diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4583 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.119138149556401,
+  "eval_steps": 500,
+  "global_step": 6500,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0063371356147021544,
+      "grad_norm": 9.841416358947754,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 1.3229,
+      "step": 10
+    },
+    {
+      "epoch": 0.012674271229404309,
+      "grad_norm": 10.776036262512207,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 1.3359,
+      "step": 20
+    },
+    {
+      "epoch": 0.019011406844106463,
+      "grad_norm": 8.647089004516602,
+      "learning_rate": 6e-06,
+      "loss": 1.2916,
+      "step": 30
+    },
+    {
+      "epoch": 0.025348542458808618,
+      "grad_norm": 3.833814859390259,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 0.8317,
+      "step": 40
+    },
+    {
+      "epoch": 0.031685678073510776,
+      "grad_norm": 4.45695686340332,
+      "learning_rate": 1e-05,
+      "loss": 0.3917,
+      "step": 50
+    },
+    {
+      "epoch": 0.03802281368821293,
+      "grad_norm": 4.198420524597168,
+      "learning_rate": 1.2e-05,
+      "loss": 0.3646,
+      "step": 60
+    },
+    {
+      "epoch": 0.044359949302915085,
+      "grad_norm": 1.619905948638916,
+      "learning_rate": 1.4000000000000001e-05,
+      "loss": 0.2154,
+      "step": 70
+    },
+    {
+      "epoch": 0.050697084917617236,
+      "grad_norm": 1.2290397882461548,
+      "learning_rate": 1.6000000000000003e-05,
+      "loss": 0.1981,
+      "step": 80
+    },
+    {
+      "epoch": 0.057034220532319393,
+      "grad_norm": 1.4010839462280273,
+      "learning_rate": 1.8e-05,
+      "loss": 0.1628,
+      "step": 90
+    },
+    {
+      "epoch": 0.06337135614702155,
+      "grad_norm": 1.5164788961410522,
+      "learning_rate": 2e-05,
+      "loss": 0.1349,
+      "step": 100
+    },
+    {
+      "epoch": 0.0697084917617237,
+      "grad_norm": 1.504683017730713,
+      "learning_rate": 2.2000000000000003e-05,
+      "loss": 0.1351,
+      "step": 110
+    },
+    {
+      "epoch": 0.07604562737642585,
+      "grad_norm": 1.1718053817749023,
+      "learning_rate": 2.4e-05,
+      "loss": 0.1315,
+      "step": 120
+    },
+    {
+      "epoch": 0.08238276299112801,
+      "grad_norm": 1.2100762128829956,
+      "learning_rate": 2.6000000000000002e-05,
+      "loss": 0.1088,
+      "step": 130
+    },
+    {
+      "epoch": 0.08871989860583017,
+      "grad_norm": 1.060626745223999,
+      "learning_rate": 2.8000000000000003e-05,
+      "loss": 0.1019,
+      "step": 140
+    },
+    {
+      "epoch": 0.09505703422053231,
+      "grad_norm": 1.1521592140197754,
+      "learning_rate": 3e-05,
+      "loss": 0.0981,
+      "step": 150
+    },
+    {
+      "epoch": 0.10139416983523447,
+      "grad_norm": 0.9289679527282715,
+      "learning_rate": 3.2000000000000005e-05,
+      "loss": 0.0904,
+      "step": 160
+    },
+    {
+      "epoch": 0.10773130544993663,
+      "grad_norm": 1.2749850749969482,
+      "learning_rate": 3.4000000000000007e-05,
+      "loss": 0.0903,
+      "step": 170
+    },
+    {
+      "epoch": 0.11406844106463879,
+      "grad_norm": 0.6532027721405029,
+      "learning_rate": 3.6e-05,
+      "loss": 0.0844,
+      "step": 180
+    },
+    {
+      "epoch": 0.12040557667934093,
+      "grad_norm": 1.0130654573440552,
+      "learning_rate": 3.8e-05,
+      "loss": 0.0952,
+      "step": 190
+    },
+    {
+      "epoch": 0.1267427122940431,
+      "grad_norm": 1.218198299407959,
+      "learning_rate": 4e-05,
+      "loss": 0.0875,
+      "step": 200
+    },
+    {
+      "epoch": 0.13307984790874525,
+      "grad_norm": 0.7836218476295471,
+      "learning_rate": 4.2e-05,
+      "loss": 0.0813,
+      "step": 210
+    },
+    {
+      "epoch": 0.1394169835234474,
+      "grad_norm": 0.7320715188980103,
+      "learning_rate": 4.4000000000000006e-05,
+      "loss": 0.0865,
+      "step": 220
+    },
+    {
+      "epoch": 0.14575411913814956,
+      "grad_norm": 0.6925137639045715,
+      "learning_rate": 4.600000000000001e-05,
+      "loss": 0.0651,
+      "step": 230
+    },
+    {
+      "epoch": 0.1520912547528517,
+      "grad_norm": 1.0975725650787354,
+      "learning_rate": 4.8e-05,
+      "loss": 0.0735,
+      "step": 240
+    },
+    {
+      "epoch": 0.15842839036755388,
+      "grad_norm": 0.7768546342849731,
+      "learning_rate": 5e-05,
+      "loss": 0.0647,
+      "step": 250
+    },
+    {
+      "epoch": 0.16476552598225602,
+      "grad_norm": 0.5613494515419006,
+      "learning_rate": 5.2000000000000004e-05,
+      "loss": 0.0608,
+      "step": 260
+    },
+    {
+      "epoch": 0.17110266159695817,
+      "grad_norm": 0.6478670239448547,
+      "learning_rate": 5.4000000000000005e-05,
+      "loss": 0.0526,
+      "step": 270
+    },
+    {
+      "epoch": 0.17743979721166034,
+      "grad_norm": 0.9302045106887817,
+      "learning_rate": 5.6000000000000006e-05,
+      "loss": 0.0657,
+      "step": 280
+    },
+    {
+      "epoch": 0.18377693282636248,
+      "grad_norm": 0.5966052412986755,
+      "learning_rate": 5.8e-05,
+      "loss": 0.0518,
+      "step": 290
+    },
+    {
+      "epoch": 0.19011406844106463,
+      "grad_norm": 0.6293818354606628,
+      "learning_rate": 6e-05,
+      "loss": 0.0617,
+      "step": 300
+    },
+    {
+      "epoch": 0.1964512040557668,
+      "grad_norm": 0.7958998084068298,
+      "learning_rate": 6.2e-05,
+      "loss": 0.0559,
+      "step": 310
+    },
+    {
+      "epoch": 0.20278833967046894,
+      "grad_norm": 0.7211484909057617,
+      "learning_rate": 6.400000000000001e-05,
+      "loss": 0.0534,
+      "step": 320
+    },
+    {
+      "epoch": 0.20912547528517111,
+      "grad_norm": 1.108765959739685,
+      "learning_rate": 6.6e-05,
+      "loss": 0.0572,
+      "step": 330
+    },
+    {
+      "epoch": 0.21546261089987326,
+      "grad_norm": 0.9579587578773499,
+      "learning_rate": 6.800000000000001e-05,
+      "loss": 0.0513,
+      "step": 340
+    },
+    {
+      "epoch": 0.2217997465145754,
+      "grad_norm": 1.0826679468154907,
+      "learning_rate": 7e-05,
+      "loss": 0.0591,
+      "step": 350
+    },
+    {
+      "epoch": 0.22813688212927757,
+      "grad_norm": 0.9620176553726196,
+      "learning_rate": 7.2e-05,
+      "loss": 0.0614,
+      "step": 360
+    },
+    {
+      "epoch": 0.23447401774397972,
+      "grad_norm": 0.3815687298774719,
+      "learning_rate": 7.4e-05,
+      "loss": 0.0452,
+      "step": 370
+    },
+    {
+      "epoch": 0.24081115335868186,
+      "grad_norm": 1.0042985677719116,
+      "learning_rate": 7.6e-05,
+      "loss": 0.0561,
+      "step": 380
+    },
+    {
+      "epoch": 0.24714828897338403,
+      "grad_norm": 0.6386390328407288,
+      "learning_rate": 7.800000000000001e-05,
+      "loss": 0.0619,
+      "step": 390
+    },
+    {
+      "epoch": 0.2534854245880862,
+      "grad_norm": 0.7774208784103394,
+      "learning_rate": 8e-05,
+      "loss": 0.0478,
+      "step": 400
+    },
+    {
+      "epoch": 0.2598225602027883,
+      "grad_norm": 0.7507523894309998,
+      "learning_rate": 8.2e-05,
+      "loss": 0.053,
+      "step": 410
+    },
+    {
+      "epoch": 0.2661596958174905,
+      "grad_norm": 0.8982664346694946,
+      "learning_rate": 8.4e-05,
+      "loss": 0.0585,
+      "step": 420
+    },
+    {
+      "epoch": 0.27249683143219267,
+      "grad_norm": 0.83918297290802,
+      "learning_rate": 8.6e-05,
+      "loss": 0.0527,
+      "step": 430
+    },
+    {
+      "epoch": 0.2788339670468948,
+      "grad_norm": 0.5620638132095337,
+      "learning_rate": 8.800000000000001e-05,
+      "loss": 0.0477,
+      "step": 440
+    },
+    {
+      "epoch": 0.28517110266159695,
+      "grad_norm": 0.7724658250808716,
+      "learning_rate": 9e-05,
+      "loss": 0.0503,
+      "step": 450
+    },
+    {
+      "epoch": 0.2915082382762991,
+      "grad_norm": 0.5999326109886169,
+      "learning_rate": 9.200000000000001e-05,
+      "loss": 0.0426,
+      "step": 460
+    },
+    {
+      "epoch": 0.29784537389100124,
+      "grad_norm": 0.6010283827781677,
+      "learning_rate": 9.4e-05,
+      "loss": 0.0464,
+      "step": 470
+    },
+    {
+      "epoch": 0.3041825095057034,
+      "grad_norm": 0.8040053248405457,
+      "learning_rate": 9.6e-05,
+      "loss": 0.0455,
+      "step": 480
+    },
+    {
+      "epoch": 0.3105196451204056,
+      "grad_norm": 1.1320879459381104,
+      "learning_rate": 9.8e-05,
+      "loss": 0.0513,
+      "step": 490
+    },
+    {
+      "epoch": 0.31685678073510776,
+      "grad_norm": 0.7393860816955566,
+      "learning_rate": 0.0001,
+      "loss": 0.0506,
+      "step": 500
+    },
+    {
+      "epoch": 0.3231939163498099,
+      "grad_norm": 0.9977770447731018,
+      "learning_rate": 9.999972660400536e-05,
+      "loss": 0.0511,
+      "step": 510
+    },
+    {
+      "epoch": 0.32953105196451205,
+      "grad_norm": 0.9153000116348267,
+      "learning_rate": 9.999890641901125e-05,
+      "loss": 0.0446,
+      "step": 520
+    },
+    {
+      "epoch": 0.3358681875792142,
+      "grad_norm": 0.7802489399909973,
+      "learning_rate": 9.999753945398704e-05,
+      "loss": 0.045,
+      "step": 530
+    },
+    {
+      "epoch": 0.34220532319391633,
+      "grad_norm": 0.7009689211845398,
+      "learning_rate": 9.99956257238817e-05,
+      "loss": 0.0537,
+      "step": 540
+    },
+    {
+      "epoch": 0.3485424588086185,
+      "grad_norm": 0.5555402636528015,
+      "learning_rate": 9.999316524962345e-05,
+      "loss": 0.0416,
+      "step": 550
+    },
+    {
+      "epoch": 0.3548795944233207,
+      "grad_norm": 0.613745391368866,
+      "learning_rate": 9.999015805811965e-05,
+      "loss": 0.0463,
+      "step": 560
+    },
+    {
+      "epoch": 0.3612167300380228,
+      "grad_norm": 0.9725404381752014,
+      "learning_rate": 9.998660418225645e-05,
+      "loss": 0.0534,
+      "step": 570
+    },
+    {
+      "epoch": 0.36755386565272496,
+      "grad_norm": 0.6908450126647949,
+      "learning_rate": 9.998250366089848e-05,
+      "loss": 0.0486,
+      "step": 580
+    },
+    {
+      "epoch": 0.37389100126742714,
+      "grad_norm": 0.8201190829277039,
+      "learning_rate": 9.997785653888835e-05,
+      "loss": 0.0381,
+      "step": 590
+    },
+    {
+      "epoch": 0.38022813688212925,
+      "grad_norm": 0.7427898645401001,
+      "learning_rate": 9.997266286704631e-05,
+      "loss": 0.0504,
+      "step": 600
+    },
+    {
+      "epoch": 0.3865652724968314,
+      "grad_norm": 0.5355698466300964,
+      "learning_rate": 9.996692270216947e-05,
+      "loss": 0.0454,
+      "step": 610
+    },
+    {
+      "epoch": 0.3929024081115336,
+      "grad_norm": 0.6287090182304382,
+      "learning_rate": 9.996063610703137e-05,
+      "loss": 0.0481,
+      "step": 620
+    },
+    {
+      "epoch": 0.39923954372623577,
+      "grad_norm": 0.5858126878738403,
+      "learning_rate": 9.995380315038119e-05,
+      "loss": 0.0379,
+      "step": 630
+    },
+    {
+      "epoch": 0.4055766793409379,
+      "grad_norm": 0.3876695930957794,
+      "learning_rate": 9.994642390694308e-05,
+      "loss": 0.0387,
+      "step": 640
+    },
+    {
+      "epoch": 0.41191381495564006,
+      "grad_norm": 0.5101262331008911,
+      "learning_rate": 9.993849845741524e-05,
+      "loss": 0.0499,
+      "step": 650
+    },
+    {
+      "epoch": 0.41825095057034223,
+      "grad_norm": 0.7811976075172424,
+      "learning_rate": 9.993002688846913e-05,
+      "loss": 0.0458,
+      "step": 660
+    },
+    {
+      "epoch": 0.42458808618504434,
+      "grad_norm": 0.6726566553115845,
+      "learning_rate": 9.992100929274846e-05,
+      "loss": 0.0359,
+      "step": 670
+    },
+    {
+      "epoch": 0.4309252217997465,
+      "grad_norm": 0.43720176815986633,
+      "learning_rate": 9.991144576886823e-05,
+      "loss": 0.0525,
+      "step": 680
+    },
+    {
+      "epoch": 0.4372623574144487,
+      "grad_norm": 1.0024137496948242,
+      "learning_rate": 9.990133642141359e-05,
+      "loss": 0.04,
+      "step": 690
+    },
+    {
+      "epoch": 0.4435994930291508,
+      "grad_norm": 0.6060113310813904,
+      "learning_rate": 9.989068136093873e-05,
+      "loss": 0.0354,
+      "step": 700
+    },
+    {
+      "epoch": 0.449936628643853,
+      "grad_norm": 0.8775745630264282,
+      "learning_rate": 9.987948070396571e-05,
+      "loss": 0.0432,
+      "step": 710
+    },
+    {
+      "epoch": 0.45627376425855515,
+      "grad_norm": 0.7044681310653687,
+      "learning_rate": 9.986773457298311e-05,
+      "loss": 0.0347,
+      "step": 720
+    },
+    {
+      "epoch": 0.46261089987325726,
+      "grad_norm": 0.5258815884590149,
+      "learning_rate": 9.985544309644475e-05,
+      "loss": 0.0428,
+      "step": 730
+    },
+    {
+      "epoch": 0.46894803548795944,
+      "grad_norm": 0.5640819072723389,
+      "learning_rate": 9.984260640876821e-05,
+      "loss": 0.0477,
+      "step": 740
+    },
+    {
+      "epoch": 0.4752851711026616,
+      "grad_norm": 0.9232666492462158,
+      "learning_rate": 9.98292246503335e-05,
+      "loss": 0.0398,
+      "step": 750
+    },
+    {
+      "epoch": 0.4816223067173637,
+      "grad_norm": 0.7809548377990723,
+      "learning_rate": 9.981529796748134e-05,
+      "loss": 0.042,
+      "step": 760
+    },
+    {
+      "epoch": 0.4879594423320659,
+      "grad_norm": 0.5453559756278992,
+      "learning_rate": 9.980082651251175e-05,
+      "loss": 0.0413,
+      "step": 770
+    },
+    {
+      "epoch": 0.49429657794676807,
+      "grad_norm": 0.43146708607673645,
+      "learning_rate": 9.97858104436822e-05,
+      "loss": 0.0395,
+      "step": 780
+    },
+    {
+      "epoch": 0.5006337135614702,
+      "grad_norm": 0.6322312951087952,
+      "learning_rate": 9.977024992520602e-05,
+      "loss": 0.0303,
+      "step": 790
+    },
+    {
+      "epoch": 0.5069708491761724,
+      "grad_norm": 0.3923812508583069,
+      "learning_rate": 9.975414512725057e-05,
+      "loss": 0.0417,
+      "step": 800
+    },
+    {
+      "epoch": 0.5133079847908745,
+      "grad_norm": 0.43715372681617737,
+      "learning_rate": 9.973749622593534e-05,
+      "loss": 0.0357,
+      "step": 810
+    },
+    {
+      "epoch": 0.5196451204055766,
+      "grad_norm": 0.4882504343986511,
+      "learning_rate": 9.972030340333001e-05,
+      "loss": 0.0483,
+      "step": 820
+    },
+    {
+      "epoch": 0.5259822560202788,
+      "grad_norm": 0.47166863083839417,
+      "learning_rate": 9.970256684745258e-05,
+      "loss": 0.033,
+      "step": 830
+    },
+    {
+      "epoch": 0.532319391634981,
+      "grad_norm": 0.6284196972846985,
+      "learning_rate": 9.968428675226714e-05,
+      "loss": 0.034,
+      "step": 840
+    },
+    {
+      "epoch": 0.5386565272496832,
+      "grad_norm": 0.34937915205955505,
+      "learning_rate": 9.966546331768191e-05,
+      "loss": 0.0288,
+      "step": 850
+    },
+    {
+      "epoch": 0.5449936628643853,
+      "grad_norm": 0.6910156607627869,
+      "learning_rate": 9.964609674954696e-05,
+      "loss": 0.0413,
+      "step": 860
+    },
+    {
+      "epoch": 0.5513307984790875,
+      "grad_norm": 0.3398362994194031,
+      "learning_rate": 9.962618725965196e-05,
+      "loss": 0.0295,
+      "step": 870
+    },
+    {
+      "epoch": 0.5576679340937896,
+      "grad_norm": 0.5590093731880188,
+      "learning_rate": 9.96057350657239e-05,
+      "loss": 0.0332,
+      "step": 880
+    },
+    {
+      "epoch": 0.5640050697084917,
+      "grad_norm": 0.444553405046463,
+      "learning_rate": 9.95847403914247e-05,
+      "loss": 0.0331,
+      "step": 890
+    },
+    {
+      "epoch": 0.5703422053231939,
+      "grad_norm": 0.4849596619606018,
+      "learning_rate": 9.956320346634876e-05,
+      "loss": 0.0388,
+      "step": 900
+    },
+    {
+      "epoch": 0.5766793409378961,
+      "grad_norm": 0.33844664692878723,
+      "learning_rate": 9.954112452602045e-05,
+      "loss": 0.0324,
+      "step": 910
+    },
+    {
+      "epoch": 0.5830164765525983,
+      "grad_norm": 0.2623671889305115,
+      "learning_rate": 9.95185038118915e-05,
+      "loss": 0.0403,
+      "step": 920
+    },
+    {
+      "epoch": 0.5893536121673004,
+      "grad_norm": 0.5219829678535461,
+      "learning_rate": 9.949534157133844e-05,
+      "loss": 0.0466,
+      "step": 930
+    },
+    {
+      "epoch": 0.5956907477820025,
+      "grad_norm": 0.40922799706459045,
+      "learning_rate": 9.94716380576598e-05,
+      "loss": 0.0295,
+      "step": 940
+    },
+    {
+      "epoch": 0.6020278833967047,
+      "grad_norm": 0.37367522716522217,
+      "learning_rate": 9.944739353007344e-05,
+      "loss": 0.0352,
+      "step": 950
+    },
+    {
+      "epoch": 0.6083650190114068,
+      "grad_norm": 0.8563676476478577,
+      "learning_rate": 9.942260825371358e-05,
+      "loss": 0.0424,
+      "step": 960
+    },
+    {
+      "epoch": 0.614702154626109,
+      "grad_norm": 0.396813303232193,
+      "learning_rate": 9.939728249962807e-05,
+      "loss": 0.0386,
+      "step": 970
+    },
+    {
+      "epoch": 0.6210392902408112,
+      "grad_norm": 0.5535157918930054,
+      "learning_rate": 9.937141654477528e-05,
+      "loss": 0.0396,
+      "step": 980
+    },
+    {
+      "epoch": 0.6273764258555133,
+      "grad_norm": 0.4769563376903534,
+      "learning_rate": 9.934501067202117e-05,
+      "loss": 0.0372,
+      "step": 990
+    },
+    {
+      "epoch": 0.6337135614702155,
+      "grad_norm": 0.5703756809234619,
+      "learning_rate": 9.931806517013612e-05,
+      "loss": 0.0325,
+      "step": 1000
+    },
+    {
+      "epoch": 0.6400506970849176,
+      "grad_norm": 0.31805068254470825,
+      "learning_rate": 9.929058033379181e-05,
+      "loss": 0.0365,
+      "step": 1010
+    },
+    {
+      "epoch": 0.6463878326996197,
+      "grad_norm": 0.5349836945533752,
+      "learning_rate": 9.926255646355804e-05,
+      "loss": 0.0265,
+      "step": 1020
+    },
+    {
+      "epoch": 0.6527249683143219,
+      "grad_norm": 0.5509829521179199,
+      "learning_rate": 9.923399386589933e-05,
+      "loss": 0.0273,
+      "step": 1030
+    },
+    {
+      "epoch": 0.6590621039290241,
+      "grad_norm": 0.5818154215812683,
+      "learning_rate": 9.92048928531717e-05,
+      "loss": 0.0334,
+      "step": 1040
+    },
+    {
+      "epoch": 0.6653992395437263,
+      "grad_norm": 0.633552074432373,
+      "learning_rate": 9.917525374361912e-05,
+      "loss": 0.0297,
+      "step": 1050
+    },
+    {
+      "epoch": 0.6717363751584284,
+      "grad_norm": 0.505277156829834,
+      "learning_rate": 9.914507686137019e-05,
+      "loss": 0.0386,
+      "step": 1060
+    },
+    {
+      "epoch": 0.6780735107731305,
+      "grad_norm": 0.30526870489120483,
+      "learning_rate": 9.911436253643445e-05,
+      "loss": 0.0315,
+      "step": 1070
+    },
+    {
+      "epoch": 0.6844106463878327,
+      "grad_norm": 0.47638586163520813,
+      "learning_rate": 9.90831111046988e-05,
+      "loss": 0.036,
+      "step": 1080
+    },
+    {
+      "epoch": 0.6907477820025348,
+      "grad_norm": 0.38751357793807983,
+      "learning_rate": 9.905132290792394e-05,
+      "loss": 0.0324,
+      "step": 1090
+    },
+    {
+      "epoch": 0.697084917617237,
+      "grad_norm": 0.20010784268379211,
+      "learning_rate": 9.901899829374047e-05,
+      "loss": 0.0231,
+      "step": 1100
+    },
+    {
+      "epoch": 0.7034220532319392,
+      "grad_norm": 0.4107750654220581,
+      "learning_rate": 9.89861376156452e-05,
+      "loss": 0.0251,
+      "step": 1110
+    },
+    {
+      "epoch": 0.7097591888466414,
+      "grad_norm": 0.24435921013355255,
+      "learning_rate": 9.895274123299723e-05,
+      "loss": 0.027,
+      "step": 1120
+    },
+    {
+      "epoch": 0.7160963244613435,
+      "grad_norm": 0.4334017336368561,
+      "learning_rate": 9.891880951101407e-05,
+      "loss": 0.0345,
+      "step": 1130
+    },
+    {
+      "epoch": 0.7224334600760456,
+      "grad_norm": 0.4201258420944214,
+      "learning_rate": 9.888434282076758e-05,
+      "loss": 0.0283,
+      "step": 1140
+    },
+    {
+      "epoch": 0.7287705956907478,
+      "grad_norm": 0.41078707575798035,
+      "learning_rate": 9.884934153917997e-05,
+      "loss": 0.0335,
+      "step": 1150
+    },
+    {
+      "epoch": 0.7351077313054499,
+      "grad_norm": 0.37457314133644104,
+      "learning_rate": 9.881380604901964e-05,
+      "loss": 0.0298,
+      "step": 1160
+    },
+    {
+      "epoch": 0.7414448669201521,
+      "grad_norm": 0.7165825366973877,
+      "learning_rate": 9.877773673889701e-05,
+      "loss": 0.0338,
+      "step": 1170
+    },
+    {
+      "epoch": 0.7477820025348543,
+      "grad_norm": 0.3498821556568146,
+      "learning_rate": 9.87411340032603e-05,
+      "loss": 0.0335,
+      "step": 1180
+    },
+    {
+      "epoch": 0.7541191381495564,
+      "grad_norm": 0.3448495864868164,
+      "learning_rate": 9.870399824239117e-05,
+      "loss": 0.0273,
+      "step": 1190
+    },
+    {
+      "epoch": 0.7604562737642585,
+      "grad_norm": 0.5568084716796875,
+      "learning_rate": 9.86663298624003e-05,
+      "loss": 0.0396,
+      "step": 1200
+    },
+    {
+      "epoch": 0.7667934093789607,
+      "grad_norm": 0.4418658912181854,
+      "learning_rate": 9.862812927522309e-05,
+      "loss": 0.0376,
+      "step": 1210
+    },
+    {
+      "epoch": 0.7731305449936628,
+      "grad_norm": 0.5665590167045593,
+      "learning_rate": 9.858939689861506e-05,
+      "loss": 0.0307,
+      "step": 1220
+    },
+    {
+      "epoch": 0.779467680608365,
+      "grad_norm": 0.45505577325820923,
+      "learning_rate": 9.855013315614725e-05,
+      "loss": 0.0263,
+      "step": 1230
+    },
+    {
+      "epoch": 0.7858048162230672,
+      "grad_norm": 0.4842069745063782,
+      "learning_rate": 9.851033847720166e-05,
+      "loss": 0.0308,
+      "step": 1240
+    },
+    {
+      "epoch": 0.7921419518377694,
+      "grad_norm": 0.4048508405685425,
+      "learning_rate": 9.847001329696653e-05,
+      "loss": 0.0243,
+      "step": 1250
+    },
+    {
+      "epoch": 0.7984790874524715,
+      "grad_norm": 0.24489720165729523,
+      "learning_rate": 9.842915805643155e-05,
+      "loss": 0.0281,
+      "step": 1260
+    },
+    {
+      "epoch": 0.8048162230671736,
+      "grad_norm": 0.533295750617981,
+      "learning_rate": 9.838777320238312e-05,
+      "loss": 0.0263,
+      "step": 1270
+    },
+    {
+      "epoch": 0.8111533586818758,
+      "grad_norm": 0.36894357204437256,
+      "learning_rate": 9.834585918739936e-05,
+      "loss": 0.0316,
+      "step": 1280
+    },
+    {
+      "epoch": 0.8174904942965779,
+      "grad_norm": 0.2535454332828522,
+      "learning_rate": 9.830341646984521e-05,
+      "loss": 0.0294,
+      "step": 1290
+    },
+    {
+      "epoch": 0.8238276299112801,
+      "grad_norm": 0.263375461101532,
+      "learning_rate": 9.826044551386744e-05,
+      "loss": 0.031,
+      "step": 1300
+    },
+    {
+      "epoch": 0.8301647655259823,
+      "grad_norm": 0.4550211429595947,
+      "learning_rate": 9.821694678938953e-05,
+      "loss": 0.0278,
+      "step": 1310
+    },
+    {
+      "epoch": 0.8365019011406845,
+      "grad_norm": 0.5714216828346252,
+      "learning_rate": 9.817292077210659e-05,
+      "loss": 0.0295,
+      "step": 1320
+    },
+    {
+      "epoch": 0.8428390367553865,
+      "grad_norm": 0.6447302103042603,
+      "learning_rate": 9.812836794348004e-05,
+      "loss": 0.0331,
+      "step": 1330
+    },
+    {
+      "epoch": 0.8491761723700887,
+      "grad_norm": 0.4367750585079193,
+      "learning_rate": 9.808328879073251e-05,
+      "loss": 0.0363,
+      "step": 1340
+    },
+    {
+      "epoch": 0.8555133079847909,
+      "grad_norm": 0.34794682264328003,
+      "learning_rate": 9.803768380684242e-05,
+      "loss": 0.0241,
+      "step": 1350
+    },
+    {
+      "epoch": 0.861850443599493,
+      "grad_norm": 0.6189941167831421,
+      "learning_rate": 9.799155349053851e-05,
+      "loss": 0.0325,
+      "step": 1360
+    },
+    {
+      "epoch": 0.8681875792141952,
+      "grad_norm": 0.4560254216194153,
+      "learning_rate": 9.794489834629455e-05,
+      "loss": 0.0242,
+      "step": 1370
+    },
+    {
+      "epoch": 0.8745247148288974,
+      "grad_norm": 0.5019411444664001,
+      "learning_rate": 9.789771888432375e-05,
+      "loss": 0.0321,
+      "step": 1380
+    },
+    {
+      "epoch": 0.8808618504435995,
+      "grad_norm": 0.7201408743858337,
+      "learning_rate": 9.785001562057309e-05,
+      "loss": 0.0334,
+      "step": 1390
+    },
+    {
+      "epoch": 0.8871989860583016,
+      "grad_norm": 0.3263881206512451,
+      "learning_rate": 9.780178907671789e-05,
+      "loss": 0.0292,
+      "step": 1400
+    },
+    {
+      "epoch": 0.8935361216730038,
+      "grad_norm": 0.552166759967804,
+      "learning_rate": 9.775303978015585e-05,
+      "loss": 0.04,
+      "step": 1410
+    },
+    {
+      "epoch": 0.899873257287706,
+      "grad_norm": 0.520951509475708,
+      "learning_rate": 9.77037682640015e-05,
+      "loss": 0.0321,
+      "step": 1420
+    },
+    {
+      "epoch": 0.9062103929024081,
+      "grad_norm": 0.6059055924415588,
+      "learning_rate": 9.765397506708023e-05,
+      "loss": 0.0344,
+      "step": 1430
+    },
+    {
+      "epoch": 0.9125475285171103,
+      "grad_norm": 0.6472659707069397,
+      "learning_rate": 9.760366073392246e-05,
+      "loss": 0.0272,
+      "step": 1440
+    },
+    {
+      "epoch": 0.9188846641318125,
+      "grad_norm": 0.2647618055343628,
+      "learning_rate": 9.755282581475769e-05,
+      "loss": 0.0316,
+      "step": 1450
+    },
+    {
+      "epoch": 0.9252217997465145,
+      "grad_norm": 0.4789205491542816,
+      "learning_rate": 9.750147086550844e-05,
+      "loss": 0.0269,
+      "step": 1460
+    },
+    {
+      "epoch": 0.9315589353612167,
+      "grad_norm": 0.3620862364768982,
+      "learning_rate": 9.744959644778422e-05,
+      "loss": 0.0198,
+      "step": 1470
+    },
+    {
+      "epoch": 0.9378960709759189,
+      "grad_norm": 0.33386313915252686,
+      "learning_rate": 9.739720312887535e-05,
+      "loss": 0.0248,
+      "step": 1480
+    },
+    {
+      "epoch": 0.944233206590621,
+      "grad_norm": 0.3961511254310608,
+      "learning_rate": 9.734429148174675e-05,
+      "loss": 0.0244,
+      "step": 1490
+    },
+    {
+      "epoch": 0.9505703422053232,
+      "grad_norm": 0.29284903407096863,
+      "learning_rate": 9.729086208503174e-05,
+      "loss": 0.0278,
+      "step": 1500
+    },
+    {
+      "epoch": 0.9569074778200254,
+      "grad_norm": 0.2947191596031189,
+      "learning_rate": 9.723691552302562e-05,
+      "loss": 0.0202,
+      "step": 1510
+    },
+    {
+      "epoch": 0.9632446134347274,
+      "grad_norm": 0.5471847653388977,
+      "learning_rate": 9.718245238567939e-05,
+      "loss": 0.0235,
+      "step": 1520
+    },
+    {
+      "epoch": 0.9695817490494296,
+      "grad_norm": 0.6192600727081299,
+      "learning_rate": 9.712747326859315e-05,
+      "loss": 0.0265,
+      "step": 1530
+    },
+    {
+      "epoch": 0.9759188846641318,
+      "grad_norm": 0.49848631024360657,
+      "learning_rate": 9.707197877300974e-05,
+      "loss": 0.0211,
+      "step": 1540
+    },
+    {
+      "epoch": 0.982256020278834,
+      "grad_norm": 0.5650535225868225,
+      "learning_rate": 9.701596950580806e-05,
+      "loss": 0.0323,
+      "step": 1550
+    },
+    {
+      "epoch": 0.9885931558935361,
+      "grad_norm": 0.4596526324748993,
+      "learning_rate": 9.695944607949649e-05,
+      "loss": 0.0214,
+      "step": 1560
+    },
+    {
+      "epoch": 0.9949302915082383,
+      "grad_norm": 0.43541160225868225,
+      "learning_rate": 9.690240911220618e-05,
+      "loss": 0.0249,
+      "step": 1570
+    },
+    {
+      "epoch": 1.0012674271229405,
+      "grad_norm": 0.3652093708515167,
+      "learning_rate": 9.684485922768422e-05,
+      "loss": 0.0217,
+      "step": 1580
+    },
+    {
+      "epoch": 1.0076045627376427,
+      "grad_norm": 0.6113967895507812,
+      "learning_rate": 9.6786797055287e-05,
+      "loss": 0.0288,
+      "step": 1590
+    },
+    {
+      "epoch": 1.0139416983523448,
+      "grad_norm": 0.35289207100868225,
+      "learning_rate": 9.672822322997305e-05,
+      "loss": 0.0233,
+      "step": 1600
+    },
+    {
+      "epoch": 1.020278833967047,
+      "grad_norm": 0.4643147587776184,
+      "learning_rate": 9.66691383922964e-05,
+      "loss": 0.0255,
+      "step": 1610
+    },
+    {
+      "epoch": 1.026615969581749,
+      "grad_norm": 0.186619833111763,
+      "learning_rate": 9.660954318839933e-05,
+      "loss": 0.0239,
+      "step": 1620
+    },
+    {
+      "epoch": 1.0329531051964511,
+      "grad_norm": 0.46518614888191223,
+      "learning_rate": 9.654943827000548e-05,
+      "loss": 0.0275,
+      "step": 1630
+    },
+    {
+      "epoch": 1.0392902408111533,
+      "grad_norm": 0.32950979471206665,
+      "learning_rate": 9.648882429441257e-05,
+      "loss": 0.0289,
+      "step": 1640
+    },
+    {
+      "epoch": 1.0456273764258555,
+      "grad_norm": 0.35503265261650085,
+      "learning_rate": 9.642770192448536e-05,
+      "loss": 0.023,
+      "step": 1650
+    },
+    {
+      "epoch": 1.0519645120405576,
+      "grad_norm": 0.29523852467536926,
+      "learning_rate": 9.636607182864827e-05,
+      "loss": 0.029,
+      "step": 1660
+    },
+    {
+      "epoch": 1.0583016476552598,
+      "grad_norm": 0.5499984622001648,
+      "learning_rate": 9.630393468087818e-05,
+      "loss": 0.0232,
+      "step": 1670
+    },
+    {
+      "epoch": 1.064638783269962,
+      "grad_norm": 0.3574666678905487,
+      "learning_rate": 9.624129116069694e-05,
+      "loss": 0.0232,
+      "step": 1680
+    },
+    {
+      "epoch": 1.0709759188846641,
+      "grad_norm": 0.4880392551422119,
+      "learning_rate": 9.617814195316411e-05,
+      "loss": 0.0348,
+      "step": 1690
+    },
+    {
+      "epoch": 1.0773130544993663,
+      "grad_norm": 0.6365160942077637,
+      "learning_rate": 9.611448774886924e-05,
+      "loss": 0.0299,
+      "step": 1700
+    },
+    {
+      "epoch": 1.0836501901140685,
+      "grad_norm": 0.27736377716064453,
+      "learning_rate": 9.605032924392457e-05,
+      "loss": 0.0219,
+      "step": 1710
+    },
+    {
+      "epoch": 1.0899873257287707,
+      "grad_norm": 0.34657785296440125,
+      "learning_rate": 9.598566713995718e-05,
+      "loss": 0.0291,
+      "step": 1720
+    },
+    {
+      "epoch": 1.0963244613434728,
+      "grad_norm": 0.3848840296268463,
+      "learning_rate": 9.59205021441015e-05,
+      "loss": 0.0244,
+      "step": 1730
+    },
+    {
+      "epoch": 1.102661596958175,
+      "grad_norm": 0.29880332946777344,
+      "learning_rate": 9.58548349689915e-05,
+      "loss": 0.0276,
+      "step": 1740
+    },
+    {
+      "epoch": 1.1089987325728772,
+      "grad_norm": 0.32027584314346313,
+      "learning_rate": 9.578866633275288e-05,
+      "loss": 0.0178,
+      "step": 1750
+    },
+    {
+      "epoch": 1.1153358681875791,
+      "grad_norm": 0.32828688621520996,
+      "learning_rate": 9.572199695899522e-05,
+      "loss": 0.0221,
+      "step": 1760
+    },
+    {
+      "epoch": 1.1216730038022813,
+      "grad_norm": 0.3575758934020996,
+      "learning_rate": 9.565482757680415e-05,
+      "loss": 0.0226,
+      "step": 1770
+    },
+    {
+      "epoch": 1.1280101394169835,
+      "grad_norm": 0.39491409063339233,
+      "learning_rate": 9.558715892073323e-05,
+      "loss": 0.0213,
+      "step": 1780
+    },
+    {
+      "epoch": 1.1343472750316856,
+      "grad_norm": 0.3913297951221466,
+      "learning_rate": 9.551899173079607e-05,
+      "loss": 0.0242,
+      "step": 1790
+    },
+    {
+      "epoch": 1.1406844106463878,
+      "grad_norm": 0.3732500374317169,
+      "learning_rate": 9.545032675245813e-05,
+      "loss": 0.0262,
+      "step": 1800
+    },
+    {
+      "epoch": 1.14702154626109,
+      "grad_norm": 0.3922960162162781,
+      "learning_rate": 9.538116473662861e-05,
+      "loss": 0.0244,
+      "step": 1810
+    },
+    {
+      "epoch": 1.1533586818757922,
+      "grad_norm": 0.30090585350990295,
+      "learning_rate": 9.531150643965223e-05,
+      "loss": 0.019,
+      "step": 1820
+    },
+    {
+      "epoch": 1.1596958174904943,
+      "grad_norm": 0.20526911318302155,
+      "learning_rate": 9.524135262330098e-05,
+      "loss": 0.021,
+      "step": 1830
+    },
+    {
+      "epoch": 1.1660329531051965,
+      "grad_norm": 0.281562864780426,
+      "learning_rate": 9.517070405476575e-05,
+      "loss": 0.0237,
+      "step": 1840
+    },
+    {
+      "epoch": 1.1723700887198987,
+      "grad_norm": 0.3962242603302002,
+      "learning_rate": 9.509956150664796e-05,
+      "loss": 0.023,
+      "step": 1850
+    },
+    {
+      "epoch": 1.1787072243346008,
+      "grad_norm": 0.4091711938381195,
+      "learning_rate": 9.502792575695112e-05,
+      "loss": 0.0227,
+      "step": 1860
+    },
+    {
+      "epoch": 1.1850443599493028,
+      "grad_norm": 0.5421469807624817,
+      "learning_rate": 9.49557975890723e-05,
+      "loss": 0.0222,
+      "step": 1870
+    },
+    {
+      "epoch": 1.1913814955640052,
+      "grad_norm": 0.3753933906555176,
+      "learning_rate": 9.488317779179361e-05,
+      "loss": 0.027,
+      "step": 1880
+    },
+    {
+      "epoch": 1.1977186311787071,
+      "grad_norm": 0.2167666256427765,
+      "learning_rate": 9.481006715927351e-05,
+      "loss": 0.0263,
+      "step": 1890
+    },
+    {
+      "epoch": 1.2040557667934093,
+      "grad_norm": 0.4261206090450287,
+      "learning_rate": 9.473646649103818e-05,
+      "loss": 0.0309,
+      "step": 1900
+    },
+    {
+      "epoch": 1.2103929024081115,
+      "grad_norm": 0.5334855318069458,
+      "learning_rate": 9.46623765919727e-05,
+      "loss": 0.0242,
+      "step": 1910
+    },
+    {
+      "epoch": 1.2167300380228137,
+      "grad_norm": 0.2908576726913452,
+      "learning_rate": 9.458779827231237e-05,
+      "loss": 0.0199,
+      "step": 1920
+    },
+    {
+      "epoch": 1.2230671736375158,
+      "grad_norm": 0.3591972589492798,
+      "learning_rate": 9.451273234763371e-05,
+      "loss": 0.0247,
+      "step": 1930
+    },
+    {
+      "epoch": 1.229404309252218,
+      "grad_norm": 0.3506092131137848,
+      "learning_rate": 9.443717963884569e-05,
+      "loss": 0.0213,
+      "step": 1940
+    },
+    {
+      "epoch": 1.2357414448669202,
+      "grad_norm": 0.2472591996192932,
+      "learning_rate": 9.43611409721806e-05,
+      "loss": 0.0237,
+      "step": 1950
+    },
+    {
+      "epoch": 1.2420785804816223,
+      "grad_norm": 0.33952316641807556,
+      "learning_rate": 9.428461717918511e-05,
+      "loss": 0.0219,
+      "step": 1960
+    },
+    {
+      "epoch": 1.2484157160963245,
+      "grad_norm": 0.17157837748527527,
+      "learning_rate": 9.420760909671118e-05,
+      "loss": 0.0268,
+      "step": 1970
+    },
+    {
+      "epoch": 1.2547528517110267,
+      "grad_norm": 0.2349107414484024,
+      "learning_rate": 9.413011756690685e-05,
+      "loss": 0.0323,
+      "step": 1980
+    },
+    {
+      "epoch": 1.2610899873257289,
+      "grad_norm": 0.3310667872428894,
+      "learning_rate": 9.405214343720707e-05,
+      "loss": 0.022,
+      "step": 1990
+    },
+    {
+      "epoch": 1.2674271229404308,
+      "grad_norm": 0.45049792528152466,
+      "learning_rate": 9.397368756032445e-05,
+      "loss": 0.0236,
+      "step": 2000
+    },
+    {
+      "epoch": 1.2737642585551332,
+      "grad_norm": 0.5270654559135437,
+      "learning_rate": 9.389475079423988e-05,
+      "loss": 0.0233,
+      "step": 2010
+    },
+    {
+      "epoch": 1.2801013941698351,
+      "grad_norm": 0.4595666229724884,
+      "learning_rate": 9.381533400219318e-05,
+      "loss": 0.0224,
+      "step": 2020
+    },
+    {
+      "epoch": 1.2864385297845373,
+      "grad_norm": 0.510229766368866,
+      "learning_rate": 9.373543805267368e-05,
+      "loss": 0.025,
+      "step": 2030
+    },
+    {
+      "epoch": 1.2927756653992395,
+      "grad_norm": 0.38161230087280273,
+      "learning_rate": 9.365506381941066e-05,
+      "loss": 0.0206,
+      "step": 2040
+    },
+    {
+      "epoch": 1.2991128010139417,
+      "grad_norm": 0.3577732443809509,
+      "learning_rate": 9.357421218136386e-05,
+      "loss": 0.0242,
+      "step": 2050
+    },
+    {
+      "epoch": 1.3054499366286438,
+      "grad_norm": 0.2776769995689392,
+      "learning_rate": 9.349288402271388e-05,
+      "loss": 0.0244,
+      "step": 2060
+    },
+    {
+      "epoch": 1.311787072243346,
+      "grad_norm": 0.2369217425584793,
+      "learning_rate": 9.341108023285238e-05,
+      "loss": 0.0274,
+      "step": 2070
+    },
+    {
+      "epoch": 1.3181242078580482,
+      "grad_norm": 0.3153173327445984,
+      "learning_rate": 9.332880170637252e-05,
+      "loss": 0.0223,
+      "step": 2080
+    },
+    {
+      "epoch": 1.3244613434727504,
+      "grad_norm": 0.3351927399635315,
+      "learning_rate": 9.32460493430591e-05,
+      "loss": 0.0191,
+      "step": 2090
+    },
+    {
+      "epoch": 1.3307984790874525,
+      "grad_norm": 0.3907555639743805,
+      "learning_rate": 9.316282404787871e-05,
+      "loss": 0.0199,
+      "step": 2100
+    },
+    {
+      "epoch": 1.3371356147021547,
+      "grad_norm": 0.3048558235168457,
+      "learning_rate": 9.30791267309698e-05,
+      "loss": 0.0257,
+      "step": 2110
+    },
+    {
+      "epoch": 1.3434727503168569,
+      "grad_norm": 0.4380377233028412,
+      "learning_rate": 9.299495830763286e-05,
+      "loss": 0.0317,
+      "step": 2120
+    },
+    {
+      "epoch": 1.3498098859315588,
+      "grad_norm": 0.47419941425323486,
+      "learning_rate": 9.291031969832026e-05,
+      "loss": 0.0294,
+      "step": 2130
+    },
+    {
+      "epoch": 1.3561470215462612,
+      "grad_norm": 0.30324944853782654,
+      "learning_rate": 9.282521182862629e-05,
+      "loss": 0.0233,
+      "step": 2140
+    },
+    {
+      "epoch": 1.3624841571609632,
+      "grad_norm": 0.3189409375190735,
+      "learning_rate": 9.273963562927695e-05,
+      "loss": 0.025,
+      "step": 2150
+    },
+    {
+      "epoch": 1.3688212927756653,
+      "grad_norm": 0.34753671288490295,
+      "learning_rate": 9.265359203611987e-05,
+      "loss": 0.0302,
+      "step": 2160
+    },
+    {
+      "epoch": 1.3751584283903675,
+      "grad_norm": 0.2331438660621643,
+      "learning_rate": 9.256708199011401e-05,
+      "loss": 0.0165,
+      "step": 2170
+    },
+    {
+      "epoch": 1.3814955640050697,
+      "grad_norm": 0.27667322754859924,
+      "learning_rate": 9.248010643731935e-05,
+      "loss": 0.0175,
+      "step": 2180
+    },
+    {
+      "epoch": 1.3878326996197718,
+      "grad_norm": 0.5424685478210449,
+      "learning_rate": 9.239266632888659e-05,
+      "loss": 0.024,
+      "step": 2190
+    },
+    {
+      "epoch": 1.394169835234474,
+      "grad_norm": 0.2840613126754761,
+      "learning_rate": 9.230476262104677e-05,
+      "loss": 0.0187,
+      "step": 2200
+    },
+    {
+      "epoch": 1.4005069708491762,
+      "grad_norm": 0.43624839186668396,
+      "learning_rate": 9.221639627510076e-05,
+      "loss": 0.0253,
+      "step": 2210
+    },
+    {
+      "epoch": 1.4068441064638784,
+      "grad_norm": 0.27831771969795227,
+      "learning_rate": 9.212756825740873e-05,
+      "loss": 0.0206,
+      "step": 2220
+    },
+    {
+      "epoch": 1.4131812420785805,
+      "grad_norm": 0.2653898298740387,
+      "learning_rate": 9.20382795393797e-05,
+      "loss": 0.0228,
+      "step": 2230
+    },
+    {
+      "epoch": 1.4195183776932827,
+      "grad_norm": 0.4191180467605591,
+      "learning_rate": 9.194853109746074e-05,
+      "loss": 0.0204,
+      "step": 2240
+    },
+    {
+      "epoch": 1.4258555133079849,
+      "grad_norm": 0.3698088526725769,
+      "learning_rate": 9.185832391312644e-05,
+      "loss": 0.0201,
+      "step": 2250
+    },
+    {
+      "epoch": 1.4321926489226868,
+      "grad_norm": 0.32586073875427246,
+      "learning_rate": 9.176765897286813e-05,
+      "loss": 0.0311,
+      "step": 2260
+    },
+    {
+      "epoch": 1.4385297845373892,
+      "grad_norm": 0.26891350746154785,
+      "learning_rate": 9.167653726818305e-05,
+      "loss": 0.0228,
+      "step": 2270
+    },
+    {
+      "epoch": 1.4448669201520912,
+      "grad_norm": 0.4289737939834595,
+      "learning_rate": 9.158495979556358e-05,
+      "loss": 0.0221,
+      "step": 2280
+    },
+    {
+      "epoch": 1.4512040557667933,
+      "grad_norm": 0.3612203598022461,
+      "learning_rate": 9.14929275564863e-05,
+      "loss": 0.0215,
+      "step": 2290
+    },
+    {
+      "epoch": 1.4575411913814955,
+      "grad_norm": 0.511667013168335,
+      "learning_rate": 9.140044155740101e-05,
+      "loss": 0.0214,
+      "step": 2300
+    },
+    {
+      "epoch": 1.4638783269961977,
+      "grad_norm": 0.31154608726501465,
+      "learning_rate": 9.130750280971978e-05,
+      "loss": 0.0228,
+      "step": 2310
+    },
+    {
+      "epoch": 1.4702154626108999,
+      "grad_norm": 0.33954188227653503,
+      "learning_rate": 9.121411232980588e-05,
+      "loss": 0.0267,
+      "step": 2320
+    },
+    {
+      "epoch": 1.476552598225602,
+      "grad_norm": 0.4101477563381195,
+      "learning_rate": 9.112027113896262e-05,
+      "loss": 0.0204,
+      "step": 2330
+    },
+    {
+      "epoch": 1.4828897338403042,
+      "grad_norm": 0.3624727427959442,
+      "learning_rate": 9.102598026342222e-05,
+      "loss": 0.0194,
+      "step": 2340
+    },
+    {
+      "epoch": 1.4892268694550064,
+      "grad_norm": 0.29435303807258606,
+      "learning_rate": 9.093124073433463e-05,
+      "loss": 0.0227,
+      "step": 2350
+    },
+    {
+      "epoch": 1.4955640050697085,
+      "grad_norm": 0.3966459035873413,
+      "learning_rate": 9.083605358775612e-05,
+      "loss": 0.02,
+      "step": 2360
+    },
+    {
+      "epoch": 1.5019011406844105,
+      "grad_norm": 0.459172785282135,
+      "learning_rate": 9.074041986463808e-05,
+      "loss": 0.025,
+      "step": 2370
+    },
+    {
+      "epoch": 1.508238276299113,
+      "grad_norm": 0.32978248596191406,
+      "learning_rate": 9.064434061081562e-05,
+      "loss": 0.0186,
+      "step": 2380
+    },
+    {
+      "epoch": 1.5145754119138148,
+      "grad_norm": 0.38167238235473633,
+      "learning_rate": 9.0547816876996e-05,
+      "loss": 0.0245,
+      "step": 2390
+    },
+    {
+      "epoch": 1.5209125475285172,
+      "grad_norm": 0.32322126626968384,
+      "learning_rate": 9.045084971874738e-05,
+      "loss": 0.022,
+      "step": 2400
+    },
+    {
+      "epoch": 1.5272496831432192,
+      "grad_norm": 0.2745392322540283,
+      "learning_rate": 9.035344019648702e-05,
+      "loss": 0.0209,
+      "step": 2410
+    },
+    {
+      "epoch": 1.5335868187579216,
+      "grad_norm": 0.3605414032936096,
+      "learning_rate": 9.025558937546988e-05,
+      "loss": 0.0213,
+      "step": 2420
+    },
+    {
+      "epoch": 1.5399239543726235,
+      "grad_norm": 0.32733994722366333,
+      "learning_rate": 9.015729832577681e-05,
+      "loss": 0.0172,
+      "step": 2430
+    },
+    {
+      "epoch": 1.5462610899873257,
+      "grad_norm": 0.6043369174003601,
+      "learning_rate": 9.005856812230304e-05,
+      "loss": 0.0239,
+      "step": 2440
+    },
+    {
+      "epoch": 1.5525982256020279,
+      "grad_norm": 0.22895248234272003,
+      "learning_rate": 8.995939984474624e-05,
+      "loss": 0.019,
+      "step": 2450
+    },
+    {
+      "epoch": 1.55893536121673,
+      "grad_norm": 0.4059270918369293,
+      "learning_rate": 8.98597945775948e-05,
+      "loss": 0.0224,
+      "step": 2460
+    },
+    {
+      "epoch": 1.5652724968314322,
+      "grad_norm": 0.3693525791168213,
+      "learning_rate": 8.975975341011596e-05,
+      "loss": 0.0187,
+      "step": 2470
+    },
+    {
+      "epoch": 1.5716096324461344,
+      "grad_norm": 0.47394832968711853,
+      "learning_rate": 8.965927743634391e-05,
+      "loss": 0.021,
+      "step": 2480
+    },
+    {
+      "epoch": 1.5779467680608366,
+      "grad_norm": 0.3060173988342285,
+      "learning_rate": 8.955836775506776e-05,
+      "loss": 0.0216,
+      "step": 2490
+    },
+    {
+      "epoch": 1.5842839036755385,
+      "grad_norm": 0.2988053560256958,
+      "learning_rate": 8.945702546981969e-05,
+      "loss": 0.024,
+      "step": 2500
+    },
+    {
+      "epoch": 1.590621039290241,
+      "grad_norm": 0.5819903612136841,
+      "learning_rate": 8.935525168886262e-05,
+      "loss": 0.0225,
+      "step": 2510
+    },
+    {
+      "epoch": 1.5969581749049429,
+      "grad_norm": 0.34001755714416504,
+      "learning_rate": 8.92530475251784e-05,
+      "loss": 0.02,
+      "step": 2520
+    },
+    {
+      "epoch": 1.6032953105196452,
+      "grad_norm": 0.45877692103385925,
+      "learning_rate": 8.91504140964553e-05,
+      "loss": 0.0215,
+      "step": 2530
+    },
+    {
+      "epoch": 1.6096324461343472,
+      "grad_norm": 0.4635033905506134,
+      "learning_rate": 8.90473525250761e-05,
+      "loss": 0.0325,
+      "step": 2540
+    },
+    {
+      "epoch": 1.6159695817490496,
+      "grad_norm": 0.42221376299858093,
+      "learning_rate": 8.894386393810563e-05,
+      "loss": 0.0196,
+      "step": 2550
+    },
+    {
+      "epoch": 1.6223067173637515,
+      "grad_norm": 0.34406083822250366,
+      "learning_rate": 8.883994946727849e-05,
+      "loss": 0.0245,
+      "step": 2560
+    },
+    {
+      "epoch": 1.6286438529784537,
+      "grad_norm": 0.4621538519859314,
+      "learning_rate": 8.873561024898668e-05,
+      "loss": 0.0236,
+      "step": 2570
+    },
+    {
+      "epoch": 1.6349809885931559,
+      "grad_norm": 0.3699153959751129,
+      "learning_rate": 8.863084742426719e-05,
+      "loss": 0.024,
+      "step": 2580
+    },
+    {
+      "epoch": 1.641318124207858,
+      "grad_norm": 0.4299221336841583,
+      "learning_rate": 8.852566213878947e-05,
+      "loss": 0.0173,
+      "step": 2590
+    },
+    {
+      "epoch": 1.6476552598225602,
+      "grad_norm": 0.25442177057266235,
+      "learning_rate": 8.842005554284296e-05,
+      "loss": 0.021,
+      "step": 2600
+    },
+    {
+      "epoch": 1.6539923954372624,
+      "grad_norm": 0.31282278895378113,
+      "learning_rate": 8.831402879132446e-05,
+      "loss": 0.0238,
+      "step": 2610
+    },
+    {
+      "epoch": 1.6603295310519646,
+      "grad_norm": 0.400104820728302,
+      "learning_rate": 8.820758304372557e-05,
+      "loss": 0.0194,
+      "step": 2620
+    },
+    {
+      "epoch": 1.6666666666666665,
+      "grad_norm": 0.5096003413200378,
+      "learning_rate": 8.810071946411989e-05,
+      "loss": 0.0289,
+      "step": 2630
+    },
+    {
+      "epoch": 1.673003802281369,
+      "grad_norm": 0.36454638838768005,
+      "learning_rate": 8.799343922115044e-05,
+      "loss": 0.0245,
+      "step": 2640
+    },
+    {
+      "epoch": 1.6793409378960709,
+      "grad_norm": 0.3085906207561493,
+      "learning_rate": 8.788574348801675e-05,
+      "loss": 0.0225,
+      "step": 2650
+    },
+    {
+      "epoch": 1.6856780735107733,
+      "grad_norm": 0.25602665543556213,
+      "learning_rate": 8.77776334424621e-05,
+      "loss": 0.0202,
+      "step": 2660
+    },
+    {
+      "epoch": 1.6920152091254752,
+      "grad_norm": 0.3288155496120453,
+      "learning_rate": 8.766911026676064e-05,
+      "loss": 0.0307,
+      "step": 2670
+    },
+    {
+      "epoch": 1.6983523447401776,
+      "grad_norm": 0.2693244218826294,
+      "learning_rate": 8.756017514770443e-05,
+      "loss": 0.0169,
+      "step": 2680
+    },
+    {
+      "epoch": 1.7046894803548795,
+      "grad_norm": 0.20038799941539764,
+      "learning_rate": 8.745082927659047e-05,
+      "loss": 0.0177,
+      "step": 2690
+    },
+    {
+      "epoch": 1.7110266159695817,
+      "grad_norm": 0.2368573248386383,
+      "learning_rate": 8.73410738492077e-05,
+      "loss": 0.02,
+      "step": 2700
+    },
+    {
+      "epoch": 1.717363751584284,
+      "grad_norm": 0.3750855326652527,
+      "learning_rate": 8.723091006582389e-05,
+      "loss": 0.0157,
+      "step": 2710
+    },
+    {
+      "epoch": 1.723700887198986,
+      "grad_norm": 0.468447744846344,
+      "learning_rate": 8.71203391311725e-05,
+      "loss": 0.0212,
+      "step": 2720
+    },
+    {
+      "epoch": 1.7300380228136882,
+      "grad_norm": 0.32794639468193054,
+      "learning_rate": 8.700936225443959e-05,
+      "loss": 0.015,
+      "step": 2730
+    },
+    {
+      "epoch": 1.7363751584283904,
+      "grad_norm": 0.5568444132804871,
+      "learning_rate": 8.689798064925049e-05,
+      "loss": 0.0182,
+      "step": 2740
+    },
+    {
+      "epoch": 1.7427122940430926,
+      "grad_norm": 0.36910462379455566,
+      "learning_rate": 8.678619553365659e-05,
+      "loss": 0.0213,
+      "step": 2750
+    },
+    {
+      "epoch": 1.7490494296577945,
+      "grad_norm": 0.21325695514678955,
+      "learning_rate": 8.6674008130122e-05,
+      "loss": 0.0195,
+      "step": 2760
+    },
+    {
+      "epoch": 1.755386565272497,
+      "grad_norm": 0.22268667817115784,
+      "learning_rate": 8.656141966551019e-05,
+      "loss": 0.0192,
+      "step": 2770
+    },
+    {
+      "epoch": 1.7617237008871989,
+      "grad_norm": 0.5589450597763062,
+      "learning_rate": 8.644843137107059e-05,
+      "loss": 0.0227,
+      "step": 2780
+    },
+    {
+      "epoch": 1.7680608365019013,
+      "grad_norm": 0.22853608429431915,
+      "learning_rate": 8.633504448242505e-05,
+      "loss": 0.0234,
+      "step": 2790
+    },
+    {
+      "epoch": 1.7743979721166032,
+      "grad_norm": 0.3304268419742584,
+      "learning_rate": 8.622126023955446e-05,
+      "loss": 0.0202,
+      "step": 2800
+    },
+    {
+      "epoch": 1.7807351077313056,
+      "grad_norm": 0.30013397336006165,
+      "learning_rate": 8.610707988678503e-05,
+      "loss": 0.0189,
+      "step": 2810
+    },
+    {
+      "epoch": 1.7870722433460076,
+      "grad_norm": 0.3276195228099823,
+      "learning_rate": 8.599250467277483e-05,
+      "loss": 0.0174,
+      "step": 2820
+    },
+    {
+      "epoch": 1.7934093789607097,
+      "grad_norm": 0.37568747997283936,
+      "learning_rate": 8.587753585050004e-05,
+      "loss": 0.0217,
+      "step": 2830
+    },
+    {
+      "epoch": 1.799746514575412,
+      "grad_norm": 0.6022403836250305,
+      "learning_rate": 8.576217467724128e-05,
+      "loss": 0.0253,
+      "step": 2840
+    },
+    {
+      "epoch": 1.806083650190114,
+      "grad_norm": 0.3828267455101013,
+      "learning_rate": 8.564642241456986e-05,
+      "loss": 0.0219,
+      "step": 2850
+    },
+    {
+      "epoch": 1.8124207858048162,
+      "grad_norm": 0.37504634261131287,
+      "learning_rate": 8.553028032833397e-05,
+      "loss": 0.021,
+      "step": 2860
+    },
+    {
+      "epoch": 1.8187579214195184,
+      "grad_norm": 0.6533935070037842,
+      "learning_rate": 8.541374968864487e-05,
+      "loss": 0.027,
+      "step": 2870
+    },
+    {
+      "epoch": 1.8250950570342206,
+      "grad_norm": 0.4715104103088379,
+      "learning_rate": 8.529683176986295e-05,
+      "loss": 0.0165,
+      "step": 2880
+    },
+    {
+      "epoch": 1.8314321926489225,
+      "grad_norm": 0.5481289029121399,
+      "learning_rate": 8.517952785058385e-05,
+      "loss": 0.0253,
+      "step": 2890
+    },
+    {
+      "epoch": 1.837769328263625,
+      "grad_norm": 0.3077695667743683,
+      "learning_rate": 8.506183921362443e-05,
+      "loss": 0.0192,
+      "step": 2900
+    },
+    {
+      "epoch": 1.8441064638783269,
+      "grad_norm": 0.5000676512718201,
+      "learning_rate": 8.494376714600878e-05,
+      "loss": 0.0234,
+      "step": 2910
+    },
+    {
+      "epoch": 1.8504435994930293,
+      "grad_norm": 0.253926545381546,
+      "learning_rate": 8.482531293895412e-05,
+      "loss": 0.0225,
+      "step": 2920
+    },
+    {
+      "epoch": 1.8567807351077312,
+      "grad_norm": 0.3240850269794464,
+      "learning_rate": 8.470647788785665e-05,
+      "loss": 0.0185,
+      "step": 2930
+    },
+    {
+      "epoch": 1.8631178707224336,
+      "grad_norm": 0.4573562741279602,
+      "learning_rate": 8.458726329227747e-05,
+      "loss": 0.0218,
+      "step": 2940
+    },
+    {
+      "epoch": 1.8694550063371356,
+      "grad_norm": 0.443369597196579,
+      "learning_rate": 8.44676704559283e-05,
+      "loss": 0.0244,
+      "step": 2950
+    },
+    {
+      "epoch": 1.8757921419518377,
+      "grad_norm": 0.3431839942932129,
+      "learning_rate": 8.434770068665723e-05,
+      "loss": 0.0253,
+      "step": 2960
+    },
+    {
+      "epoch": 1.88212927756654,
+      "grad_norm": 0.37696412205696106,
+      "learning_rate": 8.422735529643444e-05,
+      "loss": 0.0202,
+      "step": 2970
+    },
+    {
+      "epoch": 1.888466413181242,
+      "grad_norm": 0.6237473487854004,
+      "learning_rate": 8.410663560133784e-05,
+      "loss": 0.0197,
+      "step": 2980
+    },
+    {
+      "epoch": 1.8948035487959443,
+      "grad_norm": 0.39705732464790344,
+      "learning_rate": 8.398554292153866e-05,
+      "loss": 0.0239,
+      "step": 2990
+    },
+    {
+      "epoch": 1.9011406844106464,
+      "grad_norm": 0.37278616428375244,
+      "learning_rate": 8.386407858128706e-05,
+      "loss": 0.0289,
+      "step": 3000
+    },
+    {
+      "epoch": 1.9074778200253486,
+      "grad_norm": 0.3443812131881714,
+      "learning_rate": 8.37422439088976e-05,
+      "loss": 0.0217,
+      "step": 3010
+    },
+    {
+      "epoch": 1.9138149556400506,
+      "grad_norm": 0.28766191005706787,
+      "learning_rate": 8.362004023673474e-05,
+      "loss": 0.024,
+      "step": 3020
+    },
+    {
+      "epoch": 1.920152091254753,
+      "grad_norm": 0.17464351654052734,
+      "learning_rate": 8.349746890119826e-05,
+      "loss": 0.0201,
+      "step": 3030
+    },
+    {
+      "epoch": 1.926489226869455,
+      "grad_norm": 0.37782022356987,
+      "learning_rate": 8.337453124270863e-05,
+      "loss": 0.0225,
+      "step": 3040
+    },
+    {
+      "epoch": 1.9328263624841573,
+      "grad_norm": 0.28459540009498596,
+      "learning_rate": 8.32512286056924e-05,
+      "loss": 0.0247,
+      "step": 3050
+    },
+    {
+      "epoch": 1.9391634980988592,
+      "grad_norm": 0.3449212610721588,
+      "learning_rate": 8.31275623385675e-05,
+      "loss": 0.0185,
+      "step": 3060
+    },
+    {
+      "epoch": 1.9455006337135616,
+      "grad_norm": 0.451109379529953,
+      "learning_rate": 8.300353379372834e-05,
+      "loss": 0.0144,
+      "step": 3070
+    },
+    {
+      "epoch": 1.9518377693282636,
+      "grad_norm": 0.44775041937828064,
+      "learning_rate": 8.287914432753123e-05,
+      "loss": 0.0239,
+      "step": 3080
+    },
+    {
+      "epoch": 1.9581749049429658,
+      "grad_norm": 0.3843693435192108,
+      "learning_rate": 8.275439530027948e-05,
+      "loss": 0.0205,
+      "step": 3090
+    },
+    {
+      "epoch": 1.964512040557668,
+      "grad_norm": 0.32762160897254944,
+      "learning_rate": 8.262928807620843e-05,
+      "loss": 0.0222,
+      "step": 3100
+    },
+    {
+      "epoch": 1.97084917617237,
+      "grad_norm": 0.323930025100708,
+      "learning_rate": 8.250382402347065e-05,
+      "loss": 0.0191,
+      "step": 3110
+    },
+    {
+      "epoch": 1.9771863117870723,
+      "grad_norm": 0.26865267753601074,
+      "learning_rate": 8.237800451412095e-05,
+      "loss": 0.0205,
+      "step": 3120
+    },
+    {
+      "epoch": 1.9835234474017744,
+      "grad_norm": 0.27057474851608276,
+      "learning_rate": 8.225183092410128e-05,
+      "loss": 0.0181,
+      "step": 3130
+    },
+    {
+      "epoch": 1.9898605830164766,
+      "grad_norm": 0.38607391715049744,
+      "learning_rate": 8.212530463322583e-05,
+      "loss": 0.0206,
+      "step": 3140
+    },
+    {
+      "epoch": 1.9961977186311786,
+      "grad_norm": 0.3551093637943268,
+      "learning_rate": 8.199842702516583e-05,
+      "loss": 0.0226,
+      "step": 3150
+    },
+    {
+      "epoch": 2.002534854245881,
+      "grad_norm": 0.3289443552494049,
+      "learning_rate": 8.18711994874345e-05,
+      "loss": 0.0159,
+      "step": 3160
+    },
+    {
+      "epoch": 2.008871989860583,
+      "grad_norm": 0.29715219140052795,
+      "learning_rate": 8.174362341137177e-05,
+      "loss": 0.0188,
+      "step": 3170
+    },
+    {
+      "epoch": 2.0152091254752853,
+      "grad_norm": 0.25433430075645447,
+      "learning_rate": 8.161570019212921e-05,
+      "loss": 0.0185,
+      "step": 3180
+    },
+    {
+      "epoch": 2.0215462610899873,
+      "grad_norm": 0.2330443561077118,
+      "learning_rate": 8.148743122865463e-05,
+      "loss": 0.016,
+      "step": 3190
+    },
+    {
+      "epoch": 2.0278833967046896,
+      "grad_norm": 0.3372531235218048,
+      "learning_rate": 8.135881792367686e-05,
+      "loss": 0.0216,
+      "step": 3200
+    },
+    {
+      "epoch": 2.0342205323193916,
+      "grad_norm": 0.24772562086582184,
+      "learning_rate": 8.12298616836904e-05,
+      "loss": 0.0156,
+      "step": 3210
+    },
+    {
+      "epoch": 2.040557667934094,
+      "grad_norm": 0.3528910279273987,
+      "learning_rate": 8.110056391894005e-05,
+      "loss": 0.0149,
+      "step": 3220
+    },
+    {
+      "epoch": 2.046894803548796,
+      "grad_norm": 0.2773556113243103,
+      "learning_rate": 8.097092604340542e-05,
+      "loss": 0.0172,
+      "step": 3230
+    },
+    {
+      "epoch": 2.053231939163498,
+      "grad_norm": 0.39497804641723633,
+      "learning_rate": 8.084094947478556e-05,
+      "loss": 0.0193,
+      "step": 3240
+    },
+    {
+      "epoch": 2.0595690747782003,
+      "grad_norm": 0.2211255133152008,
+      "learning_rate": 8.07106356344834e-05,
+      "loss": 0.0155,
+      "step": 3250
+    },
+    {
+      "epoch": 2.0659062103929022,
+      "grad_norm": 0.3327501118183136,
+      "learning_rate": 8.057998594759022e-05,
+      "loss": 0.0154,
+      "step": 3260
+    },
+    {
+      "epoch": 2.0722433460076046,
+      "grad_norm": 0.3289473056793213,
+      "learning_rate": 8.044900184287007e-05,
+      "loss": 0.0199,
+      "step": 3270
+    },
+    {
+      "epoch": 2.0785804816223066,
+      "grad_norm": 0.36586564779281616,
+      "learning_rate": 8.031768475274413e-05,
+      "loss": 0.0164,
+      "step": 3280
+    },
+    {
+      "epoch": 2.084917617237009,
+      "grad_norm": 0.3189631402492523,
+      "learning_rate": 8.018603611327504e-05,
+      "loss": 0.0168,
+      "step": 3290
+    },
+    {
+      "epoch": 2.091254752851711,
+      "grad_norm": 0.36633414030075073,
+      "learning_rate": 8.005405736415126e-05,
+      "loss": 0.0317,
+      "step": 3300
+    },
+    {
+      "epoch": 2.0975918884664133,
+      "grad_norm": 0.2375175505876541,
+      "learning_rate": 7.992174994867123e-05,
+      "loss": 0.0123,
+      "step": 3310
+    },
+    {
+      "epoch": 2.1039290240811153,
+      "grad_norm": 0.39194151759147644,
+      "learning_rate": 7.978911531372765e-05,
+      "loss": 0.0148,
+      "step": 3320
+    },
+    {
+      "epoch": 2.1102661596958177,
+      "grad_norm": 0.4107789099216461,
+      "learning_rate": 7.965615490979163e-05,
+      "loss": 0.0205,
+      "step": 3330
+    },
+    {
+      "epoch": 2.1166032953105196,
+      "grad_norm": 0.3235548436641693,
+      "learning_rate": 7.952287019089685e-05,
+      "loss": 0.0159,
+      "step": 3340
+    },
+    {
+      "epoch": 2.122940430925222,
+      "grad_norm": 0.2603886127471924,
+      "learning_rate": 7.938926261462366e-05,
+      "loss": 0.0188,
+      "step": 3350
+    },
+    {
+      "epoch": 2.129277566539924,
+      "grad_norm": 0.387168288230896,
+      "learning_rate": 7.925533364208309e-05,
+      "loss": 0.0223,
+      "step": 3360
+    },
+    {
+      "epoch": 2.1356147021546263,
+      "grad_norm": 0.4185580015182495,
+      "learning_rate": 7.912108473790092e-05,
+      "loss": 0.0242,
+      "step": 3370
+    },
+    {
+      "epoch": 2.1419518377693283,
+      "grad_norm": 0.25904420018196106,
+      "learning_rate": 7.898651737020166e-05,
+      "loss": 0.0169,
+      "step": 3380
+    },
+    {
+      "epoch": 2.1482889733840302,
+      "grad_norm": 0.29916971921920776,
+      "learning_rate": 7.88516330105925e-05,
+      "loss": 0.0155,
+      "step": 3390
+    },
+    {
+      "epoch": 2.1546261089987326,
+      "grad_norm": 0.3949578106403351,
+      "learning_rate": 7.871643313414718e-05,
+      "loss": 0.018,
+      "step": 3400
+    },
+    {
+      "epoch": 2.1609632446134346,
+      "grad_norm": 0.31313779950141907,
+      "learning_rate": 7.858091921938988e-05,
+      "loss": 0.0139,
+      "step": 3410
+    },
+    {
+      "epoch": 2.167300380228137,
+      "grad_norm": 0.31156864762306213,
+      "learning_rate": 7.844509274827907e-05,
+      "loss": 0.0231,
+      "step": 3420
+    },
+    {
+      "epoch": 2.173637515842839,
+      "grad_norm": 0.27319496870040894,
+      "learning_rate": 7.830895520619128e-05,
+      "loss": 0.0173,
+      "step": 3430
+    },
+    {
+      "epoch": 2.1799746514575413,
+      "grad_norm": 0.5082638263702393,
+      "learning_rate": 7.817250808190483e-05,
+      "loss": 0.0206,
+      "step": 3440
+    },
+    {
+      "epoch": 2.1863117870722433,
+      "grad_norm": 0.3610362410545349,
+      "learning_rate": 7.803575286758364e-05,
+      "loss": 0.0187,
+      "step": 3450
+    },
+    {
+      "epoch": 2.1926489226869457,
+      "grad_norm": 0.6687695384025574,
+      "learning_rate": 7.789869105876083e-05,
+      "loss": 0.0188,
+      "step": 3460
+    },
+    {
+      "epoch": 2.1989860583016476,
+      "grad_norm": 0.5592470765113831,
+      "learning_rate": 7.776132415432234e-05,
+      "loss": 0.0178,
+      "step": 3470
+    },
+    {
+      "epoch": 2.20532319391635,
+      "grad_norm": 0.27790650725364685,
+      "learning_rate": 7.762365365649067e-05,
+      "loss": 0.0171,
+      "step": 3480
+    },
+    {
+      "epoch": 2.211660329531052,
+      "grad_norm": 0.4534870684146881,
+      "learning_rate": 7.748568107080832e-05,
+      "loss": 0.0219,
+      "step": 3490
+    },
+    {
+      "epoch": 2.2179974651457544,
+      "grad_norm": 0.33770209550857544,
+      "learning_rate": 7.734740790612136e-05,
+      "loss": 0.0184,
+      "step": 3500
+    },
+    {
+      "epoch": 2.2243346007604563,
+      "grad_norm": 0.29577895998954773,
+      "learning_rate": 7.720883567456298e-05,
+      "loss": 0.0194,
+      "step": 3510
+    },
+    {
+      "epoch": 2.2306717363751583,
+      "grad_norm": 0.45840632915496826,
+      "learning_rate": 7.70699658915369e-05,
+      "loss": 0.023,
+      "step": 3520
+    },
+    {
+      "epoch": 2.2370088719898606,
+      "grad_norm": 0.2918465733528137,
+      "learning_rate": 7.693080007570084e-05,
+      "loss": 0.0176,
+      "step": 3530
+    },
+    {
+      "epoch": 2.2433460076045626,
+      "grad_norm": 0.3843630254268646,
+      "learning_rate": 7.679133974894983e-05,
+      "loss": 0.0265,
+      "step": 3540
+    },
+    {
+      "epoch": 2.249683143219265,
+      "grad_norm": 0.24515217542648315,
+      "learning_rate": 7.66515864363997e-05,
+      "loss": 0.0174,
+      "step": 3550
+    },
+    {
+      "epoch": 2.256020278833967,
+      "grad_norm": 0.19862592220306396,
+      "learning_rate": 7.651154166637025e-05,
+      "loss": 0.0215,
+      "step": 3560
+    },
+    {
+      "epoch": 2.2623574144486693,
+      "grad_norm": 0.17352795600891113,
+      "learning_rate": 7.637120697036866e-05,
+      "loss": 0.0278,
+      "step": 3570
+    },
+    {
+      "epoch": 2.2686945500633713,
+      "grad_norm": 0.34404289722442627,
+      "learning_rate": 7.623058388307269e-05,
+      "loss": 0.0208,
+      "step": 3580
+    },
+    {
+      "epoch": 2.2750316856780737,
+      "grad_norm": 0.2361377477645874,
+      "learning_rate": 7.608967394231387e-05,
+      "loss": 0.0215,
+      "step": 3590
+    },
+    {
+      "epoch": 2.2813688212927756,
+      "grad_norm": 0.24867765605449677,
+      "learning_rate": 7.594847868906076e-05,
+      "loss": 0.0197,
+      "step": 3600
+    },
+    {
+      "epoch": 2.2877059569074776,
+      "grad_norm": 0.33056747913360596,
+      "learning_rate": 7.580699966740201e-05,
+      "loss": 0.0206,
+      "step": 3610
+    },
+    {
+      "epoch": 2.29404309252218,
+      "grad_norm": 0.4908096194267273,
+      "learning_rate": 7.566523842452958e-05,
+      "loss": 0.0193,
+      "step": 3620
+    },
+    {
+      "epoch": 2.3003802281368824,
+      "grad_norm": 0.3107684254646301,
+      "learning_rate": 7.552319651072164e-05,
+      "loss": 0.0179,
+      "step": 3630
+    },
+    {
+      "epoch": 2.3067173637515843,
+      "grad_norm": 0.2108030468225479,
+      "learning_rate": 7.538087547932585e-05,
+      "loss": 0.0167,
+      "step": 3640
+    },
+    {
+      "epoch": 2.3130544993662863,
+      "grad_norm": 0.3508223295211792,
+      "learning_rate": 7.52382768867422e-05,
+      "loss": 0.0226,
+      "step": 3650
+    },
+    {
+      "epoch": 2.3193916349809887,
+      "grad_norm": 0.33375847339630127,
+      "learning_rate": 7.509540229240601e-05,
+      "loss": 0.0159,
+      "step": 3660
+    },
+    {
+      "epoch": 2.3257287705956906,
+      "grad_norm": 0.3554519712924957,
+      "learning_rate": 7.495225325877103e-05,
+      "loss": 0.0182,
+      "step": 3670
+    },
+    {
+      "epoch": 2.332065906210393,
+      "grad_norm": 0.38898682594299316,
+      "learning_rate": 7.480883135129211e-05,
+      "loss": 0.0145,
+      "step": 3680
+    },
+    {
+      "epoch": 2.338403041825095,
+      "grad_norm": 0.33638787269592285,
+      "learning_rate": 7.466513813840825e-05,
+      "loss": 0.0211,
+      "step": 3690
+    },
+    {
+      "epoch": 2.3447401774397973,
+      "grad_norm": 0.40711718797683716,
+      "learning_rate": 7.452117519152542e-05,
+      "loss": 0.0196,
+      "step": 3700
+    },
+    {
+      "epoch": 2.3510773130544993,
+      "grad_norm": 0.3523763418197632,
+      "learning_rate": 7.437694408499933e-05,
+      "loss": 0.0216,
+      "step": 3710
+    },
+    {
+      "epoch": 2.3574144486692017,
+      "grad_norm": 0.2812027633190155,
+      "learning_rate": 7.423244639611826e-05,
+      "loss": 0.0207,
+      "step": 3720
+    },
+    {
+      "epoch": 2.3637515842839036,
+      "grad_norm": 0.2593352496623993,
+      "learning_rate": 7.408768370508576e-05,
+      "loss": 0.0198,
+      "step": 3730
+    },
+    {
+      "epoch": 2.3700887198986056,
+      "grad_norm": 0.5507994890213013,
+      "learning_rate": 7.394265759500348e-05,
+      "loss": 0.0249,
+      "step": 3740
+    },
+    {
+      "epoch": 2.376425855513308,
+      "grad_norm": 0.33799615502357483,
+      "learning_rate": 7.379736965185368e-05,
+      "loss": 0.0222,
+      "step": 3750
+    },
+    {
+      "epoch": 2.3827629911280104,
+      "grad_norm": 0.3209291398525238,
+      "learning_rate": 7.365182146448205e-05,
+      "loss": 0.02,
+      "step": 3760
+    },
+    {
+      "epoch": 2.3891001267427123,
+      "grad_norm": 0.34429389238357544,
+      "learning_rate": 7.350601462458024e-05,
+      "loss": 0.0156,
+      "step": 3770
+    },
+    {
+      "epoch": 2.3954372623574143,
+      "grad_norm": 0.29221874475479126,
+      "learning_rate": 7.335995072666848e-05,
+      "loss": 0.0141,
+      "step": 3780
+    },
+    {
+      "epoch": 2.4017743979721167,
+      "grad_norm": 0.37552404403686523,
+      "learning_rate": 7.32136313680782e-05,
+      "loss": 0.0193,
+      "step": 3790
+    },
+    {
+      "epoch": 2.4081115335868186,
+      "grad_norm": 0.25026583671569824,
+      "learning_rate": 7.30670581489344e-05,
+      "loss": 0.0215,
+      "step": 3800
+    },
+    {
+      "epoch": 2.414448669201521,
+      "grad_norm": 0.3728311061859131,
+      "learning_rate": 7.292023267213835e-05,
+      "loss": 0.0173,
+      "step": 3810
+    },
+    {
+      "epoch": 2.420785804816223,
+      "grad_norm": 0.35615667700767517,
+      "learning_rate": 7.277315654334997e-05,
+      "loss": 0.021,
+      "step": 3820
+    },
+    {
+      "epoch": 2.4271229404309254,
+      "grad_norm": 0.25696536898612976,
+      "learning_rate": 7.262583137097018e-05,
+      "loss": 0.018,
+      "step": 3830
+    },
+    {
+      "epoch": 2.4334600760456273,
+      "grad_norm": 0.3925075829029083,
+      "learning_rate": 7.247825876612353e-05,
+      "loss": 0.0168,
+      "step": 3840
+    },
+    {
+      "epoch": 2.4397972116603297,
+      "grad_norm": 0.2698250710964203,
+      "learning_rate": 7.233044034264034e-05,
+      "loss": 0.0178,
+      "step": 3850
+    },
+    {
+      "epoch": 2.4461343472750317,
+      "grad_norm": 0.29354268312454224,
+      "learning_rate": 7.218237771703921e-05,
+      "loss": 0.0183,
+      "step": 3860
+    },
+    {
+      "epoch": 2.4524714828897336,
+      "grad_norm": 0.3076222240924835,
+      "learning_rate": 7.203407250850928e-05,
+      "loss": 0.0167,
+      "step": 3870
+    },
+    {
+      "epoch": 2.458808618504436,
+      "grad_norm": 0.23654355108737946,
+      "learning_rate": 7.188552633889259e-05,
+      "loss": 0.0165,
+      "step": 3880
+    },
+    {
+      "epoch": 2.4651457541191384,
+      "grad_norm": 0.2536214292049408,
+      "learning_rate": 7.173674083266624e-05,
+      "loss": 0.0138,
+      "step": 3890
+    },
+    {
+      "epoch": 2.4714828897338403,
+      "grad_norm": 0.2710602879524231,
+      "learning_rate": 7.158771761692464e-05,
+      "loss": 0.0148,
+      "step": 3900
+    },
+    {
+      "epoch": 2.4778200253485423,
+      "grad_norm": 0.15324455499649048,
+      "learning_rate": 7.143845832136188e-05,
+      "loss": 0.0142,
+      "step": 3910
+    },
+    {
+      "epoch": 2.4841571609632447,
+      "grad_norm": 0.13363996148109436,
+      "learning_rate": 7.128896457825364e-05,
+      "loss": 0.0144,
+      "step": 3920
+    },
+    {
+      "epoch": 2.4904942965779466,
+      "grad_norm": 0.2499360889196396,
+      "learning_rate": 7.113923802243957e-05,
+      "loss": 0.0177,
+      "step": 3930
+    },
+    {
+      "epoch": 2.496831432192649,
+      "grad_norm": 0.1742558777332306,
+      "learning_rate": 7.09892802913053e-05,
+      "loss": 0.0161,
+      "step": 3940
+    },
+    {
+      "epoch": 2.503168567807351,
+      "grad_norm": 0.33764421939849854,
+      "learning_rate": 7.083909302476453e-05,
+      "loss": 0.0189,
+      "step": 3950
+    },
+    {
+      "epoch": 2.5095057034220534,
+      "grad_norm": 0.35038939118385315,
+      "learning_rate": 7.068867786524116e-05,
+      "loss": 0.0236,
+      "step": 3960
+    },
+    {
+      "epoch": 2.5158428390367553,
+      "grad_norm": 0.16515985131263733,
+      "learning_rate": 7.053803645765128e-05,
+      "loss": 0.0169,
+      "step": 3970
+    },
+    {
+      "epoch": 2.5221799746514577,
+      "grad_norm": 0.49210289120674133,
+      "learning_rate": 7.038717044938519e-05,
+      "loss": 0.0188,
+      "step": 3980
+    },
+    {
+      "epoch": 2.5285171102661597,
+      "grad_norm": 0.2284458875656128,
+      "learning_rate": 7.023608149028937e-05,
+      "loss": 0.0159,
+      "step": 3990
+    },
+    {
+      "epoch": 2.5348542458808616,
+      "grad_norm": 0.19401752948760986,
+      "learning_rate": 7.008477123264848e-05,
+      "loss": 0.0168,
+      "step": 4000
+    },
+    {
+      "epoch": 2.541191381495564,
+      "grad_norm": 0.14309851825237274,
+      "learning_rate": 6.993324133116726e-05,
+      "loss": 0.0118,
+      "step": 4010
+    },
+    {
+      "epoch": 2.5475285171102664,
+      "grad_norm": 0.22239387035369873,
+      "learning_rate": 6.978149344295242e-05,
+      "loss": 0.0141,
+      "step": 4020
+    },
+    {
+      "epoch": 2.5538656527249683,
+      "grad_norm": 0.4320535957813263,
+      "learning_rate": 6.962952922749457e-05,
+      "loss": 0.0178,
+      "step": 4030
+    },
+    {
+      "epoch": 2.5602027883396703,
+      "grad_norm": 0.32748010754585266,
+      "learning_rate": 6.947735034665002e-05,
+      "loss": 0.0137,
+      "step": 4040
+    },
+    {
+      "epoch": 2.5665399239543727,
+      "grad_norm": 0.1492803394794464,
+      "learning_rate": 6.932495846462261e-05,
+      "loss": 0.0219,
+      "step": 4050
+    },
+    {
+      "epoch": 2.5728770595690746,
+      "grad_norm": 0.3322237432003021,
+      "learning_rate": 6.917235524794558e-05,
+      "loss": 0.0188,
+      "step": 4060
+    },
+    {
+      "epoch": 2.579214195183777,
+      "grad_norm": 0.40152451395988464,
+      "learning_rate": 6.901954236546323e-05,
+      "loss": 0.0263,
+      "step": 4070
+    },
+    {
+      "epoch": 2.585551330798479,
+      "grad_norm": 0.43540433049201965,
+      "learning_rate": 6.886652148831279e-05,
+      "loss": 0.0146,
+      "step": 4080
+    },
+    {
+      "epoch": 2.5918884664131814,
+      "grad_norm": 0.24758180975914001,
+      "learning_rate": 6.871329428990602e-05,
+      "loss": 0.0146,
+      "step": 4090
+    },
+    {
+      "epoch": 2.5982256020278833,
+      "grad_norm": 0.34365662932395935,
+      "learning_rate": 6.855986244591104e-05,
+      "loss": 0.0166,
+      "step": 4100
+    },
+    {
+      "epoch": 2.6045627376425857,
+      "grad_norm": 0.2038048356771469,
+      "learning_rate": 6.840622763423391e-05,
+      "loss": 0.0151,
+      "step": 4110
+    },
+    {
+      "epoch": 2.6108998732572877,
+      "grad_norm": 0.182295024394989,
+      "learning_rate": 6.825239153500029e-05,
+      "loss": 0.0145,
+      "step": 4120
+    },
+    {
+      "epoch": 2.6172370088719896,
+      "grad_norm": 0.3425464928150177,
+      "learning_rate": 6.809835583053715e-05,
+      "loss": 0.0121,
+      "step": 4130
+    },
+    {
+      "epoch": 2.623574144486692,
+      "grad_norm": 0.2870509624481201,
+      "learning_rate": 6.794412220535426e-05,
+      "loss": 0.0127,
+      "step": 4140
+    },
+    {
+      "epoch": 2.6299112801013944,
+      "grad_norm": 0.34516963362693787,
+      "learning_rate": 6.778969234612584e-05,
+      "loss": 0.0158,
+      "step": 4150
+    },
+    {
+      "epoch": 2.6362484157160964,
+      "grad_norm": 0.27682995796203613,
+      "learning_rate": 6.763506794167208e-05,
+      "loss": 0.0132,
+      "step": 4160
+    },
+    {
+      "epoch": 2.6425855513307983,
+      "grad_norm": 0.3806769549846649,
+      "learning_rate": 6.748025068294067e-05,
+      "loss": 0.0158,
+      "step": 4170
+    },
+    {
+      "epoch": 2.6489226869455007,
+      "grad_norm": 0.20911303162574768,
+      "learning_rate": 6.732524226298841e-05,
+      "loss": 0.0164,
+      "step": 4180
+    },
+    {
+      "epoch": 2.6552598225602027,
+      "grad_norm": 0.263217568397522,
+      "learning_rate": 6.71700443769625e-05,
+      "loss": 0.0174,
+      "step": 4190
+    },
+    {
+      "epoch": 2.661596958174905,
+      "grad_norm": 0.17768670618534088,
+      "learning_rate": 6.701465872208216e-05,
+      "loss": 0.0156,
+      "step": 4200
+    },
+    {
+      "epoch": 2.667934093789607,
+      "grad_norm": 0.30574101209640503,
+      "learning_rate": 6.685908699762002e-05,
+      "loss": 0.0127,
+      "step": 4210
+    },
+    {
+      "epoch": 2.6742712294043094,
+      "grad_norm": 0.2742626965045929,
+      "learning_rate": 6.670333090488356e-05,
+      "loss": 0.0168,
+      "step": 4220
+    },
+    {
+      "epoch": 2.6806083650190113,
+      "grad_norm": 0.24799436330795288,
+      "learning_rate": 6.654739214719641e-05,
+      "loss": 0.0141,
+      "step": 4230
+    },
+    {
+      "epoch": 2.6869455006337137,
+      "grad_norm": 0.33699798583984375,
+      "learning_rate": 6.639127242987988e-05,
+      "loss": 0.0115,
+      "step": 4240
+    },
+    {
+      "epoch": 2.6932826362484157,
+      "grad_norm": 0.4366147816181183,
+      "learning_rate": 6.623497346023418e-05,
+      "loss": 0.0201,
+      "step": 4250
+    },
+    {
+      "epoch": 2.6996197718631176,
+      "grad_norm": 0.29953205585479736,
+      "learning_rate": 6.607849694751977e-05,
+      "loss": 0.0143,
+      "step": 4260
+    },
+    {
+      "epoch": 2.70595690747782,
+      "grad_norm": 0.33452221751213074,
+      "learning_rate": 6.592184460293877e-05,
+      "loss": 0.0144,
+      "step": 4270
+    },
+    {
+      "epoch": 2.7122940430925224,
+      "grad_norm": 0.2584233283996582,
+      "learning_rate": 6.576501813961609e-05,
+      "loss": 0.0161,
+      "step": 4280
+    },
+    {
+      "epoch": 2.7186311787072244,
+      "grad_norm": 0.3979489505290985,
+      "learning_rate": 6.56080192725808e-05,
+      "loss": 0.0282,
+      "step": 4290
+    },
+    {
+      "epoch": 2.7249683143219263,
+      "grad_norm": 0.4084644615650177,
+      "learning_rate": 6.545084971874738e-05,
+      "loss": 0.0155,
+      "step": 4300
+    },
+    {
+      "epoch": 2.7313054499366287,
+      "grad_norm": 0.3948853611946106,
+      "learning_rate": 6.529351119689688e-05,
+      "loss": 0.014,
+      "step": 4310
+    },
+    {
+      "epoch": 2.7376425855513307,
+      "grad_norm": 0.3730246424674988,
+      "learning_rate": 6.513600542765817e-05,
+      "loss": 0.0205,
+      "step": 4320
+    },
+    {
+      "epoch": 2.743979721166033,
+      "grad_norm": 0.30106091499328613,
+      "learning_rate": 6.497833413348909e-05,
+      "loss": 0.0162,
+      "step": 4330
+    },
+    {
+      "epoch": 2.750316856780735,
+      "grad_norm": 0.14317883551120758,
+      "learning_rate": 6.48204990386577e-05,
+      "loss": 0.0134,
+      "step": 4340
+    },
+    {
+      "epoch": 2.7566539923954374,
+      "grad_norm": 0.33047810196876526,
+      "learning_rate": 6.466250186922325e-05,
+      "loss": 0.0138,
+      "step": 4350
+    },
+    {
+      "epoch": 2.7629911280101394,
+      "grad_norm": 0.29450923204421997,
+      "learning_rate": 6.450434435301751e-05,
+      "loss": 0.0166,
+      "step": 4360
+    },
+    {
+      "epoch": 2.7693282636248417,
+      "grad_norm": 0.22328028082847595,
+      "learning_rate": 6.43460282196257e-05,
+      "loss": 0.0159,
+      "step": 4370
+    },
+    {
+      "epoch": 2.7756653992395437,
+      "grad_norm": 0.170170858502388,
+      "learning_rate": 6.418755520036775e-05,
+      "loss": 0.0147,
+      "step": 4380
+    },
+    {
+      "epoch": 2.7820025348542456,
+      "grad_norm": 0.3056323826313019,
+      "learning_rate": 6.402892702827916e-05,
+      "loss": 0.0193,
+      "step": 4390
+    },
+    {
+      "epoch": 2.788339670468948,
+      "grad_norm": 0.19363383948802948,
+      "learning_rate": 6.387014543809223e-05,
+      "loss": 0.0134,
+      "step": 4400
+    },
+    {
+      "epoch": 2.7946768060836504,
+      "grad_norm": 0.4496561586856842,
+      "learning_rate": 6.371121216621698e-05,
+      "loss": 0.0132,
+      "step": 4410
+    },
+    {
+      "epoch": 2.8010139416983524,
+      "grad_norm": 0.17226959764957428,
+      "learning_rate": 6.355212895072223e-05,
+      "loss": 0.0119,
+      "step": 4420
+    },
+    {
+      "epoch": 2.8073510773130543,
+      "grad_norm": 0.2828828692436218,
+      "learning_rate": 6.339289753131649e-05,
+      "loss": 0.016,
+      "step": 4430
+    },
+    {
+      "epoch": 2.8136882129277567,
+      "grad_norm": 0.28239116072654724,
+      "learning_rate": 6.323351964932908e-05,
+      "loss": 0.015,
+      "step": 4440
+    },
+    {
+      "epoch": 2.8200253485424587,
+      "grad_norm": 0.2226468324661255,
+      "learning_rate": 6.307399704769099e-05,
+      "loss": 0.0135,
+      "step": 4450
+    },
+    {
+      "epoch": 2.826362484157161,
+      "grad_norm": 0.23484410345554352,
+      "learning_rate": 6.291433147091583e-05,
+      "loss": 0.015,
+      "step": 4460
+    },
+    {
+      "epoch": 2.832699619771863,
+      "grad_norm": 0.2702759802341461,
+      "learning_rate": 6.275452466508077e-05,
+      "loss": 0.016,
+      "step": 4470
+    },
+    {
+      "epoch": 2.8390367553865654,
+      "grad_norm": 0.39190873503685,
+      "learning_rate": 6.259457837780742e-05,
+      "loss": 0.0116,
+      "step": 4480
+    },
+    {
+      "epoch": 2.8453738910012674,
+      "grad_norm": 0.4797994792461395,
+      "learning_rate": 6.243449435824276e-05,
+      "loss": 0.0148,
+      "step": 4490
+    },
+    {
+      "epoch": 2.8517110266159698,
+      "grad_norm": 0.29361021518707275,
+      "learning_rate": 6.227427435703997e-05,
+      "loss": 0.025,
+      "step": 4500
+    },
+    {
+      "epoch": 2.8580481622306717,
+      "grad_norm": 0.26431670784950256,
+      "learning_rate": 6.211392012633932e-05,
+      "loss": 0.0116,
+      "step": 4510
+    },
+    {
+      "epoch": 2.8643852978453737,
+      "grad_norm": 0.2542441189289093,
+      "learning_rate": 6.195343341974899e-05,
+      "loss": 0.0191,
+      "step": 4520
+    },
+    {
+      "epoch": 2.870722433460076,
+      "grad_norm": 0.21877507865428925,
+      "learning_rate": 6.179281599232591e-05,
+      "loss": 0.016,
+      "step": 4530
+    },
+    {
+      "epoch": 2.8770595690747784,
+      "grad_norm": 0.3358696401119232,
+      "learning_rate": 6.163206960055651e-05,
+      "loss": 0.0152,
+      "step": 4540
+    },
+    {
+      "epoch": 2.8833967046894804,
+      "grad_norm": 0.3265288174152374,
+      "learning_rate": 6.147119600233758e-05,
+      "loss": 0.0199,
+      "step": 4550
+    },
+    {
+      "epoch": 2.8897338403041823,
+      "grad_norm": 0.3462512195110321,
+      "learning_rate": 6.131019695695702e-05,
+      "loss": 0.0141,
+      "step": 4560
+    },
+    {
+      "epoch": 2.8960709759188847,
+      "grad_norm": 0.24054935574531555,
+      "learning_rate": 6.11490742250746e-05,
+      "loss": 0.0133,
+      "step": 4570
+    },
+    {
+      "epoch": 2.9024081115335867,
+      "grad_norm": 0.2769206166267395,
+      "learning_rate": 6.0987829568702656e-05,
+      "loss": 0.0177,
+      "step": 4580
+    },
+    {
+      "epoch": 2.908745247148289,
+      "grad_norm": 0.23850680887699127,
+      "learning_rate": 6.0826464751186994e-05,
+      "loss": 0.0112,
+      "step": 4590
+    },
+    {
+      "epoch": 2.915082382762991,
+      "grad_norm": 0.20703135430812836,
+      "learning_rate": 6.066498153718735e-05,
+      "loss": 0.0117,
+      "step": 4600
+    },
+    {
+      "epoch": 2.9214195183776934,
+      "grad_norm": 0.22506844997406006,
+      "learning_rate": 6.05033816926583e-05,
+      "loss": 0.0162,
+      "step": 4610
+    },
+    {
+      "epoch": 2.9277566539923954,
+      "grad_norm": 0.30291593074798584,
+      "learning_rate": 6.034166698482984e-05,
+      "loss": 0.0147,
+      "step": 4620
+    },
+    {
+      "epoch": 2.9340937896070978,
+      "grad_norm": 0.24053440988063812,
+      "learning_rate": 6.017983918218812e-05,
+      "loss": 0.0091,
+      "step": 4630
+    },
+    {
+      "epoch": 2.9404309252217997,
+      "grad_norm": 0.27635499835014343,
+      "learning_rate": 6.001790005445607e-05,
+      "loss": 0.0154,
+      "step": 4640
+    },
+    {
+      "epoch": 2.9467680608365017,
+      "grad_norm": 0.29039266705513,
+      "learning_rate": 5.985585137257401e-05,
+      "loss": 0.0165,
+      "step": 4650
+    },
+    {
+      "epoch": 2.953105196451204,
+      "grad_norm": 0.1605728417634964,
+      "learning_rate": 5.969369490868042e-05,
+      "loss": 0.0154,
+      "step": 4660
+    },
+    {
+      "epoch": 2.9594423320659065,
+      "grad_norm": 0.2104729562997818,
+      "learning_rate": 5.953143243609235e-05,
+      "loss": 0.0224,
+      "step": 4670
+    },
+    {
+      "epoch": 2.9657794676806084,
+      "grad_norm": 0.21257384121418,
+      "learning_rate": 5.9369065729286245e-05,
+      "loss": 0.0118,
+      "step": 4680
+    },
+    {
+      "epoch": 2.9721166032953104,
+      "grad_norm": 0.2738170027732849,
+      "learning_rate": 5.9206596563878357e-05,
+      "loss": 0.0183,
+      "step": 4690
+    },
+    {
+      "epoch": 2.9784537389100127,
+      "grad_norm": 0.2844460904598236,
+      "learning_rate": 5.90440267166055e-05,
+      "loss": 0.0156,
+      "step": 4700
+    },
+    {
+      "epoch": 2.9847908745247147,
+      "grad_norm": 0.3566923439502716,
+      "learning_rate": 5.888135796530544e-05,
+      "loss": 0.0134,
+      "step": 4710
+    },
+    {
+      "epoch": 2.991128010139417,
+      "grad_norm": 0.1965407133102417,
+      "learning_rate": 5.871859208889759e-05,
+      "loss": 0.013,
+      "step": 4720
+    },
+    {
+      "epoch": 2.997465145754119,
+      "grad_norm": 0.18356238305568695,
+      "learning_rate": 5.85557308673635e-05,
+      "loss": 0.0113,
+      "step": 4730
+    },
+    {
+      "epoch": 3.0038022813688214,
+      "grad_norm": 0.23230914771556854,
+      "learning_rate": 5.8392776081727385e-05,
+      "loss": 0.0164,
+      "step": 4740
+    },
+    {
+      "epoch": 3.0101394169835234,
+      "grad_norm": 0.27572014927864075,
+      "learning_rate": 5.8229729514036705e-05,
+      "loss": 0.0113,
+      "step": 4750
+    },
+    {
+      "epoch": 3.016476552598226,
+      "grad_norm": 0.36763879656791687,
+      "learning_rate": 5.8066592947342555e-05,
+      "loss": 0.0124,
+      "step": 4760
+    },
+    {
+      "epoch": 3.0228136882129277,
+      "grad_norm": 0.18714648485183716,
+      "learning_rate": 5.7903368165680327e-05,
+      "loss": 0.0143,
+      "step": 4770
+    },
+    {
+      "epoch": 3.02915082382763,
+      "grad_norm": 0.22437100112438202,
+      "learning_rate": 5.7740056954050084e-05,
+      "loss": 0.0176,
+      "step": 4780
+    },
+    {
+      "epoch": 3.035487959442332,
+      "grad_norm": 0.33260634541511536,
+      "learning_rate": 5.757666109839702e-05,
+      "loss": 0.0195,
+      "step": 4790
+    },
+    {
+      "epoch": 3.041825095057034,
+      "grad_norm": 0.29105865955352783,
+      "learning_rate": 5.74131823855921e-05,
+      "loss": 0.015,
+      "step": 4800
+    },
+    {
+      "epoch": 3.0481622306717364,
+      "grad_norm": 0.20448502898216248,
+      "learning_rate": 5.72496226034123e-05,
+      "loss": 0.0141,
+      "step": 4810
+    },
+    {
+      "epoch": 3.0544993662864384,
+      "grad_norm": 0.2469986230134964,
+      "learning_rate": 5.7085983540521216e-05,
+      "loss": 0.0117,
+      "step": 4820
+    },
+    {
+      "epoch": 3.0608365019011408,
+      "grad_norm": 0.2351512908935547,
+      "learning_rate": 5.692226698644938e-05,
+      "loss": 0.0138,
+      "step": 4830
+    },
+    {
+      "epoch": 3.0671736375158427,
+      "grad_norm": 0.2490169107913971,
+      "learning_rate": 5.675847473157485e-05,
+      "loss": 0.0144,
+      "step": 4840
+    },
+    {
+      "epoch": 3.073510773130545,
+      "grad_norm": 0.3461778461933136,
+      "learning_rate": 5.6594608567103456e-05,
+      "loss": 0.0126,
+      "step": 4850
+    },
+    {
+      "epoch": 3.079847908745247,
+      "grad_norm": 0.28383558988571167,
+      "learning_rate": 5.6430670285049314e-05,
+      "loss": 0.0117,
+      "step": 4860
+    },
+    {
+      "epoch": 3.0861850443599494,
+      "grad_norm": 0.25864022970199585,
+      "learning_rate": 5.6266661678215216e-05,
+      "loss": 0.0146,
+      "step": 4870
+    },
+    {
+      "epoch": 3.0925221799746514,
+      "grad_norm": 0.4212886095046997,
+      "learning_rate": 5.6102584540173006e-05,
+      "loss": 0.0135,
+      "step": 4880
+    },
+    {
+      "epoch": 3.098859315589354,
+      "grad_norm": 0.2258593887090683,
+      "learning_rate": 5.5938440665244006e-05,
+      "loss": 0.0164,
+      "step": 4890
+    },
+    {
+      "epoch": 3.1051964512040557,
+      "grad_norm": 0.3095914423465729,
+      "learning_rate": 5.577423184847932e-05,
+      "loss": 0.0134,
+      "step": 4900
+    },
+    {
+      "epoch": 3.111533586818758,
+      "grad_norm": 0.14614951610565186,
+      "learning_rate": 5.560995988564023e-05,
+      "loss": 0.0131,
+      "step": 4910
+    },
+    {
+      "epoch": 3.11787072243346,
+      "grad_norm": 0.30523428320884705,
+      "learning_rate": 5.544562657317863e-05,
+      "loss": 0.015,
+      "step": 4920
+    },
+    {
+      "epoch": 3.124207858048162,
+      "grad_norm": 0.17284521460533142,
+      "learning_rate": 5.52812337082173e-05,
+      "loss": 0.0126,
+      "step": 4930
+    },
+    {
+      "epoch": 3.1305449936628644,
+      "grad_norm": 0.33044707775115967,
+      "learning_rate": 5.511678308853026e-05,
+      "loss": 0.0142,
+      "step": 4940
+    },
+    {
+      "epoch": 3.1368821292775664,
+      "grad_norm": 0.313587486743927,
+      "learning_rate": 5.495227651252315e-05,
+      "loss": 0.011,
+      "step": 4950
+    },
+    {
+      "epoch": 3.1432192648922688,
+      "grad_norm": 0.3139721751213074,
+      "learning_rate": 5.478771577921351e-05,
+      "loss": 0.0132,
+      "step": 4960
+    },
+    {
+      "epoch": 3.1495564005069707,
+      "grad_norm": 0.313909649848938,
+      "learning_rate": 5.462310268821118e-05,
+      "loss": 0.0129,
+      "step": 4970
+    },
+    {
+      "epoch": 3.155893536121673,
+      "grad_norm": 0.17512694001197815,
+      "learning_rate": 5.445843903969854e-05,
+      "loss": 0.0102,
+      "step": 4980
+    },
+    {
+      "epoch": 3.162230671736375,
+      "grad_norm": 0.21401318907737732,
+      "learning_rate": 5.4293726634410855e-05,
+      "loss": 0.012,
+      "step": 4990
+    },
+    {
+      "epoch": 3.1685678073510775,
+      "grad_norm": 0.15441134572029114,
+      "learning_rate": 5.4128967273616625e-05,
+      "loss": 0.0119,
+      "step": 5000
+    },
+    {
+      "epoch": 3.1749049429657794,
+      "grad_norm": 0.23418472707271576,
+      "learning_rate": 5.396416275909779e-05,
+      "loss": 0.015,
+      "step": 5010
+    },
+    {
+      "epoch": 3.181242078580482,
+      "grad_norm": 0.2487867772579193,
+      "learning_rate": 5.379931489313016e-05,
+      "loss": 0.0133,
+      "step": 5020
+    },
+    {
+      "epoch": 3.1875792141951838,
+      "grad_norm": 0.2844524681568146,
+      "learning_rate": 5.363442547846356e-05,
+      "loss": 0.014,
+      "step": 5030
+    },
+    {
+      "epoch": 3.1939163498098857,
+      "grad_norm": 0.20346881449222565,
+      "learning_rate": 5.3469496318302204e-05,
+      "loss": 0.0114,
+      "step": 5040
+    },
+    {
+      "epoch": 3.200253485424588,
+      "grad_norm": 0.3049124479293823,
+      "learning_rate": 5.330452921628497e-05,
+      "loss": 0.0153,
+      "step": 5050
+    },
+    {
+      "epoch": 3.20659062103929,
+      "grad_norm": 0.29535266757011414,
+      "learning_rate": 5.313952597646568e-05,
+      "loss": 0.0132,
+      "step": 5060
+    },
+    {
+      "epoch": 3.2129277566539924,
+      "grad_norm": 0.3959789276123047,
+      "learning_rate": 5.297448840329329e-05,
+      "loss": 0.0157,
+      "step": 5070
+    },
+    {
+      "epoch": 3.2192648922686944,
+      "grad_norm": 0.23339158296585083,
+      "learning_rate": 5.280941830159227e-05,
+      "loss": 0.0128,
+      "step": 5080
+    },
+    {
+      "epoch": 3.225602027883397,
+      "grad_norm": 0.2398107349872589,
+      "learning_rate": 5.264431747654284e-05,
+      "loss": 0.0137,
+      "step": 5090
+    },
+    {
+      "epoch": 3.2319391634980987,
+      "grad_norm": 0.24322153627872467,
+      "learning_rate": 5.247918773366112e-05,
+      "loss": 0.0184,
+      "step": 5100
+    },
+    {
+      "epoch": 3.238276299112801,
+      "grad_norm": 0.15239882469177246,
+      "learning_rate": 5.231403087877955e-05,
+      "loss": 0.0161,
+      "step": 5110
+    },
+    {
+      "epoch": 3.244613434727503,
+      "grad_norm": 0.31975507736206055,
+      "learning_rate": 5.214884871802703e-05,
+      "loss": 0.0147,
+      "step": 5120
+    },
+    {
+      "epoch": 3.2509505703422055,
+      "grad_norm": 0.22015072405338287,
+      "learning_rate": 5.198364305780922e-05,
+      "loss": 0.0105,
+      "step": 5130
+    },
+    {
+      "epoch": 3.2572877059569074,
+      "grad_norm": 0.34388023614883423,
+      "learning_rate": 5.1818415704788725e-05,
+      "loss": 0.0156,
+      "step": 5140
+    },
+    {
+      "epoch": 3.26362484157161,
+      "grad_norm": 0.3033609092235565,
+      "learning_rate": 5.165316846586541e-05,
+      "loss": 0.0148,
+      "step": 5150
+    },
+    {
+      "epoch": 3.2699619771863118,
+      "grad_norm": 0.1909927874803543,
+      "learning_rate": 5.148790314815663e-05,
+      "loss": 0.0123,
+      "step": 5160
+    },
+    {
+      "epoch": 3.2762991128010137,
+      "grad_norm": 0.23113128542900085,
+      "learning_rate": 5.132262155897739e-05,
+      "loss": 0.0164,
+      "step": 5170
+    },
+    {
+      "epoch": 3.282636248415716,
+      "grad_norm": 0.19848641753196716,
+      "learning_rate": 5.1157325505820694e-05,
+      "loss": 0.0173,
+      "step": 5180
+    },
+    {
+      "epoch": 3.288973384030418,
+      "grad_norm": 0.27411961555480957,
+      "learning_rate": 5.0992016796337686e-05,
+      "loss": 0.0105,
+      "step": 5190
+    },
+    {
+      "epoch": 3.2953105196451205,
+      "grad_norm": 0.19896908104419708,
+      "learning_rate": 5.0826697238317935e-05,
+      "loss": 0.0153,
+      "step": 5200
+    },
+    {
+      "epoch": 3.3016476552598224,
+      "grad_norm": 0.20173603296279907,
+      "learning_rate": 5.066136863966963e-05,
+      "loss": 0.0121,
+      "step": 5210
+    },
+    {
+      "epoch": 3.307984790874525,
+      "grad_norm": 0.1532195508480072,
+      "learning_rate": 5.0496032808399815e-05,
+      "loss": 0.0116,
+      "step": 5220
+    },
+    {
+      "epoch": 3.3143219264892267,
+      "grad_norm": 0.14388404786586761,
+      "learning_rate": 5.033069155259471e-05,
+      "loss": 0.0134,
+      "step": 5230
+    },
+    {
+      "epoch": 3.320659062103929,
+      "grad_norm": 0.2755822241306305,
+      "learning_rate": 5.016534668039976e-05,
+      "loss": 0.0103,
+      "step": 5240
+    },
+    {
+      "epoch": 3.326996197718631,
+      "grad_norm": 0.2270147204399109,
+      "learning_rate": 5e-05,
+      "loss": 0.0093,
+      "step": 5250
+    },
+    {
+      "epoch": 3.3333333333333335,
+      "grad_norm": 0.17542065680027008,
+      "learning_rate": 4.9834653319600246e-05,
+      "loss": 0.0119,
+      "step": 5260
+    },
+    {
+      "epoch": 3.3396704689480354,
+      "grad_norm": 0.26830822229385376,
+      "learning_rate": 4.96693084474053e-05,
+      "loss": 0.0139,
+      "step": 5270
+    },
+    {
+      "epoch": 3.346007604562738,
+      "grad_norm": 0.23319189250469208,
+      "learning_rate": 4.950396719160018e-05,
+      "loss": 0.0115,
+      "step": 5280
+    },
+    {
+      "epoch": 3.3523447401774398,
+      "grad_norm": 0.28234973549842834,
+      "learning_rate": 4.93386313603304e-05,
+      "loss": 0.0096,
+      "step": 5290
+    },
+    {
+      "epoch": 3.3586818757921417,
+      "grad_norm": 0.2358221560716629,
+      "learning_rate": 4.917330276168208e-05,
+      "loss": 0.0126,
+      "step": 5300
+    },
+    {
+      "epoch": 3.365019011406844,
+      "grad_norm": 0.29058966040611267,
+      "learning_rate": 4.9007983203662326e-05,
+      "loss": 0.0281,
+      "step": 5310
+    },
+    {
+      "epoch": 3.371356147021546,
+      "grad_norm": 0.23800086975097656,
+      "learning_rate": 4.884267449417931e-05,
+      "loss": 0.0177,
+      "step": 5320
+    },
+    {
+      "epoch": 3.3776932826362485,
+      "grad_norm": 0.26259422302246094,
+      "learning_rate": 4.867737844102261e-05,
+      "loss": 0.014,
+      "step": 5330
+    },
+    {
+      "epoch": 3.3840304182509504,
+      "grad_norm": 0.2746807336807251,
+      "learning_rate": 4.851209685184338e-05,
+      "loss": 0.0102,
+      "step": 5340
+    },
+    {
+      "epoch": 3.390367553865653,
+      "grad_norm": 0.46735528111457825,
+      "learning_rate": 4.834683153413459e-05,
+      "loss": 0.0207,
+      "step": 5350
+    },
+    {
+      "epoch": 3.3967046894803548,
+      "grad_norm": 0.2719608247280121,
+      "learning_rate": 4.818158429521129e-05,
+      "loss": 0.0129,
+      "step": 5360
+    },
+    {
+      "epoch": 3.403041825095057,
+      "grad_norm": 0.28547102212905884,
+      "learning_rate": 4.801635694219079e-05,
+      "loss": 0.0101,
+      "step": 5370
+    },
+    {
+      "epoch": 3.409378960709759,
+      "grad_norm": 0.37190404534339905,
+      "learning_rate": 4.785115128197298e-05,
+      "loss": 0.0156,
+      "step": 5380
+    },
+    {
+      "epoch": 3.4157160963244615,
+      "grad_norm": 0.3981722295284271,
+      "learning_rate": 4.7685969121220456e-05,
+      "loss": 0.0119,
+      "step": 5390
+    },
+    {
+      "epoch": 3.4220532319391634,
+      "grad_norm": 0.325492262840271,
+      "learning_rate": 4.7520812266338885e-05,
+      "loss": 0.0135,
+      "step": 5400
+    },
+    {
+      "epoch": 3.428390367553866,
+      "grad_norm": 0.24744857847690582,
+      "learning_rate": 4.735568252345718e-05,
+      "loss": 0.0152,
+      "step": 5410
+    },
+    {
+      "epoch": 3.434727503168568,
+      "grad_norm": 0.15038323402404785,
+      "learning_rate": 4.7190581698407725e-05,
+      "loss": 0.0161,
+      "step": 5420
+    },
+    {
+      "epoch": 3.4410646387832697,
+      "grad_norm": 0.2310582995414734,
+      "learning_rate": 4.702551159670672e-05,
+      "loss": 0.0121,
+      "step": 5430
+    },
+    {
+      "epoch": 3.447401774397972,
+      "grad_norm": 0.17245414853096008,
+      "learning_rate": 4.6860474023534335e-05,
+      "loss": 0.0127,
+      "step": 5440
+    },
+    {
+      "epoch": 3.453738910012674,
+      "grad_norm": 0.2583564817905426,
+      "learning_rate": 4.669547078371504e-05,
+      "loss": 0.0171,
+      "step": 5450
+    },
+    {
+      "epoch": 3.4600760456273765,
+      "grad_norm": 0.3162192702293396,
+      "learning_rate": 4.65305036816978e-05,
+      "loss": 0.0113,
+      "step": 5460
+    },
+    {
+      "epoch": 3.4664131812420784,
+      "grad_norm": 0.29524022340774536,
+      "learning_rate": 4.6365574521536445e-05,
+      "loss": 0.0126,
+      "step": 5470
+    },
+    {
+      "epoch": 3.472750316856781,
+      "grad_norm": 0.14694812893867493,
+      "learning_rate": 4.620068510686985e-05,
+      "loss": 0.0119,
+      "step": 5480
+    },
+    {
+      "epoch": 3.4790874524714828,
+      "grad_norm": 0.1337551772594452,
+      "learning_rate": 4.60358372409022e-05,
+      "loss": 0.0088,
+      "step": 5490
+    },
+    {
+      "epoch": 3.485424588086185,
+      "grad_norm": 0.3237602114677429,
+      "learning_rate": 4.5871032726383386e-05,
+      "loss": 0.0091,
+      "step": 5500
+    },
+    {
+      "epoch": 3.491761723700887,
+      "grad_norm": 0.1705574095249176,
+      "learning_rate": 4.570627336558915e-05,
+      "loss": 0.0097,
+      "step": 5510
+    },
+    {
+      "epoch": 3.4980988593155895,
+      "grad_norm": 0.2950294613838196,
+      "learning_rate": 4.554156096030149e-05,
+      "loss": 0.0117,
+      "step": 5520
+    },
+    {
+      "epoch": 3.5044359949302915,
+      "grad_norm": 0.27918121218681335,
+      "learning_rate": 4.537689731178883e-05,
+      "loss": 0.0104,
+      "step": 5530
+    },
+    {
+      "epoch": 3.510773130544994,
+      "grad_norm": 0.2752488851547241,
+      "learning_rate": 4.5212284220786494e-05,
+      "loss": 0.009,
+      "step": 5540
+    },
+    {
+      "epoch": 3.517110266159696,
+      "grad_norm": 0.24830137193202972,
+      "learning_rate": 4.504772348747687e-05,
+      "loss": 0.0133,
+      "step": 5550
+    },
+    {
+      "epoch": 3.5234474017743977,
+      "grad_norm": 0.38127654790878296,
+      "learning_rate": 4.488321691146975e-05,
+      "loss": 0.0142,
+      "step": 5560
+    },
+    {
+      "epoch": 3.5297845373891,
+      "grad_norm": 0.3115534782409668,
+      "learning_rate": 4.471876629178273e-05,
+      "loss": 0.0158,
+      "step": 5570
+    },
+    {
+      "epoch": 3.5361216730038025,
+      "grad_norm": 0.18441027402877808,
+      "learning_rate": 4.4554373426821374e-05,
+      "loss": 0.0087,
+      "step": 5580
+    },
+    {
+      "epoch": 3.5424588086185045,
+      "grad_norm": 0.19032280147075653,
+      "learning_rate": 4.439004011435979e-05,
+      "loss": 0.0128,
+      "step": 5590
+    },
+    {
+      "epoch": 3.5487959442332064,
+      "grad_norm": 0.16672304272651672,
+      "learning_rate": 4.4225768151520694e-05,
+      "loss": 0.0182,
+      "step": 5600
+    },
+    {
+      "epoch": 3.555133079847909,
+      "grad_norm": 0.24737948179244995,
+      "learning_rate": 4.406155933475599e-05,
+      "loss": 0.0112,
+      "step": 5610
+    },
+    {
+      "epoch": 3.5614702154626108,
+      "grad_norm": 0.17746677994728088,
+      "learning_rate": 4.3897415459827e-05,
+      "loss": 0.0151,
+      "step": 5620
+    },
+    {
+      "epoch": 3.567807351077313,
+      "grad_norm": 0.2580298185348511,
+      "learning_rate": 4.373333832178478e-05,
+      "loss": 0.0092,
+      "step": 5630
+    },
+    {
+      "epoch": 3.574144486692015,
+      "grad_norm": 0.24834871292114258,
+      "learning_rate": 4.3569329714950704e-05,
+      "loss": 0.0133,
+      "step": 5640
+    },
+    {
+      "epoch": 3.5804816223067175,
+      "grad_norm": 0.23635822534561157,
+      "learning_rate": 4.3405391432896555e-05,
+      "loss": 0.0132,
+      "step": 5650
+    },
+    {
+      "epoch": 3.5868187579214195,
+      "grad_norm": 0.2585609555244446,
+      "learning_rate": 4.324152526842517e-05,
+      "loss": 0.014,
+      "step": 5660
+    },
+    {
+      "epoch": 3.593155893536122,
+      "grad_norm": 0.2859804034233093,
+      "learning_rate": 4.307773301355062e-05,
+      "loss": 0.0101,
+      "step": 5670
+    },
+    {
+      "epoch": 3.599493029150824,
+      "grad_norm": 0.18319383263587952,
+      "learning_rate": 4.291401645947879e-05,
+      "loss": 0.0202,
+      "step": 5680
+    },
+    {
+      "epoch": 3.6058301647655258,
+      "grad_norm": 0.2706187665462494,
+      "learning_rate": 4.275037739658771e-05,
+      "loss": 0.0067,
+      "step": 5690
+    },
+    {
+      "epoch": 3.612167300380228,
+      "grad_norm": 0.2773619592189789,
+      "learning_rate": 4.2586817614407895e-05,
+      "loss": 0.0114,
+      "step": 5700
+    },
+    {
+      "epoch": 3.6185044359949305,
+      "grad_norm": 0.24154767394065857,
+      "learning_rate": 4.2423338901602985e-05,
+      "loss": 0.0178,
+      "step": 5710
+    },
+    {
+      "epoch": 3.6248415716096325,
+      "grad_norm": 0.1268390715122223,
+      "learning_rate": 4.2259943045949934e-05,
+      "loss": 0.0098,
+      "step": 5720
+    },
+    {
+      "epoch": 3.6311787072243344,
+      "grad_norm": 0.14447900652885437,
+      "learning_rate": 4.209663183431969e-05,
+      "loss": 0.0121,
+      "step": 5730
+    },
+    {
+      "epoch": 3.637515842839037,
+      "grad_norm": 0.18617920577526093,
+      "learning_rate": 4.1933407052657456e-05,
+      "loss": 0.0092,
+      "step": 5740
+    },
+    {
+      "epoch": 3.643852978453739,
+      "grad_norm": 0.3276727497577667,
+      "learning_rate": 4.17702704859633e-05,
+      "loss": 0.0109,
+      "step": 5750
+    },
+    {
+      "epoch": 3.650190114068441,
+      "grad_norm": 0.20072652399539948,
+      "learning_rate": 4.160722391827262e-05,
+      "loss": 0.0106,
+      "step": 5760
+    },
+    {
+      "epoch": 3.656527249683143,
+      "grad_norm": 0.2064548283815384,
+      "learning_rate": 4.14442691326365e-05,
+      "loss": 0.0097,
+      "step": 5770
+    },
+    {
+      "epoch": 3.6628643852978455,
+      "grad_norm": 0.20051850378513336,
+      "learning_rate": 4.1281407911102425e-05,
+      "loss": 0.0127,
+      "step": 5780
+    },
+    {
+      "epoch": 3.6692015209125475,
+      "grad_norm": 0.27055758237838745,
+      "learning_rate": 4.111864203469457e-05,
+      "loss": 0.0125,
+      "step": 5790
+    },
+    {
+      "epoch": 3.67553865652725,
+      "grad_norm": 0.3249766230583191,
+      "learning_rate": 4.095597328339452e-05,
+      "loss": 0.013,
+      "step": 5800
+    },
+    {
+      "epoch": 3.681875792141952,
+      "grad_norm": 0.36412471532821655,
+      "learning_rate": 4.079340343612165e-05,
+      "loss": 0.0135,
+      "step": 5810
+    },
+    {
+      "epoch": 3.6882129277566538,
+      "grad_norm": 0.43938231468200684,
+      "learning_rate": 4.063093427071376e-05,
+      "loss": 0.0142,
+      "step": 5820
+    },
+    {
+      "epoch": 3.694550063371356,
+      "grad_norm": 0.2211577147245407,
+      "learning_rate": 4.046856756390767e-05,
+      "loss": 0.0092,
+      "step": 5830
+    },
+    {
+      "epoch": 3.7008871989860586,
+      "grad_norm": 0.2317885160446167,
+      "learning_rate": 4.0306305091319595e-05,
+      "loss": 0.0103,
+      "step": 5840
+    },
+    {
+      "epoch": 3.7072243346007605,
+      "grad_norm": 0.30124250054359436,
+      "learning_rate": 4.0144148627425993e-05,
+      "loss": 0.0128,
+      "step": 5850
+    },
+    {
+      "epoch": 3.7135614702154625,
+      "grad_norm": 0.22802747786045074,
+      "learning_rate": 3.9982099945543945e-05,
+      "loss": 0.0092,
+      "step": 5860
+    },
+    {
+      "epoch": 3.719898605830165,
+      "grad_norm": 0.3372552990913391,
+      "learning_rate": 3.982016081781189e-05,
+      "loss": 0.009,
+      "step": 5870
+    },
+    {
+      "epoch": 3.726235741444867,
+      "grad_norm": 0.3483387529850006,
+      "learning_rate": 3.965833301517017e-05,
+      "loss": 0.015,
+      "step": 5880
+    },
+    {
+      "epoch": 3.732572877059569,
+      "grad_norm": 0.17584814131259918,
+      "learning_rate": 3.949661830734172e-05,
+      "loss": 0.0077,
+      "step": 5890
+    },
+    {
+      "epoch": 3.738910012674271,
+      "grad_norm": 0.26330336928367615,
+      "learning_rate": 3.933501846281267e-05,
+      "loss": 0.0101,
+      "step": 5900
+    },
+    {
+      "epoch": 3.7452471482889735,
+      "grad_norm": 0.19230346381664276,
+      "learning_rate": 3.917353524881302e-05,
+      "loss": 0.014,
+      "step": 5910
+    },
+    {
+      "epoch": 3.7515842839036755,
+      "grad_norm": 0.19629403948783875,
+      "learning_rate": 3.901217043129735e-05,
+      "loss": 0.0124,
+      "step": 5920
+    },
+    {
+      "epoch": 3.757921419518378,
+      "grad_norm": 0.22321918606758118,
+      "learning_rate": 3.8850925774925425e-05,
+      "loss": 0.0116,
+      "step": 5930
+    },
+    {
+      "epoch": 3.76425855513308,
+      "grad_norm": 0.20382849872112274,
+      "learning_rate": 3.8689803043043e-05,
+      "loss": 0.0142,
+      "step": 5940
+    },
+    {
+      "epoch": 3.770595690747782,
+      "grad_norm": 0.16952167451381683,
+      "learning_rate": 3.852880399766243e-05,
+      "loss": 0.0125,
+      "step": 5950
+    },
+    {
+      "epoch": 3.776932826362484,
+      "grad_norm": 0.23627594113349915,
+      "learning_rate": 3.836793039944349e-05,
+      "loss": 0.0132,
+      "step": 5960
+    },
+    {
+      "epoch": 3.7832699619771866,
+      "grad_norm": 0.20818836987018585,
+      "learning_rate": 3.820718400767409e-05,
+      "loss": 0.0121,
+      "step": 5970
+    },
+    {
+      "epoch": 3.7896070975918885,
+      "grad_norm": 0.37023457884788513,
+      "learning_rate": 3.8046566580251e-05,
+      "loss": 0.0093,
+      "step": 5980
+    },
+    {
+      "epoch": 3.7959442332065905,
+      "grad_norm": 0.24624159932136536,
+      "learning_rate": 3.788607987366069e-05,
+      "loss": 0.0105,
+      "step": 5990
+    },
+    {
+      "epoch": 3.802281368821293,
+      "grad_norm": 0.2265588641166687,
+      "learning_rate": 3.772572564296005e-05,
+      "loss": 0.0148,
+      "step": 6000
+    },
+    {
+      "epoch": 3.808618504435995,
+      "grad_norm": 0.24321384727954865,
+      "learning_rate": 3.756550564175727e-05,
+      "loss": 0.014,
+      "step": 6010
+    },
+    {
+      "epoch": 3.814955640050697,
+      "grad_norm": 0.15189670026302338,
+      "learning_rate": 3.74054216221926e-05,
+      "loss": 0.0112,
+      "step": 6020
+    },
+    {
+      "epoch": 3.821292775665399,
+      "grad_norm": 0.366676926612854,
+      "learning_rate": 3.7245475334919246e-05,
+      "loss": 0.0163,
+      "step": 6030
+    },
+    {
+      "epoch": 3.8276299112801015,
+      "grad_norm": 0.22733189165592194,
+      "learning_rate": 3.7085668529084184e-05,
+      "loss": 0.014,
+      "step": 6040
+    },
+    {
+      "epoch": 3.8339670468948035,
+      "grad_norm": 0.17254877090454102,
+      "learning_rate": 3.6926002952309016e-05,
+      "loss": 0.0112,
+      "step": 6050
+    },
+    {
+      "epoch": 3.840304182509506,
+      "grad_norm": 0.22515353560447693,
+      "learning_rate": 3.676648035067093e-05,
+      "loss": 0.0113,
+      "step": 6060
+    },
+    {
+      "epoch": 3.846641318124208,
+      "grad_norm": 0.239216610789299,
+      "learning_rate": 3.6607102468683526e-05,
+      "loss": 0.011,
+      "step": 6070
+    },
+    {
+      "epoch": 3.85297845373891,
+      "grad_norm": 0.11387116461992264,
+      "learning_rate": 3.6447871049277796e-05,
+      "loss": 0.0072,
+      "step": 6080
+    },
+    {
+      "epoch": 3.859315589353612,
+      "grad_norm": 0.18705078959465027,
+      "learning_rate": 3.628878783378302e-05,
+      "loss": 0.0118,
+      "step": 6090
+    },
+    {
+      "epoch": 3.8656527249683146,
+      "grad_norm": 0.1804051548242569,
+      "learning_rate": 3.612985456190778e-05,
+      "loss": 0.0134,
+      "step": 6100
+    },
+    {
+      "epoch": 3.8719898605830165,
+      "grad_norm": 0.4679585099220276,
+      "learning_rate": 3.597107297172084e-05,
+      "loss": 0.011,
+      "step": 6110
+    },
+    {
+      "epoch": 3.8783269961977185,
+      "grad_norm": 0.14044460654258728,
+      "learning_rate": 3.581244479963225e-05,
+      "loss": 0.0106,
+      "step": 6120
+    },
+    {
+      "epoch": 3.884664131812421,
+      "grad_norm": 0.332378089427948,
+      "learning_rate": 3.5653971780374295e-05,
+      "loss": 0.0166,
+      "step": 6130
+    },
+    {
+      "epoch": 3.891001267427123,
+      "grad_norm": 0.2639565169811249,
+      "learning_rate": 3.5495655646982505e-05,
+      "loss": 0.0139,
+      "step": 6140
+    },
+    {
+      "epoch": 3.897338403041825,
+      "grad_norm": 0.2334318906068802,
+      "learning_rate": 3.533749813077677e-05,
+      "loss": 0.0137,
+      "step": 6150
+    },
+    {
+      "epoch": 3.903675538656527,
+      "grad_norm": 0.3366606831550598,
+      "learning_rate": 3.517950096134232e-05,
+      "loss": 0.0121,
+      "step": 6160
+    },
+    {
+      "epoch": 3.9100126742712296,
+      "grad_norm": 0.32218649983406067,
+      "learning_rate": 3.5021665866510925e-05,
+      "loss": 0.0113,
+      "step": 6170
+    },
+    {
+      "epoch": 3.9163498098859315,
+      "grad_norm": 0.14572682976722717,
+      "learning_rate": 3.4863994572341843e-05,
+      "loss": 0.0147,
+      "step": 6180
+    },
+    {
+      "epoch": 3.922686945500634,
+      "grad_norm": 0.2537533938884735,
+      "learning_rate": 3.470648880310313e-05,
+      "loss": 0.011,
+      "step": 6190
+    },
+    {
+      "epoch": 3.929024081115336,
+      "grad_norm": 0.25570058822631836,
+      "learning_rate": 3.4549150281252636e-05,
+      "loss": 0.0107,
+      "step": 6200
+    },
+    {
+      "epoch": 3.935361216730038,
+      "grad_norm": 0.18532606959342957,
+      "learning_rate": 3.439198072741921e-05,
+      "loss": 0.0109,
+      "step": 6210
+    },
+    {
+      "epoch": 3.94169835234474,
+      "grad_norm": 0.1890391707420349,
+      "learning_rate": 3.423498186038393e-05,
+      "loss": 0.0168,
+      "step": 6220
+    },
+    {
+      "epoch": 3.9480354879594426,
+      "grad_norm": 0.17653165757656097,
+      "learning_rate": 3.407815539706124e-05,
+      "loss": 0.01,
+      "step": 6230
+    },
+    {
+      "epoch": 3.9543726235741445,
+      "grad_norm": 0.127501979470253,
+      "learning_rate": 3.392150305248024e-05,
+      "loss": 0.0122,
+      "step": 6240
+    },
+    {
+      "epoch": 3.9607097591888465,
+      "grad_norm": 0.16189290583133698,
+      "learning_rate": 3.3765026539765834e-05,
+      "loss": 0.0114,
+      "step": 6250
+    },
+    {
+      "epoch": 3.967046894803549,
+      "grad_norm": 0.17998936772346497,
+      "learning_rate": 3.360872757012011e-05,
+      "loss": 0.0148,
+      "step": 6260
+    },
+    {
+      "epoch": 3.973384030418251,
+      "grad_norm": 0.17054511606693268,
+      "learning_rate": 3.3452607852803584e-05,
+      "loss": 0.0112,
+      "step": 6270
+    },
+    {
+      "epoch": 3.9797211660329532,
+      "grad_norm": 0.2786466181278229,
+      "learning_rate": 3.329666909511645e-05,
+      "loss": 0.0123,
+      "step": 6280
+    },
+    {
+      "epoch": 3.986058301647655,
+      "grad_norm": 0.21183429658412933,
+      "learning_rate": 3.3140913002379995e-05,
+      "loss": 0.0099,
+      "step": 6290
+    },
+    {
+      "epoch": 3.9923954372623576,
+      "grad_norm": 0.2610870897769928,
+      "learning_rate": 3.298534127791785e-05,
+      "loss": 0.0136,
+      "step": 6300
+    },
+    {
+      "epoch": 3.9987325728770595,
+      "grad_norm": 0.22637638449668884,
+      "learning_rate": 3.282995562303754e-05,
+      "loss": 0.011,
+      "step": 6310
+    },
+    {
+      "epoch": 4.005069708491762,
+      "grad_norm": 0.18758799135684967,
+      "learning_rate": 3.267475773701161e-05,
+      "loss": 0.009,
+      "step": 6320
+    },
+    {
+      "epoch": 4.011406844106464,
+      "grad_norm": 0.22371801733970642,
+      "learning_rate": 3.251974931705933e-05,
+      "loss": 0.0107,
+      "step": 6330
+    },
+    {
+      "epoch": 4.017743979721166,
+      "grad_norm": 0.22646395862102509,
+      "learning_rate": 3.236493205832795e-05,
+      "loss": 0.0117,
+      "step": 6340
+    },
+    {
+      "epoch": 4.024081115335868,
+      "grad_norm": 0.19698502123355865,
+      "learning_rate": 3.221030765387417e-05,
+      "loss": 0.0119,
+      "step": 6350
+    },
+    {
+      "epoch": 4.030418250950571,
+      "grad_norm": 0.22287245094776154,
+      "learning_rate": 3.205587779464576e-05,
+      "loss": 0.0093,
+      "step": 6360
+    },
+    {
+      "epoch": 4.0367553865652726,
+      "grad_norm": 0.27699437737464905,
+      "learning_rate": 3.190164416946285e-05,
+      "loss": 0.0093,
+      "step": 6370
+    },
+    {
+      "epoch": 4.0430925221799745,
+      "grad_norm": 0.1392444372177124,
+      "learning_rate": 3.1747608464999725e-05,
+      "loss": 0.0072,
+      "step": 6380
+    },
+    {
+      "epoch": 4.0494296577946765,
+      "grad_norm": 0.2287701815366745,
+      "learning_rate": 3.1593772365766105e-05,
+      "loss": 0.0154,
+      "step": 6390
+    },
+    {
+      "epoch": 4.055766793409379,
+      "grad_norm": 0.24777956306934357,
+      "learning_rate": 3.144013755408895e-05,
+      "loss": 0.012,
+      "step": 6400
+    },
+    {
+      "epoch": 4.062103929024081,
+      "grad_norm": 0.1426210254430771,
+      "learning_rate": 3.128670571009399e-05,
+      "loss": 0.012,
+      "step": 6410
+    },
+    {
+      "epoch": 4.068441064638783,
+      "grad_norm": 0.11875156313180923,
+      "learning_rate": 3.113347851168721e-05,
+      "loss": 0.0088,
+      "step": 6420
+    },
+    {
+      "epoch": 4.074778200253485,
+      "grad_norm": 0.1875237077474594,
+      "learning_rate": 3.098045763453678e-05,
+      "loss": 0.0074,
+      "step": 6430
+    },
+    {
+      "epoch": 4.081115335868188,
+      "grad_norm": 0.09495358169078827,
+      "learning_rate": 3.082764475205442e-05,
+      "loss": 0.0095,
+      "step": 6440
+    },
+    {
+      "epoch": 4.08745247148289,
+      "grad_norm": 0.2620109021663666,
+      "learning_rate": 3.0675041535377405e-05,
+      "loss": 0.014,
+      "step": 6450
+    },
+    {
+      "epoch": 4.093789607097592,
+      "grad_norm": 0.30358102917671204,
+      "learning_rate": 3.052264965335e-05,
+      "loss": 0.0068,
+      "step": 6460
+    },
+    {
+      "epoch": 4.100126742712294,
+      "grad_norm": 0.2840622663497925,
+      "learning_rate": 3.0370470772505433e-05,
+      "loss": 0.008,
+      "step": 6470
+    },
+    {
+      "epoch": 4.106463878326996,
+      "grad_norm": 0.12551774084568024,
+      "learning_rate": 3.0218506557047598e-05,
+      "loss": 0.0069,
+      "step": 6480
+    },
+    {
+      "epoch": 4.112801013941699,
+      "grad_norm": 0.2494884580373764,
+      "learning_rate": 3.006675866883275e-05,
+      "loss": 0.0076,
+      "step": 6490
+    },
+    {
+      "epoch": 4.119138149556401,
+      "grad_norm": 0.11765393614768982,
+      "learning_rate": 2.991522876735154e-05,
+      "loss": 0.0065,
+      "step": 6500
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 10000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 7,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.278893812003424e+17,
+  "train_batch_size": 16,
+  "trial_name": null,
+  "trial_params": null
+}