
+      "step": 330
+    },
+    {
+      "epoch": 0.041375,
+      "grad_norm": 3.369840383529663,
+      "grad_norm_var": 1.4852025101017772,
+      "learning_rate": 0.0001,
+      "loss": 1.0208,
+      "loss/crossentropy": 2.448961019515991,
+      "loss/hidden": 0.6640625,
+      "loss/logits": 0.1107659712433815,
+      "loss/reg": 0.024592852219939232,
+      "step": 331
+    },
+    {
+      "epoch": 0.0415,
+      "grad_norm": 2.90629506111145,
+      "grad_norm_var": 1.5460412234751968,
+      "learning_rate": 0.0001,
+      "loss": 0.9411,
+      "loss/crossentropy": 2.595834732055664,
+      "loss/hidden": 0.59375,
+      "loss/logits": 0.10147438943386078,
+      "loss/reg": 0.024584442377090454,
+      "step": 332
+    },
+    {
+      "epoch": 0.041625,
+      "grad_norm": 3.402386426925659,
+      "grad_norm_var": 1.5068032644485272,
+      "learning_rate": 0.0001,
+      "loss": 1.0144,
+      "loss/crossentropy": 2.363499402999878,
+      "loss/hidden": 0.62890625,
+      "loss/logits": 0.1397327035665512,
+      "loss/reg": 0.02457563206553459,
+      "step": 333
+    },
+    {
+      "epoch": 0.04175,
+      "grad_norm": 3.597465753555298,
+      "grad_norm_var": 1.5101723473758888,
+      "learning_rate": 0.0001,
+      "loss": 1.1917,
+      "loss/crossentropy": 2.3256943225860596,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.15691371262073517,
+      "loss/reg": 0.02456764318048954,
+      "step": 334
+    },
+    {
+      "epoch": 0.041875,
+      "grad_norm": 3.512730121612549,
+      "grad_norm_var": 1.506262203991567,
+      "learning_rate": 0.0001,
+      "loss": 1.0631,
+      "loss/crossentropy": 2.383385181427002,
+      "loss/hidden": 0.69140625,
+      "loss/logits": 0.12608817219734192,
+      "loss/reg": 0.024559510871767998,
+      "step": 335
+    },
+    {
+      "epoch": 0.042,
+      "grad_norm": 5.968827247619629,
+      "grad_norm_var": 0.5694964439723395,
+      "learning_rate": 0.0001,
+      "loss": 1.258,
+      "loss/crossentropy": 2.212991952896118,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.18438610434532166,
+      "loss/reg": 0.02455153875052929,
+      "step": 336
+    },
+    {
+      "epoch": 0.042125,
+      "grad_norm": 3.644115924835205,
+      "grad_norm_var": 0.5311677507970897,
+      "learning_rate": 0.0001,
+      "loss": 1.1203,
+      "loss/crossentropy": 2.3620948791503906,
+      "loss/hidden": 0.72265625,
+      "loss/logits": 0.15219135582447052,
+      "loss/reg": 0.02454366721212864,
+      "step": 337
+    },
+    {
+      "epoch": 0.04225,
+      "grad_norm": 2.716367244720459,
+      "grad_norm_var": 0.6013761572733584,
+      "learning_rate": 0.0001,
+      "loss": 0.919,
+      "loss/crossentropy": 2.3863117694854736,
+      "loss/hidden": 0.578125,
+      "loss/logits": 0.0955524817109108,
+      "loss/reg": 0.024536145851016045,
+      "step": 338
+    },
+    {
+      "epoch": 0.042375,
+      "grad_norm": 3.8117287158966064,
+      "grad_norm_var": 0.5485673092684531,
+      "learning_rate": 0.0001,
+      "loss": 1.0529,
+      "loss/crossentropy": 2.4951469898223877,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.1318708062171936,
+      "loss/reg": 0.024528514593839645,
+      "step": 339
+    },
+    {
+      "epoch": 0.0425,
+      "grad_norm": 2.5218894481658936,
+      "grad_norm_var": 0.5973356289049185,
+      "learning_rate": 0.0001,
+      "loss": 0.9673,
+      "loss/crossentropy": 2.370839834213257,
+      "loss/hidden": 0.6015625,
+      "loss/logits": 0.12049120664596558,
+      "loss/reg": 0.02452007494866848,
+      "step": 340
+    },
+    {
+      "epoch": 0.042625,
+      "grad_norm": 2.7343108654022217,
+      "grad_norm_var": 0.6285810690168129,
+      "learning_rate": 0.0001,
+      "loss": 1.1201,
+      "loss/crossentropy": 2.2990617752075195,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.16403642296791077,
+      "loss/reg": 0.024511409923434258,
+      "step": 341
+    },
+    {
+      "epoch": 0.04275,
+      "grad_norm": 2.5267186164855957,
+      "grad_norm_var": 0.6803812464423664,
+      "learning_rate": 0.0001,
+      "loss": 0.9056,
+      "loss/crossentropy": 2.5629754066467285,
+      "loss/hidden": 0.5546875,
+      "loss/logits": 0.10588675737380981,
+      "loss/reg": 0.024503152817487717,
+      "step": 342
+    },
+    {
+      "epoch": 0.042875,
+      "grad_norm": 2.2731988430023193,
+      "grad_norm_var": 0.7637691760365584,
+      "learning_rate": 0.0001,
+      "loss": 0.8154,
+      "loss/crossentropy": 2.4123644828796387,
+      "loss/hidden": 0.490234375,
+      "loss/logits": 0.08022630214691162,
+      "loss/reg": 0.024494923651218414,
+      "step": 343
+    },
+    {
+      "epoch": 0.043,
+      "grad_norm": 3.2791945934295654,
+      "grad_norm_var": 0.7306725618305314,
+      "learning_rate": 0.0001,
+      "loss": 0.9879,
+      "loss/crossentropy": 2.7643816471099854,
+      "loss/hidden": 0.6171875,
+      "loss/logits": 0.12586694955825806,
+      "loss/reg": 0.024486759677529335,
+      "step": 344
+    },
+    {
+      "epoch": 0.043125,
+      "grad_norm": 3.6740994453430176,
+      "grad_norm_var": 0.7341663188433093,
+      "learning_rate": 0.0001,
+      "loss": 0.8947,
+      "loss/crossentropy": 2.6802780628204346,
+      "loss/hidden": 0.5546875,
+      "loss/logits": 0.0951782613992691,
+      "loss/reg": 0.024478696286678314,
+      "step": 345
+    },
+    {
+      "epoch": 0.04325,
+      "grad_norm": 3.234722137451172,
+      "grad_norm_var": 0.7240869727338347,
+      "learning_rate": 0.0001,
+      "loss": 0.8517,
+      "loss/crossentropy": 2.577921152114868,
+      "loss/hidden": 0.515625,
+      "loss/logits": 0.09137749671936035,
+      "loss/reg": 0.024470962584018707,
+      "step": 346
+    },
+    {
+      "epoch": 0.043375,
+      "grad_norm": 4.024074554443359,
+      "grad_norm_var": 0.7548921970504276,
+      "learning_rate": 0.0001,
+      "loss": 1.0017,
+      "loss/crossentropy": 2.8129634857177734,
+      "loss/hidden": 0.63671875,
+      "loss/logits": 0.120377317070961,
+      "loss/reg": 0.02446298860013485,
+      "step": 347
+    },
+    {
+      "epoch": 0.0435,
+      "grad_norm": 3.4327027797698975,
+      "grad_norm_var": 0.7400679146500114,
+      "learning_rate": 0.0001,
+      "loss": 0.9864,
+      "loss/crossentropy": 2.4310224056243896,
+      "loss/hidden": 0.640625,
+      "loss/logits": 0.10121208429336548,
+      "loss/reg": 0.0244552381336689,
+      "step": 348
+    },
+    {
+      "epoch": 0.043625,
+      "grad_norm": 3.2115890979766846,
+      "grad_norm_var": 0.7422101391295163,
+      "learning_rate": 0.0001,
+      "loss": 1.0594,
+      "loss/crossentropy": 2.3658318519592285,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.13528358936309814,
+      "loss/reg": 0.02444704994559288,
+      "step": 349
+    },
+    {
+      "epoch": 0.04375,
+      "grad_norm": 3.4005136489868164,
+      "grad_norm_var": 0.739061242813568,
+      "learning_rate": 0.0001,
+      "loss": 1.0154,
+      "loss/crossentropy": 2.9171154499053955,
+      "loss/hidden": 0.6171875,
+      "loss/logits": 0.15386469662189484,
+      "loss/reg": 0.024438532069325447,
+      "step": 350
+    },
+    {
+      "epoch": 0.043875,
+      "grad_norm": 4.1175127029418945,
+      "grad_norm_var": 0.7731950105324129,
+      "learning_rate": 0.0001,
+      "loss": 0.949,
+      "loss/crossentropy": 2.3405494689941406,
+      "loss/hidden": 0.59765625,
+      "loss/logits": 0.10704682767391205,
+      "loss/reg": 0.024429937824606895,
+      "step": 351
+    },
+    {
+      "epoch": 0.044,
+      "grad_norm": 3.9496090412139893,
+      "grad_norm_var": 0.33930652052577653,
+      "learning_rate": 0.0001,
+      "loss": 0.889,
+      "loss/crossentropy": 2.9135613441467285,
+      "loss/hidden": 0.54296875,
+      "loss/logits": 0.1018136739730835,
+      "loss/reg": 0.024421829730272293,
+      "step": 352
+    },
+    {
+      "epoch": 0.044125,
+      "grad_norm": 3.794520139694214,
+      "grad_norm_var": 0.347931624130162,
+      "learning_rate": 0.0001,
+      "loss": 1.1771,
+      "loss/crossentropy": 2.5529844760894775,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.17909468710422516,
+      "loss/reg": 0.02441396936774254,
+      "step": 353
+    },
+    {
+      "epoch": 0.04425,
+      "grad_norm": 3.164349317550659,
+      "grad_norm_var": 0.3259767305032634,
+      "learning_rate": 0.0001,
+      "loss": 1.0804,
+      "loss/crossentropy": 2.3843231201171875,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.12148329615592957,
+      "loss/reg": 0.02440580353140831,
+      "step": 354
+    },
+    {
+      "epoch": 0.044375,
+      "grad_norm": 2.7847914695739746,
+      "grad_norm_var": 0.32482231475126633,
+      "learning_rate": 0.0001,
+      "loss": 1.0463,
+      "loss/crossentropy": 2.521649122238159,
+      "loss/hidden": 0.68359375,
+      "loss/logits": 0.1187494546175003,
+      "loss/reg": 0.024397339671850204,
+      "step": 355
+    },
+    {
+      "epoch": 0.0445,
+      "grad_norm": 3.2391860485076904,
+      "grad_norm_var": 0.28660331114573767,
+      "learning_rate": 0.0001,
+      "loss": 1.0225,
+      "loss/crossentropy": 2.4073832035064697,
+      "loss/hidden": 0.640625,
+      "loss/logits": 0.13795481622219086,
+      "loss/reg": 0.024389205500483513,
+      "step": 356
+    },
+    {
+      "epoch": 0.044625,
+      "grad_norm": 5.839352130889893,
+      "grad_norm_var": 0.6539216724231817,
+      "learning_rate": 0.0001,
+      "loss": 1.1058,
+      "loss/crossentropy": 2.6860220432281494,
+      "loss/hidden": 0.7265625,
+      "loss/logits": 0.13543812930583954,
+      "loss/reg": 0.02438061311841011,
+      "step": 357
+    },
+    {
+      "epoch": 0.04475,
+      "grad_norm": 4.340758800506592,
+      "grad_norm_var": 0.6249977794062299,
+      "learning_rate": 0.0001,
+      "loss": 1.1,
+      "loss/crossentropy": 2.770012378692627,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.1531440019607544,
+      "loss/reg": 0.024372335523366928,
+      "step": 358
+    },
+    {
+      "epoch": 0.044875,
+      "grad_norm": 8.795727729797363,
+      "grad_norm_var": 2.1213731683569863,
+      "learning_rate": 0.0001,
+      "loss": 1.3149,
+      "loss/crossentropy": 2.2893428802490234,
+      "loss/hidden": 0.91796875,
+      "loss/logits": 0.15326841175556183,
+      "loss/reg": 0.02436378225684166,
+      "step": 359
+    },
+    {
+      "epoch": 0.045,
+      "grad_norm": 4.104447841644287,
+      "grad_norm_var": 2.0826812332104394,
+      "learning_rate": 0.0001,
+      "loss": 1.15,
+      "loss/crossentropy": 2.625378370285034,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.14858195185661316,
+      "loss/reg": 0.024355949833989143,
+      "step": 360
+    },
+    {
+      "epoch": 0.045125,
+      "grad_norm": 3.354828357696533,
+      "grad_norm_var": 2.105873348197963,
+      "learning_rate": 0.0001,
+      "loss": 0.9494,
+      "loss/crossentropy": 2.4916443824768066,
+      "loss/hidden": 0.5859375,
+      "loss/logits": 0.11993111670017242,
+      "loss/reg": 0.024348480626940727,
+      "step": 361
+    },
+    {
+      "epoch": 0.04525,
+      "grad_norm": 3.3247056007385254,
+      "grad_norm_var": 2.096606359520402,
+      "learning_rate": 0.0001,
+      "loss": 0.9604,
+      "loss/crossentropy": 2.282543420791626,
+      "loss/hidden": 0.609375,
+      "loss/logits": 0.10765975713729858,
+      "loss/reg": 0.02434113249182701,
+      "step": 362
+    },
+    {
+      "epoch": 0.045375,
+      "grad_norm": 4.314214706420898,
+      "grad_norm_var": 2.1006745469652883,
+      "learning_rate": 0.0001,
+      "loss": 1.0684,
+      "loss/crossentropy": 2.447218179702759,
+      "loss/hidden": 0.6953125,
+      "loss/logits": 0.12974585592746735,
+      "loss/reg": 0.024332784116268158,
+      "step": 363
+    },
+    {
+      "epoch": 0.0455,
+      "grad_norm": 3.6783666610717773,
+      "grad_norm_var": 2.083471757970481,
+      "learning_rate": 0.0001,
+      "loss": 1.1509,
+      "loss/crossentropy": 2.6020665168762207,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.1576002687215805,
+      "loss/reg": 0.024325383827090263,
+      "step": 364
+    },
+    {
+      "epoch": 0.045625,
+      "grad_norm": 3.535550832748413,
+      "grad_norm_var": 2.0521572529950523,
+      "learning_rate": 0.0001,
+      "loss": 0.9672,
+      "loss/crossentropy": 2.2514841556549072,
+      "loss/hidden": 0.61328125,
+      "loss/logits": 0.110772505402565,
+      "loss/reg": 0.024317855015397072,
+      "step": 365
+    },
+    {
+      "epoch": 0.04575,
+      "grad_norm": 6.190865993499756,
+      "grad_norm_var": 2.275325586048537,
+      "learning_rate": 0.0001,
+      "loss": 1.2696,
+      "loss/crossentropy": 2.5250887870788574,
+      "loss/hidden": 0.87890625,
+      "loss/logits": 0.1475657969713211,
+      "loss/reg": 0.02430957928299904,
+      "step": 366
+    },
+    {
+      "epoch": 0.045875,
+      "grad_norm": 6.9109907150268555,
+      "grad_norm_var": 2.701389202772653,
+      "learning_rate": 0.0001,
+      "loss": 1.42,
+      "loss/crossentropy": 2.0714285373687744,
+      "loss/hidden": 0.98828125,
+      "loss/logits": 0.18868008255958557,
+      "loss/reg": 0.024301210418343544,
+      "step": 367
+    },
+    {
+      "epoch": 0.046,
+      "grad_norm": 3.939924955368042,
+      "grad_norm_var": 2.702051041555256,
+      "learning_rate": 0.0001,
+      "loss": 1.0163,
+      "loss/crossentropy": 2.677281379699707,
+      "loss/hidden": 0.640625,
+      "loss/logits": 0.1327013224363327,
+      "loss/reg": 0.024292904883623123,
+      "step": 368
+    },
+    {
+      "epoch": 0.046125,
+      "grad_norm": 2.942032814025879,
+      "grad_norm_var": 2.822776844100568,
+      "learning_rate": 0.0001,
+      "loss": 0.9582,
+      "loss/crossentropy": 2.3630495071411133,
+      "loss/hidden": 0.6015625,
+      "loss/logits": 0.11380600929260254,
+      "loss/reg": 0.02428455464541912,
+      "step": 369
+    },
+    {
+      "epoch": 0.04625,
+      "grad_norm": 3.125349521636963,
+      "grad_norm_var": 2.8293167859701627,
+      "learning_rate": 0.0001,
+      "loss": 1.0463,
+      "loss/crossentropy": 2.5113635063171387,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.12774598598480225,
+      "loss/reg": 0.02427608147263527,
+      "step": 370
+    },
+    {
+      "epoch": 0.046375,
+      "grad_norm": 3.15159273147583,
+      "grad_norm_var": 2.758666518773039,
+      "learning_rate": 0.0001,
+      "loss": 0.9198,
+      "loss/crossentropy": 2.31132435798645,
+      "loss/hidden": 0.5703125,
+      "loss/logits": 0.10678299516439438,
+      "loss/reg": 0.024267377331852913,
+      "step": 371
+    },
+    {
+      "epoch": 0.0465,
+      "grad_norm": 4.394677639007568,
+      "grad_norm_var": 2.6595375525429406,
+      "learning_rate": 0.0001,
+      "loss": 0.9954,
+      "loss/crossentropy": 2.4987642765045166,
+      "loss/hidden": 0.6328125,
+      "loss/logits": 0.11996030062437057,
+      "loss/reg": 0.0242580845952034,
+      "step": 372
+    },
+    {
+      "epoch": 0.046625,
+      "grad_norm": 3.7915477752685547,
+      "grad_norm_var": 2.5549678839666425,
+      "learning_rate": 0.0001,
+      "loss": 0.9193,
+      "loss/crossentropy": 2.3996403217315674,
+      "loss/hidden": 0.578125,
+      "loss/logits": 0.09870465099811554,
+      "loss/reg": 0.02424856275320053,
+      "step": 373
+    },
+    {
+      "epoch": 0.04675,
+      "grad_norm": 2.833364725112915,
+      "grad_norm_var": 2.7025530371611066,
+      "learning_rate": 0.0001,
+      "loss": 0.9417,
+      "loss/crossentropy": 2.336583137512207,
+      "loss/hidden": 0.59765625,
+      "loss/logits": 0.10166990756988525,
+      "loss/reg": 0.024240419268608093,
+      "step": 374
+    },
+    {
+      "epoch": 0.046875,
+      "grad_norm": 5.302695274353027,
+      "grad_norm_var": 1.359315799583595,
+      "learning_rate": 0.0001,
+      "loss": 0.9495,
+      "loss/crossentropy": 2.5733697414398193,
+      "loss/hidden": 0.578125,
+      "loss/logits": 0.1290503740310669,
+      "loss/reg": 0.0242319293320179,
+      "step": 375
+    },
+    {
+      "epoch": 0.047,
+      "grad_norm": 5.087683200836182,
+      "grad_norm_var": 1.426096117003745,
+      "learning_rate": 0.0001,
+      "loss": 0.9067,
+      "loss/crossentropy": 2.548323631286621,
+      "loss/hidden": 0.56640625,
+      "loss/logits": 0.09806410223245621,
+      "loss/reg": 0.02422359585762024,
+      "step": 376
+    },
+    {
+      "epoch": 0.047125,
+      "grad_norm": 2.840883255004883,
+      "grad_norm_var": 1.4948607984557458,
+      "learning_rate": 0.0001,
+      "loss": 1.0234,
+      "loss/crossentropy": 2.484494209289551,
+      "loss/hidden": 0.65234375,
+      "loss/logits": 0.12894591689109802,
+      "loss/reg": 0.024214565753936768,
+      "step": 377
+    },
+    {
+      "epoch": 0.04725,
+      "grad_norm": 6.647733688354492,
+      "grad_norm_var": 1.848030946106532,
+      "learning_rate": 0.0001,
+      "loss": 1.131,
+      "loss/crossentropy": 2.832048177719116,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.14676988124847412,
+      "loss/reg": 0.024205682799220085,
+      "step": 378
+    },
+    {
+      "epoch": 0.047375,
+      "grad_norm": 3.465564727783203,
+      "grad_norm_var": 1.8906396391038638,
+      "learning_rate": 0.0001,
+      "loss": 1.0051,
+      "loss/crossentropy": 2.5129964351654053,
+      "loss/hidden": 0.62890625,
+      "loss/logits": 0.134174644947052,
+      "loss/reg": 0.024197354912757874,
+      "step": 379
+    },
+    {
+      "epoch": 0.0475,
+      "grad_norm": 3.6985855102539062,
+      "grad_norm_var": 1.8891513099755568,
+      "learning_rate": 0.0001,
+      "loss": 1.4422,
+      "loss/crossentropy": 2.2384328842163086,
+      "loss/hidden": 1.0,
+      "loss/logits": 0.20027095079421997,
+      "loss/reg": 0.02418883889913559,
+      "step": 380
+    },
+    {
+      "epoch": 0.047625,
+      "grad_norm": 3.4343388080596924,
+      "grad_norm_var": 1.8993141107729303,
+      "learning_rate": 0.0001,
+      "loss": 0.8648,
+      "loss/crossentropy": 2.753706932067871,
+      "loss/hidden": 0.5234375,
+      "loss/logits": 0.09959565848112106,
+      "loss/reg": 0.024180689826607704,
+      "step": 381
+    },
+    {
+      "epoch": 0.04775,
+      "grad_norm": 3.418168067932129,
+      "grad_norm_var": 1.6566847859375398,
+      "learning_rate": 0.0001,
+      "loss": 1.0532,
+      "loss/crossentropy": 2.1205835342407227,
+      "loss/hidden": 0.69921875,
+      "loss/logits": 0.11221310496330261,
+      "loss/reg": 0.02417258359491825,
+      "step": 382
+    },
+    {
+      "epoch": 0.047875,
+      "grad_norm": 3.3881125450134277,
+      "grad_norm_var": 1.093930487598201,
+      "learning_rate": 0.0001,
+      "loss": 0.9968,
+      "loss/crossentropy": 2.540379285812378,
+      "loss/hidden": 0.62890625,
+      "loss/logits": 0.1262589991092682,
+      "loss/reg": 0.024164721369743347,
+      "step": 383
+    },
+    {
+      "epoch": 0.048,
+      "grad_norm": 3.142486095428467,
+      "grad_norm_var": 1.1231981378319982,
+      "learning_rate": 0.0001,
+      "loss": 0.9186,
+      "loss/crossentropy": 2.5265657901763916,
+      "loss/hidden": 0.58203125,
+      "loss/logits": 0.0949624702334404,
+      "loss/reg": 0.02415630966424942,
+      "step": 384
+    },
+    {
+      "epoch": 0.048125,
+      "grad_norm": 7.363763332366943,
+      "grad_norm_var": 1.8443340238906671,
+      "learning_rate": 0.0001,
+      "loss": 1.1703,
+      "loss/crossentropy": 2.442409038543701,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.12408800423145294,
+      "loss/reg": 0.024148130789399147,
+      "step": 385
+    },
+    {
+      "epoch": 0.04825,
+      "grad_norm": 3.801536798477173,
+      "grad_norm_var": 1.7879312710551798,
+      "learning_rate": 0.0001,
+      "loss": 1.0129,
+      "loss/crossentropy": 2.461942434310913,
+      "loss/hidden": 0.640625,
+      "loss/logits": 0.13092291355133057,
+      "loss/reg": 0.024139659479260445,
+      "step": 386
+    },
+    {
+      "epoch": 0.048375,
+      "grad_norm": 4.24082612991333,
+      "grad_norm_var": 1.7228677295443293,
+      "learning_rate": 0.0001,
+      "loss": 1.1161,
+      "loss/crossentropy": 2.2889018058776855,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.12870003283023834,
+      "loss/reg": 0.024130841717123985,
+      "step": 387
+    },
+    {
+      "epoch": 0.0485,
+      "grad_norm": 4.190494060516357,
+      "grad_norm_var": 1.7195812284180172,
+      "learning_rate": 0.0001,
+      "loss": 1.1649,
+      "loss/crossentropy": 2.4475483894348145,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.16198021173477173,
+      "loss/reg": 0.024122456088662148,
+      "step": 388
+    },
+    {
+      "epoch": 0.048625,
+      "grad_norm": 15.206843376159668,
+      "grad_norm_var": 9.294742605150057,
+      "learning_rate": 0.0001,
+      "loss": 1.0606,
+      "loss/crossentropy": 2.3164589405059814,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.116313636302948,
+      "loss/reg": 0.024113710969686508,
+      "step": 389
+    },
+    {
+      "epoch": 0.04875,
+      "grad_norm": 10.421854972839355,
+      "grad_norm_var": 10.824103712955843,
+      "learning_rate": 0.0001,
+      "loss": 1.0638,
+      "loss/crossentropy": 2.1764075756073,
+      "loss/hidden": 0.70703125,
+      "loss/logits": 0.11567908525466919,
+      "loss/reg": 0.024104835465550423,
+      "step": 390
+    },
+    {
+      "epoch": 0.048875,
+      "grad_norm": 4.923640727996826,
+      "grad_norm_var": 10.83563756748113,
+      "learning_rate": 0.0001,
+      "loss": 0.9465,
+      "loss/crossentropy": 2.6762068271636963,
+      "loss/hidden": 0.6015625,
+      "loss/logits": 0.10399520397186279,
+      "loss/reg": 0.02409605123102665,
+      "step": 391
+    },
+    {
+      "epoch": 0.049,
+      "grad_norm": 4.840577125549316,
+      "grad_norm_var": 10.847422220224528,
+      "learning_rate": 0.0001,
+      "loss": 1.2037,
+      "loss/crossentropy": 2.482144832611084,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.15031081438064575,
+      "loss/reg": 0.02408732660114765,
+      "step": 392
+    },
+    {
+      "epoch": 0.049125,
+      "grad_norm": 2.9535470008850098,
+      "grad_norm_var": 10.811063470934846,
+      "learning_rate": 0.0001,
+      "loss": 1.0811,
+      "loss/crossentropy": 2.5998470783233643,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.12545132637023926,
+      "loss/reg": 0.024078134447336197,
+      "step": 393
+    },
+    {
+      "epoch": 0.04925,
+      "grad_norm": 4.08555793762207,
+      "grad_norm_var": 10.768160950066,
+      "learning_rate": 0.0001,
+      "loss": 1.3211,
+      "loss/crossentropy": 2.611746072769165,
+      "loss/hidden": 0.90625,
+      "loss/logits": 0.17412351071834564,
+      "loss/reg": 0.02406897209584713,
+      "step": 394
+    },
+    {
+      "epoch": 0.049375,
+      "grad_norm": 4.730119705200195,
+      "grad_norm_var": 10.582242923890295,
+      "learning_rate": 0.0001,
+      "loss": 0.9463,
+      "loss/crossentropy": 2.5379066467285156,
+      "loss/hidden": 0.59765625,
+      "loss/logits": 0.10806328058242798,
+      "loss/reg": 0.024059420451521873,
+      "step": 395
+    },
+    {
+      "epoch": 0.0495,
+      "grad_norm": 4.012064456939697,
+      "grad_norm_var": 10.523956759484621,
+      "learning_rate": 0.0001,
+      "loss": 0.9181,
+      "loss/crossentropy": 2.567375898361206,
+      "loss/hidden": 0.57421875,
+      "loss/logits": 0.10337453335523605,
+      "loss/reg": 0.024049852043390274,
+      "step": 396
+    },
+    {
+      "epoch": 0.049625,
+      "grad_norm": 4.57706880569458,
+      "grad_norm_var": 10.32746400090784,
+      "learning_rate": 0.0001,
+      "loss": 1.051,
+      "loss/crossentropy": 2.6955151557922363,
+      "loss/hidden": 0.6875,
+      "loss/logits": 0.12309969961643219,
+      "loss/reg": 0.02403969317674637,
+      "step": 397
+    },
+    {
+      "epoch": 0.04975,
+      "grad_norm": 2.7931346893310547,
+      "grad_norm_var": 10.511295288820635,
+      "learning_rate": 0.0001,
+      "loss": 1.0373,
+      "loss/crossentropy": 2.34134578704834,
+      "loss/hidden": 0.66796875,
+      "loss/logits": 0.12901151180267334,
+      "loss/reg": 0.02403116784989834,
+      "step": 398
+    },
+    {
+      "epoch": 0.049875,
+      "grad_norm": 2.4789700508117676,
+      "grad_norm_var": 10.793738555266915,
+      "learning_rate": 0.0001,
+      "loss": 1.1463,
+      "loss/crossentropy": 2.1852893829345703,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.15216518938541412,
+      "loss/reg": 0.024022690951824188,
+      "step": 399
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.340605735778809,
+      "grad_norm_var": 10.482396698239997,
+      "learning_rate": 0.0001,
+      "loss": 1.2874,
+      "loss/crossentropy": 2.4223577976226807,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.13714072108268738,
+      "loss/reg": 0.024014031514525414,
+      "step": 400
+    },
+    {
+      "epoch": 0.050125,
+      "grad_norm": 5.600348949432373,
+      "grad_norm_var": 10.208567826877847,
+      "learning_rate": 0.0001,
+      "loss": 0.9905,
+      "loss/crossentropy": 2.494180917739868,
+      "loss/hidden": 0.64453125,
+      "loss/logits": 0.10594967007637024,
+      "loss/reg": 0.024006787687540054,
+      "step": 401
+    },
+    {
+      "epoch": 0.05025,
+      "grad_norm": 5.094600677490234,
+      "grad_norm_var": 10.061216488426146,
+      "learning_rate": 0.0001,
+      "loss": 1.2828,
+      "loss/crossentropy": 2.1899051666259766,
+      "loss/hidden": 0.89453125,
+      "loss/logits": 0.1482730507850647,
+      "loss/reg": 0.02399739809334278,
+      "step": 402
+    },
+    {
+      "epoch": 0.050375,
+      "grad_norm": 3.915607452392578,
+      "grad_norm_var": 10.115626051260246,
+      "learning_rate": 0.0001,
+      "loss": 1.0375,
+      "loss/crossentropy": 2.790240526199341,
+      "loss/hidden": 0.65625,
+      "loss/logits": 0.14136558771133423,
+      "loss/reg": 0.02398892305791378,
+      "step": 403
+    },
+    {
+      "epoch": 0.0505,
+      "grad_norm": 70.49039459228516,
+      "grad_norm_var": 274.8357269833382,
+      "learning_rate": 0.0001,
+      "loss": 1.1534,
+      "loss/crossentropy": 2.2490358352661133,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.12061962485313416,
+      "loss/reg": 0.02398114837706089,
+      "step": 404
+    },
+    {
+      "epoch": 0.050625,
+      "grad_norm": 6.5985307693481445,
+      "grad_norm_var": 272.87861181728414,
+      "learning_rate": 0.0001,
+      "loss": 1.1479,
+      "loss/crossentropy": 2.856632709503174,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.12297768890857697,
+      "loss/reg": 0.0239717997610569,
+      "step": 405
+    },
+    {
+      "epoch": 0.05075,
+      "grad_norm": 4.041878700256348,
+      "grad_norm_var": 274.1523084795167,
+      "learning_rate": 0.0001,
+      "loss": 1.1408,
+      "loss/crossentropy": 2.5744457244873047,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.13939061760902405,
+      "loss/reg": 0.02396412193775177,
+      "step": 406
+    },
+    {
+      "epoch": 0.050875,
+      "grad_norm": 3.5284969806671143,
+      "grad_norm_var": 274.94477307618513,
+      "learning_rate": 0.0001,
+      "loss": 1.0838,
+      "loss/crossentropy": 2.7705729007720947,
+      "loss/hidden": 0.72265625,
+      "loss/logits": 0.12160193920135498,
+      "loss/reg": 0.023956267163157463,
+      "step": 407
+    },
+    {
+      "epoch": 0.051,
+      "grad_norm": 3.867558240890503,
+      "grad_norm_var": 275.4712566581114,
+      "learning_rate": 0.0001,
+      "loss": 1.1238,
+      "loss/crossentropy": 2.2807302474975586,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.1304117739200592,
+      "loss/reg": 0.023948216810822487,
+      "step": 408
+    },
+    {
+      "epoch": 0.051125,
+      "grad_norm": 4.9265217781066895,
+      "grad_norm_var": 274.2865770164501,
+      "learning_rate": 0.0001,
+      "loss": 0.955,
+      "loss/crossentropy": 2.6176843643188477,
+      "loss/hidden": 0.62890625,
+      "loss/logits": 0.08674542605876923,
+      "loss/reg": 0.02393944188952446,
+      "step": 409
+    },
+    {
+      "epoch": 0.05125,
+      "grad_norm": 3.1925463676452637,
+      "grad_norm_var": 274.86264478448203,
+      "learning_rate": 0.0001,
+      "loss": 1.0329,
+      "loss/crossentropy": 2.790286064147949,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.11385629326105118,
+      "loss/reg": 0.02393159456551075,
+      "step": 410
+    },
+    {
+      "epoch": 0.051375,
+      "grad_norm": 2.6993377208709717,
+      "grad_norm_var": 276.12743945534294,
+      "learning_rate": 0.0001,
+      "loss": 0.9268,
+      "loss/crossentropy": 2.5643763542175293,
+      "loss/hidden": 0.58203125,
+      "loss/logits": 0.10557639598846436,
+      "loss/reg": 0.0239238403737545,
+      "step": 411
+    },
+    {
+      "epoch": 0.0515,
+      "grad_norm": 4.7841315269470215,
+      "grad_norm_var": 275.72098389943244,
+      "learning_rate": 0.0001,
+      "loss": 1.3399,
+      "loss/crossentropy": 2.3737518787384033,
+      "loss/hidden": 0.89453125,
+      "loss/logits": 0.2062419056892395,
+      "loss/reg": 0.023915138095617294,
+      "step": 412
+    },
+    {
+      "epoch": 0.051625,
+      "grad_norm": 2.8743667602539062,
+      "grad_norm_var": 276.7634192046356,
+      "learning_rate": 0.0001,
+      "loss": 1.105,
+      "loss/crossentropy": 2.225154399871826,
+      "loss/hidden": 0.734375,
+      "loss/logits": 0.13158033788204193,
+      "loss/reg": 0.023907041177153587,
+      "step": 413
+    },
+    {
+      "epoch": 0.05175,
+      "grad_norm": 4.873403072357178,
+      "grad_norm_var": 275.51638736026473,
+      "learning_rate": 0.0001,
+      "loss": 1.1328,
+      "loss/crossentropy": 2.444675922393799,
+      "loss/hidden": 0.734375,
+      "loss/logits": 0.15947584807872772,
+      "loss/reg": 0.023899447172880173,
+      "step": 414
+    },
+    {
+      "epoch": 0.051875,
+      "grad_norm": 3.6676955223083496,
+      "grad_norm_var": 274.6671585398764,
+      "learning_rate": 0.0001,
+      "loss": 0.9504,
+      "loss/crossentropy": 2.6828341484069824,
+      "loss/hidden": 0.6015625,
+      "loss/logits": 0.10996139049530029,
+      "loss/reg": 0.023892199620604515,
+      "step": 415
+    },
+    {
+      "epoch": 0.052,
+      "grad_norm": 2.703623056411743,
+      "grad_norm_var": 276.2015243387772,
+      "learning_rate": 0.0001,
+      "loss": 1.049,
+      "loss/crossentropy": 2.4529809951782227,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.1305149495601654,
+      "loss/reg": 0.023883724585175514,
+      "step": 416
+    },
+    {
+      "epoch": 0.052125,
+      "grad_norm": 2.9288852214813232,
+      "grad_norm_var": 277.61048629826035,
+      "learning_rate": 0.0001,
+      "loss": 1.0151,
+      "loss/crossentropy": 2.6102705001831055,
+      "loss/hidden": 0.640625,
+      "loss/logits": 0.13567429780960083,
+      "loss/reg": 0.023875238373875618,
+      "step": 417
+    },
+    {
+      "epoch": 0.05225,
+      "grad_norm": 2.5006144046783447,
+      "grad_norm_var": 279.0831974622093,
+      "learning_rate": 0.0001,
+      "loss": 1.0571,
+      "loss/crossentropy": 2.457202672958374,
+      "loss/hidden": 0.69140625,
+      "loss/logits": 0.12700514495372772,
+      "loss/reg": 0.023867420852184296,
+      "step": 418
+    },
+    {
+      "epoch": 0.052375,
+      "grad_norm": 5.180849552154541,
+      "grad_norm_var": 278.49850212580674,
+      "learning_rate": 0.0001,
+      "loss": 1.1214,
+      "loss/crossentropy": 2.7064497470855713,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.13676053285598755,
+      "loss/reg": 0.02385888434946537,
+      "step": 419
+    },
+    {
+      "epoch": 0.0525,
+      "grad_norm": 3.2954249382019043,
+      "grad_norm_var": 1.3051375489769337,
+      "learning_rate": 0.0001,
+      "loss": 1.0717,
+      "loss/crossentropy": 2.31532621383667,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.13004590570926666,
+      "loss/reg": 0.023851698264479637,
+      "step": 420
+    },
+    {
+      "epoch": 0.052625,
+      "grad_norm": 2.8389148712158203,
+      "grad_norm_var": 0.8127685868335741,
+      "learning_rate": 0.0001,
+      "loss": 1.0781,
+      "loss/crossentropy": 2.3562088012695312,
+      "loss/hidden": 0.70703125,
+      "loss/logits": 0.13266143202781677,
+      "loss/reg": 0.023843195289373398,
+      "step": 421
+    },
+    {
+      "epoch": 0.05275,
+      "grad_norm": 4.581103324890137,
+      "grad_norm_var": 0.8613437167520175,
+      "learning_rate": 0.0001,
+      "loss": 1.4299,
+      "loss/crossentropy": 2.1004514694213867,
+      "loss/hidden": 1.0546875,
+      "loss/logits": 0.13683074712753296,
+      "loss/reg": 0.023835282772779465,
+      "step": 422
+    },
+    {
+      "epoch": 0.052875,
+      "grad_norm": 4.119724750518799,
+      "grad_norm_var": 0.8733982923945914,
+      "learning_rate": 0.0001,
+      "loss": 1.2152,
+      "loss/crossentropy": 2.640662670135498,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.12537327408790588,
+      "loss/reg": 0.02382684126496315,
+      "step": 423
+    },
+    {
+      "epoch": 0.053,
+      "grad_norm": 4.544858455657959,
+      "grad_norm_var": 0.918133871994677,
+      "learning_rate": 0.0001,
+      "loss": 1.0957,
+      "loss/crossentropy": 2.8237485885620117,
+      "loss/hidden": 0.7265625,
+      "loss/logits": 0.1309557855129242,
+      "loss/reg": 0.023818302899599075,
+      "step": 424
+    },
+    {
+      "epoch": 0.053125,
+      "grad_norm": 2.757870674133301,
+      "grad_norm_var": 0.8666742418813403,
+      "learning_rate": 0.0001,
+      "loss": 0.962,
+      "loss/crossentropy": 2.5533342361450195,
+      "loss/hidden": 0.62109375,
+      "loss/logits": 0.10283903032541275,
+      "loss/reg": 0.023810207843780518,
+      "step": 425
+    },
+    {
+      "epoch": 0.05325,
+      "grad_norm": 2.8035690784454346,
+      "grad_norm_var": 0.8970790990362765,
+      "learning_rate": 0.0001,
+      "loss": 1.0877,
+      "loss/crossentropy": 2.610870361328125,
+      "loss/hidden": 0.6953125,
+      "loss/logits": 0.1544169783592224,
+      "loss/reg": 0.02380150742828846,
+      "step": 426
+    },
+    {
+      "epoch": 0.053375,
+      "grad_norm": 3.5559544563293457,
+      "grad_norm_var": 0.8432525593756196,
+      "learning_rate": 0.0001,
+      "loss": 1.1169,
+      "loss/crossentropy": 2.537252187728882,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.1328693926334381,
+      "loss/reg": 0.02379263937473297,
+      "step": 427
+    },
+    {
+      "epoch": 0.0535,
+      "grad_norm": 3.58505916595459,
+      "grad_norm_var": 0.7479056021171611,
+      "learning_rate": 0.0001,
+      "loss": 1.0427,
+      "loss/crossentropy": 2.240241289138794,
+      "loss/hidden": 0.68359375,
+      "loss/logits": 0.1212427169084549,
+      "loss/reg": 0.02378367818892002,
+      "step": 428
+    },
+    {
+      "epoch": 0.053625,
+      "grad_norm": 3.6646056175231934,
+      "grad_norm_var": 0.7156687449512967,
+      "learning_rate": 0.0001,
+      "loss": 1.0166,
+      "loss/crossentropy": 2.765550374984741,
+      "loss/hidden": 0.6328125,
+      "loss/logits": 0.14600837230682373,
+      "loss/reg": 0.023774517700076103,
+      "step": 429
+    },
+    {
+      "epoch": 0.05375,
+      "grad_norm": 3.2596421241760254,
+      "grad_norm_var": 0.6044660126437359,
+      "learning_rate": 0.0001,
+      "loss": 1.13,
+      "loss/crossentropy": 2.1970372200012207,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.13451042771339417,
+      "loss/reg": 0.023765094578266144,
+      "step": 430
+    },
+    {
+      "epoch": 0.053875,
+      "grad_norm": 4.039770603179932,
+      "grad_norm_var": 0.6214738630237046,
+      "learning_rate": 0.0001,
+      "loss": 1.2708,
+      "loss/crossentropy": 2.5254645347595215,
+      "loss/hidden": 0.83984375,
+      "loss/logits": 0.19336232542991638,
+      "loss/reg": 0.023756500333547592,
+      "step": 431
+    },
+    {
+      "epoch": 0.054,
+      "grad_norm": 2.9176530838012695,
+      "grad_norm_var": 0.6009675102137348,
+      "learning_rate": 0.0001,
+      "loss": 0.9399,
+      "loss/crossentropy": 2.611178398132324,
+      "loss/hidden": 0.59375,
+      "loss/logits": 0.10868553817272186,
+      "loss/reg": 0.023747922852635384,
+      "step": 432
+    },
+    {
+      "epoch": 0.054125,
+      "grad_norm": 3.540189027786255,
+      "grad_norm_var": 0.5748467113480954,
+      "learning_rate": 0.0001,
+      "loss": 1.164,
+      "loss/crossentropy": 2.4766628742218018,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.1258353739976883,
+      "loss/reg": 0.023739352822303772,
+      "step": 433
+    },
+    {
+      "epoch": 0.05425,
+      "grad_norm": 3.7053966522216797,
+      "grad_norm_var": 0.4931212433281331,
+      "learning_rate": 0.0001,
+      "loss": 1.0522,
+      "loss/crossentropy": 2.491478681564331,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.13909485936164856,
+      "loss/reg": 0.023731039837002754,
+      "step": 434
+    },
+    {
+      "epoch": 0.054375,
+      "grad_norm": 3.4101648330688477,
+      "grad_norm_var": 0.32751985750053336,
+      "learning_rate": 0.0001,
+      "loss": 0.9527,
+      "loss/crossentropy": 2.5495922565460205,
+      "loss/hidden": 0.59765625,
+      "loss/logits": 0.1178436130285263,
+      "loss/reg": 0.023722674697637558,
+      "step": 435
+    },
+    {
+      "epoch": 0.0545,
+      "grad_norm": 3.923790693283081,
+      "grad_norm_var": 0.33181180777142816,
+      "learning_rate": 0.0001,
+      "loss": 1.1217,
+      "loss/crossentropy": 2.350161552429199,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.13453412055969238,
+      "loss/reg": 0.02371453307569027,
+      "step": 436
+    },
+    {
+      "epoch": 0.054625,
+      "grad_norm": 3.2052547931671143,
+      "grad_norm_var": 0.3040979482718351,
+      "learning_rate": 0.0001,
+      "loss": 1.0641,
+      "loss/crossentropy": 2.5055789947509766,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.11222882568836212,
+      "loss/reg": 0.023706616833806038,
+      "step": 437
+    },
+    {
+      "epoch": 0.05475,
+      "grad_norm": 2.820063591003418,
+      "grad_norm_var": 0.26777286633351405,
+      "learning_rate": 0.0001,
+      "loss": 1.1811,
+      "loss/crossentropy": 2.535069704055786,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.1628357172012329,
+      "loss/reg": 0.023698095232248306,
+      "step": 438
+    },
+    {
+      "epoch": 0.054875,
+      "grad_norm": 3.095841884613037,
+      "grad_norm_var": 0.24744105333314317,
+      "learning_rate": 0.0001,
+      "loss": 1.1936,
+      "loss/crossentropy": 2.313276767730713,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.17543524503707886,
+      "loss/reg": 0.02368931844830513,
+      "step": 439
+    },
+    {
+      "epoch": 0.055,
+      "grad_norm": 2.955540895462036,
+      "grad_norm_var": 0.1683967569747227,
+      "learning_rate": 0.0001,
+      "loss": 0.991,
+      "loss/crossentropy": 2.747128486633301,
+      "loss/hidden": 0.6484375,
+      "loss/logits": 0.10580303519964218,
+      "loss/reg": 0.023680580779910088,
+      "step": 440
+    },
+    {
+      "epoch": 0.055125,
+      "grad_norm": 2.866759777069092,
+      "grad_norm_var": 0.16086728592038804,
+      "learning_rate": 0.0001,
+      "loss": 0.9014,
+      "loss/crossentropy": 2.4465224742889404,
+      "loss/hidden": 0.5703125,
+      "loss/logits": 0.09433356672525406,
+      "loss/reg": 0.023672088980674744,
+      "step": 441
+    },
+    {
+      "epoch": 0.05525,
+      "grad_norm": 3.1793012619018555,
+      "grad_norm_var": 0.14310091597801508,
+      "learning_rate": 0.0001,
+      "loss": 1.0742,
+      "loss/crossentropy": 2.364579677581787,
+      "loss/hidden": 0.6875,
+      "loss/logits": 0.15003816783428192,
+      "loss/reg": 0.023663459345698357,
+      "step": 442
+    },
+    {
+      "epoch": 0.055375,
+      "grad_norm": 4.3040900230407715,
+      "grad_norm_var": 0.19784760386154687,
+      "learning_rate": 0.0001,
+      "loss": 1.1406,
+      "loss/crossentropy": 2.955233573913574,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.1579587459564209,
+      "loss/reg": 0.023654496297240257,
+      "step": 443
+    },
+    {
+      "epoch": 0.0555,
+      "grad_norm": 3.6495676040649414,
+      "grad_norm_var": 0.19966009525053988,
+      "learning_rate": 0.0001,
+      "loss": 1.0678,
+      "loss/crossentropy": 2.184626340866089,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.11654888093471527,
+      "loss/reg": 0.023645464330911636,
+      "step": 444
+    },
+    {
+      "epoch": 0.055625,
+      "grad_norm": 2.7992637157440186,
+      "grad_norm_var": 0.21692371557562992,
+      "learning_rate": 0.0001,
+      "loss": 1.0746,
+      "loss/crossentropy": 1.9573417901992798,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.1234317272901535,
+      "loss/reg": 0.023636594414711,
+      "step": 445
+    },
+    {
+      "epoch": 0.05575,
+      "grad_norm": 3.4131267070770264,
+      "grad_norm_var": 0.21645445922388262,
+      "learning_rate": 0.0001,
+      "loss": 1.0938,
+      "loss/crossentropy": 2.3615517616271973,
+      "loss/hidden": 0.71875,
+      "loss/logits": 0.1387380063533783,
+      "loss/reg": 0.023627731949090958,
+      "step": 446
+    },
+    {
+      "epoch": 0.055875,
+      "grad_norm": 2.890436887741089,
+      "grad_norm_var": 0.19547383544341201,
+      "learning_rate": 0.0001,
+      "loss": 0.8903,
+      "loss/crossentropy": 2.2940785884857178,
+      "loss/hidden": 0.5546875,
+      "loss/logits": 0.09945578873157501,
+      "loss/reg": 0.023619333282113075,
+      "step": 447
+    },
+    {
+      "epoch": 0.056,
+      "grad_norm": 3.9202187061309814,
+      "grad_norm_var": 0.20821686288428035,
+      "learning_rate": 0.0001,
+      "loss": 1.1124,
+      "loss/crossentropy": 2.561680316925049,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.13801740109920502,
+      "loss/reg": 0.023610329255461693,
+      "step": 448
+    },
+    {
+      "epoch": 0.056125,
+      "grad_norm": 3.0585479736328125,
+      "grad_norm_var": 0.21081889060941586,
+      "learning_rate": 0.0001,
+      "loss": 0.9545,
+      "loss/crossentropy": 2.4095382690429688,
+      "loss/hidden": 0.609375,
+      "loss/logits": 0.10911431908607483,
+      "loss/reg": 0.02360081672668457,
+      "step": 449
+    },
+    {
+      "epoch": 0.05625,
+      "grad_norm": 3.3532161712646484,
+      "grad_norm_var": 0.2007006666523369,
+      "learning_rate": 0.0001,
+      "loss": 1.1045,
+      "loss/crossentropy": 1.8727831840515137,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.1264159381389618,
+      "loss/reg": 0.023592744022607803,
+      "step": 450
+    },
+    {
+      "epoch": 0.056375,
+      "grad_norm": 3.1456074714660645,
+      "grad_norm_var": 0.20128870800301848,
+      "learning_rate": 0.0001,
+      "loss": 0.9574,
+      "loss/crossentropy": 2.4792492389678955,
+      "loss/hidden": 0.609375,
+      "loss/logits": 0.11215440928936005,
+      "loss/reg": 0.02358343079686165,
+      "step": 451
+    },
+    {
+      "epoch": 0.0565,
+      "grad_norm": 3.402637004852295,
+      "grad_norm_var": 0.1739656178124496,
+      "learning_rate": 0.0001,
+      "loss": 1.0743,
+      "loss/crossentropy": 2.3384110927581787,
+      "loss/hidden": 0.6875,
+      "loss/logits": 0.15106429159641266,
+      "loss/reg": 0.02357417158782482,
+      "step": 452
+    },
+    {
+      "epoch": 0.056625,
+      "grad_norm": 3.259817361831665,
+      "grad_norm_var": 0.17379912081048493,
+      "learning_rate": 0.0001,
+      "loss": 1.1221,
+      "loss/crossentropy": 2.456613779067993,
+      "loss/hidden": 0.734375,
+      "loss/logits": 0.15212517976760864,
+      "loss/reg": 0.023564757779240608,
+      "step": 453
+    },
+    {
+      "epoch": 0.05675,
+      "grad_norm": 2.531987190246582,
+      "grad_norm_var": 0.195773570863138,
+      "learning_rate": 0.0001,
+      "loss": 0.9605,
+      "loss/crossentropy": 2.394343376159668,
+      "loss/hidden": 0.609375,
+      "loss/logits": 0.11555634438991547,
+      "loss/reg": 0.023556271567940712,
+      "step": 454
+    },
+    {
+      "epoch": 0.056875,
+      "grad_norm": 2.594336986541748,
+      "grad_norm_var": 0.22107356191806862,
+      "learning_rate": 0.0001,
+      "loss": 0.9207,
+      "loss/crossentropy": 2.781731367111206,
+      "loss/hidden": 0.58203125,
+      "loss/logits": 0.10317155718803406,
+      "loss/reg": 0.023547139018774033,
+      "step": 455
+    },
+    {
+      "epoch": 0.057,
+      "grad_norm": 3.1260924339294434,
+      "grad_norm_var": 0.21715561662650557,
+      "learning_rate": 0.0001,
+      "loss": 1.1637,
+      "loss/crossentropy": 2.5018725395202637,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.14706720411777496,
+      "loss/reg": 0.02353852428495884,
+      "step": 456
+    },
+    {
+      "epoch": 0.057125,
+      "grad_norm": 3.159911632537842,
+      "grad_norm_var": 0.20878072756432645,
+      "learning_rate": 0.0001,
+      "loss": 1.0676,
+      "loss/crossentropy": 2.668788433074951,
+      "loss/hidden": 0.68359375,
+      "loss/logits": 0.1487644910812378,
+      "loss/reg": 0.02352879010140896,
+      "step": 457
+    },
+    {
+      "epoch": 0.05725,
+      "grad_norm": 3.0937418937683105,
+      "grad_norm_var": 0.20989373673105333,
+      "learning_rate": 0.0001,
+      "loss": 1.0188,
+      "loss/crossentropy": 2.498286008834839,
+      "loss/hidden": 0.6640625,
+      "loss/logits": 0.11953231692314148,
+      "loss/reg": 0.023519227281212807,
+      "step": 458
+    },
+    {
+      "epoch": 0.057375,
+      "grad_norm": 2.4465949535369873,
+      "grad_norm_var": 0.15987096754079838,
+      "learning_rate": 0.0001,
+      "loss": 1.0333,
+      "loss/crossentropy": 2.401630401611328,
+      "loss/hidden": 0.671875,
+      "loss/logits": 0.12631601095199585,
+      "loss/reg": 0.023509083315730095,
+      "step": 459
+    },
+    {
+      "epoch": 0.0575,
+      "grad_norm": 3.8185830116271973,
+      "grad_norm_var": 0.17369585396981316,
+      "learning_rate": 0.0001,
+      "loss": 1.2185,
+      "loss/crossentropy": 2.3343100547790527,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.13973468542099,
+      "loss/reg": 0.023500461131334305,
+      "step": 460
+    },
+    {
+      "epoch": 0.057625,
+      "grad_norm": 2.769894599914551,
+      "grad_norm_var": 0.1750287637092998,
+      "learning_rate": 0.0001,
+      "loss": 1.0478,
+      "loss/crossentropy": 2.594421148300171,
+      "loss/hidden": 0.66796875,
+      "loss/logits": 0.14495806396007538,
+      "loss/reg": 0.02349086105823517,
+      "step": 461
+    },
+    {
+      "epoch": 0.05775,
+      "grad_norm": 3.924386501312256,
+      "grad_norm_var": 0.21107140664515026,
+      "learning_rate": 0.0001,
+      "loss": 1.1201,
+      "loss/crossentropy": 1.8543033599853516,
+      "loss/hidden": 0.71875,
+      "loss/logits": 0.16654378175735474,
+      "loss/reg": 0.023482073098421097,
+      "step": 462
+    },
+    {
+      "epoch": 0.057875,
+      "grad_norm": 2.719325304031372,
+      "grad_norm_var": 0.21896016035892957,
+      "learning_rate": 0.0001,
+      "loss": 0.9973,
+      "loss/crossentropy": 2.7110376358032227,
+      "loss/hidden": 0.6328125,
+      "loss/logits": 0.1297917366027832,
+      "loss/reg": 0.02347267046570778,
+      "step": 463
+    },
+    {
+      "epoch": 0.058,
+      "grad_norm": 3.110532522201538,
+      "grad_norm_var": 0.17627651595159174,
+      "learning_rate": 0.0001,
+      "loss": 0.9503,
+      "loss/crossentropy": 2.424137830734253,
+      "loss/hidden": 0.61328125,
+      "loss/logits": 0.10241679847240448,
+      "loss/reg": 0.023463333025574684,
+      "step": 464
+    },
+    {
+      "epoch": 0.058125,
+      "grad_norm": 3.2945988178253174,
+      "grad_norm_var": 0.17862116157392785,
+      "learning_rate": 0.0001,
+      "loss": 1.1634,
+      "loss/crossentropy": 2.8107378482818604,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.17888996005058289,
+      "loss/reg": 0.0234534852206707,
+      "step": 465
+    },
+    {
+      "epoch": 0.05825,
+      "grad_norm": 4.864523887634277,
+      "grad_norm_var": 0.37049430510921844,
+      "learning_rate": 0.0001,
+      "loss": 1.0812,
+      "loss/crossentropy": 2.5031394958496094,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.13192051649093628,
+      "loss/reg": 0.023444540798664093,
+      "step": 466
+    },
+    {
+      "epoch": 0.058375,
+      "grad_norm": 3.8722984790802,
+      "grad_norm_var": 0.3978501673810001,
+      "learning_rate": 0.0001,
+      "loss": 1.2316,
+      "loss/crossentropy": 1.9772007465362549,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.14563970267772675,
+      "loss/reg": 0.023435747250914574,
+      "step": 467
+    },
+    {
+      "epoch": 0.0585,
+      "grad_norm": 4.621346473693848,
+      "grad_norm_var": 0.5155902021721573,
+      "learning_rate": 0.0001,
+      "loss": 1.1725,
+      "loss/crossentropy": 2.4977753162384033,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.149122953414917,
+      "loss/reg": 0.023427119478583336,
+      "step": 468
+    },
+    {
+      "epoch": 0.058625,
+      "grad_norm": 3.36370849609375,
+      "grad_norm_var": 0.5153549660191058,
+      "learning_rate": 0.0001,
+      "loss": 1.2411,
+      "loss/crossentropy": 2.4750683307647705,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.18663738667964935,
+      "loss/reg": 0.023417862132191658,
+      "step": 469
+    },
+    {
+      "epoch": 0.05875,
+      "grad_norm": 3.391871690750122,
+      "grad_norm_var": 0.46984604899865395,
+      "learning_rate": 0.0001,
+      "loss": 1.1066,
+      "loss/crossentropy": 2.5189239978790283,
+      "loss/hidden": 0.71875,
+      "loss/logits": 0.1537722945213318,
+      "loss/reg": 0.023408619686961174,
+      "step": 470
+    },
+    {
+      "epoch": 0.058875,
+      "grad_norm": 3.903122901916504,
+      "grad_norm_var": 0.43880097595690587,
+      "learning_rate": 0.0001,
+      "loss": 0.9258,
+      "loss/crossentropy": 2.548199415206909,
+      "loss/hidden": 0.59375,
+      "loss/logits": 0.09802491217851639,
+      "loss/reg": 0.02339930646121502,
+      "step": 471
+    },
+    {
+      "epoch": 0.059,
+      "grad_norm": 5.576291084289551,
+      "grad_norm_var": 0.7024716555345464,
+      "learning_rate": 0.0001,
+      "loss": 1.2843,
+      "loss/crossentropy": 2.682227849960327,
+      "loss/hidden": 0.859375,
+      "loss/logits": 0.19106051325798035,
+      "loss/reg": 0.023390140384435654,
+      "step": 472
+    },
+    {
+      "epoch": 0.059125,
+      "grad_norm": 2.4201464653015137,
+      "grad_norm_var": 0.7821220779043936,
+      "learning_rate": 0.0001,
+      "loss": 0.9341,
+      "loss/crossentropy": 2.5251269340515137,
+      "loss/hidden": 0.59375,
+      "loss/logits": 0.10658347606658936,
+      "loss/reg": 0.023381320759654045,
+      "step": 473
+    },
+    {
+      "epoch": 0.05925,
+      "grad_norm": 4.325901985168457,
+      "grad_norm_var": 0.7980385459591806,
+      "learning_rate": 0.0001,
+      "loss": 0.9798,
+      "loss/crossentropy": 2.3592865467071533,
+      "loss/hidden": 0.640625,
+      "loss/logits": 0.10548710823059082,
+      "loss/reg": 0.023372096940875053,
+      "step": 474
+    },
+    {
+      "epoch": 0.059375,
+      "grad_norm": 3.1382455825805664,
+      "grad_norm_var": 0.7168259193102716,
+      "learning_rate": 0.0001,
+      "loss": 0.9839,
+      "loss/crossentropy": 2.6056907176971436,
+      "loss/hidden": 0.62890625,
+      "loss/logits": 0.121395543217659,
+      "loss/reg": 0.023363398388028145,
+      "step": 475
+    },
+    {
+      "epoch": 0.0595,
+      "grad_norm": 3.0779755115509033,
+      "grad_norm_var": 0.7388713721113239,
+      "learning_rate": 0.0001,
+      "loss": 1.0798,
+      "loss/crossentropy": 2.320854663848877,
+      "loss/hidden": 0.72265625,
+      "loss/logits": 0.12356055527925491,
+      "loss/reg": 0.023354284465312958,
+      "step": 476
+    },
+    {
+      "epoch": 0.059625,
+      "grad_norm": 3.293567657470703,
+      "grad_norm_var": 0.6946720185858983,
+      "learning_rate": 0.0001,
+      "loss": 0.9685,
+      "loss/crossentropy": 2.7064430713653564,
+      "loss/hidden": 0.625,
+      "loss/logits": 0.11008161306381226,
+      "loss/reg": 0.023345019668340683,
+      "step": 477
+    },
+    {
+      "epoch": 0.05975,
+      "grad_norm": 2.664088249206543,
+      "grad_norm_var": 0.7530647477645431,
+      "learning_rate": 0.0001,
+      "loss": 0.9101,
+      "loss/crossentropy": 2.6147005558013916,
+      "loss/hidden": 0.5703125,
+      "loss/logits": 0.10643748193979263,
+      "loss/reg": 0.02333623729646206,
+      "step": 478
+    },
+    {
+      "epoch": 0.059875,
+      "grad_norm": 2.9174795150756836,
+      "grad_norm_var": 0.7321888983535926,
+      "learning_rate": 0.0001,
+      "loss": 0.9342,
+      "loss/crossentropy": 2.566849946975708,
+      "loss/hidden": 0.58984375,
+      "loss/logits": 0.11108069121837616,
+      "loss/reg": 0.023327510803937912,
+      "step": 479
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.3596463203430176,
+      "grad_norm_var": 0.71932045702877,
+      "learning_rate": 0.0001,
+      "loss": 1.0961,
+      "loss/crossentropy": 2.645395040512085,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.15979796648025513,
+      "loss/reg": 0.023318573832511902,
+      "step": 480
+    },
+    {
+      "epoch": 0.060125,
+      "grad_norm": 4.0903096199035645,
+      "grad_norm_var": 0.7232764591548666,
+      "learning_rate": 0.0001,
+      "loss": 1.1686,
+      "loss/crossentropy": 2.077467918395996,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.14252969622612,
+      "loss/reg": 0.02330981194972992,
+      "step": 481
+    },
+    {
+      "epoch": 0.06025,
+      "grad_norm": 3.277656078338623,
+      "grad_norm_var": 0.6300433507978689,
+      "learning_rate": 0.0001,
+      "loss": 1.1136,
+      "loss/crossentropy": 2.588385581970215,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.13843819499015808,
+      "loss/reg": 0.02330118976533413,
+      "step": 482
+    },
+    {
+      "epoch": 0.060375,
+      "grad_norm": 4.117528915405273,
+      "grad_norm_var": 0.6433314640873874,
+      "learning_rate": 0.0001,
+      "loss": 1.0239,
+      "loss/crossentropy": 2.766176700592041,
+      "loss/hidden": 0.6640625,
+      "loss/logits": 0.12686920166015625,
+      "loss/reg": 0.023292165249586105,
+      "step": 483
+    },
+    {
+      "epoch": 0.0605,
+      "grad_norm": 11.29445743560791,
+      "grad_norm_var": 4.338621670503842,
+      "learning_rate": 0.0001,
+      "loss": 1.9614,
+      "loss/crossentropy": 2.5279059410095215,
+      "loss/hidden": 1.2109375,
+      "loss/logits": 0.5176718235015869,
+      "loss/reg": 0.02328311838209629,
+      "step": 484
+    },
+    {
+      "epoch": 0.060625,
+      "grad_norm": 3.5054728984832764,
+      "grad_norm_var": 4.327600163307723,
+      "learning_rate": 0.0001,
+      "loss": 1.0719,
+      "loss/crossentropy": 2.726095199584961,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.12821289896965027,
+      "loss/reg": 0.023273879662156105,
+      "step": 485
+    },
+    {
+      "epoch": 0.06075,
+      "grad_norm": 3.157118558883667,
+      "grad_norm_var": 4.350771203860321,
+      "learning_rate": 0.0001,
+      "loss": 0.9326,
+      "loss/crossentropy": 2.207131862640381,
+      "loss/hidden": 0.60546875,
+      "loss/logits": 0.09451892971992493,
+      "loss/reg": 0.02326469123363495,
+      "step": 486
+    },
+    {
+      "epoch": 0.060875,
+      "grad_norm": 5.006563186645508,
+      "grad_norm_var": 4.411522578027558,
+      "learning_rate": 0.0001,
+      "loss": 1.1884,
+      "loss/crossentropy": 3.005479335784912,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.17073442041873932,
+      "loss/reg": 0.02325539104640484,
+      "step": 487
+    },
+    {
+      "epoch": 0.061,
+      "grad_norm": 8.253157615661621,
+      "grad_norm_var": 5.3947068177792,
+      "learning_rate": 0.0001,
+      "loss": 1.4073,
+      "loss/crossentropy": 2.329857349395752,
+      "loss/hidden": 0.984375,
+      "loss/logits": 0.19045758247375488,
+      "loss/reg": 0.02324584126472473,
+      "step": 488
+    },
+    {
+      "epoch": 0.061125,
+      "grad_norm": 4.795626640319824,
+      "grad_norm_var": 5.169810789054156,
+      "learning_rate": 0.0001,
+      "loss": 1.2274,
+      "loss/crossentropy": 2.354893207550049,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.14352190494537354,
+      "loss/reg": 0.023236218839883804,
+      "step": 489
+    },
+    {
+      "epoch": 0.06125,
+      "grad_norm": 3.1360483169555664,
+      "grad_norm_var": 5.268809256909969,
+      "learning_rate": 0.0001,
+      "loss": 1.0242,
+      "loss/crossentropy": 2.2810122966766357,
+      "loss/hidden": 0.671875,
+      "loss/logits": 0.12006018310785294,
+      "loss/reg": 0.02322734333574772,
+      "step": 490
+    },
+    {
+      "epoch": 0.061375,
+      "grad_norm": 3.931467056274414,
+      "grad_norm_var": 5.1833802842946906,
+      "learning_rate": 0.0001,
+      "loss": 1.1688,
+      "loss/crossentropy": 2.3372247219085693,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.15147234499454498,
+      "loss/reg": 0.023217879235744476,
+      "step": 491
+    },
+    {
+      "epoch": 0.0615,
+      "grad_norm": 3.8858096599578857,
+      "grad_norm_var": 5.085283642122107,
+      "learning_rate": 0.0001,
+      "loss": 1.0309,
+      "loss/crossentropy": 2.597487688064575,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.11914543062448502,
+      "loss/reg": 0.02320869080722332,
+      "step": 492
+    },
+    {
+      "epoch": 0.061625,
+      "grad_norm": 6.116000175476074,
+      "grad_norm_var": 5.160062314221837,
+      "learning_rate": 0.0001,
+      "loss": 1.2295,
+      "loss/crossentropy": 2.4041733741760254,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.14599129557609558,
+      "loss/reg": 0.023199014365673065,
+      "step": 493
+    },
+    {
+      "epoch": 0.06175,
+      "grad_norm": 3.4635026454925537,
+      "grad_norm_var": 4.994267697000357,
+      "learning_rate": 0.0001,
+      "loss": 1.0957,
+      "loss/crossentropy": 2.692230224609375,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.13329669833183289,
+      "loss/reg": 0.023189352825284004,
+      "step": 494
+    },
+    {
+      "epoch": 0.061875,
+      "grad_norm": 6.833379745483398,
+      "grad_norm_var": 5.051083471594066,
+      "learning_rate": 0.0001,
+      "loss": 1.1437,
+      "loss/crossentropy": 2.470660448074341,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.12278926372528076,
+      "loss/reg": 0.023179946467280388,
+      "step": 495
+    },
+    {
+      "epoch": 0.062,
+      "grad_norm": 3.2948107719421387,
+      "grad_norm_var": 5.064566926371495,
+      "learning_rate": 0.0001,
+      "loss": 1.109,
+      "loss/crossentropy": 2.3682029247283936,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.13507232069969177,
+      "loss/reg": 0.02317013218998909,
+      "step": 496
+    },
+    {
+      "epoch": 0.062125,
+      "grad_norm": 4.057919025421143,
+      "grad_norm_var": 5.068064269732227,
+      "learning_rate": 0.0001,
+      "loss": 1.1648,
+      "loss/crossentropy": 2.4153223037719727,
+      "loss/hidden": 0.765625,
+      "loss/logits": 0.16754823923110962,
+      "loss/reg": 0.02316114492714405,
+      "step": 497
+    },
+    {
+      "epoch": 0.06225,
+      "grad_norm": 3.3985679149627686,
+      "grad_norm_var": 5.043098814178749,
+      "learning_rate": 0.0001,
+      "loss": 1.1033,
+      "loss/crossentropy": 2.6683197021484375,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.13353273272514343,
+      "loss/reg": 0.023151271045207977,
+      "step": 498
+    },
+    {
+      "epoch": 0.062375,
+      "grad_norm": 2.6574859619140625,
+      "grad_norm_var": 5.326800856327217,
+      "learning_rate": 0.0001,
+      "loss": 0.9413,
+      "loss/crossentropy": 2.490849733352661,
+      "loss/hidden": 0.60546875,
+      "loss/logits": 0.1043705940246582,
+      "loss/reg": 0.023141290992498398,
+      "step": 499
+    },
+    {
+      "epoch": 0.0625,
+      "grad_norm": 3.2627739906311035,
+      "grad_norm_var": 2.402846049322009,
+      "learning_rate": 0.0001,
+      "loss": 1.0233,
+      "loss/crossentropy": 2.3429057598114014,
+      "loss/hidden": 0.63671875,
+      "loss/logits": 0.1552838534116745,
+      "loss/reg": 0.02313125506043434,
+      "step": 500
+    },
+    {
+      "epoch": 0.062625,
+      "grad_norm": 3.2453906536102295,
+      "grad_norm_var": 2.4345300369903553,
+      "learning_rate": 0.0001,
+      "loss": 1.1719,
+      "loss/crossentropy": 2.497878313064575,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.17116406559944153,
+      "loss/reg": 0.02312229759991169,
+      "step": 501
+    },
+    {
+      "epoch": 0.06275,
+      "grad_norm": 4.79340934753418,
+      "grad_norm_var": 2.3566760840149366,
+      "learning_rate": 0.0001,
+      "loss": 1.0001,
+      "loss/crossentropy": 2.360431671142578,
+      "loss/hidden": 0.6484375,
+      "loss/logits": 0.12051868438720703,
+      "loss/reg": 0.02311263047158718,
+      "step": 502
+    },
+    {
+      "epoch": 0.062875,
+      "grad_norm": 3.1595826148986816,
+      "grad_norm_var": 2.4163836713766425,
+      "learning_rate": 0.0001,
+      "loss": 0.9158,
+      "loss/crossentropy": 2.266618490219116,
+      "loss/hidden": 0.58203125,
+      "loss/logits": 0.1027822494506836,
+      "loss/reg": 0.023102767765522003,
+      "step": 503
+    },
+    {
+      "epoch": 0.063,
+      "grad_norm": 3.59019136428833,
+      "grad_norm_var": 1.2975304557542653,
+      "learning_rate": 0.0001,
+      "loss": 0.9685,
+      "loss/crossentropy": 2.6386334896087646,
+      "loss/hidden": 0.625,
+      "loss/logits": 0.11253425478935242,
+      "loss/reg": 0.023093828931450844,
+      "step": 504
+    },
+    {
+      "epoch": 0.063125,
+      "grad_norm": 3.1218326091766357,
+      "grad_norm_var": 1.2897946661694502,
+      "learning_rate": 0.0001,
+      "loss": 0.9493,
+      "loss/crossentropy": 2.2931203842163086,
+      "loss/hidden": 0.6171875,
+      "loss/logits": 0.10128200799226761,
+      "loss/reg": 0.0230838842689991,
+      "step": 505
+    },
+    {
+      "epoch": 0.06325,
+      "grad_norm": 7.243019104003906,
+      "grad_norm_var": 1.941121973828988,
+      "learning_rate": 0.0001,
+      "loss": 1.1363,
+      "loss/crossentropy": 2.510519504547119,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.11650878190994263,
+      "loss/reg": 0.023074399679899216,
+      "step": 506
+    },
+    {
+      "epoch": 0.063375,
+      "grad_norm": 2.7458910942077637,
+      "grad_norm_var": 2.0601092371510408,
+      "learning_rate": 0.0001,
+      "loss": 0.8688,
+      "loss/crossentropy": 2.504798173904419,
+      "loss/hidden": 0.54296875,
+      "loss/logits": 0.09517204016447067,
+      "loss/reg": 0.023064618930220604,
+      "step": 507
+    },
+    {
+      "epoch": 0.0635,
+      "grad_norm": 3.834894895553589,
+      "grad_norm_var": 2.061415401484546,
+      "learning_rate": 0.0001,
+      "loss": 1.0614,
+      "loss/crossentropy": 2.504178285598755,
+      "loss/hidden": 0.69140625,
+      "loss/logits": 0.13941214978694916,
+      "loss/reg": 0.023055192083120346,
+      "step": 508
+    },
+    {
+      "epoch": 0.063625,
+      "grad_norm": 3.0524418354034424,
+      "grad_norm_var": 1.8045701590720038,
+      "learning_rate": 0.0001,
+      "loss": 0.9395,
+      "loss/crossentropy": 2.6670830249786377,
+      "loss/hidden": 0.6015625,
+      "loss/logits": 0.10751838982105255,
+      "loss/reg": 0.023046277463436127,
+      "step": 509
+    },
+    {
+      "epoch": 0.06375,
+      "grad_norm": 2.638979196548462,
+      "grad_norm_var": 1.8906158947457992,
+      "learning_rate": 0.0001,
+      "loss": 0.9649,
+      "loss/crossentropy": 2.5606770515441895,
+      "loss/hidden": 0.62109375,
+      "loss/logits": 0.11341118812561035,
+      "loss/reg": 0.02303677424788475,
+      "step": 510
+    },
+    {
+      "epoch": 0.063875,
+      "grad_norm": 4.029105186462402,
+      "grad_norm_var": 1.2509737999906814,
+      "learning_rate": 0.0001,
+      "loss": 1.0378,
+      "loss/crossentropy": 2.446560859680176,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.09661944955587387,
+      "loss/reg": 0.023027852177619934,
+      "step": 511
+    },
+    {
+      "epoch": 0.064,
+      "grad_norm": 3.203378438949585,
+      "grad_norm_var": 1.255617850639648,
+      "learning_rate": 0.0001,
+      "loss": 0.9705,
+      "loss/crossentropy": 2.08353328704834,
+      "loss/hidden": 0.63671875,
+      "loss/logits": 0.10362571477890015,
+      "loss/reg": 0.02301831543445587,
+      "step": 512
+    },
+    {
+      "epoch": 0.064125,
+      "grad_norm": 2.5737931728363037,
+      "grad_norm_var": 1.3080458668089554,
+      "learning_rate": 0.0001,
+      "loss": 1.0597,
+      "loss/crossentropy": 2.3520448207855225,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.11864635348320007,
+      "loss/reg": 0.023009376600384712,
+      "step": 513
+    },
+    {
+      "epoch": 0.06425,
+      "grad_norm": 3.6731107234954834,
+      "grad_norm_var": 1.307783724921588,
+      "learning_rate": 0.0001,
+      "loss": 1.1013,
+      "loss/crossentropy": 2.77113938331604,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.1408485472202301,
+      "loss/reg": 0.022999830543994904,
+      "step": 514
+    },
+    {
+      "epoch": 0.064375,
+      "grad_norm": 3.06605863571167,
+      "grad_norm_var": 1.269509965568249,
+      "learning_rate": 0.0001,
+      "loss": 0.9224,
+      "loss/crossentropy": 2.3147072792053223,
+      "loss/hidden": 0.59375,
+      "loss/logits": 0.0987289547920227,
+      "loss/reg": 0.022990131750702858,
+      "step": 515
+    },
+    {
+      "epoch": 0.0645,
+      "grad_norm": 3.462446689605713,
+      "grad_norm_var": 1.2636330593023397,
+      "learning_rate": 0.0001,
+      "loss": 1.0201,
+      "loss/crossentropy": 2.4042327404022217,
+      "loss/hidden": 0.66796875,
+      "loss/logits": 0.12231434136629105,
+      "loss/reg": 0.02298046089708805,
+      "step": 516
+    },
+    {
+      "epoch": 0.064625,
+      "grad_norm": 4.2029500007629395,
+      "grad_norm_var": 1.2769943636458339,
+      "learning_rate": 0.0001,
+      "loss": 1.222,
+      "loss/crossentropy": 2.514112710952759,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.16808617115020752,
+      "loss/reg": 0.022970519959926605,
+      "step": 517
+    },
+    {
+      "epoch": 0.06475,
+      "grad_norm": 3.433554172515869,
+      "grad_norm_var": 1.1851525686550586,
+      "learning_rate": 0.0001,
+      "loss": 1.2444,
+      "loss/crossentropy": 2.493224859237671,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.17107471823692322,
+      "loss/reg": 0.022961357608437538,
+      "step": 518
+    },
+    {
+      "epoch": 0.064875,
+      "grad_norm": 4.310100078582764,
+      "grad_norm_var": 1.2057753361074104,
+      "learning_rate": 0.0001,
+      "loss": 1.0108,
+      "loss/crossentropy": 2.5606930255889893,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.10546360909938812,
+      "loss/reg": 0.022952331230044365,
+      "step": 519
+    },
+    {
+      "epoch": 0.065,
+      "grad_norm": 3.674527883529663,
+      "grad_norm_var": 1.2057007253632908,
+      "learning_rate": 0.0001,
+      "loss": 0.9803,
+      "loss/crossentropy": 2.4196624755859375,
+      "loss/hidden": 0.64453125,
+      "loss/logits": 0.10631287097930908,
+      "loss/reg": 0.02294265851378441,
+      "step": 520
+    },
+    {
+      "epoch": 0.065125,
+      "grad_norm": 3.101484775543213,
+      "grad_norm_var": 1.20713683658368,
+      "learning_rate": 0.0001,
+      "loss": 1.2375,
+      "loss/crossentropy": 2.1448891162872314,
+      "loss/hidden": 0.88671875,
+      "loss/logits": 0.12145140767097473,
+      "loss/reg": 0.02293260022997856,
+      "step": 521
+    },
+    {
+      "epoch": 0.06525,
+      "grad_norm": 3.2564265727996826,
+      "grad_norm_var": 0.2854656858181736,
+      "learning_rate": 0.0001,
+      "loss": 0.9783,
+      "loss/crossentropy": 2.3986809253692627,
+      "loss/hidden": 0.609375,
+      "loss/logits": 0.1396813988685608,
+      "loss/reg": 0.022922798991203308,
+      "step": 522
+    },
+    {
+      "epoch": 0.065375,
+      "grad_norm": 3.3007333278656006,
+      "grad_norm_var": 0.2569672821287893,
+      "learning_rate": 0.0001,
+      "loss": 0.9586,
+      "loss/crossentropy": 2.7955212593078613,
+      "loss/hidden": 0.62109375,
+      "loss/logits": 0.10833179950714111,
+      "loss/reg": 0.02291307970881462,
+      "step": 523
+    },
+    {
+      "epoch": 0.0655,
+      "grad_norm": 2.9546499252319336,
+      "grad_norm_var": 0.25738909944094907,
+      "learning_rate": 0.0001,
+      "loss": 1.0937,
+      "loss/crossentropy": 2.409029006958008,
+      "loss/hidden": 0.71875,
+      "loss/logits": 0.14591118693351746,
+      "loss/reg": 0.022904111072421074,
+      "step": 524
+    },
+    {
+      "epoch": 0.065625,
+      "grad_norm": 3.2193830013275146,
+      "grad_norm_var": 0.25204334767618,
+      "learning_rate": 0.0001,
+      "loss": 1.1221,
+      "loss/crossentropy": 2.491401195526123,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.13920898735523224,
+      "loss/reg": 0.02289445698261261,
+      "step": 525
+    },
+    {
+      "epoch": 0.06575,
+      "grad_norm": 40.478694915771484,
+      "grad_norm_var": 85.9971082258447,
+      "learning_rate": 0.0001,
+      "loss": 1.1342,
+      "loss/crossentropy": 2.4311652183532715,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.10845671594142914,
+      "loss/reg": 0.022885650396347046,
+      "step": 526
+    },
+    {
+      "epoch": 0.065875,
+      "grad_norm": 3.2905385494232178,
+      "grad_norm_var": 86.20029999738617,
+      "learning_rate": 0.0001,
+      "loss": 1.0101,
+      "loss/crossentropy": 2.1737422943115234,
+      "loss/hidden": 0.671875,
+      "loss/logits": 0.10944204032421112,
+      "loss/reg": 0.022876843810081482,
+      "step": 527
+    },
+    {
+      "epoch": 0.066,
+      "grad_norm": 4.666721343994141,
+      "grad_norm_var": 85.84699165642114,
+      "learning_rate": 0.0001,
+      "loss": 1.2142,
+      "loss/crossentropy": 2.462200880050659,
+      "loss/hidden": 0.86328125,
+      "loss/logits": 0.12223749607801437,
+      "loss/reg": 0.02286742813885212,
+      "step": 528
+    },
+    {
+      "epoch": 0.066125,
+      "grad_norm": 3.0347273349761963,
+      "grad_norm_var": 85.66251244998139,
+      "learning_rate": 0.0001,
+      "loss": 1.0285,
+      "loss/crossentropy": 2.38840651512146,
+      "loss/hidden": 0.6640625,
+      "loss/logits": 0.13581448793411255,
+      "loss/reg": 0.022858494892716408,
+      "step": 529
+    },
+    {
+      "epoch": 0.06625,
+      "grad_norm": 2.5890092849731445,
+      "grad_norm_var": 86.04634847608627,
+      "learning_rate": 0.0001,
+      "loss": 0.967,
+      "loss/crossentropy": 2.5042731761932373,
+      "loss/hidden": 0.62890625,
+      "loss/logits": 0.10961504280567169,
+      "loss/reg": 0.02284966967999935,
+      "step": 530
+    },
+    {
+      "epoch": 0.066375,
+      "grad_norm": 3.3963401317596436,
+      "grad_norm_var": 85.93485657047692,
+      "learning_rate": 0.0001,
+      "loss": 1.072,
+      "loss/crossentropy": 2.7259116172790527,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.13268503546714783,
+      "loss/reg": 0.02284088172018528,
+      "step": 531
+    },
+    {
+      "epoch": 0.0665,
+      "grad_norm": 3.731293201446533,
+      "grad_norm_var": 85.85653980693046,
+      "learning_rate": 0.0001,
+      "loss": 1.0647,
+      "loss/crossentropy": 2.277968168258667,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.12150134146213531,
+      "loss/reg": 0.02283208817243576,
+      "step": 532
+    },
+    {
+      "epoch": 0.066625,
+      "grad_norm": 4.581428050994873,
+      "grad_norm_var": 85.78540060231343,
+      "learning_rate": 0.0001,
+      "loss": 1.2486,
+      "loss/crossentropy": 2.067720890045166,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.17660680413246155,
+      "loss/reg": 0.02282322198152542,
+      "step": 533
+    },
+    {
+      "epoch": 0.06675,
+      "grad_norm": 3.0526421070098877,
+      "grad_norm_var": 85.91535378874303,
+      "learning_rate": 0.0001,
+      "loss": 1.1318,
+      "loss/crossentropy": 2.4441521167755127,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.13023720681667328,
+      "loss/reg": 0.022814445197582245,
+      "step": 534
+    },
+    {
+      "epoch": 0.066875,
+      "grad_norm": 3.4852664470672607,
+      "grad_norm_var": 86.12062292738887,
+      "learning_rate": 0.0001,
+      "loss": 0.9936,
+      "loss/crossentropy": 2.418733596801758,
+      "loss/hidden": 0.66015625,
+      "loss/logits": 0.10543158650398254,
+      "loss/reg": 0.022805610671639442,
+      "step": 535
+    },
+    {
+      "epoch": 0.067,
+      "grad_norm": 2.7321274280548096,
+      "grad_norm_var": 86.43545869041343,
+      "learning_rate": 0.0001,
+      "loss": 1.0096,
+      "loss/crossentropy": 2.3275394439697266,
+      "loss/hidden": 0.65625,
+      "loss/logits": 0.12536926567554474,
+      "loss/reg": 0.022797243669629097,
+      "step": 536
+    },
+    {
+      "epoch": 0.067125,
+      "grad_norm": 3.029811382293701,
+      "grad_norm_var": 86.46041611877568,
+      "learning_rate": 0.0001,
+      "loss": 1.1049,
+      "loss/crossentropy": 2.2477617263793945,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.1309519112110138,
+      "loss/reg": 0.022788099944591522,
+      "step": 537
+    },
+    {
+      "epoch": 0.06725,
+      "grad_norm": 2.7345895767211914,
+      "grad_norm_var": 86.64571497988939,
+      "learning_rate": 0.0001,
+      "loss": 1.1272,
+      "loss/crossentropy": 2.4289236068725586,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.141631618142128,
+      "loss/reg": 0.022778736427426338,
+      "step": 538
+    },
+    {
+      "epoch": 0.067375,
+      "grad_norm": 3.2934482097625732,
+      "grad_norm_var": 86.6479928457627,
+      "learning_rate": 0.0001,
+      "loss": 1.0286,
+      "loss/crossentropy": 2.538973093032837,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.12509004771709442,
+      "loss/reg": 0.02277030609548092,
+      "step": 539
+    },
+    {
+      "epoch": 0.0675,
+      "grad_norm": 3.833656072616577,
+      "grad_norm_var": 86.38133368804911,
+      "learning_rate": 0.0001,
+      "loss": 0.999,
+      "loss/crossentropy": 2.429593563079834,
+      "loss/hidden": 0.64453125,
+      "loss/logits": 0.12682604789733887,
+      "loss/reg": 0.022761952131986618,
+      "step": 540
+    },
+    {
+      "epoch": 0.067625,
+      "grad_norm": 3.544104814529419,
+      "grad_norm_var": 86.28065873545309,
+      "learning_rate": 0.0001,
+      "loss": 1.0002,
+      "loss/crossentropy": 2.166292905807495,
+      "loss/hidden": 0.66015625,
+      "loss/logits": 0.11254848539829254,
+      "loss/reg": 0.02275264821946621,
+      "step": 541
+    },
+    {
+      "epoch": 0.06775,
+      "grad_norm": 4.461411952972412,
+      "grad_norm_var": 0.4229304647166086,
+      "learning_rate": 0.0001,
+      "loss": 1.6203,
+      "loss/crossentropy": 2.6347737312316895,
+      "loss/hidden": 1.1328125,
+      "loss/logits": 0.2600440979003906,
+      "loss/reg": 0.022744029760360718,
+      "step": 542
+    },
+    {
+      "epoch": 0.067875,
+      "grad_norm": 2.6814959049224854,
+      "grad_norm_var": 0.46036790462302574,
+      "learning_rate": 0.0001,
+      "loss": 0.9923,
+      "loss/crossentropy": 2.580264091491699,
+      "loss/hidden": 0.65234375,
+      "loss/logits": 0.11260214447975159,
+      "loss/reg": 0.022734828293323517,
+      "step": 543
+    },
+    {
+      "epoch": 0.068,
+      "grad_norm": 3.685408353805542,
+      "grad_norm_var": 0.35847800648438505,
+      "learning_rate": 0.0001,
+      "loss": 1.1452,
+      "loss/crossentropy": 2.7129862308502197,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.13279825448989868,
+      "loss/reg": 0.022726204246282578,
+      "step": 544
+    },
+    {
+      "epoch": 0.068125,
+      "grad_norm": 6.349724292755127,
+      "grad_norm_var": 0.8985836730566762,
+      "learning_rate": 0.0001,
+      "loss": 1.1353,
+      "loss/crossentropy": 2.7293214797973633,
+      "loss/hidden": 0.65625,
+      "loss/logits": 0.2518823742866516,
+      "loss/reg": 0.022716930136084557,
+      "step": 545
+    },
+    {
+      "epoch": 0.06825,
+      "grad_norm": 3.681774616241455,
+      "grad_norm_var": 0.829722440393675,
+      "learning_rate": 0.0001,
+      "loss": 1.1759,
+      "loss/crossentropy": 2.278223991394043,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.1441642791032791,
+      "loss/reg": 0.022707859054207802,
+      "step": 546
+    },
+    {
+      "epoch": 0.068375,
+      "grad_norm": 3.84778094291687,
+      "grad_norm_var": 0.8276635905853821,
+      "learning_rate": 0.0001,
+      "loss": 1.3398,
+      "loss/crossentropy": 2.1253304481506348,
+      "loss/hidden": 0.94921875,
+      "loss/logits": 0.16354900598526,
+      "loss/reg": 0.022699227556586266,
+      "step": 547
+    },
+    {
+      "epoch": 0.0685,
+      "grad_norm": 5.178676605224609,
+      "grad_norm_var": 0.9703527182714744,
+      "learning_rate": 0.0001,
+      "loss": 1.022,
+      "loss/crossentropy": 2.7832319736480713,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.11932960152626038,
+      "loss/reg": 0.02269013226032257,
+      "step": 548
+    },
+    {
+      "epoch": 0.068625,
+      "grad_norm": 3.0284383296966553,
+      "grad_norm_var": 0.9511722709094016,
+      "learning_rate": 0.0001,
+      "loss": 1.0412,
+      "loss/crossentropy": 1.9599052667617798,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.11125050485134125,
+      "loss/reg": 0.022681355476379395,
+      "step": 549
+    },
+    {
+      "epoch": 0.06875,
+      "grad_norm": 3.167809247970581,
+      "grad_norm_var": 0.942616955302132,
+      "learning_rate": 0.0001,
+      "loss": 1.0545,
+      "loss/crossentropy": 2.5844783782958984,
+      "loss/hidden": 0.69140625,
+      "loss/logits": 0.13634443283081055,
+      "loss/reg": 0.02267223782837391,
+      "step": 550
+    },
+    {
+      "epoch": 0.068875,
+      "grad_norm": 3.2949278354644775,
+      "grad_norm_var": 0.9495941353113788,
+      "learning_rate": 0.0001,
+      "loss": 1.0619,
+      "loss/crossentropy": 2.314173460006714,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.12046810984611511,
+      "loss/reg": 0.022662866860628128,
+      "step": 551
+    },
+    {
+      "epoch": 0.069,
+      "grad_norm": 2.8322856426239014,
+      "grad_norm_var": 0.9378422714313653,
+      "learning_rate": 0.0001,
+      "loss": 1.0272,
+      "loss/crossentropy": 2.768298387527466,
+      "loss/hidden": 0.66796875,
+      "loss/logits": 0.1326732635498047,
+      "loss/reg": 0.022653890773653984,
+      "step": 552
+    },
+    {
+      "epoch": 0.069125,
+      "grad_norm": 6.815005779266357,
+      "grad_norm_var": 1.5125797637252996,
+      "learning_rate": 0.0001,
+      "loss": 1.1457,
+      "loss/crossentropy": 2.8011491298675537,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.13409599661827087,
+      "loss/reg": 0.022644398733973503,
+      "step": 553
+    },
+    {
+      "epoch": 0.06925,
+      "grad_norm": 4.219581127166748,
+      "grad_norm_var": 1.4192768991356985,
+      "learning_rate": 0.0001,
+      "loss": 1.2504,
+      "loss/crossentropy": 2.5874292850494385,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.17252284288406372,
+      "loss/reg": 0.02263464592397213,
+      "step": 554
+    },
+    {
+      "epoch": 0.069375,
+      "grad_norm": 3.8279531002044678,
+      "grad_norm_var": 1.3871550629864857,
+      "learning_rate": 0.0001,
+      "loss": 1.1814,
+      "loss/crossentropy": 2.5384669303894043,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.1543978452682495,
+      "loss/reg": 0.022625621408224106,
+      "step": 555
+    },
+    {
+      "epoch": 0.0695,
+      "grad_norm": 3.563680648803711,
+      "grad_norm_var": 1.3987108056073487,
+      "learning_rate": 0.0001,
+      "loss": 1.0727,
+      "loss/crossentropy": 2.111318349838257,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.10826431214809418,
+      "loss/reg": 0.022616824135184288,
+      "step": 556
+    },
+    {
+      "epoch": 0.069625,
+      "grad_norm": 3.9599223136901855,
+      "grad_norm_var": 1.3836174934919199,
+      "learning_rate": 0.0001,
+      "loss": 1.0309,
+      "loss/crossentropy": 2.429032325744629,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.12908612191677094,
+      "loss/reg": 0.022607678547501564,
+      "step": 557
+    },
+    {
+      "epoch": 0.06975,
+      "grad_norm": 2.8072519302368164,
+      "grad_norm_var": 1.4610802306207225,
+      "learning_rate": 0.0001,
+      "loss": 1.0694,
+      "loss/crossentropy": 2.5817906856536865,
+      "loss/hidden": 0.6875,
+      "loss/logits": 0.15589380264282227,
+      "loss/reg": 0.022598396986722946,
+      "step": 558
+    },
+    {
+      "epoch": 0.069875,
+      "grad_norm": 2.764833927154541,
+      "grad_norm_var": 1.4475983977607596,
+      "learning_rate": 0.0001,
+      "loss": 1.0233,
+      "loss/crossentropy": 2.202237844467163,
+      "loss/hidden": 0.66015625,
+      "loss/logits": 0.13721255958080292,
+      "loss/reg": 0.02258932963013649,
+      "step": 559
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.113494396209717,
+      "grad_norm_var": 1.5267634464669517,
+      "learning_rate": 0.0001,
+      "loss": 1.1672,
+      "loss/crossentropy": 2.006186008453369,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.14061376452445984,
+      "loss/reg": 0.022580046206712723,
+      "step": 560
+    },
+    {
+      "epoch": 0.070125,
+      "grad_norm": 3.8133227825164795,
+      "grad_norm_var": 1.1437787263680603,
+      "learning_rate": 0.0001,
+      "loss": 1.0148,
+      "loss/crossentropy": 2.37170672416687,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.11329137533903122,
+      "loss/reg": 0.02257111482322216,
+      "step": 561
+    },
+    {
+      "epoch": 0.07025,
+      "grad_norm": 2.7208938598632812,
+      "grad_norm_var": 1.2255733087022764,
+      "learning_rate": 0.0001,
+      "loss": 1.0621,
+      "loss/crossentropy": 2.271667242050171,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.1333208978176117,
+      "loss/reg": 0.02256210334599018,
+      "step": 562
+    },
+    {
+      "epoch": 0.070375,
+      "grad_norm": 53.38179016113281,
+      "grad_norm_var": 154.8279377341773,
+      "learning_rate": 0.0001,
+      "loss": 0.9618,
+      "loss/crossentropy": 2.2667322158813477,
+      "loss/hidden": 0.640625,
+      "loss/logits": 0.09567096829414368,
+      "loss/reg": 0.02255306765437126,
+      "step": 563
+    },
+    {
+      "epoch": 0.0705,
+      "grad_norm": 5.538954257965088,
+      "grad_norm_var": 154.75309317540348,
+      "learning_rate": 0.0001,
+      "loss": 1.782,
+      "loss/crossentropy": 2.5168824195861816,
+      "loss/hidden": 1.3671875,
+      "loss/logits": 0.1893981695175171,
+      "loss/reg": 0.02254408597946167,
+      "step": 564
+    },
+    {
+      "epoch": 0.070625,
+      "grad_norm": 2.8311338424682617,
+      "grad_norm_var": 154.85811657117597,
+      "learning_rate": 0.0001,
+      "loss": 1.1104,
+      "loss/crossentropy": 2.1938281059265137,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.13508498668670654,
+      "loss/reg": 0.02253509685397148,
+      "step": 565
+    },
+    {
+      "epoch": 0.07075,
+      "grad_norm": 4.604408264160156,
+      "grad_norm_var": 154.26918998432618,
+      "learning_rate": 0.0001,
+      "loss": 1.2995,
+      "loss/crossentropy": 2.103062391281128,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.16405051946640015,
+      "loss/reg": 0.02252543345093727,
+      "step": 566
+    },
+    {
+      "epoch": 0.070875,
+      "grad_norm": 4.917901992797852,
+      "grad_norm_var": 153.63084329919155,
+      "learning_rate": 0.0001,
+      "loss": 1.2945,
+      "loss/crossentropy": 2.6932315826416016,
+      "loss/hidden": 0.87890625,
+      "loss/logits": 0.1904207468032837,
+      "loss/reg": 0.022515632212162018,
+      "step": 567
+    },
+    {
+      "epoch": 0.071,
+      "grad_norm": 4.055351257324219,
+      "grad_norm_var": 153.02723135387427,
+      "learning_rate": 0.0001,
+      "loss": 1.2253,
+      "loss/crossentropy": 2.3849422931671143,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.1642536073923111,
+      "loss/reg": 0.02250652387738228,
+      "step": 568
+    },
+    {
+      "epoch": 0.071125,
+      "grad_norm": 2.851072072982788,
+      "grad_norm_var": 154.20402053832456,
+      "learning_rate": 0.0001,
+      "loss": 1.0724,
+      "loss/crossentropy": 2.580428123474121,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.144349604845047,
+      "loss/reg": 0.022497190162539482,
+      "step": 569
+    },
+    {
+      "epoch": 0.07125,
+      "grad_norm": 3.257493495941162,
+      "grad_norm_var": 154.6102933496205,
+      "learning_rate": 0.0001,
+      "loss": 0.9328,
+      "loss/crossentropy": 2.4263837337493896,
+      "loss/hidden": 0.59375,
+      "loss/logits": 0.11414404958486557,
+      "loss/reg": 0.02248740941286087,
+      "step": 570
+    },
+    {
+      "epoch": 0.071375,
+      "grad_norm": 3.6194608211517334,
+      "grad_norm_var": 154.69773136421824,
+      "learning_rate": 0.0001,
+      "loss": 1.0417,
+      "loss/crossentropy": 2.7035937309265137,
+      "loss/hidden": 0.69140625,
+      "loss/logits": 0.12549756467342377,
+      "loss/reg": 0.022477447986602783,
+      "step": 571
+    },
+    {
+      "epoch": 0.0715,
+      "grad_norm": 4.193033695220947,
+      "grad_norm_var": 154.44566535859553,
+      "learning_rate": 0.0001,
+      "loss": 1.6505,
+      "loss/crossentropy": 2.274057149887085,
+      "loss/hidden": 1.1640625,
+      "loss/logits": 0.26177138090133667,
+      "loss/reg": 0.022467276081442833,
+      "step": 572
+    },
+    {
+      "epoch": 0.071625,
+      "grad_norm": 2.7127623558044434,
+      "grad_norm_var": 155.03209308401435,
+      "learning_rate": 0.0001,
+      "loss": 0.9118,
+      "loss/crossentropy": 2.4076664447784424,
+      "loss/hidden": 0.58203125,
+      "loss/logits": 0.10521911084651947,
+      "loss/reg": 0.022456735372543335,
+      "step": 573
+    },
+    {
+      "epoch": 0.07175,
+      "grad_norm": 3.5871214866638184,
+      "grad_norm_var": 154.65243889362196,
+      "learning_rate": 0.0001,
+      "loss": 0.9755,
+      "loss/crossentropy": 2.8088905811309814,
+      "loss/hidden": 0.62890625,
+      "loss/logits": 0.12216061353683472,
+      "loss/reg": 0.022446416318416595,
+      "step": 574
+    },
+    {
+      "epoch": 0.071875,
+      "grad_norm": 20.843276977539062,
+      "grad_norm_var": 165.17750310305152,
+      "learning_rate": 0.0001,
+      "loss": 1.1707,
+      "loss/crossentropy": 2.569218635559082,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.1221412867307663,
+      "loss/reg": 0.02243630215525627,
+      "step": 575
+    },
+    {
+      "epoch": 0.072,
+      "grad_norm": 3.375251293182373,
+      "grad_norm_var": 166.03594003131977,
+      "learning_rate": 0.0001,
+      "loss": 0.9923,
+      "loss/crossentropy": 2.7540183067321777,
+      "loss/hidden": 0.640625,
+      "loss/logits": 0.12738245725631714,
+      "loss/reg": 0.022425668314099312,
+      "step": 576
+    },
+    {
+      "epoch": 0.072125,
+      "grad_norm": 3.457340717315674,
+      "grad_norm_var": 166.23754433202586,
+      "learning_rate": 0.0001,
+      "loss": 1.0371,
+      "loss/crossentropy": 2.2943716049194336,
+      "loss/hidden": 0.66796875,
+      "loss/logits": 0.14497271180152893,
+      "loss/reg": 0.022415172308683395,
+      "step": 577
+    },
+    {
+      "epoch": 0.07225,
+      "grad_norm": 2.966371774673462,
+      "grad_norm_var": 166.07272256293106,
+      "learning_rate": 0.0001,
+      "loss": 1.2044,
+      "loss/crossentropy": 2.760359048843384,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.16780292987823486,
+      "loss/reg": 0.022406071424484253,
+      "step": 578
+    },
+    {
+      "epoch": 0.072375,
+      "grad_norm": 2.5532329082489014,
+      "grad_norm_var": 19.219812763276813,
+      "learning_rate": 0.0001,
+      "loss": 0.9996,
+      "loss/crossentropy": 2.427797794342041,
+      "loss/hidden": 0.66796875,
+      "loss/logits": 0.10767525434494019,
+      "loss/reg": 0.02239692024886608,
+      "step": 579
+    },
+    {
+      "epoch": 0.0725,
+      "grad_norm": 3.250906229019165,
+      "grad_norm_var": 19.294198335433887,
+      "learning_rate": 0.0001,
+      "loss": 1.1148,
+      "loss/crossentropy": 2.29978084564209,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.1408957540988922,
+      "loss/reg": 0.0223868228495121,
+      "step": 580
+    },
+    {
+      "epoch": 0.072625,
+      "grad_norm": 3.8051562309265137,
+      "grad_norm_var": 19.12802354300362,
+      "learning_rate": 0.0001,
+      "loss": 1.2329,
+      "loss/crossentropy": 2.402350664138794,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.17713911831378937,
+      "loss/reg": 0.022376833483576775,
+      "step": 581
+    },
+    {
+      "epoch": 0.07275,
+      "grad_norm": 4.466115951538086,
+      "grad_norm_var": 19.129656316190147,
+      "learning_rate": 0.0001,
+      "loss": 1.1846,
+      "loss/crossentropy": 2.5129504203796387,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.14841441810131073,
+      "loss/reg": 0.02236761339008808,
+      "step": 582
+    },
+    {
+      "epoch": 0.072875,
+      "grad_norm": 3.1465752124786377,
+      "grad_norm_var": 19.255278342461487,
+      "learning_rate": 0.0001,
+      "loss": 1.1857,
+      "loss/crossentropy": 2.5782837867736816,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.14178822934627533,
+      "loss/reg": 0.022358402609825134,
+      "step": 583
+    },
+    {
+      "epoch": 0.073,
+      "grad_norm": 2.7474803924560547,
+      "grad_norm_var": 19.441256858474677,
+      "learning_rate": 0.0001,
+      "loss": 1.0532,
+      "loss/crossentropy": 2.4901747703552246,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.12661507725715637,
+      "loss/reg": 0.02234930731356144,
+      "step": 584
+    },
+    {
+      "epoch": 0.073125,
+      "grad_norm": 2.7027359008789062,
+      "grad_norm_var": 19.47380183903334,
+      "learning_rate": 0.0001,
+      "loss": 0.8469,
+      "loss/crossentropy": 2.694582223892212,
+      "loss/hidden": 0.5390625,
+      "loss/logits": 0.08438676595687866,
+      "loss/reg": 0.022340187802910805,
+      "step": 585
+    },
+    {
+      "epoch": 0.07325,
+      "grad_norm": 2.847770929336548,
+      "grad_norm_var": 19.547679388785195,
+      "learning_rate": 0.0001,
+      "loss": 0.9885,
+      "loss/crossentropy": 2.5558888912200928,
+      "loss/hidden": 0.6484375,
+      "loss/logits": 0.11670757830142975,
+      "loss/reg": 0.022330984473228455,
+      "step": 586
+    },
+    {
+      "epoch": 0.073375,
+      "grad_norm": 3.20444917678833,
+      "grad_norm_var": 19.601201389954156,
+      "learning_rate": 0.0001,
+      "loss": 1.2216,
+      "loss/crossentropy": 2.0031001567840576,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.1546502560377121,
+      "loss/reg": 0.022321749478578568,
+      "step": 587
+    },
+    {
+      "epoch": 0.0735,
+      "grad_norm": 4.657837390899658,
+      "grad_norm_var": 19.6039707895662,
+      "learning_rate": 0.0001,
+      "loss": 1.1377,
+      "loss/crossentropy": 2.598100185394287,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.13329939544200897,
+      "loss/reg": 0.02231265790760517,
+      "step": 588
+    },
+    {
+      "epoch": 0.073625,
+      "grad_norm": 3.1166787147521973,
+      "grad_norm_var": 19.52355503271277,
+      "learning_rate": 0.0001,
+      "loss": 1.304,
+      "loss/crossentropy": 2.2209084033966064,
+      "loss/hidden": 0.89453125,
+      "loss/logits": 0.18648099899291992,
+      "loss/reg": 0.022303014993667603,
+      "step": 589
+    },
+    {
+      "epoch": 0.07375,
+      "grad_norm": 2.988344669342041,
+      "grad_norm_var": 19.61249925539728,
+      "learning_rate": 0.0001,
+      "loss": 0.9412,
+      "loss/crossentropy": 2.4533708095550537,
+      "loss/hidden": 0.60546875,
+      "loss/logits": 0.11275988817214966,
+      "loss/reg": 0.022293319925665855,
+      "step": 590
+    },
+    {
+      "epoch": 0.073875,
+      "grad_norm": 3.0553901195526123,
+      "grad_norm_var": 0.3491433903254536,
+      "learning_rate": 0.0001,
+      "loss": 1.0883,
+      "loss/crossentropy": 2.3986990451812744,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.13498544692993164,
+      "loss/reg": 0.022283662110567093,
+      "step": 591
+    },
+    {
+      "epoch": 0.074,
+      "grad_norm": 4.445681095123291,
+      "grad_norm_var": 0.43558600780207446,
+      "learning_rate": 0.0001,
+      "loss": 1.3195,
+      "loss/crossentropy": 2.335937976837158,
+      "loss/hidden": 0.90625,
+      "loss/logits": 0.19050292670726776,
+      "loss/reg": 0.022273709997534752,
+      "step": 592
+    },
+    {
+      "epoch": 0.074125,
+      "grad_norm": 41.39726638793945,
+      "grad_norm_var": 91.00287624901027,
+      "learning_rate": 0.0001,
+      "loss": 1.2527,
+      "loss/crossentropy": 2.3441669940948486,
+      "loss/hidden": 0.86328125,
+      "loss/logits": 0.1667976826429367,
+      "loss/reg": 0.02226419560611248,
+      "step": 593
+    },
+    {
+      "epoch": 0.07425,
+      "grad_norm": 2.8557939529418945,
+      "grad_norm_var": 91.04408434440504,
+      "learning_rate": 0.0001,
+      "loss": 1.0612,
+      "loss/crossentropy": 2.380885362625122,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.1354852318763733,
+      "loss/reg": 0.022254258394241333,
+      "step": 594
+    },
+    {
+      "epoch": 0.074375,
+      "grad_norm": 6.565737724304199,
+      "grad_norm_var": 90.36543928633743,
+      "learning_rate": 0.0001,
+      "loss": 1.2519,
+      "loss/crossentropy": 2.3877017498016357,
+      "loss/hidden": 0.90625,
+      "loss/logits": 0.1231798455119133,
+      "loss/reg": 0.022244345396757126,
+      "step": 595
+    },
+    {
+      "epoch": 0.0745,
+      "grad_norm": 3.6746349334716797,
+      "grad_norm_var": 90.22397938232942,
+      "learning_rate": 0.0001,
+      "loss": 1.1718,
+      "loss/crossentropy": 2.397080898284912,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.1486300826072693,
+      "loss/reg": 0.022234413772821426,
+      "step": 596
+    },
+    {
+      "epoch": 0.074625,
+      "grad_norm": 2.981614589691162,
+      "grad_norm_var": 90.50516196939815,
+      "learning_rate": 0.0001,
+      "loss": 1.043,
+      "loss/crossentropy": 2.37412428855896,
+      "loss/hidden": 0.69140625,
+      "loss/logits": 0.12931303679943085,
+      "loss/reg": 0.022224588319659233,
+      "step": 597
+    },
+    {
+      "epoch": 0.07475,
+      "grad_norm": 3.6531307697296143,
+      "grad_norm_var": 90.70497774366554,
+      "learning_rate": 0.0001,
+      "loss": 1.1272,
+      "loss/crossentropy": 2.6906161308288574,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.13550975918769836,
+      "loss/reg": 0.02221417799592018,
+      "step": 598
+    },
+    {
+      "epoch": 0.074875,
+      "grad_norm": 3.751652240753174,
+      "grad_norm_var": 90.50753182721607,
+      "learning_rate": 0.0001,
+      "loss": 1.1618,
+      "loss/crossentropy": 2.6506831645965576,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.1468081772327423,
+      "loss/reg": 0.022204989567399025,
+      "step": 599
+    },
+    {
+      "epoch": 0.075,
+      "grad_norm": 3.228563070297241,
+      "grad_norm_var": 90.31879350061254,
+      "learning_rate": 0.0001,
+      "loss": 1.0791,
+      "loss/crossentropy": 2.381368398666382,
+      "loss/hidden": 0.71875,
+      "loss/logits": 0.13835087418556213,
+      "loss/reg": 0.022195899859070778,
+      "step": 600
+    },
+    {
+      "epoch": 0.075125,
+      "grad_norm": 5.392886638641357,
+      "grad_norm_var": 89.60797997668053,
+      "learning_rate": 0.0001,
+      "loss": 1.2309,
+      "loss/crossentropy": 2.52596378326416,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.17307403683662415,
+      "loss/reg": 0.022185994312167168,
+      "step": 601
+    },
+    {
+      "epoch": 0.07525,
+      "grad_norm": 3.2832775115966797,
+      "grad_norm_var": 89.43019603463321,
+      "learning_rate": 0.0001,
+      "loss": 1.1462,
+      "loss/crossentropy": 2.4442057609558105,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.15487736463546753,
+      "loss/reg": 0.02217610739171505,
+      "step": 602
+    },
+    {
+      "epoch": 0.075375,
+      "grad_norm": 4.1014084815979,
+      "grad_norm_var": 89.1293068696746,
+      "learning_rate": 0.0001,
+      "loss": 1.0847,
+      "loss/crossentropy": 2.531064987182617,
+      "loss/hidden": 0.7265625,
+      "loss/logits": 0.13643184304237366,
+      "loss/reg": 0.02216634899377823,
+      "step": 603
+    },
+    {
+      "epoch": 0.0755,
+      "grad_norm": 3.656172037124634,
+      "grad_norm_var": 89.39756111673695,
+      "learning_rate": 0.0001,
+      "loss": 1.0161,
+      "loss/crossentropy": 2.470468044281006,
+      "loss/hidden": 0.68359375,
+      "loss/logits": 0.11097002029418945,
+      "loss/reg": 0.02215682342648506,
+      "step": 604
+    },
+    {
+      "epoch": 0.075625,
+      "grad_norm": 3.8061492443084717,
+      "grad_norm_var": 89.1498668494714,
+      "learning_rate": 0.0001,
+      "loss": 1.1195,
+      "loss/crossentropy": 2.469788074493408,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.12457874417304993,
+      "loss/reg": 0.022147687152028084,
+      "step": 605
+    },
+    {
+      "epoch": 0.07575,
+      "grad_norm": 2.658135414123535,
+      "grad_norm_var": 89.29708722871554,
+      "learning_rate": 0.0001,
+      "loss": 1.0657,
+      "loss/crossentropy": 2.2552011013031006,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.12950366735458374,
+      "loss/reg": 0.022138802334666252,
+      "step": 606
+    },
+    {
+      "epoch": 0.075875,
+      "grad_norm": 3.291750907897949,
+      "grad_norm_var": 89.20284122750789,
+      "learning_rate": 0.0001,
+      "loss": 1.133,
+      "loss/crossentropy": 2.225055456161499,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.1734064519405365,
+      "loss/reg": 0.022129878401756287,
+      "step": 607
+    },
+    {
+      "epoch": 0.076,
+      "grad_norm": 4.001339912414551,
+      "grad_norm_var": 89.31742762195415,
+      "learning_rate": 0.0001,
+      "loss": 0.9658,
+      "loss/crossentropy": 2.3826169967651367,
+      "loss/hidden": 0.64453125,
+      "loss/logits": 0.10007497668266296,
+      "loss/reg": 0.022121025249361992,
+      "step": 608
+    },
+    {
+      "epoch": 0.076125,
+      "grad_norm": 2.9729602336883545,
+      "grad_norm_var": 0.9817241823775095,
+      "learning_rate": 0.0001,
+      "loss": 1.1962,
+      "loss/crossentropy": 2.4926881790161133,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.1508128046989441,
+      "loss/reg": 0.022111859172582626,
+      "step": 609
+    },
+    {
+      "epoch": 0.07625,
+      "grad_norm": 2.5144121646881104,
+      "grad_norm_var": 1.0293551003726749,
+      "learning_rate": 0.0001,
+      "loss": 0.9883,
+      "loss/crossentropy": 2.478576898574829,
+      "loss/hidden": 0.65234375,
+      "loss/logits": 0.11488830298185349,
+      "loss/reg": 0.02210419811308384,
+      "step": 610
+    },
+    {
+      "epoch": 0.076375,
+      "grad_norm": 3.7777743339538574,
+      "grad_norm_var": 0.4576308797359,
+      "learning_rate": 0.0001,
+      "loss": 1.0059,
+      "loss/crossentropy": 2.4356257915496826,
+      "loss/hidden": 0.6484375,
+      "loss/logits": 0.13651525974273682,
+      "loss/reg": 0.022095149382948875,
+      "step": 611
+    },
+    {
+      "epoch": 0.0765,
+      "grad_norm": 3.232398748397827,
+      "grad_norm_var": 0.4623055923756754,
+      "learning_rate": 0.0001,
+      "loss": 1.0108,
+      "loss/crossentropy": 2.4058218002319336,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.11410736292600632,
+      "loss/reg": 0.022087210789322853,
+      "step": 612
+    },
+    {
+      "epoch": 0.076625,
+      "grad_norm": 2.61934757232666,
+      "grad_norm_var": 0.49646373584008807,
+      "learning_rate": 0.0001,
+      "loss": 0.9466,
+      "loss/crossentropy": 2.3981077671051025,
+      "loss/hidden": 0.59765625,
+      "loss/logits": 0.1281721591949463,
+      "loss/reg": 0.022078126668930054,
+      "step": 613
+    },
+    {
+      "epoch": 0.07675,
+      "grad_norm": 4.228268146514893,
+      "grad_norm_var": 0.5291615579452734,
+      "learning_rate": 0.0001,
+      "loss": 1.0729,
+      "loss/crossentropy": 2.664149761199951,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.12176868319511414,
+      "loss/reg": 0.02207016758620739,
+      "step": 614
+    },
+    {
+      "epoch": 0.076875,
+      "grad_norm": 2.9137685298919678,
+      "grad_norm_var": 0.5485319535320492,
+      "learning_rate": 0.0001,
+      "loss": 0.8708,
+      "loss/crossentropy": 2.5457890033721924,
+      "loss/hidden": 0.55078125,
+      "loss/logits": 0.09941907972097397,
+      "loss/reg": 0.022060981020331383,
+      "step": 615
+    },
+    {
+      "epoch": 0.077,
+      "grad_norm": 3.4499685764312744,
+      "grad_norm_var": 0.5441756848342263,
+      "learning_rate": 0.0001,
+      "loss": 1.1439,
+      "loss/crossentropy": 2.3822786808013916,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.14211627840995789,
+      "loss/reg": 0.02205180749297142,
+      "step": 616
+    },
+    {
+      "epoch": 0.077125,
+      "grad_norm": 2.271921157836914,
+      "grad_norm_var": 0.36266744154546565,
+      "learning_rate": 0.0001,
+      "loss": 0.9061,
+      "loss/crossentropy": 2.707848072052002,
+      "loss/hidden": 0.58203125,
+      "loss/logits": 0.10360105335712433,
+      "loss/reg": 0.022042402997612953,
+      "step": 617
+    },
+    {
+      "epoch": 0.07725,
+      "grad_norm": 3.16980242729187,
+      "grad_norm_var": 0.36370543210803513,
+      "learning_rate": 0.0001,
+      "loss": 1.2733,
+      "loss/crossentropy": 2.1980466842651367,
+      "loss/hidden": 0.86328125,
+      "loss/logits": 0.18972548842430115,
+      "loss/reg": 0.02203306369483471,
+      "step": 618
+    },
+    {
+      "epoch": 0.077375,
+      "grad_norm": 2.4408295154571533,
+      "grad_norm_var": 0.3567501583972517,
+      "learning_rate": 0.0001,
+      "loss": 1.1352,
+      "loss/crossentropy": 2.3986692428588867,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.1454274207353592,
+      "loss/reg": 0.022023871541023254,
+      "step": 619
+    },
+    {
+      "epoch": 0.0775,
+      "grad_norm": 2.558687925338745,
+      "grad_norm_var": 0.36349398943808237,
+      "learning_rate": 0.0001,
+      "loss": 0.9776,
+      "loss/crossentropy": 2.4290616512298584,
+      "loss/hidden": 0.65625,
+      "loss/logits": 0.1012360006570816,
+      "loss/reg": 0.0220141913741827,
+      "step": 620
+    },
+    {
+      "epoch": 0.077625,
+      "grad_norm": 3.7468619346618652,
+      "grad_norm_var": 0.3582835152003213,
+      "learning_rate": 0.0001,
+      "loss": 1.1635,
+      "loss/crossentropy": 2.4566454887390137,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.15830263495445251,
+      "loss/reg": 0.02200442925095558,
+      "step": 621
+    },
+    {
+      "epoch": 0.07775,
+      "grad_norm": 3.8364651203155518,
+      "grad_norm_var": 0.3732032502257139,
+      "learning_rate": 0.0001,
+      "loss": 1.3213,
+      "loss/crossentropy": 2.5934860706329346,
+      "loss/hidden": 0.9296875,
+      "loss/logits": 0.17167437076568604,
+      "loss/reg": 0.021995313465595245,
+      "step": 622
+    },
+    {
+      "epoch": 0.077875,
+      "grad_norm": 2.9136173725128174,
+      "grad_norm_var": 0.37696739372618043,
+      "learning_rate": 0.0001,
+      "loss": 1.1191,
+      "loss/crossentropy": 2.1625287532806396,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.1492336541414261,
+      "loss/reg": 0.021986283361911774,
+      "step": 623
+    },
+    {
+      "epoch": 0.078,
+      "grad_norm": 3.3136067390441895,
+      "grad_norm_var": 0.3298862344756988,
+      "learning_rate": 0.0001,
+      "loss": 1.1487,
+      "loss/crossentropy": 2.3915464878082275,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.1398892104625702,
+      "loss/reg": 0.021976841613650322,
+      "step": 624
+    },
+    {
+      "epoch": 0.078125,
+      "grad_norm": 3.1731748580932617,
+      "grad_norm_var": 0.3283984444798058,
+      "learning_rate": 0.0001,
+      "loss": 1.0945,
+      "loss/crossentropy": 2.53873610496521,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.16392138600349426,
+      "loss/reg": 0.021966535598039627,
+      "step": 625
+    },
+    {
+      "epoch": 0.07825,
+      "grad_norm": 4.172556400299072,
+      "grad_norm_var": 0.3630228628347131,
+      "learning_rate": 0.0001,
+      "loss": 1.1888,
+      "loss/crossentropy": 2.4219956398010254,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.17627781629562378,
+      "loss/reg": 0.021956363692879677,
+      "step": 626
+    },
+    {
+      "epoch": 0.078375,
+      "grad_norm": 3.99817156791687,
+      "grad_norm_var": 0.3819004722530487,
+      "learning_rate": 0.0001,
+      "loss": 1.3074,
+      "loss/crossentropy": 2.2407050132751465,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.20513707399368286,
+      "loss/reg": 0.02194611169397831,
+      "step": 627
+    },
+    {
+      "epoch": 0.0785,
+      "grad_norm": 3.2869558334350586,
+      "grad_norm_var": 0.3819405314837089,
+      "learning_rate": 0.0001,
+      "loss": 0.923,
+      "loss/crossentropy": 2.5824477672576904,
+      "loss/hidden": 0.59375,
+      "loss/logits": 0.10985252261161804,
+      "loss/reg": 0.021935785189270973,
+      "step": 628
+    },
+    {
+      "epoch": 0.078625,
+      "grad_norm": 3.0037360191345215,
+      "grad_norm_var": 0.35855200267849247,
+      "learning_rate": 0.0001,
+      "loss": 1.0636,
+      "loss/crossentropy": 2.4916722774505615,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.13343587517738342,
+      "loss/reg": 0.021925168111920357,
+      "step": 629
+    },
+    {
+      "epoch": 0.07875,
+      "grad_norm": 3.5274088382720947,
+      "grad_norm_var": 0.3006291732182412,
+      "learning_rate": 0.0001,
+      "loss": 1.1441,
+      "loss/crossentropy": 1.7279773950576782,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.13975940644741058,
+      "loss/reg": 0.02191445231437683,
+      "step": 630
+    },
+    {
+      "epoch": 0.078875,
+      "grad_norm": 3.2895989418029785,
+      "grad_norm_var": 0.29330515223301745,
+      "learning_rate": 0.0001,
+      "loss": 1.0394,
+      "loss/crossentropy": 2.370850086212158,
+      "loss/hidden": 0.68359375,
+      "loss/logits": 0.13678821921348572,
+      "loss/reg": 0.02190525084733963,
+      "step": 631
+    },
+    {
+      "epoch": 0.079,
+      "grad_norm": 3.741135597229004,
+      "grad_norm_var": 0.30599490652712474,
+      "learning_rate": 0.0001,
+      "loss": 1.2655,
+      "loss/crossentropy": 2.365464210510254,
+      "loss/hidden": 0.87890625,
+      "loss/logits": 0.16766133904457092,
+      "loss/reg": 0.02189476415514946,
+      "step": 632
+    },
+    {
+      "epoch": 0.079125,
+      "grad_norm": 4.110158443450928,
+      "grad_norm_var": 0.2706546096444164,
+      "learning_rate": 0.0001,
+      "loss": 1.1804,
+      "loss/crossentropy": 2.658433437347412,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.1646716445684433,
+      "loss/reg": 0.021885616704821587,
+      "step": 633
+    },
+    {
+      "epoch": 0.07925,
+      "grad_norm": 3.7503654956817627,
+      "grad_norm_var": 0.27446839769864156,
+      "learning_rate": 0.0001,
+      "loss": 1.1,
+      "loss/crossentropy": 2.4457831382751465,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.13510534167289734,
+      "loss/reg": 0.021875550970435143,
+      "step": 634
+    },
+    {
+      "epoch": 0.079375,
+      "grad_norm": 2.917835235595703,
+      "grad_norm_var": 0.22584356567046956,
+      "learning_rate": 0.0001,
+      "loss": 1.1498,
+      "loss/crossentropy": 2.3328001499176025,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.1537725031375885,
+      "loss/reg": 0.021865583956241608,
+      "step": 635
+    },
+    {
+      "epoch": 0.0795,
+      "grad_norm": 2.907240152359009,
+      "grad_norm_var": 0.19160647764443486,
+      "learning_rate": 0.0001,
+      "loss": 0.9073,
+      "loss/crossentropy": 2.556042432785034,
+      "loss/hidden": 0.59375,
+      "loss/logits": 0.09503352642059326,
+      "loss/reg": 0.02185530960559845,
+      "step": 636
+    },
+    {
+      "epoch": 0.079625,
+      "grad_norm": 2.5937955379486084,
+      "grad_norm_var": 0.2337615816576618,
+      "learning_rate": 0.0001,
+      "loss": 0.9198,
+      "loss/crossentropy": 2.363370180130005,
+      "loss/hidden": 0.5859375,
+      "loss/logits": 0.11537887156009674,
+      "loss/reg": 0.021845519542694092,
+      "step": 637
+    },
+    {
+      "epoch": 0.07975,
+      "grad_norm": 3.9080708026885986,
+      "grad_norm_var": 0.2381681132369368,
+      "learning_rate": 0.0001,
+      "loss": 1.1552,
+      "loss/crossentropy": 2.5054194927215576,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.12043754756450653,
+      "loss/reg": 0.021835271269083023,
+      "step": 638
+    },
+    {
+      "epoch": 0.079875,
+      "grad_norm": 2.8192946910858154,
+      "grad_norm_var": 0.24500412598165416,
+      "learning_rate": 0.0001,
+      "loss": 0.9375,
+      "loss/crossentropy": 2.6611719131469727,
+      "loss/hidden": 0.60546875,
+      "loss/logits": 0.11382012814283371,
+      "loss/reg": 0.021824965253472328,
+      "step": 639
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.111716270446777,
+      "grad_norm_var": 0.27486954530744445,
+      "learning_rate": 0.0001,
+      "loss": 1.2972,
+      "loss/crossentropy": 2.472568988800049,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.16890740394592285,
+      "loss/reg": 0.02181575633585453,
+      "step": 640
+    },
+    {
+      "epoch": 0.080125,
+      "grad_norm": 5.715061187744141,
+      "grad_norm_var": 0.5825168124345791,
+      "learning_rate": 0.0001,
+      "loss": 1.0307,
+      "loss/crossentropy": 2.2123966217041016,
+      "loss/hidden": 0.68359375,
+      "loss/logits": 0.12899596989154816,
+      "loss/reg": 0.021806620061397552,
+      "step": 641
+    },
+    {
+      "epoch": 0.08025,
+      "grad_norm": 3.8680315017700195,
+      "grad_norm_var": 0.5657073815126407,
+      "learning_rate": 0.0001,
+      "loss": 1.053,
+      "loss/crossentropy": 2.4526453018188477,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.13190723955631256,
+      "loss/reg": 0.021797508001327515,
+      "step": 642
+    },
+    {
+      "epoch": 0.080375,
+      "grad_norm": 4.992859363555908,
+      "grad_norm_var": 0.680778895488037,
+      "learning_rate": 0.0001,
+      "loss": 1.1683,
+      "loss/crossentropy": 2.3445968627929688,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.12617573142051697,
+      "loss/reg": 0.02178841643035412,
+      "step": 643
+    },
+    {
+      "epoch": 0.0805,
+      "grad_norm": 4.339104175567627,
+      "grad_norm_var": 0.6977811040599325,
+      "learning_rate": 0.0001,
+      "loss": 1.3433,
+      "loss/crossentropy": 1.8577097654342651,
+      "loss/hidden": 0.9453125,
+      "loss/logits": 0.18017446994781494,
+      "loss/reg": 0.02177964523434639,
+      "step": 644
+    },
+    {
+      "epoch": 0.080625,
+      "grad_norm": 4.026562213897705,
+      "grad_norm_var": 0.6648423545945282,
+      "learning_rate": 0.0001,
+      "loss": 1.3266,
+      "loss/crossentropy": 2.415410041809082,
+      "loss/hidden": 0.859375,
+      "loss/logits": 0.24951061606407166,
+      "loss/reg": 0.021770501509308815,
+      "step": 645
+    },
+    {
+      "epoch": 0.08075,
+      "grad_norm": 2.8517775535583496,
+      "grad_norm_var": 0.7169049906385166,
+      "learning_rate": 0.0001,
+      "loss": 0.9283,
+      "loss/crossentropy": 2.6426851749420166,
+      "loss/hidden": 0.6015625,
+      "loss/logits": 0.10916159301996231,
+      "loss/reg": 0.021761184558272362,
+      "step": 646
+    },
+    {
+      "epoch": 0.080875,
+      "grad_norm": 3.3171377182006836,
+      "grad_norm_var": 0.7152750431492562,
+      "learning_rate": 0.0001,
+      "loss": 1.0088,
+      "loss/crossentropy": 2.638533115386963,
+      "loss/hidden": 0.62890625,
+      "loss/logits": 0.1623907834291458,
+      "loss/reg": 0.02175196446478367,
+      "step": 647
+    },
+    {
+      "epoch": 0.081,
+      "grad_norm": 3.653881311416626,
+      "grad_norm_var": 0.7158322952113887,
+      "learning_rate": 0.0001,
+      "loss": 1.1132,
+      "loss/crossentropy": 2.745229482650757,
+      "loss/hidden": 0.765625,
+      "loss/logits": 0.1301097273826599,
+      "loss/reg": 0.021742329001426697,
+      "step": 648
+    },
+    {
+      "epoch": 0.081125,
+      "grad_norm": 4.31439733505249,
+      "grad_norm_var": 0.728446489341123,
+      "learning_rate": 0.0001,
+      "loss": 1.2706,
+      "loss/crossentropy": 2.7004082202911377,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.25250470638275146,
+      "loss/reg": 0.021732579916715622,
+      "step": 649
+    },
+    {
+      "epoch": 0.08125,
+      "grad_norm": 4.621687889099121,
+      "grad_norm_var": 0.7753064642270201,
+      "learning_rate": 0.0001,
+      "loss": 1.4091,
+      "loss/crossentropy": 2.38405179977417,
+      "loss/hidden": 0.89453125,
+      "loss/logits": 0.29737186431884766,
+      "loss/reg": 0.02172265760600567,
+      "step": 650
+    },
+    {
+      "epoch": 0.081375,
+      "grad_norm": 2.87424898147583,
+      "grad_norm_var": 0.7806094534209419,
+      "learning_rate": 0.0001,
+      "loss": 1.2103,
+      "loss/crossentropy": 2.0905954837799072,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.18066659569740295,
+      "loss/reg": 0.021712414920330048,
+      "step": 651
+    },
+    {
+      "epoch": 0.0815,
+      "grad_norm": 4.724517345428467,
+      "grad_norm_var": 0.7689569917943563,
+      "learning_rate": 0.0001,
+      "loss": 1.2928,
+      "loss/crossentropy": 2.500594139099121,
+      "loss/hidden": 0.89453125,
+      "loss/logits": 0.1812204122543335,
+      "loss/reg": 0.021703310310840607,
+      "step": 652
+    },
+    {
+      "epoch": 0.081625,
+      "grad_norm": 2.8673081398010254,
+      "grad_norm_var": 0.7252403996552116,
+      "learning_rate": 0.0001,
+      "loss": 1.0588,
+      "loss/crossentropy": 2.3599724769592285,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.13873916864395142,
+      "loss/reg": 0.021693557500839233,
+      "step": 653
+    },
+    {
+      "epoch": 0.08175,
+      "grad_norm": 3.099315643310547,
+      "grad_norm_var": 0.7693322976491117,
+      "learning_rate": 0.0001,
+      "loss": 1.1961,
+      "loss/crossentropy": 2.3167355060577393,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.14727652072906494,
+      "loss/reg": 0.021683741360902786,
+      "step": 654
+    },
+    {
+      "epoch": 0.081875,
+      "grad_norm": 3.287602186203003,
+      "grad_norm_var": 0.7163515778113151,
+      "learning_rate": 0.0001,
+      "loss": 1.0429,
+      "loss/crossentropy": 2.7967636585235596,
+      "loss/hidden": 0.6875,
+      "loss/logits": 0.13863959908485413,
+      "loss/reg": 0.021674364805221558,
+      "step": 655
+    },
+    {
+      "epoch": 0.082,
+      "grad_norm": 3.487874746322632,
+      "grad_norm_var": 0.7244436337536264,
+      "learning_rate": 0.0001,
+      "loss": 1.0883,
+      "loss/crossentropy": 2.5409998893737793,
+      "loss/hidden": 0.71875,
+      "loss/logits": 0.1528949737548828,
+      "loss/reg": 0.02166520059108734,
+      "step": 656
+    },
+    {
+      "epoch": 0.082125,
+      "grad_norm": 3.6202752590179443,
+      "grad_norm_var": 0.4854858648423857,
+      "learning_rate": 0.0001,
+      "loss": 1.2516,
+      "loss/crossentropy": 2.3633673191070557,
+      "loss/hidden": 0.88671875,
+      "loss/logits": 0.14828172326087952,
+      "loss/reg": 0.021655315533280373,
+      "step": 657
+    },
+    {
+      "epoch": 0.08225,
+      "grad_norm": 3.7921783924102783,
+      "grad_norm_var": 0.484617963461113,
+      "learning_rate": 0.0001,
+      "loss": 1.07,
+      "loss/crossentropy": 2.497013807296753,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.13874131441116333,
+      "loss/reg": 0.021646033972501755,
+      "step": 658
+    },
+    {
+      "epoch": 0.082375,
+      "grad_norm": 5.818857669830322,
+      "grad_norm_var": 0.6650298211735747,
+      "learning_rate": 0.0001,
+      "loss": 1.2961,
+      "loss/crossentropy": 2.6599833965301514,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.16955840587615967,
+      "loss/reg": 0.021636882796883583,
+      "step": 659
+    },
+    {
+      "epoch": 0.0825,
+      "grad_norm": 3.465527057647705,
+      "grad_norm_var": 0.6491808619433999,
+      "learning_rate": 0.0001,
+      "loss": 1.1365,
+      "loss/crossentropy": 2.5972306728363037,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.14674408733844757,
+      "loss/reg": 0.02162766456604004,
+      "step": 660
+    },
+    {
+      "epoch": 0.082625,
+      "grad_norm": 3.143159866333008,
+      "grad_norm_var": 0.664078497493661,
+      "learning_rate": 0.0001,
+      "loss": 1.2241,
+      "loss/crossentropy": 2.5945792198181152,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.17588043212890625,
+      "loss/reg": 0.021618474274873734,
+      "step": 661
+    },
+    {
+      "epoch": 0.08275,
+      "grad_norm": 3.7091660499572754,
+      "grad_norm_var": 0.6149151800980365,
+      "learning_rate": 0.0001,
+      "loss": 1.1901,
+      "loss/crossentropy": 2.631565809249878,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.16149985790252686,
+      "loss/reg": 0.021609637886285782,
+      "step": 662
+    },
+    {
+      "epoch": 0.082875,
+      "grad_norm": 3.1449224948883057,
+      "grad_norm_var": 0.6264170707357067,
+      "learning_rate": 0.0001,
+      "loss": 1.1491,
+      "loss/crossentropy": 2.2890501022338867,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.14797118306159973,
+      "loss/reg": 0.021600957959890366,
+      "step": 663
+    },
+    {
+      "epoch": 0.083,
+      "grad_norm": 3.096752405166626,
+      "grad_norm_var": 0.6512152784754629,
+      "learning_rate": 0.0001,
+      "loss": 1.0925,
+      "loss/crossentropy": 2.767069101333618,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.1343650072813034,
+      "loss/reg": 0.021592585369944572,
+      "step": 664
+    },
+    {
+      "epoch": 0.083125,
+      "grad_norm": 3.8935883045196533,
+      "grad_norm_var": 0.6273466460071402,
+      "learning_rate": 0.0001,
+      "loss": 1.0996,
+      "loss/crossentropy": 2.4779109954833984,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.12599951028823853,
+      "loss/reg": 0.021583350375294685,
+      "step": 665
+    },
+    {
+      "epoch": 0.08325,
+      "grad_norm": 3.6881141662597656,
+      "grad_norm_var": 0.5627883047301658,
+      "learning_rate": 0.0001,
+      "loss": 1.1184,
+      "loss/crossentropy": 2.1603968143463135,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.14095276594161987,
+      "loss/reg": 0.021574225276708603,
+      "step": 666
+    },
+    {
+      "epoch": 0.083375,
+      "grad_norm": 2.9750800132751465,
+      "grad_norm_var": 0.5535713466115603,
+      "learning_rate": 0.0001,
+      "loss": 1.1215,
+      "loss/crossentropy": 2.4817895889282227,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.15194162726402283,
+      "loss/reg": 0.02156493254005909,
+      "step": 667
+    },
+    {
+      "epoch": 0.0835,
+      "grad_norm": 2.649543046951294,
+      "grad_norm_var": 0.5152581471177806,
+      "learning_rate": 0.0001,
+      "loss": 0.918,
+      "loss/crossentropy": 2.334226608276367,
+      "loss/hidden": 0.59375,
+      "loss/logits": 0.10870292782783508,
+      "loss/reg": 0.02155502513051033,
+      "step": 668
+    },
+    {
+      "epoch": 0.083625,
+      "grad_norm": 4.535495281219482,
+      "grad_norm_var": 0.5520843285134825,
+      "learning_rate": 0.0001,
+      "loss": 1.168,
+      "loss/crossentropy": 2.4633467197418213,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.15570059418678284,
+      "loss/reg": 0.021545063704252243,
+      "step": 669
+    },
+    {
+      "epoch": 0.08375,
+      "grad_norm": 3.3291449546813965,
+      "grad_norm_var": 0.5404115229162234,
+      "learning_rate": 0.0001,
+      "loss": 1.1241,
+      "loss/crossentropy": 2.317607879638672,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.12363065779209137,
+      "loss/reg": 0.021535001695156097,
+      "step": 670
+    },
+    {
+      "epoch": 0.083875,
+      "grad_norm": 4.071917533874512,
+      "grad_norm_var": 0.5459456401930327,
+      "learning_rate": 0.0001,
+      "loss": 1.1652,
+      "loss/crossentropy": 2.4951958656311035,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.11796893179416656,
+      "loss/reg": 0.021524924784898758,
+      "step": 671
+    },
+    {
+      "epoch": 0.084,
+      "grad_norm": 4.647782802581787,
+      "grad_norm_var": 0.6047501670354971,
+      "learning_rate": 0.0001,
+      "loss": 1.157,
+      "loss/crossentropy": 2.6994447708129883,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.16059228777885437,
+      "loss/reg": 0.021514689549803734,
+      "step": 672
+    },
+    {
+      "epoch": 0.084125,
+      "grad_norm": 4.240062713623047,
+      "grad_norm_var": 0.6201999433703328,
+      "learning_rate": 0.0001,
+      "loss": 1.0984,
+      "loss/crossentropy": 2.592426300048828,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.10989370942115784,
+      "loss/reg": 0.021505359560251236,
+      "step": 673
+    },
+    {
+      "epoch": 0.08425,
+      "grad_norm": 15.647865295410156,
+      "grad_norm_var": 9.451818582857973,
+      "learning_rate": 0.0001,
+      "loss": 1.6418,
+      "loss/crossentropy": 2.723198413848877,
+      "loss/hidden": 1.078125,
+      "loss/logits": 0.3486996293067932,
+      "loss/reg": 0.02149534970521927,
+      "step": 674
+    },
+    {
+      "epoch": 0.084375,
+      "grad_norm": 3.6480963230133057,
+      "grad_norm_var": 9.365638761154676,
+      "learning_rate": 0.0001,
+      "loss": 1.1052,
+      "loss/crossentropy": 2.618795871734619,
+      "loss/hidden": 0.734375,
+      "loss/logits": 0.1559816300868988,
+      "loss/reg": 0.02148519456386566,
+      "step": 675
+    },
+    {
+      "epoch": 0.0845,
+      "grad_norm": 3.5789527893066406,
+      "grad_norm_var": 9.352796045350159,
+      "learning_rate": 0.0001,
+      "loss": 1.1092,
+      "loss/crossentropy": 2.4817607402801514,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.14444658160209656,
+      "loss/reg": 0.021475963294506073,
+      "step": 676
+    },
+    {
+      "epoch": 0.084625,
+      "grad_norm": 3.143718719482422,
+      "grad_norm_var": 9.352704277495931,
+      "learning_rate": 0.0001,
+      "loss": 1.0394,
+      "loss/crossentropy": 2.28035044670105,
+      "loss/hidden": 0.70703125,
+      "loss/logits": 0.11767329275608063,
+      "loss/reg": 0.02146601676940918,
+      "step": 677
+    },
+    {
+      "epoch": 0.08475,
+      "grad_norm": 4.2889580726623535,
+      "grad_norm_var": 9.322240526517621,
+      "learning_rate": 0.0001,
+      "loss": 1.1067,
+      "loss/crossentropy": 2.28658390045166,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.18121060729026794,
+      "loss/reg": 0.021456118673086166,
+      "step": 678
+    },
+    {
+      "epoch": 0.084875,
+      "grad_norm": 2.800516128540039,
+      "grad_norm_var": 9.387804829955044,
+      "learning_rate": 0.0001,
+      "loss": 0.975,
+      "loss/crossentropy": 2.699044704437256,
+      "loss/hidden": 0.63671875,
+      "loss/logits": 0.12379397451877594,
+      "loss/reg": 0.02144702896475792,
+      "step": 679
+    },
+    {
+      "epoch": 0.085,
+      "grad_norm": 3.064215898513794,
+      "grad_norm_var": 9.393480165725077,
+      "learning_rate": 0.0001,
+      "loss": 1.1186,
+      "loss/crossentropy": 2.547557830810547,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.14249341189861298,
+      "loss/reg": 0.02143782004714012,
+      "step": 680
+    },
+    {
+      "epoch": 0.085125,
+      "grad_norm": 3.7314915657043457,
+      "grad_norm_var": 9.405801361337378,
+      "learning_rate": 0.0001,
+      "loss": 1.1376,
+      "loss/crossentropy": 2.705601930618286,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.14988452196121216,
+      "loss/reg": 0.02142806351184845,
+      "step": 681
+    },
+    {
+      "epoch": 0.08525,
+      "grad_norm": 2.256387948989868,
+      "grad_norm_var": 9.665529326305519,
+      "learning_rate": 0.0001,
+      "loss": 0.945,
+      "loss/crossentropy": 2.464989423751831,
+      "loss/hidden": 0.62109375,
+      "loss/logits": 0.10975323617458344,
+      "loss/reg": 0.02141808532178402,
+      "step": 682
+    },
+    {
+      "epoch": 0.085375,
+      "grad_norm": 3.150348424911499,
+      "grad_norm_var": 9.636765682886749,
+      "learning_rate": 0.0001,
+      "loss": 1.0573,
+      "loss/crossentropy": 2.2520275115966797,
+      "loss/hidden": 0.72265625,
+      "loss/logits": 0.12059713900089264,
+      "loss/reg": 0.021408328786492348,
+      "step": 683
+    },
+    {
+      "epoch": 0.0855,
+      "grad_norm": 2.973684549331665,
+      "grad_norm_var": 9.572043410499656,
+      "learning_rate": 0.0001,
+      "loss": 1.0663,
+      "loss/crossentropy": 2.4148662090301514,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.11397817730903625,
+      "loss/reg": 0.021399127319455147,
+      "step": 684
+    },
+    {
+      "epoch": 0.085625,
+      "grad_norm": 4.39288330078125,
+      "grad_norm_var": 9.56920341692891,
+      "learning_rate": 0.0001,
+      "loss": 1.206,
+      "loss/crossentropy": 2.5025415420532227,
+      "loss/hidden": 0.84765625,
+      "loss/logits": 0.1444375216960907,
+      "loss/reg": 0.021389208734035492,
+      "step": 685
+    },
+    {
+      "epoch": 0.08575,
+      "grad_norm": 2.815019369125366,
+      "grad_norm_var": 9.652987248771876,
+      "learning_rate": 0.0001,
+      "loss": 1.0631,
+      "loss/crossentropy": 2.236358165740967,
+      "loss/hidden": 0.72265625,
+      "loss/logits": 0.1266387552022934,
+      "loss/reg": 0.021379247307777405,
+      "step": 686
+    },
+    {
+      "epoch": 0.085875,
+      "grad_norm": 3.6744375228881836,
+      "grad_norm_var": 9.673796390527396,
+      "learning_rate": 0.0001,
+      "loss": 1.2523,
+      "loss/crossentropy": 2.1673667430877686,
+      "loss/hidden": 0.8671875,
+      "loss/logits": 0.1714528650045395,
+      "loss/reg": 0.02136901021003723,
+      "step": 687
+    },
+    {
+      "epoch": 0.086,
+      "grad_norm": 3.169265031814575,
+      "grad_norm_var": 9.732675648460468,
+      "learning_rate": 0.0001,
+      "loss": 1.1389,
+      "loss/crossentropy": 2.45385479927063,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.13236522674560547,
+      "loss/reg": 0.021359853446483612,
+      "step": 688
+    },
+    {
+      "epoch": 0.086125,
+      "grad_norm": 7.017651557922363,
+      "grad_norm_var": 10.2441458601344,
+      "learning_rate": 0.0001,
+      "loss": 1.4071,
+      "loss/crossentropy": 2.344740629196167,
+      "loss/hidden": 1.0625,
+      "loss/logits": 0.13107535243034363,
+      "loss/reg": 0.021349839866161346,
+      "step": 689
+    },
+    {
+      "epoch": 0.08625,
+      "grad_norm": 3.5523998737335205,
+      "grad_norm_var": 1.142674868302701,
+      "learning_rate": 0.0001,
+      "loss": 1.1408,
+      "loss/crossentropy": 2.323866367340088,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.13442741334438324,
+      "loss/reg": 0.021340306848287582,
+      "step": 690
+    },
+    {
+      "epoch": 0.086375,
+      "grad_norm": 3.2477400302886963,
+      "grad_norm_var": 1.1489843436981233,
+      "learning_rate": 0.0001,
+      "loss": 1.0067,
+      "loss/crossentropy": 2.672182083129883,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.11757320165634155,
+      "loss/reg": 0.021331045776605606,
+      "step": 691
+    },
+    {
+      "epoch": 0.0865,
+      "grad_norm": 3.1238696575164795,
+      "grad_norm_var": 1.1603900529546722,
+      "learning_rate": 0.0001,
+      "loss": 1.1122,
+      "loss/crossentropy": 2.510279893875122,
+      "loss/hidden": 0.765625,
+      "loss/logits": 0.13339656591415405,
+      "loss/reg": 0.021321000531315804,
+      "step": 692
+    },
+    {
+      "epoch": 0.086625,
+      "grad_norm": 2.552560329437256,
+      "grad_norm_var": 1.2122975827491755,
+      "learning_rate": 0.0001,
+      "loss": 1.0396,
+      "loss/crossentropy": 2.5303587913513184,
+      "loss/hidden": 0.6953125,
+      "loss/logits": 0.13117440044879913,
+      "loss/reg": 0.021313220262527466,
+      "step": 693
+    },
+    {
+      "epoch": 0.08675,
+      "grad_norm": 2.7075002193450928,
+      "grad_norm_var": 1.1997649773340213,
+      "learning_rate": 0.0001,
+      "loss": 1.0787,
+      "loss/crossentropy": 2.4859445095062256,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.12738223373889923,
+      "loss/reg": 0.021303845569491386,
+      "step": 694
+    },
+    {
+      "epoch": 0.086875,
+      "grad_norm": 3.2640225887298584,
+      "grad_norm_var": 1.1768004922088529,
+      "learning_rate": 0.0001,
+      "loss": 1.3383,
+      "loss/crossentropy": 2.4017555713653564,
+      "loss/hidden": 0.96875,
+      "loss/logits": 0.15659019351005554,
+      "loss/reg": 0.021294469013810158,
+      "step": 695
+    },
+    {
+      "epoch": 0.087,
+      "grad_norm": 3.5289857387542725,
+      "grad_norm_var": 1.1683562063707291,
+      "learning_rate": 0.0001,
+      "loss": 1.2607,
+      "loss/crossentropy": 2.2716236114501953,
+      "loss/hidden": 0.87109375,
+      "loss/logits": 0.17679363489151,
+      "loss/reg": 0.021284854039549828,
+      "step": 696
+    },
+    {
+      "epoch": 0.087125,
+      "grad_norm": 8.924003601074219,
+      "grad_norm_var": 3.0501856400161764,
+      "learning_rate": 0.0001,
+      "loss": 1.474,
+      "loss/crossentropy": 2.3684239387512207,
+      "loss/hidden": 1.0546875,
+      "loss/logits": 0.2065410166978836,
+      "loss/reg": 0.02127666585147381,
+      "step": 697
+    },
+    {
+      "epoch": 0.08725,
+      "grad_norm": 4.2463788986206055,
+      "grad_norm_var": 2.8955696375948605,
+      "learning_rate": 0.0001,
+      "loss": 1.1835,
+      "loss/crossentropy": 2.234137535095215,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.1427059769630432,
+      "loss/reg": 0.021268585696816444,
+      "step": 698
+    },
+    {
+      "epoch": 0.087375,
+      "grad_norm": 3.0776546001434326,
+      "grad_norm_var": 2.903130025314302,
+      "learning_rate": 0.0001,
+      "loss": 1.0559,
+      "loss/crossentropy": 2.4638020992279053,
+      "loss/hidden": 0.7265625,
+      "loss/logits": 0.1167045459151268,
+      "loss/reg": 0.02125934511423111,
+      "step": 699
+    },
+    {
+      "epoch": 0.0875,
+      "grad_norm": 3.9374022483825684,
+      "grad_norm_var": 2.843209099820052,
+      "learning_rate": 0.0001,
+      "loss": 0.9629,
+      "loss/crossentropy": 2.88035249710083,
+      "loss/hidden": 0.63671875,
+      "loss/logits": 0.11369955539703369,
+      "loss/reg": 0.021249722689390182,
+      "step": 700
+    },
+    {
+      "epoch": 0.087625,
+      "grad_norm": 4.235450744628906,
+      "grad_norm_var": 2.8355032825089417,
+      "learning_rate": 0.0001,
+      "loss": 1.278,
+      "loss/crossentropy": 2.2560718059539795,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.1828073114156723,
+      "loss/reg": 0.0212401133030653,
+      "step": 701
+    },
+    {
+      "epoch": 0.08775,
+      "grad_norm": 2.3705031871795654,
+      "grad_norm_var": 2.9146564397348773,
+      "learning_rate": 0.0001,
+      "loss": 0.9899,
+      "loss/crossentropy": 2.531632661819458,
+      "loss/hidden": 0.64453125,
+      "loss/logits": 0.13305732607841492,
+      "loss/reg": 0.021231388673186302,
+      "step": 702
+    },
+    {
+      "epoch": 0.087875,
+      "grad_norm": 3.5063636302948,
+      "grad_norm_var": 2.921798711310286,
+      "learning_rate": 0.0001,
+      "loss": 1.0524,
+      "loss/crossentropy": 2.548243999481201,
+      "loss/hidden": 0.72265625,
+      "loss/logits": 0.11747653782367706,
+      "loss/reg": 0.02122276835143566,
+      "step": 703
+    },
+    {
+      "epoch": 0.088,
+      "grad_norm": 3.141618013381958,
+      "grad_norm_var": 2.924554396554724,
+      "learning_rate": 0.0001,
+      "loss": 0.9087,
+      "loss/crossentropy": 2.3723928928375244,
+      "loss/hidden": 0.59375,
+      "loss/logits": 0.10285645723342896,
+      "loss/reg": 0.021213354542851448,
+      "step": 704
+    },
+    {
+      "epoch": 0.088125,
+      "grad_norm": 4.195517063140869,
+      "grad_norm_var": 2.2500098957229806,
+      "learning_rate": 0.0001,
+      "loss": 1.1263,
+      "loss/crossentropy": 2.383502244949341,
+      "loss/hidden": 0.765625,
+      "loss/logits": 0.14867368340492249,
+      "loss/reg": 0.021204529330134392,
+      "step": 705
+    },
+    {
+      "epoch": 0.08825,
+      "grad_norm": 3.646482467651367,
+      "grad_norm_var": 2.2483885758775686,
+      "learning_rate": 0.0001,
+      "loss": 1.1459,
+      "loss/crossentropy": 2.239716053009033,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.14099720120429993,
+      "loss/reg": 0.021195242181420326,
+      "step": 706
+    },
+    {
+      "epoch": 0.088375,
+      "grad_norm": 3.3426883220672607,
+      "grad_norm_var": 2.242826109053838,
+      "learning_rate": 0.0001,
+      "loss": 1.0323,
+      "loss/crossentropy": 2.270444869995117,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.10951918363571167,
+      "loss/reg": 0.021186839789152145,
+      "step": 707
+    },
+    {
+      "epoch": 0.0885,
+      "grad_norm": 3.083806037902832,
+      "grad_norm_var": 2.2462046620558,
+      "learning_rate": 0.0001,
+      "loss": 1.086,
+      "loss/crossentropy": 2.5013253688812256,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.12026840448379517,
+      "loss/reg": 0.021178435534238815,
+      "step": 708
+    },
+    {
+      "epoch": 0.088625,
+      "grad_norm": 3.71588134765625,
+      "grad_norm_var": 2.147370219186798,
+      "learning_rate": 0.0001,
+      "loss": 1.1125,
+      "loss/crossentropy": 2.342475175857544,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.13124068081378937,
+      "loss/reg": 0.021169869229197502,
+      "step": 709
+    },
+    {
+      "epoch": 0.08875,
+      "grad_norm": 3.4402713775634766,
+      "grad_norm_var": 2.0734307300644255,
+      "learning_rate": 0.0001,
+      "loss": 1.1229,
+      "loss/crossentropy": 2.4113690853118896,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.14960095286369324,
+      "loss/reg": 0.021161576732993126,
+      "step": 710
+    },
+    {
+      "epoch": 0.088875,
+      "grad_norm": 10.640914916992188,
+      "grad_norm_var": 4.894724677276531,
+      "learning_rate": 0.0001,
+      "loss": 1.7804,
+      "loss/crossentropy": 2.6555004119873047,
+      "loss/hidden": 1.328125,
+      "loss/logits": 0.240725576877594,
+      "loss/reg": 0.02115357480943203,
+      "step": 711
+    },
+    {
+      "epoch": 0.089,
+      "grad_norm": 3.23919415473938,
+      "grad_norm_var": 4.930329406483421,
+      "learning_rate": 0.0001,
+      "loss": 1.1368,
+      "loss/crossentropy": 2.4652106761932373,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.14798855781555176,
+      "loss/reg": 0.021144360303878784,
+      "step": 712
+    },
+    {
+      "epoch": 0.089125,
+      "grad_norm": 2.8362622261047363,
+      "grad_norm_var": 3.4904838717428524,
+      "learning_rate": 0.0001,
+      "loss": 1.0668,
+      "loss/crossentropy": 2.216999053955078,
+      "loss/hidden": 0.7265625,
+      "loss/logits": 0.1288391500711441,
+      "loss/reg": 0.021136239171028137,
+      "step": 713
+    },
+    {
+      "epoch": 0.08925,
+      "grad_norm": 5.395811080932617,
+      "grad_norm_var": 3.623687874884585,
+      "learning_rate": 0.0001,
+      "loss": 1.2022,
+      "loss/crossentropy": 2.463663101196289,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.1667168289422989,
+      "loss/reg": 0.02112707309424877,
+      "step": 714
+    },
+    {
+      "epoch": 0.089375,
+      "grad_norm": 3.475656032562256,
+      "grad_norm_var": 3.585286252049487,
+      "learning_rate": 0.0001,
+      "loss": 1.1524,
+      "loss/crossentropy": 2.4688560962677,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.14434993267059326,
+      "loss/reg": 0.02111782319843769,
+      "step": 715
+    },
+    {
+      "epoch": 0.0895,
+      "grad_norm": 4.29054594039917,
+      "grad_norm_var": 3.5895333664829043,
+      "learning_rate": 0.0001,
+      "loss": 1.1214,
+      "loss/crossentropy": 2.4820876121520996,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.15247562527656555,
+      "loss/reg": 0.021108638495206833,
+      "step": 716
+    },
+    {
+      "epoch": 0.089625,
+      "grad_norm": 2.927002429962158,
+      "grad_norm_var": 3.661532010616088,
+      "learning_rate": 0.0001,
+      "loss": 1.0904,
+      "loss/crossentropy": 2.499390125274658,
+      "loss/hidden": 0.734375,
+      "loss/logits": 0.14508014917373657,
+      "loss/reg": 0.021099381148815155,
+      "step": 717
+    },
+    {
+      "epoch": 0.08975,
+      "grad_norm": 2.529557943344116,
+      "grad_norm_var": 3.629551988733732,
+      "learning_rate": 0.0001,
+      "loss": 1.1194,
+      "loss/crossentropy": 2.395061492919922,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.13898837566375732,
+      "loss/reg": 0.021089982241392136,
+      "step": 718
+    },
+    {
+      "epoch": 0.089875,
+      "grad_norm": 3.2681446075439453,
+      "grad_norm_var": 3.6476018392648397,
+      "learning_rate": 0.0001,
+      "loss": 1.0036,
+      "loss/crossentropy": 2.65079927444458,
+      "loss/hidden": 0.6640625,
+      "loss/logits": 0.12875254452228546,
+      "loss/reg": 0.021080130711197853,
+      "step": 719
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.6700246334075928,
+      "grad_norm_var": 3.7122117675621022,
+      "learning_rate": 0.0001,
+      "loss": 0.9929,
+      "loss/crossentropy": 2.428039789199829,
+      "loss/hidden": 0.65234375,
+      "loss/logits": 0.12986770272254944,
+      "loss/reg": 0.021070368587970734,
+      "step": 720
+    },
+    {
+      "epoch": 0.090125,
+      "grad_norm": 3.296482563018799,
+      "grad_norm_var": 3.7295350110356558,
+      "learning_rate": 0.0001,
+      "loss": 1.1571,
+      "loss/crossentropy": 2.317190408706665,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.14570315182209015,
+      "loss/reg": 0.02106117643415928,
+      "step": 721
+    },
+    {
+      "epoch": 0.09025,
+      "grad_norm": 4.451722145080566,
+      "grad_norm_var": 3.74687645800365,
+      "learning_rate": 0.0001,
+      "loss": 1.1385,
+      "loss/crossentropy": 2.2958080768585205,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.146757572889328,
+      "loss/reg": 0.021051928400993347,
+      "step": 722
+    },
+    {
+      "epoch": 0.090375,
+      "grad_norm": 2.4288058280944824,
+      "grad_norm_var": 3.8685376080960414,
+      "learning_rate": 0.0001,
+      "loss": 0.9788,
+      "loss/crossentropy": 2.4692001342773438,
+      "loss/hidden": 0.65234375,
+      "loss/logits": 0.11606692522764206,
+      "loss/reg": 0.021042969077825546,
+      "step": 723
+    },
+    {
+      "epoch": 0.0905,
+      "grad_norm": 45.21147918701172,
+      "grad_norm_var": 110.45448625779909,
+      "learning_rate": 0.0001,
+      "loss": 1.0823,
+      "loss/crossentropy": 2.4428551197052,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.12983539700508118,
+      "loss/reg": 0.021031970158219337,
+      "step": 724
+    },
+    {
+      "epoch": 0.090625,
+      "grad_norm": 3.7111551761627197,
+      "grad_norm_var": 110.45623490585022,
+      "learning_rate": 0.0001,
+      "loss": 1.4512,
+      "loss/crossentropy": 2.1149277687072754,
+      "loss/hidden": 1.046875,
+      "loss/logits": 0.1941366195678711,
+      "loss/reg": 0.021020574495196342,
+      "step": 725
+    },
+    {
+      "epoch": 0.09075,
+      "grad_norm": 3.3139920234680176,
+      "grad_norm_var": 110.50855221427315,
+      "learning_rate": 0.0001,
+      "loss": 1.0686,
+      "loss/crossentropy": 2.798910140991211,
+      "loss/hidden": 0.71875,
+      "loss/logits": 0.13971024751663208,
+      "loss/reg": 0.021009519696235657,
+      "step": 726
+    },
+    {
+      "epoch": 0.090875,
+      "grad_norm": 2.5206596851348877,
+      "grad_norm_var": 110.12514261997971,
+      "learning_rate": 0.0001,
+      "loss": 1.0391,
+      "loss/crossentropy": 2.400813579559326,
+      "loss/hidden": 0.6953125,
+      "loss/logits": 0.13384617865085602,
+      "loss/reg": 0.020998528227210045,
+      "step": 727
+    },
+    {
+      "epoch": 0.091,
+      "grad_norm": 2.698018789291382,
+      "grad_norm_var": 110.34070270952832,
+      "learning_rate": 0.0001,
+      "loss": 0.928,
+      "loss/crossentropy": 2.505309820175171,
+      "loss/hidden": 0.61328125,
+      "loss/logits": 0.10482652485370636,
+      "loss/reg": 0.020989248529076576,
+      "step": 728
+    },
+    {
+      "epoch": 0.091125,
+      "grad_norm": 6.086211204528809,
+      "grad_norm_var": 109.65630388036335,
+      "learning_rate": 0.0001,
+      "loss": 1.4543,
+      "loss/crossentropy": 2.065880537033081,
+      "loss/hidden": 1.1171875,
+      "loss/logits": 0.12730881571769714,
+      "loss/reg": 0.020978538319468498,
+      "step": 729
+    },
+    {
+      "epoch": 0.09125,
+      "grad_norm": 2.527377128601074,
+      "grad_norm_var": 110.45601242879228,
+      "learning_rate": 0.0001,
+      "loss": 0.9281,
+      "loss/crossentropy": 2.575125217437744,
+      "loss/hidden": 0.60546875,
+      "loss/logits": 0.11296658217906952,
+      "loss/reg": 0.020967954769730568,
+      "step": 730
+    },
+    {
+      "epoch": 0.091375,
+      "grad_norm": 2.405383825302124,
+      "grad_norm_var": 110.88254605251709,
+      "learning_rate": 0.0001,
+      "loss": 1.0005,
+      "loss/crossentropy": 2.1985392570495605,
+      "loss/hidden": 0.671875,
+      "loss/logits": 0.1190965548157692,
+      "loss/reg": 0.02095715142786503,
+      "step": 731
+    },
+    {
+      "epoch": 0.0915,
+      "grad_norm": 5.095945835113525,
+      "grad_norm_var": 110.75067974759948,
+      "learning_rate": 0.0001,
+      "loss": 1.2535,
+      "loss/crossentropy": 2.451446056365967,
+      "loss/hidden": 0.88671875,
+      "loss/logits": 0.15732397139072418,
+      "loss/reg": 0.020946422591805458,
+      "step": 732
+    },
+    {
+      "epoch": 0.091625,
+      "grad_norm": 2.497173547744751,
+      "grad_norm_var": 110.93526847423995,
+      "learning_rate": 0.0001,
+      "loss": 1.1301,
+      "loss/crossentropy": 2.4981319904327393,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.15116086602210999,
+      "loss/reg": 0.020936597138643265,
+      "step": 733
+    },
+    {
+      "epoch": 0.09175,
+      "grad_norm": 4.977592468261719,
+      "grad_norm_var": 110.20332761050602,
+      "learning_rate": 0.0001,
+      "loss": 1.0306,
+      "loss/crossentropy": 2.37424898147583,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.11818103492259979,
+      "loss/reg": 0.020926134660840034,
+      "step": 734
+    },
+    {
+      "epoch": 0.091875,
+      "grad_norm": 2.8039772510528564,
+      "grad_norm_var": 110.39035266849663,
+      "learning_rate": 0.0001,
+      "loss": 0.9389,
+      "loss/crossentropy": 2.5503880977630615,
+      "loss/hidden": 0.62109375,
+      "loss/logits": 0.1085958182811737,
+      "loss/reg": 0.02091672271490097,
+      "step": 735
+    },
+    {
+      "epoch": 0.092,
+      "grad_norm": 3.418717861175537,
+      "grad_norm_var": 110.08862675247053,
+      "learning_rate": 0.0001,
+      "loss": 0.9544,
+      "loss/crossentropy": 2.6695592403411865,
+      "loss/hidden": 0.62890625,
+      "loss/logits": 0.11641087383031845,
+      "loss/reg": 0.02090657874941826,
+      "step": 736
+    },
+    {
+      "epoch": 0.092125,
+      "grad_norm": 2.807041883468628,
+      "grad_norm_var": 110.28591938740921,
+      "learning_rate": 0.0001,
+      "loss": 1.1135,
+      "loss/crossentropy": 2.4328622817993164,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.13500146567821503,
+      "loss/reg": 0.02089635282754898,
+      "step": 737
+    },
+    {
+      "epoch": 0.09225,
+      "grad_norm": 3.713057518005371,
+      "grad_norm_var": 110.4783888232826,
+      "learning_rate": 0.0001,
+      "loss": 1.1755,
+      "loss/crossentropy": 2.0374624729156494,
+      "loss/hidden": 0.83984375,
+      "loss/logits": 0.12679257988929749,
+      "loss/reg": 0.02088700234889984,
+      "step": 738
+    },
+    {
+      "epoch": 0.092375,
+      "grad_norm": 3.618948459625244,
+      "grad_norm_var": 109.99807079993953,
+      "learning_rate": 0.0001,
+      "loss": 0.9255,
+      "loss/crossentropy": 2.536527395248413,
+      "loss/hidden": 0.609375,
+      "loss/logits": 0.10737244784832001,
+      "loss/reg": 0.020877836272120476,
+      "step": 739
+    },
+    {
+      "epoch": 0.0925,
+      "grad_norm": 2.815113067626953,
+      "grad_norm_var": 1.1792510177041378,
+      "learning_rate": 0.0001,
+      "loss": 0.9755,
+      "loss/crossentropy": 2.464996337890625,
+      "loss/hidden": 0.640625,
+      "loss/logits": 0.12616491317749023,
+      "loss/reg": 0.020868681371212006,
+      "step": 740
+    },
+    {
+      "epoch": 0.092625,
+      "grad_norm": 3.194117546081543,
+      "grad_norm_var": 1.1771383378848062,
+      "learning_rate": 0.0001,
+      "loss": 1.1989,
+      "loss/crossentropy": 2.325310707092285,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.14658081531524658,
+      "loss/reg": 0.020859118551015854,
+      "step": 741
+    },
+    {
+      "epoch": 0.09275,
+      "grad_norm": 2.970301628112793,
+      "grad_norm_var": 1.1887296793511715,
+      "learning_rate": 0.0001,
+      "loss": 0.9939,
+      "loss/crossentropy": 2.4747390747070312,
+      "loss/hidden": 0.66015625,
+      "loss/logits": 0.1252739280462265,
+      "loss/reg": 0.020850006490945816,
+      "step": 742
+    },
+    {
+      "epoch": 0.092875,
+      "grad_norm": 3.7745604515075684,
+      "grad_norm_var": 1.1425983881417718,
+      "learning_rate": 0.0001,
+      "loss": 0.9835,
+      "loss/crossentropy": 2.4191653728485107,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.09930374473333359,
+      "loss/reg": 0.020840618759393692,
+      "step": 743
+    },
+    {
+      "epoch": 0.093,
+      "grad_norm": 2.8048055171966553,
+      "grad_norm_var": 1.132423092522494,
+      "learning_rate": 0.0001,
+      "loss": 1.083,
+      "loss/crossentropy": 2.764143228530884,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.12075912207365036,
+      "loss/reg": 0.020830942317843437,
+      "step": 744
+    },
+    {
+      "epoch": 0.093125,
+      "grad_norm": 9.04628849029541,
+      "grad_norm_var": 2.712848654929009,
+      "learning_rate": 0.0001,
+      "loss": 1.4779,
+      "loss/crossentropy": 2.2979469299316406,
+      "loss/hidden": 1.125,
+      "loss/logits": 0.1446615606546402,
+      "loss/reg": 0.02082117274403572,
+      "step": 745
+    },
+    {
+      "epoch": 0.09325,
+      "grad_norm": 3.1999406814575195,
+      "grad_norm_var": 2.6400540651182967,
+      "learning_rate": 0.0001,
+      "loss": 1.0424,
+      "loss/crossentropy": 2.6157429218292236,
+      "loss/hidden": 0.71875,
+      "loss/logits": 0.11555971205234528,
+      "loss/reg": 0.020811092108488083,
+      "step": 746
+    },
+    {
+      "epoch": 0.093375,
+      "grad_norm": 3.6833674907684326,
+      "grad_norm_var": 2.522139333113606,
+      "learning_rate": 0.0001,
+      "loss": 1.1046,
+      "loss/crossentropy": 2.6110498905181885,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.13881847262382507,
+      "loss/reg": 0.020801017060875893,
+      "step": 747
+    },
+    {
+      "epoch": 0.0935,
+      "grad_norm": 3.073922872543335,
+      "grad_norm_var": 2.4218973518929827,
+      "learning_rate": 0.0001,
+      "loss": 0.9672,
+      "loss/crossentropy": 2.0898826122283936,
+      "loss/hidden": 0.65625,
+      "loss/logits": 0.10303568840026855,
+      "loss/reg": 0.020791731774806976,
+      "step": 748
+    },
+    {
+      "epoch": 0.093625,
+      "grad_norm": 2.9481358528137207,
+      "grad_norm_var": 2.365294319547022,
+      "learning_rate": 0.0001,
+      "loss": 0.9589,
+      "loss/crossentropy": 2.477987766265869,
+      "loss/hidden": 0.640625,
+      "loss/logits": 0.11042475700378418,
+      "loss/reg": 0.02078239433467388,
+      "step": 749
+    },
+    {
+      "epoch": 0.09375,
+      "grad_norm": 5.792114734649658,
+      "grad_norm_var": 2.5478865053407236,
+      "learning_rate": 0.0001,
+      "loss": 1.2395,
+      "loss/crossentropy": 2.0092225074768066,
+      "loss/hidden": 0.87109375,
+      "loss/logits": 0.16063663363456726,
+      "loss/reg": 0.020772725343704224,
+      "step": 750
+    },
+    {
+      "epoch": 0.093875,
+      "grad_norm": 3.148350954055786,
+      "grad_norm_var": 2.512823601683457,
+      "learning_rate": 0.0001,
+      "loss": 1.14,
+      "loss/crossentropy": 2.669532537460327,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.15109741687774658,
+      "loss/reg": 0.020762871950864792,
+      "step": 751
+    },
+    {
+      "epoch": 0.094,
+      "grad_norm": 3.2779595851898193,
+      "grad_norm_var": 2.5202896391695124,
+      "learning_rate": 0.0001,
+      "loss": 1.1492,
+      "loss/crossentropy": 2.477522373199463,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.1291784942150116,
+      "loss/reg": 0.020753389224410057,
+      "step": 752
+    },
+    {
+      "epoch": 0.094125,
+      "grad_norm": 2.3718457221984863,
+      "grad_norm_var": 2.586364485192132,
+      "learning_rate": 0.0001,
+      "loss": 1.0399,
+      "loss/crossentropy": 2.497688055038452,
+      "loss/hidden": 0.69921875,
+      "loss/logits": 0.13327988982200623,
+      "loss/reg": 0.02074403502047062,
+      "step": 753
+    },
+    {
+      "epoch": 0.09425,
+      "grad_norm": 4.972538948059082,
+      "grad_norm_var": 2.6852568725766104,
+      "learning_rate": 0.0001,
+      "loss": 1.2511,
+      "loss/crossentropy": 2.232241630554199,
+      "loss/hidden": 0.90234375,
+      "loss/logits": 0.1414394974708557,
+      "loss/reg": 0.020734604448080063,
+      "step": 754
+    },
+    {
+      "epoch": 0.094375,
+      "grad_norm": 2.6991426944732666,
+      "grad_norm_var": 2.7595134043336267,
+      "learning_rate": 0.0001,
+      "loss": 1.001,
+      "loss/crossentropy": 2.596348285675049,
+      "loss/hidden": 0.6640625,
+      "loss/logits": 0.1297152191400528,
+      "loss/reg": 0.020725268870592117,
+      "step": 755
+    },
+    {
+      "epoch": 0.0945,
+      "grad_norm": 2.8633017539978027,
+      "grad_norm_var": 2.753743097466793,
+      "learning_rate": 0.0001,
+      "loss": 1.065,
+      "loss/crossentropy": 2.3198070526123047,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.11169925332069397,
+      "loss/reg": 0.020716087892651558,
+      "step": 756
+    },
+    {
+      "epoch": 0.094625,
+      "grad_norm": 3.628239154815674,
+      "grad_norm_var": 2.733994536045853,
+      "learning_rate": 0.0001,
+      "loss": 1.3141,
+      "loss/crossentropy": 2.1950411796569824,
+      "loss/hidden": 0.92578125,
+      "loss/logits": 0.1812862753868103,
+      "loss/reg": 0.020707255229353905,
+      "step": 757
+    },
+    {
+      "epoch": 0.09475,
+      "grad_norm": 2.805727958679199,
+      "grad_norm_var": 2.753145827217212,
+      "learning_rate": 0.0001,
+      "loss": 1.109,
+      "loss/crossentropy": 2.40027117729187,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.13251210749149323,
+      "loss/reg": 0.02069801278412342,
+      "step": 758
+    },
+    {
+      "epoch": 0.094875,
+      "grad_norm": 3.1892051696777344,
+      "grad_norm_var": 2.7730842000576295,
+      "learning_rate": 0.0001,
+      "loss": 1.1224,
+      "loss/crossentropy": 2.2343788146972656,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.1382053792476654,
+      "loss/reg": 0.020689615979790688,
+      "step": 759
+    },
+    {
+      "epoch": 0.095,
+      "grad_norm": 2.925913095474243,
+      "grad_norm_var": 2.759237877311041,
+      "learning_rate": 0.0001,
+      "loss": 0.9857,
+      "loss/crossentropy": 2.5283889770507812,
+      "loss/hidden": 0.6640625,
+      "loss/logits": 0.11487281322479248,
+      "loss/reg": 0.02068025805056095,
+      "step": 760
+    },
+    {
+      "epoch": 0.095125,
+      "grad_norm": 4.760406970977783,
+      "grad_norm_var": 0.8673601536624308,
+      "learning_rate": 0.0001,
+      "loss": 1.3807,
+      "loss/crossentropy": 2.8629026412963867,
+      "loss/hidden": 0.9765625,
+      "loss/logits": 0.1973596215248108,
+      "loss/reg": 0.020673030987381935,
+      "step": 761
+    },
+    {
+      "epoch": 0.09525,
+      "grad_norm": 2.6182596683502197,
+      "grad_norm_var": 0.9085803501248163,
+      "learning_rate": 0.0001,
+      "loss": 0.9554,
+      "loss/crossentropy": 2.6935925483703613,
+      "loss/hidden": 0.6328125,
+      "loss/logits": 0.11598716676235199,
+      "loss/reg": 0.020663931965827942,
+      "step": 762
+    },
+    {
+      "epoch": 0.095375,
+      "grad_norm": 3.404240131378174,
+      "grad_norm_var": 0.9037375089777697,
+      "learning_rate": 0.0001,
+      "loss": 1.0173,
+      "loss/crossentropy": 2.5958027839660645,
+      "loss/hidden": 0.6953125,
+      "loss/logits": 0.11542315781116486,
+      "loss/reg": 0.020654823631048203,
+      "step": 763
+    },
+    {
+      "epoch": 0.0955,
+      "grad_norm": 3.0021934509277344,
+      "grad_norm_var": 0.9072250591900151,
+      "learning_rate": 0.0001,
+      "loss": 0.9811,
+      "loss/crossentropy": 2.4611666202545166,
+      "loss/hidden": 0.6640625,
+      "loss/logits": 0.11060373485088348,
+      "loss/reg": 0.020645687356591225,
+      "step": 764
+    },
+    {
+      "epoch": 0.095625,
+      "grad_norm": 4.123987674713135,
+      "grad_norm_var": 0.9227216736856043,
+      "learning_rate": 0.0001,
+      "loss": 1.2112,
+      "loss/crossentropy": 2.327146530151367,
+      "loss/hidden": 0.87109375,
+      "loss/logits": 0.13372407853603363,
+      "loss/reg": 0.020637821406126022,
+      "step": 765
+    },
+    {
+      "epoch": 0.09575,
+      "grad_norm": 3.835116147994995,
+      "grad_norm_var": 0.5572045887439032,
+      "learning_rate": 0.0001,
+      "loss": 1.0062,
+      "loss/crossentropy": 2.5438876152038574,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.12415439635515213,
+      "loss/reg": 0.02063015103340149,
+      "step": 766
+    },
+    {
+      "epoch": 0.095875,
+      "grad_norm": 3.2649309635162354,
+      "grad_norm_var": 0.5548939110280107,
+      "learning_rate": 0.0001,
+      "loss": 1.1664,
+      "loss/crossentropy": 2.6621615886688232,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.15938454866409302,
+      "loss/reg": 0.020622732117772102,
+      "step": 767
+    },
+    {
+      "epoch": 0.096,
+      "grad_norm": 3.398061752319336,
+      "grad_norm_var": 0.554498685348062,
+      "learning_rate": 0.0001,
+      "loss": 0.852,
+      "loss/crossentropy": 2.3231897354125977,
+      "loss/hidden": 0.5625,
+      "loss/logits": 0.08329755067825317,
+      "loss/reg": 0.0206154715269804,
+      "step": 768
+    },
+    {
+      "epoch": 0.096125,
+      "grad_norm": 2.8744146823883057,
+      "grad_norm_var": 0.5036373977995244,
+      "learning_rate": 0.0001,
+      "loss": 0.9094,
+      "loss/crossentropy": 2.5140726566314697,
+      "loss/hidden": 0.6171875,
+      "loss/logits": 0.0860939770936966,
+      "loss/reg": 0.020607706159353256,
+      "step": 769
+    },
+    {
+      "epoch": 0.09625,
+      "grad_norm": 8.404803276062012,
+      "grad_norm_var": 1.9605456650252857,
+      "learning_rate": 0.0001,
+      "loss": 1.614,
+      "loss/crossentropy": 2.4072842597961426,
+      "loss/hidden": 1.1328125,
+      "loss/logits": 0.2751774787902832,
+      "loss/reg": 0.020600339397788048,
+      "step": 770
+    },
+    {
+      "epoch": 0.096375,
+      "grad_norm": 2.581511974334717,
+      "grad_norm_var": 1.975733645477993,
+      "learning_rate": 0.0001,
+      "loss": 1.08,
+      "loss/crossentropy": 2.3724427223205566,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.1358003169298172,
+      "loss/reg": 0.02059323526918888,
+      "step": 771
+    },
+    {
+      "epoch": 0.0965,
+      "grad_norm": 3.9817845821380615,
+      "grad_norm_var": 1.9433082266341482,
+      "learning_rate": 0.0001,
+      "loss": 1.0926,
+      "loss/crossentropy": 2.459740400314331,
+      "loss/hidden": 0.765625,
+      "loss/logits": 0.12109124660491943,
+      "loss/reg": 0.02058546058833599,
+      "step": 772
+    },
+    {
+      "epoch": 0.096625,
+      "grad_norm": 7.001491069793701,
+      "grad_norm_var": 2.633487351928299,
+      "learning_rate": 0.0001,
+      "loss": 1.1253,
+      "loss/crossentropy": 2.6342828273773193,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.10705184936523438,
+      "loss/reg": 0.020576275885105133,
+      "step": 773
+    },
+    {
+      "epoch": 0.09675,
+      "grad_norm": 2.4826319217681885,
+      "grad_norm_var": 2.6865387021083524,
+      "learning_rate": 0.0001,
+      "loss": 1.1239,
+      "loss/crossentropy": 2.537883996963501,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.1447601616382599,
+      "loss/reg": 0.02056770585477352,
+      "step": 774
+    },
+    {
+      "epoch": 0.096875,
+      "grad_norm": 3.5470290184020996,
+      "grad_norm_var": 2.6622723084153237,
+      "learning_rate": 0.0001,
+      "loss": 1.1572,
+      "loss/crossentropy": 2.572312831878662,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.15083444118499756,
+      "loss/reg": 0.020558428019285202,
+      "step": 775
+    },
+    {
+      "epoch": 0.097,
+      "grad_norm": 2.8388142585754395,
+      "grad_norm_var": 2.673918444962514,
+      "learning_rate": 0.0001,
+      "loss": 1.1426,
+      "loss/crossentropy": 2.2697503566741943,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.14412574470043182,
+      "loss/reg": 0.020549749955534935,
+      "step": 776
+    },
+    {
+      "epoch": 0.097125,
+      "grad_norm": 2.9166338443756104,
+      "grad_norm_var": 2.670560695292122,
+      "learning_rate": 0.0001,
+      "loss": 0.8926,
+      "loss/crossentropy": 2.4288480281829834,
+      "loss/hidden": 0.58203125,
+      "loss/logits": 0.10516718029975891,
+      "loss/reg": 0.02054043672978878,
+      "step": 777
+    },
+    {
+      "epoch": 0.09725,
+      "grad_norm": 3.565744161605835,
+      "grad_norm_var": 2.58151597609506,
+      "learning_rate": 0.0001,
+      "loss": 1.1299,
+      "loss/crossentropy": 2.444842576980591,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.1472683846950531,
+      "loss/reg": 0.020531047135591507,
+      "step": 778
+    },
+    {
+      "epoch": 0.097375,
+      "grad_norm": 2.8829805850982666,
+      "grad_norm_var": 2.627842889624617,
+      "learning_rate": 0.0001,
+      "loss": 1.1382,
+      "loss/crossentropy": 2.548234462738037,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.1478501409292221,
+      "loss/reg": 0.02052178978919983,
+      "step": 779
+    },
+    {
+      "epoch": 0.0975,
+      "grad_norm": 2.908299207687378,
+      "grad_norm_var": 2.6383052442278556,
+      "learning_rate": 0.0001,
+      "loss": 1.069,
+      "loss/crossentropy": 2.3388137817382812,
+      "loss/hidden": 0.734375,
+      "loss/logits": 0.12949424982070923,
+      "loss/reg": 0.020512979477643967,
+      "step": 780
+    },
+    {
+      "epoch": 0.097625,
+      "grad_norm": 2.5852208137512207,
+      "grad_norm_var": 2.717361748364273,
+      "learning_rate": 0.0001,
+      "loss": 0.9782,
+      "loss/crossentropy": 2.4337894916534424,
+      "loss/hidden": 0.65625,
+      "loss/logits": 0.11689651757478714,
+      "loss/reg": 0.020503604784607887,
+      "step": 781
+    },
+    {
+      "epoch": 0.09775,
+      "grad_norm": 5.328097820281982,
+      "grad_norm_var": 2.885194693951976,
+      "learning_rate": 0.0001,
+      "loss": 1.2053,
+      "loss/crossentropy": 2.5564181804656982,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.19961750507354736,
+      "loss/reg": 0.020494818687438965,
+      "step": 782
+    },
+    {
+      "epoch": 0.097875,
+      "grad_norm": 3.647054672241211,
+      "grad_norm_var": 2.8678156226553613,
+      "learning_rate": 0.0001,
+      "loss": 1.0039,
+      "loss/crossentropy": 2.5830650329589844,
+      "loss/hidden": 0.66796875,
+      "loss/logits": 0.13103139400482178,
+      "loss/reg": 0.020485466346144676,
+      "step": 783
+    },
+    {
+      "epoch": 0.098,
+      "grad_norm": 4.480771541595459,
+      "grad_norm_var": 2.8817531456144723,
+      "learning_rate": 0.0001,
+      "loss": 1.3811,
+      "loss/crossentropy": 2.4550743103027344,
+      "loss/hidden": 0.9765625,
+      "loss/logits": 0.1997724324464798,
+      "loss/reg": 0.020477164536714554,
+      "step": 784
+    },
+    {
+      "epoch": 0.098125,
+      "grad_norm": 3.1495656967163086,
+      "grad_norm_var": 2.8497140664534366,
+      "learning_rate": 0.0001,
+      "loss": 1.1386,
+      "loss/crossentropy": 2.846536636352539,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.1722334325313568,
+      "loss/reg": 0.020469149574637413,
+      "step": 785
+    },
+    {
+      "epoch": 0.09825,
+      "grad_norm": 3.611919641494751,
+      "grad_norm_var": 1.4027508562338329,
+      "learning_rate": 0.0001,
+      "loss": 1.0783,
+      "loss/crossentropy": 2.7328994274139404,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.14321856200695038,
+      "loss/reg": 0.020459884777665138,
+      "step": 786
+    },
+    {
+      "epoch": 0.098375,
+      "grad_norm": 4.535567283630371,
+      "grad_norm_var": 1.3775118805215696,
+      "learning_rate": 0.0001,
+      "loss": 1.3035,
+      "loss/crossentropy": 2.616641044616699,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.24746158719062805,
+      "loss/reg": 0.020450593903660774,
+      "step": 787
+    },
+    {
+      "epoch": 0.0985,
+      "grad_norm": 2.7010657787323,
+      "grad_norm_var": 1.4347220572574786,
+      "learning_rate": 0.0001,
+      "loss": 0.9439,
+      "loss/crossentropy": 2.7851712703704834,
+      "loss/hidden": 0.62890625,
+      "loss/logits": 0.11055716872215271,
+      "loss/reg": 0.020441319793462753,
+      "step": 788
+    },
+    {
+      "epoch": 0.098625,
+      "grad_norm": 3.3690221309661865,
+      "grad_norm_var": 0.6296018822433316,
+      "learning_rate": 0.0001,
+      "loss": 0.9858,
+      "loss/crossentropy": 2.5199151039123535,
+      "loss/hidden": 0.6640625,
+      "loss/logits": 0.11739970743656158,
+      "loss/reg": 0.020432572811841965,
+      "step": 789
+    },
+    {
+      "epoch": 0.09875,
+      "grad_norm": 3.5332016944885254,
+      "grad_norm_var": 0.5687648370759459,
+      "learning_rate": 0.0001,
+      "loss": 1.2114,
+      "loss/crossentropy": 2.3861191272735596,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.19853942096233368,
+      "loss/reg": 0.020423252135515213,
+      "step": 790
+    },
+    {
+      "epoch": 0.098875,
+      "grad_norm": 3.4778192043304443,
+      "grad_norm_var": 0.5684000998912779,
+      "learning_rate": 0.0001,
+      "loss": 1.3337,
+      "loss/crossentropy": 2.177149772644043,
+      "loss/hidden": 0.94921875,
+      "loss/logits": 0.18037351965904236,
+      "loss/reg": 0.020413951948285103,
+      "step": 791
+    },
+    {
+      "epoch": 0.099,
+      "grad_norm": 3.1103479862213135,
+      "grad_norm_var": 0.5501298461305394,
+      "learning_rate": 0.0001,
+      "loss": 1.1811,
+      "loss/crossentropy": 2.381289005279541,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.16453775763511658,
+      "loss/reg": 0.020404649898409843,
+      "step": 792
+    },
+    {
+      "epoch": 0.099125,
+      "grad_norm": 2.7654690742492676,
+      "grad_norm_var": 0.563068172749172,
+      "learning_rate": 0.0001,
+      "loss": 0.9788,
+      "loss/crossentropy": 2.6645448207855225,
+      "loss/hidden": 0.65234375,
+      "loss/logits": 0.12248219549655914,
+      "loss/reg": 0.020395854488015175,
+      "step": 793
+    },
+    {
+      "epoch": 0.09925,
+      "grad_norm": 2.9942195415496826,
+      "grad_norm_var": 0.5768165563916947,
+      "learning_rate": 0.0001,
+      "loss": 1.1054,
+      "loss/crossentropy": 2.388904571533203,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.16325941681861877,
+      "loss/reg": 0.020387381315231323,
+      "step": 794
+    },
+    {
+      "epoch": 0.099375,
+      "grad_norm": 3.1070520877838135,
+      "grad_norm_var": 0.5632370819485725,
+      "learning_rate": 0.0001,
+      "loss": 1.0819,
+      "loss/crossentropy": 2.4645302295684814,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.13203924894332886,
+      "loss/reg": 0.020378144457936287,
+      "step": 795
+    },
+    {
+      "epoch": 0.0995,
+      "grad_norm": 3.5182595252990723,
+      "grad_norm_var": 0.5419026805153273,
+      "learning_rate": 0.0001,
+      "loss": 1.3242,
+      "loss/crossentropy": 2.337388038635254,
+      "loss/hidden": 0.96875,
+      "loss/logits": 0.15180304646492004,
+      "loss/reg": 0.020368557423353195,
+      "step": 796
+    },
+    {
+      "epoch": 0.099625,
+      "grad_norm": 2.4217121601104736,
+      "grad_norm_var": 0.5634005753459926,
+      "learning_rate": 0.0001,
+      "loss": 0.976,
+      "loss/crossentropy": 2.520176887512207,
+      "loss/hidden": 0.6640625,
+      "loss/logits": 0.1083153486251831,
+      "loss/reg": 0.020359758287668228,
+      "step": 797
+    },
+    {
+      "epoch": 0.09975,
+      "grad_norm": 3.818143129348755,
+      "grad_norm_var": 0.33472096860269646,
+      "learning_rate": 0.0001,
+      "loss": 1.2273,
+      "loss/crossentropy": 2.452592134475708,
+      "loss/hidden": 0.86328125,
+      "loss/logits": 0.1604856252670288,
+      "loss/reg": 0.02035023830831051,
+      "step": 798
+    },
+    {
+      "epoch": 0.099875,
+      "grad_norm": 2.8298075199127197,
+      "grad_norm_var": 0.3484620943588491,
+      "learning_rate": 0.0001,
+      "loss": 1.094,
+      "loss/crossentropy": 2.6919286251068115,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.15227138996124268,
+      "loss/reg": 0.020340625196695328,
+      "step": 799
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 45.50175094604492,
+      "grad_norm_var": 111.76340644728633,
+      "learning_rate": 0.0001,
+      "loss": 1.6213,
+      "loss/crossentropy": 2.2076241970062256,
+      "loss/hidden": 1.125,
+      "loss/logits": 0.2930048406124115,
+      "loss/reg": 0.020331410691142082,
+      "step": 800
+    },
+    {
+      "epoch": 0.100125,
+      "grad_norm": 3.073981523513794,
+      "grad_norm_var": 111.7915103772579,
+      "learning_rate": 0.0001,
+      "loss": 1.1146,
+      "loss/crossentropy": 2.5723347663879395,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.13403840363025665,
+      "loss/reg": 0.020322071388363838,
+      "step": 801
+    },
+    {
+      "epoch": 0.10025,
+      "grad_norm": 4.337286472320557,
+      "grad_norm_var": 111.60328751499571,
+      "learning_rate": 0.0001,
+      "loss": 1.2228,
+      "loss/crossentropy": 2.7784037590026855,
+      "loss/hidden": 0.83984375,
+      "loss/logits": 0.17983263731002808,
+      "loss/reg": 0.020312372595071793,
+      "step": 802
+    },
+    {
+      "epoch": 0.100375,
+      "grad_norm": 3.656367778778076,
+      "grad_norm_var": 111.81663718658596,
+      "learning_rate": 0.0001,
+      "loss": 1.1003,
+      "loss/crossentropy": 2.4638426303863525,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.12776704132556915,
+      "loss/reg": 0.020302986726164818,
+      "step": 803
+    },
+    {
+      "epoch": 0.1005,
+      "grad_norm": 3.38935923576355,
+      "grad_norm_var": 111.55373057700994,
+      "learning_rate": 0.0001,
+      "loss": 0.9749,
+      "loss/crossentropy": 2.4947969913482666,
+      "loss/hidden": 0.6640625,
+      "loss/logits": 0.10794391483068466,
+      "loss/reg": 0.020293867215514183,
+      "step": 804
+    },
+    {
+      "epoch": 0.100625,
+      "grad_norm": 3.2775380611419678,
+      "grad_norm_var": 111.58551029522327,
+      "learning_rate": 0.0001,
+      "loss": 1.133,
+      "loss/crossentropy": 2.4812419414520264,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.15277597308158875,
+      "loss/reg": 0.020284701138734818,
+      "step": 805
+    },
+    {
+      "epoch": 0.10075,
+      "grad_norm": 2.996157646179199,
+      "grad_norm_var": 111.77485823890761,
+      "learning_rate": 0.0001,
+      "loss": 1.0998,
+      "loss/crossentropy": 2.516984224319458,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.14709463715553284,
+      "loss/reg": 0.020275365561246872,
+      "step": 806
+    },
+    {
+      "epoch": 0.100875,
+      "grad_norm": 5.575490951538086,
+      "grad_norm_var": 111.37459403701224,
+      "learning_rate": 0.0001,
+      "loss": 1.0629,
+      "loss/crossentropy": 2.6265506744384766,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.12198775261640549,
+      "loss/reg": 0.02026602067053318,
+      "step": 807
+    },
+    {
+      "epoch": 0.101,
+      "grad_norm": 2.653712749481201,
+      "grad_norm_var": 111.56498102164149,
+      "learning_rate": 0.0001,
+      "loss": 0.957,
+      "loss/crossentropy": 2.755121946334839,
+      "loss/hidden": 0.6328125,
+      "loss/logits": 0.12166983634233475,
+      "loss/reg": 0.02025618776679039,
+      "step": 808
+    },
+    {
+      "epoch": 0.101125,
+      "grad_norm": 2.5454790592193604,
+      "grad_norm_var": 111.66272758702647,
+      "learning_rate": 0.0001,
+      "loss": 1.0209,
+      "loss/crossentropy": 2.475360870361328,
+      "loss/hidden": 0.6953125,
+      "loss/logits": 0.1231006383895874,
+      "loss/reg": 0.02024705521762371,
+      "step": 809
+    },
+    {
+      "epoch": 0.10125,
+      "grad_norm": 12.002355575561523,
+      "grad_norm_var": 113.1469842386682,
+      "learning_rate": 0.0001,
+      "loss": 1.1289,
+      "loss/crossentropy": 2.485873222351074,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.1492297500371933,
+      "loss/reg": 0.020237451419234276,
+      "step": 810
+    },
+    {
+      "epoch": 0.101375,
+      "grad_norm": 3.0118675231933594,
+      "grad_norm_var": 113.19117010752396,
+      "learning_rate": 0.0001,
+      "loss": 1.0673,
+      "loss/crossentropy": 2.610520124435425,
+      "loss/hidden": 0.7265625,
+      "loss/logits": 0.13847574591636658,
+      "loss/reg": 0.020227529108524323,
+      "step": 811
+    },
+    {
+      "epoch": 0.1015,
+      "grad_norm": 37.33988952636719,
+      "grad_norm_var": 171.06705552642327,
+      "learning_rate": 0.0001,
+      "loss": 4.5438,
+      "loss/crossentropy": 3.938170909881592,
+      "loss/hidden": 2.203125,
+      "loss/logits": 2.1385304927825928,
+      "loss/reg": 0.02021711878478527,
+      "step": 812
+    },
+    {
+      "epoch": 0.101625,
+      "grad_norm": 2.306735038757324,
+      "grad_norm_var": 171.16339278078715,
+      "learning_rate": 0.0001,
+      "loss": 0.928,
+      "loss/crossentropy": 2.3164937496185303,
+      "loss/hidden": 0.625,
+      "loss/logits": 0.10087703168392181,
+      "loss/reg": 0.020207591354846954,
+      "step": 813
+    },
+    {
+      "epoch": 0.10175,
+      "grad_norm": 3.00903058052063,
+      "grad_norm_var": 171.72501112959992,
+      "learning_rate": 0.0001,
+      "loss": 1.1753,
+      "loss/crossentropy": 2.4386990070343018,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.1452336609363556,
+      "loss/reg": 0.02019745111465454,
+      "step": 814
+    },
+    {
+      "epoch": 0.101875,
+      "grad_norm": 2.680140972137451,
+      "grad_norm_var": 171.84144221114087,
+      "learning_rate": 0.0001,
+      "loss": 1.157,
+      "loss/crossentropy": 2.531343460083008,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.15824541449546814,
+      "loss/reg": 0.020187031477689743,
+      "step": 815
+    },
+    {
+      "epoch": 0.102,
+      "grad_norm": 2.582740068435669,
+      "grad_norm_var": 75.71062264039246,
+      "learning_rate": 0.0001,
+      "loss": 1.0065,
+      "loss/crossentropy": 2.5258586406707764,
+      "loss/hidden": 0.69140625,
+      "loss/logits": 0.11330173909664154,
+      "loss/reg": 0.02017681486904621,
+      "step": 816
+    },
+    {
+      "epoch": 0.102125,
+      "grad_norm": 3.5809519290924072,
+      "grad_norm_var": 75.53549752812218,
+      "learning_rate": 0.0001,
+      "loss": 1.2756,
+      "loss/crossentropy": 2.080437183380127,
+      "loss/hidden": 0.90625,
+      "loss/logits": 0.16763544082641602,
+      "loss/reg": 0.02016652189195156,
+      "step": 817
+    },
+    {
+      "epoch": 0.10225,
+      "grad_norm": 3.093592405319214,
+      "grad_norm_var": 75.89695881356819,
+      "learning_rate": 0.0001,
+      "loss": 1.1484,
+      "loss/crossentropy": 2.630833625793457,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.14600682258605957,
+      "loss/reg": 0.020157409831881523,
+      "step": 818
+    },
+    {
+      "epoch": 0.102375,
+      "grad_norm": 3.1120946407318115,
+      "grad_norm_var": 76.0751246894024,
+      "learning_rate": 0.0001,
+      "loss": 1.038,
+      "loss/crossentropy": 2.5411901473999023,
+      "loss/hidden": 0.69140625,
+      "loss/logits": 0.14511412382125854,
+      "loss/reg": 0.020148303359746933,
+      "step": 819
+    },
+    {
+      "epoch": 0.1025,
+      "grad_norm": 2.9903945922851562,
+      "grad_norm_var": 76.21449508483448,
+      "learning_rate": 0.0001,
+      "loss": 1.0769,
+      "loss/crossentropy": 2.556246042251587,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.13721047341823578,
+      "loss/reg": 0.020139139145612717,
+      "step": 820
+    },
+    {
+      "epoch": 0.102625,
+      "grad_norm": 3.297760248184204,
+      "grad_norm_var": 76.2077263993312,
+      "learning_rate": 0.0001,
+      "loss": 1.025,
+      "loss/crossentropy": 2.5704898834228516,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.14399868249893188,
+      "loss/reg": 0.020129989832639694,
+      "step": 821
+    },
+    {
+      "epoch": 0.10275,
+      "grad_norm": 3.810601234436035,
+      "grad_norm_var": 75.94485425030823,
+      "learning_rate": 0.0001,
+      "loss": 1.2142,
+      "loss/crossentropy": 2.4684388637542725,
+      "loss/hidden": 0.83984375,
+      "loss/logits": 0.17318351566791534,
+      "loss/reg": 0.020120643079280853,
+      "step": 822
+    },
+    {
+      "epoch": 0.102875,
+      "grad_norm": 2.798835515975952,
+      "grad_norm_var": 76.52818091118122,
+      "learning_rate": 0.0001,
+      "loss": 0.9558,
+      "loss/crossentropy": 2.476940393447876,
+      "loss/hidden": 0.6484375,
+      "loss/logits": 0.10626688599586487,
+      "loss/reg": 0.020111503079533577,
+      "step": 823
+    },
+    {
+      "epoch": 0.103,
+      "grad_norm": 4.541812419891357,
+      "grad_norm_var": 75.99013496754353,
+      "learning_rate": 0.0001,
+      "loss": 1.1311,
+      "loss/crossentropy": 2.339744806289673,
+      "loss/hidden": 0.7265625,
+      "loss/logits": 0.2034740447998047,
+      "loss/reg": 0.020102351903915405,
+      "step": 824
+    },
+    {
+      "epoch": 0.103125,
+      "grad_norm": 2.6360268592834473,
+      "grad_norm_var": 75.95142766348106,
+      "learning_rate": 0.0001,
+      "loss": 1.0849,
+      "loss/crossentropy": 2.3795337677001953,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.1378898024559021,
+      "loss/reg": 0.02009383775293827,
+      "step": 825
+    },
+    {
+      "epoch": 0.10325,
+      "grad_norm": 3.2206642627716064,
+      "grad_norm_var": 73.50864103963063,
+      "learning_rate": 0.0001,
+      "loss": 1.09,
+      "loss/crossentropy": 2.645775556564331,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.13137856125831604,
+      "loss/reg": 0.020084405317902565,
+      "step": 826
+    },
+    {
+      "epoch": 0.103375,
+      "grad_norm": 3.2071568965911865,
+      "grad_norm_var": 73.45272548167614,
+      "learning_rate": 0.0001,
+      "loss": 0.9383,
+      "loss/crossentropy": 2.7224762439727783,
+      "loss/hidden": 0.625,
+      "loss/logits": 0.11251779645681381,
+      "loss/reg": 0.020074598491191864,
+      "step": 827
+    },
+    {
+      "epoch": 0.1035,
+      "grad_norm": 3.2850613594055176,
+      "grad_norm_var": 0.2863261800221416,
+      "learning_rate": 0.0001,
+      "loss": 1.1194,
+      "loss/crossentropy": 2.522642135620117,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.17652815580368042,
+      "loss/reg": 0.02006435953080654,
+      "step": 828
+    },
+    {
+      "epoch": 0.103625,
+      "grad_norm": 3.5068256855010986,
+      "grad_norm_var": 0.24387138774257647,
+      "learning_rate": 0.0001,
+      "loss": 0.9457,
+      "loss/crossentropy": 2.6463444232940674,
+      "loss/hidden": 0.6328125,
+      "loss/logits": 0.11229754984378815,
+      "loss/reg": 0.020054515451192856,
+      "step": 829
+    },
+    {
+      "epoch": 0.10375,
+      "grad_norm": 4.16710090637207,
+      "grad_norm_var": 0.29672115328222404,
+      "learning_rate": 0.0001,
+      "loss": 1.301,
+      "loss/crossentropy": 2.519270420074463,
+      "loss/hidden": 0.921875,
+      "loss/logits": 0.1786651611328125,
+      "loss/reg": 0.020044928416609764,
+      "step": 830
+    },
+    {
+      "epoch": 0.103875,
+      "grad_norm": 3.5839273929595947,
+      "grad_norm_var": 0.27524789373512704,
+      "learning_rate": 0.0001,
+      "loss": 0.9949,
+      "loss/crossentropy": 2.0604701042175293,
+      "loss/hidden": 0.68359375,
+      "loss/logits": 0.1109282597899437,
+      "loss/reg": 0.02003585919737816,
+      "step": 831
+    },
+    {
+      "epoch": 0.104,
+      "grad_norm": 2.8121650218963623,
+      "grad_norm_var": 0.25541980739101955,
+      "learning_rate": 0.0001,
+      "loss": 1.1057,
+      "loss/crossentropy": 2.670191764831543,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.1515616476535797,
+      "loss/reg": 0.020026110112667084,
+      "step": 832
+    },
+    {
+      "epoch": 0.104125,
+      "grad_norm": 3.116018295288086,
+      "grad_norm_var": 0.2547872758708628,
+      "learning_rate": 0.0001,
+      "loss": 1.0301,
+      "loss/crossentropy": 3.016143321990967,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.1502160131931305,
+      "loss/reg": 0.020016156136989594,
+      "step": 833
+    },
+    {
+      "epoch": 0.10425,
+      "grad_norm": 6.202447414398193,
+      "grad_norm_var": 0.7634439694535559,
+      "learning_rate": 0.0001,
+      "loss": 1.2045,
+      "loss/crossentropy": 2.8383448123931885,
+      "loss/hidden": 0.84765625,
+      "loss/logits": 0.15678462386131287,
+      "loss/reg": 0.020007088780403137,
+      "step": 834
+    },
+    {
+      "epoch": 0.104375,
+      "grad_norm": 3.4904069900512695,
+      "grad_norm_var": 0.7519116349075012,
+      "learning_rate": 0.0001,
+      "loss": 1.2658,
+      "loss/crossentropy": 2.597175359725952,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.22209098935127258,
+      "loss/reg": 0.019998185336589813,
+      "step": 835
+    },
+    {
+      "epoch": 0.1045,
+      "grad_norm": 4.6242594718933105,
+      "grad_norm_var": 0.7986550791863064,
+      "learning_rate": 0.0001,
+      "loss": 1.1411,
+      "loss/crossentropy": 2.6017813682556152,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.14435435831546783,
+      "loss/reg": 0.019988389685750008,
+      "step": 836
+    },
+    {
+      "epoch": 0.104625,
+      "grad_norm": 3.2676827907562256,
+      "grad_norm_var": 0.800099420481778,
+      "learning_rate": 0.0001,
+      "loss": 1.1908,
+      "loss/crossentropy": 2.454334259033203,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.16681547462940216,
+      "loss/reg": 0.019979091361165047,
+      "step": 837
+    },
+    {
+      "epoch": 0.10475,
+      "grad_norm": 3.4931089878082275,
+      "grad_norm_var": 0.7992595598721048,
+      "learning_rate": 0.0001,
+      "loss": 1.1803,
+      "loss/crossentropy": 2.2890915870666504,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.15247204899787903,
+      "loss/reg": 0.019969170913100243,
+      "step": 838
+    },
+    {
+      "epoch": 0.104875,
+      "grad_norm": 2.7106313705444336,
+      "grad_norm_var": 0.8094277801425143,
+      "learning_rate": 0.0001,
+      "loss": 1.097,
+      "loss/crossentropy": 2.56706166267395,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.12783397734165192,
+      "loss/reg": 0.019960079342126846,
+      "step": 839
+    },
+    {
+      "epoch": 0.105,
+      "grad_norm": 3.2825114727020264,
+      "grad_norm_var": 0.7531900707246374,
+      "learning_rate": 0.0001,
+      "loss": 1.2726,
+      "loss/crossentropy": 2.324145555496216,
+      "loss/hidden": 0.88671875,
+      "loss/logits": 0.18636029958724976,
+      "loss/reg": 0.019951237365603447,
+      "step": 840
+    },
+    {
+      "epoch": 0.105125,
+      "grad_norm": 2.752570152282715,
+      "grad_norm_var": 0.7400250579900473,
+      "learning_rate": 0.0001,
+      "loss": 1.0885,
+      "loss/crossentropy": 2.4273452758789062,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.1312946379184723,
+      "loss/reg": 0.01994233950972557,
+      "step": 841
+    },
+    {
+      "epoch": 0.10525,
+      "grad_norm": 3.451720714569092,
+      "grad_norm_var": 0.733364881032247,
+      "learning_rate": 0.0001,
+      "loss": 1.1042,
+      "loss/crossentropy": 2.5271100997924805,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.1431449055671692,
+      "loss/reg": 0.019932815805077553,
+      "step": 842
+    },
+    {
+      "epoch": 0.105375,
+      "grad_norm": 3.0586564540863037,
+      "grad_norm_var": 0.741721542830341,
+      "learning_rate": 0.0001,
+      "loss": 1.0412,
+      "loss/crossentropy": 2.9447247982025146,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.13100013136863708,
+      "loss/reg": 0.019923273473978043,
+      "step": 843
+    },
+    {
+      "epoch": 0.1055,
+      "grad_norm": 3.4214084148406982,
+      "grad_norm_var": 0.7380611813534226,
+      "learning_rate": 0.0001,
+      "loss": 1.2281,
+      "loss/crossentropy": 2.3005340099334717,
+      "loss/hidden": 0.86328125,
+      "loss/logits": 0.1657242327928543,
+      "loss/reg": 0.01991339959204197,
+      "step": 844
+    },
+    {
+      "epoch": 0.105625,
+      "grad_norm": 2.6607367992401123,
+      "grad_norm_var": 0.7886706735221035,
+      "learning_rate": 0.0001,
+      "loss": 1.1345,
+      "loss/crossentropy": 2.5435853004455566,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.1347101330757141,
+      "loss/reg": 0.019903138279914856,
+      "step": 845
+    },
+    {
+      "epoch": 0.10575,
+      "grad_norm": 3.0965163707733154,
+      "grad_norm_var": 0.7659307635756494,
+      "learning_rate": 0.0001,
+      "loss": 1.0956,
+      "loss/crossentropy": 2.6083555221557617,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.1505199819803238,
+      "loss/reg": 0.019893797114491463,
+      "step": 846
+    },
+    {
+      "epoch": 0.105875,
+      "grad_norm": 4.309004306793213,
+      "grad_norm_var": 0.8127957898222188,
+      "learning_rate": 0.0001,
+      "loss": 1.078,
+      "loss/crossentropy": 2.515653371810913,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.11742238700389862,
+      "loss/reg": 0.019883660599589348,
+      "step": 847
+    },
+    {
+      "epoch": 0.106,
+      "grad_norm": 3.248624563217163,
+      "grad_norm_var": 0.7855834171862691,
+      "learning_rate": 0.0001,
+      "loss": 1.0838,
+      "loss/crossentropy": 2.593562126159668,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.13119123876094818,
+      "loss/reg": 0.019874349236488342,
+      "step": 848
+    },
+    {
+      "epoch": 0.106125,
+      "grad_norm": 3.347926378250122,
+      "grad_norm_var": 0.7767115778534122,
+      "learning_rate": 0.0001,
+      "loss": 1.0816,
+      "loss/crossentropy": 2.3947036266326904,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.13295108079910278,
+      "loss/reg": 0.019864298403263092,
+      "step": 849
+    },
+    {
+      "epoch": 0.10625,
+      "grad_norm": 4.369658946990967,
+      "grad_norm_var": 0.33264170947598637,
+      "learning_rate": 0.0001,
+      "loss": 1.0708,
+      "loss/crossentropy": 2.449446201324463,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.13007891178131104,
+      "loss/reg": 0.019854165613651276,
+      "step": 850
+    },
+    {
+      "epoch": 0.106375,
+      "grad_norm": 3.361130475997925,
+      "grad_norm_var": 0.332327660409797,
+      "learning_rate": 0.0001,
+      "loss": 0.9761,
+      "loss/crossentropy": 2.566408157348633,
+      "loss/hidden": 0.66796875,
+      "loss/logits": 0.10966208577156067,
+      "loss/reg": 0.019843947142362595,
+      "step": 851
+    },
+    {
+      "epoch": 0.1065,
+      "grad_norm": 2.6885204315185547,
+      "grad_norm_var": 0.25144665871681204,
+      "learning_rate": 0.0001,
+      "loss": 0.9331,
+      "loss/crossentropy": 2.3620035648345947,
+      "loss/hidden": 0.625,
+      "loss/logits": 0.10973039269447327,
+      "loss/reg": 0.019834715873003006,
+      "step": 852
+    },
+    {
+      "epoch": 0.106625,
+      "grad_norm": 2.615734338760376,
+      "grad_norm_var": 0.2793016853206145,
+      "learning_rate": 0.0001,
+      "loss": 1.0168,
+      "loss/crossentropy": 2.605545997619629,
+      "loss/hidden": 0.69140625,
+      "loss/logits": 0.12715153396129608,
+      "loss/reg": 0.01982559822499752,
+      "step": 853
+    },
+    {
+      "epoch": 0.10675,
+      "grad_norm": 3.073012113571167,
+      "grad_norm_var": 0.276254032788457,
+      "learning_rate": 0.0001,
+      "loss": 0.9855,
+      "loss/crossentropy": 2.634042739868164,
+      "loss/hidden": 0.671875,
+      "loss/logits": 0.11542729288339615,
+      "loss/reg": 0.019816165789961815,
+      "step": 854
+    },
+    {
+      "epoch": 0.106875,
+      "grad_norm": 2.759152412414551,
+      "grad_norm_var": 0.27313479552050995,
+      "learning_rate": 0.0001,
+      "loss": 0.9145,
+      "loss/crossentropy": 2.4920616149902344,
+      "loss/hidden": 0.61328125,
+      "loss/logits": 0.10316716134548187,
+      "loss/reg": 0.019806833937764168,
+      "step": 855
+    },
+    {
+      "epoch": 0.107,
+      "grad_norm": 3.647084951400757,
+      "grad_norm_var": 0.28455080731760823,
+      "learning_rate": 0.0001,
+      "loss": 1.1336,
+      "loss/crossentropy": 2.34993314743042,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.150485098361969,
+      "loss/reg": 0.019797123968601227,
+      "step": 856
+    },
+    {
+      "epoch": 0.107125,
+      "grad_norm": 5.309474468231201,
+      "grad_norm_var": 0.5265287098230971,
+      "learning_rate": 0.0001,
+      "loss": 1.3513,
+      "loss/crossentropy": 2.4511518478393555,
+      "loss/hidden": 0.94140625,
+      "loss/logits": 0.21205759048461914,
+      "loss/reg": 0.01978708617389202,
+      "step": 857
+    },
+    {
+      "epoch": 0.10725,
+      "grad_norm": 3.1982548236846924,
+      "grad_norm_var": 0.5288348795583182,
+      "learning_rate": 0.0001,
+      "loss": 1.1847,
+      "loss/crossentropy": 2.2767844200134277,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.15097308158874512,
+      "loss/reg": 0.01977648213505745,
+      "step": 858
+    },
+    {
+      "epoch": 0.107375,
+      "grad_norm": 3.7261335849761963,
+      "grad_norm_var": 0.5276094221235986,
+      "learning_rate": 0.0001,
+      "loss": 1.1138,
+      "loss/crossentropy": 2.660844087600708,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.13101539015769958,
+      "loss/reg": 0.01976662687957287,
+      "step": 859
+    },
+    {
+      "epoch": 0.1075,
+      "grad_norm": 3.3435637950897217,
+      "grad_norm_var": 0.5280464375318101,
+      "learning_rate": 0.0001,
+      "loss": 1.2877,
+      "loss/crossentropy": 2.4701263904571533,
+      "loss/hidden": 0.90625,
+      "loss/logits": 0.18388135731220245,
+      "loss/reg": 0.019757471978664398,
+      "step": 860
+    },
+    {
+      "epoch": 0.107625,
+      "grad_norm": 2.630688190460205,
+      "grad_norm_var": 0.5311534898567278,
+      "learning_rate": 0.0001,
+      "loss": 1.063,
+      "loss/crossentropy": 2.5359132289886475,
+      "loss/hidden": 0.734375,
+      "loss/logits": 0.1311398446559906,
+      "loss/reg": 0.019748201593756676,
+      "step": 861
+    },
+    {
+      "epoch": 0.10775,
+      "grad_norm": 2.5543456077575684,
+      "grad_norm_var": 0.5729300014134933,
+      "learning_rate": 0.0001,
+      "loss": 0.987,
+      "loss/crossentropy": 2.3763418197631836,
+      "loss/hidden": 0.671875,
+      "loss/logits": 0.11773102730512619,
+      "loss/reg": 0.01973855495452881,
+      "step": 862
+    },
+    {
+      "epoch": 0.107875,
+      "grad_norm": 2.8768351078033447,
+      "grad_norm_var": 0.5249464789316676,
+      "learning_rate": 0.0001,
+      "loss": 1.11,
+      "loss/crossentropy": 2.467682361602783,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.13924749195575714,
+      "loss/reg": 0.019729435443878174,
+      "step": 863
+    },
+    {
+      "epoch": 0.108,
+      "grad_norm": 9.608988761901855,
+      "grad_norm_var": 3.012409881249372,
+      "learning_rate": 0.0001,
+      "loss": 1.8371,
+      "loss/crossentropy": 2.4410533905029297,
+      "loss/hidden": 1.359375,
+      "loss/logits": 0.280517578125,
+      "loss/reg": 0.019719891250133514,
+      "step": 864
+    },
+    {
+      "epoch": 0.108125,
+      "grad_norm": 2.6161296367645264,
+      "grad_norm_var": 3.079687357926654,
+      "learning_rate": 0.0001,
+      "loss": 0.9254,
+      "loss/crossentropy": 2.656090497970581,
+      "loss/hidden": 0.6171875,
+      "loss/logits": 0.11112320423126221,
+      "loss/reg": 0.01971041038632393,
+      "step": 865
+    },
+    {
+      "epoch": 0.10825,
+      "grad_norm": 6.282230377197266,
+      "grad_norm_var": 3.4921671952336317,
+      "learning_rate": 0.0001,
+      "loss": 1.1269,
+      "loss/crossentropy": 2.4799087047576904,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.12914146482944489,
+      "loss/reg": 0.01970127783715725,
+      "step": 866
+    },
+    {
+      "epoch": 0.108375,
+      "grad_norm": 3.052783727645874,
+      "grad_norm_var": 3.514845564297898,
+      "learning_rate": 0.0001,
+      "loss": 1.1383,
+      "loss/crossentropy": 2.4700145721435547,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.15620394051074982,
+      "loss/reg": 0.019691679626703262,
+      "step": 867
+    },
+    {
+      "epoch": 0.1085,
+      "grad_norm": 3.2105300426483154,
+      "grad_norm_var": 3.458070348929603,
+      "learning_rate": 0.0001,
+      "loss": 1.067,
+      "loss/crossentropy": 2.135777473449707,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.11233663558959961,
+      "loss/reg": 0.019682079553604126,
+      "step": 868
+    },
+    {
+      "epoch": 0.108625,
+      "grad_norm": 2.7341737747192383,
+      "grad_norm_var": 3.4405364793380135,
+      "learning_rate": 0.0001,
+      "loss": 1.0495,
+      "loss/crossentropy": 2.4592809677124023,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.1223248764872551,
+      "loss/reg": 0.019672293215990067,
+      "step": 869
+    },
+    {
+      "epoch": 0.10875,
+      "grad_norm": 4.007035732269287,
+      "grad_norm_var": 3.4058996890488658,
+      "learning_rate": 0.0001,
+      "loss": 1.1341,
+      "loss/crossentropy": 2.5008175373077393,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.13670845329761505,
+      "loss/reg": 0.01966211199760437,
+      "step": 870
+    },
+    {
+      "epoch": 0.108875,
+      "grad_norm": 3.146348476409912,
+      "grad_norm_var": 3.359090924722083,
+      "learning_rate": 0.0001,
+      "loss": 1.2459,
+      "loss/crossentropy": 2.0440711975097656,
+      "loss/hidden": 0.8984375,
+      "loss/logits": 0.15090960264205933,
+      "loss/reg": 0.019652366638183594,
+      "step": 871
+    },
+    {
+      "epoch": 0.109,
+      "grad_norm": 4.632335662841797,
+      "grad_norm_var": 3.3902752468766453,
+      "learning_rate": 0.0001,
+      "loss": 1.1446,
+      "loss/crossentropy": 2.4574947357177734,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.13958214223384857,
+      "loss/reg": 0.01964336633682251,
+      "step": 872
+    },
+    {
+      "epoch": 0.109125,
+      "grad_norm": 5.440892219543457,
+      "grad_norm_var": 3.415471723579617,
+      "learning_rate": 0.0001,
+      "loss": 1.1098,
+      "loss/crossentropy": 2.565347671508789,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.10876456648111343,
+      "loss/reg": 0.01963435485959053,
+      "step": 873
+    },
+    {
+      "epoch": 0.10925,
+      "grad_norm": 2.873284101486206,
+      "grad_norm_var": 3.454269091905695,
+      "learning_rate": 0.0001,
+      "loss": 1.0607,
+      "loss/crossentropy": 2.561459541320801,
+      "loss/hidden": 0.72265625,
+      "loss/logits": 0.14177533984184265,
+      "loss/reg": 0.01962495781481266,
+      "step": 874
+    },
+    {
+      "epoch": 0.109375,
+      "grad_norm": 3.0819554328918457,
+      "grad_norm_var": 3.496943197417559,
+      "learning_rate": 0.0001,
+      "loss": 1.2422,
+      "loss/crossentropy": 2.2944936752319336,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.16319133341312408,
+      "loss/reg": 0.019616009667515755,
+      "step": 875
+    },
+    {
+      "epoch": 0.1095,
+      "grad_norm": 4.361453533172607,
+      "grad_norm_var": 3.488792217244569,
+      "learning_rate": 0.0001,
+      "loss": 1.3828,
+      "loss/crossentropy": 2.4624311923980713,
+      "loss/hidden": 1.0078125,
+      "loss/logits": 0.1789003610610962,
+      "loss/reg": 0.019607286900281906,
+      "step": 876
+    },
+    {
+      "epoch": 0.109625,
+      "grad_norm": 3.34078049659729,
+      "grad_norm_var": 3.3959280790074216,
+      "learning_rate": 0.0001,
+      "loss": 0.9589,
+      "loss/crossentropy": 2.4549739360809326,
+      "loss/hidden": 0.64453125,
+      "loss/logits": 0.11839590966701508,
+      "loss/reg": 0.019598115235567093,
+      "step": 877
+    },
+    {
+      "epoch": 0.10975,
+      "grad_norm": 3.434715986251831,
+      "grad_norm_var": 3.2759937907981884,
+      "learning_rate": 0.0001,
+      "loss": 1.1192,
+      "loss/crossentropy": 2.571798086166382,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.16550706326961517,
+      "loss/reg": 0.01958884485065937,
+      "step": 878
+    },
+    {
+      "epoch": 0.109875,
+      "grad_norm": 3.593993663787842,
+      "grad_norm_var": 3.196554005024426,
+      "learning_rate": 0.0001,
+      "loss": 1.1792,
+      "loss/crossentropy": 2.447843551635742,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.15137381851673126,
+      "loss/reg": 0.019579457119107246,
+      "step": 879
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.3654563426971436,
+      "grad_norm_var": 1.0373482238282006,
+      "learning_rate": 0.0001,
+      "loss": 1.0326,
+      "loss/crossentropy": 2.6210246086120605,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.12209475785493851,
+      "loss/reg": 0.019569827243685722,
+      "step": 880
+    },
+    {
+      "epoch": 0.110125,
+      "grad_norm": 2.6370863914489746,
+      "grad_norm_var": 1.0343516088559113,
+      "learning_rate": 0.0001,
+      "loss": 1.1442,
+      "loss/crossentropy": 2.2681713104248047,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.132216215133667,
+      "loss/reg": 0.019560784101486206,
+      "step": 881
+    },
+    {
+      "epoch": 0.11025,
+      "grad_norm": 2.152343273162842,
+      "grad_norm_var": 0.6782700998492743,
+      "learning_rate": 0.0001,
+      "loss": 1.0136,
+      "loss/crossentropy": 2.437594175338745,
+      "loss/hidden": 0.70703125,
+      "loss/logits": 0.11108942329883575,
+      "loss/reg": 0.019551947712898254,
+      "step": 882
+    },
+    {
+      "epoch": 0.110375,
+      "grad_norm": 3.2755074501037598,
+      "grad_norm_var": 0.6698247850929626,
+      "learning_rate": 0.0001,
+      "loss": 0.9972,
+      "loss/crossentropy": 2.481198310852051,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.12207823246717453,
+      "loss/reg": 0.01954270713031292,
+      "step": 883
+    },
+    {
+      "epoch": 0.1105,
+      "grad_norm": 3.770535707473755,
+      "grad_norm_var": 0.6711344077538037,
+      "learning_rate": 0.0001,
+      "loss": 1.0475,
+      "loss/crossentropy": 2.434851884841919,
+      "loss/hidden": 0.71875,
+      "loss/logits": 0.13345816731452942,
+      "loss/reg": 0.019533507525920868,
+      "step": 884
+    },
+    {
+      "epoch": 0.110625,
+      "grad_norm": 2.7666234970092773,
+      "grad_norm_var": 0.6679279033368438,
+      "learning_rate": 0.0001,
+      "loss": 1.1166,
+      "loss/crossentropy": 2.433852434158325,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.1401294767856598,
+      "loss/reg": 0.01952442154288292,
+      "step": 885
+    },
+    {
+      "epoch": 0.11075,
+      "grad_norm": 2.7105534076690674,
+      "grad_norm_var": 0.6840409496040507,
+      "learning_rate": 0.0001,
+      "loss": 1.0924,
+      "loss/crossentropy": 2.219019889831543,
+      "loss/hidden": 0.765625,
+      "loss/logits": 0.13160449266433716,
+      "loss/reg": 0.019515201449394226,
+      "step": 886
+    },
+    {
+      "epoch": 0.110875,
+      "grad_norm": 2.8931920528411865,
+      "grad_norm_var": 0.6969961519386968,
+      "learning_rate": 0.0001,
+      "loss": 1.1134,
+      "loss/crossentropy": 2.511371612548828,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.13708999752998352,
+      "loss/reg": 0.019505700096488,
+      "step": 887
+    },
+    {
+      "epoch": 0.111,
+      "grad_norm": 2.821718215942383,
+      "grad_norm_var": 0.6033415037749854,
+      "learning_rate": 0.0001,
+      "loss": 1.2446,
+      "loss/crossentropy": 2.389580011367798,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.16678079962730408,
+      "loss/reg": 0.019496839493513107,
+      "step": 888
+    },
+    {
+      "epoch": 0.111125,
+      "grad_norm": 2.833461284637451,
+      "grad_norm_var": 0.2778808504856213,
+      "learning_rate": 0.0001,
+      "loss": 1.029,
+      "loss/crossentropy": 2.1618683338165283,
+      "loss/hidden": 0.70703125,
+      "loss/logits": 0.12710769474506378,
+      "loss/reg": 0.019487854093313217,
+      "step": 889
+    },
+    {
+      "epoch": 0.11125,
+      "grad_norm": 3.3806753158569336,
+      "grad_norm_var": 0.2773113837378702,
+      "learning_rate": 0.0001,
+      "loss": 1.1349,
+      "loss/crossentropy": 2.656121253967285,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.1471368372440338,
+      "loss/reg": 0.019478676840662956,
+      "step": 890
+    },
+    {
+      "epoch": 0.111375,
+      "grad_norm": 2.932758092880249,
+      "grad_norm_var": 0.2800811641911004,
+      "learning_rate": 0.0001,
+      "loss": 1.1,
+      "loss/crossentropy": 2.5325851440429688,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.14360320568084717,
+      "loss/reg": 0.019469575956463814,
+      "step": 891
+    },
+    {
+      "epoch": 0.1115,
+      "grad_norm": 2.3603265285491943,
+      "grad_norm_var": 0.20497304301798067,
+      "learning_rate": 0.0001,
+      "loss": 1.0739,
+      "loss/crossentropy": 2.3697421550750732,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.1292482614517212,
+      "loss/reg": 0.01946048066020012,
+      "step": 892
+    },
+    {
+      "epoch": 0.111625,
+      "grad_norm": 4.056443214416504,
+      "grad_norm_var": 0.2678930990243863,
+      "learning_rate": 0.0001,
+      "loss": 1.0008,
+      "loss/crossentropy": 2.7553114891052246,
+      "loss/hidden": 0.68359375,
+      "loss/logits": 0.12268239259719849,
+      "loss/reg": 0.019451187923550606,
+      "step": 893
+    },
+    {
+      "epoch": 0.11175,
+      "grad_norm": 3.346064329147339,
+      "grad_norm_var": 0.2639738255705176,
+      "learning_rate": 0.0001,
+      "loss": 1.1791,
+      "loss/crossentropy": 2.4348855018615723,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.19565626978874207,
+      "loss/reg": 0.019442636519670486,
+      "step": 894
+    },
+    {
+      "epoch": 0.111875,
+      "grad_norm": 6.878971576690674,
+      "grad_norm_var": 1.1740357353356365,
+      "learning_rate": 0.0001,
+      "loss": 1.7085,
+      "loss/crossentropy": 2.9299001693725586,
+      "loss/hidden": 1.1640625,
+      "loss/logits": 0.3501082956790924,
+      "loss/reg": 0.019433531910181046,
+      "step": 895
+    },
+    {
+      "epoch": 0.112,
+      "grad_norm": 3.1834142208099365,
+      "grad_norm_var": 1.1735802221223497,
+      "learning_rate": 0.0001,
+      "loss": 1.2953,
+      "loss/crossentropy": 2.4165709018707275,
+      "loss/hidden": 0.92578125,
+      "loss/logits": 0.17523059248924255,
+      "loss/reg": 0.019424354657530785,
+      "step": 896
+    },
+    {
+      "epoch": 0.112125,
+      "grad_norm": 2.5551555156707764,
+      "grad_norm_var": 1.180695081530242,
+      "learning_rate": 0.0001,
+      "loss": 1.0171,
+      "loss/crossentropy": 2.4345083236694336,
+      "loss/hidden": 0.6953125,
+      "loss/logits": 0.1276446282863617,
+      "loss/reg": 0.019415004178881645,
+      "step": 897
+    },
+    {
+      "epoch": 0.11225,
+      "grad_norm": 3.5786855220794678,
+      "grad_norm_var": 1.1000748366509354,
+      "learning_rate": 0.0001,
+      "loss": 0.9925,
+      "loss/crossentropy": 2.464219808578491,
+      "loss/hidden": 0.69140625,
+      "loss/logits": 0.10706112533807755,
+      "loss/reg": 0.01940576173365116,
+      "step": 898
+    },
+    {
+      "epoch": 0.112375,
+      "grad_norm": 8.784457206726074,
+      "grad_norm_var": 2.9538895197120096,
+      "learning_rate": 0.0001,
+      "loss": 1.6799,
+      "loss/crossentropy": 2.3273468017578125,
+      "loss/hidden": 1.2265625,
+      "loss/logits": 0.25934016704559326,
+      "loss/reg": 0.019396713003516197,
+      "step": 899
+    },
+    {
+      "epoch": 0.1125,
+      "grad_norm": 4.44765567779541,
+      "grad_norm_var": 2.990871190956643,
+      "learning_rate": 0.0001,
+      "loss": 1.0685,
+      "loss/crossentropy": 2.5760438442230225,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.14414295554161072,
+      "loss/reg": 0.019387517124414444,
+      "step": 900
+    },
+    {
+      "epoch": 0.112625,
+      "grad_norm": 4.9651265144348145,
+      "grad_norm_var": 3.013306784613239,
+      "learning_rate": 0.0001,
+      "loss": 1.1917,
+      "loss/crossentropy": 2.5148589611053467,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.1697455495595932,
+      "loss/reg": 0.01937839388847351,
+      "step": 901
+    },
+    {
+      "epoch": 0.11275,
+      "grad_norm": 3.8322513103485107,
+      "grad_norm_var": 2.9203267227302745,
+      "learning_rate": 0.0001,
+      "loss": 1.2309,
+      "loss/crossentropy": 2.536973476409912,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.15436521172523499,
+      "loss/reg": 0.019369108602404594,
+      "step": 902
+    },
+    {
+      "epoch": 0.112875,
+      "grad_norm": 3.5994656085968018,
+      "grad_norm_var": 2.8540415836766555,
+      "learning_rate": 0.0001,
+      "loss": 1.0761,
+      "loss/crossentropy": 2.3160693645477295,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.12077254056930542,
+      "loss/reg": 0.019359666854143143,
+      "step": 903
+    },
+    {
+      "epoch": 0.113,
+      "grad_norm": 4.031139373779297,
+      "grad_norm_var": 2.7599236229360753,
+      "learning_rate": 0.0001,
+      "loss": 1.0988,
+      "loss/crossentropy": 2.5636203289031982,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.12018904089927673,
+      "loss/reg": 0.019350115209817886,
+      "step": 904
+    },
+    {
+      "epoch": 0.113125,
+      "grad_norm": 3.2178378105163574,
+      "grad_norm_var": 2.7069185907568794,
+      "learning_rate": 0.0001,
+      "loss": 1.0472,
+      "loss/crossentropy": 2.3457703590393066,
+      "loss/hidden": 0.72265625,
+      "loss/logits": 0.13114379346370697,
+      "loss/reg": 0.019340479746460915,
+      "step": 905
+    },
+    {
+      "epoch": 0.11325,
+      "grad_norm": 3.096679925918579,
+      "grad_norm_var": 2.7381334427643536,
+      "learning_rate": 0.0001,
+      "loss": 0.9662,
+      "loss/crossentropy": 2.3510820865631104,
+      "loss/hidden": 0.671875,
+      "loss/logits": 0.10106582939624786,
+      "loss/reg": 0.019330844283103943,
+      "step": 906
+    },
+    {
+      "epoch": 0.113375,
+      "grad_norm": 7.590158462524414,
+      "grad_norm_var": 3.3974738441650887,
+      "learning_rate": 0.0001,
+      "loss": 1.6666,
+      "loss/crossentropy": 2.3520147800445557,
+      "loss/hidden": 1.2265625,
+      "loss/logits": 0.2468346357345581,
+      "loss/reg": 0.019321195781230927,
+      "step": 907
+    },
+    {
+      "epoch": 0.1135,
+      "grad_norm": 3.3216323852539062,
+      "grad_norm_var": 3.20081618522182,
+      "learning_rate": 0.0001,
+      "loss": 1.2612,
+      "loss/crossentropy": 2.330040216445923,
+      "loss/hidden": 0.90234375,
+      "loss/logits": 0.16571447253227234,
+      "loss/reg": 0.019311606884002686,
+      "step": 908
+    },
+    {
+      "epoch": 0.113625,
+      "grad_norm": 3.4920501708984375,
+      "grad_norm_var": 3.246978809627046,
+      "learning_rate": 0.0001,
+      "loss": 1.1161,
+      "loss/crossentropy": 2.598465919494629,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.15353354811668396,
+      "loss/reg": 0.019302019849419594,
+      "step": 909
+    },
+    {
+      "epoch": 0.11375,
+      "grad_norm": 3.0414321422576904,
+      "grad_norm_var": 3.294370585536838,
+      "learning_rate": 0.0001,
+      "loss": 1.0693,
+      "loss/crossentropy": 2.459712028503418,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.13418710231781006,
+      "loss/reg": 0.019292324781417847,
+      "step": 910
+    },
+    {
+      "epoch": 0.113875,
+      "grad_norm": 3.131361722946167,
+      "grad_norm_var": 2.908980195007492,
+      "learning_rate": 0.0001,
+      "loss": 1.2301,
+      "loss/crossentropy": 2.268738269805908,
+      "loss/hidden": 0.875,
+      "loss/logits": 0.16223490238189697,
+      "loss/reg": 0.019283456727862358,
+      "step": 911
+    },
+    {
+      "epoch": 0.114,
+      "grad_norm": 3.160707950592041,
+      "grad_norm_var": 2.9118381902992776,
+      "learning_rate": 0.0001,
+      "loss": 1.3173,
+      "loss/crossentropy": 2.2185730934143066,
+      "loss/hidden": 0.9453125,
+      "loss/logits": 0.17927365005016327,
+      "loss/reg": 0.019274268299341202,
+      "step": 912
+    },
+    {
+      "epoch": 0.114125,
+      "grad_norm": 14.604296684265137,
+      "grad_norm_var": 9.479147248477696,
+      "learning_rate": 0.0001,
+      "loss": 1.4278,
+      "loss/crossentropy": 2.2313976287841797,
+      "loss/hidden": 1.078125,
+      "loss/logits": 0.15699920058250427,
+      "loss/reg": 0.019266733899712563,
+      "step": 913
+    },
+    {
+      "epoch": 0.11425,
+      "grad_norm": 3.744290590286255,
+      "grad_norm_var": 9.452382803070204,
+      "learning_rate": 0.0001,
+      "loss": 1.1776,
+      "loss/crossentropy": 2.689699172973633,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.15302729606628418,
+      "loss/reg": 0.019258547574281693,
+      "step": 914
+    },
+    {
+      "epoch": 0.114375,
+      "grad_norm": 2.999354839324951,
+      "grad_norm_var": 8.531466626401157,
+      "learning_rate": 0.0001,
+      "loss": 1.3107,
+      "loss/crossentropy": 1.8664510250091553,
+      "loss/hidden": 0.9453125,
+      "loss/logits": 0.1728517711162567,
+      "loss/reg": 0.01925109326839447,
+      "step": 915
+    },
+    {
+      "epoch": 0.1145,
+      "grad_norm": 3.194913148880005,
+      "grad_norm_var": 8.64117053500835,
+      "learning_rate": 0.0001,
+      "loss": 1.3016,
+      "loss/crossentropy": 2.25707745552063,
+      "loss/hidden": 0.95703125,
+      "loss/logits": 0.15214993059635162,
+      "loss/reg": 0.019243914633989334,
+      "step": 916
+    },
+    {
+      "epoch": 0.114625,
+      "grad_norm": 2.335845708847046,
+      "grad_norm_var": 8.88876728908843,
+      "learning_rate": 0.0001,
+      "loss": 0.9484,
+      "loss/crossentropy": 2.6386678218841553,
+      "loss/hidden": 0.63671875,
+      "loss/logits": 0.11931365728378296,
+      "loss/reg": 0.019234785810112953,
+      "step": 917
+    },
+    {
+      "epoch": 0.11475,
+      "grad_norm": 3.6922056674957275,
+      "grad_norm_var": 8.898252742921356,
+      "learning_rate": 0.0001,
+      "loss": 1.0667,
+      "loss/crossentropy": 2.5539944171905518,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.14395025372505188,
+      "loss/reg": 0.019226964563131332,
+      "step": 918
+    },
+    {
+      "epoch": 0.114875,
+      "grad_norm": 2.946389675140381,
+      "grad_norm_var": 8.982934878513694,
+      "learning_rate": 0.0001,
+      "loss": 1.0911,
+      "loss/crossentropy": 2.3651351928710938,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.1371677815914154,
+      "loss/reg": 0.019219111651182175,
+      "step": 919
+    },
+    {
+      "epoch": 0.115,
+      "grad_norm": 3.050600051879883,
+      "grad_norm_var": 9.068373446668678,
+      "learning_rate": 0.0001,
+      "loss": 1.0419,
+      "loss/crossentropy": 2.3772196769714355,
+      "loss/hidden": 0.734375,
+      "loss/logits": 0.11545050889253616,
+      "loss/reg": 0.019211286678910255,
+      "step": 920
+    },
+    {
+      "epoch": 0.115125,
+      "grad_norm": 2.954526424407959,
+      "grad_norm_var": 9.105915478669973,
+      "learning_rate": 0.0001,
+      "loss": 1.0807,
+      "loss/crossentropy": 2.2472290992736816,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.13482055068016052,
+      "loss/reg": 0.019202249124646187,
+      "step": 921
+    },
+    {
+      "epoch": 0.11525,
+      "grad_norm": 6.214420795440674,
+      "grad_norm_var": 9.27670245999236,
+      "learning_rate": 0.0001,
+      "loss": 1.8562,
+      "loss/crossentropy": 2.9924590587615967,
+      "loss/hidden": 1.2734375,
+      "loss/logits": 0.3907894492149353,
+      "loss/reg": 0.019193273037672043,
+      "step": 922
+    },
+    {
+      "epoch": 0.115375,
+      "grad_norm": 4.534095287322998,
+      "grad_norm_var": 8.536934613221737,
+      "learning_rate": 0.0001,
+      "loss": 1.4885,
+      "loss/crossentropy": 2.394090414047241,
+      "loss/hidden": 1.078125,
+      "loss/logits": 0.2185368537902832,
+      "loss/reg": 0.019184142351150513,
+      "step": 923
+    },
+    {
+      "epoch": 0.1155,
+      "grad_norm": 3.6761627197265625,
+      "grad_norm_var": 8.505579278095963,
+      "learning_rate": 0.0001,
+      "loss": 1.1306,
+      "loss/crossentropy": 2.3566806316375732,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.14977289736270905,
+      "loss/reg": 0.019175738096237183,
+      "step": 924
+    },
+    {
+      "epoch": 0.115625,
+      "grad_norm": 4.592898368835449,
+      "grad_norm_var": 8.481328607270026,
+      "learning_rate": 0.0001,
+      "loss": 1.2942,
+      "loss/crossentropy": 2.7458250522613525,
+      "loss/hidden": 0.953125,
+      "loss/logits": 0.14940449595451355,
+      "loss/reg": 0.01916695386171341,
+      "step": 925
+    },
+    {
+      "epoch": 0.11575,
+      "grad_norm": 2.7483372688293457,
+      "grad_norm_var": 8.533618684340597,
+      "learning_rate": 0.0001,
+      "loss": 1.0727,
+      "loss/crossentropy": 2.4450652599334717,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.12724441289901733,
+      "loss/reg": 0.019158538430929184,
+      "step": 926
+    },
+    {
+      "epoch": 0.115875,
+      "grad_norm": 3.075195074081421,
+      "grad_norm_var": 8.541996814909611,
+      "learning_rate": 0.0001,
+      "loss": 1.1251,
+      "loss/crossentropy": 2.3960814476013184,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.1718631386756897,
+      "loss/reg": 0.019150495529174805,
+      "step": 927
+    },
+    {
+      "epoch": 0.116,
+      "grad_norm": 3.313359022140503,
+      "grad_norm_var": 8.521887542243068,
+      "learning_rate": 0.0001,
+      "loss": 1.1316,
+      "loss/crossentropy": 2.3464736938476562,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.1276848018169403,
+      "loss/reg": 0.019141457974910736,
+      "step": 928
+    },
+    {
+      "epoch": 0.116125,
+      "grad_norm": 4.252639293670654,
+      "grad_norm_var": 0.9000980544993648,
+      "learning_rate": 0.0001,
+      "loss": 1.1818,
+      "loss/crossentropy": 2.523444414138794,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.15845248103141785,
+      "loss/reg": 0.01913331262767315,
+      "step": 929
+    },
+    {
+      "epoch": 0.11625,
+      "grad_norm": 3.097456455230713,
+      "grad_norm_var": 0.9123223599265882,
+      "learning_rate": 0.0001,
+      "loss": 1.1264,
+      "loss/crossentropy": 2.595120668411255,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.16165900230407715,
+      "loss/reg": 0.019125619903206825,
+      "step": 930
+    },
+    {
+      "epoch": 0.116375,
+      "grad_norm": 3.292982816696167,
+      "grad_norm_var": 0.8964505136113113,
+      "learning_rate": 0.0001,
+      "loss": 1.1132,
+      "loss/crossentropy": 2.4093523025512695,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.14468303322792053,
+      "loss/reg": 0.019116582348942757,
+      "step": 931
+    },
+    {
+      "epoch": 0.1165,
+      "grad_norm": 2.559980869293213,
+      "grad_norm_var": 0.952617731514821,
+      "learning_rate": 0.0001,
+      "loss": 0.9697,
+      "loss/crossentropy": 2.5491104125976562,
+      "loss/hidden": 0.6484375,
+      "loss/logits": 0.13018551468849182,
+      "loss/reg": 0.019108334556221962,
+      "step": 932
+    },
+    {
+      "epoch": 0.116625,
+      "grad_norm": 3.058579683303833,
+      "grad_norm_var": 0.871050822267735,
+      "learning_rate": 0.0001,
+      "loss": 1.2919,
+      "loss/crossentropy": 2.303837537765503,
+      "loss/hidden": 0.9296875,
+      "loss/logits": 0.17123734951019287,
+      "loss/reg": 0.019099365919828415,
+      "step": 933
+    },
+    {
+      "epoch": 0.11675,
+      "grad_norm": 4.866446495056152,
+      "grad_norm_var": 0.9769503909617764,
+      "learning_rate": 0.0001,
+      "loss": 1.1876,
+      "loss/crossentropy": 2.364741325378418,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.1607905626296997,
+      "loss/reg": 0.019090238958597183,
+      "step": 934
+    },
+    {
+      "epoch": 0.116875,
+      "grad_norm": 5.912527561187744,
+      "grad_norm_var": 1.252657817578581,
+      "learning_rate": 0.0001,
+      "loss": 1.4831,
+      "loss/crossentropy": 2.641162633895874,
+      "loss/hidden": 1.09375,
+      "loss/logits": 0.19855040311813354,
+      "loss/reg": 0.019081177189946175,
+      "step": 935
+    },
+    {
+      "epoch": 0.117,
+      "grad_norm": 2.8301663398742676,
+      "grad_norm_var": 1.2784556528629765,
+      "learning_rate": 0.0001,
+      "loss": 1.0878,
+      "loss/crossentropy": 2.277157783508301,
+      "loss/hidden": 0.765625,
+      "loss/logits": 0.13142800331115723,
+      "loss/reg": 0.019072722643613815,
+      "step": 936
+    },
+    {
+      "epoch": 0.117125,
+      "grad_norm": 2.9044833183288574,
+      "grad_norm_var": 1.2843284928455583,
+      "learning_rate": 0.0001,
+      "loss": 1.1381,
+      "loss/crossentropy": 2.3311665058135986,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.1506001055240631,
+      "loss/reg": 0.01906409114599228,
+      "step": 937
+    },
+    {
+      "epoch": 0.11725,
+      "grad_norm": 3.78202223777771,
+      "grad_norm_var": 0.8736988295376342,
+      "learning_rate": 0.0001,
+      "loss": 1.2869,
+      "loss/crossentropy": 2.4287989139556885,
+      "loss/hidden": 0.8984375,
+      "loss/logits": 0.197871595621109,
+      "loss/reg": 0.019055521115660667,
+      "step": 938
+    },
+    {
+      "epoch": 0.117375,
+      "grad_norm": 3.1086387634277344,
+      "grad_norm_var": 0.8338185014655397,
+      "learning_rate": 0.0001,
+      "loss": 1.0576,
+      "loss/crossentropy": 2.335935354232788,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.1288556456565857,
+      "loss/reg": 0.019047552719712257,
+      "step": 939
+    },
+    {
+      "epoch": 0.1175,
+      "grad_norm": 4.869723320007324,
+      "grad_norm_var": 0.9402287231159085,
+      "learning_rate": 0.0001,
+      "loss": 1.0378,
+      "loss/crossentropy": 2.9672138690948486,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.13644808530807495,
+      "loss/reg": 0.01903851516544819,
+      "step": 940
+    },
+    {
+      "epoch": 0.117625,
+      "grad_norm": 2.8750882148742676,
+      "grad_norm_var": 0.9067692046415797,
+      "learning_rate": 0.0001,
+      "loss": 1.0037,
+      "loss/crossentropy": 2.3138959407806396,
+      "loss/hidden": 0.69140625,
+      "loss/logits": 0.12202918529510498,
+      "loss/reg": 0.019029438495635986,
+      "step": 941
+    },
+    {
+      "epoch": 0.11775,
+      "grad_norm": 3.044121742248535,
+      "grad_norm_var": 0.8812433820018912,
+      "learning_rate": 0.0001,
+      "loss": 1.0173,
+      "loss/crossentropy": 2.4543912410736084,
+      "loss/hidden": 0.70703125,
+      "loss/logits": 0.12005805224180222,
+      "loss/reg": 0.01902030035853386,
+      "step": 942
+    },
+    {
+      "epoch": 0.117875,
+      "grad_norm": 2.945160150527954,
+      "grad_norm_var": 0.8905794039935603,
+      "learning_rate": 0.0001,
+      "loss": 1.1688,
+      "loss/crossentropy": 2.3482954502105713,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.15057498216629028,
+      "loss/reg": 0.019011201336979866,
+      "step": 943
+    },
+    {
+      "epoch": 0.118,
+      "grad_norm": 3.0109965801239014,
+      "grad_norm_var": 0.9056152589294107,
+      "learning_rate": 0.0001,
+      "loss": 1.3721,
+      "loss/crossentropy": 2.353516101837158,
+      "loss/hidden": 0.984375,
+      "loss/logits": 0.1977054923772812,
+      "loss/reg": 0.0190016757696867,
+      "step": 944
+    },
+    {
+      "epoch": 0.118125,
+      "grad_norm": 3.424039363861084,
+      "grad_norm_var": 0.8682128423744849,
+      "learning_rate": 0.0001,
+      "loss": 1.1647,
+      "loss/crossentropy": 2.513850450515747,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.15838034451007843,
+      "loss/reg": 0.018992552533745766,
+      "step": 945
+    },
+    {
+      "epoch": 0.11825,
+      "grad_norm": 2.7656776905059814,
+      "grad_norm_var": 0.8917454992029661,
+      "learning_rate": 0.0001,
+      "loss": 1.0819,
+      "loss/crossentropy": 2.650881767272949,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.1420745849609375,
+      "loss/reg": 0.01898341253399849,
+      "step": 946
+    },
+    {
+      "epoch": 0.118375,
+      "grad_norm": 4.2130560874938965,
+      "grad_norm_var": 0.9250033835133629,
+      "learning_rate": 0.0001,
+      "loss": 1.2469,
+      "loss/crossentropy": 2.683312177658081,
+      "loss/hidden": 0.90625,
+      "loss/logits": 0.15088841319084167,
+      "loss/reg": 0.018973875790834427,
+      "step": 947
+    },
+    {
+      "epoch": 0.1185,
+      "grad_norm": 3.834226608276367,
+      "grad_norm_var": 0.8649633510209883,
+      "learning_rate": 0.0001,
+      "loss": 0.9767,
+      "loss/crossentropy": 2.5706145763397217,
+      "loss/hidden": 0.6640625,
+      "loss/logits": 0.12295837700366974,
+      "loss/reg": 0.018963845446705818,
+      "step": 948
+    },
+    {
+      "epoch": 0.118625,
+      "grad_norm": 3.3009235858917236,
+      "grad_norm_var": 0.8514524765901378,
+      "learning_rate": 0.0001,
+      "loss": 1.1857,
+      "loss/crossentropy": 2.6797893047332764,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.17191748321056366,
+      "loss/reg": 0.018954817205667496,
+      "step": 949
+    },
+    {
+      "epoch": 0.11875,
+      "grad_norm": 6.452317237854004,
+      "grad_norm_var": 1.2752747995844325,
+      "learning_rate": 0.0001,
+      "loss": 1.4528,
+      "loss/crossentropy": 2.7505640983581543,
+      "loss/hidden": 1.0546875,
+      "loss/logits": 0.20869939029216766,
+      "loss/reg": 0.018945740535855293,
+      "step": 950
+    },
+    {
+      "epoch": 0.118875,
+      "grad_norm": 3.405714273452759,
+      "grad_norm_var": 0.9300412257064863,
+      "learning_rate": 0.0001,
+      "loss": 1.0922,
+      "loss/crossentropy": 2.802401065826416,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.14506830275058746,
+      "loss/reg": 0.018936749547719955,
+      "step": 951
+    },
+    {
+      "epoch": 0.119,
+      "grad_norm": 9.662891387939453,
+      "grad_norm_var": 3.194050081601077,
+      "learning_rate": 0.0001,
+      "loss": 1.5003,
+      "loss/crossentropy": 2.700052499771118,
+      "loss/hidden": 1.1171875,
+      "loss/logits": 0.19381779432296753,
+      "loss/reg": 0.01892753876745701,
+      "step": 952
+    },
+    {
+      "epoch": 0.119125,
+      "grad_norm": 2.8670711517333984,
+      "grad_norm_var": 3.199477320796336,
+      "learning_rate": 0.0001,
+      "loss": 1.0153,
+      "loss/crossentropy": 2.490739107131958,
+      "loss/hidden": 0.69921875,
+      "loss/logits": 0.12687504291534424,
+      "loss/reg": 0.01891852729022503,
+      "step": 953
+    },
+    {
+      "epoch": 0.11925,
+      "grad_norm": 4.066836357116699,
+      "grad_norm_var": 3.1973098694543274,
+      "learning_rate": 0.0001,
+      "loss": 1.2704,
+      "loss/crossentropy": 2.4122891426086426,
+      "loss/hidden": 0.9140625,
+      "loss/logits": 0.1672634333372116,
+      "loss/reg": 0.018909232690930367,
+      "step": 954
+    },
+    {
+      "epoch": 0.119375,
+      "grad_norm": 2.8044497966766357,
+      "grad_norm_var": 3.2388562001879793,
+      "learning_rate": 0.0001,
+      "loss": 1.0385,
+      "loss/crossentropy": 2.473945379257202,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.13464638590812683,
+      "loss/reg": 0.018900100141763687,
+      "step": 955
+    },
+    {
+      "epoch": 0.1195,
+      "grad_norm": 3.7214255332946777,
+      "grad_norm_var": 3.1837278954586044,
+      "learning_rate": 0.0001,
+      "loss": 1.0832,
+      "loss/crossentropy": 2.6148641109466553,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.1403505802154541,
+      "loss/reg": 0.01889113523066044,
+      "step": 956
+    },
+    {
+      "epoch": 0.119625,
+      "grad_norm": 3.0579674243927,
+      "grad_norm_var": 3.160836005262268,
+      "learning_rate": 0.0001,
+      "loss": 1.1761,
+      "loss/crossentropy": 2.5016398429870605,
+      "loss/hidden": 0.83984375,
+      "loss/logits": 0.1474056839942932,
+      "loss/reg": 0.0188821442425251,
+      "step": 957
+    },
+    {
+      "epoch": 0.11975,
+      "grad_norm": 2.429112672805786,
+      "grad_norm_var": 3.255565314691306,
+      "learning_rate": 0.0001,
+      "loss": 0.9853,
+      "loss/crossentropy": 2.3262507915496826,
+      "loss/hidden": 0.66796875,
+      "loss/logits": 0.1286056935787201,
+      "loss/reg": 0.018873048946261406,
+      "step": 958
+    },
+    {
+      "epoch": 0.119875,
+      "grad_norm": 2.9561123847961426,
+      "grad_norm_var": 3.2542184489207098,
+      "learning_rate": 0.0001,
+      "loss": 0.9739,
+      "loss/crossentropy": 2.5669143199920654,
+      "loss/hidden": 0.65234375,
+      "loss/logits": 0.13291960954666138,
+      "loss/reg": 0.018863873556256294,
+      "step": 959
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.1904752254486084,
+      "grad_norm_var": 3.2355963683487268,
+      "learning_rate": 0.0001,
+      "loss": 1.074,
+      "loss/crossentropy": 2.5522408485412598,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.1315881609916687,
+      "loss/reg": 0.018855126574635506,
+      "step": 960
+    },
+    {
+      "epoch": 0.120125,
+      "grad_norm": 7.435352802276611,
+      "grad_norm_var": 3.9949775747956586,
+      "learning_rate": 0.0001,
+      "loss": 1.6007,
+      "loss/crossentropy": 2.7422168254852295,
+      "loss/hidden": 1.1875,
+      "loss/logits": 0.2246999442577362,
+      "loss/reg": 0.018846556544303894,
+      "step": 961
+    },
+    {
+      "epoch": 0.12025,
+      "grad_norm": 4.614249229431152,
+      "grad_norm_var": 3.8709926395951384,
+      "learning_rate": 0.0001,
+      "loss": 1.3908,
+      "loss/crossentropy": 2.356748104095459,
+      "loss/hidden": 0.953125,
+      "loss/logits": 0.24928607046604156,
+      "loss/reg": 0.018837420269846916,
+      "step": 962
+    },
+    {
+      "epoch": 0.120375,
+      "grad_norm": 4.978577136993408,
+      "grad_norm_var": 3.903770487124885,
+      "learning_rate": 0.0001,
+      "loss": 1.4333,
+      "loss/crossentropy": 2.2456307411193848,
+      "loss/hidden": 1.0703125,
+      "loss/logits": 0.17467612028121948,
+      "loss/reg": 0.01882883533835411,
+      "step": 963
+    },
+    {
+      "epoch": 0.1205,
+      "grad_norm": 5.51161003112793,
+      "grad_norm_var": 3.97576236618078,
+      "learning_rate": 0.0001,
+      "loss": 1.2552,
+      "loss/crossentropy": 2.7595887184143066,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.18415382504463196,
+      "loss/reg": 0.01882052607834339,
+      "step": 964
+    },
+    {
+      "epoch": 0.120625,
+      "grad_norm": 3.2307283878326416,
+      "grad_norm_var": 3.9863892013288393,
+      "learning_rate": 0.0001,
+      "loss": 1.2728,
+      "loss/crossentropy": 2.48502779006958,
+      "loss/hidden": 0.88671875,
+      "loss/logits": 0.19795754551887512,
+      "loss/reg": 0.01881156861782074,
+      "step": 965
+    },
+    {
+      "epoch": 0.12075,
+      "grad_norm": 3.441767454147339,
+      "grad_norm_var": 3.7286595116638916,
+      "learning_rate": 0.0001,
+      "loss": 1.182,
+      "loss/crossentropy": 2.508680582046509,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.15802894532680511,
+      "loss/reg": 0.018803071230649948,
+      "step": 966
+    },
+    {
+      "epoch": 0.120875,
+      "grad_norm": 4.063246726989746,
+      "grad_norm_var": 3.685090208705704,
+      "learning_rate": 0.0001,
+      "loss": 1.3354,
+      "loss/crossentropy": 2.47729229927063,
+      "loss/hidden": 0.984375,
+      "loss/logits": 0.16310644149780273,
+      "loss/reg": 0.018794314935803413,
+      "step": 967
+    },
+    {
+      "epoch": 0.121,
+      "grad_norm": 5.8381171226501465,
+      "grad_norm_var": 1.8400005684538645,
+      "learning_rate": 0.0001,
+      "loss": 1.1025,
+      "loss/crossentropy": 2.5666096210479736,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.1373094618320465,
+      "loss/reg": 0.018785255029797554,
+      "step": 968
+    },
+    {
+      "epoch": 0.121125,
+      "grad_norm": 2.7132527828216553,
+      "grad_norm_var": 1.8649801572693356,
+      "learning_rate": 0.0001,
+      "loss": 1.0912,
+      "loss/crossentropy": 2.4335010051727295,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.13391214609146118,
+      "loss/reg": 0.018776265904307365,
+      "step": 969
+    },
+    {
+      "epoch": 0.12125,
+      "grad_norm": 2.9154088497161865,
+      "grad_norm_var": 1.938092020210782,
+      "learning_rate": 0.0001,
+      "loss": 1.1314,
+      "loss/crossentropy": 2.523137331008911,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.1429774910211563,
+      "loss/reg": 0.01876768097281456,
+      "step": 970
+    },
+    {
+      "epoch": 0.121375,
+      "grad_norm": 3.3754584789276123,
+      "grad_norm_var": 1.8726730225127388,
+      "learning_rate": 0.0001,
+      "loss": 1.0747,
+      "loss/crossentropy": 2.6033432483673096,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.14106187224388123,
+      "loss/reg": 0.018759164959192276,
+      "step": 971
+    },
+    {
+      "epoch": 0.1215,
+      "grad_norm": 2.7394940853118896,
+      "grad_norm_var": 1.9650935524715956,
+      "learning_rate": 0.0001,
+      "loss": 1.0871,
+      "loss/crossentropy": 2.2976291179656982,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.1300249695777893,
+      "loss/reg": 0.01874978095293045,
+      "step": 972
+    },
+    {
+      "epoch": 0.121625,
+      "grad_norm": 2.9742624759674072,
+      "grad_norm_var": 1.9749925269591906,
+      "learning_rate": 0.0001,
+      "loss": 1.1221,
+      "loss/crossentropy": 2.7049221992492676,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.1416921317577362,
+      "loss/reg": 0.018740687519311905,
+      "step": 973
+    },
+    {
+      "epoch": 0.12175,
+      "grad_norm": 2.927666187286377,
+      "grad_norm_var": 1.892721758937742,
+      "learning_rate": 0.0001,
+      "loss": 1.0252,
+      "loss/crossentropy": 2.6492226123809814,
+      "loss/hidden": 0.71875,
+      "loss/logits": 0.11909263581037521,
+      "loss/reg": 0.018731672316789627,
+      "step": 974
+    },
+    {
+      "epoch": 0.121875,
+      "grad_norm": 16.453693389892578,
+      "grad_norm_var": 11.523681815422924,
+      "learning_rate": 0.0001,
+      "loss": 1.5788,
+      "loss/crossentropy": 2.375779867172241,
+      "loss/hidden": 1.171875,
+      "loss/logits": 0.2197396457195282,
+      "loss/reg": 0.01872306317090988,
+      "step": 975
+    },
+    {
+      "epoch": 0.122,
+      "grad_norm": 2.928443193435669,
+      "grad_norm_var": 11.583339951760102,
+      "learning_rate": 0.0001,
+      "loss": 1.0995,
+      "loss/crossentropy": 2.8136162757873535,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.13505280017852783,
+      "loss/reg": 0.018714020028710365,
+      "step": 976
+    },
+    {
+      "epoch": 0.122125,
+      "grad_norm": 4.540535926818848,
+      "grad_norm_var": 11.07401646408874,
+      "learning_rate": 0.0001,
+      "loss": 1.1806,
+      "loss/crossentropy": 2.5312135219573975,
+      "loss/hidden": 0.83984375,
+      "loss/logits": 0.15369677543640137,
+      "loss/reg": 0.01870504766702652,
+      "step": 977
+    },
+    {
+      "epoch": 0.12225,
+      "grad_norm": 2.4433298110961914,
+      "grad_norm_var": 11.358052675820652,
+      "learning_rate": 0.0001,
+      "loss": 1.0167,
+      "loss/crossentropy": 2.578800678253174,
+      "loss/hidden": 0.6953125,
+      "loss/logits": 0.13442841172218323,
+      "loss/reg": 0.018696293234825134,
+      "step": 978
+    },
+    {
+      "epoch": 0.122375,
+      "grad_norm": 3.341182231903076,
+      "grad_norm_var": 11.408522912728658,
+      "learning_rate": 0.0001,
+      "loss": 1.1346,
+      "loss/crossentropy": 2.4482579231262207,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.14306378364562988,
+      "loss/reg": 0.018688105046749115,
+      "step": 979
+    },
+    {
+      "epoch": 0.1225,
+      "grad_norm": 2.433337450027466,
+      "grad_norm_var": 11.51984045745032,
+      "learning_rate": 0.0001,
+      "loss": 0.9764,
+      "loss/crossentropy": 2.321781873703003,
+      "loss/hidden": 0.65625,
+      "loss/logits": 0.13331879675388336,
+      "loss/reg": 0.018679112195968628,
+      "step": 980
+    },
+    {
+      "epoch": 0.122625,
+      "grad_norm": 3.6000678539276123,
+      "grad_norm_var": 11.483219758864427,
+      "learning_rate": 0.0001,
+      "loss": 1.1731,
+      "loss/crossentropy": 2.5233397483825684,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.16610421240329742,
+      "loss/reg": 0.018670594319701195,
+      "step": 981
+    },
+    {
+      "epoch": 0.12275,
+      "grad_norm": 2.8748631477355957,
+      "grad_norm_var": 11.558394893606714,
+      "learning_rate": 0.0001,
+      "loss": 1.0362,
+      "loss/crossentropy": 2.4155187606811523,
+      "loss/hidden": 0.7265625,
+      "loss/logits": 0.12304510176181793,
+      "loss/reg": 0.018661517649888992,
+      "step": 982
+    },
+    {
+      "epoch": 0.122875,
+      "grad_norm": 2.2293035984039307,
+      "grad_norm_var": 11.786185692154314,
+      "learning_rate": 0.0001,
+      "loss": 1.0077,
+      "loss/crossentropy": 2.618363618850708,
+      "loss/hidden": 0.6875,
+      "loss/logits": 0.13366106152534485,
+      "loss/reg": 0.01865258812904358,
+      "step": 983
+    },
+    {
+      "epoch": 0.123,
+      "grad_norm": 3.615424633026123,
+      "grad_norm_var": 11.556298836968558,
+      "learning_rate": 0.0001,
+      "loss": 1.0544,
+      "loss/crossentropy": 2.224966526031494,
+      "loss/hidden": 0.734375,
+      "loss/logits": 0.1335442066192627,
+      "loss/reg": 0.01864360086619854,
+      "step": 984
+    },
+    {
+      "epoch": 0.123125,
+      "grad_norm": 2.93835186958313,
+      "grad_norm_var": 11.524399601900045,
+      "learning_rate": 0.0001,
+      "loss": 1.232,
+      "loss/crossentropy": 2.348299026489258,
+      "loss/hidden": 0.87890625,
+      "loss/logits": 0.1667439341545105,
+      "loss/reg": 0.018634630367159843,
+      "step": 985
+    },
+    {
+      "epoch": 0.12325,
+      "grad_norm": 3.2750635147094727,
+      "grad_norm_var": 11.485476360611186,
+      "learning_rate": 0.0001,
+      "loss": 1.0639,
+      "loss/crossentropy": 2.5712990760803223,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.12762659788131714,
+      "loss/reg": 0.018625380471348763,
+      "step": 986
+    },
+    {
+      "epoch": 0.123375,
+      "grad_norm": 3.81864595413208,
+      "grad_norm_var": 11.465683474564775,
+      "learning_rate": 0.0001,
+      "loss": 1.1284,
+      "loss/crossentropy": 2.4274511337280273,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.15321871638298035,
+      "loss/reg": 0.01861615665256977,
+      "step": 987
+    },
+    {
+      "epoch": 0.1235,
+      "grad_norm": 3.2677698135375977,
+      "grad_norm_var": 11.398153583229371,
+      "learning_rate": 0.0001,
+      "loss": 1.1315,
+      "loss/crossentropy": 2.656604290008545,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.14076048135757446,
+      "loss/reg": 0.018606893718242645,
+      "step": 988
+    },
+    {
+      "epoch": 0.123625,
+      "grad_norm": 3.561713457107544,
+      "grad_norm_var": 11.341034456038914,
+      "learning_rate": 0.0001,
+      "loss": 1.0152,
+      "loss/crossentropy": 2.6205251216888428,
+      "loss/hidden": 0.70703125,
+      "loss/logits": 0.12218683958053589,
+      "loss/reg": 0.01859763078391552,
+      "step": 989
+    },
+    {
+      "epoch": 0.12375,
+      "grad_norm": 2.146240472793579,
+      "grad_norm_var": 11.49254916357394,
+      "learning_rate": 0.0001,
+      "loss": 1.0449,
+      "loss/crossentropy": 2.4804821014404297,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.14422178268432617,
+      "loss/reg": 0.018588390201330185,
+      "step": 990
+    },
+    {
+      "epoch": 0.123875,
+      "grad_norm": 2.6142220497131348,
+      "grad_norm_var": 0.4215380256098586,
+      "learning_rate": 0.0001,
+      "loss": 1.0371,
+      "loss/crossentropy": 2.3303298950195312,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.11297546327114105,
+      "loss/reg": 0.018579507246613503,
+      "step": 991
+    },
+    {
+      "epoch": 0.124,
+      "grad_norm": 4.17028284072876,
+      "grad_norm_var": 0.4892223582938262,
+      "learning_rate": 0.0001,
+      "loss": 1.6898,
+      "loss/crossentropy": 2.250098943710327,
+      "loss/hidden": 1.203125,
+      "loss/logits": 0.30099910497665405,
+      "loss/reg": 0.018569782376289368,
+      "step": 992
+    },
+    {
+      "epoch": 0.124125,
+      "grad_norm": 2.3814339637756348,
+      "grad_norm_var": 0.3887345955884323,
+      "learning_rate": 0.0001,
+      "loss": 1.0837,
+      "loss/crossentropy": 2.6447057723999023,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.15198630094528198,
+      "loss/reg": 0.01855996623635292,
+      "step": 993
+    },
+    {
+      "epoch": 0.12425,
+      "grad_norm": 4.670334815979004,
+      "grad_norm_var": 0.5202129226035586,
+      "learning_rate": 0.0001,
+      "loss": 1.1908,
+      "loss/crossentropy": 2.3941245079040527,
+      "loss/hidden": 0.85546875,
+      "loss/logits": 0.1498216688632965,
+      "loss/reg": 0.018550006672739983,
+      "step": 994
+    },
+    {
+      "epoch": 0.124375,
+      "grad_norm": 2.498332977294922,
+      "grad_norm_var": 0.5469080049785059,
+      "learning_rate": 0.0001,
+      "loss": 0.9226,
+      "loss/crossentropy": 2.489830255508423,
+      "loss/hidden": 0.63671875,
+      "loss/logits": 0.10051175206899643,
+      "loss/reg": 0.01854090392589569,
+      "step": 995
+    },
+    {
+      "epoch": 0.1245,
+      "grad_norm": 2.987192392349243,
+      "grad_norm_var": 0.5145625202891589,
+      "learning_rate": 0.0001,
+      "loss": 1.1638,
+      "loss/crossentropy": 2.094637632369995,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.12690997123718262,
+      "loss/reg": 0.018531804904341698,
+      "step": 996
+    },
+    {
+      "epoch": 0.124625,
+      "grad_norm": 2.738851547241211,
+      "grad_norm_var": 0.5110263660772335,
+      "learning_rate": 0.0001,
+      "loss": 1.1959,
+      "loss/crossentropy": 2.1406211853027344,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.16688010096549988,
+      "loss/reg": 0.018522722646594048,
+      "step": 997
+    },
+    {
+      "epoch": 0.12475,
+      "grad_norm": 5.02874231338501,
+      "grad_norm_var": 0.732945509426732,
+      "learning_rate": 0.0001,
+      "loss": 1.089,
+      "loss/crossentropy": 2.442840337753296,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.142162024974823,
+      "loss/reg": 0.018513953313231468,
+      "step": 998
+    },
+    {
+      "epoch": 0.124875,
+      "grad_norm": 2.947786331176758,
+      "grad_norm_var": 0.6677765621166297,
+      "learning_rate": 0.0001,
+      "loss": 1.3459,
+      "loss/crossentropy": 2.483854055404663,
+      "loss/hidden": 0.96875,
+      "loss/logits": 0.19213837385177612,
+      "loss/reg": 0.018505612388253212,
+      "step": 999
+    },
+    {
+      "epoch": 0.125,
+      "grad_norm": 4.166240692138672,
+      "grad_norm_var": 0.7105452516630361,
+      "learning_rate": 0.0001,
+      "loss": 1.2402,
+      "loss/crossentropy": 2.454993963241577,
+      "loss/hidden": 0.87890625,
+      "loss/logits": 0.17627671360969543,
+      "loss/reg": 0.018497284501791,
+      "step": 1000
+    },
+    {
+      "epoch": 0.125125,
+      "grad_norm": 2.7403693199157715,
+      "grad_norm_var": 0.723220167440765,
+      "learning_rate": 0.0001,
+      "loss": 1.0711,
+      "loss/crossentropy": 2.3623507022857666,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.1362101435661316,
+      "loss/reg": 0.018489044159650803,
+      "step": 1001
+    },
+    {
+      "epoch": 0.12525,
+      "grad_norm": 4.017945766448975,
+      "grad_norm_var": 0.7539223188067105,
+      "learning_rate": 0.0001,
+      "loss": 1.2595,
+      "loss/crossentropy": 2.199939727783203,
+      "loss/hidden": 0.93359375,
+      "loss/logits": 0.14108574390411377,
+      "loss/reg": 0.018480053171515465,
+      "step": 1002
+    },
+    {
+      "epoch": 0.125375,
+      "grad_norm": 3.8252949714660645,
+      "grad_norm_var": 0.7543319037149537,
+      "learning_rate": 0.0001,
+      "loss": 1.2766,
+      "loss/crossentropy": 2.4151408672332764,
+      "loss/hidden": 0.91796875,
+      "loss/logits": 0.1738898754119873,
+      "loss/reg": 0.018471699208021164,
+      "step": 1003
+    },
+    {
+      "epoch": 0.1255,
+      "grad_norm": 3.2934703826904297,
+      "grad_norm_var": 0.754056547294514,
+      "learning_rate": 0.0001,
+      "loss": 1.1773,
+      "loss/crossentropy": 2.4902610778808594,
+      "loss/hidden": 0.83984375,
+      "loss/logits": 0.1528603434562683,
+      "loss/reg": 0.018463551998138428,
+      "step": 1004
+    },
+    {
+      "epoch": 0.125625,
+      "grad_norm": 2.8537371158599854,
+      "grad_norm_var": 0.7665102142099857,
+      "learning_rate": 0.0001,
+      "loss": 1.0491,
+      "loss/crossentropy": 2.5475528240203857,
+      "loss/hidden": 0.734375,
+      "loss/logits": 0.1301451176404953,
+      "loss/reg": 0.01845443621277809,
+      "step": 1005
+    },
+    {
+      "epoch": 0.12575,
+      "grad_norm": 10.271280288696289,
+      "grad_norm_var": 3.623624147504192,
+      "learning_rate": 0.0001,
+      "loss": 1.9301,
+      "loss/crossentropy": 2.789032459259033,
+      "loss/hidden": 1.359375,
+      "loss/logits": 0.386294960975647,
+      "loss/reg": 0.018445348367094994,
+      "step": 1006
+    },
+    {
+      "epoch": 0.125875,
+      "grad_norm": 3.5450735092163086,
+      "grad_norm_var": 3.5274627001684156,
+      "learning_rate": 0.0001,
+      "loss": 1.0646,
+      "loss/crossentropy": 2.7634024620056152,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.1419609636068344,
+      "loss/reg": 0.01843627728521824,
+      "step": 1007
+    },
+    {
+      "epoch": 0.126,
+      "grad_norm": 3.8805432319641113,
+      "grad_norm_var": 3.521631426981449,
+      "learning_rate": 0.0001,
+      "loss": 1.2255,
+      "loss/crossentropy": 2.313877820968628,
+      "loss/hidden": 0.859375,
+      "loss/logits": 0.18188925087451935,
+      "loss/reg": 0.018427575007081032,
+      "step": 1008
+    },
+    {
+      "epoch": 0.126125,
+      "grad_norm": 3.56695556640625,
+      "grad_norm_var": 3.3749006612486068,
+      "learning_rate": 0.0001,
+      "loss": 1.1632,
+      "loss/crossentropy": 3.036876678466797,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.1664760708808899,
+      "loss/reg": 0.018418410792946815,
+      "step": 1009
+    },
+    {
+      "epoch": 0.12625,
+      "grad_norm": 3.4208462238311768,
+      "grad_norm_var": 3.350722625996883,
+      "learning_rate": 0.0001,
+      "loss": 1.3539,
+      "loss/crossentropy": 2.335106372833252,
+      "loss/hidden": 1.0,
+      "loss/logits": 0.16979742050170898,
+      "loss/reg": 0.018409088253974915,
+      "step": 1010
+    },
+    {
+      "epoch": 0.126375,
+      "grad_norm": 4.440703392028809,
+      "grad_norm_var": 3.2335077439479125,
+      "learning_rate": 0.0001,
+      "loss": 1.5031,
+      "loss/crossentropy": 1.8046929836273193,
+      "loss/hidden": 1.0703125,
+      "loss/logits": 0.2487386018037796,
+      "loss/reg": 0.018400251865386963,
+      "step": 1011
+    },
+    {
+      "epoch": 0.1265,
+      "grad_norm": 3.6042182445526123,
+      "grad_norm_var": 3.175392851042456,
+      "learning_rate": 0.0001,
+      "loss": 1.0341,
+      "loss/crossentropy": 2.732485294342041,
+      "loss/hidden": 0.70703125,
+      "loss/logits": 0.14311768114566803,
+      "loss/reg": 0.018391618505120277,
+      "step": 1012
+    },
+    {
+      "epoch": 0.126625,
+      "grad_norm": 3.0008747577667236,
+      "grad_norm_var": 3.1348769442621385,
+      "learning_rate": 0.0001,
+      "loss": 1.2818,
+      "loss/crossentropy": 2.4243671894073486,
+      "loss/hidden": 0.8984375,
+      "loss/logits": 0.19954946637153625,
+      "loss/reg": 0.01838279701769352,
+      "step": 1013
+    },
+    {
+      "epoch": 0.12675,
+      "grad_norm": 3.548919677734375,
+      "grad_norm_var": 3.076212765414932,
+      "learning_rate": 0.0001,
+      "loss": 1.0597,
+      "loss/crossentropy": 2.646744966506958,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.11818855255842209,
+      "loss/reg": 0.018373781815171242,
+      "step": 1014
+    },
+    {
+      "epoch": 0.126875,
+      "grad_norm": 3.4126574993133545,
+      "grad_norm_var": 3.027892721971917,
+      "learning_rate": 0.0001,
+      "loss": 1.2446,
+      "loss/crossentropy": 2.2376763820648193,
+      "loss/hidden": 0.90234375,
+      "loss/logits": 0.15857847034931183,
+      "loss/reg": 0.018364954739809036,
+      "step": 1015
+    },
+    {
+      "epoch": 0.127,
+      "grad_norm": 2.8473548889160156,
+      "grad_norm_var": 3.1028595438739974,
+      "learning_rate": 0.0001,
+      "loss": 1.171,
+      "loss/crossentropy": 2.4969823360443115,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.17886783182621002,
+      "loss/reg": 0.01835593394935131,
+      "step": 1016
+    },
+    {
+      "epoch": 0.127125,
+      "grad_norm": 3.4492313861846924,
+      "grad_norm_var": 3.0254289441294913,
+      "learning_rate": 0.0001,
+      "loss": 1.1521,
+      "loss/crossentropy": 2.6258046627044678,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.14048799872398376,
+      "loss/reg": 0.018347129225730896,
+      "step": 1017
+    },
+    {
+      "epoch": 0.12725,
+      "grad_norm": 2.7578630447387695,
+      "grad_norm_var": 3.1109318052612775,
+      "learning_rate": 0.0001,
+      "loss": 1.1909,
+      "loss/crossentropy": 2.5233936309814453,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.16377520561218262,
+      "loss/reg": 0.018338393419981003,
+      "step": 1018
+    },
+    {
+      "epoch": 0.127375,
+      "grad_norm": 4.195294380187988,
+      "grad_norm_var": 3.117902257815278,
+      "learning_rate": 0.0001,
+      "loss": 1.3224,
+      "loss/crossentropy": 1.8772588968276978,
+      "loss/hidden": 0.984375,
+      "loss/logits": 0.1546936333179474,
+      "loss/reg": 0.01832934282720089,
+      "step": 1019
+    },
+    {
+      "epoch": 0.1275,
+      "grad_norm": 3.7895658016204834,
+      "grad_norm_var": 3.0944502488485717,
+      "learning_rate": 0.0001,
+      "loss": 1.1973,
+      "loss/crossentropy": 2.3982863426208496,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.17036345601081848,
+      "loss/reg": 0.01832025870680809,
+      "step": 1020
+    },
+    {
+      "epoch": 0.127625,
+      "grad_norm": 6.021730422973633,
+      "grad_norm_var": 3.274883958363541,
+      "learning_rate": 0.0001,
+      "loss": 1.5031,
+      "loss/crossentropy": 2.428899049758911,
+      "loss/hidden": 1.109375,
+      "loss/logits": 0.21065470576286316,
+      "loss/reg": 0.01831124909222126,
+      "step": 1021
+    },
+    {
+      "epoch": 0.12775,
+      "grad_norm": 4.54056453704834,
+      "grad_norm_var": 0.6193178360798726,
+      "learning_rate": 0.0001,
+      "loss": 1.6077,
+      "loss/crossentropy": 2.3732786178588867,
+      "loss/hidden": 1.1953125,
+      "loss/logits": 0.22934843599796295,
+      "loss/reg": 0.01830223761498928,
+      "step": 1022
+    },
+    {
+      "epoch": 0.127875,
+      "grad_norm": 4.612347602844238,
+      "grad_norm_var": 0.6611490686092556,
+      "learning_rate": 0.0001,
+      "loss": 1.1245,
+      "loss/crossentropy": 2.393007516860962,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.13689909875392914,
+      "loss/reg": 0.0182929839938879,
+      "step": 1023
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 4.8808794021606445,
+      "grad_norm_var": 0.7320190710671227,
+      "learning_rate": 0.0001,
+      "loss": 1.6498,
+      "loss/crossentropy": 1.8295843601226807,
+      "loss/hidden": 1.265625,
+      "loss/logits": 0.2013380229473114,
+      "loss/reg": 0.01828295737504959,
+      "step": 1024
+    },
+    {
+      "epoch": 0.128125,
+      "grad_norm": 6.514127731323242,
+      "grad_norm_var": 1.1516245124796,
+      "learning_rate": 0.0001,
+      "loss": 1.4269,
+      "loss/crossentropy": 2.6622331142425537,
+      "loss/hidden": 1.0390625,
+      "loss/logits": 0.20514041185379028,
+      "loss/reg": 0.018273649737238884,
+      "step": 1025
+    },
+    {
+      "epoch": 0.12825,
+      "grad_norm": 3.6572396755218506,
+      "grad_norm_var": 1.1348195216000618,
+      "learning_rate": 0.0001,
+      "loss": 1.2673,
+      "loss/crossentropy": 2.410374164581299,
+      "loss/hidden": 0.9140625,
+      "loss/logits": 0.17059262096881866,
+      "loss/reg": 0.018263790756464005,
+      "step": 1026
+    },
+    {
+      "epoch": 0.128375,
+      "grad_norm": 2.8987033367156982,
+      "grad_norm_var": 1.209186568114529,
+      "learning_rate": 0.0001,
+      "loss": 1.2157,
+      "loss/crossentropy": 2.236950397491455,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.15036781132221222,
+      "loss/reg": 0.018253570422530174,
+      "step": 1027
+    },
+    {
+      "epoch": 0.1285,
+      "grad_norm": 3.276984453201294,
+      "grad_norm_var": 1.2324156239644137,
+      "learning_rate": 0.0001,
+      "loss": 1.2494,
+      "loss/crossentropy": 2.683966636657715,
+      "loss/hidden": 0.90234375,
+      "loss/logits": 0.16460160911083221,
+      "loss/reg": 0.0182446651160717,
+      "step": 1028
+    },
+    {
+      "epoch": 0.128625,
+      "grad_norm": 5.419056415557861,
+      "grad_norm_var": 1.2877520831133287,
+      "learning_rate": 0.0001,
+      "loss": 1.4861,
+      "loss/crossentropy": 2.306896209716797,
+      "loss/hidden": 1.1015625,
+      "loss/logits": 0.20220112800598145,
+      "loss/reg": 0.018234653398394585,
+      "step": 1029
+    },
+    {
+      "epoch": 0.12875,
+      "grad_norm": 3.4227371215820312,
+      "grad_norm_var": 1.298252758406007,
+      "learning_rate": 0.0001,
+      "loss": 1.1322,
+      "loss/crossentropy": 2.6958277225494385,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.16871951520442963,
+      "loss/reg": 0.018225453794002533,
+      "step": 1030
+    },
+    {
+      "epoch": 0.128875,
+      "grad_norm": 2.5001378059387207,
+      "grad_norm_var": 1.4346570797964828,
+      "learning_rate": 0.0001,
+      "loss": 1.0835,
+      "loss/crossentropy": 2.743760824203491,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.14742383360862732,
+      "loss/reg": 0.01821640320122242,
+      "step": 1031
+    },
+    {
+      "epoch": 0.129,
+      "grad_norm": 3.422490119934082,
+      "grad_norm_var": 1.3631839436174078,
+      "learning_rate": 0.0001,
+      "loss": 1.213,
+      "loss/crossentropy": 2.6142783164978027,
+      "loss/hidden": 0.8671875,
+      "loss/logits": 0.1637578308582306,
+      "loss/reg": 0.01820731721818447,
+      "step": 1032
+    },
+    {
+      "epoch": 0.129125,
+      "grad_norm": 3.200204849243164,
+      "grad_norm_var": 1.3881674273527314,
+      "learning_rate": 0.0001,
+      "loss": 1.28,
+      "loss/crossentropy": 2.4325146675109863,
+      "loss/hidden": 0.91796875,
+      "loss/logits": 0.18004098534584045,
+      "loss/reg": 0.018198398873209953,
+      "step": 1033
+    },
+    {
+      "epoch": 0.12925,
+      "grad_norm": 3.3802099227905273,
+      "grad_norm_var": 1.3035463186707592,
+      "learning_rate": 0.0001,
+      "loss": 1.1984,
+      "loss/crossentropy": 2.424464225769043,
+      "loss/hidden": 0.8671875,
+      "loss/logits": 0.14932216703891754,
+      "loss/reg": 0.018189454451203346,
+      "step": 1034
+    },
+    {
+      "epoch": 0.129375,
+      "grad_norm": 3.2020263671875,
+      "grad_norm_var": 1.3536821307745555,
+      "learning_rate": 0.0001,
+      "loss": 1.085,
+      "loss/crossentropy": 2.539475440979004,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.1258580982685089,
+      "loss/reg": 0.018180513754487038,
+      "step": 1035
+    },
+    {
+      "epoch": 0.1295,
+      "grad_norm": 3.5899553298950195,
+      "grad_norm_var": 1.3630023284114352,
+      "learning_rate": 0.0001,
+      "loss": 1.1137,
+      "loss/crossentropy": 2.277602434158325,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.15075945854187012,
+      "loss/reg": 0.018171606585383415,
+      "step": 1036
+    },
+    {
+      "epoch": 0.129625,
+      "grad_norm": 3.1815500259399414,
+      "grad_norm_var": 1.1143223174368395,
+      "learning_rate": 0.0001,
+      "loss": 1.2827,
+      "loss/crossentropy": 2.5251145362854004,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.19092242419719696,
+      "loss/reg": 0.018162967637181282,
+      "step": 1037
+    },
+    {
+      "epoch": 0.12975,
+      "grad_norm": 3.6182315349578857,
+      "grad_norm_var": 1.083329466089586,
+      "learning_rate": 0.0001,
+      "loss": 1.5649,
+      "loss/crossentropy": 2.0126724243164062,
+      "loss/hidden": 1.1796875,
+      "loss/logits": 0.20370075106620789,
+      "loss/reg": 0.018154380843043327,
+      "step": 1038
+    },
+    {
+      "epoch": 0.129875,
+      "grad_norm": 3.239778518676758,
+      "grad_norm_var": 1.0521445613054454,
+      "learning_rate": 0.0001,
+      "loss": 1.1521,
+      "loss/crossentropy": 2.286173105239868,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.1503629982471466,
+      "loss/reg": 0.01814563386142254,
+      "step": 1039
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.980926990509033,
+      "grad_norm_var": 1.2991062966869322,
+      "learning_rate": 0.0001,
+      "loss": 1.4125,
+      "loss/crossentropy": 2.5735340118408203,
+      "loss/hidden": 0.87109375,
+      "loss/logits": 0.36005836725234985,
+      "loss/reg": 0.018136821687221527,
+      "step": 1040
+    },
+    {
+      "epoch": 0.130125,
+      "grad_norm": 3.8425281047821045,
+      "grad_norm_var": 0.7718063043351868,
+      "learning_rate": 0.0001,
+      "loss": 1.3336,
+      "loss/crossentropy": 2.002030372619629,
+      "loss/hidden": 0.9921875,
+      "loss/logits": 0.16016829013824463,
+      "loss/reg": 0.01812821812927723,
+      "step": 1041
+    },
+    {
+      "epoch": 0.13025,
+      "grad_norm": 2.6303024291992188,
+      "grad_norm_var": 0.8318731912874341,
+      "learning_rate": 0.0001,
+      "loss": 1.0674,
+      "loss/crossentropy": 2.5561156272888184,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.13234438002109528,
+      "loss/reg": 0.01811978593468666,
+      "step": 1042
+    },
+    {
+      "epoch": 0.130375,
+      "grad_norm": 4.081605911254883,
+      "grad_norm_var": 0.816546710723537,
+      "learning_rate": 0.0001,
+      "loss": 1.1508,
+      "loss/crossentropy": 2.5789449214935303,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.13762518763542175,
+      "loss/reg": 0.01811092346906662,
+      "step": 1043
+    },
+    {
+      "epoch": 0.1305,
+      "grad_norm": 3.746345281600952,
+      "grad_norm_var": 0.808580216385354,
+      "learning_rate": 0.0001,
+      "loss": 1.0868,
+      "loss/crossentropy": 2.2884302139282227,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.14801663160324097,
+      "loss/reg": 0.018101971596479416,
+      "step": 1044
+    },
+    {
+      "epoch": 0.130625,
+      "grad_norm": 3.1717352867126465,
+      "grad_norm_var": 0.5952362637338294,
+      "learning_rate": 0.0001,
+      "loss": 1.0943,
+      "loss/crossentropy": 2.4526915550231934,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.1438664197921753,
+      "loss/reg": 0.018093010410666466,
+      "step": 1045
+    },
+    {
+      "epoch": 0.13075,
+      "grad_norm": 5.281366348266602,
+      "grad_norm_var": 0.7887311446436568,
+      "learning_rate": 0.0001,
+      "loss": 1.1712,
+      "loss/crossentropy": 2.400421142578125,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.1543923020362854,
+      "loss/reg": 0.01808418706059456,
+      "step": 1046
+    },
+    {
+      "epoch": 0.130875,
+      "grad_norm": 5.1911540031433105,
+      "grad_norm_var": 0.8361699826228097,
+      "learning_rate": 0.0001,
+      "loss": 1.2646,
+      "loss/crossentropy": 2.6032376289367676,
+      "loss/hidden": 0.91796875,
+      "loss/logits": 0.16584512591362,
+      "loss/reg": 0.01807507872581482,
+      "step": 1047
+    },
+    {
+      "epoch": 0.131,
+      "grad_norm": 2.902432680130005,
+      "grad_norm_var": 0.8790790548012068,
+      "learning_rate": 0.0001,
+      "loss": 1.2484,
+      "loss/crossentropy": 2.4166929721832275,
+      "loss/hidden": 0.90234375,
+      "loss/logits": 0.1653715819120407,
+      "loss/reg": 0.018065867945551872,
+      "step": 1048
+    },
+    {
+      "epoch": 0.131125,
+      "grad_norm": 3.408224582672119,
+      "grad_norm_var": 0.8661178167301301,
+      "learning_rate": 0.0001,
+      "loss": 1.1045,
+      "loss/crossentropy": 2.5960233211517334,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.13876128196716309,
+      "loss/reg": 0.01805703714489937,
+      "step": 1049
+    },
+    {
+      "epoch": 0.13125,
+      "grad_norm": 2.8764824867248535,
+      "grad_norm_var": 0.9086952536899503,
+      "learning_rate": 0.0001,
+      "loss": 1.2047,
+      "loss/crossentropy": 2.280080556869507,
+      "loss/hidden": 0.875,
+      "loss/logits": 0.1491965353488922,
+      "loss/reg": 0.018048083409667015,
+      "step": 1050
+    },
+    {
+      "epoch": 0.131375,
+      "grad_norm": 4.48457670211792,
+      "grad_norm_var": 0.9183881653108216,
+      "learning_rate": 0.0001,
+      "loss": 1.0941,
+      "loss/crossentropy": 2.5797057151794434,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.17152699828147888,
+      "loss/reg": 0.018038896843791008,
+      "step": 1051
+    },
+    {
+      "epoch": 0.1315,
+      "grad_norm": 3.204270839691162,
+      "grad_norm_var": 0.9398596856778226,
+      "learning_rate": 0.0001,
+      "loss": 1.423,
+      "loss/crossentropy": 2.1369965076446533,
+      "loss/hidden": 1.0625,
+      "loss/logits": 0.18023288249969482,
+      "loss/reg": 0.018029624596238136,
+      "step": 1052
+    },
+    {
+      "epoch": 0.131625,
+      "grad_norm": 3.225297212600708,
+      "grad_norm_var": 0.9363567728280238,
+      "learning_rate": 0.0001,
+      "loss": 1.1299,
+      "loss/crossentropy": 2.5327794551849365,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.1567188799381256,
+      "loss/reg": 0.018020475283265114,
+      "step": 1053
+    },
+    {
+      "epoch": 0.13175,
+      "grad_norm": 2.6159298419952393,
+      "grad_norm_var": 1.0241485205327874,
+      "learning_rate": 0.0001,
+      "loss": 1.0505,
+      "loss/crossentropy": 2.5744223594665527,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.1320829838514328,
+      "loss/reg": 0.018011758103966713,
+      "step": 1054
+    },
+    {
+      "epoch": 0.131875,
+      "grad_norm": 4.383845806121826,
+      "grad_norm_var": 1.0292396555670902,
+      "learning_rate": 0.0001,
+      "loss": 1.3117,
+      "loss/crossentropy": 2.4903180599212646,
+      "loss/hidden": 0.90625,
+      "loss/logits": 0.22538328170776367,
+      "loss/reg": 0.0180036798119545,
+      "step": 1055
+    },
+    {
+      "epoch": 0.132,
+      "grad_norm": 2.4342551231384277,
+      "grad_norm_var": 0.7907919306719045,
+      "learning_rate": 0.0001,
+      "loss": 1.0574,
+      "loss/crossentropy": 2.3163890838623047,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.12748895585536957,
+      "loss/reg": 0.01799560710787773,
+      "step": 1056
+    },
+    {
+      "epoch": 0.132125,
+      "grad_norm": 3.542881727218628,
+      "grad_norm_var": 0.7864152227556851,
+      "learning_rate": 0.0001,
+      "loss": 1.294,
+      "loss/crossentropy": 2.3338446617126465,
+      "loss/hidden": 0.93359375,
+      "loss/logits": 0.18051280081272125,
+      "loss/reg": 0.017986783757805824,
+      "step": 1057
+    },
+    {
+      "epoch": 0.13225,
+      "grad_norm": 4.0569915771484375,
+      "grad_norm_var": 0.7341544247946139,
+      "learning_rate": 0.0001,
+      "loss": 1.3818,
+      "loss/crossentropy": 2.330078363418579,
+      "loss/hidden": 1.015625,
+      "loss/logits": 0.18636882305145264,
+      "loss/reg": 0.017977885901927948,
+      "step": 1058
+    },
+    {
+      "epoch": 0.132375,
+      "grad_norm": 3.663426637649536,
+      "grad_norm_var": 0.7217416281732681,
+      "learning_rate": 0.0001,
+      "loss": 1.1508,
+      "loss/crossentropy": 2.5138370990753174,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.1742265522480011,
+      "loss/reg": 0.017969388514757156,
+      "step": 1059
+    },
+    {
+      "epoch": 0.1325,
+      "grad_norm": 2.9652159214019775,
+      "grad_norm_var": 0.7484703245510673,
+      "learning_rate": 0.0001,
+      "loss": 1.1313,
+      "loss/crossentropy": 2.2722136974334717,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.1430921107530594,
+      "loss/reg": 0.0179609302431345,
+      "step": 1060
+    },
+    {
+      "epoch": 0.132625,
+      "grad_norm": 5.577633857727051,
+      "grad_norm_var": 0.9767082401115801,
+      "learning_rate": 0.0001,
+      "loss": 1.7627,
+      "loss/crossentropy": 2.562012195587158,
+      "loss/hidden": 1.390625,
+      "loss/logits": 0.19251598417758942,
+      "loss/reg": 0.01795242354273796,
+      "step": 1061
+    },
+    {
+      "epoch": 0.13275,
+      "grad_norm": 5.364041328430176,
+      "grad_norm_var": 0.9941443511093354,
+      "learning_rate": 0.0001,
+      "loss": 1.5266,
+      "loss/crossentropy": 2.507934808731079,
+      "loss/hidden": 1.125,
+      "loss/logits": 0.22217199206352234,
+      "loss/reg": 0.017944158986210823,
+      "step": 1062
+    },
+    {
+      "epoch": 0.132875,
+      "grad_norm": 6.644399642944336,
+      "grad_norm_var": 1.4066377839550304,
+      "learning_rate": 0.0001,
+      "loss": 1.5347,
+      "loss/crossentropy": 2.4112086296081543,
+      "loss/hidden": 1.109375,
+      "loss/logits": 0.24593724310398102,
+      "loss/reg": 0.017935609444975853,
+      "step": 1063
+    },
+    {
+      "epoch": 0.133,
+      "grad_norm": 2.846679210662842,
+      "grad_norm_var": 1.4137598873748003,
+      "learning_rate": 0.0001,
+      "loss": 1.1476,
+      "loss/crossentropy": 2.2182581424713135,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.1519462913274765,
+      "loss/reg": 0.01792830042541027,
+      "step": 1064
+    },
+    {
+      "epoch": 0.133125,
+      "grad_norm": 6.004873752593994,
+      "grad_norm_var": 1.688838288773405,
+      "learning_rate": 0.0001,
+      "loss": 1.6923,
+      "loss/crossentropy": 2.5660815238952637,
+      "loss/hidden": 1.2578125,
+      "loss/logits": 0.25529351830482483,
+      "loss/reg": 0.01792137697339058,
+      "step": 1065
+    },
+    {
+      "epoch": 0.13325,
+      "grad_norm": 2.905362844467163,
+      "grad_norm_var": 1.684590354160513,
+      "learning_rate": 0.0001,
+      "loss": 1.1266,
+      "loss/crossentropy": 2.679647207260132,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.16615256667137146,
+      "loss/reg": 0.017914744094014168,
+      "step": 1066
+    },
+    {
+      "epoch": 0.133375,
+      "grad_norm": 2.774467945098877,
+      "grad_norm_var": 1.7557347328903679,
+      "learning_rate": 0.0001,
+      "loss": 0.9917,
+      "loss/crossentropy": 2.4071364402770996,
+      "loss/hidden": 0.69921875,
+      "loss/logits": 0.11337558180093765,
+      "loss/reg": 0.017907986417412758,
+      "step": 1067
+    },
+    {
+      "epoch": 0.1335,
+      "grad_norm": 3.0080766677856445,
+      "grad_norm_var": 1.7760288881986261,
+      "learning_rate": 0.0001,
+      "loss": 1.1718,
+      "loss/crossentropy": 2.4299240112304688,
+      "loss/hidden": 0.83984375,
+      "loss/logits": 0.15295377373695374,
+      "loss/reg": 0.01789918728172779,
+      "step": 1068
+    },
+    {
+      "epoch": 0.133625,
+      "grad_norm": 3.646827220916748,
+      "grad_norm_var": 1.750571466335814,
+      "learning_rate": 0.0001,
+      "loss": 1.2924,
+      "loss/crossentropy": 2.461158514022827,
+      "loss/hidden": 0.93359375,
+      "loss/logits": 0.17990370094776154,
+      "loss/reg": 0.01789136230945587,
+      "step": 1069
+    },
+    {
+      "epoch": 0.13375,
+      "grad_norm": 4.164073944091797,
+      "grad_norm_var": 1.6348612297712255,
+      "learning_rate": 0.0001,
+      "loss": 1.171,
+      "loss/crossentropy": 2.4830503463745117,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.16796636581420898,
+      "loss/reg": 0.017882652580738068,
+      "step": 1070
+    },
+    {
+      "epoch": 0.133875,
+      "grad_norm": 3.7383811473846436,
+      "grad_norm_var": 1.6277745939444626,
+      "learning_rate": 0.0001,
+      "loss": 1.2749,
+      "loss/crossentropy": 2.3813910484313965,
+      "loss/hidden": 0.94921875,
+      "loss/logits": 0.14690269529819489,
+      "loss/reg": 0.017873771488666534,
+      "step": 1071
+    },
+    {
+      "epoch": 0.134,
+      "grad_norm": 4.526171684265137,
+      "grad_norm_var": 1.476108335704568,
+      "learning_rate": 0.0001,
+      "loss": 1.6415,
+      "loss/crossentropy": 2.1495041847229004,
+      "loss/hidden": 1.234375,
+      "loss/logits": 0.22847777605056763,
+      "loss/reg": 0.017865996807813644,
+      "step": 1072
+    },
+    {
+      "epoch": 0.134125,
+      "grad_norm": 3.382908582687378,
+      "grad_norm_var": 1.4893637052056738,
+      "learning_rate": 0.0001,
+      "loss": 1.0784,
+      "loss/crossentropy": 2.468477725982666,
+      "loss/hidden": 0.765625,
+      "loss/logits": 0.13421444594860077,
+      "loss/reg": 0.01785840094089508,
+      "step": 1073
+    },
+    {
+      "epoch": 0.13425,
+      "grad_norm": 5.163125991821289,
+      "grad_norm_var": 1.5625376434966463,
+      "learning_rate": 0.0001,
+      "loss": 1.2499,
+      "loss/crossentropy": 2.2978320121765137,
+      "loss/hidden": 0.91796875,
+      "loss/logits": 0.1534654200077057,
+      "loss/reg": 0.01785091683268547,
+      "step": 1074
+    },
+    {
+      "epoch": 0.134375,
+      "grad_norm": 3.1810903549194336,
+      "grad_norm_var": 1.6082726182831408,
+      "learning_rate": 0.0001,
+      "loss": 1.0638,
+      "loss/crossentropy": 2.5080108642578125,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.1275557279586792,
+      "loss/reg": 0.017842529341578484,
+      "step": 1075
+    },
+    {
+      "epoch": 0.1345,
+      "grad_norm": 2.6275007724761963,
+      "grad_norm_var": 1.6673241917556567,
+      "learning_rate": 0.0001,
+      "loss": 1.023,
+      "loss/crossentropy": 2.725398302078247,
+      "loss/hidden": 0.7265625,
+      "loss/logits": 0.11811243742704391,
+      "loss/reg": 0.01783500798046589,
+      "step": 1076
+    },
+    {
+      "epoch": 0.134625,
+      "grad_norm": 2.4731173515319824,
+      "grad_norm_var": 1.656907168261868,
+      "learning_rate": 0.0001,
+      "loss": 1.1398,
+      "loss/crossentropy": 2.2775444984436035,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.14902615547180176,
+      "loss/reg": 0.01782614178955555,
+      "step": 1077
+    },
+    {
+      "epoch": 0.13475,
+      "grad_norm": 2.6200878620147705,
+      "grad_norm_var": 1.593020801522063,
+      "learning_rate": 0.0001,
+      "loss": 1.0744,
+      "loss/crossentropy": 2.6091089248657227,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.1501138061285019,
+      "loss/reg": 0.017817262560129166,
+      "step": 1078
+    },
+    {
+      "epoch": 0.134875,
+      "grad_norm": 2.637160062789917,
+      "grad_norm_var": 1.040390657280697,
+      "learning_rate": 0.0001,
+      "loss": 1.0533,
+      "loss/crossentropy": 2.6109707355499268,
+      "loss/hidden": 0.734375,
+      "loss/logits": 0.1408311128616333,
+      "loss/reg": 0.01780843175947666,
+      "step": 1079
+    },
+    {
+      "epoch": 0.135,
+      "grad_norm": 2.4874284267425537,
+      "grad_norm_var": 1.0788527015533946,
+      "learning_rate": 0.0001,
+      "loss": 1.0129,
+      "loss/crossentropy": 2.4900012016296387,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.12004341185092926,
+      "loss/reg": 0.017799846827983856,
+      "step": 1080
+    },
+    {
+      "epoch": 0.135125,
+      "grad_norm": 2.8923330307006836,
+      "grad_norm_var": 0.6277088581300307,
+      "learning_rate": 0.0001,
+      "loss": 1.0085,
+      "loss/crossentropy": 2.877713441848755,
+      "loss/hidden": 0.69140625,
+      "loss/logits": 0.1391412615776062,
+      "loss/reg": 0.017792070284485817,
+      "step": 1081
+    },
+    {
+      "epoch": 0.13525,
+      "grad_norm": 3.0681328773498535,
+      "grad_norm_var": 0.6215757739924422,
+      "learning_rate": 0.0001,
+      "loss": 1.1841,
+      "loss/crossentropy": 2.4870967864990234,
+      "loss/hidden": 0.83984375,
+      "loss/logits": 0.16639447212219238,
+      "loss/reg": 0.017783144488930702,
+      "step": 1082
+    },
+    {
+      "epoch": 0.135375,
+      "grad_norm": 2.943787097930908,
+      "grad_norm_var": 0.6120804925495794,
+      "learning_rate": 0.0001,
+      "loss": 0.9977,
+      "loss/crossentropy": 2.4002597332000732,
+      "loss/hidden": 0.69921875,
+      "loss/logits": 0.12076494097709656,
+      "loss/reg": 0.017775090411305428,
+      "step": 1083
+    },
+    {
+      "epoch": 0.1355,
+      "grad_norm": 3.1173973083496094,
+      "grad_norm_var": 0.6087907870581594,
+      "learning_rate": 0.0001,
+      "loss": 1.1729,
+      "loss/crossentropy": 2.6328957080841064,
+      "loss/hidden": 0.85546875,
+      "loss/logits": 0.13975805044174194,
+      "loss/reg": 0.017766445875167847,
+      "step": 1084
+    },
+    {
+      "epoch": 0.135625,
+      "grad_norm": 4.231754779815674,
+      "grad_norm_var": 0.6578597190419477,
+      "learning_rate": 0.0001,
+      "loss": 1.5044,
+      "loss/crossentropy": 2.6625559329986572,
+      "loss/hidden": 1.109375,
+      "loss/logits": 0.2174825519323349,
+      "loss/reg": 0.01775727979838848,
+      "step": 1085
+    },
+    {
+      "epoch": 0.13575,
+      "grad_norm": 3.1539108753204346,
+      "grad_norm_var": 0.6090813956553401,
+      "learning_rate": 0.0001,
+      "loss": 1.0553,
+      "loss/crossentropy": 2.685743808746338,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.1317606419324875,
+      "loss/reg": 0.017748642712831497,
+      "step": 1086
+    },
+    {
+      "epoch": 0.135875,
+      "grad_norm": 3.386591672897339,
+      "grad_norm_var": 0.5946246391748462,
+      "learning_rate": 0.0001,
+      "loss": 1.4794,
+      "loss/crossentropy": 2.3431785106658936,
+      "loss/hidden": 1.0625,
+      "loss/logits": 0.23952066898345947,
+      "loss/reg": 0.01773969829082489,
+      "step": 1087
+    },
+    {
+      "epoch": 0.136,
+      "grad_norm": 2.8175418376922607,
+      "grad_norm_var": 0.4848234667031941,
+      "learning_rate": 0.0001,
+      "loss": 1.0416,
+      "loss/crossentropy": 2.676400899887085,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.13378942012786865,
+      "loss/reg": 0.01773088052868843,
+      "step": 1088
+    },
+    {
+      "epoch": 0.136125,
+      "grad_norm": 3.5386815071105957,
+      "grad_norm_var": 0.4914580502239192,
+      "learning_rate": 0.0001,
+      "loss": 1.1391,
+      "loss/crossentropy": 2.6673851013183594,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.14545656740665436,
+      "loss/reg": 0.01772209256887436,
+      "step": 1089
+    },
+    {
+      "epoch": 0.13625,
+      "grad_norm": 2.944378614425659,
+      "grad_norm_var": 0.2024704804136301,
+      "learning_rate": 0.0001,
+      "loss": 1.0518,
+      "loss/crossentropy": 2.3069140911102295,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.11681585013866425,
+      "loss/reg": 0.017712950706481934,
+      "step": 1090
+    },
+    {
+      "epoch": 0.136375,
+      "grad_norm": 2.4177603721618652,
+      "grad_norm_var": 0.221225648364459,
+      "learning_rate": 0.0001,
+      "loss": 1.0819,
+      "loss/crossentropy": 2.5605850219726562,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.15096938610076904,
+      "loss/reg": 0.01770433411002159,
+      "step": 1091
+    },
+    {
+      "epoch": 0.1365,
+      "grad_norm": 2.8017842769622803,
+      "grad_norm_var": 0.2154010561648003,
+      "learning_rate": 0.0001,
+      "loss": 1.0446,
+      "loss/crossentropy": 2.837974786758423,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.12545213103294373,
+      "loss/reg": 0.017695914953947067,
+      "step": 1092
+    },
+    {
+      "epoch": 0.136625,
+      "grad_norm": 2.30322265625,
+      "grad_norm_var": 0.2284775401570935,
+      "learning_rate": 0.0001,
+      "loss": 1.0164,
+      "loss/crossentropy": 2.645153760910034,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.1285744607448578,
+      "loss/reg": 0.017688019201159477,
+      "step": 1093
+    },
+    {
+      "epoch": 0.13675,
+      "grad_norm": 3.5715980529785156,
+      "grad_norm_var": 0.24192379822145516,
+      "learning_rate": 0.0001,
+      "loss": 1.1766,
+      "loss/crossentropy": 2.2281081676483154,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.16386280953884125,
+      "loss/reg": 0.017679255455732346,
+      "step": 1094
+    },
+    {
+      "epoch": 0.136875,
+      "grad_norm": 3.5261693000793457,
+      "grad_norm_var": 0.24598854725778285,
+      "learning_rate": 0.0001,
+      "loss": 1.4345,
+      "loss/crossentropy": 2.311053514480591,
+      "loss/hidden": 1.03125,
+      "loss/logits": 0.22651731967926025,
+      "loss/reg": 0.017671290785074234,
+      "step": 1095
+    },
+    {
+      "epoch": 0.137,
+      "grad_norm": 4.736113548278809,
+      "grad_norm_var": 0.3858102993473473,
+      "learning_rate": 0.0001,
+      "loss": 1.1132,
+      "loss/crossentropy": 2.483963966369629,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.1318715214729309,
+      "loss/reg": 0.01766252890229225,
+      "step": 1096
+    },
+    {
+      "epoch": 0.137125,
+      "grad_norm": 3.013045310974121,
+      "grad_norm_var": 0.3815164758052994,
+      "learning_rate": 0.0001,
+      "loss": 1.0531,
+      "loss/crossentropy": 2.3578484058380127,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.12263330817222595,
+      "loss/reg": 0.017653891816735268,
+      "step": 1097
+    },
+    {
+      "epoch": 0.13725,
+      "grad_norm": 3.655559539794922,
+      "grad_norm_var": 0.3909346674988434,
+      "learning_rate": 0.0001,
+      "loss": 1.0558,
+      "loss/crossentropy": 2.7702715396881104,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.13715983927249908,
+      "loss/reg": 0.01764553412795067,
+      "step": 1098
+    },
+    {
+      "epoch": 0.137375,
+      "grad_norm": 4.493204593658447,
+      "grad_norm_var": 0.47566105167650385,
+      "learning_rate": 0.0001,
+      "loss": 1.1521,
+      "loss/crossentropy": 3.274940013885498,
+      "loss/hidden": 0.765625,
+      "loss/logits": 0.21013642847537994,
+      "loss/reg": 0.01763724535703659,
+      "step": 1099
+    },
+    {
+      "epoch": 0.1375,
+      "grad_norm": 2.9585206508636475,
+      "grad_norm_var": 0.482309950085633,
+      "learning_rate": 0.0001,
+      "loss": 1.0561,
+      "loss/crossentropy": 2.3324267864227295,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.12980124354362488,
+      "loss/reg": 0.017628395929932594,
+      "step": 1100
+    },
+    {
+      "epoch": 0.137625,
+      "grad_norm": 4.262050628662109,
+      "grad_norm_var": 0.4859417805331835,
+      "learning_rate": 0.0001,
+      "loss": 1.0991,
+      "loss/crossentropy": 2.4208176136016846,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.12208649516105652,
+      "loss/reg": 0.017619585618376732,
+      "step": 1101
+    },
+    {
+      "epoch": 0.13775,
+      "grad_norm": 2.6123526096343994,
+      "grad_norm_var": 0.5183416158931512,
+      "learning_rate": 0.0001,
+      "loss": 0.9988,
+      "loss/crossentropy": 2.6298210620880127,
+      "loss/hidden": 0.69140625,
+      "loss/logits": 0.13125354051589966,
+      "loss/reg": 0.017610682174563408,
+      "step": 1102
+    },
+    {
+      "epoch": 0.137875,
+      "grad_norm": 3.968358039855957,
+      "grad_norm_var": 0.5450550638000642,
+      "learning_rate": 0.0001,
+      "loss": 1.1401,
+      "loss/crossentropy": 2.4403319358825684,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.15549315512180328,
+      "loss/reg": 0.01760167069733143,
+      "step": 1103
+    },
+    {
+      "epoch": 0.138,
+      "grad_norm": 2.6694836616516113,
+      "grad_norm_var": 0.5569615426057339,
+      "learning_rate": 0.0001,
+      "loss": 1.2515,
+      "loss/crossentropy": 2.35309100151062,
+      "loss/hidden": 0.8984375,
+      "loss/logits": 0.1771024614572525,
+      "loss/reg": 0.01759263686835766,
+      "step": 1104
+    },
+    {
+      "epoch": 0.138125,
+      "grad_norm": 2.5527474880218506,
+      "grad_norm_var": 0.5918626570387079,
+      "learning_rate": 0.0001,
+      "loss": 0.9445,
+      "loss/crossentropy": 2.5089659690856934,
+      "loss/hidden": 0.65625,
+      "loss/logits": 0.11239723861217499,
+      "loss/reg": 0.017583860084414482,
+      "step": 1105
+    },
+    {
+      "epoch": 0.13825,
+      "grad_norm": 3.055800676345825,
+      "grad_norm_var": 0.5876466077321024,
+      "learning_rate": 0.0001,
+      "loss": 0.9616,
+      "loss/crossentropy": 3.0698087215423584,
+      "loss/hidden": 0.66796875,
+      "loss/logits": 0.11786890029907227,
+      "loss/reg": 0.01757502555847168,
+      "step": 1106
+    },
+    {
+      "epoch": 0.138375,
+      "grad_norm": 2.3708860874176025,
+      "grad_norm_var": 0.5932188518407192,
+      "learning_rate": 0.0001,
+      "loss": 1.1363,
+      "loss/crossentropy": 2.364453077316284,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.13644903898239136,
+      "loss/reg": 0.017566362395882607,
+      "step": 1107
+    },
+    {
+      "epoch": 0.1385,
+      "grad_norm": 3.3530497550964355,
+      "grad_norm_var": 0.5767366681943767,
+      "learning_rate": 0.0001,
+      "loss": 1.1719,
+      "loss/crossentropy": 2.339834451675415,
+      "loss/hidden": 0.86328125,
+      "loss/logits": 0.13302001357078552,
+      "loss/reg": 0.017557917162775993,
+      "step": 1108
+    },
+    {
+      "epoch": 0.138625,
+      "grad_norm": 4.9073686599731445,
+      "grad_norm_var": 0.6479273995103992,
+      "learning_rate": 0.0001,
+      "loss": 1.144,
+      "loss/crossentropy": 2.608236074447632,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.1325811743736267,
+      "loss/reg": 0.017549151554703712,
+      "step": 1109
+    },
+    {
+      "epoch": 0.13875,
+      "grad_norm": 4.318938732147217,
+      "grad_norm_var": 0.6917982612527337,
+      "learning_rate": 0.0001,
+      "loss": 1.0295,
+      "loss/crossentropy": 2.6003918647766113,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.12366551160812378,
+      "loss/reg": 0.017540371045470238,
+      "step": 1110
+    },
+    {
+      "epoch": 0.138875,
+      "grad_norm": 3.439058303833008,
+      "grad_norm_var": 0.692297895774635,
+      "learning_rate": 0.0001,
+      "loss": 1.1891,
+      "loss/crossentropy": 2.3828976154327393,
+      "loss/hidden": 0.859375,
+      "loss/logits": 0.1544424593448639,
+      "loss/reg": 0.01753184385597706,
+      "step": 1111
+    },
+    {
+      "epoch": 0.139,
+      "grad_norm": 3.090857744216919,
+      "grad_norm_var": 0.5953394071265308,
+      "learning_rate": 0.0001,
+      "loss": 1.1475,
+      "loss/crossentropy": 2.2759549617767334,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.1597966104745865,
+      "loss/reg": 0.017523299902677536,
+      "step": 1112
+    },
+    {
+      "epoch": 0.139125,
+      "grad_norm": 2.421586513519287,
+      "grad_norm_var": 0.6493026217043355,
+      "learning_rate": 0.0001,
+      "loss": 1.0266,
+      "loss/crossentropy": 2.580230236053467,
+      "loss/hidden": 0.7265625,
+      "loss/logits": 0.12484898418188095,
+      "loss/reg": 0.017514871433377266,
+      "step": 1113
+    },
+    {
+      "epoch": 0.13925,
+      "grad_norm": 2.3141372203826904,
+      "grad_norm_var": 0.7130373793975477,
+      "learning_rate": 0.0001,
+      "loss": 1.0674,
+      "loss/crossentropy": 2.7415308952331543,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.14231345057487488,
+      "loss/reg": 0.01750599592924118,
+      "step": 1114
+    },
+    {
+      "epoch": 0.139375,
+      "grad_norm": 3.1447672843933105,
+      "grad_norm_var": 0.6120215321394992,
+      "learning_rate": 0.0001,
+      "loss": 1.0712,
+      "loss/crossentropy": 2.4831013679504395,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.13447736203670502,
+      "loss/reg": 0.017497511580586433,
+      "step": 1115
+    },
+    {
+      "epoch": 0.1395,
+      "grad_norm": 3.2429728507995605,
+      "grad_norm_var": 0.6073512012070741,
+      "learning_rate": 0.0001,
+      "loss": 1.1311,
+      "loss/crossentropy": 2.567009925842285,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.1554429829120636,
+      "loss/reg": 0.01748875342309475,
+      "step": 1116
+    },
+    {
+      "epoch": 0.139625,
+      "grad_norm": 3.240558624267578,
+      "grad_norm_var": 0.532380465942029,
+      "learning_rate": 0.0001,
+      "loss": 1.1099,
+      "loss/crossentropy": 2.860992431640625,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.1421348601579666,
+      "loss/reg": 0.01748001016676426,
+      "step": 1117
+    },
+    {
+      "epoch": 0.13975,
+      "grad_norm": 3.012918710708618,
+      "grad_norm_var": 0.5126825052838881,
+      "learning_rate": 0.0001,
+      "loss": 1.1268,
+      "loss/crossentropy": 2.4657137393951416,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.13955920934677124,
+      "loss/reg": 0.017471779137849808,
+      "step": 1118
+    },
+    {
+      "epoch": 0.139875,
+      "grad_norm": 2.7599992752075195,
+      "grad_norm_var": 0.47917524489163554,
+      "learning_rate": 0.0001,
+      "loss": 1.0584,
+      "loss/crossentropy": 2.366868019104004,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.1455003321170807,
+      "loss/reg": 0.017463646829128265,
+      "step": 1119
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.73720645904541,
+      "grad_norm_var": 0.48651163922628105,
+      "learning_rate": 0.0001,
+      "loss": 1.2863,
+      "loss/crossentropy": 2.9208223819732666,
+      "loss/hidden": 0.93359375,
+      "loss/logits": 0.17812727391719818,
+      "loss/reg": 0.017455046996474266,
+      "step": 1120
+    },
+    {
+      "epoch": 0.140125,
+      "grad_norm": 2.549912691116333,
+      "grad_norm_var": 0.48675118323949296,
+      "learning_rate": 0.0001,
+      "loss": 1.0037,
+      "loss/crossentropy": 2.217289447784424,
+      "loss/hidden": 0.70703125,
+      "loss/logits": 0.12221944332122803,
+      "loss/reg": 0.017446177080273628,
+      "step": 1121
+    },
+    {
+      "epoch": 0.14025,
+      "grad_norm": 3.35829758644104,
+      "grad_norm_var": 0.48725917149039616,
+      "learning_rate": 0.0001,
+      "loss": 0.9516,
+      "loss/crossentropy": 2.5682172775268555,
+      "loss/hidden": 0.66015625,
+      "loss/logits": 0.1170571893453598,
+      "loss/reg": 0.017437297850847244,
+      "step": 1122
+    },
+    {
+      "epoch": 0.140375,
+      "grad_norm": 3.384793519973755,
+      "grad_norm_var": 0.4388955051274242,
+      "learning_rate": 0.0001,
+      "loss": 1.107,
+      "loss/crossentropy": 2.8147690296173096,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.14758194983005524,
+      "loss/reg": 0.017428115010261536,
+      "step": 1123
+    },
+    {
+      "epoch": 0.1405,
+      "grad_norm": 2.5823116302490234,
+      "grad_norm_var": 0.4672083375473524,
+      "learning_rate": 0.0001,
+      "loss": 1.0724,
+      "loss/crossentropy": 2.696993827819824,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.14429491758346558,
+      "loss/reg": 0.017419347539544106,
+      "step": 1124
+    },
+    {
+      "epoch": 0.140625,
+      "grad_norm": 5.868249416351318,
+      "grad_norm_var": 0.7412100386445967,
+      "learning_rate": 0.0001,
+      "loss": 1.2337,
+      "loss/crossentropy": 2.6103923320770264,
+      "loss/hidden": 0.8984375,
+      "loss/logits": 0.16118671000003815,
+      "loss/reg": 0.017410660162568092,
+      "step": 1125
+    },
+    {
+      "epoch": 0.14075,
+      "grad_norm": 2.4435040950775146,
+      "grad_norm_var": 0.7010336436922195,
+      "learning_rate": 0.0001,
+      "loss": 0.9667,
+      "loss/crossentropy": 2.3706815242767334,
+      "loss/hidden": 0.6640625,
+      "loss/logits": 0.12864476442337036,
+      "loss/reg": 0.01740197278559208,
+      "step": 1126
+    },
+    {
+      "epoch": 0.140875,
+      "grad_norm": 2.7513833045959473,
+      "grad_norm_var": 0.7051812497689844,
+      "learning_rate": 0.0001,
+      "loss": 1.1451,
+      "loss/crossentropy": 2.4533605575561523,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.15479305386543274,
+      "loss/reg": 0.017393220216035843,
+      "step": 1127
+    },
+    {
+      "epoch": 0.141,
+      "grad_norm": 2.8634872436523438,
+      "grad_norm_var": 0.7092644673154382,
+      "learning_rate": 0.0001,
+      "loss": 1.1193,
+      "loss/crossentropy": 2.1886956691741943,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.14075767993927002,
+      "loss/reg": 0.01738525740802288,
+      "step": 1128
+    },
+    {
+      "epoch": 0.141125,
+      "grad_norm": 2.704045057296753,
+      "grad_norm_var": 0.6885219755537205,
+      "learning_rate": 0.0001,
+      "loss": 0.973,
+      "loss/crossentropy": 2.4812088012695312,
+      "loss/hidden": 0.65625,
+      "loss/logits": 0.1429394781589508,
+      "loss/reg": 0.01737692952156067,
+      "step": 1129
+    },
+    {
+      "epoch": 0.14125,
+      "grad_norm": 2.6875903606414795,
+      "grad_norm_var": 0.656991790963667,
+      "learning_rate": 0.0001,
+      "loss": 1.0697,
+      "loss/crossentropy": 2.3832883834838867,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.126431405544281,
+      "loss/reg": 0.017368923872709274,
+      "step": 1130
+    },
+    {
+      "epoch": 0.141375,
+      "grad_norm": 7.774624824523926,
+      "grad_norm_var": 1.9961090220751256,
+      "learning_rate": 0.0001,
+      "loss": 1.383,
+      "loss/crossentropy": 2.5491535663604736,
+      "loss/hidden": 1.0390625,
+      "loss/logits": 0.1702873259782791,
+      "loss/reg": 0.01736014150083065,
+      "step": 1131
+    },
+    {
+      "epoch": 0.1415,
+      "grad_norm": 3.5188465118408203,
+      "grad_norm_var": 1.9937980339815395,
+      "learning_rate": 0.0001,
+      "loss": 1.1816,
+      "loss/crossentropy": 2.6813807487487793,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.16434520483016968,
+      "loss/reg": 0.017351284623146057,
+      "step": 1132
+    },
+    {
+      "epoch": 0.141625,
+      "grad_norm": 30.482587814331055,
+      "grad_norm_var": 47.607494749063044,
+      "learning_rate": 0.0001,
+      "loss": 1.0592,
+      "loss/crossentropy": 2.607267379760742,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.13182219862937927,
+      "loss/reg": 0.01734289340674877,
+      "step": 1133
+    },
+    {
+      "epoch": 0.14175,
+      "grad_norm": 3.333674907684326,
+      "grad_norm_var": 47.52231423180526,
+      "learning_rate": 0.0001,
+      "loss": 1.1791,
+      "loss/crossentropy": 2.5332696437835693,
+      "loss/hidden": 0.85546875,
+      "loss/logits": 0.15026560425758362,
+      "loss/reg": 0.017334245145320892,
+      "step": 1134
+    },
+    {
+      "epoch": 0.141875,
+      "grad_norm": 2.8526668548583984,
+      "grad_norm_var": 47.4930115697571,
+      "learning_rate": 0.0001,
+      "loss": 1.1235,
+      "loss/crossentropy": 2.583221673965454,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.13772843778133392,
+      "loss/reg": 0.017325541004538536,
+      "step": 1135
+    },
+    {
+      "epoch": 0.142,
+      "grad_norm": 3.022386074066162,
+      "grad_norm_var": 47.66253737043978,
+      "learning_rate": 0.0001,
+      "loss": 1.2714,
+      "loss/crossentropy": 2.3197121620178223,
+      "loss/hidden": 0.9453125,
+      "loss/logits": 0.15295147895812988,
+      "loss/reg": 0.017316767945885658,
+      "step": 1136
+    },
+    {
+      "epoch": 0.142125,
+      "grad_norm": 3.3717892169952393,
+      "grad_norm_var": 47.421346164152666,
+      "learning_rate": 0.0001,
+      "loss": 1.131,
+      "loss/crossentropy": 2.543901205062866,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.13759344816207886,
+      "loss/reg": 0.017307989299297333,
+      "step": 1137
+    },
+    {
+      "epoch": 0.14225,
+      "grad_norm": 3.5415263175964355,
+      "grad_norm_var": 47.37875577313657,
+      "learning_rate": 0.0001,
+      "loss": 1.0293,
+      "loss/crossentropy": 2.4758806228637695,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.12582623958587646,
+      "loss/reg": 0.01730157807469368,
+      "step": 1138
+    },
+    {
+      "epoch": 0.142375,
+      "grad_norm": 2.9773669242858887,
+      "grad_norm_var": 47.48768287025903,
+      "learning_rate": 0.0001,
+      "loss": 1.0163,
+      "loss/crossentropy": 2.638218402862549,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.12850065529346466,
+      "loss/reg": 0.017293203622102737,
+      "step": 1139
+    },
+    {
+      "epoch": 0.1425,
+      "grad_norm": 2.8083419799804688,
+      "grad_norm_var": 47.41278427285148,
+      "learning_rate": 0.0001,
+      "loss": 1.0945,
+      "loss/crossentropy": 2.503974676132202,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.1638229787349701,
+      "loss/reg": 0.01728462241590023,
+      "step": 1140
+    },
+    {
+      "epoch": 0.142625,
+      "grad_norm": 2.425929069519043,
+      "grad_norm_var": 47.84099408884167,
+      "learning_rate": 0.0001,
+      "loss": 1.0449,
+      "loss/crossentropy": 2.3581292629241943,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.12217840552330017,
+      "loss/reg": 0.017276806756854057,
+      "step": 1141
+    },
+    {
+      "epoch": 0.14275,
+      "grad_norm": 39.945011138916016,
+      "grad_norm_var": 123.09327375389867,
+      "learning_rate": 0.0001,
+      "loss": 1.0453,
+      "loss/crossentropy": 2.5906147956848145,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.11872614920139313,
+      "loss/reg": 0.017268478870391846,
+      "step": 1142
+    },
+    {
+      "epoch": 0.142875,
+      "grad_norm": 3.0662145614624023,
+      "grad_norm_var": 122.90784367859821,
+      "learning_rate": 0.0001,
+      "loss": 1.0595,
+      "loss/crossentropy": 2.2695305347442627,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.1291281282901764,
+      "loss/reg": 0.017260266467928886,
+      "step": 1143
+    },
+    {
+      "epoch": 0.143,
+      "grad_norm": 4.583246231079102,
+      "grad_norm_var": 122.06713805652959,
+      "learning_rate": 0.0001,
+      "loss": 1.1495,
+      "loss/crossentropy": 2.480984926223755,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.16442613303661346,
+      "loss/reg": 0.017252640798687935,
+      "step": 1144
+    },
+    {
+      "epoch": 0.143125,
+      "grad_norm": 2.981353521347046,
+      "grad_norm_var": 121.89670586108006,
+      "learning_rate": 0.0001,
+      "loss": 1.1551,
+      "loss/crossentropy": 2.2689995765686035,
+      "loss/hidden": 0.85546875,
+      "loss/logits": 0.12721604108810425,
+      "loss/reg": 0.01724405400454998,
+      "step": 1145
+    },
+    {
+      "epoch": 0.14325,
+      "grad_norm": 2.3794004917144775,
+      "grad_norm_var": 122.09878373545942,
+      "learning_rate": 0.0001,
+      "loss": 1.0441,
+      "loss/crossentropy": 2.5703084468841553,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.13347238302230835,
+      "loss/reg": 0.017235582694411278,
+      "step": 1146
+    },
+    {
+      "epoch": 0.143375,
+      "grad_norm": 2.4713571071624756,
+      "grad_norm_var": 123.62106362597764,
+      "learning_rate": 0.0001,
+      "loss": 0.9492,
+      "loss/crossentropy": 2.447471857070923,
+      "loss/hidden": 0.65625,
+      "loss/logits": 0.12064240872859955,
+      "loss/reg": 0.017227739095687866,
+      "step": 1147
+    },
+    {
+      "epoch": 0.1435,
+      "grad_norm": 2.6587135791778564,
+      "grad_norm_var": 124.07916434426059,
+      "learning_rate": 0.0001,
+      "loss": 1.0818,
+      "loss/crossentropy": 2.6435859203338623,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.15571220219135284,
+      "loss/reg": 0.017219962552189827,
+      "step": 1148
+    },
+    {
+      "epoch": 0.143625,
+      "grad_norm": 2.5300068855285645,
+      "grad_norm_var": 85.6034890468201,
+      "learning_rate": 0.0001,
+      "loss": 1.0955,
+      "loss/crossentropy": 2.370417356491089,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.13428199291229248,
+      "loss/reg": 0.017211301252245903,
+      "step": 1149
+    },
+    {
+      "epoch": 0.14375,
+      "grad_norm": 2.821120023727417,
+      "grad_norm_var": 85.75492487355808,
+      "learning_rate": 0.0001,
+      "loss": 1.1666,
+      "loss/crossentropy": 2.331489324569702,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.15861350297927856,
+      "loss/reg": 0.017202915623784065,
+      "step": 1150
+    },
+    {
+      "epoch": 0.143875,
+      "grad_norm": 2.5067081451416016,
+      "grad_norm_var": 85.8742473316794,
+      "learning_rate": 0.0001,
+      "loss": 0.9844,
+      "loss/crossentropy": 2.466968297958374,
+      "loss/hidden": 0.69140625,
+      "loss/logits": 0.12101612985134125,
+      "loss/reg": 0.017194462940096855,
+      "step": 1151
+    },
+    {
+      "epoch": 0.144,
+      "grad_norm": 3.062662124633789,
+      "grad_norm_var": 85.8623557526669,
+      "learning_rate": 0.0001,
+      "loss": 1.0339,
+      "loss/crossentropy": 2.409207820892334,
+      "loss/hidden": 0.7265625,
+      "loss/logits": 0.13548848032951355,
+      "loss/reg": 0.017186442390084267,
+      "step": 1152
+    },
+    {
+      "epoch": 0.144125,
+      "grad_norm": 3.1716883182525635,
+      "grad_norm_var": 85.9151871866652,
+      "learning_rate": 0.0001,
+      "loss": 1.0846,
+      "loss/crossentropy": 2.4929358959198,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.15896344184875488,
+      "loss/reg": 0.0171778816729784,
+      "step": 1153
+    },
+    {
+      "epoch": 0.14425,
+      "grad_norm": 9.434453010559082,
+      "grad_norm_var": 86.74661652378359,
+      "learning_rate": 0.0001,
+      "loss": 2.0602,
+      "loss/crossentropy": 2.4357495307922363,
+      "loss/hidden": 1.6015625,
+      "loss/logits": 0.28698208928108215,
+      "loss/reg": 0.017169814556837082,
+      "step": 1154
+    },
+    {
+      "epoch": 0.144375,
+      "grad_norm": 3.7739198207855225,
+      "grad_norm_var": 86.506246361283,
+      "learning_rate": 0.0001,
+      "loss": 1.0704,
+      "loss/crossentropy": 2.6996419429779053,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.1448526829481125,
+      "loss/reg": 0.017161287367343903,
+      "step": 1155
+    },
+    {
+      "epoch": 0.1445,
+      "grad_norm": 4.359219551086426,
+      "grad_norm_var": 86.06611929299979,
+      "learning_rate": 0.0001,
+      "loss": 1.2325,
+      "loss/crossentropy": 2.506620168685913,
+      "loss/hidden": 0.87109375,
+      "loss/logits": 0.18992647528648376,
+      "loss/reg": 0.017152708023786545,
+      "step": 1156
+    },
+    {
+      "epoch": 0.144625,
+      "grad_norm": 3.83707594871521,
+      "grad_norm_var": 85.56313319362654,
+      "learning_rate": 0.0001,
+      "loss": 1.1268,
+      "loss/crossentropy": 2.392063856124878,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.14678636193275452,
+      "loss/reg": 0.017143724486231804,
+      "step": 1157
+    },
+    {
+      "epoch": 0.14475,
+      "grad_norm": 2.9818544387817383,
+      "grad_norm_var": 2.9152543868814065,
+      "learning_rate": 0.0001,
+      "loss": 0.9894,
+      "loss/crossentropy": 2.590273380279541,
+      "loss/hidden": 0.6953125,
+      "loss/logits": 0.1227588877081871,
+      "loss/reg": 0.01713474653661251,
+      "step": 1158
+    },
+    {
+      "epoch": 0.144875,
+      "grad_norm": 2.5520756244659424,
+      "grad_norm_var": 2.964164435968295,
+      "learning_rate": 0.0001,
+      "loss": 1.1502,
+      "loss/crossentropy": 2.3040196895599365,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.1469496786594391,
+      "loss/reg": 0.017125625163316727,
+      "step": 1159
+    },
+    {
+      "epoch": 0.145,
+      "grad_norm": 3.3840601444244385,
+      "grad_norm_var": 2.881888386237869,
+      "learning_rate": 0.0001,
+      "loss": 1.0954,
+      "loss/crossentropy": 2.4384047985076904,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.13907559216022491,
+      "loss/reg": 0.017117124050855637,
+      "step": 1160
+    },
+    {
+      "epoch": 0.145125,
+      "grad_norm": 6.576678276062012,
+      "grad_norm_var": 3.47394619120245,
+      "learning_rate": 0.0001,
+      "loss": 1.7801,
+      "loss/crossentropy": 2.66701078414917,
+      "loss/hidden": 1.2578125,
+      "loss/logits": 0.35119134187698364,
+      "loss/reg": 0.017108280211687088,
+      "step": 1161
+    },
+    {
+      "epoch": 0.14525,
+      "grad_norm": 2.326526165008545,
+      "grad_norm_var": 3.483123034262428,
+      "learning_rate": 0.0001,
+      "loss": 1.0016,
+      "loss/crossentropy": 2.351609706878662,
+      "loss/hidden": 0.70703125,
+      "loss/logits": 0.12352467328310013,
+      "loss/reg": 0.01709994673728943,
+      "step": 1162
+    },
+    {
+      "epoch": 0.145375,
+      "grad_norm": 9.206579208374023,
+      "grad_norm_var": 5.257167082686383,
+      "learning_rate": 0.0001,
+      "loss": 1.8176,
+      "loss/crossentropy": 2.4763855934143066,
+      "loss/hidden": 1.3515625,
+      "loss/logits": 0.29511886835098267,
+      "loss/reg": 0.017091669142246246,
+      "step": 1163
+    },
+    {
+      "epoch": 0.1455,
+      "grad_norm": 2.9634292125701904,
+      "grad_norm_var": 5.205470661734309,
+      "learning_rate": 0.0001,
+      "loss": 1.1448,
+      "loss/crossentropy": 2.5422465801239014,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.14583711326122284,
+      "loss/reg": 0.01708296500146389,
+      "step": 1164
+    },
+    {
+      "epoch": 0.145625,
+      "grad_norm": 3.018620014190674,
+      "grad_norm_var": 5.118565326969832,
+      "learning_rate": 0.0001,
+      "loss": 1.1524,
+      "loss/crossentropy": 2.559739828109741,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.15742075443267822,
+      "loss/reg": 0.01707414537668228,
+      "step": 1165
+    },
+    {
+      "epoch": 0.14575,
+      "grad_norm": 8.845293045043945,
+      "grad_norm_var": 6.340596335373259,
+      "learning_rate": 0.0001,
+      "loss": 1.558,
+      "loss/crossentropy": 2.8445377349853516,
+      "loss/hidden": 1.140625,
+      "loss/logits": 0.24674071371555328,
+      "loss/reg": 0.017065750434994698,
+      "step": 1166
+    },
+    {
+      "epoch": 0.145875,
+      "grad_norm": 4.030366897583008,
+      "grad_norm_var": 6.0807354199013615,
+      "learning_rate": 0.0001,
+      "loss": 1.2387,
+      "loss/crossentropy": 2.525634765625,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.18527851998806,
+      "loss/reg": 0.017056919634342194,
+      "step": 1167
+    },
+    {
+      "epoch": 0.146,
+      "grad_norm": 4.10527229309082,
+      "grad_norm_var": 5.935618580704363,
+      "learning_rate": 0.0001,
+      "loss": 1.3889,
+      "loss/crossentropy": 2.657388687133789,
+      "loss/hidden": 0.9765625,
+      "loss/logits": 0.24190323054790497,
+      "loss/reg": 0.01704811304807663,
+      "step": 1168
+    },
+    {
+      "epoch": 0.146125,
+      "grad_norm": 3.8508963584899902,
+      "grad_norm_var": 5.829627947687252,
+      "learning_rate": 0.0001,
+      "loss": 1.2453,
+      "loss/crossentropy": 2.405853509902954,
+      "loss/hidden": 0.90234375,
+      "loss/logits": 0.17257630825042725,
+      "loss/reg": 0.017038943246006966,
+      "step": 1169
+    },
+    {
+      "epoch": 0.14625,
+      "grad_norm": 13.325060844421387,
+      "grad_norm_var": 9.23016466799555,
+      "learning_rate": 0.0001,
+      "loss": 1.3331,
+      "loss/crossentropy": 2.4716832637786865,
+      "loss/hidden": 0.94140625,
+      "loss/logits": 0.2213602215051651,
+      "loss/reg": 0.01702967844903469,
+      "step": 1170
+    },
+    {
+      "epoch": 0.146375,
+      "grad_norm": 6.264571189880371,
+      "grad_norm_var": 9.228622011623397,
+      "learning_rate": 0.0001,
+      "loss": 1.1745,
+      "loss/crossentropy": 2.685593605041504,
+      "loss/hidden": 0.83984375,
+      "loss/logits": 0.16446860134601593,
+      "loss/reg": 0.017020443454384804,
+      "step": 1171
+    },
+    {
+      "epoch": 0.1465,
+      "grad_norm": 4.161008358001709,
+      "grad_norm_var": 9.25070050922357,
+      "learning_rate": 0.0001,
+      "loss": 1.2714,
+      "loss/crossentropy": 2.329845428466797,
+      "loss/hidden": 0.91796875,
+      "loss/logits": 0.18332120776176453,
+      "loss/reg": 0.01701117865741253,
+      "step": 1172
+    },
+    {
+      "epoch": 0.146625,
+      "grad_norm": 2.7158303260803223,
+      "grad_norm_var": 9.516487065581881,
+      "learning_rate": 0.0001,
+      "loss": 1.1028,
+      "loss/crossentropy": 2.3517086505889893,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.11249354481697083,
+      "loss/reg": 0.01700259931385517,
+      "step": 1173
+    },
+    {
+      "epoch": 0.14675,
+      "grad_norm": 4.040686130523682,
+      "grad_norm_var": 9.298921738225228,
+      "learning_rate": 0.0001,
+      "loss": 1.4656,
+      "loss/crossentropy": 1.98357093334198,
+      "loss/hidden": 1.1328125,
+      "loss/logits": 0.16282862424850464,
+      "loss/reg": 0.01699363812804222,
+      "step": 1174
+    },
+    {
+      "epoch": 0.146875,
+      "grad_norm": 3.484989643096924,
+      "grad_norm_var": 9.038196456147912,
+      "learning_rate": 0.0001,
+      "loss": 1.1281,
+      "loss/crossentropy": 2.7924225330352783,
+      "loss/hidden": 0.83984375,
+      "loss/logits": 0.11844426393508911,
+      "loss/reg": 0.016985056921839714,
+      "step": 1175
+    },
+    {
+      "epoch": 0.147,
+      "grad_norm": 2.9667844772338867,
+      "grad_norm_var": 9.146981868389197,
+      "learning_rate": 0.0001,
+      "loss": 1.202,
+      "loss/crossentropy": 2.405456781387329,
+      "loss/hidden": 0.875,
+      "loss/logits": 0.15722399950027466,
+      "loss/reg": 0.01697635091841221,
+      "step": 1176
+    },
+    {
+      "epoch": 0.147125,
+      "grad_norm": 5.164290904998779,
+      "grad_norm_var": 8.996899765603606,
+      "learning_rate": 0.0001,
+      "loss": 1.1927,
+      "loss/crossentropy": 2.6594996452331543,
+      "loss/hidden": 0.85546875,
+      "loss/logits": 0.16758760809898376,
+      "loss/reg": 0.01696733944118023,
+      "step": 1177
+    },
+    {
+      "epoch": 0.14725,
+      "grad_norm": 3.2219250202178955,
+      "grad_norm_var": 8.724323229467464,
+      "learning_rate": 0.0001,
+      "loss": 1.216,
+      "loss/crossentropy": 2.5696029663085938,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.16358155012130737,
+      "loss/reg": 0.016958681866526604,
+      "step": 1178
+    },
+    {
+      "epoch": 0.147375,
+      "grad_norm": 2.848184108734131,
+      "grad_norm_var": 7.7572272221691225,
+      "learning_rate": 0.0001,
+      "loss": 1.0605,
+      "loss/crossentropy": 2.317523241043091,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.1332237273454666,
+      "loss/reg": 0.016949651762843132,
+      "step": 1179
+    },
+    {
+      "epoch": 0.1475,
+      "grad_norm": 5.728630065917969,
+      "grad_norm_var": 7.5993034900551715,
+      "learning_rate": 0.0001,
+      "loss": 1.1138,
+      "loss/crossentropy": 2.771470308303833,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.13583886623382568,
+      "loss/reg": 0.016941089183092117,
+      "step": 1180
+    },
+    {
+      "epoch": 0.147625,
+      "grad_norm": 3.538257122039795,
+      "grad_norm_var": 7.488546256518004,
+      "learning_rate": 0.0001,
+      "loss": 1.0898,
+      "loss/crossentropy": 2.7069919109344482,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.1431659609079361,
+      "loss/reg": 0.01693253219127655,
+      "step": 1181
+    },
+    {
+      "epoch": 0.14775,
+      "grad_norm": 3.1879353523254395,
+      "grad_norm_var": 6.50782164977926,
+      "learning_rate": 0.0001,
+      "loss": 1.0471,
+      "loss/crossentropy": 2.6199276447296143,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.13955026865005493,
+      "loss/reg": 0.01692408137023449,
+      "step": 1182
+    },
+    {
+      "epoch": 0.147875,
+      "grad_norm": 4.523275375366211,
+      "grad_norm_var": 6.489534724108,
+      "learning_rate": 0.0001,
+      "loss": 1.2453,
+      "loss/crossentropy": 2.120800018310547,
+      "loss/hidden": 0.92578125,
+      "loss/logits": 0.1503942906856537,
+      "loss/reg": 0.016915684565901756,
+      "step": 1183
+    },
+    {
+      "epoch": 0.148,
+      "grad_norm": 3.735788345336914,
+      "grad_norm_var": 6.520985106875844,
+      "learning_rate": 0.0001,
+      "loss": 1.0708,
+      "loss/crossentropy": 2.6711180210113525,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.11657284200191498,
+      "loss/reg": 0.016907010227441788,
+      "step": 1184
+    },
+    {
+      "epoch": 0.148125,
+      "grad_norm": 5.269720077514648,
+      "grad_norm_var": 6.515042671209793,
+      "learning_rate": 0.0001,
+      "loss": 1.2068,
+      "loss/crossentropy": 2.2848961353302,
+      "loss/hidden": 0.875,
+      "loss/logits": 0.1628110110759735,
+      "loss/reg": 0.01689821295440197,
+      "step": 1185
+    },
+    {
+      "epoch": 0.14825,
+      "grad_norm": 3.4472532272338867,
+      "grad_norm_var": 1.169463016821809,
+      "learning_rate": 0.0001,
+      "loss": 1.0442,
+      "loss/crossentropy": 2.6003897190093994,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.13315384089946747,
+      "loss/reg": 0.016889235004782677,
+      "step": 1186
+    },
+    {
+      "epoch": 0.148375,
+      "grad_norm": 3.267479658126831,
+      "grad_norm_var": 0.8333935781311614,
+      "learning_rate": 0.0001,
+      "loss": 1.135,
+      "loss/crossentropy": 2.3998661041259766,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.1341383457183838,
+      "loss/reg": 0.01688062585890293,
+      "step": 1187
+    },
+    {
+      "epoch": 0.1485,
+      "grad_norm": 3.262617826461792,
+      "grad_norm_var": 0.8443526957342724,
+      "learning_rate": 0.0001,
+      "loss": 1.0085,
+      "loss/crossentropy": 2.5633139610290527,
+      "loss/hidden": 0.72265625,
+      "loss/logits": 0.11711567640304565,
+      "loss/reg": 0.016872059553861618,
+      "step": 1188
+    },
+    {
+      "epoch": 0.148625,
+      "grad_norm": 4.104423522949219,
+      "grad_norm_var": 0.7687216542662267,
+      "learning_rate": 0.0001,
+      "loss": 1.2662,
+      "loss/crossentropy": 2.4028804302215576,
+      "loss/hidden": 0.9296875,
+      "loss/logits": 0.16792353987693787,
+      "loss/reg": 0.016863279044628143,
+      "step": 1189
+    },
+    {
+      "epoch": 0.14875,
+      "grad_norm": 3.4790539741516113,
+      "grad_norm_var": 0.7750564154152146,
+      "learning_rate": 0.0001,
+      "loss": 1.3253,
+      "loss/crossentropy": 2.6870651245117188,
+      "loss/hidden": 0.9140625,
+      "loss/logits": 0.24266795814037323,
+      "loss/reg": 0.016854623332619667,
+      "step": 1190
+    },
+    {
+      "epoch": 0.148875,
+      "grad_norm": 3.920074701309204,
+      "grad_norm_var": 0.7670521683778655,
+      "learning_rate": 0.0001,
+      "loss": 1.7384,
+      "loss/crossentropy": 2.006319761276245,
+      "loss/hidden": 1.296875,
+      "loss/logits": 0.27303898334503174,
+      "loss/reg": 0.016845691949129105,
+      "step": 1191
+    },
+    {
+      "epoch": 0.149,
+      "grad_norm": 3.1460444927215576,
+      "grad_norm_var": 0.7478523869744016,
+      "learning_rate": 0.0001,
+      "loss": 1.1471,
+      "loss/crossentropy": 2.322908401489258,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.15057553350925446,
+      "loss/reg": 0.016836855560541153,
+      "step": 1192
+    },
+    {
+      "epoch": 0.149125,
+      "grad_norm": 2.8855838775634766,
+      "grad_norm_var": 0.6777176205160399,
+      "learning_rate": 0.0001,
+      "loss": 1.1942,
+      "loss/crossentropy": 2.3052327632904053,
+      "loss/hidden": 0.859375,
+      "loss/logits": 0.16649743914604187,
+      "loss/reg": 0.016828451305627823,
+      "step": 1193
+    },
+    {
+      "epoch": 0.14925,
+      "grad_norm": 5.441497325897217,
+      "grad_norm_var": 0.837366755929357,
+      "learning_rate": 0.0001,
+      "loss": 1.2583,
+      "loss/crossentropy": 2.8180274963378906,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.1799638569355011,
+      "loss/reg": 0.016819985583424568,
+      "step": 1194
+    },
+    {
+      "epoch": 0.149375,
+      "grad_norm": 2.6638519763946533,
+      "grad_norm_var": 0.864398086647378,
+      "learning_rate": 0.0001,
+      "loss": 1.2269,
+      "loss/crossentropy": 2.1777124404907227,
+      "loss/hidden": 0.8984375,
+      "loss/logits": 0.1603836864233017,
+      "loss/reg": 0.016811655834317207,
+      "step": 1195
+    },
+    {
+      "epoch": 0.1495,
+      "grad_norm": 3.1290552616119385,
+      "grad_norm_var": 0.6356402025764699,
+      "learning_rate": 0.0001,
+      "loss": 1.063,
+      "loss/crossentropy": 2.31772518157959,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.13322719931602478,
+      "loss/reg": 0.016803618520498276,
+      "step": 1196
+    },
+    {
+      "epoch": 0.149625,
+      "grad_norm": 2.8640289306640625,
+      "grad_norm_var": 0.6774789250719541,
+      "learning_rate": 0.0001,
+      "loss": 1.2119,
+      "loss/crossentropy": 2.603073835372925,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.16111129522323608,
+      "loss/reg": 0.016795063391327858,
+      "step": 1197
+    },
+    {
+      "epoch": 0.14975,
+      "grad_norm": 2.704012155532837,
+      "grad_norm_var": 0.7216374904878362,
+      "learning_rate": 0.0001,
+      "loss": 1.0581,
+      "loss/crossentropy": 2.5079526901245117,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.12854108214378357,
+      "loss/reg": 0.016786765307188034,
+      "step": 1198
+    },
+    {
+      "epoch": 0.149875,
+      "grad_norm": 3.260575532913208,
+      "grad_norm_var": 0.6684105203930727,
+      "learning_rate": 0.0001,
+      "loss": 1.1244,
+      "loss/crossentropy": 2.43037486076355,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.14800330996513367,
+      "loss/reg": 0.01677859202027321,
+      "step": 1199
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.9981093406677246,
+      "grad_norm_var": 0.6828016535629103,
+      "learning_rate": 0.0001,
+      "loss": 1.4566,
+      "loss/crossentropy": 2.5041277408599854,
+      "loss/hidden": 1.0625,
+      "loss/logits": 0.2263844907283783,
+      "loss/reg": 0.016770560294389725,
+      "step": 1200
+    },
+    {
+      "epoch": 0.150125,
+      "grad_norm": 3.0788891315460205,
+      "grad_norm_var": 0.4629717181381217,
+      "learning_rate": 0.0001,
+      "loss": 1.085,
+      "loss/crossentropy": 2.586244821548462,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.1478727161884308,
+      "loss/reg": 0.016761835664510727,
+      "step": 1201
+    },
+    {
+      "epoch": 0.15025,
+      "grad_norm": 2.979407787322998,
+      "grad_norm_var": 0.47078996164599,
+      "learning_rate": 0.0001,
+      "loss": 1.1119,
+      "loss/crossentropy": 2.566535472869873,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.1514027863740921,
+      "loss/reg": 0.016753433272242546,
+      "step": 1202
+    },
+    {
+      "epoch": 0.150375,
+      "grad_norm": 3.7025930881500244,
+      "grad_norm_var": 0.4793410999759106,
+      "learning_rate": 0.0001,
+      "loss": 1.1125,
+      "loss/crossentropy": 2.696070909500122,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.14427737891674042,
+      "loss/reg": 0.016745014116168022,
+      "step": 1203
+    },
+    {
+      "epoch": 0.1505,
+      "grad_norm": 3.7526400089263916,
+      "grad_norm_var": 0.48855855062033865,
+      "learning_rate": 0.0001,
+      "loss": 1.4721,
+      "loss/crossentropy": 2.3625986576080322,
+      "loss/hidden": 1.0859375,
+      "loss/logits": 0.21884310245513916,
+      "loss/reg": 0.016736432909965515,
+      "step": 1204
+    },
+    {
+      "epoch": 0.150625,
+      "grad_norm": 2.6353213787078857,
+      "grad_norm_var": 0.4819149135234577,
+      "learning_rate": 0.0001,
+      "loss": 1.0125,
+      "loss/crossentropy": 2.5642147064208984,
+      "loss/hidden": 0.72265625,
+      "loss/logits": 0.1225152313709259,
+      "loss/reg": 0.01672798953950405,
+      "step": 1205
+    },
+    {
+      "epoch": 0.15075,
+      "grad_norm": 2.7348835468292236,
+      "grad_norm_var": 0.4977728974757267,
+      "learning_rate": 0.0001,
+      "loss": 1.0998,
+      "loss/crossentropy": 2.2675602436065674,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.13179215788841248,
+      "loss/reg": 0.016719957813620567,
+      "step": 1206
+    },
+    {
+      "epoch": 0.150875,
+      "grad_norm": 2.585419178009033,
+      "grad_norm_var": 0.4887115845786629,
+      "learning_rate": 0.0001,
+      "loss": 1.0439,
+      "loss/crossentropy": 2.3835902214050293,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.13845473527908325,
+      "loss/reg": 0.01671142503619194,
+      "step": 1207
+    },
+    {
+      "epoch": 0.151,
+      "grad_norm": 2.457369804382324,
+      "grad_norm_var": 0.519646055542415,
+      "learning_rate": 0.0001,
+      "loss": 1.0538,
+      "loss/crossentropy": 2.719109296798706,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.129006028175354,
+      "loss/reg": 0.01670280657708645,
+      "step": 1208
+    },
+    {
+      "epoch": 0.151125,
+      "grad_norm": 2.794466733932495,
+      "grad_norm_var": 0.5229773551564895,
+      "learning_rate": 0.0001,
+      "loss": 1.1287,
+      "loss/crossentropy": 2.4790573120117188,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.1414172500371933,
+      "loss/reg": 0.016694119200110435,
+      "step": 1209
+    },
+    {
+      "epoch": 0.15125,
+      "grad_norm": 2.713892698287964,
+      "grad_norm_var": 0.14054897219938098,
+      "learning_rate": 0.0001,
+      "loss": 1.1539,
+      "loss/crossentropy": 2.6969070434570312,
+      "loss/hidden": 0.84765625,
+      "loss/logits": 0.1394297033548355,
+      "loss/reg": 0.01668514870107174,
+      "step": 1210
+    },
+    {
+      "epoch": 0.151375,
+      "grad_norm": 3.024172782897949,
+      "grad_norm_var": 0.13535290931035035,
+      "learning_rate": 0.0001,
+      "loss": 0.9908,
+      "loss/crossentropy": 2.757359027862549,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.1209060549736023,
+      "loss/reg": 0.016675440594553947,
+      "step": 1211
+    },
+    {
+      "epoch": 0.1515,
+      "grad_norm": 2.4915895462036133,
+      "grad_norm_var": 0.14667295132687527,
+      "learning_rate": 0.0001,
+      "loss": 1.0311,
+      "loss/crossentropy": 2.311927080154419,
+      "loss/hidden": 0.734375,
+      "loss/logits": 0.13000784814357758,
+      "loss/reg": 0.01666680909693241,
+      "step": 1212
+    },
+    {
+      "epoch": 0.151625,
+      "grad_norm": 3.0987884998321533,
+      "grad_norm_var": 0.1482532510455627,
+      "learning_rate": 0.0001,
+      "loss": 0.9663,
+      "loss/crossentropy": 2.665640354156494,
+      "loss/hidden": 0.68359375,
+      "loss/logits": 0.11612722277641296,
+      "loss/reg": 0.016658229753375053,
+      "step": 1213
+    },
+    {
+      "epoch": 0.15175,
+      "grad_norm": 3.144266128540039,
+      "grad_norm_var": 0.1466168566420831,
+      "learning_rate": 0.0001,
+      "loss": 1.2371,
+      "loss/crossentropy": 2.5783543586730957,
+      "loss/hidden": 0.9140625,
+      "loss/logits": 0.1565355509519577,
+      "loss/reg": 0.01664978824555874,
+      "step": 1214
+    },
+    {
+      "epoch": 0.151875,
+      "grad_norm": 2.5145692825317383,
+      "grad_norm_var": 0.15207652538272973,
+      "learning_rate": 0.0001,
+      "loss": 0.9714,
+      "loss/crossentropy": 2.4898011684417725,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.12530556321144104,
+      "loss/reg": 0.01664099656045437,
+      "step": 1215
+    },
+    {
+      "epoch": 0.152,
+      "grad_norm": 2.5222647190093994,
+      "grad_norm_var": 0.16121854801189647,
+      "learning_rate": 0.0001,
+      "loss": 1.073,
+      "loss/crossentropy": 2.417036533355713,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.14500710368156433,
+      "loss/reg": 0.01663181744515896,
+      "step": 1216
+    },
+    {
+      "epoch": 0.152125,
+      "grad_norm": 2.7117919921875,
+      "grad_norm_var": 0.16036668917481525,
+      "learning_rate": 0.0001,
+      "loss": 1.0801,
+      "loss/crossentropy": 2.658942461013794,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.14434798061847687,
+      "loss/reg": 0.016622193157672882,
+      "step": 1217
+    },
+    {
+      "epoch": 0.15225,
+      "grad_norm": 2.9744222164154053,
+      "grad_norm_var": 0.16029316464901414,
+      "learning_rate": 0.0001,
+      "loss": 1.1878,
+      "loss/crossentropy": 2.491722822189331,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.17787563800811768,
+      "loss/reg": 0.016612496227025986,
+      "step": 1218
+    },
+    {
+      "epoch": 0.152375,
+      "grad_norm": 3.3779635429382324,
+      "grad_norm_var": 0.1306752736974545,
+      "learning_rate": 0.0001,
+      "loss": 1.0725,
+      "loss/crossentropy": 2.4862918853759766,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.14869090914726257,
+      "loss/reg": 0.01660403423011303,
+      "step": 1219
+    },
+    {
+      "epoch": 0.1525,
+      "grad_norm": 2.5530524253845215,
+      "grad_norm_var": 0.07557910361394207,
+      "learning_rate": 0.0001,
+      "loss": 1.0191,
+      "loss/crossentropy": 2.614441394805908,
+      "loss/hidden": 0.72265625,
+      "loss/logits": 0.13050062954425812,
+      "loss/reg": 0.016595516353845596,
+      "step": 1220
+    },
+    {
+      "epoch": 0.152625,
+      "grad_norm": 2.378697156906128,
+      "grad_norm_var": 0.08433378351046841,
+      "learning_rate": 0.0001,
+      "loss": 1.0027,
+      "loss/crossentropy": 2.529618501663208,
+      "loss/hidden": 0.72265625,
+      "loss/logits": 0.11422193050384521,
+      "loss/reg": 0.016587061807513237,
+      "step": 1221
+    },
+    {
+      "epoch": 0.15275,
+      "grad_norm": 3.0336711406707764,
+      "grad_norm_var": 0.08911795415122749,
+      "learning_rate": 0.0001,
+      "loss": 1.0727,
+      "loss/crossentropy": 2.5050387382507324,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.15298862755298615,
+      "loss/reg": 0.016578199341893196,
+      "step": 1222
+    },
+    {
+      "epoch": 0.152875,
+      "grad_norm": 5.338172435760498,
+      "grad_norm_var": 0.4936799710714484,
+      "learning_rate": 0.0001,
+      "loss": 1.1124,
+      "loss/crossentropy": 2.6853761672973633,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.13811737298965454,
+      "loss/reg": 0.016569815576076508,
+      "step": 1223
+    },
+    {
+      "epoch": 0.153,
+      "grad_norm": 2.962609052658081,
+      "grad_norm_var": 0.47674628875148056,
+      "learning_rate": 0.0001,
+      "loss": 1.0689,
+      "loss/crossentropy": 2.441126585006714,
+      "loss/hidden": 0.765625,
+      "loss/logits": 0.13769929111003876,
+      "loss/reg": 0.016561318188905716,
+      "step": 1224
+    },
+    {
+      "epoch": 0.153125,
+      "grad_norm": 2.7154529094696045,
+      "grad_norm_var": 0.4790610818976868,
+      "learning_rate": 0.0001,
+      "loss": 1.1047,
+      "loss/crossentropy": 2.4226393699645996,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.14224863052368164,
+      "loss/reg": 0.016552859917283058,
+      "step": 1225
+    },
+    {
+      "epoch": 0.15325,
+      "grad_norm": 3.163189172744751,
+      "grad_norm_var": 0.4762029205706464,
+      "learning_rate": 0.0001,
+      "loss": 1.1356,
+      "loss/crossentropy": 2.642491340637207,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.1537851244211197,
+      "loss/reg": 0.016544297337532043,
+      "step": 1226
+    },
+    {
+      "epoch": 0.153375,
+      "grad_norm": 8.438536643981934,
+      "grad_norm_var": 2.3256512762465564,
+      "learning_rate": 0.0001,
+      "loss": 1.4483,
+      "loss/crossentropy": 2.159141778945923,
+      "loss/hidden": 1.0703125,
+      "loss/logits": 0.21258942782878876,
+      "loss/reg": 0.016535377129912376,
+      "step": 1227
+    },
+    {
+      "epoch": 0.1535,
+      "grad_norm": 16.965858459472656,
+      "grad_norm_var": 13.784859138236738,
+      "learning_rate": 0.0001,
+      "loss": 3.1548,
+      "loss/crossentropy": 2.990753173828125,
+      "loss/hidden": 1.8203125,
+      "loss/logits": 1.1692469120025635,
+      "loss/reg": 0.016526464372873306,
+      "step": 1228
+    },
+    {
+      "epoch": 0.153625,
+      "grad_norm": 3.789302110671997,
+      "grad_norm_var": 13.709283357450609,
+      "learning_rate": 0.0001,
+      "loss": 1.0345,
+      "loss/crossentropy": 2.7094292640686035,
+      "loss/hidden": 0.7265625,
+      "loss/logits": 0.1427900195121765,
+      "loss/reg": 0.016517426818609238,
+      "step": 1229
+    },
+    {
+      "epoch": 0.15375,
+      "grad_norm": 2.838264226913452,
+      "grad_norm_var": 13.761738651197222,
+      "learning_rate": 0.0001,
+      "loss": 1.1093,
+      "loss/crossentropy": 2.543126106262207,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.1551959365606308,
+      "loss/reg": 0.01650911010801792,
+      "step": 1230
+    },
+    {
+      "epoch": 0.153875,
+      "grad_norm": 3.218651533126831,
+      "grad_norm_var": 13.628173707948106,
+      "learning_rate": 0.0001,
+      "loss": 1.0906,
+      "loss/crossentropy": 2.112255096435547,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.1404722034931183,
+      "loss/reg": 0.016500860452651978,
+      "step": 1231
+    },
+    {
+      "epoch": 0.154,
+      "grad_norm": 4.796919345855713,
+      "grad_norm_var": 13.40893956577282,
+      "learning_rate": 0.0001,
+      "loss": 1.2598,
+      "loss/crossentropy": 2.6373000144958496,
+      "loss/hidden": 0.93359375,
+      "loss/logits": 0.16131797432899475,
+      "loss/reg": 0.01649186760187149,
+      "step": 1232
+    },
+    {
+      "epoch": 0.154125,
+      "grad_norm": 3.640801191329956,
+      "grad_norm_var": 13.24713470324538,
+      "learning_rate": 0.0001,
+      "loss": 1.4181,
+      "loss/crossentropy": 2.4901435375213623,
+      "loss/hidden": 1.03125,
+      "loss/logits": 0.2219938188791275,
+      "loss/reg": 0.01648273505270481,
+      "step": 1233
+    },
+    {
+      "epoch": 0.15425,
+      "grad_norm": 3.2018325328826904,
+      "grad_norm_var": 13.203757643215535,
+      "learning_rate": 0.0001,
+      "loss": 1.1419,
+      "loss/crossentropy": 2.504349708557129,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.1725234091281891,
+      "loss/reg": 0.016473697498440742,
+      "step": 1234
+    },
+    {
+      "epoch": 0.154375,
+      "grad_norm": 2.982257127761841,
+      "grad_norm_var": 13.274105522819255,
+      "learning_rate": 0.0001,
+      "loss": 1.0759,
+      "loss/crossentropy": 2.3660614490509033,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.14176063239574432,
+      "loss/reg": 0.01646505668759346,
+      "step": 1235
+    },
+    {
+      "epoch": 0.1545,
+      "grad_norm": 4.535284519195557,
+      "grad_norm_var": 13.004824447407662,
+      "learning_rate": 0.0001,
+      "loss": 1.4385,
+      "loss/crossentropy": 2.8622822761535645,
+      "loss/hidden": 1.0,
+      "loss/logits": 0.2738988697528839,
+      "loss/reg": 0.016456691548228264,
+      "step": 1236
+    },
+    {
+      "epoch": 0.154625,
+      "grad_norm": 3.0818183422088623,
+      "grad_norm_var": 12.825136343225884,
+      "learning_rate": 0.0001,
+      "loss": 1.0988,
+      "loss/crossentropy": 2.449317693710327,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.14135484397411346,
+      "loss/reg": 0.01644827052950859,
+      "step": 1237
+    },
+    {
+      "epoch": 0.15475,
+      "grad_norm": 4.102293491363525,
+      "grad_norm_var": 12.663514204467905,
+      "learning_rate": 0.0001,
+      "loss": 1.1842,
+      "loss/crossentropy": 2.7557554244995117,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.16828957200050354,
+      "loss/reg": 0.01643945835530758,
+      "step": 1238
+    },
+    {
+      "epoch": 0.154875,
+      "grad_norm": 4.042688369750977,
+      "grad_norm_var": 12.664341312944845,
+      "learning_rate": 0.0001,
+      "loss": 1.2182,
+      "loss/crossentropy": 2.4376473426818848,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.17108118534088135,
+      "loss/reg": 0.016430867835879326,
+      "step": 1239
+    },
+    {
+      "epoch": 0.155,
+      "grad_norm": 3.330106735229492,
+      "grad_norm_var": 12.589868576516837,
+      "learning_rate": 0.0001,
+      "loss": 1.1139,
+      "loss/crossentropy": 2.705000638961792,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.14893847703933716,
+      "loss/reg": 0.01642204262316227,
+      "step": 1240
+    },
+    {
+      "epoch": 0.155125,
+      "grad_norm": 2.7572247982025146,
+      "grad_norm_var": 12.579048710159194,
+      "learning_rate": 0.0001,
+      "loss": 1.0319,
+      "loss/crossentropy": 2.726616859436035,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.12947078049182892,
+      "loss/reg": 0.016413651406764984,
+      "step": 1241
+    },
+    {
+      "epoch": 0.15525,
+      "grad_norm": 2.501053810119629,
+      "grad_norm_var": 12.740389120966254,
+      "learning_rate": 0.0001,
+      "loss": 1.0246,
+      "loss/crossentropy": 2.1374268531799316,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.11834150552749634,
+      "loss/reg": 0.016405310481786728,
+      "step": 1242
+    },
+    {
+      "epoch": 0.155375,
+      "grad_norm": 3.1334445476531982,
+      "grad_norm_var": 11.811754750464528,
+      "learning_rate": 0.0001,
+      "loss": 1.1323,
+      "loss/crossentropy": 2.204963445663452,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.14021529257297516,
+      "loss/reg": 0.01639643684029579,
+      "step": 1243
+    },
+    {
+      "epoch": 0.1555,
+      "grad_norm": 2.888942003250122,
+      "grad_norm_var": 0.43771643055907344,
+      "learning_rate": 0.0001,
+      "loss": 1.0436,
+      "loss/crossentropy": 2.6868066787719727,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.12971553206443787,
+      "loss/reg": 0.016387417912483215,
+      "step": 1244
+    },
+    {
+      "epoch": 0.155625,
+      "grad_norm": 3.695841073989868,
+      "grad_norm_var": 0.43375446821376706,
+      "learning_rate": 0.0001,
+      "loss": 1.1949,
+      "loss/crossentropy": 2.679568290710449,
+      "loss/hidden": 0.8671875,
+      "loss/logits": 0.16394542157649994,
+      "loss/reg": 0.01637819968163967,
+      "step": 1245
+    },
+    {
+      "epoch": 0.15575,
+      "grad_norm": 3.01936411857605,
+      "grad_norm_var": 0.4217159331567084,
+      "learning_rate": 0.0001,
+      "loss": 1.2859,
+      "loss/crossentropy": 2.3360393047332764,
+      "loss/hidden": 0.921875,
+      "loss/logits": 0.20038416981697083,
+      "loss/reg": 0.01636892557144165,
+      "step": 1246
+    },
+    {
+      "epoch": 0.155875,
+      "grad_norm": 2.5124776363372803,
+      "grad_norm_var": 0.47306891797669076,
+      "learning_rate": 0.0001,
+      "loss": 0.9599,
+      "loss/crossentropy": 2.558067560195923,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.11665983498096466,
+      "loss/reg": 0.016359377652406693,
+      "step": 1247
+    },
+    {
+      "epoch": 0.156,
+      "grad_norm": 3.540679693222046,
+      "grad_norm_var": 0.3358607220080972,
+      "learning_rate": 0.0001,
+      "loss": 1.1319,
+      "loss/crossentropy": 2.760929584503174,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.1480470895767212,
+      "loss/reg": 0.01635100692510605,
+      "step": 1248
+    },
+    {
+      "epoch": 0.156125,
+      "grad_norm": 3.4042372703552246,
+      "grad_norm_var": 0.3289363389964431,
+      "learning_rate": 0.0001,
+      "loss": 1.1718,
+      "loss/crossentropy": 2.771622896194458,
+      "loss/hidden": 0.859375,
+      "loss/logits": 0.14904874563217163,
+      "loss/reg": 0.01634254865348339,
+      "step": 1249
+    },
+    {
+      "epoch": 0.15625,
+      "grad_norm": 3.164729118347168,
+      "grad_norm_var": 0.32948624287587125,
+      "learning_rate": 0.0001,
+      "loss": 1.2872,
+      "loss/crossentropy": 2.3246641159057617,
+      "loss/hidden": 0.93359375,
+      "loss/logits": 0.19026914238929749,
+      "loss/reg": 0.016334179788827896,
+      "step": 1250
+    },
+    {
+      "epoch": 0.156375,
+      "grad_norm": 3.409289598464966,
+      "grad_norm_var": 0.32317475604933343,
+      "learning_rate": 0.0001,
+      "loss": 1.0262,
+      "loss/crossentropy": 2.498500347137451,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.12468535453081131,
+      "loss/reg": 0.01632508635520935,
+      "step": 1251
+    },
+    {
+      "epoch": 0.1565,
+      "grad_norm": 7.274641990661621,
+      "grad_norm_var": 1.2360715279843908,
+      "learning_rate": 0.0001,
+      "loss": 1.5887,
+      "loss/crossentropy": 2.7439510822296143,
+      "loss/hidden": 1.2734375,
+      "loss/logits": 0.15212371945381165,
+      "loss/reg": 0.016316639259457588,
+      "step": 1252
+    },
+    {
+      "epoch": 0.156625,
+      "grad_norm": 3.9700396060943604,
+      "grad_norm_var": 1.2368999449904527,
+      "learning_rate": 0.0001,
+      "loss": 1.3172,
+      "loss/crossentropy": 2.9382898807525635,
+      "loss/hidden": 0.99609375,
+      "loss/logits": 0.15806305408477783,
+      "loss/reg": 0.016308104619383812,
+      "step": 1253
+    },
+    {
+      "epoch": 0.15675,
+      "grad_norm": 3.035047769546509,
+      "grad_norm_var": 1.2290263478015266,
+      "learning_rate": 0.0001,
+      "loss": 1.0668,
+      "loss/crossentropy": 2.486111879348755,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.15766200423240662,
+      "loss/reg": 0.016299735754728317,
+      "step": 1254
+    },
+    {
+      "epoch": 0.156875,
+      "grad_norm": 4.703797340393066,
+      "grad_norm_var": 1.30594374893626,
+      "learning_rate": 0.0001,
+      "loss": 1.1914,
+      "loss/crossentropy": 2.6573047637939453,
+      "loss/hidden": 0.87890625,
+      "loss/logits": 0.14956605434417725,
+      "loss/reg": 0.0162909384816885,
+      "step": 1255
+    },
+    {
+      "epoch": 0.157,
+      "grad_norm": 2.9516775608062744,
+      "grad_norm_var": 1.3245417395020618,
+      "learning_rate": 0.0001,
+      "loss": 1.1959,
+      "loss/crossentropy": 2.3165104389190674,
+      "loss/hidden": 0.87890625,
+      "loss/logits": 0.1542087197303772,
+      "loss/reg": 0.01628235913813114,
+      "step": 1256
+    },
+    {
+      "epoch": 0.157125,
+      "grad_norm": 2.8121533393859863,
+      "grad_norm_var": 1.3193075406315065,
+      "learning_rate": 0.0001,
+      "loss": 1.2046,
+      "loss/crossentropy": 2.269421100616455,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.15904302895069122,
+      "loss/reg": 0.01627359353005886,
+      "step": 1257
+    },
+    {
+      "epoch": 0.15725,
+      "grad_norm": 2.7715342044830322,
+      "grad_norm_var": 1.287814713649868,
+      "learning_rate": 0.0001,
+      "loss": 1.0294,
+      "loss/crossentropy": 2.6235570907592773,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.15194407105445862,
+      "loss/reg": 0.01626538671553135,
+      "step": 1258
+    },
+    {
+      "epoch": 0.157375,
+      "grad_norm": 1.9729806184768677,
+      "grad_norm_var": 1.4314826970209784,
+      "learning_rate": 0.0001,
+      "loss": 0.9952,
+      "loss/crossentropy": 2.2795357704162598,
+      "loss/hidden": 0.71875,
+      "loss/logits": 0.11390725523233414,
+      "loss/reg": 0.016256939619779587,
+      "step": 1259
+    },
+    {
+      "epoch": 0.1575,
+      "grad_norm": 4.472729206085205,
+      "grad_norm_var": 1.4707347924489687,
+      "learning_rate": 0.0001,
+      "loss": 1.5205,
+      "loss/crossentropy": 2.3265631198883057,
+      "loss/hidden": 1.1328125,
+      "loss/logits": 0.22519487142562866,
+      "loss/reg": 0.016248464584350586,
+      "step": 1260
+    },
+    {
+      "epoch": 0.157625,
+      "grad_norm": 2.59954571723938,
+      "grad_norm_var": 1.5237222016228462,
+      "learning_rate": 0.0001,
+      "loss": 0.9663,
+      "loss/crossentropy": 2.6342809200286865,
+      "loss/hidden": 0.671875,
+      "loss/logits": 0.13200603425502777,
+      "loss/reg": 0.01624012365937233,
+      "step": 1261
+    },
+    {
+      "epoch": 0.15775,
+      "grad_norm": 3.7313549518585205,
+      "grad_norm_var": 1.5120623570669227,
+      "learning_rate": 0.0001,
+      "loss": 1.1542,
+      "loss/crossentropy": 2.5860178470611572,
+      "loss/hidden": 0.83984375,
+      "loss/logits": 0.15201479196548462,
+      "loss/reg": 0.016231315210461617,
+      "step": 1262
+    },
+    {
+      "epoch": 0.157875,
+      "grad_norm": 12.554633140563965,
+      "grad_norm_var": 6.465262907364731,
+      "learning_rate": 0.0001,
+      "loss": 1.0956,
+      "loss/crossentropy": 2.696977376937866,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.13646942377090454,
+      "loss/reg": 0.016222581267356873,
+      "step": 1263
+    },
+    {
+      "epoch": 0.158,
+      "grad_norm": 3.071765422821045,
+      "grad_norm_var": 6.516980451118831,
+      "learning_rate": 0.0001,
+      "loss": 0.9936,
+      "loss/crossentropy": 2.407457113265991,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.1166088730096817,
+      "loss/reg": 0.01621370017528534,
+      "step": 1264
+    },
+    {
+      "epoch": 0.158125,
+      "grad_norm": 3.9861507415771484,
+      "grad_norm_var": 6.482705701616392,
+      "learning_rate": 0.0001,
+      "loss": 1.2125,
+      "loss/crossentropy": 2.4362969398498535,
+      "loss/hidden": 0.89453125,
+      "loss/logits": 0.15587326884269714,
+      "loss/reg": 0.01620490849018097,
+      "step": 1265
+    },
+    {
+      "epoch": 0.15825,
+      "grad_norm": 4.503188610076904,
+      "grad_norm_var": 6.417924727941712,
+      "learning_rate": 0.0001,
+      "loss": 1.1854,
+      "loss/crossentropy": 2.550220012664795,
+      "loss/hidden": 0.875,
+      "loss/logits": 0.14846490323543549,
+      "loss/reg": 0.016196196898818016,
+      "step": 1266
+    },
+    {
+      "epoch": 0.158375,
+      "grad_norm": 3.923295021057129,
+      "grad_norm_var": 6.3775887710365,
+      "learning_rate": 0.0001,
+      "loss": 1.5051,
+      "loss/crossentropy": 2.2857823371887207,
+      "loss/hidden": 1.109375,
+      "loss/logits": 0.23387819528579712,
+      "loss/reg": 0.01618727669119835,
+      "step": 1267
+    },
+    {
+      "epoch": 0.1585,
+      "grad_norm": 2.639031171798706,
+      "grad_norm_var": 5.8640922918796825,
+      "learning_rate": 0.0001,
+      "loss": 1.0417,
+      "loss/crossentropy": 2.541623115539551,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.13770245015621185,
+      "loss/reg": 0.01617906242609024,
+      "step": 1268
+    },
+    {
+      "epoch": 0.158625,
+      "grad_norm": 3.2652931213378906,
+      "grad_norm_var": 5.896181098711848,
+      "learning_rate": 0.0001,
+      "loss": 1.0331,
+      "loss/crossentropy": 2.8000566959381104,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.12530115246772766,
+      "loss/reg": 0.01617092825472355,
+      "step": 1269
+    },
+    {
+      "epoch": 0.15875,
+      "grad_norm": 7.7591872215271,
+      "grad_norm_var": 6.722812290225987,
+      "learning_rate": 0.0001,
+      "loss": 1.3254,
+      "loss/crossentropy": 2.975308895111084,
+      "loss/hidden": 0.9453125,
+      "loss/logits": 0.21848054230213165,
+      "loss/reg": 0.016162917017936707,
+      "step": 1270
+    },
+    {
+      "epoch": 0.158875,
+      "grad_norm": 20.061229705810547,
+      "grad_norm_var": 22.42875378589538,
+      "learning_rate": 0.0001,
+      "loss": 1.5526,
+      "loss/crossentropy": 2.5683257579803467,
+      "loss/hidden": 1.1796875,
+      "loss/logits": 0.21141059696674347,
+      "loss/reg": 0.016154874116182327,
+      "step": 1271
+    },
+    {
+      "epoch": 0.159,
+      "grad_norm": 3.1273183822631836,
+      "grad_norm_var": 22.37821079380713,
+      "learning_rate": 0.0001,
+      "loss": 1.101,
+      "loss/crossentropy": 2.2700870037078857,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.13093486428260803,
+      "loss/reg": 0.01614651456475258,
+      "step": 1272
+    },
+    {
+      "epoch": 0.159125,
+      "grad_norm": 5.0759968757629395,
+      "grad_norm_var": 21.976791517132078,
+      "learning_rate": 0.0001,
+      "loss": 1.342,
+      "loss/crossentropy": 2.440864324569702,
+      "loss/hidden": 1.0234375,
+      "loss/logits": 0.157205730676651,
+      "loss/reg": 0.016138622537255287,
+      "step": 1273
+    },
+    {
+      "epoch": 0.15925,
+      "grad_norm": 3.844823122024536,
+      "grad_norm_var": 21.680554653297506,
+      "learning_rate": 0.0001,
+      "loss": 1.1638,
+      "loss/crossentropy": 2.461087942123413,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.11964771896600723,
+      "loss/reg": 0.016129910945892334,
+      "step": 1274
+    },
+    {
+      "epoch": 0.159375,
+      "grad_norm": 3.4437410831451416,
+      "grad_norm_var": 21.141396790594833,
+      "learning_rate": 0.0001,
+      "loss": 1.1072,
+      "loss/crossentropy": 2.2671289443969727,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.12177357822656631,
+      "loss/reg": 0.01612204499542713,
+      "step": 1275
+    },
+    {
+      "epoch": 0.1595,
+      "grad_norm": 2.6474483013153076,
+      "grad_norm_var": 21.600534383242742,
+      "learning_rate": 0.0001,
+      "loss": 1.0477,
+      "loss/crossentropy": 2.425645112991333,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.1326972246170044,
+      "loss/reg": 0.016113854944705963,
+      "step": 1276
+    },
+    {
+      "epoch": 0.159625,
+      "grad_norm": 3.2777504920959473,
+      "grad_norm_var": 21.37698263032081,
+      "learning_rate": 0.0001,
+      "loss": 1.1394,
+      "loss/crossentropy": 2.5559988021850586,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.16583070158958435,
+      "loss/reg": 0.016105838119983673,
+      "step": 1277
+    },
+    {
+      "epoch": 0.15975,
+      "grad_norm": 2.806304693222046,
+      "grad_norm_var": 21.640224221543477,
+      "learning_rate": 0.0001,
+      "loss": 1.2464,
+      "loss/crossentropy": 2.5195603370666504,
+      "loss/hidden": 0.90625,
+      "loss/logits": 0.17914444208145142,
+      "loss/reg": 0.016097450628876686,
+      "step": 1278
+    },
+    {
+      "epoch": 0.159875,
+      "grad_norm": 2.830876350402832,
+      "grad_norm_var": 18.240248060375865,
+      "learning_rate": 0.0001,
+      "loss": 1.0662,
+      "loss/crossentropy": 2.4745335578918457,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.12796689569950104,
+      "loss/reg": 0.016089415177702904,
+      "step": 1279
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.508394479751587,
+      "grad_norm_var": 18.153502836014415,
+      "learning_rate": 0.0001,
+      "loss": 1.344,
+      "loss/crossentropy": 2.2748706340789795,
+      "loss/hidden": 0.9921875,
+      "loss/logits": 0.19095824658870697,
+      "loss/reg": 0.016081009060144424,
+      "step": 1280
+    },
+    {
+      "epoch": 0.160125,
+      "grad_norm": 2.568115234375,
+      "grad_norm_var": 18.431873650050928,
+      "learning_rate": 0.0001,
+      "loss": 1.1068,
+      "loss/crossentropy": 2.393493413925171,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.14136919379234314,
+      "loss/reg": 0.016072595492005348,
+      "step": 1281
+    },
+    {
+      "epoch": 0.16025,
+      "grad_norm": 2.587556838989258,
+      "grad_norm_var": 18.712804471683103,
+      "learning_rate": 0.0001,
+      "loss": 1.0643,
+      "loss/crossentropy": 2.619617462158203,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.14193680882453918,
+      "loss/reg": 0.016064899042248726,
+      "step": 1282
+    },
+    {
+      "epoch": 0.160375,
+      "grad_norm": 4.345118522644043,
+      "grad_norm_var": 18.686686687831273,
+      "learning_rate": 0.0001,
+      "loss": 1.087,
+      "loss/crossentropy": 2.4879989624023438,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.14906466007232666,
+      "loss/reg": 0.016057275235652924,
+      "step": 1283
+    },
+    {
+      "epoch": 0.1605,
+      "grad_norm": 3.326270818710327,
+      "grad_norm_var": 18.535440191895614,
+      "learning_rate": 0.0001,
+      "loss": 1.2065,
+      "loss/crossentropy": 2.540086507797241,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.16315226256847382,
+      "loss/reg": 0.01604924537241459,
+      "step": 1284
+    },
+    {
+      "epoch": 0.160625,
+      "grad_norm": 12.10493278503418,
+      "grad_norm_var": 21.781544615658156,
+      "learning_rate": 0.0001,
+      "loss": 1.4967,
+      "loss/crossentropy": 2.0798144340515137,
+      "loss/hidden": 1.140625,
+      "loss/logits": 0.19571265578269958,
+      "loss/reg": 0.016041060909628868,
+      "step": 1285
+    },
+    {
+      "epoch": 0.16075,
+      "grad_norm": 3.7908778190612793,
+      "grad_norm_var": 21.41548096635543,
+      "learning_rate": 0.0001,
+      "loss": 1.2948,
+      "loss/crossentropy": 2.554609775543213,
+      "loss/hidden": 0.95703125,
+      "loss/logits": 0.17741422355175018,
+      "loss/reg": 0.01603337749838829,
+      "step": 1286
+    },
+    {
+      "epoch": 0.160875,
+      "grad_norm": 2.7631125450134277,
+      "grad_norm_var": 5.285413244775635,
+      "learning_rate": 0.0001,
+      "loss": 1.1375,
+      "loss/crossentropy": 2.4389936923980713,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.17257535457611084,
+      "loss/reg": 0.016025440767407417,
+      "step": 1287
+    },
+    {
+      "epoch": 0.161,
+      "grad_norm": 3.0926973819732666,
+      "grad_norm_var": 5.288953588764335,
+      "learning_rate": 0.0001,
+      "loss": 1.1024,
+      "loss/crossentropy": 2.507418394088745,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.12186664342880249,
+      "loss/reg": 0.016017207875847816,
+      "step": 1288
+    },
+    {
+      "epoch": 0.161125,
+      "grad_norm": 3.1981067657470703,
+      "grad_norm_var": 5.208865380747667,
+      "learning_rate": 0.0001,
+      "loss": 1.0863,
+      "loss/crossentropy": 2.4531667232513428,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.13326312601566315,
+      "loss/reg": 0.01600871980190277,
+      "step": 1289
+    },
+    {
+      "epoch": 0.16125,
+      "grad_norm": 2.445035457611084,
+      "grad_norm_var": 5.315218503488533,
+      "learning_rate": 0.0001,
+      "loss": 1.0727,
+      "loss/crossentropy": 2.083653450012207,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.1353735327720642,
+      "loss/reg": 0.016000347211956978,
+      "step": 1290
+    },
+    {
+      "epoch": 0.161375,
+      "grad_norm": 3.601950168609619,
+      "grad_norm_var": 5.3119885145812225,
+      "learning_rate": 0.0001,
+      "loss": 1.3367,
+      "loss/crossentropy": 2.5701797008514404,
+      "loss/hidden": 0.9765625,
+      "loss/logits": 0.20016901195049286,
+      "loss/reg": 0.015991859138011932,
+      "step": 1291
+    },
+    {
+      "epoch": 0.1615,
+      "grad_norm": 2.633711099624634,
+      "grad_norm_var": 5.313893223941083,
+      "learning_rate": 0.0001,
+      "loss": 0.9594,
+      "loss/crossentropy": 2.5569510459899902,
+      "loss/hidden": 0.69140625,
+      "loss/logits": 0.10816207528114319,
+      "loss/reg": 0.015983374789357185,
+      "step": 1292
+    },
+    {
+      "epoch": 0.161625,
+      "grad_norm": 3.0227303504943848,
+      "grad_norm_var": 5.331637216482173,
+      "learning_rate": 0.0001,
+      "loss": 1.0746,
+      "loss/crossentropy": 2.440446138381958,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.1414303183555603,
+      "loss/reg": 0.01597476750612259,
+      "step": 1293
+    },
+    {
+      "epoch": 0.16175,
+      "grad_norm": 3.4858055114746094,
+      "grad_norm_var": 5.282777369926362,
+      "learning_rate": 0.0001,
+      "loss": 1.221,
+      "loss/crossentropy": 2.63417649269104,
+      "loss/hidden": 0.90234375,
+      "loss/logits": 0.15903231501579285,
+      "loss/reg": 0.015966549515724182,
+      "step": 1294
+    },
+    {
+      "epoch": 0.161875,
+      "grad_norm": 2.918198823928833,
+      "grad_norm_var": 5.273058122497643,
+      "learning_rate": 0.0001,
+      "loss": 1.097,
+      "loss/crossentropy": 2.473909854888916,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.14054538309574127,
+      "loss/reg": 0.015958301723003387,
+      "step": 1295
+    },
+    {
+      "epoch": 0.162,
+      "grad_norm": 2.7234513759613037,
+      "grad_norm_var": 5.332879789031237,
+      "learning_rate": 0.0001,
+      "loss": 1.279,
+      "loss/crossentropy": 2.241210460662842,
+      "loss/hidden": 0.9375,
+      "loss/logits": 0.182043194770813,
+      "loss/reg": 0.015949726104736328,
+      "step": 1296
+    },
+    {
+      "epoch": 0.162125,
+      "grad_norm": 3.3515610694885254,
+      "grad_norm_var": 5.256872590146133,
+      "learning_rate": 0.0001,
+      "loss": 1.1463,
+      "loss/crossentropy": 2.6752541065216064,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.15490111708641052,
+      "loss/reg": 0.015941519290208817,
+      "step": 1297
+    },
+    {
+      "epoch": 0.16225,
+      "grad_norm": 3.6638059616088867,
+      "grad_norm_var": 5.167917555355179,
+      "learning_rate": 0.0001,
+      "loss": 1.1893,
+      "loss/crossentropy": 2.2823452949523926,
+      "loss/hidden": 0.86328125,
+      "loss/logits": 0.16667145490646362,
+      "loss/reg": 0.015933820977807045,
+      "step": 1298
+    },
+    {
+      "epoch": 0.162375,
+      "grad_norm": 4.22912073135376,
+      "grad_norm_var": 5.1600059777442855,
+      "learning_rate": 0.0001,
+      "loss": 1.1752,
+      "loss/crossentropy": 2.6061580181121826,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.1916988492012024,
+      "loss/reg": 0.01592625491321087,
+      "step": 1299
+    },
+    {
+      "epoch": 0.1625,
+      "grad_norm": 6.90070915222168,
+      "grad_norm_var": 5.7461320078663105,
+      "learning_rate": 0.0001,
+      "loss": 1.5428,
+      "loss/crossentropy": 1.8802989721298218,
+      "loss/hidden": 1.1015625,
+      "loss/logits": 0.2820611596107483,
+      "loss/reg": 0.015918578952550888,
+      "step": 1300
+    },
+    {
+      "epoch": 0.162625,
+      "grad_norm": 3.0886518955230713,
+      "grad_norm_var": 1.077876623419837,
+      "learning_rate": 0.0001,
+      "loss": 1.2216,
+      "loss/crossentropy": 2.5785911083221436,
+      "loss/hidden": 0.8984375,
+      "loss/logits": 0.16405634582042694,
+      "loss/reg": 0.01591094397008419,
+      "step": 1301
+    },
+    {
+      "epoch": 0.16275,
+      "grad_norm": 3.582451581954956,
+      "grad_norm_var": 1.0706141462880387,
+      "learning_rate": 0.0001,
+      "loss": 1.2677,
+      "loss/crossentropy": 2.2569782733917236,
+      "loss/hidden": 0.92578125,
+      "loss/logits": 0.18291030824184418,
+      "loss/reg": 0.01590333878993988,
+      "step": 1302
+    },
+    {
+      "epoch": 0.162875,
+      "grad_norm": 3.291792392730713,
+      "grad_norm_var": 1.0418618914724893,
+      "learning_rate": 0.0001,
+      "loss": 1.1212,
+      "loss/crossentropy": 2.5871994495391846,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.15368321537971497,
+      "loss/reg": 0.01589547097682953,
+      "step": 1303
+    },
+    {
+      "epoch": 0.163,
+      "grad_norm": 12.274004936218262,
+      "grad_norm_var": 5.870708246927601,
+      "learning_rate": 0.0001,
+      "loss": 1.5367,
+      "loss/crossentropy": 2.2327094078063965,
+      "loss/hidden": 1.1171875,
+      "loss/logits": 0.2606501579284668,
+      "loss/reg": 0.015888074412941933,
+      "step": 1304
+    },
+    {
+      "epoch": 0.163125,
+      "grad_norm": 3.356614112854004,
+      "grad_norm_var": 5.854788067612952,
+      "learning_rate": 0.0001,
+      "loss": 1.1269,
+      "loss/crossentropy": 2.7387585639953613,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.16729386150836945,
+      "loss/reg": 0.015879716724157333,
+      "step": 1305
+    },
+    {
+      "epoch": 0.16325,
+      "grad_norm": 2.274139404296875,
+      "grad_norm_var": 5.892856228313144,
+      "learning_rate": 0.0001,
+      "loss": 1.013,
+      "loss/crossentropy": 2.458984136581421,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.12379375100135803,
+      "loss/reg": 0.015871398150920868,
+      "step": 1306
+    },
+    {
+      "epoch": 0.163375,
+      "grad_norm": 3.3769636154174805,
+      "grad_norm_var": 5.908708209046858,
+      "learning_rate": 0.0001,
+      "loss": 1.096,
+      "loss/crossentropy": 2.4599368572235107,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.1365453004837036,
+      "loss/reg": 0.015863511711359024,
+      "step": 1307
+    },
+    {
+      "epoch": 0.1635,
+      "grad_norm": 5.5954132080078125,
+      "grad_norm_var": 5.91311204762225,
+      "learning_rate": 0.0001,
+      "loss": 1.4203,
+      "loss/crossentropy": 3.034935474395752,
+      "loss/hidden": 1.03125,
+      "loss/logits": 0.23048382997512817,
+      "loss/reg": 0.015855222940444946,
+      "step": 1308
+    },
+    {
+      "epoch": 0.163625,
+      "grad_norm": 2.820237636566162,
+      "grad_norm_var": 5.947350905923982,
+      "learning_rate": 0.0001,
+      "loss": 1.1673,
+      "loss/crossentropy": 2.4714202880859375,
+      "loss/hidden": 0.84765625,
+      "loss/logits": 0.16117262840270996,
+      "loss/reg": 0.01584673300385475,
+      "step": 1309
+    },
+    {
+      "epoch": 0.16375,
+      "grad_norm": 2.900300979614258,
+      "grad_norm_var": 6.0232289618053185,
+      "learning_rate": 0.0001,
+      "loss": 1.0603,
+      "loss/crossentropy": 2.634578227996826,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.14015381038188934,
+      "loss/reg": 0.015838829800486565,
+      "step": 1310
+    },
+    {
+      "epoch": 0.163875,
+      "grad_norm": 3.875929594039917,
+      "grad_norm_var": 5.923678794810188,
+      "learning_rate": 0.0001,
+      "loss": 1.1148,
+      "loss/crossentropy": 2.8410253524780273,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.15175345540046692,
+      "loss/reg": 0.01583097316324711,
+      "step": 1311
+    },
+    {
+      "epoch": 0.164,
+      "grad_norm": 3.126546621322632,
+      "grad_norm_var": 5.854122320902001,
+      "learning_rate": 0.0001,
+      "loss": 1.1639,
+      "loss/crossentropy": 2.5248920917510986,
+      "loss/hidden": 0.85546875,
+      "loss/logits": 0.15023675560951233,
+      "loss/reg": 0.015823420137166977,
+      "step": 1312
+    },
+    {
+      "epoch": 0.164125,
+      "grad_norm": 3.225511312484741,
+      "grad_norm_var": 5.869908623309258,
+      "learning_rate": 0.0001,
+      "loss": 1.2341,
+      "loss/crossentropy": 2.382220506668091,
+      "loss/hidden": 0.9140625,
+      "loss/logits": 0.16187481582164764,
+      "loss/reg": 0.01581561006605625,
+      "step": 1313
+    },
+    {
+      "epoch": 0.16425,
+      "grad_norm": 5.516601085662842,
+      "grad_norm_var": 5.946099660897706,
+      "learning_rate": 0.0001,
+      "loss": 1.5097,
+      "loss/crossentropy": 2.2973639965057373,
+      "loss/hidden": 1.1328125,
+      "loss/logits": 0.2187841385602951,
+      "loss/reg": 0.015808099880814552,
+      "step": 1314
+    },
+    {
+      "epoch": 0.164375,
+      "grad_norm": 3.7441492080688477,
+      "grad_norm_var": 5.967949014164251,
+      "learning_rate": 0.0001,
+      "loss": 1.0893,
+      "loss/crossentropy": 2.7098612785339355,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.15003027021884918,
+      "loss/reg": 0.015800559893250465,
+      "step": 1315
+    },
+    {
+      "epoch": 0.1645,
+      "grad_norm": 3.2912096977233887,
+      "grad_norm_var": 5.5351073509179205,
+      "learning_rate": 0.0001,
+      "loss": 1.1089,
+      "loss/crossentropy": 2.5598583221435547,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.15022039413452148,
+      "loss/reg": 0.015792248770594597,
+      "step": 1316
+    },
+    {
+      "epoch": 0.164625,
+      "grad_norm": 2.697364091873169,
+      "grad_norm_var": 5.596594138613236,
+      "learning_rate": 0.0001,
+      "loss": 1.1302,
+      "loss/crossentropy": 2.364313840866089,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.1559445559978485,
+      "loss/reg": 0.015783967450261116,
+      "step": 1317
+    },
+    {
+      "epoch": 0.16475,
+      "grad_norm": 3.0158493518829346,
+      "grad_norm_var": 5.652685497198369,
+      "learning_rate": 0.0001,
+      "loss": 1.1872,
+      "loss/crossentropy": 2.7219552993774414,
+      "loss/hidden": 0.87109375,
+      "loss/logits": 0.15835720300674438,
+      "loss/reg": 0.015775663778185844,
+      "step": 1318
+    },
+    {
+      "epoch": 0.164875,
+      "grad_norm": 307.651611328125,
+      "grad_norm_var": 5765.62343534609,
+      "learning_rate": 0.0001,
+      "loss": 3.7364,
+      "loss/crossentropy": 3.080000400543213,
+      "loss/hidden": 3.09375,
+      "loss/logits": 0.48493900895118713,
+      "loss/reg": 0.015767192468047142,
+      "step": 1319
+    },
+    {
+      "epoch": 0.165,
+      "grad_norm": 2.4901864528656006,
+      "grad_norm_var": 5785.658820843435,
+      "learning_rate": 0.0001,
+      "loss": 1.1422,
+      "loss/crossentropy": 2.167736053466797,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.1487167328596115,
+      "loss/reg": 0.015759125351905823,
+      "step": 1320
+    },
+    {
+      "epoch": 0.165125,
+      "grad_norm": 2.829414129257202,
+      "grad_norm_var": 5787.017269350288,
+      "learning_rate": 0.0001,
+      "loss": 1.1765,
+      "loss/crossentropy": 2.35046648979187,
+      "loss/hidden": 0.83984375,
+      "loss/logits": 0.17911039292812347,
+      "loss/reg": 0.01575067639350891,
+      "step": 1321
+    },
+    {
+      "epoch": 0.16525,
+      "grad_norm": 3.047290325164795,
+      "grad_norm_var": 5784.979716656826,
+      "learning_rate": 0.0001,
+      "loss": 1.1083,
+      "loss/crossentropy": 2.5440683364868164,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.15790146589279175,
+      "loss/reg": 0.015742652118206024,
+      "step": 1322
+    },
+    {
+      "epoch": 0.165375,
+      "grad_norm": 2.9854726791381836,
+      "grad_norm_var": 5785.984900115946,
+      "learning_rate": 0.0001,
+      "loss": 1.0518,
+      "loss/crossentropy": 2.0691964626312256,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.12100932002067566,
+      "loss/reg": 0.015734149143099785,
+      "step": 1323
+    },
+    {
+      "epoch": 0.1655,
+      "grad_norm": 2.8492424488067627,
+      "grad_norm_var": 5792.618796374745,
+      "learning_rate": 0.0001,
+      "loss": 1.321,
+      "loss/crossentropy": 2.045762538909912,
+      "loss/hidden": 1.0390625,
+      "loss/logits": 0.12468324601650238,
+      "loss/reg": 0.01572607271373272,
+      "step": 1324
+    },
+    {
+      "epoch": 0.165625,
+      "grad_norm": 3.2983663082122803,
+      "grad_norm_var": 5791.394160827106,
+      "learning_rate": 0.0001,
+      "loss": 1.1712,
+      "loss/crossentropy": 2.6199800968170166,
+      "loss/hidden": 0.87109375,
+      "loss/logits": 0.14290478825569153,
+      "loss/reg": 0.01571841724216938,
+      "step": 1325
+    },
+    {
+      "epoch": 0.16575,
+      "grad_norm": 3.7790138721466064,
+      "grad_norm_var": 5789.171384194312,
+      "learning_rate": 0.0001,
+      "loss": 1.3719,
+      "loss/crossentropy": 2.5326523780822754,
+      "loss/hidden": 0.984375,
+      "loss/logits": 0.23037347197532654,
+      "loss/reg": 0.015710193663835526,
+      "step": 1326
+    },
+    {
+      "epoch": 0.165875,
+      "grad_norm": 3.0544376373291016,
+      "grad_norm_var": 5791.235862450414,
+      "learning_rate": 0.0001,
+      "loss": 1.3355,
+      "loss/crossentropy": 2.679947853088379,
+      "loss/hidden": 0.94140625,
+      "loss/logits": 0.2371068298816681,
+      "loss/reg": 0.015702618286013603,
+      "step": 1327
+    },
+    {
+      "epoch": 0.166,
+      "grad_norm": 3.107823133468628,
+      "grad_norm_var": 5791.283719365005,
+      "learning_rate": 0.0001,
+      "loss": 1.184,
+      "loss/crossentropy": 2.564558506011963,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.17545363306999207,
+      "loss/reg": 0.01569523848593235,
+      "step": 1328
+    },
+    {
+      "epoch": 0.166125,
+      "grad_norm": 3.589580774307251,
+      "grad_norm_var": 5790.3667350596925,
+      "learning_rate": 0.0001,
+      "loss": 1.4278,
+      "loss/crossentropy": 2.2362802028656006,
+      "loss/hidden": 1.03125,
+      "loss/logits": 0.2396482676267624,
+      "loss/reg": 0.01568804495036602,
+      "step": 1329
+    },
+    {
+      "epoch": 0.16625,
+      "grad_norm": 3.2111618518829346,
+      "grad_norm_var": 5795.860842463791,
+      "learning_rate": 0.0001,
+      "loss": 1.0855,
+      "loss/crossentropy": 2.5137100219726562,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.1356830894947052,
+      "loss/reg": 0.01568055897951126,
+      "step": 1330
+    },
+    {
+      "epoch": 0.166375,
+      "grad_norm": 2.7476325035095215,
+      "grad_norm_var": 5798.370483928043,
+      "learning_rate": 0.0001,
+      "loss": 1.2698,
+      "loss/crossentropy": 2.456242322921753,
+      "loss/hidden": 0.9453125,
+      "loss/logits": 0.16777516901493073,
+      "loss/reg": 0.01567363552749157,
+      "step": 1331
+    },
+    {
+      "epoch": 0.1665,
+      "grad_norm": 2.6416473388671875,
+      "grad_norm_var": 5800.026099397797,
+      "learning_rate": 0.0001,
+      "loss": 1.1194,
+      "loss/crossentropy": 2.7138493061065674,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.15024441480636597,
+      "loss/reg": 0.015665553510189056,
+      "step": 1332
+    },
+    {
+      "epoch": 0.166625,
+      "grad_norm": 3.000983476638794,
+      "grad_norm_var": 5799.247920072332,
+      "learning_rate": 0.0001,
+      "loss": 1.2997,
+      "loss/crossentropy": 2.4222257137298584,
+      "loss/hidden": 0.9609375,
+      "loss/logits": 0.1821441352367401,
+      "loss/reg": 0.015658436343073845,
+      "step": 1333
+    },
+    {
+      "epoch": 0.16675,
+      "grad_norm": 4.0280961990356445,
+      "grad_norm_var": 5796.738777158668,
+      "learning_rate": 0.0001,
+      "loss": 1.4751,
+      "loss/crossentropy": 2.4873077869415283,
+      "loss/hidden": 1.1015625,
+      "loss/logits": 0.2170398086309433,
+      "loss/reg": 0.015650250017642975,
+      "step": 1334
+    },
+    {
+      "epoch": 0.166875,
+      "grad_norm": 3.3132269382476807,
+      "grad_norm_var": 0.16804106807275332,
+      "learning_rate": 0.0001,
+      "loss": 1.0521,
+      "loss/crossentropy": 2.3228578567504883,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.13784348964691162,
+      "loss/reg": 0.015642434358596802,
+      "step": 1335
+    },
+    {
+      "epoch": 0.167,
+      "grad_norm": 2.9326908588409424,
+      "grad_norm_var": 0.14292226941004174,
+      "learning_rate": 0.0001,
+      "loss": 1.0604,
+      "loss/crossentropy": 2.6680498123168945,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.1462496519088745,
+      "loss/reg": 0.015634268522262573,
+      "step": 1336
+    },
+    {
+      "epoch": 0.167125,
+      "grad_norm": 3.7127490043640137,
+      "grad_norm_var": 0.1538134730442567,
+      "learning_rate": 0.0001,
+      "loss": 1.1028,
+      "loss/crossentropy": 2.439518690109253,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.14183643460273743,
+      "loss/reg": 0.015626052394509315,
+      "step": 1337
+    },
+    {
+      "epoch": 0.16725,
+      "grad_norm": 3.0613186359405518,
+      "grad_norm_var": 0.15352851622272493,
+      "learning_rate": 0.0001,
+      "loss": 1.1624,
+      "loss/crossentropy": 2.4576408863067627,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.1624375730752945,
+      "loss/reg": 0.015617795288562775,
+      "step": 1338
+    },
+    {
+      "epoch": 0.167375,
+      "grad_norm": 2.435100793838501,
+      "grad_norm_var": 0.18872328446350153,
+      "learning_rate": 0.0001,
+      "loss": 1.1516,
+      "loss/crossentropy": 2.6004011631011963,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.1517818719148636,
+      "loss/reg": 0.015609413385391235,
+      "step": 1339
+    },
+    {
+      "epoch": 0.1675,
+      "grad_norm": 2.6485419273376465,
+      "grad_norm_var": 0.19989636027441596,
+      "learning_rate": 0.0001,
+      "loss": 1.0442,
+      "loss/crossentropy": 2.560258626937866,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.13042010366916656,
+      "loss/reg": 0.015600843355059624,
+      "step": 1340
+    },
+    {
+      "epoch": 0.167625,
+      "grad_norm": 2.89918851852417,
+      "grad_norm_var": 0.20249881233273162,
+      "learning_rate": 0.0001,
+      "loss": 1.0632,
+      "loss/crossentropy": 2.5941762924194336,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.12995409965515137,
+      "loss/reg": 0.015592261217534542,
+      "step": 1341
+    },
+    {
+      "epoch": 0.16775,
+      "grad_norm": 2.8898985385894775,
+      "grad_norm_var": 0.175583338922907,
+      "learning_rate": 0.0001,
+      "loss": 1.11,
+      "loss/crossentropy": 2.4292283058166504,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.14161787927150726,
+      "loss/reg": 0.015583738684654236,
+      "step": 1342
+    },
+    {
+      "epoch": 0.167875,
+      "grad_norm": 2.9340474605560303,
+      "grad_norm_var": 0.1768935876133876,
+      "learning_rate": 0.0001,
+      "loss": 1.0755,
+      "loss/crossentropy": 2.2534961700439453,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.14238594472408295,
+      "loss/reg": 0.015575552359223366,
+      "step": 1343
+    },
+    {
+      "epoch": 0.168,
+      "grad_norm": 2.4977829456329346,
+      "grad_norm_var": 0.19724767622532605,
+      "learning_rate": 0.0001,
+      "loss": 1.1106,
+      "loss/crossentropy": 2.5671305656433105,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.15805330872535706,
+      "loss/reg": 0.015566708520054817,
+      "step": 1344
+    },
+    {
+      "epoch": 0.168125,
+      "grad_norm": 3.79781174659729,
+      "grad_norm_var": 0.2153835126984556,
+      "learning_rate": 0.0001,
+      "loss": 1.2096,
+      "loss/crossentropy": 2.341862440109253,
+      "loss/hidden": 0.90234375,
+      "loss/logits": 0.15165603160858154,
+      "loss/reg": 0.015558542683720589,
+      "step": 1345
+    },
+    {
+      "epoch": 0.16825,
+      "grad_norm": 4.044692039489746,
+      "grad_norm_var": 0.2770521554047564,
+      "learning_rate": 0.0001,
+      "loss": 1.2426,
+      "loss/crossentropy": 2.053048849105835,
+      "loss/hidden": 0.96484375,
+      "loss/logits": 0.1222817674279213,
+      "loss/reg": 0.015550050884485245,
+      "step": 1346
+    },
+    {
+      "epoch": 0.168375,
+      "grad_norm": 2.795452117919922,
+      "grad_norm_var": 0.27495421257927977,
+      "learning_rate": 0.0001,
+      "loss": 1.1152,
+      "loss/crossentropy": 2.381235361099243,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.16290614008903503,
+      "loss/reg": 0.015541622415184975,
+      "step": 1347
+    },
+    {
+      "epoch": 0.1685,
+      "grad_norm": 2.5820398330688477,
+      "grad_norm_var": 0.27883561860861783,
+      "learning_rate": 0.0001,
+      "loss": 0.9562,
+      "loss/crossentropy": 2.6334848403930664,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.12118068337440491,
+      "loss/reg": 0.01553329173475504,
+      "step": 1348
+    },
+    {
+      "epoch": 0.168625,
+      "grad_norm": 3.5735538005828857,
+      "grad_norm_var": 0.29189209249198955,
+      "learning_rate": 0.0001,
+      "loss": 1.1534,
+      "loss/crossentropy": 2.587275505065918,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.1934923231601715,
+      "loss/reg": 0.01552544254809618,
+      "step": 1349
+    },
+    {
+      "epoch": 0.16875,
+      "grad_norm": 2.8559420108795166,
+      "grad_norm_var": 0.23804927371853257,
+      "learning_rate": 0.0001,
+      "loss": 1.1208,
+      "loss/crossentropy": 2.4297351837158203,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.14923590421676636,
+      "loss/reg": 0.01551780290901661,
+      "step": 1350
+    },
+    {
+      "epoch": 0.168875,
+      "grad_norm": 2.5194828510284424,
+      "grad_norm_var": 0.25071932648220735,
+      "learning_rate": 0.0001,
+      "loss": 1.0241,
+      "loss/crossentropy": 2.4934616088867188,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.12285730242729187,
+      "loss/reg": 0.015510031953454018,
+      "step": 1351
+    },
+    {
+      "epoch": 0.169,
+      "grad_norm": 2.714785575866699,
+      "grad_norm_var": 0.2559699884583568,
+      "learning_rate": 0.0001,
+      "loss": 0.9928,
+      "loss/crossentropy": 2.7447316646575928,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.13469095528125763,
+      "loss/reg": 0.015501690097153187,
+      "step": 1352
+    },
+    {
+      "epoch": 0.169125,
+      "grad_norm": 2.226409912109375,
+      "grad_norm_var": 0.2523278002535238,
+      "learning_rate": 0.0001,
+      "loss": 1.0071,
+      "loss/crossentropy": 2.3574349880218506,
+      "loss/hidden": 0.72265625,
+      "loss/logits": 0.1294814646244049,
+      "loss/reg": 0.015494225546717644,
+      "step": 1353
+    },
+    {
+      "epoch": 0.16925,
+      "grad_norm": 2.2927169799804688,
+      "grad_norm_var": 0.27320470544451986,
+      "learning_rate": 0.0001,
+      "loss": 0.9263,
+      "loss/crossentropy": 2.6081109046936035,
+      "loss/hidden": 0.66015625,
+      "loss/logits": 0.11125902831554413,
+      "loss/reg": 0.015486053191125393,
+      "step": 1354
+    },
+    {
+      "epoch": 0.169375,
+      "grad_norm": 2.4285078048706055,
+      "grad_norm_var": 0.27357804892559606,
+      "learning_rate": 0.0001,
+      "loss": 1.0351,
+      "loss/crossentropy": 2.654952049255371,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.13033051788806915,
+      "loss/reg": 0.015478008426725864,
+      "step": 1355
+    },
+    {
+      "epoch": 0.1695,
+      "grad_norm": 2.5576510429382324,
+      "grad_norm_var": 0.27661218725713915,
+      "learning_rate": 0.0001,
+      "loss": 0.9944,
+      "loss/crossentropy": 2.525481939315796,
+      "loss/hidden": 0.6953125,
+      "loss/logits": 0.14441752433776855,
+      "loss/reg": 0.015470432117581367,
+      "step": 1356
+    },
+    {
+      "epoch": 0.169625,
+      "grad_norm": 3.3350884914398193,
+      "grad_norm_var": 0.2913103816812319,
+      "learning_rate": 0.0001,
+      "loss": 1.0548,
+      "loss/crossentropy": 2.649492025375366,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.15016667544841766,
+      "loss/reg": 0.015463395975530148,
+      "step": 1357
+    },
+    {
+      "epoch": 0.16975,
+      "grad_norm": 39.05031967163086,
+      "grad_norm_var": 82.07282531411327,
+      "learning_rate": 0.0001,
+      "loss": 1.0599,
+      "loss/crossentropy": 2.447631359100342,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.1085047721862793,
+      "loss/reg": 0.015455886721611023,
+      "step": 1358
+    },
+    {
+      "epoch": 0.169875,
+      "grad_norm": 2.6993017196655273,
+      "grad_norm_var": 82.14524851838627,
+      "learning_rate": 0.0001,
+      "loss": 0.9833,
+      "loss/crossentropy": 2.4683852195739746,
+      "loss/hidden": 0.70703125,
+      "loss/logits": 0.12176641821861267,
+      "loss/reg": 0.015447665005922318,
+      "step": 1359
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.8238816261291504,
+      "grad_norm_var": 82.03774119861214,
+      "learning_rate": 0.0001,
+      "loss": 0.9911,
+      "loss/crossentropy": 2.6632418632507324,
+      "loss/hidden": 0.71875,
+      "loss/logits": 0.11792933940887451,
+      "loss/reg": 0.015439565293490887,
+      "step": 1360
+    },
+    {
+      "epoch": 0.170125,
+      "grad_norm": 3.259418249130249,
+      "grad_norm_var": 82.15246657395147,
+      "learning_rate": 0.0001,
+      "loss": 1.1184,
+      "loss/crossentropy": 2.7344815731048584,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.15935808420181274,
+      "loss/reg": 0.015431756153702736,
+      "step": 1361
+    },
+    {
+      "epoch": 0.17025,
+      "grad_norm": 3.7520382404327393,
+      "grad_norm_var": 82.19938647618869,
+      "learning_rate": 0.0001,
+      "loss": 0.9883,
+      "loss/crossentropy": 2.7563743591308594,
+      "loss/hidden": 0.6953125,
+      "loss/logits": 0.13872626423835754,
+      "loss/reg": 0.015424099750816822,
+      "step": 1362
+    },
+    {
+      "epoch": 0.170375,
+      "grad_norm": 2.766862154006958,
+      "grad_norm_var": 82.20819070334429,
+      "learning_rate": 0.0001,
+      "loss": 0.9816,
+      "loss/crossentropy": 2.5083751678466797,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.12434659898281097,
+      "loss/reg": 0.01541648618876934,
+      "step": 1363
+    },
+    {
+      "epoch": 0.1705,
+      "grad_norm": 4.0485734939575195,
+      "grad_norm_var": 81.85223413984265,
+      "learning_rate": 0.0001,
+      "loss": 1.2789,
+      "loss/crossentropy": 2.2484068870544434,
+      "loss/hidden": 0.9296875,
+      "loss/logits": 0.19513945281505585,
+      "loss/reg": 0.015408649109303951,
+      "step": 1364
+    },
+    {
+      "epoch": 0.170625,
+      "grad_norm": 3.177393913269043,
+      "grad_norm_var": 81.94697865555716,
+      "learning_rate": 0.0001,
+      "loss": 1.0776,
+      "loss/crossentropy": 2.5238068103790283,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.1267687976360321,
+      "loss/reg": 0.01540052518248558,
+      "step": 1365
+    },
+    {
+      "epoch": 0.17075,
+      "grad_norm": 8.407809257507324,
+      "grad_norm_var": 82.17024249924071,
+      "learning_rate": 0.0001,
+      "loss": 1.3745,
+      "loss/crossentropy": 2.6552045345306396,
+      "loss/hidden": 1.0,
+      "loss/logits": 0.2205539047718048,
+      "loss/reg": 0.01539271418005228,
+      "step": 1366
+    },
+    {
+      "epoch": 0.170875,
+      "grad_norm": 3.5715489387512207,
+      "grad_norm_var": 81.82079857540536,
+      "learning_rate": 0.0001,
+      "loss": 1.1968,
+      "loss/crossentropy": 2.3991222381591797,
+      "loss/hidden": 0.88671875,
+      "loss/logits": 0.1562199890613556,
+      "loss/reg": 0.01538482028990984,
+      "step": 1367
+    },
+    {
+      "epoch": 0.171,
+      "grad_norm": 3.431995153427124,
+      "grad_norm_var": 81.57995561859133,
+      "learning_rate": 0.0001,
+      "loss": 1.1711,
+      "loss/crossentropy": 2.475353479385376,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.16577771306037903,
+      "loss/reg": 0.015377094969153404,
+      "step": 1368
+    },
+    {
+      "epoch": 0.171125,
+      "grad_norm": 2.35178279876709,
+      "grad_norm_var": 81.52430399273271,
+      "learning_rate": 0.0001,
+      "loss": 0.9426,
+      "loss/crossentropy": 2.6334617137908936,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.113157257437706,
+      "loss/reg": 0.015369528904557228,
+      "step": 1369
+    },
+    {
+      "epoch": 0.17125,
+      "grad_norm": 2.277183771133423,
+      "grad_norm_var": 81.53121470659494,
+      "learning_rate": 0.0001,
+      "loss": 1.0905,
+      "loss/crossentropy": 2.4363012313842773,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.14004433155059814,
+      "loss/reg": 0.015361781232059002,
+      "step": 1370
+    },
+    {
+      "epoch": 0.171375,
+      "grad_norm": 2.6015970706939697,
+      "grad_norm_var": 81.45940420807128,
+      "learning_rate": 0.0001,
+      "loss": 1.1042,
+      "loss/crossentropy": 2.4151899814605713,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.14207510650157928,
+      "loss/reg": 0.015353736467659473,
+      "step": 1371
+    },
+    {
+      "epoch": 0.1715,
+      "grad_norm": 2.7182395458221436,
+      "grad_norm_var": 81.39518805728484,
+      "learning_rate": 0.0001,
+      "loss": 1.0125,
+      "loss/crossentropy": 2.546245813369751,
+      "loss/hidden": 0.734375,
+      "loss/logits": 0.12465515732765198,
+      "loss/reg": 0.015345688909292221,
+      "step": 1372
+    },
+    {
+      "epoch": 0.171625,
+      "grad_norm": 2.4946951866149902,
+      "grad_norm_var": 81.69783167090145,
+      "learning_rate": 0.0001,
+      "loss": 1.0365,
+      "loss/crossentropy": 2.4872238636016846,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.121379055082798,
+      "loss/reg": 0.015337930992245674,
+      "step": 1373
+    },
+    {
+      "epoch": 0.17175,
+      "grad_norm": 2.35099720954895,
+      "grad_norm_var": 2.1436230027036345,
+      "learning_rate": 0.0001,
+      "loss": 1.0129,
+      "loss/crossentropy": 2.5447275638580322,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.12912125885486603,
+      "loss/reg": 0.015329563058912754,
+      "step": 1374
+    },
+    {
+      "epoch": 0.171875,
+      "grad_norm": 3.531275987625122,
+      "grad_norm_var": 2.120711236961161,
+      "learning_rate": 0.0001,
+      "loss": 1.1698,
+      "loss/crossentropy": 2.6748950481414795,
+      "loss/hidden": 0.88671875,
+      "loss/logits": 0.1298878937959671,
+      "loss/reg": 0.015320966020226479,
+      "step": 1375
+    },
+    {
+      "epoch": 0.172,
+      "grad_norm": 2.5129640102386475,
+      "grad_norm_var": 2.1484737670750205,
+      "learning_rate": 0.0001,
+      "loss": 0.9549,
+      "loss/crossentropy": 2.4481074810028076,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.1260184943675995,
+      "loss/reg": 0.015312742441892624,
+      "step": 1376
+    },
+    {
+      "epoch": 0.172125,
+      "grad_norm": 2.2456185817718506,
+      "grad_norm_var": 2.222034906196358,
+      "learning_rate": 0.0001,
+      "loss": 0.9583,
+      "loss/crossentropy": 2.697751760482788,
+      "loss/hidden": 0.6875,
+      "loss/logits": 0.11780545860528946,
+      "loss/reg": 0.01530410535633564,
+      "step": 1377
+    },
+    {
+      "epoch": 0.17225,
+      "grad_norm": 3.371694564819336,
+      "grad_norm_var": 2.206379139706632,
+      "learning_rate": 0.0001,
+      "loss": 1.0815,
+      "loss/crossentropy": 2.5198638439178467,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.1512106955051422,
+      "loss/reg": 0.015296036377549171,
+      "step": 1378
+    },
+    {
+      "epoch": 0.172375,
+      "grad_norm": 3.3469138145446777,
+      "grad_norm_var": 2.190717504538292,
+      "learning_rate": 0.0001,
+      "loss": 1.1007,
+      "loss/crossentropy": 2.500404119491577,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.10408136248588562,
+      "loss/reg": 0.015288051217794418,
+      "step": 1379
+    },
+    {
+      "epoch": 0.1725,
+      "grad_norm": 2.7227180004119873,
+      "grad_norm_var": 2.1642779969536603,
+      "learning_rate": 0.0001,
+      "loss": 1.0844,
+      "loss/crossentropy": 2.3428802490234375,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.1386614292860031,
+      "loss/reg": 0.015280190855264664,
+      "step": 1380
+    },
+    {
+      "epoch": 0.172625,
+      "grad_norm": 2.9242119789123535,
+      "grad_norm_var": 2.1688668977830954,
+      "learning_rate": 0.0001,
+      "loss": 1.239,
+      "loss/crossentropy": 2.668134927749634,
+      "loss/hidden": 0.88671875,
+      "loss/logits": 0.19956323504447937,
+      "loss/reg": 0.015271900221705437,
+      "step": 1381
+    },
+    {
+      "epoch": 0.17275,
+      "grad_norm": 2.492800712585449,
+      "grad_norm_var": 0.23164549730800724,
+      "learning_rate": 0.0001,
+      "loss": 0.9934,
+      "loss/crossentropy": 2.5607941150665283,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.12592297792434692,
+      "loss/reg": 0.01526356115937233,
+      "step": 1382
+    },
+    {
+      "epoch": 0.172875,
+      "grad_norm": 3.9195237159729004,
+      "grad_norm_var": 0.2745866186604149,
+      "learning_rate": 0.0001,
+      "loss": 1.2783,
+      "loss/crossentropy": 2.569718837738037,
+      "loss/hidden": 0.9453125,
+      "loss/logits": 0.18045136332511902,
+      "loss/reg": 0.015255567617714405,
+      "step": 1383
+    },
+    {
+      "epoch": 0.173,
+      "grad_norm": 3.3682620525360107,
+      "grad_norm_var": 0.2697324337180568,
+      "learning_rate": 0.0001,
+      "loss": 1.0247,
+      "loss/crossentropy": 2.756709337234497,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.1299922913312912,
+      "loss/reg": 0.015247280709445477,
+      "step": 1384
+    },
+    {
+      "epoch": 0.173125,
+      "grad_norm": 75.24736022949219,
+      "grad_norm_var": 327.7621509720236,
+      "learning_rate": 0.0001,
+      "loss": 1.1005,
+      "loss/crossentropy": 2.492774248123169,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.11999952048063278,
+      "loss/reg": 0.015239723026752472,
+      "step": 1385
+    },
+    {
+      "epoch": 0.17325,
+      "grad_norm": 2.639838218688965,
+      "grad_norm_var": 327.5234904743987,
+      "learning_rate": 0.0001,
+      "loss": 1.123,
+      "loss/crossentropy": 2.783590793609619,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.15431414544582367,
+      "loss/reg": 0.015231656841933727,
+      "step": 1386
+    },
+    {
+      "epoch": 0.173375,
+      "grad_norm": 8.296279907226562,
+      "grad_norm_var": 325.9027345524763,
+      "learning_rate": 0.0001,
+      "loss": 1.0279,
+      "loss/crossentropy": 2.591702938079834,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.12174837291240692,
+      "loss/reg": 0.01522404421120882,
+      "step": 1387
+    },
+    {
+      "epoch": 0.1735,
+      "grad_norm": 5.515318393707275,
+      "grad_norm_var": 324.5108738623049,
+      "learning_rate": 0.0001,
+      "loss": 1.2457,
+      "loss/crossentropy": 2.321329355239868,
+      "loss/hidden": 0.9453125,
+      "loss/logits": 0.14820685982704163,
+      "loss/reg": 0.01521630771458149,
+      "step": 1388
+    },
+    {
+      "epoch": 0.173625,
+      "grad_norm": 2.965728759765625,
+      "grad_norm_var": 324.1829850455801,
+      "learning_rate": 0.0001,
+      "loss": 1.0633,
+      "loss/crossentropy": 2.864595413208008,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.14945918321609497,
+      "loss/reg": 0.015208802185952663,
+      "step": 1389
+    },
+    {
+      "epoch": 0.17375,
+      "grad_norm": 2.8366498947143555,
+      "grad_norm_var": 323.8341522332258,
+      "learning_rate": 0.0001,
+      "loss": 1.178,
+      "loss/crossentropy": 2.634251594543457,
+      "loss/hidden": 0.8671875,
+      "loss/logits": 0.1587602198123932,
+      "loss/reg": 0.015202116221189499,
+      "step": 1390
+    },
+    {
+      "epoch": 0.173875,
+      "grad_norm": 3.016205310821533,
+      "grad_norm_var": 324.1573581089945,
+      "learning_rate": 0.0001,
+      "loss": 1.0881,
+      "loss/crossentropy": 2.1916353702545166,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.14321959018707275,
+      "loss/reg": 0.015194511041045189,
+      "step": 1391
+    },
+    {
+      "epoch": 0.174,
+      "grad_norm": 34.42859649658203,
+      "grad_norm_var": 364.624406562687,
+      "learning_rate": 0.0001,
+      "loss": 1.0421,
+      "loss/crossentropy": 2.4397382736206055,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.12064860761165619,
+      "loss/reg": 0.015187704935669899,
+      "step": 1392
+    },
+    {
+      "epoch": 0.174125,
+      "grad_norm": 3.928023099899292,
+      "grad_norm_var": 363.0711295434095,
+      "learning_rate": 0.0001,
+      "loss": 1.2257,
+      "loss/crossentropy": 2.1384785175323486,
+      "loss/hidden": 0.921875,
+      "loss/logits": 0.15198791027069092,
+      "loss/reg": 0.015179669484496117,
+      "step": 1393
+    },
+    {
+      "epoch": 0.17425,
+      "grad_norm": 5.9230194091796875,
+      "grad_norm_var": 361.2014745642014,
+      "learning_rate": 0.0001,
+      "loss": 1.2374,
+      "loss/crossentropy": 2.474677801132202,
+      "loss/hidden": 0.9453125,
+      "loss/logits": 0.14034368097782135,
+      "loss/reg": 0.015171775594353676,
+      "step": 1394
+    },
+    {
+      "epoch": 0.174375,
+      "grad_norm": 5.671228885650635,
+      "grad_norm_var": 359.4081015077895,
+      "learning_rate": 0.0001,
+      "loss": 1.6994,
+      "loss/crossentropy": 2.1997835636138916,
+      "loss/hidden": 1.34375,
+      "loss/logits": 0.20398974418640137,
+      "loss/reg": 0.015164447948336601,
+      "step": 1395
+    },
+    {
+      "epoch": 0.1745,
+      "grad_norm": 2.8046653270721436,
+      "grad_norm_var": 359.32498119248385,
+      "learning_rate": 0.0001,
+      "loss": 1.2227,
+      "loss/crossentropy": 2.2905519008636475,
+      "loss/hidden": 0.92578125,
+      "loss/logits": 0.14532050490379333,
+      "loss/reg": 0.015156446024775505,
+      "step": 1396
+    },
+    {
+      "epoch": 0.174625,
+      "grad_norm": 3.8924100399017334,
+      "grad_norm_var": 358.42190384848453,
+      "learning_rate": 0.0001,
+      "loss": 1.2919,
+      "loss/crossentropy": 2.5372776985168457,
+      "loss/hidden": 0.89453125,
+      "loss/logits": 0.2458799183368683,
+      "loss/reg": 0.015149053186178207,
+      "step": 1397
+    },
+    {
+      "epoch": 0.17475,
+      "grad_norm": 3.204949378967285,
+      "grad_norm_var": 357.6995478125139,
+      "learning_rate": 0.0001,
+      "loss": 1.2475,
+      "loss/crossentropy": 2.3744595050811768,
+      "loss/hidden": 0.9296875,
+      "loss/logits": 0.16641047596931458,
+      "loss/reg": 0.015141867101192474,
+      "step": 1398
+    },
+    {
+      "epoch": 0.174875,
+      "grad_norm": 2.5903260707855225,
+      "grad_norm_var": 358.97241696361874,
+      "learning_rate": 0.0001,
+      "loss": 1.079,
+      "loss/crossentropy": 2.613074541091919,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.1385442167520523,
+      "loss/reg": 0.015134819783270359,
+      "step": 1399
+    },
+    {
+      "epoch": 0.175,
+      "grad_norm": 2.877967357635498,
+      "grad_norm_var": 359.4468337869737,
+      "learning_rate": 0.0001,
+      "loss": 1.0731,
+      "loss/crossentropy": 2.320415735244751,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.12884414196014404,
+      "loss/reg": 0.015128130093216896,
+      "step": 1400
+    },
+    {
+      "epoch": 0.175125,
+      "grad_norm": 2.7650208473205566,
+      "grad_norm_var": 60.75819602934475,
+      "learning_rate": 0.0001,
+      "loss": 1.2011,
+      "loss/crossentropy": 2.242631673812866,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.1670370101928711,
+      "loss/reg": 0.015121539123356342,
+      "step": 1401
+    },
+    {
+      "epoch": 0.17525,
+      "grad_norm": 3.245616912841797,
+      "grad_norm_var": 60.52307577230558,
+      "learning_rate": 0.0001,
+      "loss": 1.2625,
+      "loss/crossentropy": 2.2215981483459473,
+      "loss/hidden": 0.9453125,
+      "loss/logits": 0.16608265042304993,
+      "loss/reg": 0.015114927664399147,
+      "step": 1402
+    },
+    {
+      "epoch": 0.175375,
+      "grad_norm": 62.871517181396484,
+      "grad_norm_var": 264.3128112734296,
+      "learning_rate": 0.0001,
+      "loss": 3.0131,
+      "loss/crossentropy": 2.5125503540039062,
+      "loss/hidden": 2.4375,
+      "loss/logits": 0.42451030015945435,
+      "loss/reg": 0.015108034946024418,
+      "step": 1403
+    },
+    {
+      "epoch": 0.1755,
+      "grad_norm": 3.219738006591797,
+      "grad_norm_var": 265.7955458129954,
+      "learning_rate": 0.0001,
+      "loss": 1.2222,
+      "loss/crossentropy": 2.3459861278533936,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.16100311279296875,
+      "loss/reg": 0.015101352706551552,
+      "step": 1404
+    },
+    {
+      "epoch": 0.175625,
+      "grad_norm": 2.7814700603485107,
+      "grad_norm_var": 265.9493587458945,
+      "learning_rate": 0.0001,
+      "loss": 1.0905,
+      "loss/crossentropy": 2.5293545722961426,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.1387513130903244,
+      "loss/reg": 0.01509346254169941,
+      "step": 1405
+    },
+    {
+      "epoch": 0.17575,
+      "grad_norm": 2.6544785499572754,
+      "grad_norm_var": 266.1042610002773,
+      "learning_rate": 0.0001,
+      "loss": 1.0535,
+      "loss/crossentropy": 2.5097312927246094,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.1291828155517578,
+      "loss/reg": 0.015085602179169655,
+      "step": 1406
+    },
+    {
+      "epoch": 0.175875,
+      "grad_norm": 3.1223692893981934,
+      "grad_norm_var": 266.0186046129393,
+      "learning_rate": 0.0001,
+      "loss": 1.4849,
+      "loss/crossentropy": 2.2004942893981934,
+      "loss/hidden": 1.140625,
+      "loss/logits": 0.1934625506401062,
+      "loss/reg": 0.015078413300216198,
+      "step": 1407
+    },
+    {
+      "epoch": 0.176,
+      "grad_norm": 4.406020164489746,
+      "grad_norm_var": 221.05808913641562,
+      "learning_rate": 0.0001,
+      "loss": 1.2206,
+      "loss/crossentropy": 2.239046096801758,
+      "loss/hidden": 0.90625,
+      "loss/logits": 0.1636815071105957,
+      "loss/reg": 0.015070472843945026,
+      "step": 1408
+    },
+    {
+      "epoch": 0.176125,
+      "grad_norm": 2.3968393802642822,
+      "grad_norm_var": 221.88230400943462,
+      "learning_rate": 0.0001,
+      "loss": 1.0135,
+      "loss/crossentropy": 2.5016496181488037,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.13240215182304382,
+      "loss/reg": 0.015062601305544376,
+      "step": 1409
+    },
+    {
+      "epoch": 0.17625,
+      "grad_norm": 4.053096771240234,
+      "grad_norm_var": 222.40718733745135,
+      "learning_rate": 0.0001,
+      "loss": 1.3875,
+      "loss/crossentropy": 2.5696771144866943,
+      "loss/hidden": 1.046875,
+      "loss/logits": 0.19007228314876556,
+      "loss/reg": 0.015055070631206036,
+      "step": 1410
+    },
+    {
+      "epoch": 0.176375,
+      "grad_norm": 3.213609218597412,
+      "grad_norm_var": 223.23151802105554,
+      "learning_rate": 0.0001,
+      "loss": 1.101,
+      "loss/crossentropy": 2.580714225769043,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.13412834703922272,
+      "loss/reg": 0.01504768431186676,
+      "step": 1411
+    },
+    {
+      "epoch": 0.1765,
+      "grad_norm": 7.05836820602417,
+      "grad_norm_var": 222.05031160271446,
+      "learning_rate": 0.0001,
+      "loss": 1.3444,
+      "loss/crossentropy": 2.6128039360046387,
+      "loss/hidden": 1.046875,
+      "loss/logits": 0.14714105427265167,
+      "loss/reg": 0.015039726160466671,
+      "step": 1412
+    },
+    {
+      "epoch": 0.176625,
+      "grad_norm": 2.905400037765503,
+      "grad_norm_var": 222.53952156242244,
+      "learning_rate": 0.0001,
+      "loss": 1.0393,
+      "loss/crossentropy": 2.390307664871216,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.13505280017852783,
+      "loss/reg": 0.015031633898615837,
+      "step": 1413
+    },
+    {
+      "epoch": 0.17675,
+      "grad_norm": 2.7498996257781982,
+      "grad_norm_var": 222.7879046702847,
+      "learning_rate": 0.0001,
+      "loss": 1.1323,
+      "loss/crossentropy": 2.4936277866363525,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.15391477942466736,
+      "loss/reg": 0.015023862943053246,
+      "step": 1414
+    },
+    {
+      "epoch": 0.176875,
+      "grad_norm": 3.2711455821990967,
+      "grad_norm_var": 222.41140935738122,
+      "learning_rate": 0.0001,
+      "loss": 1.159,
+      "loss/crossentropy": 2.586298704147339,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.15724779665470123,
+      "loss/reg": 0.015016157180070877,
+      "step": 1415
+    },
+    {
+      "epoch": 0.177,
+      "grad_norm": 2.5050208568573,
+      "grad_norm_var": 222.63002493426723,
+      "learning_rate": 0.0001,
+      "loss": 0.9875,
+      "loss/crossentropy": 2.4593968391418457,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.12256527692079544,
+      "loss/reg": 0.015008063055574894,
+      "step": 1416
+    },
+    {
+      "epoch": 0.177125,
+      "grad_norm": 5.0899248123168945,
+      "grad_norm_var": 221.63143029624337,
+      "learning_rate": 0.0001,
+      "loss": 1.3226,
+      "loss/crossentropy": 2.529581069946289,
+      "loss/hidden": 0.9921875,
+      "loss/logits": 0.18042267858982086,
+      "loss/reg": 0.014999698847532272,
+      "step": 1417
+    },
+    {
+      "epoch": 0.17725,
+      "grad_norm": 2.873887300491333,
+      "grad_norm_var": 221.83712878589319,
+      "learning_rate": 0.0001,
+      "loss": 1.1234,
+      "loss/crossentropy": 2.637040376663208,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.1570313721895218,
+      "loss/reg": 0.014991391450166702,
+      "step": 1418
+    },
+    {
+      "epoch": 0.177375,
+      "grad_norm": 2.4938929080963135,
+      "grad_norm_var": 1.4893372742809399,
+      "learning_rate": 0.0001,
+      "loss": 1.0818,
+      "loss/crossentropy": 2.3521924018859863,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.14290779829025269,
+      "loss/reg": 0.01498338021337986,
+      "step": 1419
+    },
+    {
+      "epoch": 0.1775,
+      "grad_norm": 7.30941104888916,
+      "grad_norm_var": 2.4229140389199517,
+      "learning_rate": 0.0001,
+      "loss": 1.2235,
+      "loss/crossentropy": 2.295588254928589,
+      "loss/hidden": 0.9140625,
+      "loss/logits": 0.15963850915431976,
+      "loss/reg": 0.014975651167333126,
+      "step": 1420
+    },
+    {
+      "epoch": 0.177625,
+      "grad_norm": 2.976419448852539,
+      "grad_norm_var": 2.4019258000462114,
+      "learning_rate": 0.0001,
+      "loss": 1.1336,
+      "loss/crossentropy": 2.398700714111328,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.17146307229995728,
+      "loss/reg": 0.014968100003898144,
+      "step": 1421
+    },
+    {
+      "epoch": 0.17775,
+      "grad_norm": 6.012148380279541,
+      "grad_norm_var": 2.641842426821879,
+      "learning_rate": 0.0001,
+      "loss": 1.4281,
+      "loss/crossentropy": 2.3960044384002686,
+      "loss/hidden": 1.0625,
+      "loss/logits": 0.21602004766464233,
+      "loss/reg": 0.01496051624417305,
+      "step": 1422
+    },
+    {
+      "epoch": 0.177875,
+      "grad_norm": 3.721160888671875,
+      "grad_norm_var": 2.601979835113735,
+      "learning_rate": 0.0001,
+      "loss": 1.4417,
+      "loss/crossentropy": 2.231612205505371,
+      "loss/hidden": 1.078125,
+      "loss/logits": 0.21401020884513855,
+      "loss/reg": 0.01495243888348341,
+      "step": 1423
+    },
+    {
+      "epoch": 0.178,
+      "grad_norm": 2.6747055053710938,
+      "grad_norm_var": 2.68168930149086,
+      "learning_rate": 0.0001,
+      "loss": 1.1718,
+      "loss/crossentropy": 2.2853996753692627,
+      "loss/hidden": 0.87109375,
+      "loss/logits": 0.15127936005592346,
+      "loss/reg": 0.014944901689887047,
+      "step": 1424
+    },
+    {
+      "epoch": 0.178125,
+      "grad_norm": 11.191994667053223,
+      "grad_norm_var": 5.833885032277132,
+      "learning_rate": 0.0001,
+      "loss": 1.2679,
+      "loss/crossentropy": 2.5116124153137207,
+      "loss/hidden": 0.9296875,
+      "loss/logits": 0.18879596889019012,
+      "loss/reg": 0.014937575906515121,
+      "step": 1425
+    },
+    {
+      "epoch": 0.17825,
+      "grad_norm": 3.66359543800354,
+      "grad_norm_var": 5.860409413897694,
+      "learning_rate": 0.0001,
+      "loss": 1.1213,
+      "loss/crossentropy": 2.5135722160339355,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.1477985680103302,
+      "loss/reg": 0.014929663389921188,
+      "step": 1426
+    },
+    {
+      "epoch": 0.178375,
+      "grad_norm": 7.735219478607178,
+      "grad_norm_var": 6.448943732227483,
+      "learning_rate": 0.0001,
+      "loss": 1.1287,
+      "loss/crossentropy": 2.675562620162964,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.16307833790779114,
+      "loss/reg": 0.01492208894342184,
+      "step": 1427
+    },
+    {
+      "epoch": 0.1785,
+      "grad_norm": 3.957063913345337,
+      "grad_norm_var": 6.0498597570917205,
+      "learning_rate": 0.0001,
+      "loss": 1.247,
+      "loss/crossentropy": 2.4751713275909424,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.18771307170391083,
+      "loss/reg": 0.014914041385054588,
+      "step": 1428
+    },
+    {
+      "epoch": 0.178625,
+      "grad_norm": 4.195860862731934,
+      "grad_norm_var": 5.888917428574246,
+      "learning_rate": 0.0001,
+      "loss": 1.4162,
+      "loss/crossentropy": 2.572279930114746,
+      "loss/hidden": 1.03125,
+      "loss/logits": 0.23586444556713104,
+      "loss/reg": 0.014906428754329681,
+      "step": 1429
+    },
+    {
+      "epoch": 0.17875,
+      "grad_norm": 3.2409324645996094,
+      "grad_norm_var": 5.787681963969308,
+      "learning_rate": 0.0001,
+      "loss": 1.0805,
+      "loss/crossentropy": 2.401996374130249,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.13070350885391235,
+      "loss/reg": 0.014898893423378468,
+      "step": 1430
+    },
+    {
+      "epoch": 0.178875,
+      "grad_norm": 3.5550079345703125,
+      "grad_norm_var": 5.744049750040046,
+      "learning_rate": 0.0001,
+      "loss": 1.159,
+      "loss/crossentropy": 2.4993062019348145,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.1741950362920761,
+      "loss/reg": 0.014890996739268303,
+      "step": 1431
+    },
+    {
+      "epoch": 0.179,
+      "grad_norm": 2.403970718383789,
+      "grad_norm_var": 5.772574341640287,
+      "learning_rate": 0.0001,
+      "loss": 0.9385,
+      "loss/crossentropy": 2.338052749633789,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.11390648782253265,
+      "loss/reg": 0.014883420430123806,
+      "step": 1432
+    },
+    {
+      "epoch": 0.179125,
+      "grad_norm": 3.128432512283325,
+      "grad_norm_var": 5.876657514659041,
+      "learning_rate": 0.0001,
+      "loss": 1.1205,
+      "loss/crossentropy": 2.323333740234375,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.1436302661895752,
+      "loss/reg": 0.014875980094075203,
+      "step": 1433
+    },
+    {
+      "epoch": 0.17925,
+      "grad_norm": 2.75893235206604,
+      "grad_norm_var": 5.901577514262766,
+      "learning_rate": 0.0001,
+      "loss": 1.0911,
+      "loss/crossentropy": 2.403975009918213,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.14556562900543213,
+      "loss/reg": 0.014869259670376778,
+      "step": 1434
+    },
+    {
+      "epoch": 0.179375,
+      "grad_norm": 4.4465861320495605,
+      "grad_norm_var": 5.633549820228222,
+      "learning_rate": 0.0001,
+      "loss": 1.4989,
+      "loss/crossentropy": 2.418936014175415,
+      "loss/hidden": 1.109375,
+      "loss/logits": 0.24089978635311127,
+      "loss/reg": 0.014861512929201126,
+      "step": 1435
+    },
+    {
+      "epoch": 0.1795,
+      "grad_norm": 2.976722478866577,
+      "grad_norm_var": 5.218912579761936,
+      "learning_rate": 0.0001,
+      "loss": 1.1243,
+      "loss/crossentropy": 2.423041582107544,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.14372695982456207,
+      "loss/reg": 0.014853725209832191,
+      "step": 1436
+    },
+    {
+      "epoch": 0.179625,
+      "grad_norm": 2.8299055099487305,
+      "grad_norm_var": 5.245913751427944,
+      "learning_rate": 0.0001,
+      "loss": 0.9575,
+      "loss/crossentropy": 2.5777714252471924,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.1293865293264389,
+      "loss/reg": 0.014846443198621273,
+      "step": 1437
+    },
+    {
+      "epoch": 0.17975,
+      "grad_norm": 3.8410592079162598,
+      "grad_norm_var": 5.039317138416171,
+      "learning_rate": 0.0001,
+      "loss": 1.1673,
+      "loss/crossentropy": 2.2890384197235107,
+      "loss/hidden": 0.859375,
+      "loss/logits": 0.15949448943138123,
+      "loss/reg": 0.014838694594800472,
+      "step": 1438
+    },
+    {
+      "epoch": 0.179875,
+      "grad_norm": 3.1295011043548584,
+      "grad_norm_var": 5.094637447706395,
+      "learning_rate": 0.0001,
+      "loss": 1.0505,
+      "loss/crossentropy": 2.4979283809661865,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.12488513439893723,
+      "loss/reg": 0.014831180684268475,
+      "step": 1439
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.2620816230773926,
+      "grad_norm_var": 5.0039422612885565,
+      "learning_rate": 0.0001,
+      "loss": 1.0724,
+      "loss/crossentropy": 2.5473530292510986,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.1195274293422699,
+      "loss/reg": 0.014823324047029018,
+      "step": 1440
+    },
+    {
+      "epoch": 0.180125,
+      "grad_norm": 2.8635447025299072,
+      "grad_norm_var": 1.5135115386307687,
+      "learning_rate": 0.0001,
+      "loss": 1.0522,
+      "loss/crossentropy": 2.709258794784546,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.13453327119350433,
+      "loss/reg": 0.01481538638472557,
+      "step": 1441
+    },
+    {
+      "epoch": 0.18025,
+      "grad_norm": 2.694793462753296,
+      "grad_norm_var": 1.5670935881051355,
+      "learning_rate": 0.0001,
+      "loss": 1.042,
+      "loss/crossentropy": 2.6498730182647705,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.13220643997192383,
+      "loss/reg": 0.01480773463845253,
+      "step": 1442
+    },
+    {
+      "epoch": 0.180375,
+      "grad_norm": 3.662919282913208,
+      "grad_norm_var": 0.33856051311708674,
+      "learning_rate": 0.0001,
+      "loss": 1.195,
+      "loss/crossentropy": 2.589240550994873,
+      "loss/hidden": 0.86328125,
+      "loss/logits": 0.18375417590141296,
+      "loss/reg": 0.014799892902374268,
+      "step": 1443
+    },
+    {
+      "epoch": 0.1805,
+      "grad_norm": 3.0655100345611572,
+      "grad_norm_var": 0.3112265539838631,
+      "learning_rate": 0.0001,
+      "loss": 1.0221,
+      "loss/crossentropy": 2.322566509246826,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.11638704687356949,
+      "loss/reg": 0.014792241156101227,
+      "step": 1444
+    },
+    {
+      "epoch": 0.180625,
+      "grad_norm": 3.0010573863983154,
+      "grad_norm_var": 0.2503215727538245,
+      "learning_rate": 0.0001,
+      "loss": 1.1143,
+      "loss/crossentropy": 2.2516565322875977,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.13441051542758942,
+      "loss/reg": 0.014784677885472775,
+      "step": 1445
+    },
+    {
+      "epoch": 0.18075,
+      "grad_norm": 2.6158626079559326,
+      "grad_norm_var": 0.26956362632713443,
+      "learning_rate": 0.0001,
+      "loss": 1.1134,
+      "loss/crossentropy": 2.476480007171631,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.1648055911064148,
+      "loss/reg": 0.014777293428778648,
+      "step": 1446
+    },
+    {
+      "epoch": 0.180875,
+      "grad_norm": 2.593005895614624,
+      "grad_norm_var": 0.2741393520658806,
+      "learning_rate": 0.0001,
+      "loss": 1.1448,
+      "loss/crossentropy": 2.5278983116149902,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.1651032269001007,
+      "loss/reg": 0.014769522473216057,
+      "step": 1447
+    },
+    {
+      "epoch": 0.181,
+      "grad_norm": 3.3676040172576904,
+      "grad_norm_var": 0.24536603446710312,
+      "learning_rate": 0.0001,
+      "loss": 1.0625,
+      "loss/crossentropy": 2.5347743034362793,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.13756276667118073,
+      "loss/reg": 0.01476323138922453,
+      "step": 1448
+    },
+    {
+      "epoch": 0.181125,
+      "grad_norm": 2.7923572063446045,
+      "grad_norm_var": 0.2529365869795574,
+      "learning_rate": 0.0001,
+      "loss": 1.3644,
+      "loss/crossentropy": 2.110182046890259,
+      "loss/hidden": 1.0234375,
+      "loss/logits": 0.1933690905570984,
+      "loss/reg": 0.014755993150174618,
+      "step": 1449
+    },
+    {
+      "epoch": 0.18125,
+      "grad_norm": 70.97599792480469,
+      "grad_norm_var": 287.8273579394089,
+      "learning_rate": 0.0001,
+      "loss": 1.1063,
+      "loss/crossentropy": 2.6005754470825195,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.14632144570350647,
+      "loss/reg": 0.014748108573257923,
+      "step": 1450
+    },
+    {
+      "epoch": 0.181375,
+      "grad_norm": 5.7473530769348145,
+      "grad_norm_var": 287.42393180966496,
+      "learning_rate": 0.0001,
+      "loss": 1.6866,
+      "loss/crossentropy": 2.3792290687561035,
+      "loss/hidden": 1.2265625,
+      "loss/logits": 0.31258517503738403,
+      "loss/reg": 0.014740433543920517,
+      "step": 1451
+    },
+    {
+      "epoch": 0.1815,
+      "grad_norm": 3.4155495166778564,
+      "grad_norm_var": 287.17343283264796,
+      "learning_rate": 0.0001,
+      "loss": 1.1977,
+      "loss/crossentropy": 2.4722485542297363,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.1675756871700287,
+      "loss/reg": 0.014733717776834965,
+      "step": 1452
+    },
+    {
+      "epoch": 0.181625,
+      "grad_norm": 2.5721304416656494,
+      "grad_norm_var": 287.3377922083848,
+      "learning_rate": 0.0001,
+      "loss": 1.0864,
+      "loss/crossentropy": 2.65114426612854,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.1461598426103592,
+      "loss/reg": 0.01472744531929493,
+      "step": 1453
+    },
+    {
+      "epoch": 0.18175,
+      "grad_norm": 3.40112566947937,
+      "grad_norm_var": 287.5630487447142,
+      "learning_rate": 0.0001,
+      "loss": 1.3854,
+      "loss/crossentropy": 2.49849796295166,
+      "loss/hidden": 1.0390625,
+      "loss/logits": 0.19910715520381927,
+      "loss/reg": 0.014719628728926182,
+      "step": 1454
+    },
+    {
+      "epoch": 0.181875,
+      "grad_norm": 3.4560482501983643,
+      "grad_norm_var": 287.381708208898,
+      "learning_rate": 0.0001,
+      "loss": 1.2084,
+      "loss/crossentropy": 2.630333423614502,
+      "loss/hidden": 0.90625,
+      "loss/logits": 0.1550079882144928,
+      "loss/reg": 0.014712607488036156,
+      "step": 1455
+    },
+    {
+      "epoch": 0.182,
+      "grad_norm": 2.8808276653289795,
+      "grad_norm_var": 287.60459257620465,
+      "learning_rate": 0.0001,
+      "loss": 1.1446,
+      "loss/crossentropy": 2.7043960094451904,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.14598789811134338,
+      "loss/reg": 0.01470634713768959,
+      "step": 1456
+    },
+    {
+      "epoch": 0.182125,
+      "grad_norm": 2.3799662590026855,
+      "grad_norm_var": 287.91454947447465,
+      "learning_rate": 0.0001,
+      "loss": 1.0639,
+      "loss/crossentropy": 2.4705262184143066,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.13952219486236572,
+      "loss/reg": 0.014699905179440975,
+      "step": 1457
+    },
+    {
+      "epoch": 0.18225,
+      "grad_norm": 3.0028903484344482,
+      "grad_norm_var": 287.7266240160942,
+      "learning_rate": 0.0001,
+      "loss": 1.1176,
+      "loss/crossentropy": 2.4257349967956543,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.1581314206123352,
+      "loss/reg": 0.014692124910652637,
+      "step": 1458
+    },
+    {
+      "epoch": 0.182375,
+      "grad_norm": 2.5650155544281006,
+      "grad_norm_var": 288.3538726561922,
+      "learning_rate": 0.0001,
+      "loss": 1.0762,
+      "loss/crossentropy": 2.4188382625579834,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.14031586050987244,
+      "loss/reg": 0.014684327878057957,
+      "step": 1459
+    },
+    {
+      "epoch": 0.1825,
+      "grad_norm": 3.0308585166931152,
+      "grad_norm_var": 288.3738099925176,
+      "learning_rate": 0.0001,
+      "loss": 1.2435,
+      "loss/crossentropy": 2.5064949989318848,
+      "loss/hidden": 0.9296875,
+      "loss/logits": 0.16703946888446808,
+      "loss/reg": 0.014676612801849842,
+      "step": 1460
+    },
+    {
+      "epoch": 0.182625,
+      "grad_norm": 3.903977394104004,
+      "grad_norm_var": 287.89971053282926,
+      "learning_rate": 0.0001,
+      "loss": 1.0425,
+      "loss/crossentropy": 2.5762782096862793,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.14189936220645905,
+      "loss/reg": 0.014669781550765038,
+      "step": 1461
+    },
+    {
+      "epoch": 0.18275,
+      "grad_norm": 3.6422464847564697,
+      "grad_norm_var": 287.3082663217944,
+      "learning_rate": 0.0001,
+      "loss": 1.0475,
+      "loss/crossentropy": 2.6891908645629883,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.12741170823574066,
+      "loss/reg": 0.0146620599552989,
+      "step": 1462
+    },
+    {
+      "epoch": 0.182875,
+      "grad_norm": 2.7266409397125244,
+      "grad_norm_var": 287.22225368800906,
+      "learning_rate": 0.0001,
+      "loss": 1.1596,
+      "loss/crossentropy": 2.4353747367858887,
+      "loss/hidden": 0.84765625,
+      "loss/logits": 0.16537800431251526,
+      "loss/reg": 0.014655125327408314,
+      "step": 1463
+    },
+    {
+      "epoch": 0.183,
+      "grad_norm": 8.853667259216309,
+      "grad_norm_var": 286.08693801368264,
+      "learning_rate": 0.0001,
+      "loss": 1.2411,
+      "loss/crossentropy": 2.9862468242645264,
+      "loss/hidden": 0.90625,
+      "loss/logits": 0.1883150041103363,
+      "loss/reg": 0.014648628421127796,
+      "step": 1464
+    },
+    {
+      "epoch": 0.183125,
+      "grad_norm": 3.3163163661956787,
+      "grad_norm_var": 285.7518694340515,
+      "learning_rate": 0.0001,
+      "loss": 1.1675,
+      "loss/crossentropy": 2.642627477645874,
+      "loss/hidden": 0.8671875,
+      "loss/logits": 0.15386328101158142,
+      "loss/reg": 0.014640781097114086,
+      "step": 1465
+    },
+    {
+      "epoch": 0.18325,
+      "grad_norm": 3.663252830505371,
+      "grad_norm_var": 2.5336251924554736,
+      "learning_rate": 0.0001,
+      "loss": 1.2303,
+      "loss/crossentropy": 2.353497266769409,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.17375797033309937,
+      "loss/reg": 0.014634395018219948,
+      "step": 1466
+    },
+    {
+      "epoch": 0.183375,
+      "grad_norm": 3.4295380115509033,
+      "grad_norm_var": 2.224270864584783,
+      "learning_rate": 0.0001,
+      "loss": 1.186,
+      "loss/crossentropy": 2.5201594829559326,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.1568823754787445,
+      "loss/reg": 0.01462656632065773,
+      "step": 1467
+    },
+    {
+      "epoch": 0.1835,
+      "grad_norm": 2.4592854976654053,
+      "grad_norm_var": 2.294103952189973,
+      "learning_rate": 0.0001,
+      "loss": 1.0183,
+      "loss/crossentropy": 2.5786993503570557,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.1260080635547638,
+      "loss/reg": 0.01461972575634718,
+      "step": 1468
+    },
+    {
+      "epoch": 0.183625,
+      "grad_norm": 4.478452205657959,
+      "grad_norm_var": 2.296768240317753,
+      "learning_rate": 0.0001,
+      "loss": 1.1604,
+      "loss/crossentropy": 2.472398042678833,
+      "loss/hidden": 0.87890625,
+      "loss/logits": 0.13542431592941284,
+      "loss/reg": 0.014611870981752872,
+      "step": 1469
+    },
+    {
+      "epoch": 0.18375,
+      "grad_norm": 2.8013250827789307,
+      "grad_norm_var": 2.3331091729009654,
+      "learning_rate": 0.0001,
+      "loss": 1.2052,
+      "loss/crossentropy": 2.448435068130493,
+      "loss/hidden": 0.8671875,
+      "loss/logits": 0.19197094440460205,
+      "loss/reg": 0.014604009687900543,
+      "step": 1470
+    },
+    {
+      "epoch": 0.183875,
+      "grad_norm": 2.5477471351623535,
+      "grad_norm_var": 2.394463361736054,
+      "learning_rate": 0.0001,
+      "loss": 1.0754,
+      "loss/crossentropy": 2.4608027935028076,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.13251473009586334,
+      "loss/reg": 0.014596136286854744,
+      "step": 1471
+    },
+    {
+      "epoch": 0.184,
+      "grad_norm": 2.9321022033691406,
+      "grad_norm_var": 2.3905305167023623,
+      "learning_rate": 0.0001,
+      "loss": 0.997,
+      "loss/crossentropy": 2.4788243770599365,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.10894551128149033,
+      "loss/reg": 0.014588245190680027,
+      "step": 1472
+    },
+    {
+      "epoch": 0.184125,
+      "grad_norm": 3.2976138591766357,
+      "grad_norm_var": 2.3081604420680737,
+      "learning_rate": 0.0001,
+      "loss": 1.0611,
+      "loss/crossentropy": 2.440394878387451,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.13798683881759644,
+      "loss/reg": 0.014581209979951382,
+      "step": 1473
+    },
+    {
+      "epoch": 0.18425,
+      "grad_norm": 2.3695755004882812,
+      "grad_norm_var": 2.378640708203821,
+      "learning_rate": 0.0001,
+      "loss": 1.0597,
+      "loss/crossentropy": 2.457941770553589,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.1365838199853897,
+      "loss/reg": 0.014573454856872559,
+      "step": 1474
+    },
+    {
+      "epoch": 0.184375,
+      "grad_norm": 2.909522771835327,
+      "grad_norm_var": 2.343060112343133,
+      "learning_rate": 0.0001,
+      "loss": 1.0217,
+      "loss/crossentropy": 2.291848659515381,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.10653392970561981,
+      "loss/reg": 0.014565936289727688,
+      "step": 1475
+    },
+    {
+      "epoch": 0.1845,
+      "grad_norm": 4.1714019775390625,
+      "grad_norm_var": 2.349577274287818,
+      "learning_rate": 0.0001,
+      "loss": 1.2516,
+      "loss/crossentropy": 2.4249794483184814,
+      "loss/hidden": 0.9375,
+      "loss/logits": 0.16853143274784088,
+      "loss/reg": 0.0145582165569067,
+      "step": 1476
+    },
+    {
+      "epoch": 0.184625,
+      "grad_norm": 2.30051326751709,
+      "grad_norm_var": 2.4439813338223115,
+      "learning_rate": 0.0001,
+      "loss": 1.0722,
+      "loss/crossentropy": 2.621089220046997,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.13760125637054443,
+      "loss/reg": 0.014550295658409595,
+      "step": 1477
+    },
+    {
+      "epoch": 0.18475,
+      "grad_norm": 2.7384438514709473,
+      "grad_norm_var": 2.4771341504323514,
+      "learning_rate": 0.0001,
+      "loss": 1.1117,
+      "loss/crossentropy": 2.1965439319610596,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.1342887282371521,
+      "loss/reg": 0.014541985467076302,
+      "step": 1478
+    },
+    {
+      "epoch": 0.184875,
+      "grad_norm": 37.360321044921875,
+      "grad_norm_var": 74.16407744545192,
+      "learning_rate": 0.0001,
+      "loss": 0.9989,
+      "loss/crossentropy": 2.9801506996154785,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.10352309793233871,
+      "loss/reg": 0.014533448964357376,
+      "step": 1479
+    },
+    {
+      "epoch": 0.185,
+      "grad_norm": 3.2833774089813232,
+      "grad_norm_var": 73.68816936181202,
+      "learning_rate": 0.0001,
+      "loss": 1.3522,
+      "loss/crossentropy": 2.354642868041992,
+      "loss/hidden": 1.0078125,
+      "loss/logits": 0.19911861419677734,
+      "loss/reg": 0.014524821192026138,
+      "step": 1480
+    },
+    {
+      "epoch": 0.185125,
+      "grad_norm": 2.9734416007995605,
+      "grad_norm_var": 73.78408654274459,
+      "learning_rate": 0.0001,
+      "loss": 1.1487,
+      "loss/crossentropy": 2.5445892810821533,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.1597684621810913,
+      "loss/reg": 0.014516538009047508,
+      "step": 1481
+    },
+    {
+      "epoch": 0.18525,
+      "grad_norm": 3.2212953567504883,
+      "grad_norm_var": 73.88875146417945,
+      "learning_rate": 0.0001,
+      "loss": 1.1972,
+      "loss/crossentropy": 2.4549801349639893,
+      "loss/hidden": 0.875,
+      "loss/logits": 0.17715027928352356,
+      "loss/reg": 0.014508497901260853,
+      "step": 1482
+    },
+    {
+      "epoch": 0.185375,
+      "grad_norm": 3.029637098312378,
+      "grad_norm_var": 73.9933942207774,
+      "learning_rate": 0.0001,
+      "loss": 1.0726,
+      "loss/crossentropy": 2.6153388023376465,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.13069182634353638,
+      "loss/reg": 0.014500816352665424,
+      "step": 1483
+    },
+    {
+      "epoch": 0.1855,
+      "grad_norm": 3.7203376293182373,
+      "grad_norm_var": 73.63538575655532,
+      "learning_rate": 0.0001,
+      "loss": 1.1901,
+      "loss/crossentropy": 2.437302350997925,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.16235879063606262,
+      "loss/reg": 0.014492850750684738,
+      "step": 1484
+    },
+    {
+      "epoch": 0.185625,
+      "grad_norm": 2.606203079223633,
+      "grad_norm_var": 74.04917997908026,
+      "learning_rate": 0.0001,
+      "loss": 0.9748,
+      "loss/crossentropy": 2.3537402153015137,
+      "loss/hidden": 0.71484375,
+      "loss/logits": 0.11513984948396683,
+      "loss/reg": 0.014484620653092861,
+      "step": 1485
+    },
+    {
+      "epoch": 0.18575,
+      "grad_norm": 3.775780200958252,
+      "grad_norm_var": 73.80448419578045,
+      "learning_rate": 0.0001,
+      "loss": 1.1872,
+      "loss/crossentropy": 2.6341981887817383,
+      "loss/hidden": 0.890625,
+      "loss/logits": 0.1517730951309204,
+      "loss/reg": 0.014477075077593327,
+      "step": 1486
+    },
+    {
+      "epoch": 0.185875,
+      "grad_norm": 3.1672568321228027,
+      "grad_norm_var": 73.60919906004487,
+      "learning_rate": 0.0001,
+      "loss": 1.0533,
+      "loss/crossentropy": 2.3641412258148193,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.12350372225046158,
+      "loss/reg": 0.014468939043581486,
+      "step": 1487
+    },
+    {
+      "epoch": 0.186,
+      "grad_norm": 2.9723494052886963,
+      "grad_norm_var": 73.59690980017169,
+      "learning_rate": 0.0001,
+      "loss": 1.1189,
+      "loss/crossentropy": 2.4746789932250977,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.1578400433063507,
+      "loss/reg": 0.014461321756243706,
+      "step": 1488
+    },
+    {
+      "epoch": 0.186125,
+      "grad_norm": 4.781861782073975,
+      "grad_norm_var": 73.34949321986252,
+      "learning_rate": 0.0001,
+      "loss": 1.0247,
+      "loss/crossentropy": 2.6176137924194336,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.11844634264707565,
+      "loss/reg": 0.014453292824327946,
+      "step": 1489
+    },
+    {
+      "epoch": 0.18625,
+      "grad_norm": 3.498753786087036,
+      "grad_norm_var": 72.98251711179161,
+      "learning_rate": 0.0001,
+      "loss": 1.0433,
+      "loss/crossentropy": 2.575167417526245,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.1371680647134781,
+      "loss/reg": 0.014445292763411999,
+      "step": 1490
+    },
+    {
+      "epoch": 0.186375,
+      "grad_norm": 3.3169898986816406,
+      "grad_norm_var": 72.85721374014727,
+      "learning_rate": 0.0001,
+      "loss": 1.2038,
+      "loss/crossentropy": 2.556975841522217,
+      "loss/hidden": 0.8984375,
+      "loss/logits": 0.1610018014907837,
+      "loss/reg": 0.01443762518465519,
+      "step": 1491
+    },
+    {
+      "epoch": 0.1865,
+      "grad_norm": 3.731511354446411,
+      "grad_norm_var": 72.9432662884783,
+      "learning_rate": 0.0001,
+      "loss": 0.967,
+      "loss/crossentropy": 2.7008392810821533,
+      "loss/hidden": 0.6953125,
+      "loss/logits": 0.12734538316726685,
+      "loss/reg": 0.014430060982704163,
+      "step": 1492
+    },
+    {
+      "epoch": 0.186625,
+      "grad_norm": 2.6620421409606934,
+      "grad_norm_var": 72.80179282549148,
+      "learning_rate": 0.0001,
+      "loss": 1.0512,
+      "loss/crossentropy": 2.7235186100006104,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.1257205456495285,
+      "loss/reg": 0.014422285370528698,
+      "step": 1493
+    },
+    {
+      "epoch": 0.18675,
+      "grad_norm": 3.3065989017486572,
+      "grad_norm_var": 72.61826294021522,
+      "learning_rate": 0.0001,
+      "loss": 1.1709,
+      "loss/crossentropy": 2.5472517013549805,
+      "loss/hidden": 0.87109375,
+      "loss/logits": 0.15567487478256226,
+      "loss/reg": 0.014414286240935326,
+      "step": 1494
+    },
+    {
+      "epoch": 0.186875,
+      "grad_norm": 3.0994396209716797,
+      "grad_norm_var": 0.27048224897859136,
+      "learning_rate": 0.0001,
+      "loss": 1.1325,
+      "loss/crossentropy": 2.8537662029266357,
+      "loss/hidden": 0.84765625,
+      "loss/logits": 0.1407473087310791,
+      "loss/reg": 0.014406588859856129,
+      "step": 1495
+    },
+    {
+      "epoch": 0.187,
+      "grad_norm": 2.676715612411499,
+      "grad_norm_var": 0.29658286686653407,
+      "learning_rate": 0.0001,
+      "loss": 1.1095,
+      "loss/crossentropy": 2.4631519317626953,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.1335071474313736,
+      "loss/reg": 0.01439889520406723,
+      "step": 1496
+    },
+    {
+      "epoch": 0.187125,
+      "grad_norm": 3.475898265838623,
+      "grad_norm_var": 0.29157201854104237,
+      "learning_rate": 0.0001,
+      "loss": 1.3724,
+      "loss/crossentropy": 2.6677818298339844,
+      "loss/hidden": 1.03125,
+      "loss/logits": 0.1972627192735672,
+      "loss/reg": 0.014391305856406689,
+      "step": 1497
+    },
+    {
+      "epoch": 0.18725,
+      "grad_norm": 2.769845962524414,
+      "grad_norm_var": 0.30996036390038334,
+      "learning_rate": 0.0001,
+      "loss": 1.0487,
+      "loss/crossentropy": 2.2365801334381104,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.12756550312042236,
+      "loss/reg": 0.014383896254003048,
+      "step": 1498
+    },
+    {
+      "epoch": 0.187375,
+      "grad_norm": 2.8853700160980225,
+      "grad_norm_var": 0.31621077264406544,
+      "learning_rate": 0.0001,
+      "loss": 1.1072,
+      "loss/crossentropy": 2.6836471557617188,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.12754273414611816,
+      "loss/reg": 0.014376661740243435,
+      "step": 1499
+    },
+    {
+      "epoch": 0.1875,
+      "grad_norm": 2.84495210647583,
+      "grad_norm_var": 0.31246808986064234,
+      "learning_rate": 0.0001,
+      "loss": 1.2259,
+      "loss/crossentropy": 2.2098467350006104,
+      "loss/hidden": 0.9140625,
+      "loss/logits": 0.16809730231761932,
+      "loss/reg": 0.014369050972163677,
+      "step": 1500
+    },
+    {
+      "epoch": 0.187625,
+      "grad_norm": 6.990378379821777,
+      "grad_norm_var": 1.1530969883337083,
+      "learning_rate": 0.0001,
+      "loss": 1.3325,
+      "loss/crossentropy": 2.6205296516418457,
+      "loss/hidden": 0.98828125,
+      "loss/logits": 0.2006484866142273,
+      "loss/reg": 0.014361603185534477,
+      "step": 1501
+    },
+    {
+      "epoch": 0.18775,
+      "grad_norm": 3.3288161754608154,
+      "grad_norm_var": 1.1489830243296237,
+      "learning_rate": 0.0001,
+      "loss": 1.2901,
+      "loss/crossentropy": 2.6651721000671387,
+      "loss/hidden": 0.97265625,
+      "loss/logits": 0.17393462359905243,
+      "loss/reg": 0.014354297891259193,
+      "step": 1502
+    },
+    {
+      "epoch": 0.187875,
+      "grad_norm": 2.990226984024048,
+      "grad_norm_var": 1.1580711389422116,
+      "learning_rate": 0.0001,
+      "loss": 1.0356,
+      "loss/crossentropy": 2.6010048389434814,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.13436806201934814,
+      "loss/reg": 0.014346664771437645,
+      "step": 1503
+    },
+    {
+      "epoch": 0.188,
+      "grad_norm": 2.692655086517334,
+      "grad_norm_var": 1.1810803489356902,
+      "learning_rate": 0.0001,
+      "loss": 1.2967,
+      "loss/crossentropy": 2.152822971343994,
+      "loss/hidden": 0.97265625,
+      "loss/logits": 0.1806175410747528,
+      "loss/reg": 0.014339223504066467,
+      "step": 1504
+    },
+    {
+      "epoch": 0.188125,
+      "grad_norm": 87.7072982788086,
+      "grad_norm_var": 445.7985967242888,
+      "learning_rate": 0.0001,
+      "loss": 1.5753,
+      "loss/crossentropy": 2.5603487491607666,
+      "loss/hidden": 1.2890625,
+      "loss/logits": 0.1428820788860321,
+      "loss/reg": 0.014331568963825703,
+      "step": 1505
+    },
+    {
+      "epoch": 0.18825,
+      "grad_norm": 2.2747116088867188,
+      "grad_norm_var": 446.72864180402945,
+      "learning_rate": 0.0001,
+      "loss": 0.9904,
+      "loss/crossentropy": 2.3886001110076904,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.11668366193771362,
+      "loss/reg": 0.014323906973004341,
+      "step": 1506
+    },
+    {
+      "epoch": 0.188375,
+      "grad_norm": 2.267629623413086,
+      "grad_norm_var": 447.52923211089256,
+      "learning_rate": 0.0001,
+      "loss": 1.0243,
+      "loss/crossentropy": 2.423229932785034,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.12335064262151718,
+      "loss/reg": 0.014316298067569733,
+      "step": 1507
+    },
+    {
+      "epoch": 0.1885,
+      "grad_norm": 3.617304801940918,
+      "grad_norm_var": 447.6023780363864,
+      "learning_rate": 0.0001,
+      "loss": 1.3432,
+      "loss/crossentropy": 2.423637866973877,
+      "loss/hidden": 0.98046875,
+      "loss/logits": 0.21964287757873535,
+      "loss/reg": 0.014309005811810493,
+      "step": 1508
+    },
+    {
+      "epoch": 0.188625,
+      "grad_norm": 4.585550308227539,
+      "grad_norm_var": 446.3429466687175,
+      "learning_rate": 0.0001,
+      "loss": 1.3114,
+      "loss/crossentropy": 2.488982677459717,
+      "loss/hidden": 0.984375,
+      "loss/logits": 0.18399442732334137,
+      "loss/reg": 0.014301459304988384,
+      "step": 1509
+    },
+    {
+      "epoch": 0.18875,
+      "grad_norm": 3.1016576290130615,
+      "grad_norm_var": 446.4900686609001,
+      "learning_rate": 0.0001,
+      "loss": 1.0954,
+      "loss/crossentropy": 2.5978121757507324,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.1477963924407959,
+      "loss/reg": 0.014294042252004147,
+      "step": 1510
+    },
+    {
+      "epoch": 0.188875,
+      "grad_norm": 2.6194772720336914,
+      "grad_norm_var": 446.85530854590877,
+      "learning_rate": 0.0001,
+      "loss": 1.1325,
+      "loss/crossentropy": 2.6663565635681152,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.15372540056705475,
+      "loss/reg": 0.014286375604569912,
+      "step": 1511
+    },
+    {
+      "epoch": 0.189,
+      "grad_norm": 7.802513122558594,
+      "grad_norm_var": 444.48216865196844,
+      "learning_rate": 0.0001,
+      "loss": 1.4337,
+      "loss/crossentropy": 2.6862032413482666,
+      "loss/hidden": 1.09375,
+      "loss/logits": 0.19720105826854706,
+      "loss/reg": 0.014278349466621876,
+      "step": 1512
+    },
+    {
+      "epoch": 0.189125,
+      "grad_norm": 3.310786247253418,
+      "grad_norm_var": 444.6026705038085,
+      "learning_rate": 0.0001,
+      "loss": 1.3428,
+      "loss/crossentropy": 2.1258671283721924,
+      "loss/hidden": 1.0234375,
+      "loss/logits": 0.17663809657096863,
+      "loss/reg": 0.014270216226577759,
+      "step": 1513
+    },
+    {
+      "epoch": 0.18925,
+      "grad_norm": 5.729246616363525,
+      "grad_norm_var": 442.7462351862821,
+      "learning_rate": 0.0001,
+      "loss": 1.0873,
+      "loss/crossentropy": 2.9488282203674316,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.1438678652048111,
+      "loss/reg": 0.01426264550536871,
+      "step": 1514
+    },
+    {
+      "epoch": 0.189375,
+      "grad_norm": 3.574831247329712,
+      "grad_norm_var": 442.2095373355806,
+      "learning_rate": 0.0001,
+      "loss": 1.0687,
+      "loss/crossentropy": 2.5907766819000244,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.13321569561958313,
+      "loss/reg": 0.014254805631935596,
+      "step": 1515
+    },
+    {
+      "epoch": 0.1895,
+      "grad_norm": 3.7363853454589844,
+      "grad_norm_var": 441.5169453192193,
+      "learning_rate": 0.0001,
+      "loss": 1.1045,
+      "loss/crossentropy": 2.552393913269043,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.1338808685541153,
+      "loss/reg": 0.014247224666178226,
+      "step": 1516
+    },
+    {
+      "epoch": 0.189625,
+      "grad_norm": 2.745471954345703,
+      "grad_norm_var": 443.8629711327847,
+      "learning_rate": 0.0001,
+      "loss": 1.1382,
+      "loss/crossentropy": 2.7987189292907715,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.15207618474960327,
+      "loss/reg": 0.01423969492316246,
+      "step": 1517
+    },
+    {
+      "epoch": 0.18975,
+      "grad_norm": 2.548691987991333,
+      "grad_norm_var": 444.4784529377907,
+      "learning_rate": 0.0001,
+      "loss": 1.1242,
+      "loss/crossentropy": 2.413057565689087,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.15380065143108368,
+      "loss/reg": 0.014231679029762745,
+      "step": 1518
+    },
+    {
+      "epoch": 0.189875,
+      "grad_norm": 3.604227304458618,
+      "grad_norm_var": 444.0238071702247,
+      "learning_rate": 0.0001,
+      "loss": 1.3671,
+      "loss/crossentropy": 2.107539415359497,
+      "loss/hidden": 1.015625,
+      "loss/logits": 0.20924213528633118,
+      "loss/reg": 0.014223325997591019,
+      "step": 1519
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 2.764598846435547,
+      "grad_norm_var": 443.96487541121735,
+      "learning_rate": 0.0001,
+      "loss": 0.9063,
+      "loss/crossentropy": 2.645019054412842,
+      "loss/hidden": 0.6484375,
+      "loss/logits": 0.11573326587677002,
+      "loss/reg": 0.014215711504220963,
+      "step": 1520
+    },
+    {
+      "epoch": 0.190125,
+      "grad_norm": 3.417343854904175,
+      "grad_norm_var": 2.0384518833410272,
+      "learning_rate": 0.0001,
+      "loss": 1.1809,
+      "loss/crossentropy": 2.0993950366973877,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.1286228746175766,
+      "loss/reg": 0.014208083041012287,
+      "step": 1521
+    },
+    {
+      "epoch": 0.19025,
+      "grad_norm": 2.5339038372039795,
+      "grad_norm_var": 1.996633160561107,
+      "learning_rate": 0.0001,
+      "loss": 1.1835,
+      "loss/crossentropy": 2.3847148418426514,
+      "loss/hidden": 0.890625,
+      "loss/logits": 0.15089885890483856,
+      "loss/reg": 0.014200469478964806,
+      "step": 1522
+    },
+    {
+      "epoch": 0.190375,
+      "grad_norm": 5.412919044494629,
+      "grad_norm_var": 2.046751372081492,
+      "learning_rate": 0.0001,
+      "loss": 1.2804,
+      "loss/crossentropy": 2.647505760192871,
+      "loss/hidden": 1.0078125,
+      "loss/logits": 0.1306157112121582,
+      "loss/reg": 0.014192642644047737,
+      "step": 1523
+    },
+    {
+      "epoch": 0.1905,
+      "grad_norm": 3.2071800231933594,
+      "grad_norm_var": 2.068296485893218,
+      "learning_rate": 0.0001,
+      "loss": 1.0836,
+      "loss/crossentropy": 2.7921669483184814,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.1526501327753067,
+      "loss/reg": 0.01418515294790268,
+      "step": 1524
+    },
+    {
+      "epoch": 0.190625,
+      "grad_norm": 3.779355049133301,
+      "grad_norm_var": 2.0237706183651416,
+      "learning_rate": 0.0001,
+      "loss": 1.0808,
+      "loss/crossentropy": 2.626934289932251,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.1460772454738617,
+      "loss/reg": 0.014177562668919563,
+      "step": 1525
+    },
+    {
+      "epoch": 0.19075,
+      "grad_norm": 2.429072618484497,
+      "grad_norm_var": 2.109561386098513,
+      "learning_rate": 0.0001,
+      "loss": 1.0907,
+      "loss/crossentropy": 2.4317626953125,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.14040398597717285,
+      "loss/reg": 0.014170153997838497,
+      "step": 1526
+    },
+    {
+      "epoch": 0.190875,
+      "grad_norm": 9.4876070022583,
+      "grad_norm_var": 4.06735639009191,
+      "learning_rate": 0.0001,
+      "loss": 1.4633,
+      "loss/crossentropy": 2.2017152309417725,
+      "loss/hidden": 1.15625,
+      "loss/logits": 0.16541144251823425,
+      "loss/reg": 0.014162125997245312,
+      "step": 1527
+    },
+    {
+      "epoch": 0.191,
+      "grad_norm": 3.5674562454223633,
+      "grad_norm_var": 3.1147103692906586,
+      "learning_rate": 0.0001,
+      "loss": 1.2965,
+      "loss/crossentropy": 2.85056734085083,
+      "loss/hidden": 0.98828125,
+      "loss/logits": 0.1666402667760849,
+      "loss/reg": 0.014154158532619476,
+      "step": 1528
+    },
+    {
+      "epoch": 0.191125,
+      "grad_norm": 2.6774816513061523,
+      "grad_norm_var": 3.1866235930450406,
+      "learning_rate": 0.0001,
+      "loss": 1.1206,
+      "loss/crossentropy": 2.367767095565796,
+      "loss/hidden": 0.83984375,
+      "loss/logits": 0.13927701115608215,
+      "loss/reg": 0.014145908877253532,
+      "step": 1529
+    },
+    {
+      "epoch": 0.19125,
+      "grad_norm": 8.835766792297363,
+      "grad_norm_var": 4.578113572841516,
+      "learning_rate": 0.0001,
+      "loss": 1.3634,
+      "loss/crossentropy": 2.98734974861145,
+      "loss/hidden": 1.0625,
+      "loss/logits": 0.15954606235027313,
+      "loss/reg": 0.014136978425085545,
+      "step": 1530
+    },
+    {
+      "epoch": 0.191375,
+      "grad_norm": 2.618042469024658,
+      "grad_norm_var": 4.692138147417478,
+      "learning_rate": 0.0001,
+      "loss": 1.162,
+      "loss/crossentropy": 2.5602381229400635,
+      "loss/hidden": 0.86328125,
+      "loss/logits": 0.15742892026901245,
+      "loss/reg": 0.014127855189144611,
+      "step": 1531
+    },
+    {
+      "epoch": 0.1915,
+      "grad_norm": 2.5733799934387207,
+      "grad_norm_var": 4.811403170073528,
+      "learning_rate": 0.0001,
+      "loss": 1.1001,
+      "loss/crossentropy": 2.511143684387207,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.15818098187446594,
+      "loss/reg": 0.014118584804236889,
+      "step": 1532
+    },
+    {
+      "epoch": 0.191625,
+      "grad_norm": 3.53314208984375,
+      "grad_norm_var": 4.730224432732357,
+      "learning_rate": 0.0001,
+      "loss": 1.1633,
+      "loss/crossentropy": 2.384641647338867,
+      "loss/hidden": 0.88671875,
+      "loss/logits": 0.13548484444618225,
+      "loss/reg": 0.014110967516899109,
+      "step": 1533
+    },
+    {
+      "epoch": 0.19175,
+      "grad_norm": 2.8533082008361816,
+      "grad_norm_var": 4.679641703787122,
+      "learning_rate": 0.0001,
+      "loss": 1.0351,
+      "loss/crossentropy": 2.5344479084014893,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.13237014412879944,
+      "loss/reg": 0.014102863147854805,
+      "step": 1534
+    },
+    {
+      "epoch": 0.191875,
+      "grad_norm": 3.1153576374053955,
+      "grad_norm_var": 4.717503317774359,
+      "learning_rate": 0.0001,
+      "loss": 1.1134,
+      "loss/crossentropy": 2.35892915725708,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.1481958031654358,
+      "loss/reg": 0.014095306396484375,
+      "step": 1535
+    },
+    {
+      "epoch": 0.192,
+      "grad_norm": 2.1978485584259033,
+      "grad_norm_var": 4.825294315312353,
+      "learning_rate": 0.0001,
+      "loss": 1.0504,
+      "loss/crossentropy": 1.9747709035873413,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.12044215202331543,
+      "loss/reg": 0.014087379910051823,
+      "step": 1536
+    },
+    {
+      "epoch": 0.192125,
+      "grad_norm": 3.004746913909912,
+      "grad_norm_var": 4.861933406571506,
+      "learning_rate": 0.0001,
+      "loss": 1.1657,
+      "loss/crossentropy": 2.6981475353240967,
+      "loss/hidden": 0.8671875,
+      "loss/logits": 0.1577630490064621,
+      "loss/reg": 0.014079651795327663,
+      "step": 1537
+    },
+    {
+      "epoch": 0.19225,
+      "grad_norm": 2.8707988262176514,
+      "grad_norm_var": 4.809272805310865,
+      "learning_rate": 0.0001,
+      "loss": 1.0948,
+      "loss/crossentropy": 2.7565841674804688,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.1454852670431137,
+      "loss/reg": 0.01407123077660799,
+      "step": 1538
+    },
+    {
+      "epoch": 0.192375,
+      "grad_norm": 3.2437527179718018,
+      "grad_norm_var": 4.661507493305176,
+      "learning_rate": 0.0001,
+      "loss": 1.268,
+      "loss/crossentropy": 2.3115715980529785,
+      "loss/hidden": 0.953125,
+      "loss/logits": 0.1742616444826126,
+      "loss/reg": 0.014062810689210892,
+      "step": 1539
+    },
+    {
+      "epoch": 0.1925,
+      "grad_norm": 3.202810049057007,
+      "grad_norm_var": 4.661824760391179,
+      "learning_rate": 0.0001,
+      "loss": 1.1887,
+      "loss/crossentropy": 2.351571798324585,
+      "loss/hidden": 0.89453125,
+      "loss/logits": 0.15359902381896973,
+      "loss/reg": 0.014054255560040474,
+      "step": 1540
+    },
+    {
+      "epoch": 0.192625,
+      "grad_norm": 2.8404643535614014,
+      "grad_norm_var": 4.713165856200402,
+      "learning_rate": 0.0001,
+      "loss": 1.0457,
+      "loss/crossentropy": 2.3900203704833984,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.13183440268039703,
+      "loss/reg": 0.014045719988644123,
+      "step": 1541
+    },
+    {
+      "epoch": 0.19275,
+      "grad_norm": 2.3016951084136963,
+      "grad_norm_var": 4.735606807223379,
+      "learning_rate": 0.0001,
+      "loss": 1.1063,
+      "loss/crossentropy": 2.413910150527954,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.14167185127735138,
+      "loss/reg": 0.014036578126251698,
+      "step": 1542
+    },
+    {
+      "epoch": 0.192875,
+      "grad_norm": 2.4764444828033447,
+      "grad_norm_var": 2.381355740482232,
+      "learning_rate": 0.0001,
+      "loss": 1.1498,
+      "loss/crossentropy": 2.508134603500366,
+      "loss/hidden": 0.85546875,
+      "loss/logits": 0.15407797694206238,
+      "loss/reg": 0.014029039070010185,
+      "step": 1543
+    },
+    {
+      "epoch": 0.193,
+      "grad_norm": 3.122025966644287,
+      "grad_norm_var": 2.374577491531939,
+      "learning_rate": 0.0001,
+      "loss": 1.0871,
+      "loss/crossentropy": 2.517812490463257,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.1304890513420105,
+      "loss/reg": 0.014020491391420364,
+      "step": 1544
+    },
+    {
+      "epoch": 0.193125,
+      "grad_norm": 3.5309674739837646,
+      "grad_norm_var": 2.358743795236999,
+      "learning_rate": 0.0001,
+      "loss": 1.3379,
+      "loss/crossentropy": 2.173165798187256,
+      "loss/hidden": 1.015625,
+      "loss/logits": 0.18210922181606293,
+      "loss/reg": 0.014012150466442108,
+      "step": 1545
+    },
+    {
+      "epoch": 0.19325,
+      "grad_norm": 2.8506641387939453,
+      "grad_norm_var": 0.15605408960923697,
+      "learning_rate": 0.0001,
+      "loss": 1.0831,
+      "loss/crossentropy": 2.6542272567749023,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.14616048336029053,
+      "loss/reg": 0.014004606753587723,
+      "step": 1546
+    },
+    {
+      "epoch": 0.193375,
+      "grad_norm": 2.6514549255371094,
+      "grad_norm_var": 0.1548857183604106,
+      "learning_rate": 0.0001,
+      "loss": 1.2542,
+      "loss/crossentropy": 2.267397403717041,
+      "loss/hidden": 0.953125,
+      "loss/logits": 0.16108238697052002,
+      "loss/reg": 0.013996096327900887,
+      "step": 1547
+    },
+    {
+      "epoch": 0.1935,
+      "grad_norm": 3.0364744663238525,
+      "grad_norm_var": 0.1482419605375867,
+      "learning_rate": 0.0001,
+      "loss": 1.1003,
+      "loss/crossentropy": 2.34012770652771,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.14404839277267456,
+      "loss/reg": 0.013987723737955093,
+      "step": 1548
+    },
+    {
+      "epoch": 0.193625,
+      "grad_norm": 3.133777141571045,
+      "grad_norm_var": 0.12593383250846937,
+      "learning_rate": 0.0001,
+      "loss": 1.0985,
+      "loss/crossentropy": 2.618865966796875,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.1383739411830902,
+      "loss/reg": 0.013980153016746044,
+      "step": 1549
+    },
+    {
+      "epoch": 0.19375,
+      "grad_norm": 3.1407878398895264,
+      "grad_norm_var": 0.12923131391317852,
+      "learning_rate": 0.0001,
+      "loss": 1.101,
+      "loss/crossentropy": 2.688640832901001,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.1566149890422821,
+      "loss/reg": 0.013972623273730278,
+      "step": 1550
+    },
+    {
+      "epoch": 0.193875,
+      "grad_norm": 3.174556016921997,
+      "grad_norm_var": 0.1309922878345463,
+      "learning_rate": 0.0001,
+      "loss": 1.1717,
+      "loss/crossentropy": 2.1879935264587402,
+      "loss/hidden": 0.875,
+      "loss/logits": 0.15705443918704987,
+      "loss/reg": 0.013964612036943436,
+      "step": 1551
+    },
+    {
+      "epoch": 0.194,
+      "grad_norm": 3.0509793758392334,
+      "grad_norm_var": 0.09391514491316547,
+      "learning_rate": 0.0001,
+      "loss": 0.9586,
+      "loss/crossentropy": 2.3660058975219727,
+      "loss/hidden": 0.70703125,
+      "loss/logits": 0.11203601211309433,
+      "loss/reg": 0.013956940732896328,
+      "step": 1552
+    },
+    {
+      "epoch": 0.194125,
+      "grad_norm": 3.3187499046325684,
+      "grad_norm_var": 0.10123814801312937,
+      "learning_rate": 0.0001,
+      "loss": 1.1062,
+      "loss/crossentropy": 2.394197463989258,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.150349423289299,
+      "loss/reg": 0.013949030078947544,
+      "step": 1553
+    },
+    {
+      "epoch": 0.19425,
+      "grad_norm": 2.903859853744507,
+      "grad_norm_var": 0.10075169250019347,
+      "learning_rate": 0.0001,
+      "loss": 1.0472,
+      "loss/crossentropy": 2.6103944778442383,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.1500079333782196,
+      "loss/reg": 0.013941409066319466,
+      "step": 1554
+    },
+    {
+      "epoch": 0.194375,
+      "grad_norm": 3.239140272140503,
+      "grad_norm_var": 0.1006023266548631,
+      "learning_rate": 0.0001,
+      "loss": 1.1446,
+      "loss/crossentropy": 2.4417059421539307,
+      "loss/hidden": 0.859375,
+      "loss/logits": 0.1459202617406845,
+      "loss/reg": 0.01393379457294941,
+      "step": 1555
+    },
+    {
+      "epoch": 0.1945,
+      "grad_norm": 2.5746965408325195,
+      "grad_norm_var": 0.10814357204767854,
+      "learning_rate": 0.0001,
+      "loss": 1.0554,
+      "loss/crossentropy": 2.532374620437622,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.1466185450553894,
+      "loss/reg": 0.013926350511610508,
+      "step": 1556
+    },
+    {
+      "epoch": 0.194625,
+      "grad_norm": 3.628814697265625,
+      "grad_norm_var": 0.13450941960769924,
+      "learning_rate": 0.0001,
+      "loss": 1.3249,
+      "loss/crossentropy": 2.3864905834198,
+      "loss/hidden": 1.0234375,
+      "loss/logits": 0.16223978996276855,
+      "loss/reg": 0.013918645679950714,
+      "step": 1557
+    },
+    {
+      "epoch": 0.19475,
+      "grad_norm": 2.513291358947754,
+      "grad_norm_var": 0.11736836954879663,
+      "learning_rate": 0.0001,
+      "loss": 1.0366,
+      "loss/crossentropy": 2.7693920135498047,
+      "loss/hidden": 0.765625,
+      "loss/logits": 0.1318259835243225,
+      "loss/reg": 0.013910802081227303,
+      "step": 1558
+    },
+    {
+      "epoch": 0.194875,
+      "grad_norm": 3.2579643726348877,
+      "grad_norm_var": 0.09872798985575174,
+      "learning_rate": 0.0001,
+      "loss": 1.0427,
+      "loss/crossentropy": 2.501122236251831,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.1263553649187088,
+      "loss/reg": 0.013903032056987286,
+      "step": 1559
+    },
+    {
+      "epoch": 0.195,
+      "grad_norm": 20.420337677001953,
+      "grad_norm_var": 18.91951441564413,
+      "learning_rate": 0.0001,
+      "loss": 1.3618,
+      "loss/crossentropy": 2.46478533744812,
+      "loss/hidden": 1.046875,
+      "loss/logits": 0.1759488880634308,
+      "loss/reg": 0.013895487412810326,
+      "step": 1560
+    },
+    {
+      "epoch": 0.195125,
+      "grad_norm": 2.696916341781616,
+      "grad_norm_var": 19.032016931453608,
+      "learning_rate": 0.0001,
+      "loss": 1.174,
+      "loss/crossentropy": 2.2702722549438477,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.2030649483203888,
+      "loss/reg": 0.013887940905988216,
+      "step": 1561
+    },
+    {
+      "epoch": 0.19525,
+      "grad_norm": 4.7133378982543945,
+      "grad_norm_var": 18.93870030552219,
+      "learning_rate": 0.0001,
+      "loss": 1.2984,
+      "loss/crossentropy": 1.9574859142303467,
+      "loss/hidden": 1.015625,
+      "loss/logits": 0.144018292427063,
+      "loss/reg": 0.013880369253456593,
+      "step": 1562
+    },
+    {
+      "epoch": 0.195375,
+      "grad_norm": 2.689742088317871,
+      "grad_norm_var": 18.930805267251564,
+      "learning_rate": 0.0001,
+      "loss": 1.0472,
+      "loss/crossentropy": 2.5726308822631836,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.1389167606830597,
+      "loss/reg": 0.013872841373085976,
+      "step": 1563
+    },
+    {
+      "epoch": 0.1955,
+      "grad_norm": 3.9613354206085205,
+      "grad_norm_var": 18.838524358177413,
+      "learning_rate": 0.0001,
+      "loss": 1.0574,
+      "loss/crossentropy": 2.2799181938171387,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.14136341214179993,
+      "loss/reg": 0.013865368440747261,
+      "step": 1564
+    },
+    {
+      "epoch": 0.195625,
+      "grad_norm": 2.84377121925354,
+      "grad_norm_var": 18.88795320188418,
+      "learning_rate": 0.0001,
+      "loss": 1.2585,
+      "loss/crossentropy": 2.306931495666504,
+      "loss/hidden": 0.9296875,
+      "loss/logits": 0.19025813043117523,
+      "loss/reg": 0.01385766826570034,
+      "step": 1565
+    },
+    {
+      "epoch": 0.19575,
+      "grad_norm": 3.5409600734710693,
+      "grad_norm_var": 18.83835057402114,
+      "learning_rate": 0.0001,
+      "loss": 1.0235,
+      "loss/crossentropy": 2.611679792404175,
+      "loss/hidden": 0.734375,
+      "loss/logits": 0.15062148869037628,
+      "loss/reg": 0.013849391601979733,
+      "step": 1566
+    },
+    {
+      "epoch": 0.195875,
+      "grad_norm": 2.4071547985076904,
+      "grad_norm_var": 18.988576179472293,
+      "learning_rate": 0.0001,
+      "loss": 0.9748,
+      "loss/crossentropy": 2.4327828884124756,
+      "loss/hidden": 0.7265625,
+      "loss/logits": 0.10979950428009033,
+      "loss/reg": 0.013841108419001102,
+      "step": 1567
+    },
+    {
+      "epoch": 0.196,
+      "grad_norm": 2.5137085914611816,
+      "grad_norm_var": 19.091440757813977,
+      "learning_rate": 0.0001,
+      "loss": 1.0107,
+      "loss/crossentropy": 2.6318089962005615,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.12240324914455414,
+      "loss/reg": 0.013833708129823208,
+      "step": 1568
+    },
+    {
+      "epoch": 0.196125,
+      "grad_norm": 3.096564769744873,
+      "grad_norm_var": 19.120676935364486,
+      "learning_rate": 0.0001,
+      "loss": 1.0895,
+      "loss/crossentropy": 2.374494791030884,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.15435993671417236,
+      "loss/reg": 0.013826224021613598,
+      "step": 1569
+    },
+    {
+      "epoch": 0.19625,
+      "grad_norm": 2.169402837753296,
+      "grad_norm_var": 19.28010469927802,
+      "learning_rate": 0.0001,
+      "loss": 0.9576,
+      "loss/crossentropy": 2.391505479812622,
+      "loss/hidden": 0.6953125,
+      "loss/logits": 0.12412133812904358,
+      "loss/reg": 0.013818818144500256,
+      "step": 1570
+    },
+    {
+      "epoch": 0.196375,
+      "grad_norm": 3.0437140464782715,
+      "grad_norm_var": 19.30600940844608,
+      "learning_rate": 0.0001,
+      "loss": 1.239,
+      "loss/crossentropy": 2.501621723175049,
+      "loss/hidden": 0.9140625,
+      "loss/logits": 0.18683794140815735,
+      "loss/reg": 0.013811350800096989,
+      "step": 1571
+    },
+    {
+      "epoch": 0.1965,
+      "grad_norm": 2.5897915363311768,
+      "grad_norm_var": 19.302894385600194,
+      "learning_rate": 0.0001,
+      "loss": 1.035,
+      "loss/crossentropy": 2.4975814819335938,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.1391228437423706,
+      "loss/reg": 0.013803965412080288,
+      "step": 1572
+    },
+    {
+      "epoch": 0.196625,
+      "grad_norm": 2.516031503677368,
+      "grad_norm_var": 19.454711828122623,
+      "learning_rate": 0.0001,
+      "loss": 0.9646,
+      "loss/crossentropy": 2.3849472999572754,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.12352485954761505,
+      "loss/reg": 0.013796493411064148,
+      "step": 1573
+    },
+    {
+      "epoch": 0.19675,
+      "grad_norm": 2.0846610069274902,
+      "grad_norm_var": 19.55464017386997,
+      "learning_rate": 0.0001,
+      "loss": 1.0915,
+      "loss/crossentropy": 2.333425998687744,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.15285080671310425,
+      "loss/reg": 0.013788769952952862,
+      "step": 1574
+    },
+    {
+      "epoch": 0.196875,
+      "grad_norm": 2.8610305786132812,
+      "grad_norm_var": 19.605563364937975,
+      "learning_rate": 0.0001,
+      "loss": 0.9948,
+      "loss/crossentropy": 2.2638580799102783,
+      "loss/hidden": 0.69140625,
+      "loss/logits": 0.16560885310173035,
+      "loss/reg": 0.013781173154711723,
+      "step": 1575
+    },
+    {
+      "epoch": 0.197,
+      "grad_norm": 2.6086127758026123,
+      "grad_norm_var": 0.45959099378937346,
+      "learning_rate": 0.0001,
+      "loss": 1.1686,
+      "loss/crossentropy": 2.385111093521118,
+      "loss/hidden": 0.87890625,
+      "loss/logits": 0.15191304683685303,
+      "loss/reg": 0.01377350464463234,
+      "step": 1576
+    },
+    {
+      "epoch": 0.197125,
+      "grad_norm": 4.653759002685547,
+      "grad_norm_var": 0.6469626890922541,
+      "learning_rate": 0.0001,
+      "loss": 1.3708,
+      "loss/crossentropy": 2.556640625,
+      "loss/hidden": 1.0390625,
+      "loss/logits": 0.1940462738275528,
+      "loss/reg": 0.013765547424554825,
+      "step": 1577
+    },
+    {
+      "epoch": 0.19725,
+      "grad_norm": 3.648402690887451,
+      "grad_norm_var": 0.4771692938880648,
+      "learning_rate": 0.0001,
+      "loss": 1.0714,
+      "loss/crossentropy": 2.9089317321777344,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.13300660252571106,
+      "loss/reg": 0.013758075423538685,
+      "step": 1578
+    },
+    {
+      "epoch": 0.197375,
+      "grad_norm": 3.6522624492645264,
+      "grad_norm_var": 0.5014419172244267,
+      "learning_rate": 0.0001,
+      "loss": 1.1563,
+      "loss/crossentropy": 2.547022581100464,
+      "loss/hidden": 0.875,
+      "loss/logits": 0.14384308457374573,
+      "loss/reg": 0.013750174082815647,
+      "step": 1579
+    },
+    {
+      "epoch": 0.1975,
+      "grad_norm": 3.541172981262207,
+      "grad_norm_var": 0.4592891725510337,
+      "learning_rate": 0.0001,
+      "loss": 1.3161,
+      "loss/crossentropy": 2.342548370361328,
+      "loss/hidden": 1.015625,
+      "loss/logits": 0.16299855709075928,
+      "loss/reg": 0.013742712326347828,
+      "step": 1580
+    },
+    {
+      "epoch": 0.197625,
+      "grad_norm": 3.876798391342163,
+      "grad_norm_var": 0.506438619715117,
+      "learning_rate": 0.0001,
+      "loss": 1.6102,
+      "loss/crossentropy": 2.066164970397949,
+      "loss/hidden": 1.2265625,
+      "loss/logits": 0.24627353250980377,
+      "loss/reg": 0.01373511552810669,
+      "step": 1581
+    },
+    {
+      "epoch": 0.19775,
+      "grad_norm": 2.5165457725524902,
+      "grad_norm_var": 0.5050025113667628,
+      "learning_rate": 0.0001,
+      "loss": 0.9853,
+      "loss/crossentropy": 2.4709391593933105,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.1175711378455162,
+      "loss/reg": 0.013726145029067993,
+      "step": 1582
+    },
+    {
+      "epoch": 0.197875,
+      "grad_norm": 2.9556968212127686,
+      "grad_norm_var": 0.48145601689114204,
+      "learning_rate": 0.0001,
+      "loss": 1.0748,
+      "loss/crossentropy": 2.4822633266448975,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.1524786353111267,
+      "loss/reg": 0.013718714937567711,
+      "step": 1583
+    },
+    {
+      "epoch": 0.198,
+      "grad_norm": 3.161695957183838,
+      "grad_norm_var": 0.46391222848667796,
+      "learning_rate": 0.0001,
+      "loss": 1.0736,
+      "loss/crossentropy": 2.565737724304199,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.13177385926246643,
+      "loss/reg": 0.013710074126720428,
+      "step": 1584
+    },
+    {
+      "epoch": 0.198125,
+      "grad_norm": 7.7486958503723145,
+      "grad_norm_var": 1.8386121671978246,
+      "learning_rate": 0.0001,
+      "loss": 1.5153,
+      "loss/crossentropy": 2.1480207443237305,
+      "loss/hidden": 1.1484375,
+      "loss/logits": 0.22985613346099854,
+      "loss/reg": 0.013702766969799995,
+      "step": 1585
+    },
+    {
+      "epoch": 0.19825,
+      "grad_norm": 3.0685408115386963,
+      "grad_norm_var": 1.7473924169074466,
+      "learning_rate": 0.0001,
+      "loss": 1.1056,
+      "loss/crossentropy": 2.5632662773132324,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.15228858590126038,
+      "loss/reg": 0.013695274479687214,
+      "step": 1586
+    },
+    {
+      "epoch": 0.198375,
+      "grad_norm": 4.356388568878174,
+      "grad_norm_var": 1.7913349785216173,
+      "learning_rate": 0.0001,
+      "loss": 1.2621,
+      "loss/crossentropy": 2.306581497192383,
+      "loss/hidden": 0.92578125,
+      "loss/logits": 0.19942906498908997,
+      "loss/reg": 0.013687408529222012,
+      "step": 1587
+    },
+    {
+      "epoch": 0.1985,
+      "grad_norm": 2.662060260772705,
+      "grad_norm_var": 1.7829870936881813,
+      "learning_rate": 0.0001,
+      "loss": 1.2003,
+      "loss/crossentropy": 2.5029749870300293,
+      "loss/hidden": 0.8984375,
+      "loss/logits": 0.16509464383125305,
+      "loss/reg": 0.013679537922143936,
+      "step": 1588
+    },
+    {
+      "epoch": 0.198625,
+      "grad_norm": 4.810679912567139,
+      "grad_norm_var": 1.8127030143600467,
+      "learning_rate": 0.0001,
+      "loss": 1.3583,
+      "loss/crossentropy": 2.612169027328491,
+      "loss/hidden": 1.03125,
+      "loss/logits": 0.1903519481420517,
+      "loss/reg": 0.01367125939577818,
+      "step": 1589
+    },
+    {
+      "epoch": 0.19875,
+      "grad_norm": 2.754514217376709,
+      "grad_norm_var": 1.7020179846119938,
+      "learning_rate": 0.0001,
+      "loss": 1.0334,
+      "loss/crossentropy": 2.3626105785369873,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.1428566873073578,
+      "loss/reg": 0.013663833029568195,
+      "step": 1590
+    },
+    {
+      "epoch": 0.198875,
+      "grad_norm": 3.2916183471679688,
+      "grad_norm_var": 1.6665986976673535,
+      "learning_rate": 0.0001,
+      "loss": 1.2137,
+      "loss/crossentropy": 2.2759227752685547,
+      "loss/hidden": 0.87109375,
+      "loss/logits": 0.20607063174247742,
+      "loss/reg": 0.013655973598361015,
+      "step": 1591
+    },
+    {
+      "epoch": 0.199,
+      "grad_norm": 2.276350975036621,
+      "grad_norm_var": 1.7221462363283602,
+      "learning_rate": 0.0001,
+      "loss": 0.9194,
+      "loss/crossentropy": 2.429999351501465,
+      "loss/hidden": 0.671875,
+      "loss/logits": 0.11101213842630386,
+      "loss/reg": 0.013647787272930145,
+      "step": 1592
+    },
+    {
+      "epoch": 0.199125,
+      "grad_norm": 2.7854268550872803,
+      "grad_norm_var": 1.6992207121709098,
+      "learning_rate": 0.0001,
+      "loss": 1.1081,
+      "loss/crossentropy": 2.6592743396759033,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.15529003739356995,
+      "loss/reg": 0.013640275225043297,
+      "step": 1593
+    },
+    {
+      "epoch": 0.19925,
+      "grad_norm": 2.5592005252838135,
+      "grad_norm_var": 1.7618627623489904,
+      "learning_rate": 0.0001,
+      "loss": 1.1146,
+      "loss/crossentropy": 2.27114200592041,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.1267244815826416,
+      "loss/reg": 0.013632478192448616,
+      "step": 1594
+    },
+    {
+      "epoch": 0.199375,
+      "grad_norm": 3.1051416397094727,
+      "grad_norm_var": 1.7695445919247692,
+      "learning_rate": 0.0001,
+      "loss": 1.2779,
+      "loss/crossentropy": 2.096179246902466,
+      "loss/hidden": 0.96875,
+      "loss/logits": 0.1729447841644287,
+      "loss/reg": 0.013624493032693863,
+      "step": 1595
+    },
+    {
+      "epoch": 0.1995,
+      "grad_norm": 2.740203857421875,
+      "grad_norm_var": 1.8017103679937707,
+      "learning_rate": 0.0001,
+      "loss": 0.9738,
+      "loss/crossentropy": 2.5383737087249756,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.12667007744312286,
+      "loss/reg": 0.013617145828902721,
+      "step": 1596
+    },
+    {
+      "epoch": 0.199625,
+      "grad_norm": 2.787034034729004,
+      "grad_norm_var": 1.8091027588667723,
+      "learning_rate": 0.0001,
+      "loss": 1.0191,
+      "loss/crossentropy": 2.452460527420044,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.12520015239715576,
+      "loss/reg": 0.013609787449240685,
+      "step": 1597
+    },
+    {
+      "epoch": 0.19975,
+      "grad_norm": 2.521711587905884,
+      "grad_norm_var": 1.8085312337868815,
+      "learning_rate": 0.0001,
+      "loss": 1.1116,
+      "loss/crossentropy": 2.4604299068450928,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.14743748307228088,
+      "loss/reg": 0.013602245599031448,
+      "step": 1598
+    },
+    {
+      "epoch": 0.199875,
+      "grad_norm": 2.541367530822754,
+      "grad_norm_var": 1.840991450339924,
+      "learning_rate": 0.0001,
+      "loss": 1.1753,
+      "loss/crossentropy": 2.3595387935638428,
+      "loss/hidden": 0.88671875,
+      "loss/logits": 0.1526535153388977,
+      "loss/reg": 0.013594499789178371,
+      "step": 1599
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 3.170604944229126,
+      "grad_norm_var": 1.840804608226538,
+      "learning_rate": 0.0001,
+      "loss": 1.1393,
+      "loss/crossentropy": 2.679455041885376,
+      "loss/hidden": 0.83984375,
+      "loss/logits": 0.16357922554016113,
+      "loss/reg": 0.01358658354729414,
+      "step": 1600
+    },
+    {
+      "epoch": 0.200125,
+      "grad_norm": 3.085852861404419,
+      "grad_norm_var": 0.44862457908304804,
+      "learning_rate": 0.0001,
+      "loss": 1.162,
+      "loss/crossentropy": 2.183800220489502,
+      "loss/hidden": 0.87109375,
+      "loss/logits": 0.15514615178108215,
+      "loss/reg": 0.013579203747212887,
+      "step": 1601
+    },
+    {
+      "epoch": 0.20025,
+      "grad_norm": 2.978541851043701,
+      "grad_norm_var": 0.44869585537156215,
+      "learning_rate": 0.0001,
+      "loss": 1.1595,
+      "loss/crossentropy": 2.7346818447113037,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.1722511649131775,
+      "loss/reg": 0.01357145607471466,
+      "step": 1602
+    },
+    {
+      "epoch": 0.200375,
+      "grad_norm": 2.433056116104126,
+      "grad_norm_var": 0.3388972014701608,
+      "learning_rate": 0.0001,
+      "loss": 1.0578,
+      "loss/crossentropy": 2.400810718536377,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.13697531819343567,
+      "loss/reg": 0.013563701882958412,
+      "step": 1603
+    },
+    {
+      "epoch": 0.2005,
+      "grad_norm": 2.651841640472412,
+      "grad_norm_var": 0.3392367186207063,
+      "learning_rate": 0.0001,
+      "loss": 1.0811,
+      "loss/crossentropy": 2.490342617034912,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.1603870689868927,
+      "loss/reg": 0.013556394726037979,
+      "step": 1604
+    },
+    {
+      "epoch": 0.200625,
+      "grad_norm": 2.8456075191497803,
+      "grad_norm_var": 0.08148981985714249,
+      "learning_rate": 0.0001,
+      "loss": 1.1882,
+      "loss/crossentropy": 2.4219372272491455,
+      "loss/hidden": 0.87890625,
+      "loss/logits": 0.17379163205623627,
+      "loss/reg": 0.0135487737134099,
+      "step": 1605
+    },
+    {
+      "epoch": 0.20075,
+      "grad_norm": 2.8339900970458984,
+      "grad_norm_var": 0.08158268879521771,
+      "learning_rate": 0.0001,
+      "loss": 0.9949,
+      "loss/crossentropy": 2.611527442932129,
+      "loss/hidden": 0.734375,
+      "loss/logits": 0.12512920796871185,
+      "loss/reg": 0.013541026972234249,
+      "step": 1606
+    },
+    {
+      "epoch": 0.200875,
+      "grad_norm": 3.8811702728271484,
+      "grad_norm_var": 0.14289600365006525,
+      "learning_rate": 0.0001,
+      "loss": 1.0861,
+      "loss/crossentropy": 2.5142064094543457,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.13831937313079834,
+      "loss/reg": 0.013533033430576324,
+      "step": 1607
+    },
+    {
+      "epoch": 0.201,
+      "grad_norm": 3.6074378490448,
+      "grad_norm_var": 0.15629189387123668,
+      "learning_rate": 0.0001,
+      "loss": 1.0655,
+      "loss/crossentropy": 2.4579451084136963,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.13723407685756683,
+      "loss/reg": 0.013525336049497128,
+      "step": 1608
+    },
+    {
+      "epoch": 0.201125,
+      "grad_norm": 2.8290226459503174,
+      "grad_norm_var": 0.1556981224441491,
+      "learning_rate": 0.0001,
+      "loss": 1.0043,
+      "loss/crossentropy": 2.7764294147491455,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.1308579295873642,
+      "loss/reg": 0.01351715624332428,
+      "step": 1609
+    },
+    {
+      "epoch": 0.20125,
+      "grad_norm": 2.59934401512146,
+      "grad_norm_var": 0.15391725674414733,
+      "learning_rate": 0.0001,
+      "loss": 1.0636,
+      "loss/crossentropy": 2.5147807598114014,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.13549384474754333,
+      "loss/reg": 0.013509852811694145,
+      "step": 1610
+    },
+    {
+      "epoch": 0.201375,
+      "grad_norm": 2.7972164154052734,
+      "grad_norm_var": 0.15196475783560098,
+      "learning_rate": 0.0001,
+      "loss": 1.0975,
+      "loss/crossentropy": 2.507530689239502,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.14603210985660553,
+      "loss/reg": 0.013502140529453754,
+      "step": 1611
+    },
+    {
+      "epoch": 0.2015,
+      "grad_norm": 3.2658374309539795,
+      "grad_norm_var": 0.15845418736495442,
+      "learning_rate": 0.0001,
+      "loss": 1.2549,
+      "loss/crossentropy": 2.164236307144165,
+      "loss/hidden": 0.9453125,
+      "loss/logits": 0.17467688024044037,
+      "loss/reg": 0.01349355187267065,
+      "step": 1612
+    },
+    {
+      "epoch": 0.201625,
+      "grad_norm": 2.7505931854248047,
+      "grad_norm_var": 0.15921652951525,
+      "learning_rate": 0.0001,
+      "loss": 1.0594,
+      "loss/crossentropy": 2.0924651622772217,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.12380584329366684,
+      "loss/reg": 0.013486127369105816,
+      "step": 1613
+    },
+    {
+      "epoch": 0.20175,
+      "grad_norm": 3.203188180923462,
+      "grad_norm_var": 0.15163660104867607,
+      "learning_rate": 0.0001,
+      "loss": 1.1208,
+      "loss/crossentropy": 2.62121319770813,
+      "loss/hidden": 0.84765625,
+      "loss/logits": 0.1383514702320099,
+      "loss/reg": 0.013478783890604973,
+      "step": 1614
+    },
+    {
+      "epoch": 0.201875,
+      "grad_norm": 2.6951982975006104,
+      "grad_norm_var": 0.1443821198786272,
+      "learning_rate": 0.0001,
+      "loss": 1.4372,
+      "loss/crossentropy": 2.364288091659546,
+      "loss/hidden": 1.109375,
+      "loss/logits": 0.19316649436950684,
+      "loss/reg": 0.013470477424561977,
+      "step": 1615
+    },
+    {
+      "epoch": 0.202,
+      "grad_norm": 2.867556095123291,
+      "grad_norm_var": 0.1422903014164942,
+      "learning_rate": 0.0001,
+      "loss": 1.0695,
+      "loss/crossentropy": 2.412200689315796,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.1301771104335785,
+      "loss/reg": 0.01346264686435461,
+      "step": 1616
+    },
+    {
+      "epoch": 0.202125,
+      "grad_norm": 3.35479736328125,
+      "grad_norm_var": 0.15140141291244522,
+      "learning_rate": 0.0001,
+      "loss": 1.1988,
+      "loss/crossentropy": 2.4715232849121094,
+      "loss/hidden": 0.90234375,
+      "loss/logits": 0.16192206740379333,
+      "loss/reg": 0.013454729691147804,
+      "step": 1617
+    },
+    {
+      "epoch": 0.20225,
+      "grad_norm": 22.905237197875977,
+      "grad_norm_var": 24.978816029077244,
+      "learning_rate": 0.0001,
+      "loss": 1.579,
+      "loss/crossentropy": 2.2932474613189697,
+      "loss/hidden": 1.21875,
+      "loss/logits": 0.22579804062843323,
+      "loss/reg": 0.013446752913296223,
+      "step": 1618
+    },
+    {
+      "epoch": 0.202375,
+      "grad_norm": 2.676539182662964,
+      "grad_norm_var": 24.92450698201823,
+      "learning_rate": 0.0001,
+      "loss": 1.0158,
+      "loss/crossentropy": 2.443284034729004,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.13141301274299622,
+      "loss/reg": 0.01343945600092411,
+      "step": 1619
+    },
+    {
+      "epoch": 0.2025,
+      "grad_norm": 3.8005261421203613,
+      "grad_norm_var": 24.764457157993366,
+      "learning_rate": 0.0001,
+      "loss": 1.2057,
+      "loss/crossentropy": 2.7221662998199463,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.1611778438091278,
+      "loss/reg": 0.013432328589260578,
+      "step": 1620
+    },
+    {
+      "epoch": 0.202625,
+      "grad_norm": 2.675564765930176,
+      "grad_norm_var": 24.79939933153828,
+      "learning_rate": 0.0001,
+      "loss": 1.0176,
+      "loss/crossentropy": 2.7226831912994385,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.12948712706565857,
+      "loss/reg": 0.013424508273601532,
+      "step": 1621
+    },
+    {
+      "epoch": 0.20275,
+      "grad_norm": 5.532473564147949,
+      "grad_norm_var": 24.728322365826084,
+      "learning_rate": 0.0001,
+      "loss": 1.393,
+      "loss/crossentropy": 2.501004219055176,
+      "loss/hidden": 1.0546875,
+      "loss/logits": 0.20409145951271057,
+      "loss/reg": 0.013417219743132591,
+      "step": 1622
+    },
+    {
+      "epoch": 0.202875,
+      "grad_norm": 2.9599437713623047,
+      "grad_norm_var": 24.853088500483913,
+      "learning_rate": 0.0001,
+      "loss": 1.047,
+      "loss/crossentropy": 2.583350419998169,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.1238275021314621,
+      "loss/reg": 0.013409611769020557,
+      "step": 1623
+    },
+    {
+      "epoch": 0.203,
+      "grad_norm": 2.609393358230591,
+      "grad_norm_var": 25.021814610211056,
+      "learning_rate": 0.0001,
+      "loss": 1.0136,
+      "loss/crossentropy": 2.800400495529175,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.13346146047115326,
+      "loss/reg": 0.013402425684034824,
+      "step": 1624
+    },
+    {
+      "epoch": 0.203125,
+      "grad_norm": 3.3347935676574707,
+      "grad_norm_var": 24.93556049048077,
+      "learning_rate": 0.0001,
+      "loss": 1.3162,
+      "loss/crossentropy": 2.306166648864746,
+      "loss/hidden": 1.0,
+      "loss/logits": 0.1822924166917801,
+      "loss/reg": 0.013394915498793125,
+      "step": 1625
+    },
+    {
+      "epoch": 0.20325,
+      "grad_norm": 2.487501859664917,
+      "grad_norm_var": 24.9628476598421,
+      "learning_rate": 0.0001,
+      "loss": 1.163,
+      "loss/crossentropy": 2.2325022220611572,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.1463523954153061,
+      "loss/reg": 0.013387140817940235,
+      "step": 1626
+    },
+    {
+      "epoch": 0.203375,
+      "grad_norm": 2.961826801300049,
+      "grad_norm_var": 24.930026653662896,
+      "learning_rate": 0.0001,
+      "loss": 1.1983,
+      "loss/crossentropy": 2.5841550827026367,
+      "loss/hidden": 0.85546875,
+      "loss/logits": 0.2090195268392563,
+      "loss/reg": 0.013379319570958614,
+      "step": 1627
+    },
+    {
+      "epoch": 0.2035,
+      "grad_norm": 2.5991106033325195,
+      "grad_norm_var": 25.056860448715973,
+      "learning_rate": 0.0001,
+      "loss": 1.0763,
+      "loss/crossentropy": 2.617375135421753,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.1457059681415558,
+      "loss/reg": 0.01337137445807457,
+      "step": 1628
+    },
+    {
+      "epoch": 0.203625,
+      "grad_norm": 3.130793571472168,
+      "grad_norm_var": 24.985404162175104,
+      "learning_rate": 0.0001,
+      "loss": 1.4514,
+      "loss/crossentropy": 2.28818678855896,
+      "loss/hidden": 1.09375,
+      "loss/logits": 0.2240520417690277,
+      "loss/reg": 0.013363334350287914,
+      "step": 1629
+    },
+    {
+      "epoch": 0.20375,
+      "grad_norm": 3.3310132026672363,
+      "grad_norm_var": 24.96667274126064,
+      "learning_rate": 0.0001,
+      "loss": 1.2145,
+      "loss/crossentropy": 2.4561591148376465,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.1981697976589203,
+      "loss/reg": 0.013356066308915615,
+      "step": 1630
+    },
+    {
+      "epoch": 0.203875,
+      "grad_norm": 2.9992592334747314,
+      "grad_norm_var": 24.904546403803987,
+      "learning_rate": 0.0001,
+      "loss": 1.3327,
+      "loss/crossentropy": 2.304450035095215,
+      "loss/hidden": 1.03125,
+      "loss/logits": 0.16797912120819092,
+      "loss/reg": 0.013348528183996677,
+      "step": 1631
+    },
+    {
+      "epoch": 0.204,
+      "grad_norm": 3.1939244270324707,
+      "grad_norm_var": 24.844990519484032,
+      "learning_rate": 0.0001,
+      "loss": 1.0774,
+      "loss/crossentropy": 2.4736955165863037,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.15497320890426636,
+      "loss/reg": 0.013341384008526802,
+      "step": 1632
+    },
+    {
+      "epoch": 0.204125,
+      "grad_norm": 2.8169233798980713,
+      "grad_norm_var": 24.938715041703063,
+      "learning_rate": 0.0001,
+      "loss": 0.8683,
+      "loss/crossentropy": 2.676666736602783,
+      "loss/hidden": 0.62109375,
+      "loss/logits": 0.11391356587409973,
+      "loss/reg": 0.013333736918866634,
+      "step": 1633
+    },
+    {
+      "epoch": 0.20425,
+      "grad_norm": 3.69463849067688,
+      "grad_norm_var": 0.5429387753821554,
+      "learning_rate": 0.0001,
+      "loss": 1.7185,
+      "loss/crossentropy": 1.9171524047851562,
+      "loss/hidden": 1.3125,
+      "loss/logits": 0.27270910143852234,
+      "loss/reg": 0.013325790874660015,
+      "step": 1634
+    },
+    {
+      "epoch": 0.204375,
+      "grad_norm": 2.5067968368530273,
+      "grad_norm_var": 0.5560268531523621,
+      "learning_rate": 0.0001,
+      "loss": 0.9318,
+      "loss/crossentropy": 2.6572604179382324,
+      "loss/hidden": 0.6875,
+      "loss/logits": 0.11111121624708176,
+      "loss/reg": 0.013317680917680264,
+      "step": 1635
+    },
+    {
+      "epoch": 0.2045,
+      "grad_norm": 2.7445805072784424,
+      "grad_norm_var": 0.5361896610226305,
+      "learning_rate": 0.0001,
+      "loss": 1.1343,
+      "loss/crossentropy": 2.34808087348938,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.16526776552200317,
+      "loss/reg": 0.013309688307344913,
+      "step": 1636
+    },
+    {
+      "epoch": 0.204625,
+      "grad_norm": 2.5069262981414795,
+      "grad_norm_var": 0.5474804142573985,
+      "learning_rate": 0.0001,
+      "loss": 0.9285,
+      "loss/crossentropy": 2.678297758102417,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.11579211056232452,
+      "loss/reg": 0.013302477076649666,
+      "step": 1637
+    },
+    {
+      "epoch": 0.20475,
+      "grad_norm": 3.034487247467041,
+      "grad_norm_var": 0.12334773017239489,
+      "learning_rate": 0.0001,
+      "loss": 1.1895,
+      "loss/crossentropy": 2.0104122161865234,
+      "loss/hidden": 0.90234375,
+      "loss/logits": 0.15417572855949402,
+      "loss/reg": 0.013295282609760761,
+      "step": 1638
+    },
+    {
+      "epoch": 0.204875,
+      "grad_norm": 2.7470128536224365,
+      "grad_norm_var": 0.12538795384078488,
+      "learning_rate": 0.0001,
+      "loss": 1.0896,
+      "loss/crossentropy": 2.2556824684143066,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.13637785613536835,
+      "loss/reg": 0.013288102112710476,
+      "step": 1639
+    },
+    {
+      "epoch": 0.205,
+      "grad_norm": 2.6958811283111572,
+      "grad_norm_var": 0.12228878695745683,
+      "learning_rate": 0.0001,
+      "loss": 0.9741,
+      "loss/crossentropy": 2.2803361415863037,
+      "loss/hidden": 0.72265625,
+      "loss/logits": 0.11862225830554962,
+      "loss/reg": 0.013280958868563175,
+      "step": 1640
+    },
+    {
+      "epoch": 0.205125,
+      "grad_norm": 2.923374652862549,
+      "grad_norm_var": 0.11033848957714554,
+      "learning_rate": 0.0001,
+      "loss": 1.1012,
+      "loss/crossentropy": 2.5998592376708984,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.14425665140151978,
+      "loss/reg": 0.013273877091705799,
+      "step": 1641
+    },
+    {
+      "epoch": 0.20525,
+      "grad_norm": 16.231281280517578,
+      "grad_norm_var": 11.163123044335308,
+      "learning_rate": 0.0001,
+      "loss": 1.573,
+      "loss/crossentropy": 2.28498911857605,
+      "loss/hidden": 1.1875,
+      "loss/logits": 0.25285494327545166,
+      "loss/reg": 0.013266587629914284,
+      "step": 1642
+    },
+    {
+      "epoch": 0.205375,
+      "grad_norm": 2.8149280548095703,
+      "grad_norm_var": 11.180053543888178,
+      "learning_rate": 0.0001,
+      "loss": 1.1024,
+      "loss/crossentropy": 2.148362636566162,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.15731269121170044,
+      "loss/reg": 0.013259139843285084,
+      "step": 1643
+    },
+    {
+      "epoch": 0.2055,
+      "grad_norm": 3.9090774059295654,
+      "grad_norm_var": 11.086604757622807,
+      "learning_rate": 0.0001,
+      "loss": 1.4303,
+      "loss/crossentropy": 2.6009087562561035,
+      "loss/hidden": 1.0703125,
+      "loss/logits": 0.22749975323677063,
+      "loss/reg": 0.01325154397636652,
+      "step": 1644
+    },
+    {
+      "epoch": 0.205625,
+      "grad_norm": 2.7664973735809326,
+      "grad_norm_var": 11.128864400454058,
+      "learning_rate": 0.0001,
+      "loss": 0.9888,
+      "loss/crossentropy": 2.576356887817383,
+      "loss/hidden": 0.734375,
+      "loss/logits": 0.12194088101387024,
+      "loss/reg": 0.013244031928479671,
+      "step": 1645
+    },
+    {
+      "epoch": 0.20575,
+      "grad_norm": 2.7838664054870605,
+      "grad_norm_var": 11.182320606560461,
+      "learning_rate": 0.0001,
+      "loss": 1.0593,
+      "loss/crossentropy": 2.458827018737793,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.12612512707710266,
+      "loss/reg": 0.013236268423497677,
+      "step": 1646
+    },
+    {
+      "epoch": 0.205875,
+      "grad_norm": 3.0026252269744873,
+      "grad_norm_var": 11.181974019665732,
+      "learning_rate": 0.0001,
+      "loss": 1.0012,
+      "loss/crossentropy": 2.8010799884796143,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.1305844783782959,
+      "loss/reg": 0.01322907768189907,
+      "step": 1647
+    },
+    {
+      "epoch": 0.206,
+      "grad_norm": 3.544288158416748,
+      "grad_norm_var": 11.162580503190616,
+      "learning_rate": 0.0001,
+      "loss": 1.2099,
+      "loss/crossentropy": 2.5440893173217773,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.19490712881088257,
+      "loss/reg": 0.013221465051174164,
+      "step": 1648
+    },
+    {
+      "epoch": 0.206125,
+      "grad_norm": 3.2531473636627197,
+      "grad_norm_var": 11.117574070283142,
+      "learning_rate": 0.0001,
+      "loss": 1.3131,
+      "loss/crossentropy": 2.267230272293091,
+      "loss/hidden": 0.98828125,
+      "loss/logits": 0.19262857735157013,
+      "loss/reg": 0.013214114122092724,
+      "step": 1649
+    },
+    {
+      "epoch": 0.20625,
+      "grad_norm": 2.65549635887146,
+      "grad_norm_var": 11.202772982286575,
+      "learning_rate": 0.0001,
+      "loss": 1.1064,
+      "loss/crossentropy": 2.623307228088379,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.16180500388145447,
+      "loss/reg": 0.013206473551690578,
+      "step": 1650
+    },
+    {
+      "epoch": 0.206375,
+      "grad_norm": 3.4816501140594482,
+      "grad_norm_var": 11.09960005034767,
+      "learning_rate": 0.0001,
+      "loss": 1.1543,
+      "loss/crossentropy": 2.644336462020874,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.17855095863342285,
+      "loss/reg": 0.013198580592870712,
+      "step": 1651
+    },
+    {
+      "epoch": 0.2065,
+      "grad_norm": 2.8406617641448975,
+      "grad_norm_var": 11.086419925476456,
+      "learning_rate": 0.0001,
+      "loss": 0.9424,
+      "loss/crossentropy": 2.5468101501464844,
+      "loss/hidden": 0.69921875,
+      "loss/logits": 0.11126389354467392,
+      "loss/reg": 0.01318978238850832,
+      "step": 1652
+    },
+    {
+      "epoch": 0.206625,
+      "grad_norm": 2.7195427417755127,
+      "grad_norm_var": 11.051894988141786,
+      "learning_rate": 0.0001,
+      "loss": 1.2921,
+      "loss/crossentropy": 2.5399060249328613,
+      "loss/hidden": 0.96484375,
+      "loss/logits": 0.19546636939048767,
+      "loss/reg": 0.013182584196329117,
+      "step": 1653
+    },
+    {
+      "epoch": 0.20675,
+      "grad_norm": 2.4805049896240234,
+      "grad_norm_var": 11.13040761168501,
+      "learning_rate": 0.0001,
+      "loss": 0.9454,
+      "loss/crossentropy": 2.6860992908477783,
+      "loss/hidden": 0.69921875,
+      "loss/logits": 0.11445553600788116,
+      "loss/reg": 0.013174121268093586,
+      "step": 1654
+    },
+    {
+      "epoch": 0.206875,
+      "grad_norm": 2.9009270668029785,
+      "grad_norm_var": 11.110214998589447,
+      "learning_rate": 0.0001,
+      "loss": 1.1267,
+      "loss/crossentropy": 2.6687192916870117,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.15907853841781616,
+      "loss/reg": 0.013165648095309734,
+      "step": 1655
+    },
+    {
+      "epoch": 0.207,
+      "grad_norm": 5.767265796661377,
+      "grad_norm_var": 11.242431274213367,
+      "learning_rate": 0.0001,
+      "loss": 1.3322,
+      "loss/crossentropy": 2.3070762157440186,
+      "loss/hidden": 1.0234375,
+      "loss/logits": 0.17719680070877075,
+      "loss/reg": 0.013158419169485569,
+      "step": 1656
+    },
+    {
+      "epoch": 0.207125,
+      "grad_norm": 2.9326095581054688,
+      "grad_norm_var": 11.241105151169359,
+      "learning_rate": 0.0001,
+      "loss": 1.2628,
+      "loss/crossentropy": 2.429621934890747,
+      "loss/hidden": 0.94921875,
+      "loss/logits": 0.18210232257843018,
+      "loss/reg": 0.013150298967957497,
+      "step": 1657
+    },
+    {
+      "epoch": 0.20725,
+      "grad_norm": 3.11118483543396,
+      "grad_norm_var": 0.6121424659486439,
+      "learning_rate": 0.0001,
+      "loss": 1.0707,
+      "loss/crossentropy": 2.6597509384155273,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.15797501802444458,
+      "loss/reg": 0.013143090531229973,
+      "step": 1658
+    },
+    {
+      "epoch": 0.207375,
+      "grad_norm": 3.273897886276245,
+      "grad_norm_var": 0.6026450391734357,
+      "learning_rate": 0.0001,
+      "loss": 1.5868,
+      "loss/crossentropy": 2.4786179065704346,
+      "loss/hidden": 1.1953125,
+      "loss/logits": 0.2601345479488373,
+      "loss/reg": 0.013135283254086971,
+      "step": 1659
+    },
+    {
+      "epoch": 0.2075,
+      "grad_norm": 2.313868761062622,
+      "grad_norm_var": 0.6138390088935656,
+      "learning_rate": 0.0001,
+      "loss": 1.0272,
+      "loss/crossentropy": 2.646815538406372,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.1224367693066597,
+      "loss/reg": 0.013128082267940044,
+      "step": 1660
+    },
+    {
+      "epoch": 0.207625,
+      "grad_norm": 3.9266319274902344,
+      "grad_norm_var": 0.6441662093447725,
+      "learning_rate": 0.0001,
+      "loss": 1.3995,
+      "loss/crossentropy": 2.480496644973755,
+      "loss/hidden": 1.0859375,
+      "loss/logits": 0.1823263317346573,
+      "loss/reg": 0.013120830059051514,
+      "step": 1661
+    },
+    {
+      "epoch": 0.20775,
+      "grad_norm": 4.4009881019592285,
+      "grad_norm_var": 0.7207383535866235,
+      "learning_rate": 0.0001,
+      "loss": 1.5143,
+      "loss/crossentropy": 2.490128993988037,
+      "loss/hidden": 1.15625,
+      "loss/logits": 0.22691452503204346,
+      "loss/reg": 0.013114040717482567,
+      "step": 1662
+    },
+    {
+      "epoch": 0.207875,
+      "grad_norm": 2.803945302963257,
+      "grad_norm_var": 0.7307607399187871,
+      "learning_rate": 0.0001,
+      "loss": 1.1357,
+      "loss/crossentropy": 2.6528525352478027,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.1609269231557846,
+      "loss/reg": 0.013106826692819595,
+      "step": 1663
+    },
+    {
+      "epoch": 0.208,
+      "grad_norm": 3.9250094890594482,
+      "grad_norm_var": 0.753468894736533,
+      "learning_rate": 0.0001,
+      "loss": 1.5158,
+      "loss/crossentropy": 2.8934035301208496,
+      "loss/hidden": 1.1484375,
+      "loss/logits": 0.23635676503181458,
+      "loss/reg": 0.013099766336381435,
+      "step": 1664
+    },
+    {
+      "epoch": 0.208125,
+      "grad_norm": 2.537315607070923,
+      "grad_norm_var": 0.789891085089917,
+      "learning_rate": 0.0001,
+      "loss": 1.1371,
+      "loss/crossentropy": 2.4187686443328857,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.15466302633285522,
+      "loss/reg": 0.013092422857880592,
+      "step": 1665
+    },
+    {
+      "epoch": 0.20825,
+      "grad_norm": 2.553142547607422,
+      "grad_norm_var": 0.7987201352076717,
+      "learning_rate": 0.0001,
+      "loss": 1.0203,
+      "loss/crossentropy": 2.137171745300293,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.12768462300300598,
+      "loss/reg": 0.013086005114018917,
+      "step": 1666
+    },
+    {
+      "epoch": 0.208375,
+      "grad_norm": 2.1522605419158936,
+      "grad_norm_var": 0.8677726892771486,
+      "learning_rate": 0.0001,
+      "loss": 0.9736,
+      "loss/crossentropy": 2.4442384243011475,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.13187186419963837,
+      "loss/reg": 0.013079563155770302,
+      "step": 1667
+    },
+    {
+      "epoch": 0.2085,
+      "grad_norm": 2.8897452354431152,
+      "grad_norm_var": 0.8658007433698373,
+      "learning_rate": 0.0001,
+      "loss": 0.9685,
+      "loss/crossentropy": 2.4599223136901855,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.12683042883872986,
+      "loss/reg": 0.01307291816920042,
+      "step": 1668
+    },
+    {
+      "epoch": 0.208625,
+      "grad_norm": 3.2847912311553955,
+      "grad_norm_var": 0.8519672623501406,
+      "learning_rate": 0.0001,
+      "loss": 0.9332,
+      "loss/crossentropy": 2.2193775177001953,
+      "loss/hidden": 0.6796875,
+      "loss/logits": 0.12289424240589142,
+      "loss/reg": 0.013066742569208145,
+      "step": 1669
+    },
+    {
+      "epoch": 0.20875,
+      "grad_norm": 3.1088175773620605,
+      "grad_norm_var": 0.8160818976558214,
+      "learning_rate": 0.0001,
+      "loss": 1.2098,
+      "loss/crossentropy": 2.402890920639038,
+      "loss/hidden": 0.93359375,
+      "loss/logits": 0.14563772082328796,
+      "loss/reg": 0.013059607706964016,
+      "step": 1670
+    },
+    {
+      "epoch": 0.208875,
+      "grad_norm": 3.5626771450042725,
+      "grad_norm_var": 0.8133001054088592,
+      "learning_rate": 0.0001,
+      "loss": 1.1257,
+      "loss/crossentropy": 2.4826173782348633,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.15928024053573608,
+      "loss/reg": 0.013052698224782944,
+      "step": 1671
+    },
+    {
+      "epoch": 0.209,
+      "grad_norm": 3.8422510623931885,
+      "grad_norm_var": 0.40753121137815923,
+      "learning_rate": 0.0001,
+      "loss": 1.0944,
+      "loss/crossentropy": 2.2613954544067383,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.14750930666923523,
+      "loss/reg": 0.013046123087406158,
+      "step": 1672
+    },
+    {
+      "epoch": 0.209125,
+      "grad_norm": 3.1732969284057617,
+      "grad_norm_var": 0.40373591532357217,
+      "learning_rate": 0.0001,
+      "loss": 1.148,
+      "loss/crossentropy": 2.6913723945617676,
+      "loss/hidden": 0.84765625,
+      "loss/logits": 0.1699780523777008,
+      "loss/reg": 0.01303982362151146,
+      "step": 1673
+    },
+    {
+      "epoch": 0.20925,
+      "grad_norm": 3.5349199771881104,
+      "grad_norm_var": 0.41114120511671504,
+      "learning_rate": 0.0001,
+      "loss": 1.1601,
+      "loss/crossentropy": 2.636218547821045,
+      "loss/hidden": 0.859375,
+      "loss/logits": 0.1703943908214569,
+      "loss/reg": 0.013032618910074234,
+      "step": 1674
+    },
+    {
+      "epoch": 0.209375,
+      "grad_norm": 3.3266332149505615,
+      "grad_norm_var": 0.4117979013874437,
+      "learning_rate": 0.0001,
+      "loss": 1.2931,
+      "loss/crossentropy": 2.5062716007232666,
+      "loss/hidden": 0.96875,
+      "loss/logits": 0.1940596103668213,
+      "loss/reg": 0.013026205822825432,
+      "step": 1675
+    },
+    {
+      "epoch": 0.2095,
+      "grad_norm": 2.690976142883301,
+      "grad_norm_var": 0.3757021597893411,
+      "learning_rate": 0.0001,
+      "loss": 1.0795,
+      "loss/crossentropy": 2.6262998580932617,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.14850963652133942,
+      "loss/reg": 0.013019601814448833,
+      "step": 1676
+    },
+    {
+      "epoch": 0.209625,
+      "grad_norm": 2.262477397918701,
+      "grad_norm_var": 0.3946797642913109,
+      "learning_rate": 0.0001,
+      "loss": 1.0682,
+      "loss/crossentropy": 2.522778272628784,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.14901478588581085,
+      "loss/reg": 0.013013252057135105,
+      "step": 1677
+    },
+    {
+      "epoch": 0.20975,
+      "grad_norm": 4.060770034790039,
+      "grad_norm_var": 0.3441717651006473,
+      "learning_rate": 0.0001,
+      "loss": 1.1066,
+      "loss/crossentropy": 2.75931715965271,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.1445423662662506,
+      "loss/reg": 0.0130059365183115,
+      "step": 1678
+    },
+    {
+      "epoch": 0.209875,
+      "grad_norm": 3.035757064819336,
+      "grad_norm_var": 0.3381691610716331,
+      "learning_rate": 0.0001,
+      "loss": 1.2199,
+      "loss/crossentropy": 2.2356534004211426,
+      "loss/hidden": 0.9140625,
+      "loss/logits": 0.1757938712835312,
+      "loss/reg": 0.012999509461224079,
+      "step": 1679
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.4442572593688965,
+      "grad_norm_var": 0.3010964780836121,
+      "learning_rate": 0.0001,
+      "loss": 1.3513,
+      "loss/crossentropy": 2.327789545059204,
+      "loss/hidden": 1.03125,
+      "loss/logits": 0.19014956057071686,
+      "loss/reg": 0.012993012554943562,
+      "step": 1680
+    },
+    {
+      "epoch": 0.210125,
+      "grad_norm": 3.582969903945923,
+      "grad_norm_var": 0.29220290919850694,
+      "learning_rate": 0.0001,
+      "loss": 1.1276,
+      "loss/crossentropy": 2.5549631118774414,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.18918761610984802,
+      "loss/reg": 0.0129857761785388,
+      "step": 1681
+    },
+    {
+      "epoch": 0.21025,
+      "grad_norm": 2.9530413150787354,
+      "grad_norm_var": 0.27002111859183725,
+      "learning_rate": 0.0001,
+      "loss": 1.051,
+      "loss/crossentropy": 2.4788119792938232,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.14384132623672485,
+      "loss/reg": 0.012978510931134224,
+      "step": 1682
+    },
+    {
+      "epoch": 0.210375,
+      "grad_norm": 2.191362142562866,
+      "grad_norm_var": 0.2647501539546122,
+      "learning_rate": 0.0001,
+      "loss": 1.0957,
+      "loss/crossentropy": 2.532787561416626,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.15740235149860382,
+      "loss/reg": 0.012971782125532627,
+      "step": 1683
+    },
+    {
+      "epoch": 0.2105,
+      "grad_norm": 5.248608112335205,
+      "grad_norm_var": 0.5199526136507975,
+      "learning_rate": 0.0001,
+      "loss": 1.3987,
+      "loss/crossentropy": 2.741248369216919,
+      "loss/hidden": 1.0625,
+      "loss/logits": 0.206559956073761,
+      "loss/reg": 0.012965119443833828,
+      "step": 1684
+    },
+    {
+      "epoch": 0.210625,
+      "grad_norm": 9.923733711242676,
+      "grad_norm_var": 3.2333504676999985,
+      "learning_rate": 0.0001,
+      "loss": 1.2908,
+      "loss/crossentropy": 2.4856460094451904,
+      "loss/hidden": 1.0078125,
+      "loss/logits": 0.1534312665462494,
+      "loss/reg": 0.012957965023815632,
+      "step": 1685
+    },
+    {
+      "epoch": 0.21075,
+      "grad_norm": 2.7674450874328613,
+      "grad_norm_var": 3.2696547533182656,
+      "learning_rate": 0.0001,
+      "loss": 1.094,
+      "loss/crossentropy": 2.563523054122925,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.13640597462654114,
+      "loss/reg": 0.012951286509633064,
+      "step": 1686
+    },
+    {
+      "epoch": 0.210875,
+      "grad_norm": 2.63311505317688,
+      "grad_norm_var": 3.3437877784147436,
+      "learning_rate": 0.0001,
+      "loss": 1.0298,
+      "loss/crossentropy": 2.604543685913086,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.13861128687858582,
+      "loss/reg": 0.012944560497999191,
+      "step": 1687
+    },
+    {
+      "epoch": 0.211,
+      "grad_norm": 10.589808464050293,
+      "grad_norm_var": 6.347074021055989,
+      "learning_rate": 0.0001,
+      "loss": 1.3073,
+      "loss/crossentropy": 2.7445929050445557,
+      "loss/hidden": 1.0078125,
+      "loss/logits": 0.1701403260231018,
+      "loss/reg": 0.01293735858052969,
+      "step": 1688
+    },
+    {
+      "epoch": 0.211125,
+      "grad_norm": 2.85196590423584,
+      "grad_norm_var": 6.392746951466915,
+      "learning_rate": 0.0001,
+      "loss": 1.0699,
+      "loss/crossentropy": 2.5519180297851562,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.12032581865787506,
+      "loss/reg": 0.012930169701576233,
+      "step": 1689
+    },
+    {
+      "epoch": 0.21125,
+      "grad_norm": 3.8103320598602295,
+      "grad_norm_var": 6.377889547085541,
+      "learning_rate": 0.0001,
+      "loss": 1.1586,
+      "loss/crossentropy": 2.479715585708618,
+      "loss/hidden": 0.87890625,
+      "loss/logits": 0.15049707889556885,
+      "loss/reg": 0.012922849506139755,
+      "step": 1690
+    },
+    {
+      "epoch": 0.211375,
+      "grad_norm": 2.617124080657959,
+      "grad_norm_var": 6.481173027892926,
+      "learning_rate": 0.0001,
+      "loss": 1.0754,
+      "loss/crossentropy": 2.60794734954834,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.14545781910419464,
+      "loss/reg": 0.012915358878672123,
+      "step": 1691
+    },
+    {
+      "epoch": 0.2115,
+      "grad_norm": 3.9162254333496094,
+      "grad_norm_var": 6.354372430431797,
+      "learning_rate": 0.0001,
+      "loss": 1.1995,
+      "loss/crossentropy": 2.4997646808624268,
+      "loss/hidden": 0.921875,
+      "loss/logits": 0.1485539972782135,
+      "loss/reg": 0.012908329255878925,
+      "step": 1692
+    },
+    {
+      "epoch": 0.211625,
+      "grad_norm": 2.7217462062835693,
+      "grad_norm_var": 6.2539271325365196,
+      "learning_rate": 0.0001,
+      "loss": 1.1512,
+      "loss/crossentropy": 2.617048501968384,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.22927230596542358,
+      "loss/reg": 0.012900839559733868,
+      "step": 1693
+    },
+    {
+      "epoch": 0.21175,
+      "grad_norm": 2.224349021911621,
+      "grad_norm_var": 6.485761495009887,
+      "learning_rate": 0.0001,
+      "loss": 0.9654,
+      "loss/crossentropy": 2.703951358795166,
+      "loss/hidden": 0.72265625,
+      "loss/logits": 0.11382674425840378,
+      "loss/reg": 0.012893814593553543,
+      "step": 1694
+    },
+    {
+      "epoch": 0.211875,
+      "grad_norm": 2.9117870330810547,
+      "grad_norm_var": 6.503189101884489,
+      "learning_rate": 0.0001,
+      "loss": 1.2238,
+      "loss/crossentropy": 2.5064597129821777,
+      "loss/hidden": 0.8984375,
+      "loss/logits": 0.1964527666568756,
+      "loss/reg": 0.012886927463114262,
+      "step": 1695
+    },
+    {
+      "epoch": 0.212,
+      "grad_norm": 2.943073034286499,
+      "grad_norm_var": 6.557645425580934,
+      "learning_rate": 0.0001,
+      "loss": 1.0268,
+      "loss/crossentropy": 2.681680679321289,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.14798036217689514,
+      "loss/reg": 0.012880067341029644,
+      "step": 1696
+    },
+    {
+      "epoch": 0.212125,
+      "grad_norm": 2.8364195823669434,
+      "grad_norm_var": 6.633285254118307,
+      "learning_rate": 0.0001,
+      "loss": 1.2396,
+      "loss/crossentropy": 2.267443895339966,
+      "loss/hidden": 0.94140625,
+      "loss/logits": 0.16948428750038147,
+      "loss/reg": 0.012873291037976742,
+      "step": 1697
+    },
+    {
+      "epoch": 0.21225,
+      "grad_norm": 2.667727470397949,
+      "grad_norm_var": 6.676156819217446,
+      "learning_rate": 0.0001,
+      "loss": 1.1879,
+      "loss/crossentropy": 2.5819482803344727,
+      "loss/hidden": 0.8671875,
+      "loss/logits": 0.1920512467622757,
+      "loss/reg": 0.012866680510342121,
+      "step": 1698
+    },
+    {
+      "epoch": 0.212375,
+      "grad_norm": 3.2915964126586914,
+      "grad_norm_var": 6.496990351425703,
+      "learning_rate": 0.0001,
+      "loss": 1.0907,
+      "loss/crossentropy": 2.5162808895111084,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.13395392894744873,
+      "loss/reg": 0.012859686277806759,
+      "step": 1699
+    },
+    {
+      "epoch": 0.2125,
+      "grad_norm": 2.9825503826141357,
+      "grad_norm_var": 6.439824510110865,
+      "learning_rate": 0.0001,
+      "loss": 1.2178,
+      "loss/crossentropy": 2.2996246814727783,
+      "loss/hidden": 0.9140625,
+      "loss/logits": 0.1752283126115799,
+      "loss/reg": 0.01285255141556263,
+      "step": 1700
+    },
+    {
+      "epoch": 0.212625,
+      "grad_norm": 2.9510700702667236,
+      "grad_norm_var": 3.836942936103522,
+      "learning_rate": 0.0001,
+      "loss": 1.3336,
+      "loss/crossentropy": 2.3370251655578613,
+      "loss/hidden": 1.015625,
+      "loss/logits": 0.1895022988319397,
+      "loss/reg": 0.01284568477421999,
+      "step": 1701
+    },
+    {
+      "epoch": 0.21275,
+      "grad_norm": 2.5308520793914795,
+      "grad_norm_var": 3.8610195504163127,
+      "learning_rate": 0.0001,
+      "loss": 1.2202,
+      "loss/crossentropy": 2.513655185699463,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.18170946836471558,
+      "loss/reg": 0.01283828355371952,
+      "step": 1702
+    },
+    {
+      "epoch": 0.212875,
+      "grad_norm": 4.751686096191406,
+      "grad_norm_var": 3.923506474684037,
+      "learning_rate": 0.0001,
+      "loss": 1.2352,
+      "loss/crossentropy": 2.173074245452881,
+      "loss/hidden": 0.953125,
+      "loss/logits": 0.15377816557884216,
+      "loss/reg": 0.012831181287765503,
+      "step": 1703
+    },
+    {
+      "epoch": 0.213,
+      "grad_norm": 3.511111259460449,
+      "grad_norm_var": 0.3989999503182124,
+      "learning_rate": 0.0001,
+      "loss": 1.2827,
+      "loss/crossentropy": 2.21486759185791,
+      "loss/hidden": 0.9765625,
+      "loss/logits": 0.17789113521575928,
+      "loss/reg": 0.01282414235174656,
+      "step": 1704
+    },
+    {
+      "epoch": 0.213125,
+      "grad_norm": 2.6096677780151367,
+      "grad_norm_var": 0.4105200098953181,
+      "learning_rate": 0.0001,
+      "loss": 1.0013,
+      "loss/crossentropy": 2.316843271255493,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.12707841396331787,
+      "loss/reg": 0.012817141599953175,
+      "step": 1705
+    },
+    {
+      "epoch": 0.21325,
+      "grad_norm": 2.8105530738830566,
+      "grad_norm_var": 0.375613954977198,
+      "learning_rate": 0.0001,
+      "loss": 1.0661,
+      "loss/crossentropy": 2.369736909866333,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.13333451747894287,
+      "loss/reg": 0.012809785082936287,
+      "step": 1706
+    },
+    {
+      "epoch": 0.213375,
+      "grad_norm": 3.1986374855041504,
+      "grad_norm_var": 0.36571755056712973,
+      "learning_rate": 0.0001,
+      "loss": 1.057,
+      "loss/crossentropy": 2.320181131362915,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.12431928515434265,
+      "loss/reg": 0.012802771292626858,
+      "step": 1707
+    },
+    {
+      "epoch": 0.2135,
+      "grad_norm": 2.6183996200561523,
+      "grad_norm_var": 0.32173357495411,
+      "learning_rate": 0.0001,
+      "loss": 1.0431,
+      "loss/crossentropy": 2.4588027000427246,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.13782772421836853,
+      "loss/reg": 0.01279559638351202,
+      "step": 1708
+    },
+    {
+      "epoch": 0.213625,
+      "grad_norm": 2.6844642162323,
+      "grad_norm_var": 0.3230673077730707,
+      "learning_rate": 0.0001,
+      "loss": 1.12,
+      "loss/crossentropy": 2.0865495204925537,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.1561318039894104,
+      "loss/reg": 0.012788429856300354,
+      "step": 1709
+    },
+    {
+      "epoch": 0.21375,
+      "grad_norm": 15.15482234954834,
+      "grad_norm_var": 9.486914195081253,
+      "learning_rate": 0.0001,
+      "loss": 1.095,
+      "loss/crossentropy": 2.5281543731689453,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.14299717545509338,
+      "loss/reg": 0.012780962511897087,
+      "step": 1710
+    },
+    {
+      "epoch": 0.213875,
+      "grad_norm": 3.1808626651763916,
+      "grad_norm_var": 9.460348003251534,
+      "learning_rate": 0.0001,
+      "loss": 1.1502,
+      "loss/crossentropy": 2.383219003677368,
+      "loss/hidden": 0.89453125,
+      "loss/logits": 0.12796291708946228,
+      "loss/reg": 0.012773919850587845,
+      "step": 1711
+    },
+    {
+      "epoch": 0.214,
+      "grad_norm": 4.163309574127197,
+      "grad_norm_var": 9.414766565786032,
+      "learning_rate": 0.0001,
+      "loss": 1.3702,
+      "loss/crossentropy": 2.455665111541748,
+      "loss/hidden": 1.03125,
+      "loss/logits": 0.21129587292671204,
+      "loss/reg": 0.012766973115503788,
+      "step": 1712
+    },
+    {
+      "epoch": 0.214125,
+      "grad_norm": 2.4835665225982666,
+      "grad_norm_var": 9.471244857981635,
+      "learning_rate": 0.0001,
+      "loss": 1.0306,
+      "loss/crossentropy": 2.1449975967407227,
+      "loss/hidden": 0.765625,
+      "loss/logits": 0.13736189901828766,
+      "loss/reg": 0.012759552337229252,
+      "step": 1713
+    },
+    {
+      "epoch": 0.21425,
+      "grad_norm": 3.456458806991577,
+      "grad_norm_var": 9.385853171996628,
+      "learning_rate": 0.0001,
+      "loss": 1.0432,
+      "loss/crossentropy": 2.746030330657959,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.1344452202320099,
+      "loss/reg": 0.01275256834924221,
+      "step": 1714
+    },
+    {
+      "epoch": 0.214375,
+      "grad_norm": 2.5408084392547607,
+      "grad_norm_var": 9.481860031377286,
+      "learning_rate": 0.0001,
+      "loss": 1.1865,
+      "loss/crossentropy": 2.4518415927886963,
+      "loss/hidden": 0.875,
+      "loss/logits": 0.18404340744018555,
+      "loss/reg": 0.01274561882019043,
+      "step": 1715
+    },
+    {
+      "epoch": 0.2145,
+      "grad_norm": 2.634857654571533,
+      "grad_norm_var": 9.529713299554377,
+      "learning_rate": 0.0001,
+      "loss": 1.0157,
+      "loss/crossentropy": 2.7000675201416016,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.14611287415027618,
+      "loss/reg": 0.012738562189042568,
+      "step": 1716
+    },
+    {
+      "epoch": 0.214625,
+      "grad_norm": 2.735212564468384,
+      "grad_norm_var": 9.557923964972339,
+      "learning_rate": 0.0001,
+      "loss": 1.1027,
+      "loss/crossentropy": 2.5324275493621826,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.14335589110851288,
+      "loss/reg": 0.012731647118926048,
+      "step": 1717
+    },
+    {
+      "epoch": 0.21475,
+      "grad_norm": 2.8685081005096436,
+      "grad_norm_var": 9.50716521368737,
+      "learning_rate": 0.0001,
+      "loss": 1.0448,
+      "loss/crossentropy": 2.559062957763672,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.14023086428642273,
+      "loss/reg": 0.012724741362035275,
+      "step": 1718
+    },
+    {
+      "epoch": 0.214875,
+      "grad_norm": 2.7883400917053223,
+      "grad_norm_var": 9.508818411578824,
+      "learning_rate": 0.0001,
+      "loss": 1.1709,
+      "loss/crossentropy": 2.7375028133392334,
+      "loss/hidden": 0.86328125,
+      "loss/logits": 0.18043699860572815,
+      "loss/reg": 0.012717757374048233,
+      "step": 1719
+    },
+    {
+      "epoch": 0.215,
+      "grad_norm": 4.499538421630859,
+      "grad_norm_var": 9.54301307944679,
+      "learning_rate": 0.0001,
+      "loss": 1.394,
+      "loss/crossentropy": 2.3297677040100098,
+      "loss/hidden": 1.046875,
+      "loss/logits": 0.2200247049331665,
+      "loss/reg": 0.012710826471447945,
+      "step": 1720
+    },
+    {
+      "epoch": 0.215125,
+      "grad_norm": 2.6797218322753906,
+      "grad_norm_var": 9.53241861946436,
+      "learning_rate": 0.0001,
+      "loss": 1.2027,
+      "loss/crossentropy": 2.094529628753662,
+      "loss/hidden": 0.92578125,
+      "loss/logits": 0.1498618870973587,
+      "loss/reg": 0.012703881599009037,
+      "step": 1721
+    },
+    {
+      "epoch": 0.21525,
+      "grad_norm": 2.300706148147583,
+      "grad_norm_var": 9.614644455093018,
+      "learning_rate": 0.0001,
+      "loss": 1.056,
+      "loss/crossentropy": 2.341108560562134,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.14775308966636658,
+      "loss/reg": 0.012696735560894012,
+      "step": 1722
+    },
+    {
+      "epoch": 0.215375,
+      "grad_norm": 2.584787607192993,
+      "grad_norm_var": 9.683262071884398,
+      "learning_rate": 0.0001,
+      "loss": 1.2349,
+      "loss/crossentropy": 2.511127471923828,
+      "loss/hidden": 0.94140625,
+      "loss/logits": 0.16660267114639282,
+      "loss/reg": 0.012690168805420399,
+      "step": 1723
+    },
+    {
+      "epoch": 0.2155,
+      "grad_norm": 2.7051329612731934,
+      "grad_norm_var": 9.67109810339239,
+      "learning_rate": 0.0001,
+      "loss": 1.0733,
+      "loss/crossentropy": 2.2387161254882812,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.14958390593528748,
+      "loss/reg": 0.012683309614658356,
+      "step": 1724
+    },
+    {
+      "epoch": 0.215625,
+      "grad_norm": 2.6227893829345703,
+      "grad_norm_var": 9.679821099755788,
+      "learning_rate": 0.0001,
+      "loss": 1.0231,
+      "loss/crossentropy": 2.2900543212890625,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.1268245130777359,
+      "loss/reg": 0.012676582671701908,
+      "step": 1725
+    },
+    {
+      "epoch": 0.21575,
+      "grad_norm": 3.2791683673858643,
+      "grad_norm_var": 0.3762032236898106,
+      "learning_rate": 0.0001,
+      "loss": 1.3265,
+      "loss/crossentropy": 2.2019684314727783,
+      "loss/hidden": 1.0078125,
+      "loss/logits": 0.19199709594249725,
+      "loss/reg": 0.01266949437558651,
+      "step": 1726
+    },
+    {
+      "epoch": 0.215875,
+      "grad_norm": 3.5843939781188965,
+      "grad_norm_var": 0.3977131857555946,
+      "learning_rate": 0.0001,
+      "loss": 1.1419,
+      "loss/crossentropy": 2.474095106124878,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.16370782256126404,
+      "loss/reg": 0.012662646360695362,
+      "step": 1727
+    },
+    {
+      "epoch": 0.216,
+      "grad_norm": 4.510242938995361,
+      "grad_norm_var": 0.45925816137897757,
+      "learning_rate": 0.0001,
+      "loss": 1.0694,
+      "loss/crossentropy": 2.829896926879883,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.13811606168746948,
+      "loss/reg": 0.012655883096158504,
+      "step": 1728
+    },
+    {
+      "epoch": 0.216125,
+      "grad_norm": 3.1752772331237793,
+      "grad_norm_var": 0.4399517134814609,
+      "learning_rate": 0.0001,
+      "loss": 1.2447,
+      "loss/crossentropy": 2.6229610443115234,
+      "loss/hidden": 0.92578125,
+      "loss/logits": 0.19241932034492493,
+      "loss/reg": 0.012648857198655605,
+      "step": 1729
+    },
+    {
+      "epoch": 0.21625,
+      "grad_norm": 3.531876802444458,
+      "grad_norm_var": 0.44429015337232514,
+      "learning_rate": 0.0001,
+      "loss": 1.2492,
+      "loss/crossentropy": 2.466808795928955,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.2125817835330963,
+      "loss/reg": 0.01264151744544506,
+      "step": 1730
+    },
+    {
+      "epoch": 0.216375,
+      "grad_norm": 7.509683609008789,
+      "grad_norm_var": 1.6400556058435427,
+      "learning_rate": 0.0001,
+      "loss": 1.6584,
+      "loss/crossentropy": 2.631517171859741,
+      "loss/hidden": 1.2890625,
+      "loss/logits": 0.24304136633872986,
+      "loss/reg": 0.012634233571588993,
+      "step": 1731
+    },
+    {
+      "epoch": 0.2165,
+      "grad_norm": 2.4990897178649902,
+      "grad_norm_var": 1.6546175936024687,
+      "learning_rate": 0.0001,
+      "loss": 1.041,
+      "loss/crossentropy": 2.5932819843292236,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.13742858171463013,
+      "loss/reg": 0.012627250514924526,
+      "step": 1732
+    },
+    {
+      "epoch": 0.216625,
+      "grad_norm": 2.389944553375244,
+      "grad_norm_var": 1.6911601234778078,
+      "learning_rate": 0.0001,
+      "loss": 0.9812,
+      "loss/crossentropy": 2.318802833557129,
+      "loss/hidden": 0.73046875,
+      "loss/logits": 0.12450764328241348,
+      "loss/reg": 0.012620464898645878,
+      "step": 1733
+    },
+    {
+      "epoch": 0.21675,
+      "grad_norm": 2.6159985065460205,
+      "grad_norm_var": 1.7112070581365304,
+      "learning_rate": 0.0001,
+      "loss": 1.141,
+      "loss/crossentropy": 2.716378688812256,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.17106443643569946,
+      "loss/reg": 0.012613578699529171,
+      "step": 1734
+    },
+    {
+      "epoch": 0.216875,
+      "grad_norm": 3.3391213417053223,
+      "grad_norm_var": 1.6904040902109745,
+      "learning_rate": 0.0001,
+      "loss": 1.1009,
+      "loss/crossentropy": 2.4364097118377686,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.14670366048812866,
+      "loss/reg": 0.012606512755155563,
+      "step": 1735
+    },
+    {
+      "epoch": 0.217,
+      "grad_norm": 3.596266984939575,
+      "grad_norm_var": 1.604664018069381,
+      "learning_rate": 0.0001,
+      "loss": 1.2259,
+      "loss/crossentropy": 2.486934185028076,
+      "loss/hidden": 0.94140625,
+      "loss/logits": 0.15848477184772491,
+      "loss/reg": 0.012599713169038296,
+      "step": 1736
+    },
+    {
+      "epoch": 0.217125,
+      "grad_norm": 2.6140332221984863,
+      "grad_norm_var": 1.6104343887098584,
+      "learning_rate": 0.0001,
+      "loss": 1.0859,
+      "loss/crossentropy": 2.3189902305603027,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.13571876287460327,
+      "loss/reg": 0.012592630460858345,
+      "step": 1737
+    },
+    {
+      "epoch": 0.21725,
+      "grad_norm": 2.87418532371521,
+      "grad_norm_var": 1.554299756346355,
+      "learning_rate": 0.0001,
+      "loss": 1.2125,
+      "loss/crossentropy": 2.095885992050171,
+      "loss/hidden": 0.91796875,
+      "loss/logits": 0.16862446069717407,
+      "loss/reg": 0.012585851363837719,
+      "step": 1738
+    },
+    {
+      "epoch": 0.217375,
+      "grad_norm": 4.191568851470947,
+      "grad_norm_var": 1.5539712836144928,
+      "learning_rate": 0.0001,
+      "loss": 1.7198,
+      "loss/crossentropy": 2.424238681793213,
+      "loss/hidden": 1.3046875,
+      "loss/logits": 0.2893022298812866,
+      "loss/reg": 0.012578519992530346,
+      "step": 1739
+    },
+    {
+      "epoch": 0.2175,
+      "grad_norm": 2.666806936264038,
+      "grad_norm_var": 1.5578179682411237,
+      "learning_rate": 0.0001,
+      "loss": 1.1622,
+      "loss/crossentropy": 2.2649426460266113,
+      "loss/hidden": 0.875,
+      "loss/logits": 0.16145509481430054,
+      "loss/reg": 0.01257177721709013,
+      "step": 1740
+    },
+    {
+      "epoch": 0.217625,
+      "grad_norm": 2.6806042194366455,
+      "grad_norm_var": 1.5517463474106703,
+      "learning_rate": 0.0001,
+      "loss": 1.0853,
+      "loss/crossentropy": 2.4488718509674072,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.15884113311767578,
+      "loss/reg": 0.012565212324261665,
+      "step": 1741
+    },
+    {
+      "epoch": 0.21775,
+      "grad_norm": 3.704366683959961,
+      "grad_norm_var": 1.5538631925025423,
+      "learning_rate": 0.0001,
+      "loss": 1.2354,
+      "loss/crossentropy": 2.6138522624969482,
+      "loss/hidden": 0.94140625,
+      "loss/logits": 0.16840386390686035,
+      "loss/reg": 0.012558743357658386,
+      "step": 1742
+    },
+    {
+      "epoch": 0.217875,
+      "grad_norm": 3.2077255249023438,
+      "grad_norm_var": 1.556870797982151,
+      "learning_rate": 0.0001,
+      "loss": 1.316,
+      "loss/crossentropy": 2.141016721725464,
+      "loss/hidden": 1.0390625,
+      "loss/logits": 0.15139830112457275,
+      "loss/reg": 0.012552441097795963,
+      "step": 1743
+    },
+    {
+      "epoch": 0.218,
+      "grad_norm": 2.951503038406372,
+      "grad_norm_var": 1.4871620619995438,
+      "learning_rate": 0.0001,
+      "loss": 1.0989,
+      "loss/crossentropy": 2.5490455627441406,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.1530894637107849,
+      "loss/reg": 0.012546169571578503,
+      "step": 1744
+    },
+    {
+      "epoch": 0.218125,
+      "grad_norm": 3.119734525680542,
+      "grad_norm_var": 1.488624773240887,
+      "learning_rate": 0.0001,
+      "loss": 1.0844,
+      "loss/crossentropy": 2.4211771488189697,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.14257250726222992,
+      "loss/reg": 0.012539266608655453,
+      "step": 1745
+    },
+    {
+      "epoch": 0.21825,
+      "grad_norm": 3.663892984390259,
+      "grad_norm_var": 1.4930337180673443,
+      "learning_rate": 0.0001,
+      "loss": 1.0072,
+      "loss/crossentropy": 2.431640148162842,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.11231502890586853,
+      "loss/reg": 0.012532477267086506,
+      "step": 1746
+    },
+    {
+      "epoch": 0.218375,
+      "grad_norm": 2.77474045753479,
+      "grad_norm_var": 0.26911648905078756,
+      "learning_rate": 0.0001,
+      "loss": 1.0839,
+      "loss/crossentropy": 2.7117726802825928,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.15006102621555328,
+      "loss/reg": 0.012525715865194798,
+      "step": 1747
+    },
+    {
+      "epoch": 0.2185,
+      "grad_norm": 2.6455776691436768,
+      "grad_norm_var": 0.25958807313815746,
+      "learning_rate": 0.0001,
+      "loss": 1.0041,
+      "loss/crossentropy": 2.7506678104400635,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.1328078657388687,
+      "loss/reg": 0.012519202195107937,
+      "step": 1748
+    },
+    {
+      "epoch": 0.218625,
+      "grad_norm": 24.219947814941406,
+      "grad_norm_var": 28.07975632569756,
+      "learning_rate": 0.0001,
+      "loss": 1.4577,
+      "loss/crossentropy": 2.543936014175415,
+      "loss/hidden": 1.203125,
+      "loss/logits": 0.12944281101226807,
+      "loss/reg": 0.01251268945634365,
+      "step": 1749
+    },
+    {
+      "epoch": 0.21875,
+      "grad_norm": 37.90777587890625,
+      "grad_norm_var": 97.39228721131126,
+      "learning_rate": 0.0001,
+      "loss": 1.1124,
+      "loss/crossentropy": 2.2267796993255615,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.1357845962047577,
+      "loss/reg": 0.012506171129643917,
+      "step": 1750
+    },
+    {
+      "epoch": 0.218875,
+      "grad_norm": 3.590038537979126,
+      "grad_norm_var": 97.28596098453178,
+      "learning_rate": 0.0001,
+      "loss": 1.0614,
+      "loss/crossentropy": 2.42260479927063,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.1316903978586197,
+      "loss/reg": 0.012499329634010792,
+      "step": 1751
+    },
+    {
+      "epoch": 0.219,
+      "grad_norm": 2.845022439956665,
+      "grad_norm_var": 97.6271689383777,
+      "learning_rate": 0.0001,
+      "loss": 1.092,
+      "loss/crossentropy": 2.804971218109131,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.14280739426612854,
+      "loss/reg": 0.012492484413087368,
+      "step": 1752
+    },
+    {
+      "epoch": 0.219125,
+      "grad_norm": 3.189939260482788,
+      "grad_norm_var": 97.34154979157398,
+      "learning_rate": 0.0001,
+      "loss": 1.1496,
+      "loss/crossentropy": 2.448812484741211,
+      "loss/hidden": 0.875,
+      "loss/logits": 0.14971600472927094,
+      "loss/reg": 0.012485613115131855,
+      "step": 1753
+    },
+    {
+      "epoch": 0.21925,
+      "grad_norm": 2.670297145843506,
+      "grad_norm_var": 97.44651079060357,
+      "learning_rate": 0.0001,
+      "loss": 1.1915,
+      "loss/crossentropy": 2.4728586673736572,
+      "loss/hidden": 0.87890625,
+      "loss/logits": 0.1877635270357132,
+      "loss/reg": 0.012478794902563095,
+      "step": 1754
+    },
+    {
+      "epoch": 0.219375,
+      "grad_norm": 3.3437023162841797,
+      "grad_norm_var": 97.76674601970879,
+      "learning_rate": 0.0001,
+      "loss": 1.1638,
+      "loss/crossentropy": 2.3619091510772705,
+      "loss/hidden": 0.87890625,
+      "loss/logits": 0.1601634919643402,
+      "loss/reg": 0.012471921741962433,
+      "step": 1755
+    },
+    {
+      "epoch": 0.2195,
+      "grad_norm": 3.5836129188537598,
+      "grad_norm_var": 97.34167870831958,
+      "learning_rate": 0.0001,
+      "loss": 1.3367,
+      "loss/crossentropy": 2.671821355819702,
+      "loss/hidden": 1.03125,
+      "loss/logits": 0.1808364987373352,
+      "loss/reg": 0.012465192936360836,
+      "step": 1756
+    },
+    {
+      "epoch": 0.219625,
+      "grad_norm": 2.6130692958831787,
+      "grad_norm_var": 97.37753712214686,
+      "learning_rate": 0.0001,
+      "loss": 0.9876,
+      "loss/crossentropy": 2.5102763175964355,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.12473314255475998,
+      "loss/reg": 0.01245811302214861,
+      "step": 1757
+    },
+    {
+      "epoch": 0.21975,
+      "grad_norm": 3.395615339279175,
+      "grad_norm_var": 97.50380796911385,
+      "learning_rate": 0.0001,
+      "loss": 1.2064,
+      "loss/crossentropy": 2.485377073287964,
+      "loss/hidden": 0.9140625,
+      "loss/logits": 0.16786439716815948,
+      "loss/reg": 0.01245111133903265,
+      "step": 1758
+    },
+    {
+      "epoch": 0.219875,
+      "grad_norm": 4.825221061706543,
+      "grad_norm_var": 96.93408061336578,
+      "learning_rate": 0.0001,
+      "loss": 1.0737,
+      "loss/crossentropy": 2.5381832122802734,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.13672339916229248,
+      "loss/reg": 0.012444333173334599,
+      "step": 1759
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 2.695854663848877,
+      "grad_norm_var": 97.06623592058641,
+      "learning_rate": 0.0001,
+      "loss": 1.2057,
+      "loss/crossentropy": 2.5472490787506104,
+      "loss/hidden": 0.90625,
+      "loss/logits": 0.17507830262184143,
+      "loss/reg": 0.012437481433153152,
+      "step": 1760
+    },
+    {
+      "epoch": 0.220125,
+      "grad_norm": 2.619779109954834,
+      "grad_norm_var": 97.32003810205347,
+      "learning_rate": 0.0001,
+      "loss": 1.0803,
+      "loss/crossentropy": 2.5513741970062256,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.13957199454307556,
+      "loss/reg": 0.012430677190423012,
+      "step": 1761
+    },
+    {
+      "epoch": 0.22025,
+      "grad_norm": 2.6490063667297363,
+      "grad_norm_var": 97.79004434756365,
+      "learning_rate": 0.0001,
+      "loss": 1.1991,
+      "loss/crossentropy": 2.408766031265259,
+      "loss/hidden": 0.921875,
+      "loss/logits": 0.15302732586860657,
+      "loss/reg": 0.012424097396433353,
+      "step": 1762
+    },
+    {
+      "epoch": 0.220375,
+      "grad_norm": 3.600940465927124,
+      "grad_norm_var": 97.41152871154476,
+      "learning_rate": 0.0001,
+      "loss": 1.0448,
+      "loss/crossentropy": 2.567258358001709,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.12761420011520386,
+      "loss/reg": 0.012417309917509556,
+      "step": 1763
+    },
+    {
+      "epoch": 0.2205,
+      "grad_norm": 2.783167600631714,
+      "grad_norm_var": 97.33925474643891,
+      "learning_rate": 0.0001,
+      "loss": 1.2647,
+      "loss/crossentropy": 2.4639530181884766,
+      "loss/hidden": 0.98828125,
+      "loss/logits": 0.15236002206802368,
+      "loss/reg": 0.012410394847393036,
+      "step": 1764
+    },
+    {
+      "epoch": 0.220625,
+      "grad_norm": 2.8482487201690674,
+      "grad_norm_var": 75.84316673308427,
+      "learning_rate": 0.0001,
+      "loss": 1.1159,
+      "loss/crossentropy": 2.614351511001587,
+      "loss/hidden": 0.84765625,
+      "loss/logits": 0.14418500661849976,
+      "loss/reg": 0.012403246946632862,
+      "step": 1765
+    },
+    {
+      "epoch": 0.22075,
+      "grad_norm": 4.04698371887207,
+      "grad_norm_var": 0.388008374816809,
+      "learning_rate": 0.0001,
+      "loss": 1.3027,
+      "loss/crossentropy": 2.6768503189086914,
+      "loss/hidden": 1.0,
+      "loss/logits": 0.17873653769493103,
+      "loss/reg": 0.012396049685776234,
+      "step": 1766
+    },
+    {
+      "epoch": 0.220875,
+      "grad_norm": 16.276508331298828,
+      "grad_norm_var": 11.096302421984861,
+      "learning_rate": 0.0001,
+      "loss": 1.1526,
+      "loss/crossentropy": 2.6857051849365234,
+      "loss/hidden": 0.87890625,
+      "loss/logits": 0.14977185428142548,
+      "loss/reg": 0.01238931342959404,
+      "step": 1767
+    },
+    {
+      "epoch": 0.221,
+      "grad_norm": 3.4267630577087402,
+      "grad_norm_var": 11.027930664929489,
+      "learning_rate": 0.0001,
+      "loss": 1.0696,
+      "loss/crossentropy": 2.5231096744537354,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.1450020670890808,
+      "loss/reg": 0.01238264236599207,
+      "step": 1768
+    },
+    {
+      "epoch": 0.221125,
+      "grad_norm": 2.5508995056152344,
+      "grad_norm_var": 11.12550393762463,
+      "learning_rate": 0.0001,
+      "loss": 1.1015,
+      "loss/crossentropy": 2.5338897705078125,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.1652265191078186,
+      "loss/reg": 0.012375940568745136,
+      "step": 1769
+    },
+    {
+      "epoch": 0.22125,
+      "grad_norm": 2.664597511291504,
+      "grad_norm_var": 11.126513136887851,
+      "learning_rate": 0.0001,
+      "loss": 1.0946,
+      "loss/crossentropy": 2.4849798679351807,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.16228261590003967,
+      "loss/reg": 0.012369180098176003,
+      "step": 1770
+    },
+    {
+      "epoch": 0.221375,
+      "grad_norm": 4.765702247619629,
+      "grad_norm_var": 11.129360295504242,
+      "learning_rate": 0.0001,
+      "loss": 1.1274,
+      "loss/crossentropy": 2.440256357192993,
+      "loss/hidden": 0.85546875,
+      "loss/logits": 0.14826975762844086,
+      "loss/reg": 0.01236215140670538,
+      "step": 1771
+    },
+    {
+      "epoch": 0.2215,
+      "grad_norm": 3.133784294128418,
+      "grad_norm_var": 11.172026082855703,
+      "learning_rate": 0.0001,
+      "loss": 1.0617,
+      "loss/crossentropy": 2.7009589672088623,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.14906959235668182,
+      "loss/reg": 0.012355109676718712,
+      "step": 1772
+    },
+    {
+      "epoch": 0.221625,
+      "grad_norm": 3.329200506210327,
+      "grad_norm_var": 11.066300955859553,
+      "learning_rate": 0.0001,
+      "loss": 1.064,
+      "loss/crossentropy": 2.5220742225646973,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.12804913520812988,
+      "loss/reg": 0.01234830915927887,
+      "step": 1773
+    },
+    {
+      "epoch": 0.22175,
+      "grad_norm": 4.502491474151611,
+      "grad_norm_var": 11.038805635564632,
+      "learning_rate": 0.0001,
+      "loss": 1.1216,
+      "loss/crossentropy": 2.4427030086517334,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.14662238955497742,
+      "loss/reg": 0.012341534718871117,
+      "step": 1774
+    },
+    {
+      "epoch": 0.221875,
+      "grad_norm": 4.261502265930176,
+      "grad_norm_var": 11.009414759143056,
+      "learning_rate": 0.0001,
+      "loss": 1.2303,
+      "loss/crossentropy": 2.463496208190918,
+      "loss/hidden": 0.921875,
+      "loss/logits": 0.18511801958084106,
+      "loss/reg": 0.012334545142948627,
+      "step": 1775
+    },
+    {
+      "epoch": 0.222,
+      "grad_norm": 2.7611560821533203,
+      "grad_norm_var": 10.997153332448207,
+      "learning_rate": 0.0001,
+      "loss": 1.103,
+      "loss/crossentropy": 2.643156051635742,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.17114627361297607,
+      "loss/reg": 0.012327339500188828,
+      "step": 1776
+    },
+    {
+      "epoch": 0.222125,
+      "grad_norm": 2.4437193870544434,
+      "grad_norm_var": 11.034748998650539,
+      "learning_rate": 0.0001,
+      "loss": 1.0003,
+      "loss/crossentropy": 2.910154342651367,
+      "loss/hidden": 0.73828125,
+      "loss/logits": 0.13876628875732422,
+      "loss/reg": 0.012320267036557198,
+      "step": 1777
+    },
+    {
+      "epoch": 0.22225,
+      "grad_norm": 4.437688827514648,
+      "grad_norm_var": 10.88203350793716,
+      "learning_rate": 0.0001,
+      "loss": 1.3096,
+      "loss/crossentropy": 2.2305355072021484,
+      "loss/hidden": 1.0,
+      "loss/logits": 0.1865069717168808,
+      "loss/reg": 0.012313010171055794,
+      "step": 1778
+    },
+    {
+      "epoch": 0.222375,
+      "grad_norm": 3.549940586090088,
+      "grad_norm_var": 10.886538839719051,
+      "learning_rate": 0.0001,
+      "loss": 1.2606,
+      "loss/crossentropy": 2.4338810443878174,
+      "loss/hidden": 0.9609375,
+      "loss/logits": 0.1766429841518402,
+      "loss/reg": 0.012306897900998592,
+      "step": 1779
+    },
+    {
+      "epoch": 0.2225,
+      "grad_norm": 3.347196578979492,
+      "grad_norm_var": 10.797133407006523,
+      "learning_rate": 0.0001,
+      "loss": 1.1304,
+      "loss/crossentropy": 2.835353374481201,
+      "loss/hidden": 0.84765625,
+      "loss/logits": 0.1597260981798172,
+      "loss/reg": 0.012300064787268639,
+      "step": 1780
+    },
+    {
+      "epoch": 0.222625,
+      "grad_norm": 3.352013111114502,
+      "grad_norm_var": 10.717386787566932,
+      "learning_rate": 0.0001,
+      "loss": 0.9526,
+      "loss/crossentropy": 2.4358811378479004,
+      "loss/hidden": 0.7265625,
+      "loss/logits": 0.10308530926704407,
+      "loss/reg": 0.012293456122279167,
+      "step": 1781
+    },
+    {
+      "epoch": 0.22275,
+      "grad_norm": 3.733184576034546,
+      "grad_norm_var": 10.73425846404735,
+      "learning_rate": 0.0001,
+      "loss": 1.0594,
+      "loss/crossentropy": 2.7393431663513184,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.139676034450531,
+      "loss/reg": 0.012287287972867489,
+      "step": 1782
+    },
+    {
+      "epoch": 0.222875,
+      "grad_norm": 2.7327675819396973,
+      "grad_norm_var": 0.5414954532598804,
+      "learning_rate": 0.0001,
+      "loss": 0.9985,
+      "loss/crossentropy": 2.4664580821990967,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.1256864368915558,
+      "loss/reg": 0.012280437164008617,
+      "step": 1783
+    },
+    {
+      "epoch": 0.223,
+      "grad_norm": 2.7125771045684814,
+      "grad_norm_var": 0.5743527285802199,
+      "learning_rate": 0.0001,
+      "loss": 1.0003,
+      "loss/crossentropy": 2.3793299198150635,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.12756191194057465,
+      "loss/reg": 0.012273788452148438,
+      "step": 1784
+    },
+    {
+      "epoch": 0.223125,
+      "grad_norm": 4.580681800842285,
+      "grad_norm_var": 0.6041116655885861,
+      "learning_rate": 0.0001,
+      "loss": 1.5202,
+      "loss/crossentropy": 1.8176430463790894,
+      "loss/hidden": 1.171875,
+      "loss/logits": 0.22560694813728333,
+      "loss/reg": 0.012266921810805798,
+      "step": 1785
+    },
+    {
+      "epoch": 0.22325,
+      "grad_norm": 3.310520887374878,
+      "grad_norm_var": 0.5565812947330983,
+      "learning_rate": 0.0001,
+      "loss": 1.0981,
+      "loss/crossentropy": 2.3666412830352783,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.13953973352909088,
+      "loss/reg": 0.012260017916560173,
+      "step": 1786
+    },
+    {
+      "epoch": 0.223375,
+      "grad_norm": 3.3059277534484863,
+      "grad_norm_var": 0.45501991794078456,
+      "learning_rate": 0.0001,
+      "loss": 1.198,
+      "loss/crossentropy": 2.3159430027008057,
+      "loss/hidden": 0.90625,
+      "loss/logits": 0.16920451819896698,
+      "loss/reg": 0.012253600172698498,
+      "step": 1787
+    },
+    {
+      "epoch": 0.2235,
+      "grad_norm": 2.687586784362793,
+      "grad_norm_var": 0.4873702987343137,
+      "learning_rate": 0.0001,
+      "loss": 1.0853,
+      "loss/crossentropy": 2.2798428535461426,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.12685424089431763,
+      "loss/reg": 0.012246805243194103,
+      "step": 1788
+    },
+    {
+      "epoch": 0.223625,
+      "grad_norm": 2.5434353351593018,
+      "grad_norm_var": 0.5376211993019903,
+      "learning_rate": 0.0001,
+      "loss": 0.9544,
+      "loss/crossentropy": 2.6347262859344482,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.12888537347316742,
+      "loss/reg": 0.012239991687238216,
+      "step": 1789
+    },
+    {
+      "epoch": 0.22375,
+      "grad_norm": 2.1676807403564453,
+      "grad_norm_var": 0.5324380567161892,
+      "learning_rate": 0.0001,
+      "loss": 1.0395,
+      "loss/crossentropy": 2.4155385494232178,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.13590013980865479,
+      "loss/reg": 0.012233071029186249,
+      "step": 1790
+    },
+    {
+      "epoch": 0.223875,
+      "grad_norm": 2.7618322372436523,
+      "grad_norm_var": 0.4698401846206546,
+      "learning_rate": 0.0001,
+      "loss": 0.9624,
+      "loss/crossentropy": 2.396468162536621,
+      "loss/hidden": 0.7109375,
+      "loss/logits": 0.12916944921016693,
+      "loss/reg": 0.012226960621774197,
+      "step": 1791
+    },
+    {
+      "epoch": 0.224,
+      "grad_norm": 2.605301856994629,
+      "grad_norm_var": 0.4794749872916848,
+      "learning_rate": 0.0001,
+      "loss": 1.0749,
+      "loss/crossentropy": 2.55946946144104,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.1441141664981842,
+      "loss/reg": 0.012221286073327065,
+      "step": 1792
+    },
+    {
+      "epoch": 0.224125,
+      "grad_norm": 2.7405450344085693,
+      "grad_norm_var": 0.4573457631061404,
+      "learning_rate": 0.0001,
+      "loss": 1.2526,
+      "loss/crossentropy": 2.6217191219329834,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.22033345699310303,
+      "loss/reg": 0.012214362621307373,
+      "step": 1793
+    },
+    {
+      "epoch": 0.22425,
+      "grad_norm": 4.141777515411377,
+      "grad_norm_var": 0.412429371225325,
+      "learning_rate": 0.0001,
+      "loss": 1.3743,
+      "loss/crossentropy": 1.6800086498260498,
+      "loss/hidden": 1.109375,
+      "loss/logits": 0.14288680255413055,
+      "loss/reg": 0.012208379805088043,
+      "step": 1794
+    },
+    {
+      "epoch": 0.224375,
+      "grad_norm": 3.810920476913452,
+      "grad_norm_var": 0.43087940783878576,
+      "learning_rate": 0.0001,
+      "loss": 1.0222,
+      "loss/crossentropy": 2.799727201461792,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.12675243616104126,
+      "loss/reg": 0.012202396057546139,
+      "step": 1795
+    },
+    {
+      "epoch": 0.2245,
+      "grad_norm": 5.86505126953125,
+      "grad_norm_var": 0.8904950250011618,
+      "learning_rate": 0.0001,
+      "loss": 1.3993,
+      "loss/crossentropy": 2.3400473594665527,
+      "loss/hidden": 1.109375,
+      "loss/logits": 0.16792380809783936,
+      "loss/reg": 0.012196486815810204,
+      "step": 1796
+    },
+    {
+      "epoch": 0.224625,
+      "grad_norm": 4.985130310058594,
+      "grad_norm_var": 1.0650859328511084,
+      "learning_rate": 0.0001,
+      "loss": 1.2963,
+      "loss/crossentropy": 2.8352878093719482,
+      "loss/hidden": 0.9609375,
+      "loss/logits": 0.21343833208084106,
+      "loss/reg": 0.012190500274300575,
+      "step": 1797
+    },
+    {
+      "epoch": 0.22475,
+      "grad_norm": 3.3390769958496094,
+      "grad_norm_var": 1.0582211532143861,
+      "learning_rate": 0.0001,
+      "loss": 1.2406,
+      "loss/crossentropy": 2.443668842315674,
+      "loss/hidden": 0.94921875,
+      "loss/logits": 0.16958631575107574,
+      "loss/reg": 0.012183710001409054,
+      "step": 1798
+    },
+    {
+      "epoch": 0.224875,
+      "grad_norm": 6.64145040512085,
+      "grad_norm_var": 1.6689068782404564,
+      "learning_rate": 0.0001,
+      "loss": 1.0831,
+      "loss/crossentropy": 2.9244539737701416,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.1449125111103058,
+      "loss/reg": 0.012176886200904846,
+      "step": 1799
+    },
+    {
+      "epoch": 0.225,
+      "grad_norm": 5.070358753204346,
+      "grad_norm_var": 1.7255938342974626,
+      "learning_rate": 0.0001,
+      "loss": 1.6577,
+      "loss/crossentropy": 2.39843487739563,
+      "loss/hidden": 1.296875,
+      "loss/logits": 0.23911422491073608,
+      "loss/reg": 0.012170514091849327,
+      "step": 1800
+    },
+    {
+      "epoch": 0.225125,
+      "grad_norm": 4.553746223449707,
+      "grad_norm_var": 1.7227809488467594,
+      "learning_rate": 0.0001,
+      "loss": 1.2558,
+      "loss/crossentropy": 2.6131913661956787,
+      "loss/hidden": 0.953125,
+      "loss/logits": 0.181059330701828,
+      "loss/reg": 0.012163708917796612,
+      "step": 1801
+    },
+    {
+      "epoch": 0.22525,
+      "grad_norm": 4.960707187652588,
+      "grad_norm_var": 1.7889862701755266,
+      "learning_rate": 0.0001,
+      "loss": 1.1394,
+      "loss/crossentropy": 2.8065645694732666,
+      "loss/hidden": 0.87890625,
+      "loss/logits": 0.13889148831367493,
+      "loss/reg": 0.012157038785517216,
+      "step": 1802
+    },
+    {
+      "epoch": 0.225375,
+      "grad_norm": 3.1211495399475098,
+      "grad_norm_var": 1.8054184757113145,
+      "learning_rate": 0.0001,
+      "loss": 1.1695,
+      "loss/crossentropy": 2.45322585105896,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.16519255936145782,
+      "loss/reg": 0.012150485999882221,
+      "step": 1803
+    },
+    {
+      "epoch": 0.2255,
+      "grad_norm": 3.1585381031036377,
+      "grad_norm_var": 1.74473550652442,
+      "learning_rate": 0.0001,
+      "loss": 1.1105,
+      "loss/crossentropy": 2.432692050933838,
+      "loss/hidden": 0.85546875,
+      "loss/logits": 0.1336173415184021,
+      "loss/reg": 0.012143667787313461,
+      "step": 1804
+    },
+    {
+      "epoch": 0.225625,
+      "grad_norm": 3.253127098083496,
+      "grad_norm_var": 1.647454221879684,
+      "learning_rate": 0.0001,
+      "loss": 1.1977,
+      "loss/crossentropy": 2.515928268432617,
+      "loss/hidden": 0.9140625,
+      "loss/logits": 0.1622786521911621,
+      "loss/reg": 0.012136753648519516,
+      "step": 1805
+    },
+    {
+      "epoch": 0.22575,
+      "grad_norm": 2.5755836963653564,
+      "grad_norm_var": 1.560998409452076,
+      "learning_rate": 0.0001,
+      "loss": 1.0782,
+      "loss/crossentropy": 2.3887364864349365,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.14443060755729675,
+      "loss/reg": 0.01213008351624012,
+      "step": 1806
+    },
+    {
+      "epoch": 0.225875,
+      "grad_norm": 3.9604837894439697,
+      "grad_norm_var": 1.4570643895132678,
+      "learning_rate": 0.0001,
+      "loss": 1.3124,
+      "loss/crossentropy": 2.5280768871307373,
+      "loss/hidden": 0.96484375,
+      "loss/logits": 0.2262798249721527,
+      "loss/reg": 0.012123593129217625,
+      "step": 1807
+    },
+    {
+      "epoch": 0.226,
+      "grad_norm": 3.0762956142425537,
+      "grad_norm_var": 1.380270170821843,
+      "learning_rate": 0.0001,
+      "loss": 1.0253,
+      "loss/crossentropy": 2.928957939147949,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.13464616239070892,
+      "loss/reg": 0.012116815894842148,
+      "step": 1808
+    },
+    {
+      "epoch": 0.226125,
+      "grad_norm": 3.482069730758667,
+      "grad_norm_var": 1.2823655143187342,
+      "learning_rate": 0.0001,
+      "loss": 1.3007,
+      "loss/crossentropy": 2.523679733276367,
+      "loss/hidden": 0.99609375,
+      "loss/logits": 0.18347124755382538,
+      "loss/reg": 0.012109990231692791,
+      "step": 1809
+    },
+    {
+      "epoch": 0.22625,
+      "grad_norm": 2.7978577613830566,
+      "grad_norm_var": 1.3921909123204605,
+      "learning_rate": 0.0001,
+      "loss": 1.1764,
+      "loss/crossentropy": 2.5396063327789307,
+      "loss/hidden": 0.8984375,
+      "loss/logits": 0.15695957839488983,
+      "loss/reg": 0.012103389948606491,
+      "step": 1810
+    },
+    {
+      "epoch": 0.226375,
+      "grad_norm": 3.3859915733337402,
+      "grad_norm_var": 1.4164960881741988,
+      "learning_rate": 0.0001,
+      "loss": 1.1827,
+      "loss/crossentropy": 2.421473264694214,
+      "loss/hidden": 0.890625,
+      "loss/logits": 0.1710580289363861,
+      "loss/reg": 0.01209702342748642,
+      "step": 1811
+    },
+    {
+      "epoch": 0.2265,
+      "grad_norm": 2.7233433723449707,
+      "grad_norm_var": 1.2580651775254919,
+      "learning_rate": 0.0001,
+      "loss": 1.1054,
+      "loss/crossentropy": 2.460507392883301,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.14855897426605225,
+      "loss/reg": 0.012090392410755157,
+      "step": 1812
+    },
+    {
+      "epoch": 0.226625,
+      "grad_norm": 2.803961992263794,
+      "grad_norm_var": 1.2159247798178134,
+      "learning_rate": 0.0001,
+      "loss": 1.0841,
+      "loss/crossentropy": 2.4394664764404297,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.13119429349899292,
+      "loss/reg": 0.012084128335118294,
+      "step": 1813
+    },
+    {
+      "epoch": 0.22675,
+      "grad_norm": 2.345188856124878,
+      "grad_norm_var": 1.323038348759911,
+      "learning_rate": 0.0001,
+      "loss": 1.0238,
+      "loss/crossentropy": 2.4309263229370117,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.13351188600063324,
+      "loss/reg": 0.01207789871841669,
+      "step": 1814
+    },
+    {
+      "epoch": 0.226875,
+      "grad_norm": 2.409219741821289,
+      "grad_norm_var": 0.7371698535500113,
+      "learning_rate": 0.0001,
+      "loss": 1.0862,
+      "loss/crossentropy": 2.6133697032928467,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.14128842949867249,
+      "loss/reg": 0.012071667239069939,
+      "step": 1815
+    },
+    {
+      "epoch": 0.227,
+      "grad_norm": 2.506357192993164,
+      "grad_norm_var": 0.5615762297781826,
+      "learning_rate": 0.0001,
+      "loss": 1.1755,
+      "loss/crossentropy": 2.237166166305542,
+      "loss/hidden": 0.875,
+      "loss/logits": 0.17987734079360962,
+      "loss/reg": 0.01206506323069334,
+      "step": 1816
+    },
+    {
+      "epoch": 0.227125,
+      "grad_norm": 3.0118813514709473,
+      "grad_norm_var": 0.43074473519190304,
+      "learning_rate": 0.0001,
+      "loss": 1.2733,
+      "loss/crossentropy": 2.306960344314575,
+      "loss/hidden": 0.97265625,
+      "loss/logits": 0.18007943034172058,
+      "loss/reg": 0.012058539316058159,
+      "step": 1817
+    },
+    {
+      "epoch": 0.22725,
+      "grad_norm": 2.3795273303985596,
+      "grad_norm_var": 0.20616682699356942,
+      "learning_rate": 0.0001,
+      "loss": 1.1529,
+      "loss/crossentropy": 2.451083183288574,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.1495814025402069,
+      "loss/reg": 0.012052006088197231,
+      "step": 1818
+    },
+    {
+      "epoch": 0.227375,
+      "grad_norm": 2.332099676132202,
+      "grad_norm_var": 0.22569619304637148,
+      "learning_rate": 0.0001,
+      "loss": 1.0095,
+      "loss/crossentropy": 2.538863182067871,
+      "loss/hidden": 0.765625,
+      "loss/logits": 0.1233830600976944,
+      "loss/reg": 0.012045500800013542,
+      "step": 1819
+    },
+    {
+      "epoch": 0.2275,
+      "grad_norm": 2.9021058082580566,
+      "grad_norm_var": 0.22054224463718033,
+      "learning_rate": 0.0001,
+      "loss": 1.4194,
+      "loss/crossentropy": 2.32419753074646,
+      "loss/hidden": 1.1171875,
+      "loss/logits": 0.18185210227966309,
+      "loss/reg": 0.012038588523864746,
+      "step": 1820
+    },
+    {
+      "epoch": 0.227625,
+      "grad_norm": 2.590693235397339,
+      "grad_norm_var": 0.21426742260109843,
+      "learning_rate": 0.0001,
+      "loss": 1.0937,
+      "loss/crossentropy": 2.387709856033325,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.14921152591705322,
+      "loss/reg": 0.012031761929392815,
+      "step": 1821
+    },
+    {
+      "epoch": 0.22775,
+      "grad_norm": 3.300173044204712,
+      "grad_norm_var": 0.22248606839011675,
+      "learning_rate": 0.0001,
+      "loss": 1.2738,
+      "loss/crossentropy": 2.198507308959961,
+      "loss/hidden": 0.921875,
+      "loss/logits": 0.2317131608724594,
+      "loss/reg": 0.012025104835629463,
+      "step": 1822
+    },
+    {
+      "epoch": 0.227875,
+      "grad_norm": 2.4631481170654297,
+      "grad_norm_var": 0.14599126890724012,
+      "learning_rate": 0.0001,
+      "loss": 1.0501,
+      "loss/crossentropy": 2.370026111602783,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.14089564979076385,
+      "loss/reg": 0.012018561363220215,
+      "step": 1823
+    },
+    {
+      "epoch": 0.228,
+      "grad_norm": 3.0455853939056396,
+      "grad_norm_var": 0.14484462892754285,
+      "learning_rate": 0.0001,
+      "loss": 1.1923,
+      "loss/crossentropy": 2.4721148014068604,
+      "loss/hidden": 0.890625,
+      "loss/logits": 0.18152591586112976,
+      "loss/reg": 0.01201186515390873,
+      "step": 1824
+    },
+    {
+      "epoch": 0.228125,
+      "grad_norm": 3.3358211517333984,
+      "grad_norm_var": 0.13249022272456326,
+      "learning_rate": 0.0001,
+      "loss": 1.1468,
+      "loss/crossentropy": 2.6852481365203857,
+      "loss/hidden": 0.86328125,
+      "loss/logits": 0.16346678137779236,
+      "loss/reg": 0.012004739604890347,
+      "step": 1825
+    },
+    {
+      "epoch": 0.22825,
+      "grad_norm": 2.6543102264404297,
+      "grad_norm_var": 0.13326040062104477,
+      "learning_rate": 0.0001,
+      "loss": 1.034,
+      "loss/crossentropy": 2.3123793601989746,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.14057926833629608,
+      "loss/reg": 0.011998065747320652,
+      "step": 1826
+    },
+    {
+      "epoch": 0.228375,
+      "grad_norm": 2.4901700019836426,
+      "grad_norm_var": 0.10886572110197955,
+      "learning_rate": 0.0001,
+      "loss": 1.0816,
+      "loss/crossentropy": 2.3769681453704834,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.1531093716621399,
+      "loss/reg": 0.011991064064204693,
+      "step": 1827
+    },
+    {
+      "epoch": 0.2285,
+      "grad_norm": 2.7322981357574463,
+      "grad_norm_var": 0.10889162038143449,
+      "learning_rate": 0.0001,
+      "loss": 1.2769,
+      "loss/crossentropy": 2.432474136352539,
+      "loss/hidden": 0.96484375,
+      "loss/logits": 0.19221791625022888,
+      "loss/reg": 0.011984668672084808,
+      "step": 1828
+    },
+    {
+      "epoch": 0.228625,
+      "grad_norm": 3.040367603302002,
+      "grad_norm_var": 0.11545954489478329,
+      "learning_rate": 0.0001,
+      "loss": 1.1185,
+      "loss/crossentropy": 2.5954103469848633,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.17058409750461578,
+      "loss/reg": 0.011977394111454487,
+      "step": 1829
+    },
+    {
+      "epoch": 0.22875,
+      "grad_norm": 4.54348611831665,
+      "grad_norm_var": 0.30728487463357607,
+      "learning_rate": 0.0001,
+      "loss": 1.4038,
+      "loss/crossentropy": 2.3670899868011475,
+      "loss/hidden": 1.1171875,
+      "loss/logits": 0.16688917577266693,
+      "loss/reg": 0.011970371939241886,
+      "step": 1830
+    },
+    {
+      "epoch": 0.228875,
+      "grad_norm": 2.4918017387390137,
+      "grad_norm_var": 0.3027632602969258,
+      "learning_rate": 0.0001,
+      "loss": 1.1499,
+      "loss/crossentropy": 2.4433906078338623,
+      "loss/hidden": 0.859375,
+      "loss/logits": 0.17090028524398804,
+      "loss/reg": 0.011963790282607079,
+      "step": 1831
+    },
+    {
+      "epoch": 0.229,
+      "grad_norm": 2.9950368404388428,
+      "grad_norm_var": 0.2944027102760967,
+      "learning_rate": 0.0001,
+      "loss": 1.1104,
+      "loss/crossentropy": 2.791016101837158,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.16274631023406982,
+      "loss/reg": 0.011956454254686832,
+      "step": 1832
+    },
+    {
+      "epoch": 0.229125,
+      "grad_norm": 3.249807596206665,
+      "grad_norm_var": 0.3016714416861295,
+      "learning_rate": 0.0001,
+      "loss": 1.1681,
+      "loss/crossentropy": 2.459493637084961,
+      "loss/hidden": 0.890625,
+      "loss/logits": 0.1580270528793335,
+      "loss/reg": 0.011949477717280388,
+      "step": 1833
+    },
+    {
+      "epoch": 0.22925,
+      "grad_norm": 2.811523199081421,
+      "grad_norm_var": 0.2828291293008836,
+      "learning_rate": 0.0001,
+      "loss": 1.1852,
+      "loss/crossentropy": 2.3925209045410156,
+      "loss/hidden": 0.90234375,
+      "loss/logits": 0.16343240439891815,
+      "loss/reg": 0.011942173354327679,
+      "step": 1834
+    },
+    {
+      "epoch": 0.229375,
+      "grad_norm": 2.8339016437530518,
+      "grad_norm_var": 0.25815168646451997,
+      "learning_rate": 0.0001,
+      "loss": 1.2253,
+      "loss/crossentropy": 2.3892555236816406,
+      "loss/hidden": 0.93359375,
+      "loss/logits": 0.17232704162597656,
+      "loss/reg": 0.011935535818338394,
+      "step": 1835
+    },
+    {
+      "epoch": 0.2295,
+      "grad_norm": 3.5241827964782715,
+      "grad_norm_var": 0.2769127015292412,
+      "learning_rate": 0.0001,
+      "loss": 1.0834,
+      "loss/crossentropy": 2.4555206298828125,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.15549984574317932,
+      "loss/reg": 0.011928863823413849,
+      "step": 1836
+    },
+    {
+      "epoch": 0.229625,
+      "grad_norm": 3.1511709690093994,
+      "grad_norm_var": 0.26548067421464716,
+      "learning_rate": 0.0001,
+      "loss": 1.178,
+      "loss/crossentropy": 2.634005069732666,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.14858701825141907,
+      "loss/reg": 0.011921834200620651,
+      "step": 1837
+    },
+    {
+      "epoch": 0.22975,
+      "grad_norm": 3.1957132816314697,
+      "grad_norm_var": 0.2625588163447295,
+      "learning_rate": 0.0001,
+      "loss": 1.4165,
+      "loss/crossentropy": 2.364680051803589,
+      "loss/hidden": 1.0703125,
+      "loss/logits": 0.22706648707389832,
+      "loss/reg": 0.01191527210175991,
+      "step": 1838
+    },
+    {
+      "epoch": 0.229875,
+      "grad_norm": 3.4130618572235107,
+      "grad_norm_var": 0.24654008934847388,
+      "learning_rate": 0.0001,
+      "loss": 1.1042,
+      "loss/crossentropy": 2.80603289604187,
+      "loss/hidden": 0.84765625,
+      "loss/logits": 0.1374989151954651,
+      "loss/reg": 0.011908802203834057,
+      "step": 1839
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 10.125365257263184,
+      "grad_norm_var": 3.3332932374288853,
+      "learning_rate": 0.0001,
+      "loss": 1.2822,
+      "loss/crossentropy": 2.7263705730438232,
+      "loss/hidden": 0.96875,
+      "loss/logits": 0.1943938136100769,
+      "loss/reg": 0.011902069672942162,
+      "step": 1840
+    },
+    {
+      "epoch": 0.230125,
+      "grad_norm": 3.0267136096954346,
+      "grad_norm_var": 3.3475461515590967,
+      "learning_rate": 0.0001,
+      "loss": 1.4492,
+      "loss/crossentropy": 2.527991771697998,
+      "loss/hidden": 1.125,
+      "loss/logits": 0.20524314045906067,
+      "loss/reg": 0.01189569290727377,
+      "step": 1841
+    },
+    {
+      "epoch": 0.23025,
+      "grad_norm": 3.0950570106506348,
+      "grad_norm_var": 3.308964844199806,
+      "learning_rate": 0.0001,
+      "loss": 1.1675,
+      "loss/crossentropy": 2.9109840393066406,
+      "loss/hidden": 0.87890625,
+      "loss/logits": 0.16966286301612854,
+      "loss/reg": 0.011889121495187283,
+      "step": 1842
+    },
+    {
+      "epoch": 0.230375,
+      "grad_norm": 2.6885743141174316,
+      "grad_norm_var": 3.283521301133828,
+      "learning_rate": 0.0001,
+      "loss": 0.9502,
+      "loss/crossentropy": 2.463144540786743,
+      "loss/hidden": 0.7265625,
+      "loss/logits": 0.10477050393819809,
+      "loss/reg": 0.011882682330906391,
+      "step": 1843
+    },
+    {
+      "epoch": 0.2305,
+      "grad_norm": 7.153027534484863,
+      "grad_norm_var": 4.018621504186891,
+      "learning_rate": 0.0001,
+      "loss": 1.5744,
+      "loss/crossentropy": 2.408703088760376,
+      "loss/hidden": 1.1796875,
+      "loss/logits": 0.2759320139884949,
+      "loss/reg": 0.01187664270401001,
+      "step": 1844
+    },
+    {
+      "epoch": 0.230625,
+      "grad_norm": 2.739917516708374,
+      "grad_norm_var": 4.056043276959989,
+      "learning_rate": 0.0001,
+      "loss": 1.1633,
+      "loss/crossentropy": 2.417595863342285,
+      "loss/hidden": 0.87109375,
+      "loss/logits": 0.17353396117687225,
+      "loss/reg": 0.011870414949953556,
+      "step": 1845
+    },
+    {
+      "epoch": 0.23075,
+      "grad_norm": 3.544043779373169,
+      "grad_norm_var": 4.021382457840771,
+      "learning_rate": 0.0001,
+      "loss": 1.256,
+      "loss/crossentropy": 2.5209176540374756,
+      "loss/hidden": 0.96484375,
+      "loss/logits": 0.1725134551525116,
+      "loss/reg": 0.0118643743917346,
+      "step": 1846
+    },
+    {
+      "epoch": 0.230875,
+      "grad_norm": 2.5454280376434326,
+      "grad_norm_var": 4.012548475227705,
+      "learning_rate": 0.0001,
+      "loss": 1.125,
+      "loss/crossentropy": 2.788163185119629,
+      "loss/hidden": 0.84765625,
+      "loss/logits": 0.15880800783634186,
+      "loss/reg": 0.011857859790325165,
+      "step": 1847
+    },
+    {
+      "epoch": 0.231,
+      "grad_norm": 2.663525104522705,
+      "grad_norm_var": 4.053043390213641,
+      "learning_rate": 0.0001,
+      "loss": 1.1004,
+      "loss/crossentropy": 2.458189010620117,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.14984336495399475,
+      "loss/reg": 0.011851528659462929,
+      "step": 1848
+    },
+    {
+      "epoch": 0.231125,
+      "grad_norm": 3.5077850818634033,
+      "grad_norm_var": 4.0405115731206624,
+      "learning_rate": 0.0001,
+      "loss": 1.0857,
+      "loss/crossentropy": 2.5606067180633545,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.15474531054496765,
+      "loss/reg": 0.011845485307276249,
+      "step": 1849
+    },
+    {
+      "epoch": 0.23125,
+      "grad_norm": 3.274979829788208,
+      "grad_norm_var": 3.995870290819255,
+      "learning_rate": 0.0001,
+      "loss": 1.0438,
+      "loss/crossentropy": 2.5572433471679688,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.12853728234767914,
+      "loss/reg": 0.011839361861348152,
+      "step": 1850
+    },
+    {
+      "epoch": 0.231375,
+      "grad_norm": 2.723388433456421,
+      "grad_norm_var": 4.010576716926485,
+      "learning_rate": 0.0001,
+      "loss": 1.0091,
+      "loss/crossentropy": 2.6289405822753906,
+      "loss/hidden": 0.7578125,
+      "loss/logits": 0.1329689919948578,
+      "loss/reg": 0.011833377182483673,
+      "step": 1851
+    },
+    {
+      "epoch": 0.2315,
+      "grad_norm": 5.253420829772949,
+      "grad_norm_var": 4.140042975539734,
+      "learning_rate": 0.0001,
+      "loss": 1.2997,
+      "loss/crossentropy": 2.522901773452759,
+      "loss/hidden": 0.984375,
+      "loss/logits": 0.19709725677967072,
+      "loss/reg": 0.011827518232166767,
+      "step": 1852
+    },
+    {
+      "epoch": 0.231625,
+      "grad_norm": 2.8319897651672363,
+      "grad_norm_var": 4.177483717989236,
+      "learning_rate": 0.0001,
+      "loss": 1.3684,
+      "loss/crossentropy": 2.2863683700561523,
+      "loss/hidden": 1.078125,
+      "loss/logits": 0.17204830050468445,
+      "loss/reg": 0.011821010150015354,
+      "step": 1853
+    },
+    {
+      "epoch": 0.23175,
+      "grad_norm": 3.758786201477051,
+      "grad_norm_var": 4.14732397532542,
+      "learning_rate": 0.0001,
+      "loss": 1.091,
+      "loss/crossentropy": 2.6834845542907715,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.1564285159111023,
+      "loss/reg": 0.011815285310149193,
+      "step": 1854
+    },
+    {
+      "epoch": 0.231875,
+      "grad_norm": 2.970064163208008,
+      "grad_norm_var": 4.18814826964047,
+      "learning_rate": 0.0001,
+      "loss": 1.1222,
+      "loss/crossentropy": 2.350801944732666,
+      "loss/hidden": 0.85546875,
+      "loss/logits": 0.14867839217185974,
+      "loss/reg": 0.011809652671217918,
+      "step": 1855
+    },
+    {
+      "epoch": 0.232,
+      "grad_norm": 2.991593599319458,
+      "grad_norm_var": 1.4178378002755372,
+      "learning_rate": 0.0001,
+      "loss": 1.0011,
+      "loss/crossentropy": 2.438445568084717,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.10962004959583282,
+      "loss/reg": 0.011803114786744118,
+      "step": 1856
+    },
+    {
+      "epoch": 0.232125,
+      "grad_norm": 2.449436664581299,
+      "grad_norm_var": 1.4691695267047424,
+      "learning_rate": 0.0001,
+      "loss": 0.9734,
+      "loss/crossentropy": 2.3218026161193848,
+      "loss/hidden": 0.7265625,
+      "loss/logits": 0.12885171175003052,
+      "loss/reg": 0.011797088198363781,
+      "step": 1857
+    },
+    {
+      "epoch": 0.23225,
+      "grad_norm": 3.538789987564087,
+      "grad_norm_var": 1.4642067121892361,
+      "learning_rate": 0.0001,
+      "loss": 1.2534,
+      "loss/crossentropy": 2.3883466720581055,
+      "loss/hidden": 0.95703125,
+      "loss/logits": 0.17844754457473755,
+      "loss/reg": 0.01179055217653513,
+      "step": 1858
+    },
+    {
+      "epoch": 0.232375,
+      "grad_norm": 2.6442153453826904,
+      "grad_norm_var": 1.4686242200827846,
+      "learning_rate": 0.0001,
+      "loss": 1.1711,
+      "loss/crossentropy": 2.5905346870422363,
+      "loss/hidden": 0.890625,
+      "loss/logits": 0.1626666784286499,
+      "loss/reg": 0.011784548871219158,
+      "step": 1859
+    },
+    {
+      "epoch": 0.2325,
+      "grad_norm": 2.63577938079834,
+      "grad_norm_var": 0.4906894012779368,
+      "learning_rate": 0.0001,
+      "loss": 0.9256,
+      "loss/crossentropy": 2.1434152126312256,
+      "loss/hidden": 0.703125,
+      "loss/logits": 0.10471002757549286,
+      "loss/reg": 0.011777986772358418,
+      "step": 1860
+    },
+    {
+      "epoch": 0.232625,
+      "grad_norm": 2.578632354736328,
+      "grad_norm_var": 0.5006945948105681,
+      "learning_rate": 0.0001,
+      "loss": 1.0467,
+      "loss/crossentropy": 2.7150838375091553,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.13211926817893982,
+      "loss/reg": 0.011771964840590954,
+      "step": 1861
+    },
+    {
+      "epoch": 0.23275,
+      "grad_norm": 3.5398929119110107,
+      "grad_norm_var": 0.5004607034463172,
+      "learning_rate": 0.0001,
+      "loss": 1.2583,
+      "loss/crossentropy": 2.7141544818878174,
+      "loss/hidden": 0.94140625,
+      "loss/logits": 0.19919423758983612,
+      "loss/reg": 0.01176582369953394,
+      "step": 1862
+    },
+    {
+      "epoch": 0.232875,
+      "grad_norm": 3.1188647747039795,
+      "grad_norm_var": 0.4771405434982313,
+      "learning_rate": 0.0001,
+      "loss": 1.0285,
+      "loss/crossentropy": 2.2451930046081543,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.12186941504478455,
+      "loss/reg": 0.011759834364056587,
+      "step": 1863
+    },
+    {
+      "epoch": 0.233,
+      "grad_norm": 3.0794904232025146,
+      "grad_norm_var": 0.4606925715881855,
+      "learning_rate": 0.0001,
+      "loss": 1.0381,
+      "loss/crossentropy": 2.338972330093384,
+      "loss/hidden": 0.7890625,
+      "loss/logits": 0.1315452754497528,
+      "loss/reg": 0.011753806844353676,
+      "step": 1864
+    },
+    {
+      "epoch": 0.233125,
+      "grad_norm": 3.019987106323242,
+      "grad_norm_var": 0.45431474823360435,
+      "learning_rate": 0.0001,
+      "loss": 1.0534,
+      "loss/crossentropy": 2.5802109241485596,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.1351454257965088,
+      "loss/reg": 0.011747847311198711,
+      "step": 1865
+    },
+    {
+      "epoch": 0.23325,
+      "grad_norm": 3.0198545455932617,
+      "grad_norm_var": 0.45415120043857066,
+      "learning_rate": 0.0001,
+      "loss": 1.3085,
+      "loss/crossentropy": 2.2313005924224854,
+      "loss/hidden": 0.96875,
+      "loss/logits": 0.22236411273479462,
+      "loss/reg": 0.01174143236130476,
+      "step": 1866
+    },
+    {
+      "epoch": 0.233375,
+      "grad_norm": 2.4949746131896973,
+      "grad_norm_var": 0.46993664201424584,
+      "learning_rate": 0.0001,
+      "loss": 1.0675,
+      "loss/crossentropy": 2.530468225479126,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.14545810222625732,
+      "loss/reg": 0.01173495315015316,
+      "step": 1867
+    },
+    {
+      "epoch": 0.2335,
+      "grad_norm": 3.2556769847869873,
+      "grad_norm_var": 0.15119857978987525,
+      "learning_rate": 0.0001,
+      "loss": 1.2375,
+      "loss/crossentropy": 2.5190608501434326,
+      "loss/hidden": 0.9296875,
+      "loss/logits": 0.1905001997947693,
+      "loss/reg": 0.011728293262422085,
+      "step": 1868
+    },
+    {
+      "epoch": 0.233625,
+      "grad_norm": 5.818284034729004,
+      "grad_norm_var": 0.6434646637531138,
+      "learning_rate": 0.0001,
+      "loss": 1.7541,
+      "loss/crossentropy": 2.4520504474639893,
+      "loss/hidden": 1.328125,
+      "loss/logits": 0.30873966217041016,
+      "loss/reg": 0.011721369810402393,
+      "step": 1869
+    },
+    {
+      "epoch": 0.23375,
+      "grad_norm": 18.115583419799805,
+      "grad_norm_var": 14.629645381532592,
+      "learning_rate": 0.0001,
+      "loss": 1.626,
+      "loss/crossentropy": 2.765774965286255,
+      "loss/hidden": 1.265625,
+      "loss/logits": 0.24319738149642944,
+      "loss/reg": 0.011714870110154152,
+      "step": 1870
+    },
+    {
+      "epoch": 0.233875,
+      "grad_norm": 4.950761795043945,
+      "grad_norm_var": 14.58186333788439,
+      "learning_rate": 0.0001,
+      "loss": 1.3825,
+      "loss/crossentropy": 2.3341286182403564,
+      "loss/hidden": 1.0859375,
+      "loss/logits": 0.17949844896793365,
+      "loss/reg": 0.011708397418260574,
+      "step": 1871
+    },
+    {
+      "epoch": 0.234,
+      "grad_norm": 3.149343729019165,
+      "grad_norm_var": 14.557933702264693,
+      "learning_rate": 0.0001,
+      "loss": 1.2519,
+      "loss/crossentropy": 2.2633235454559326,
+      "loss/hidden": 0.98046875,
+      "loss/logits": 0.1543847918510437,
+      "loss/reg": 0.011701937764883041,
+      "step": 1872
+    },
+    {
+      "epoch": 0.234125,
+      "grad_norm": 3.2733466625213623,
+      "grad_norm_var": 14.4066140044379,
+      "learning_rate": 0.0001,
+      "loss": 0.9954,
+      "loss/crossentropy": 2.3169283866882324,
+      "loss/hidden": 0.75,
+      "loss/logits": 0.12840038537979126,
+      "loss/reg": 0.011695554479956627,
+      "step": 1873
+    },
+    {
+      "epoch": 0.23425,
+      "grad_norm": 2.9650564193725586,
+      "grad_norm_var": 14.482709435196355,
+      "learning_rate": 0.0001,
+      "loss": 1.0411,
+      "loss/crossentropy": 2.499809741973877,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.14294317364692688,
+      "loss/reg": 0.011689374223351479,
+      "step": 1874
+    },
+    {
+      "epoch": 0.234375,
+      "grad_norm": 3.3756163120269775,
+      "grad_norm_var": 14.361621179597174,
+      "learning_rate": 0.0001,
+      "loss": 1.2119,
+      "loss/crossentropy": 2.484644889831543,
+      "loss/hidden": 0.92578125,
+      "loss/logits": 0.16932502388954163,
+      "loss/reg": 0.011682825163006783,
+      "step": 1875
+    },
+    {
+      "epoch": 0.2345,
+      "grad_norm": 4.519603729248047,
+      "grad_norm_var": 14.171825990122963,
+      "learning_rate": 0.0001,
+      "loss": 1.2405,
+      "loss/crossentropy": 2.1961562633514404,
+      "loss/hidden": 0.95703125,
+      "loss/logits": 0.1666729897260666,
+      "loss/reg": 0.01167618203908205,
+      "step": 1876
+    },
+    {
+      "epoch": 0.234625,
+      "grad_norm": 7.144575595855713,
+      "grad_norm_var": 14.370738345950354,
+      "learning_rate": 0.0001,
+      "loss": 1.5206,
+      "loss/crossentropy": 3.1038763523101807,
+      "loss/hidden": 1.1171875,
+      "loss/logits": 0.28672921657562256,
+      "loss/reg": 0.01166961807757616,
+      "step": 1877
+    },
+    {
+      "epoch": 0.23475,
+      "grad_norm": 3.3410837650299072,
+      "grad_norm_var": 14.403365735245997,
+      "learning_rate": 0.0001,
+      "loss": 1.0188,
+      "loss/crossentropy": 2.6396772861480713,
+      "loss/hidden": 0.76953125,
+      "loss/logits": 0.1326141357421875,
+      "loss/reg": 0.011662835255265236,
+      "step": 1878
+    },
+    {
+      "epoch": 0.234875,
+      "grad_norm": 3.955864429473877,
+      "grad_norm_var": 14.274587966883617,
+      "learning_rate": 0.0001,
+      "loss": 1.2218,
+      "loss/crossentropy": 2.5625252723693848,
+      "loss/hidden": 0.9453125,
+      "loss/logits": 0.1599578559398651,
+      "loss/reg": 0.011656397022306919,
+      "step": 1879
+    },
+    {
+      "epoch": 0.235,
+      "grad_norm": 3.702489137649536,
+      "grad_norm_var": 14.162786868506032,
+      "learning_rate": 0.0001,
+      "loss": 1.1474,
+      "loss/crossentropy": 2.570443868637085,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.14810335636138916,
+      "loss/reg": 0.011648759245872498,
+      "step": 1880
+    },
+    {
+      "epoch": 0.235125,
+      "grad_norm": 3.006218194961548,
+      "grad_norm_var": 14.165986485307208,
+      "learning_rate": 0.0001,
+      "loss": 1.1945,
+      "loss/crossentropy": 2.175870895385742,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.1679525077342987,
+      "loss/reg": 0.01164237316697836,
+      "step": 1881
+    },
+    {
+      "epoch": 0.23525,
+      "grad_norm": 3.0096826553344727,
+      "grad_norm_var": 14.168346952953607,
+      "learning_rate": 0.0001,
+      "loss": 1.0846,
+      "loss/crossentropy": 2.7216742038726807,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.14791440963745117,
+      "loss/reg": 0.01163501013070345,
+      "step": 1882
+    },
+    {
+      "epoch": 0.235375,
+      "grad_norm": 3.3620381355285645,
+      "grad_norm_var": 13.954069607905243,
+      "learning_rate": 0.0001,
+      "loss": 1.339,
+      "loss/crossentropy": 2.6900413036346436,
+      "loss/hidden": 1.0234375,
+      "loss/logits": 0.19929756224155426,
+      "loss/reg": 0.01162752602249384,
+      "step": 1883
+    },
+    {
+      "epoch": 0.2355,
+      "grad_norm": 2.5984811782836914,
+      "grad_norm_var": 14.117182111852275,
+      "learning_rate": 0.0001,
+      "loss": 1.0653,
+      "loss/crossentropy": 2.392869472503662,
+      "loss/hidden": 0.8046875,
+      "loss/logits": 0.14441151916980743,
+      "loss/reg": 0.011621098034083843,
+      "step": 1884
+    },
+    {
+      "epoch": 0.235625,
+      "grad_norm": 3.761037588119507,
+      "grad_norm_var": 14.093606633107498,
+      "learning_rate": 0.0001,
+      "loss": 1.0553,
+      "loss/crossentropy": 2.8546535968780518,
+      "loss/hidden": 0.77734375,
+      "loss/logits": 0.16185110807418823,
+      "loss/reg": 0.011613850481808186,
+      "step": 1885
+    },
+    {
+      "epoch": 0.23575,
+      "grad_norm": 5.239340782165527,
+      "grad_norm_var": 1.319651559267754,
+      "learning_rate": 0.0001,
+      "loss": 1.0582,
+      "loss/crossentropy": 2.560152769088745,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.14137983322143555,
+      "loss/reg": 0.011606672778725624,
+      "step": 1886
+    },
+    {
+      "epoch": 0.235875,
+      "grad_norm": 2.783051013946533,
+      "grad_norm_var": 1.2907520410126019,
+      "learning_rate": 0.0001,
+      "loss": 1.0424,
+      "loss/crossentropy": 2.37328839302063,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.13344892859458923,
+      "loss/reg": 0.011599456891417503,
+      "step": 1887
+    },
+    {
+      "epoch": 0.236,
+      "grad_norm": 3.554786443710327,
+      "grad_norm_var": 1.2713025846844552,
+      "learning_rate": 0.0001,
+      "loss": 1.1632,
+      "loss/crossentropy": 2.494732141494751,
+      "loss/hidden": 0.85546875,
+      "loss/logits": 0.1918359100818634,
+      "loss/reg": 0.011593029834330082,
+      "step": 1888
+    },
+    {
+      "epoch": 0.236125,
+      "grad_norm": 3.321483612060547,
+      "grad_norm_var": 1.2685516790340434,
+      "learning_rate": 0.0001,
+      "loss": 1.2471,
+      "loss/crossentropy": 2.4095706939697266,
+      "loss/hidden": 0.9765625,
+      "loss/logits": 0.15465524792671204,
+      "loss/reg": 0.011586642824113369,
+      "step": 1889
+    },
+    {
+      "epoch": 0.23625,
+      "grad_norm": 2.9616763591766357,
+      "grad_norm_var": 1.2688960186311133,
+      "learning_rate": 0.0001,
+      "loss": 1.4451,
+      "loss/crossentropy": 2.239219903945923,
+      "loss/hidden": 1.140625,
+      "loss/logits": 0.1887097954750061,
+      "loss/reg": 0.011579960584640503,
+      "step": 1890
+    },
+    {
+      "epoch": 0.236375,
+      "grad_norm": 3.4970099925994873,
+      "grad_norm_var": 1.2641245233408538,
+      "learning_rate": 0.0001,
+      "loss": 1.1386,
+      "loss/crossentropy": 2.5096256732940674,
+      "loss/hidden": 0.87109375,
+      "loss/logits": 0.15174484252929688,
+      "loss/reg": 0.011572892777621746,
+      "step": 1891
+    },
+    {
+      "epoch": 0.2365,
+      "grad_norm": 2.6197335720062256,
+      "grad_norm_var": 1.2909410184264962,
+      "learning_rate": 0.0001,
+      "loss": 1.0721,
+      "loss/crossentropy": 2.704296112060547,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.14390236139297485,
+      "loss/reg": 0.01156578678637743,
+      "step": 1892
+    },
+    {
+      "epoch": 0.236625,
+      "grad_norm": 5.325662136077881,
+      "grad_norm_var": 0.6420011074287828,
+      "learning_rate": 0.0001,
+      "loss": 1.5009,
+      "loss/crossentropy": 2.394350528717041,
+      "loss/hidden": 1.2109375,
+      "loss/logits": 0.1743205487728119,
+      "loss/reg": 0.011559294536709785,
+      "step": 1893
+    },
+    {
+      "epoch": 0.23675,
+      "grad_norm": 3.4110569953918457,
+      "grad_norm_var": 0.6408013583584702,
+      "learning_rate": 0.0001,
+      "loss": 1.3052,
+      "loss/crossentropy": 2.3642685413360596,
+      "loss/hidden": 1.0,
+      "loss/logits": 0.18962880969047546,
+      "loss/reg": 0.011552795767784119,
+      "step": 1894
+    },
+    {
+      "epoch": 0.236875,
+      "grad_norm": 4.094106674194336,
+      "grad_norm_var": 0.6502721450147265,
+      "learning_rate": 0.0001,
+      "loss": 1.3815,
+      "loss/crossentropy": 2.5077645778656006,
+      "loss/hidden": 1.0859375,
+      "loss/logits": 0.18015095591545105,
+      "loss/reg": 0.01154585275799036,
+      "step": 1895
+    },
+    {
+      "epoch": 0.237,
+      "grad_norm": 3.3601646423339844,
+      "grad_norm_var": 0.6490610636632368,
+      "learning_rate": 0.0001,
+      "loss": 1.1372,
+      "loss/crossentropy": 2.4785971641540527,
+      "loss/hidden": 0.8671875,
+      "loss/logits": 0.15463027358055115,
+      "loss/reg": 0.011538945138454437,
+      "step": 1896
+    },
+    {
+      "epoch": 0.237125,
+      "grad_norm": 2.699127197265625,
+      "grad_norm_var": 0.6749314875548348,
+      "learning_rate": 0.0001,
+      "loss": 1.0438,
+      "loss/crossentropy": 2.562880754470825,
+      "loss/hidden": 0.80078125,
+      "loss/logits": 0.12774690985679626,
+      "loss/reg": 0.011531817726790905,
+      "step": 1897
+    },
+    {
+      "epoch": 0.23725,
+      "grad_norm": 3.408806324005127,
+      "grad_norm_var": 0.6601303555653694,
+      "learning_rate": 0.0001,
+      "loss": 1.2514,
+      "loss/crossentropy": 2.7521603107452393,
+      "loss/hidden": 0.95703125,
+      "loss/logits": 0.17915445566177368,
+      "loss/reg": 0.01152490172535181,
+      "step": 1898
+    },
+    {
+      "epoch": 0.237375,
+      "grad_norm": 2.782313346862793,
+      "grad_norm_var": 0.6917876208904978,
+      "learning_rate": 0.0001,
+      "loss": 1.2561,
+      "loss/crossentropy": 2.768892765045166,
+      "loss/hidden": 0.96484375,
+      "loss/logits": 0.17611053586006165,
+      "loss/reg": 0.011517412029206753,
+      "step": 1899
+    },
+    {
+      "epoch": 0.2375,
+      "grad_norm": 2.913452625274658,
+      "grad_norm_var": 0.6616557378995289,
+      "learning_rate": 0.0001,
+      "loss": 1.1729,
+      "loss/crossentropy": 2.365873336791992,
+      "loss/hidden": 0.90234375,
+      "loss/logits": 0.15543469786643982,
+      "loss/reg": 0.011509752832353115,
+      "step": 1900
+    },
+    {
+      "epoch": 0.237625,
+      "grad_norm": 2.3335838317871094,
+      "grad_norm_var": 0.736146354285043,
+      "learning_rate": 0.0001,
+      "loss": 0.9944,
+      "loss/crossentropy": 2.6142027378082275,
+      "loss/hidden": 0.75390625,
+      "loss/logits": 0.12545104324817657,
+      "loss/reg": 0.011503461748361588,
+      "step": 1901
+    },
+    {
+      "epoch": 0.23775,
+      "grad_norm": 3.5080623626708984,
+      "grad_norm_var": 0.4975255652152091,
+      "learning_rate": 0.0001,
+      "loss": 1.204,
+      "loss/crossentropy": 2.880540609359741,
+      "loss/hidden": 0.9140625,
+      "loss/logits": 0.17492809891700745,
+      "loss/reg": 0.01149703934788704,
+      "step": 1902
+    },
+    {
+      "epoch": 0.237875,
+      "grad_norm": 2.856538772583008,
+      "grad_norm_var": 0.49293619178501635,
+      "learning_rate": 0.0001,
+      "loss": 1.1438,
+      "loss/crossentropy": 2.411937713623047,
+      "loss/hidden": 0.875,
+      "loss/logits": 0.153866246342659,
+      "loss/reg": 0.01148985605686903,
+      "step": 1903
+    },
+    {
+      "epoch": 0.238,
+      "grad_norm": 2.922060012817383,
+      "grad_norm_var": 0.49565918281477345,
+      "learning_rate": 0.0001,
+      "loss": 1.0083,
+      "loss/crossentropy": 2.714773416519165,
+      "loss/hidden": 0.76171875,
+      "loss/logits": 0.1317380964756012,
+      "loss/reg": 0.011483217589557171,
+      "step": 1904
+    },
+    {
+      "epoch": 0.238125,
+      "grad_norm": 3.1696555614471436,
+      "grad_norm_var": 0.49567159607848527,
+      "learning_rate": 0.0001,
+      "loss": 1.2106,
+      "loss/crossentropy": 2.5127952098846436,
+      "loss/hidden": 0.95703125,
+      "loss/logits": 0.1388104110956192,
+      "loss/reg": 0.011476212181150913,
+      "step": 1905
+    },
+    {
+      "epoch": 0.23825,
+      "grad_norm": 2.5425992012023926,
+      "grad_norm_var": 0.5222804369498891,
+      "learning_rate": 0.0001,
+      "loss": 1.0714,
+      "loss/crossentropy": 2.6032824516296387,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.14816129207611084,
+      "loss/reg": 0.011469176970422268,
+      "step": 1906
+    },
+    {
+      "epoch": 0.238375,
+      "grad_norm": 2.748033285140991,
+      "grad_norm_var": 0.5292028458853089,
+      "learning_rate": 0.0001,
+      "loss": 1.1389,
+      "loss/crossentropy": 2.697582721710205,
+      "loss/hidden": 0.859375,
+      "loss/logits": 0.16493722796440125,
+      "loss/reg": 0.011462134309113026,
+      "step": 1907
+    },
+    {
+      "epoch": 0.2385,
+      "grad_norm": 4.973997592926025,
+      "grad_norm_var": 0.7033744522376982,
+      "learning_rate": 0.0001,
+      "loss": 1.1519,
+      "loss/crossentropy": 2.451634168624878,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.15450610220432281,
+      "loss/reg": 0.011455683968961239,
+      "step": 1908
+    },
+    {
+      "epoch": 0.238625,
+      "grad_norm": 2.706803321838379,
+      "grad_norm_var": 0.43014165554197537,
+      "learning_rate": 0.0001,
+      "loss": 1.0876,
+      "loss/crossentropy": 2.553027629852295,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.15277239680290222,
+      "loss/reg": 0.011448659934103489,
+      "step": 1909
+    },
+    {
+      "epoch": 0.23875,
+      "grad_norm": 2.650634288787842,
+      "grad_norm_var": 0.44000573292169826,
+      "learning_rate": 0.0001,
+      "loss": 1.2477,
+      "loss/crossentropy": 2.413283586502075,
+      "loss/hidden": 0.94921875,
+      "loss/logits": 0.1840301901102066,
+      "loss/reg": 0.011441958136856556,
+      "step": 1910
+    },
+    {
+      "epoch": 0.238875,
+      "grad_norm": 3.073920965194702,
+      "grad_norm_var": 0.37042588009566063,
+      "learning_rate": 0.0001,
+      "loss": 1.24,
+      "loss/crossentropy": 2.647343158721924,
+      "loss/hidden": 0.92578125,
+      "loss/logits": 0.19982083141803741,
+      "loss/reg": 0.011434913612902164,
+      "step": 1911
+    },
+    {
+      "epoch": 0.239,
+      "grad_norm": 4.049685478210449,
+      "grad_norm_var": 0.42951946606552244,
+      "learning_rate": 0.0001,
+      "loss": 1.3637,
+      "loss/crossentropy": 2.2610855102539062,
+      "loss/hidden": 1.046875,
+      "loss/logits": 0.20253784954547882,
+      "loss/reg": 0.01142791099846363,
+      "step": 1912
+    },
+    {
+      "epoch": 0.239125,
+      "grad_norm": 2.926657199859619,
+      "grad_norm_var": 0.42108803087631347,
+      "learning_rate": 0.0001,
+      "loss": 1.2487,
+      "loss/crossentropy": 2.795015335083008,
+      "loss/hidden": 0.9609375,
+      "loss/logits": 0.17352721095085144,
+      "loss/reg": 0.011420897208154202,
+      "step": 1913
+    },
+    {
+      "epoch": 0.23925,
+      "grad_norm": 2.98282790184021,
+      "grad_norm_var": 0.4147719819065732,
+      "learning_rate": 0.0001,
+      "loss": 1.1135,
+      "loss/crossentropy": 2.5007472038269043,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.1634053736925125,
+      "loss/reg": 0.011413791216909885,
+      "step": 1914
+    },
+    {
+      "epoch": 0.239375,
+      "grad_norm": 2.743699073791504,
+      "grad_norm_var": 0.4163530495107797,
+      "learning_rate": 0.0001,
+      "loss": 1.1331,
+      "loss/crossentropy": 2.5021727085113525,
+      "loss/hidden": 0.875,
+      "loss/logits": 0.14406922459602356,
+      "loss/reg": 0.011406795121729374,
+      "step": 1915
+    },
+    {
+      "epoch": 0.2395,
+      "grad_norm": 2.8699593544006348,
+      "grad_norm_var": 0.41737266552279284,
+      "learning_rate": 0.0001,
+      "loss": 1.1846,
+      "loss/crossentropy": 2.409477949142456,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.1604822278022766,
+      "loss/reg": 0.011399611830711365,
+      "step": 1916
+    },
+    {
+      "epoch": 0.239625,
+      "grad_norm": 2.5404951572418213,
+      "grad_norm_var": 0.3998377204796666,
+      "learning_rate": 0.0001,
+      "loss": 1.0893,
+      "loss/crossentropy": 2.4783318042755127,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.1551009714603424,
+      "loss/reg": 0.011393279768526554,
+      "step": 1917
+    },
+    {
+      "epoch": 0.23975,
+      "grad_norm": 2.9057133197784424,
+      "grad_norm_var": 0.38806304932068847,
+      "learning_rate": 0.0001,
+      "loss": 1.055,
+      "loss/crossentropy": 2.5464046001434326,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.1442166268825531,
+      "loss/reg": 0.011386997997760773,
+      "step": 1918
+    },
+    {
+      "epoch": 0.239875,
+      "grad_norm": 3.536914587020874,
+      "grad_norm_var": 0.4002199957694375,
+      "learning_rate": 0.0001,
+      "loss": 1.388,
+      "loss/crossentropy": 2.1978702545166016,
+      "loss/hidden": 1.0625,
+      "loss/logits": 0.21171514689922333,
+      "loss/reg": 0.011380769312381744,
+      "step": 1919
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 2.766721725463867,
+      "grad_norm_var": 0.4050817388615691,
+      "learning_rate": 0.0001,
+      "loss": 1.1325,
+      "loss/crossentropy": 2.6648097038269043,
+      "loss/hidden": 0.859375,
+      "loss/logits": 0.1593654751777649,
+      "loss/reg": 0.011374091729521751,
+      "step": 1920
+    },
+    {
+      "epoch": 0.240125,
+      "grad_norm": 3.2392373085021973,
+      "grad_norm_var": 0.4062692871011743,
+      "learning_rate": 0.0001,
+      "loss": 1.2079,
+      "loss/crossentropy": 2.6759071350097656,
+      "loss/hidden": 0.89453125,
+      "loss/logits": 0.19964686036109924,
+      "loss/reg": 0.01136783231049776,
+      "step": 1921
+    },
+    {
+      "epoch": 0.24025,
+      "grad_norm": 3.552058696746826,
+      "grad_norm_var": 0.39781198223739217,
+      "learning_rate": 0.0001,
+      "loss": 1.2685,
+      "loss/crossentropy": 2.4821066856384277,
+      "loss/hidden": 0.984375,
+      "loss/logits": 0.170503631234169,
+      "loss/reg": 0.011361614800989628,
+      "step": 1922
+    },
+    {
+      "epoch": 0.240375,
+      "grad_norm": 3.387380361557007,
+      "grad_norm_var": 0.38980030472856225,
+      "learning_rate": 0.0001,
+      "loss": 1.1798,
+      "loss/crossentropy": 2.5588643550872803,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.1560502052307129,
+      "loss/reg": 0.011355074122548103,
+      "step": 1923
+    },
+    {
+      "epoch": 0.2405,
+      "grad_norm": 2.437252998352051,
+      "grad_norm_var": 0.18576844254103358,
+      "learning_rate": 0.0001,
+      "loss": 1.2037,
+      "loss/crossentropy": 2.2251975536346436,
+      "loss/hidden": 0.9453125,
+      "loss/logits": 0.1448952853679657,
+      "loss/reg": 0.011348758824169636,
+      "step": 1924
+    },
+    {
+      "epoch": 0.240625,
+      "grad_norm": 2.870936393737793,
+      "grad_norm_var": 0.18052971078875166,
+      "learning_rate": 0.0001,
+      "loss": 1.0809,
+      "loss/crossentropy": 2.538954973220825,
+      "loss/hidden": 0.8203125,
+      "loss/logits": 0.1471145749092102,
+      "loss/reg": 0.011342531070113182,
+      "step": 1925
+    },
+    {
+      "epoch": 0.24075,
+      "grad_norm": 2.4371042251586914,
+      "grad_norm_var": 0.19427645895656553,
+      "learning_rate": 0.0001,
+      "loss": 1.0965,
+      "loss/crossentropy": 2.3479714393615723,
+      "loss/hidden": 0.828125,
+      "loss/logits": 0.15495869517326355,
+      "loss/reg": 0.01133667305111885,
+      "step": 1926
+    },
+    {
+      "epoch": 0.240875,
+      "grad_norm": 2.7412068843841553,
+      "grad_norm_var": 0.19880465575710388,
+      "learning_rate": 0.0001,
+      "loss": 1.0415,
+      "loss/crossentropy": 2.402764081954956,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.13523459434509277,
+      "loss/reg": 0.01133043598383665,
+      "step": 1927
+    },
+    {
+      "epoch": 0.241,
+      "grad_norm": 2.7866973876953125,
+      "grad_norm_var": 0.12160759981975294,
+      "learning_rate": 0.0001,
+      "loss": 1.3153,
+      "loss/crossentropy": 2.327364683151245,
+      "loss/hidden": 0.9921875,
+      "loss/logits": 0.20987921953201294,
+      "loss/reg": 0.011324429884552956,
+      "step": 1928
+    },
+    {
+      "epoch": 0.241125,
+      "grad_norm": 2.642634153366089,
+      "grad_norm_var": 0.1264088206080712,
+      "learning_rate": 0.0001,
+      "loss": 1.1514,
+      "loss/crossentropy": 2.319239616394043,
+      "loss/hidden": 0.8828125,
+      "loss/logits": 0.15535815060138702,
+      "loss/reg": 0.011318519711494446,
+      "step": 1929
+    },
+    {
+      "epoch": 0.24125,
+      "grad_norm": 2.6170005798339844,
+      "grad_norm_var": 0.13085757964727993,
+      "learning_rate": 0.0001,
+      "loss": 1.1138,
+      "loss/crossentropy": 2.5678954124450684,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.15691521763801575,
+      "loss/reg": 0.011312729679048061,
+      "step": 1930
+    },
+    {
+      "epoch": 0.241375,
+      "grad_norm": 3.271930456161499,
+      "grad_norm_var": 0.13871901991367938,
+      "learning_rate": 0.0001,
+      "loss": 1.0944,
+      "loss/crossentropy": 2.6121537685394287,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.15707923471927643,
+      "loss/reg": 0.01130701508373022,
+      "step": 1931
+    },
+    {
+      "epoch": 0.2415,
+      "grad_norm": 4.04671573638916,
+      "grad_norm_var": 0.21855977270830043,
+      "learning_rate": 0.0001,
+      "loss": 1.0816,
+      "loss/crossentropy": 2.6852362155914307,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.1443198323249817,
+      "loss/reg": 0.011301231570541859,
+      "step": 1932
+    },
+    {
+      "epoch": 0.241625,
+      "grad_norm": 2.44777250289917,
+      "grad_norm_var": 0.22460799214468258,
+      "learning_rate": 0.0001,
+      "loss": 1.1546,
+      "loss/crossentropy": 2.544693946838379,
+      "loss/hidden": 0.87890625,
+      "loss/logits": 0.1627037525177002,
+      "loss/reg": 0.011295530013740063,
+      "step": 1933
+    },
+    {
+      "epoch": 0.24175,
+      "grad_norm": 3.0974671840667725,
+      "grad_norm_var": 0.22499515882713012,
+      "learning_rate": 0.0001,
+      "loss": 1.2202,
+      "loss/crossentropy": 2.6100382804870605,
+      "loss/hidden": 0.9296875,
+      "loss/logits": 0.1776452660560608,
+      "loss/reg": 0.011289400048553944,
+      "step": 1934
+    },
+    {
+      "epoch": 0.241875,
+      "grad_norm": 2.6123459339141846,
+      "grad_norm_var": 0.21130123911614854,
+      "learning_rate": 0.0001,
+      "loss": 1.087,
+      "loss/crossentropy": 2.5938217639923096,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.14210623502731323,
+      "loss/reg": 0.011283115483820438,
+      "step": 1935
+    },
+    {
+      "epoch": 0.242,
+      "grad_norm": 2.2436466217041016,
+      "grad_norm_var": 0.24011386438526852,
+      "learning_rate": 0.0001,
+      "loss": 1.0124,
+      "loss/crossentropy": 2.5183980464935303,
+      "loss/hidden": 0.7734375,
+      "loss/logits": 0.12614327669143677,
+      "loss/reg": 0.011277486570179462,
+      "step": 1936
+    },
+    {
+      "epoch": 0.242125,
+      "grad_norm": 5.555136680603027,
+      "grad_norm_var": 0.6794719115699573,
+      "learning_rate": 0.0001,
+      "loss": 1.1126,
+      "loss/crossentropy": 2.543240547180176,
+      "loss/hidden": 0.8671875,
+      "loss/logits": 0.13274727761745453,
+      "loss/reg": 0.011271213181316853,
+      "step": 1937
+    },
+    {
+      "epoch": 0.24225,
+      "grad_norm": 2.610562562942505,
+      "grad_norm_var": 0.6714344269587417,
+      "learning_rate": 0.0001,
+      "loss": 0.8577,
+      "loss/crossentropy": 2.4972715377807617,
+      "loss/hidden": 0.6484375,
+      "loss/logits": 0.09660603851079941,
+      "loss/reg": 0.01126556284725666,
+      "step": 1938
+    },
+    {
+      "epoch": 0.242375,
+      "grad_norm": 3.4906604290008545,
+      "grad_norm_var": 0.6776027391572443,
+      "learning_rate": 0.0001,
+      "loss": 1.2131,
+      "loss/crossentropy": 2.576024293899536,
+      "loss/hidden": 0.91796875,
+      "loss/logits": 0.18251577019691467,
+      "loss/reg": 0.01125932577997446,
+      "step": 1939
+    },
+    {
+      "epoch": 0.2425,
+      "grad_norm": 2.422001361846924,
+      "grad_norm_var": 0.6787500956269599,
+      "learning_rate": 0.0001,
+      "loss": 0.9879,
+      "loss/crossentropy": 2.3907670974731445,
+      "loss/hidden": 0.74609375,
+      "loss/logits": 0.1292981505393982,
+      "loss/reg": 0.011253675445914268,
+      "step": 1940
+    },
+    {
+      "epoch": 0.242625,
+      "grad_norm": 3.6441028118133545,
+      "grad_norm_var": 0.7034908497749977,
+      "learning_rate": 0.0001,
+      "loss": 1.2193,
+      "loss/crossentropy": 2.268623113632202,
+      "loss/hidden": 0.921875,
+      "loss/logits": 0.18494915962219238,
+      "loss/reg": 0.011247408576309681,
+      "step": 1941
+    },
+    {
+      "epoch": 0.24275,
+      "grad_norm": 3.2040059566497803,
+      "grad_norm_var": 0.6784287892697456,
+      "learning_rate": 0.0001,
+      "loss": 1.3474,
+      "loss/crossentropy": 2.369084596633911,
+      "loss/hidden": 1.0234375,
+      "loss/logits": 0.21159708499908447,
+      "loss/reg": 0.011241290718317032,
+      "step": 1942
+    },
+    {
+      "epoch": 0.242875,
+      "grad_norm": 3.0895087718963623,
+      "grad_norm_var": 0.6698306293757182,
+      "learning_rate": 0.0001,
+      "loss": 1.4156,
+      "loss/crossentropy": 2.0874648094177246,
+      "loss/hidden": 1.09375,
+      "loss/logits": 0.20947730541229248,
+      "loss/reg": 0.011235379613935947,
+      "step": 1943
+    },
+    {
+      "epoch": 0.243,
+      "grad_norm": 2.4965028762817383,
+      "grad_norm_var": 0.687657011627713,
+      "learning_rate": 0.0001,
+      "loss": 1.1098,
+      "loss/crossentropy": 2.210761547088623,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.14598006010055542,
+      "loss/reg": 0.011229581199586391,
+      "step": 1944
+    },
+    {
+      "epoch": 0.243125,
+      "grad_norm": 2.9449803829193115,
+      "grad_norm_var": 0.6752047525480802,
+      "learning_rate": 0.0001,
+      "loss": 1.1668,
+      "loss/crossentropy": 2.4518120288848877,
+      "loss/hidden": 0.8671875,
+      "loss/logits": 0.18732941150665283,
+      "loss/reg": 0.011223935522139072,
+      "step": 1945
+    },
+    {
+      "epoch": 0.24325,
+      "grad_norm": 3.066868305206299,
+      "grad_norm_var": 0.6581535524958798,
+      "learning_rate": 0.0001,
+      "loss": 1.29,
+      "loss/crossentropy": 2.822751522064209,
+      "loss/hidden": 0.96484375,
+      "loss/logits": 0.21297214925289154,
+      "loss/reg": 0.01121827308088541,
+      "step": 1946
+    },
+    {
+      "epoch": 0.243375,
+      "grad_norm": 2.5983641147613525,
+      "grad_norm_var": 0.6746843795057048,
+      "learning_rate": 0.0001,
+      "loss": 1.1594,
+      "loss/crossentropy": 2.397672414779663,
+      "loss/hidden": 0.890625,
+      "loss/logits": 0.156642884016037,
+      "loss/reg": 0.01121209841221571,
+      "step": 1947
+    },
+    {
+      "epoch": 0.2435,
+      "grad_norm": 2.9095447063446045,
+      "grad_norm_var": 0.6116848502456292,
+      "learning_rate": 0.0001,
+      "loss": 1.0915,
+      "loss/crossentropy": 2.527217149734497,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.16299590468406677,
+      "loss/reg": 0.01120635587722063,
+      "step": 1948
+    },
+    {
+      "epoch": 0.243625,
+      "grad_norm": 5.4240403175354,
+      "grad_norm_var": 0.9354258383958637,
+      "learning_rate": 0.0001,
+      "loss": 1.4684,
+      "loss/crossentropy": 2.551121473312378,
+      "loss/hidden": 1.1015625,
+      "loss/logits": 0.25484058260917664,
+      "loss/reg": 0.011200698092579842,
+      "step": 1949
+    },
+    {
+      "epoch": 0.24375,
+      "grad_norm": 3.1100833415985107,
+      "grad_norm_var": 0.9352412595018517,
+      "learning_rate": 0.0001,
+      "loss": 1.2134,
+      "loss/crossentropy": 2.401557683944702,
+      "loss/hidden": 0.9140625,
+      "loss/logits": 0.18738284707069397,
+      "loss/reg": 0.011194508522748947,
+      "step": 1950
+    },
+    {
+      "epoch": 0.243875,
+      "grad_norm": 3.456061363220215,
+      "grad_norm_var": 0.9120604979018725,
+      "learning_rate": 0.0001,
+      "loss": 1.1817,
+      "loss/crossentropy": 2.688356637954712,
+      "loss/hidden": 0.875,
+      "loss/logits": 0.19481763243675232,
+      "loss/reg": 0.011188787408173084,
+      "step": 1951
+    },
+    {
+      "epoch": 0.244,
+      "grad_norm": 3.2770729064941406,
+      "grad_norm_var": 0.8378516417593855,
+      "learning_rate": 0.0001,
+      "loss": 1.4059,
+      "loss/crossentropy": 2.156545877456665,
+      "loss/hidden": 1.09375,
+      "loss/logits": 0.2003558874130249,
+      "loss/reg": 0.01118260808289051,
+      "step": 1952
+    },
+    {
+      "epoch": 0.244125,
+      "grad_norm": 3.4741783142089844,
+      "grad_norm_var": 0.4914500706308291,
+      "learning_rate": 0.0001,
+      "loss": 1.259,
+      "loss/crossentropy": 2.633277177810669,
+      "loss/hidden": 0.95703125,
+      "loss/logits": 0.19024121761322021,
+      "loss/reg": 0.011176753789186478,
+      "step": 1953
+    },
+    {
+      "epoch": 0.24425,
+      "grad_norm": 3.2239580154418945,
+      "grad_norm_var": 0.46666341661654465,
+      "learning_rate": 0.0001,
+      "loss": 1.1213,
+      "loss/crossentropy": 2.5291545391082764,
+      "loss/hidden": 0.8671875,
+      "loss/logits": 0.14238449931144714,
+      "loss/reg": 0.011171228252351284,
+      "step": 1954
+    },
+    {
+      "epoch": 0.244375,
+      "grad_norm": 2.7568724155426025,
+      "grad_norm_var": 0.4757426809575984,
+      "learning_rate": 0.0001,
+      "loss": 1.0901,
+      "loss/crossentropy": 2.6687417030334473,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.15422269701957703,
+      "loss/reg": 0.01116592064499855,
+      "step": 1955
+    },
+    {
+      "epoch": 0.2445,
+      "grad_norm": 3.1235272884368896,
+      "grad_norm_var": 0.4343252933982605,
+      "learning_rate": 0.0001,
+      "loss": 1.2777,
+      "loss/crossentropy": 2.0929863452911377,
+      "loss/hidden": 1.0078125,
+      "loss/logits": 0.15828515589237213,
+      "loss/reg": 0.011160776950418949,
+      "step": 1956
+    },
+    {
+      "epoch": 0.244625,
+      "grad_norm": 2.8191616535186768,
+      "grad_norm_var": 0.43213291318467645,
+      "learning_rate": 0.0001,
+      "loss": 1.1685,
+      "loss/crossentropy": 2.562255620956421,
+      "loss/hidden": 0.87109375,
+      "loss/logits": 0.1858426034450531,
+      "loss/reg": 0.011155403219163418,
+      "step": 1957
+    },
+    {
+      "epoch": 0.24475,
+      "grad_norm": 7.025396823883057,
+      "grad_norm_var": 1.35403696610659,
+      "learning_rate": 0.0001,
+      "loss": 1.313,
+      "loss/crossentropy": 2.0798208713531494,
+      "loss/hidden": 1.0546875,
+      "loss/logits": 0.1467897593975067,
+      "loss/reg": 0.01114996150135994,
+      "step": 1958
+    },
+    {
+      "epoch": 0.244875,
+      "grad_norm": 3.6586244106292725,
+      "grad_norm_var": 1.348840874134669,
+      "learning_rate": 0.0001,
+      "loss": 1.0528,
+      "loss/crossentropy": 2.80499267578125,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.14840048551559448,
+      "loss/reg": 0.011143793351948261,
+      "step": 1959
+    },
+    {
+      "epoch": 0.245,
+      "grad_norm": 3.549360752105713,
+      "grad_norm_var": 1.2828200422537663,
+      "learning_rate": 0.0001,
+      "loss": 1.1735,
+      "loss/crossentropy": 2.612431764602661,
+      "loss/hidden": 0.89453125,
+      "loss/logits": 0.16759154200553894,
+      "loss/reg": 0.01113795768469572,
+      "step": 1960
+    },
+    {
+      "epoch": 0.245125,
+      "grad_norm": 3.2019731998443604,
+      "grad_norm_var": 1.2670343380172766,
+      "learning_rate": 0.0001,
+      "loss": 1.2087,
+      "loss/crossentropy": 2.242535352706909,
+      "loss/hidden": 0.93359375,
+      "loss/logits": 0.1637311577796936,
+      "loss/reg": 0.011132647283375263,
+      "step": 1961
+    },
+    {
+      "epoch": 0.24525,
+      "grad_norm": 2.65508770942688,
+      "grad_norm_var": 1.3037293062655784,
+      "learning_rate": 0.0001,
+      "loss": 1.066,
+      "loss/crossentropy": 2.53322172164917,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.1461862325668335,
+      "loss/reg": 0.011126426979899406,
+      "step": 1962
+    },
+    {
+      "epoch": 0.245375,
+      "grad_norm": 2.4140734672546387,
+      "grad_norm_var": 1.3284114469095243,
+      "learning_rate": 0.0001,
+      "loss": 1.1738,
+      "loss/crossentropy": 2.441091299057007,
+      "loss/hidden": 0.91015625,
+      "loss/logits": 0.1524820178747177,
+      "loss/reg": 0.011120946146547794,
+      "step": 1963
+    },
+    {
+      "epoch": 0.2455,
+      "grad_norm": 4.649234771728516,
+      "grad_norm_var": 1.3794622764879967,
+      "learning_rate": 0.0001,
+      "loss": 1.2865,
+      "loss/crossentropy": 2.4274308681488037,
+      "loss/hidden": 1.015625,
+      "loss/logits": 0.15970450639724731,
+      "loss/reg": 0.011115475557744503,
+      "step": 1964
+    },
+    {
+      "epoch": 0.245625,
+      "grad_norm": 2.690500259399414,
+      "grad_norm_var": 1.1866477483632742,
+      "learning_rate": 0.0001,
+      "loss": 1.0634,
+      "loss/crossentropy": 2.594061851501465,
+      "loss/hidden": 0.8125,
+      "loss/logits": 0.13979078829288483,
+      "loss/reg": 0.011109288781881332,
+      "step": 1965
+    },
+    {
+      "epoch": 0.24575,
+      "grad_norm": 3.1549224853515625,
+      "grad_norm_var": 1.1847841066358076,
+      "learning_rate": 0.0001,
+      "loss": 1.037,
+      "loss/crossentropy": 2.6761364936828613,
+      "loss/hidden": 0.796875,
+      "loss/logits": 0.12911373376846313,
+      "loss/reg": 0.011103102937340736,
+      "step": 1966
+    },
+    {
+      "epoch": 0.245875,
+      "grad_norm": 2.8074398040771484,
+      "grad_norm_var": 1.210175941928484,
+      "learning_rate": 0.0001,
+      "loss": 1.0844,
+      "loss/crossentropy": 2.2973666191101074,
+      "loss/hidden": 0.81640625,
+      "loss/logits": 0.15705451369285583,
+      "loss/reg": 0.011097206734120846,
+      "step": 1967
+    },
+    {
+      "epoch": 0.246,
+      "grad_norm": 2.813947916030884,
+      "grad_norm_var": 1.2314860795345413,
+      "learning_rate": 0.0001,
+      "loss": 1.0939,
+      "loss/crossentropy": 2.4095160961151123,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.14708581566810608,
+      "loss/reg": 0.01109134592115879,
+      "step": 1968
+    },
+    {
+      "epoch": 0.246125,
+      "grad_norm": 3.996295928955078,
+      "grad_norm_var": 1.2553489249469367,
+      "learning_rate": 0.0001,
+      "loss": 1.203,
+      "loss/crossentropy": 2.5568830966949463,
+      "loss/hidden": 0.921875,
+      "loss/logits": 0.17023617029190063,
+      "loss/reg": 0.011085476726293564,
+      "step": 1969
+    },
+    {
+      "epoch": 0.24625,
+      "grad_norm": 3.1737494468688965,
+      "grad_norm_var": 1.2567437243872042,
+      "learning_rate": 0.0001,
+      "loss": 1.19,
+      "loss/crossentropy": 2.820736885070801,
+      "loss/hidden": 0.8984375,
+      "loss/logits": 0.18073034286499023,
+      "loss/reg": 0.011080092750489712,
+      "step": 1970
+    },
+    {
+      "epoch": 0.246375,
+      "grad_norm": 3.1411452293395996,
+      "grad_norm_var": 1.2327325542410335,
+      "learning_rate": 0.0001,
+      "loss": 1.0159,
+      "loss/crossentropy": 2.550328493118286,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.1200067549943924,
+      "loss/reg": 0.011074939742684364,
+      "step": 1971
+    },
+    {
+      "epoch": 0.2465,
+      "grad_norm": 2.7992095947265625,
+      "grad_norm_var": 1.2525440065906799,
+      "learning_rate": 0.0001,
+      "loss": 1.0909,
+      "loss/crossentropy": 2.4028799533843994,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.13644427061080933,
+      "loss/reg": 0.011069980449974537,
+      "step": 1972
+    },
+    {
+      "epoch": 0.246625,
+      "grad_norm": 3.7964675426483154,
+      "grad_norm_var": 1.2353292289717575,
+      "learning_rate": 0.0001,
+      "loss": 1.1583,
+      "loss/crossentropy": 2.6295735836029053,
+      "loss/hidden": 0.890625,
+      "loss/logits": 0.1569831818342209,
+      "loss/reg": 0.01106490008533001,
+      "step": 1973
+    },
+    {
+      "epoch": 0.24675,
+      "grad_norm": 2.4510080814361572,
+      "grad_norm_var": 0.3749246635942266,
+      "learning_rate": 0.0001,
+      "loss": 1.0464,
+      "loss/crossentropy": 2.5560388565063477,
+      "loss/hidden": 0.79296875,
+      "loss/logits": 0.14279651641845703,
+      "loss/reg": 0.011059917509555817,
+      "step": 1974
+    },
+    {
+      "epoch": 0.246875,
+      "grad_norm": 2.3039627075195312,
+      "grad_norm_var": 0.4039935905054485,
+      "learning_rate": 0.0001,
+      "loss": 0.9695,
+      "loss/crossentropy": 2.500410556793213,
+      "loss/hidden": 0.7421875,
+      "loss/logits": 0.11676067113876343,
+      "loss/reg": 0.011055227369070053,
+      "step": 1975
+    },
+    {
+      "epoch": 0.247,
+      "grad_norm": 2.6313462257385254,
+      "grad_norm_var": 0.4016504793812447,
+      "learning_rate": 0.0001,
+      "loss": 1.1421,
+      "loss/crossentropy": 2.510258197784424,
+      "loss/hidden": 0.87109375,
+      "loss/logits": 0.16046884655952454,
+      "loss/reg": 0.011049150489270687,
+      "step": 1976
+    },
+    {
+      "epoch": 0.247125,
+      "grad_norm": 3.447317361831665,
+      "grad_norm_var": 0.41062862008729606,
+      "learning_rate": 0.0001,
+      "loss": 1.2887,
+      "loss/crossentropy": 2.2219176292419434,
+      "loss/hidden": 1.0,
+      "loss/logits": 0.17826291918754578,
+      "loss/reg": 0.011043058708310127,
+      "step": 1977
+    },
+    {
+      "epoch": 0.24725,
+      "grad_norm": 2.774780035018921,
+      "grad_norm_var": 0.40509622860432326,
+      "learning_rate": 0.0001,
+      "loss": 1.1503,
+      "loss/crossentropy": 2.6890835762023926,
+      "loss/hidden": 0.8671875,
+      "loss/logits": 0.1727658063173294,
+      "loss/reg": 0.011036898009479046,
+      "step": 1978
+    },
+    {
+      "epoch": 0.247375,
+      "grad_norm": 2.6805171966552734,
+      "grad_norm_var": 0.38639654731341744,
+      "learning_rate": 0.0001,
+      "loss": 1.063,
+      "loss/crossentropy": 2.730665683746338,
+      "loss/hidden": 0.80859375,
+      "loss/logits": 0.14413277804851532,
+      "loss/reg": 0.01103129331022501,
+      "step": 1979
+    },
+    {
+      "epoch": 0.2475,
+      "grad_norm": 2.5546305179595947,
+      "grad_norm_var": 0.2229060548882482,
+      "learning_rate": 0.0001,
+      "loss": 1.0903,
+      "loss/crossentropy": 2.460444688796997,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.13627831637859344,
+      "loss/reg": 0.011025946587324142,
+      "step": 1980
+    },
+    {
+      "epoch": 0.247625,
+      "grad_norm": 3.147676706314087,
+      "grad_norm_var": 0.22008522732602553,
+      "learning_rate": 0.0001,
+      "loss": 1.0293,
+      "loss/crossentropy": 2.780017614364624,
+      "loss/hidden": 0.78515625,
+      "loss/logits": 0.13396324217319489,
+      "loss/reg": 0.011021067388355732,
+      "step": 1981
+    },
+    {
+      "epoch": 0.24775,
+      "grad_norm": 2.545849561691284,
+      "grad_norm_var": 0.22903709663108504,
+      "learning_rate": 0.0001,
+      "loss": 1.0831,
+      "loss/crossentropy": 2.723945379257202,
+      "loss/hidden": 0.84375,
+      "loss/logits": 0.1291472613811493,
+      "loss/reg": 0.011016342788934708,
+      "step": 1982
+    },
+    {
+      "epoch": 0.247875,
+      "grad_norm": 3.5800068378448486,
+      "grad_norm_var": 0.2525227852681437,
+      "learning_rate": 0.0001,
+      "loss": 1.7206,
+      "loss/crossentropy": 1.9947932958602905,
+      "loss/hidden": 1.3203125,
+      "loss/logits": 0.29013922810554504,
+      "loss/reg": 0.01101152878254652,
+      "step": 1983
+    },
+    {
+      "epoch": 0.248,
+      "grad_norm": 2.3426930904388428,
+      "grad_norm_var": 0.27745670304271336,
+      "learning_rate": 0.0001,
+      "loss": 1.0741,
+      "loss/crossentropy": 2.498370885848999,
+      "loss/hidden": 0.82421875,
+      "loss/logits": 0.13984939455986023,
+      "loss/reg": 0.011006931774318218,
+      "step": 1984
+    },
+    {
+      "epoch": 0.248125,
+      "grad_norm": 2.4072375297546387,
+      "grad_norm_var": 0.21579937260172882,
+      "learning_rate": 0.0001,
+      "loss": 1.1224,
+      "loss/crossentropy": 2.3923802375793457,
+      "loss/hidden": 0.83984375,
+      "loss/logits": 0.17257341742515564,
+      "loss/reg": 0.011000873520970345,
+      "step": 1985
+    },
+    {
+      "epoch": 0.24825,
+      "grad_norm": 4.206304550170898,
+      "grad_norm_var": 0.32547872452598553,
+      "learning_rate": 0.0001,
+      "loss": 1.3701,
+      "loss/crossentropy": 2.5781798362731934,
+      "loss/hidden": 1.0234375,
+      "loss/logits": 0.23670879006385803,
+      "loss/reg": 0.010995452292263508,
+      "step": 1986
+    },
+    {
+      "epoch": 0.248375,
+      "grad_norm": 5.367579460144043,
+      "grad_norm_var": 0.6992678587503592,
+      "learning_rate": 0.0001,
+      "loss": 1.381,
+      "loss/crossentropy": 2.560682535171509,
+      "loss/hidden": 1.109375,
+      "loss/logits": 0.16169926524162292,
+      "loss/reg": 0.010989362373948097,
+      "step": 1987
+    },
+    {
+      "epoch": 0.2485,
+      "grad_norm": 2.6005303859710693,
+      "grad_norm_var": 0.708770234220439,
+      "learning_rate": 0.0001,
+      "loss": 1.0953,
+      "loss/crossentropy": 2.4954795837402344,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.149558424949646,
+      "loss/reg": 0.010983383283019066,
+      "step": 1988
+    },
+    {
+      "epoch": 0.248625,
+      "grad_norm": 3.366856098175049,
+      "grad_norm_var": 0.6776825224044613,
+      "learning_rate": 0.0001,
+      "loss": 1.1398,
+      "loss/crossentropy": 2.572629451751709,
+      "loss/hidden": 0.8515625,
+      "loss/logits": 0.17850361764431,
+      "loss/reg": 0.01097831316292286,
+      "step": 1989
+    },
+    {
+      "epoch": 0.24875,
+      "grad_norm": 2.9385433197021484,
+      "grad_norm_var": 0.6551923075237018,
+      "learning_rate": 0.0001,
+      "loss": 1.114,
+      "loss/crossentropy": 2.4854507446289062,
+      "loss/hidden": 0.83203125,
+      "loss/logits": 0.17227290570735931,
+      "loss/reg": 0.010973170399665833,
+      "step": 1990
+    },
+    {
+      "epoch": 0.248875,
+      "grad_norm": 2.1237387657165527,
+      "grad_norm_var": 0.675293446442898,
+      "learning_rate": 0.0001,
+      "loss": 1.0324,
+      "loss/crossentropy": 2.4779257774353027,
+      "loss/hidden": 0.78125,
+      "loss/logits": 0.14143508672714233,
+      "loss/reg": 0.01096703764051199,
+      "step": 1991
+    },
+    {
+      "epoch": 0.249,
+      "grad_norm": 17.35088539123535,
+      "grad_norm_var": 13.405545245645559,
+      "learning_rate": 0.0001,
+      "loss": 1.5438,
+      "loss/crossentropy": 2.1613028049468994,
+      "loss/hidden": 1.2265625,
+      "loss/logits": 0.20763814449310303,
+      "loss/reg": 0.010961621068418026,
+      "step": 1992
+    },
+    {
+      "epoch": 0.249125,
+      "grad_norm": 3.643308162689209,
+      "grad_norm_var": 13.394425808799774,
+      "learning_rate": 0.0001,
+      "loss": 1.5124,
+      "loss/crossentropy": 2.740511894226074,
+      "loss/hidden": 1.125,
+      "loss/logits": 0.27779725193977356,
+      "loss/reg": 0.010955520905554295,
+      "step": 1993
+    },
+    {
+      "epoch": 0.24925,
+      "grad_norm": 2.6261227130889893,
+      "grad_norm_var": 13.419635101303081,
+      "learning_rate": 0.0001,
+      "loss": 1.1103,
+      "loss/crossentropy": 2.4874839782714844,
+      "loss/hidden": 0.859375,
+      "loss/logits": 0.14138638973236084,
+      "loss/reg": 0.010949719697237015,
+      "step": 1994
+    },
+    {
+      "epoch": 0.249375,
+      "grad_norm": 2.8772454261779785,
+      "grad_norm_var": 13.388291796772698,
+      "learning_rate": 0.0001,
+      "loss": 1.1465,
+      "loss/crossentropy": 2.4892168045043945,
+      "loss/hidden": 0.875,
+      "loss/logits": 0.1620655208826065,
+      "loss/reg": 0.010943672619760036,
+      "step": 1995
+    },
+    {
+      "epoch": 0.2495,
+      "grad_norm": 4.019129753112793,
+      "grad_norm_var": 13.244021829599607,
+      "learning_rate": 0.0001,
+      "loss": 1.1853,
+      "loss/crossentropy": 2.284893751144409,
+      "loss/hidden": 0.90234375,
+      "loss/logits": 0.17355592548847198,
+      "loss/reg": 0.010938170365989208,
+      "step": 1996
+    },
+    {
+      "epoch": 0.249625,
+      "grad_norm": 3.0385830402374268,
+      "grad_norm_var": 13.258203172483755,
+      "learning_rate": 0.0001,
+      "loss": 1.1565,
+      "loss/crossentropy": 2.5010361671447754,
+      "loss/hidden": 0.8671875,
+      "loss/logits": 0.18000578880310059,
+      "loss/reg": 0.010932603850960732,
+      "step": 1997
+    },
+    {
+      "epoch": 0.24975,
+      "grad_norm": 2.5925393104553223,
+      "grad_norm_var": 13.248884346858516,
+      "learning_rate": 0.0001,
+      "loss": 1.083,
+      "loss/crossentropy": 2.6594510078430176,
+      "loss/hidden": 0.8359375,
+      "loss/logits": 0.1377822607755661,
+      "loss/reg": 0.010927069000899792,
+      "step": 1998
+    },
+    {
+      "epoch": 0.249875,
+      "grad_norm": 2.673877239227295,
+      "grad_norm_var": 13.359108718093754,
+      "learning_rate": 0.0001,
+      "loss": 0.9031,
+      "loss/crossentropy": 2.683049440383911,
+      "loss/hidden": 0.67578125,
+      "loss/logits": 0.11807288229465485,
+      "loss/reg": 0.010921536944806576,
+      "step": 1999
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 2.3662562370300293,
+      "grad_norm_var": 13.353902173571798,
+      "learning_rate": 0.0001,
+      "loss": 1.0005,
+      "loss/crossentropy": 2.5584921836853027,
+      "loss/hidden": 0.765625,
+      "loss/logits": 0.1257750689983368,
+      "loss/reg": 0.010914883576333523,
+      "step": 2000
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 8000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 9223372036854775807,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": true,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.28811723128832e+17,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}