diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5498 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 3898,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.7940189116741517,
+      "learning_rate": 5.128205128205128e-07,
+      "loss": 1.2901,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.8549670354423652,
+      "learning_rate": 2.564102564102564e-06,
+      "loss": 1.1983,
+      "step": 5
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.5576375802504129,
+      "learning_rate": 5.128205128205128e-06,
+      "loss": 1.0245,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.659060642425238,
+      "learning_rate": 7.692307692307694e-06,
+      "loss": 1.059,
+      "step": 15
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.5974849589264373,
+      "learning_rate": 1.0256410256410256e-05,
+      "loss": 1.1426,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.8598589383620435,
+      "learning_rate": 1.282051282051282e-05,
+      "loss": 1.0725,
+      "step": 25
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.7508232225220164,
+      "learning_rate": 1.5384615384615387e-05,
+      "loss": 1.1266,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.7318286527682681,
+      "learning_rate": 1.794871794871795e-05,
+      "loss": 0.8731,
+      "step": 35
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.4400452199237334,
+      "learning_rate": 2.0512820512820512e-05,
+      "loss": 0.9001,
+      "step": 40
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.3581907713644131,
+      "learning_rate": 2.307692307692308e-05,
+      "loss": 0.8492,
+      "step": 45
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.4752977404545827,
+      "learning_rate": 2.564102564102564e-05,
+      "loss": 0.8143,
+      "step": 50
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.47540987111824407,
+      "learning_rate": 2.8205128205128207e-05,
+      "loss": 0.7851,
+      "step": 55
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.3486905588773928,
+      "learning_rate": 3.0769230769230774e-05,
+      "loss": 0.7796,
+      "step": 60
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.4379263022871022,
+      "learning_rate": 3.3333333333333335e-05,
+      "loss": 0.7896,
+      "step": 65
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.46208045268944636,
+      "learning_rate": 3.58974358974359e-05,
+      "loss": 0.8126,
+      "step": 70
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.38482216190878704,
+      "learning_rate": 3.846153846153846e-05,
+      "loss": 0.7794,
+      "step": 75
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.36177681119574645,
+      "learning_rate": 4.1025641025641023e-05,
+      "loss": 0.7081,
+      "step": 80
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.4222837002069195,
+      "learning_rate": 4.358974358974359e-05,
+      "loss": 0.7807,
+      "step": 85
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.37505732650970885,
+      "learning_rate": 4.615384615384616e-05,
+      "loss": 0.6835,
+      "step": 90
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.5376404476864999,
+      "learning_rate": 4.871794871794872e-05,
+      "loss": 0.6775,
+      "step": 95
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.33127308677570055,
+      "learning_rate": 5.128205128205128e-05,
+      "loss": 0.7702,
+      "step": 100
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.30805670386128153,
+      "learning_rate": 5.384615384615385e-05,
+      "loss": 0.7311,
+      "step": 105
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.3064832637009184,
+      "learning_rate": 5.6410256410256414e-05,
+      "loss": 0.7068,
+      "step": 110
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.36300239603665035,
+      "learning_rate": 5.897435897435898e-05,
+      "loss": 0.8024,
+      "step": 115
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.546235320721914,
+      "learning_rate": 6.153846153846155e-05,
+      "loss": 0.8347,
+      "step": 120
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.4262756501942376,
+      "learning_rate": 6.410256410256412e-05,
+      "loss": 0.7733,
+      "step": 125
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.31870027810040585,
+      "learning_rate": 6.666666666666667e-05,
+      "loss": 0.7072,
+      "step": 130
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.5080881100002653,
+      "learning_rate": 6.923076923076924e-05,
+      "loss": 0.7419,
+      "step": 135
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.45131101725332007,
+      "learning_rate": 7.17948717948718e-05,
+      "loss": 0.6721,
+      "step": 140
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.4015662904289768,
+      "learning_rate": 7.435897435897436e-05,
+      "loss": 0.8274,
+      "step": 145
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.30635979357766263,
+      "learning_rate": 7.692307692307693e-05,
+      "loss": 0.7501,
+      "step": 150
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.2563663658930907,
+      "learning_rate": 7.948717948717948e-05,
+      "loss": 0.6621,
+      "step": 155
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.3851611498390434,
+      "learning_rate": 8.205128205128205e-05,
+      "loss": 0.6981,
+      "step": 160
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.32054712548206815,
+      "learning_rate": 8.461538461538461e-05,
+      "loss": 0.7648,
+      "step": 165
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.3604034886379123,
+      "learning_rate": 8.717948717948718e-05,
+      "loss": 0.721,
+      "step": 170
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.2978365211793088,
+      "learning_rate": 8.974358974358975e-05,
+      "loss": 0.7665,
+      "step": 175
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.28010588660358143,
+      "learning_rate": 9.230769230769232e-05,
+      "loss": 0.6919,
+      "step": 180
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.3041954902881447,
+      "learning_rate": 9.487179487179487e-05,
+      "loss": 0.6963,
+      "step": 185
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.3624614794713746,
+      "learning_rate": 9.743589743589744e-05,
+      "loss": 0.7866,
+      "step": 190
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.29529594264100556,
+      "learning_rate": 0.0001,
+      "loss": 0.7542,
+      "step": 195
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.29313493488432457,
+      "learning_rate": 0.00010256410256410256,
+      "loss": 0.8368,
+      "step": 200
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.3309957222386564,
+      "learning_rate": 0.00010512820512820514,
+      "loss": 0.6755,
+      "step": 205
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.39864413284171824,
+      "learning_rate": 0.0001076923076923077,
+      "loss": 0.7935,
+      "step": 210
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.3022145052274676,
+      "learning_rate": 0.00011025641025641027,
+      "loss": 0.77,
+      "step": 215
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.3188687614270182,
+      "learning_rate": 0.00011282051282051283,
+      "loss": 0.764,
+      "step": 220
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.2771960353584736,
+      "learning_rate": 0.00011538461538461538,
+      "loss": 0.7025,
+      "step": 225
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.34181074966932906,
+      "learning_rate": 0.00011794871794871796,
+      "loss": 0.7206,
+      "step": 230
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.28461361041254846,
+      "learning_rate": 0.00012051282051282052,
+      "loss": 0.752,
+      "step": 235
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.7008172335009641,
+      "learning_rate": 0.0001230769230769231,
+      "loss": 0.8087,
+      "step": 240
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.21894206670691171,
+      "learning_rate": 0.00012564102564102564,
+      "loss": 0.7153,
+      "step": 245
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.29593725228870665,
+      "learning_rate": 0.00012820512820512823,
+      "loss": 0.7168,
+      "step": 250
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.2519334320641686,
+      "learning_rate": 0.00013076923076923077,
+      "loss": 0.7517,
+      "step": 255
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.32763786349227375,
+      "learning_rate": 0.00013333333333333334,
+      "loss": 0.6877,
+      "step": 260
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.3792543190404934,
+      "learning_rate": 0.0001358974358974359,
+      "loss": 0.7357,
+      "step": 265
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.21121164349273067,
+      "learning_rate": 0.00013846153846153847,
+      "loss": 0.7251,
+      "step": 270
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.25702919707855326,
+      "learning_rate": 0.00014102564102564104,
+      "loss": 0.728,
+      "step": 275
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.2523953056455061,
+      "learning_rate": 0.0001435897435897436,
+      "loss": 0.6635,
+      "step": 280
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.29400188903129665,
+      "learning_rate": 0.00014615384615384615,
+      "loss": 0.6737,
+      "step": 285
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.28480340987225256,
+      "learning_rate": 0.00014871794871794872,
+      "loss": 0.6627,
+      "step": 290
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.6318376196782489,
+      "learning_rate": 0.00015128205128205128,
+      "loss": 0.7294,
+      "step": 295
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.30039963893256977,
+      "learning_rate": 0.00015384615384615385,
+      "loss": 0.8311,
+      "step": 300
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.28222145819043015,
+      "learning_rate": 0.00015641025641025642,
+      "loss": 0.7052,
+      "step": 305
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.23714903546035276,
+      "learning_rate": 0.00015897435897435896,
+      "loss": 0.6598,
+      "step": 310
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.24757276162013622,
+      "learning_rate": 0.00016153846153846155,
+      "loss": 0.7401,
+      "step": 315
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.25140380741663676,
+      "learning_rate": 0.0001641025641025641,
+      "loss": 0.6531,
+      "step": 320
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.30309172865038203,
+      "learning_rate": 0.0001666666666666667,
+      "loss": 0.6746,
+      "step": 325
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.27958455276568867,
+      "learning_rate": 0.00016923076923076923,
+      "loss": 0.7966,
+      "step": 330
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.23922247511746272,
+      "learning_rate": 0.0001717948717948718,
+      "loss": 0.8222,
+      "step": 335
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.2525786116177449,
+      "learning_rate": 0.00017435897435897436,
+      "loss": 0.7323,
+      "step": 340
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.2657739010257245,
+      "learning_rate": 0.00017692307692307693,
+      "loss": 0.7402,
+      "step": 345
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.28236890518445834,
+      "learning_rate": 0.0001794871794871795,
+      "loss": 0.8582,
+      "step": 350
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.33868871224317904,
+      "learning_rate": 0.00018205128205128207,
+      "loss": 0.7021,
+      "step": 355
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.3770994222569293,
+      "learning_rate": 0.00018461538461538463,
+      "loss": 0.6761,
+      "step": 360
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.4018353610014071,
+      "learning_rate": 0.0001871794871794872,
+      "loss": 0.7644,
+      "step": 365
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.29301791588126386,
+      "learning_rate": 0.00018974358974358974,
+      "loss": 0.7497,
+      "step": 370
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.2525840897139569,
+      "learning_rate": 0.00019230769230769233,
+      "loss": 0.6905,
+      "step": 375
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.2282690464906889,
+      "learning_rate": 0.00019487179487179487,
+      "loss": 0.628,
+      "step": 380
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.29122371361878807,
+      "learning_rate": 0.00019743589743589744,
+      "loss": 0.7005,
+      "step": 385
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.2641474899247467,
+      "learning_rate": 0.0002,
+      "loss": 0.8084,
+      "step": 390
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.21454542770751026,
+      "learning_rate": 0.00019999899748734544,
+      "loss": 0.7702,
+      "step": 395
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.22924120222036806,
+      "learning_rate": 0.00019999598996948235,
+      "loss": 0.6556,
+      "step": 400
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.32163392243239314,
+      "learning_rate": 0.00019999097750671223,
+      "loss": 0.731,
+      "step": 405
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.25259388958326623,
+      "learning_rate": 0.00019998396019953624,
+      "loss": 0.6915,
+      "step": 410
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.258567962978177,
+      "learning_rate": 0.00019997493818865318,
+      "loss": 0.7234,
+      "step": 415
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.2743302123872774,
+      "learning_rate": 0.0001999639116549566,
+      "loss": 0.5963,
+      "step": 420
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.26201706655388224,
+      "learning_rate": 0.00019995088081953136,
+      "loss": 0.7192,
+      "step": 425
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.33838871762466366,
+      "learning_rate": 0.00019993584594364894,
+      "loss": 0.6471,
+      "step": 430
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.24855520937031114,
+      "learning_rate": 0.00019991880732876246,
+      "loss": 0.7001,
+      "step": 435
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.23979280174747938,
+      "learning_rate": 0.0001998997653165004,
+      "loss": 0.7427,
+      "step": 440
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.2491200817907366,
+      "learning_rate": 0.00019987872028866003,
+      "loss": 0.8069,
+      "step": 445
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.2799825613310986,
+      "learning_rate": 0.00019985567266719934,
+      "loss": 0.7842,
+      "step": 450
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.2171357139053752,
+      "learning_rate": 0.00019983062291422908,
+      "loss": 0.7458,
+      "step": 455
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.2571958093636779,
+      "learning_rate": 0.00019980357153200315,
+      "loss": 0.7086,
+      "step": 460
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.32814189406723127,
+      "learning_rate": 0.00019977451906290854,
+      "loss": 0.7168,
+      "step": 465
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.232636717309027,
+      "learning_rate": 0.00019974346608945466,
+      "loss": 0.7444,
+      "step": 470
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.2564556626752832,
+      "learning_rate": 0.0001997104132342614,
+      "loss": 0.6684,
+      "step": 475
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.2328343453656683,
+      "learning_rate": 0.00019967536116004698,
+      "loss": 0.814,
+      "step": 480
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.2743961290431054,
+      "learning_rate": 0.00019963831056961433,
+      "loss": 0.6833,
+      "step": 485
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.26151241758992244,
+      "learning_rate": 0.00019959926220583713,
+      "loss": 0.7213,
+      "step": 490
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.23479176176270686,
+      "learning_rate": 0.000199558216851645,
+      "loss": 0.7367,
+      "step": 495
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.2620759960364289,
+      "learning_rate": 0.00019951517533000764,
+      "loss": 0.6078,
+      "step": 500
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.2558927478335512,
+      "learning_rate": 0.00019947013850391847,
+      "loss": 0.712,
+      "step": 505
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.25988357901337633,
+      "learning_rate": 0.00019942310727637724,
+      "loss": 0.7728,
+      "step": 510
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.2750603881771427,
+      "learning_rate": 0.000199374082590372,
+      "loss": 0.7573,
+      "step": 515
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.22168568020157275,
+      "learning_rate": 0.00019932306542886009,
+      "loss": 0.741,
+      "step": 520
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.2776057859080138,
+      "learning_rate": 0.0001992700568147485,
+      "loss": 0.7221,
+      "step": 525
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.309528370273256,
+      "learning_rate": 0.00019921505781087334,
+      "loss": 0.7362,
+      "step": 530
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.3228391807985929,
+      "learning_rate": 0.00019915806951997862,
+      "loss": 0.7958,
+      "step": 535
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.23972118913650053,
+      "learning_rate": 0.00019909909308469398,
+      "loss": 0.76,
+      "step": 540
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.24351868304057644,
+      "learning_rate": 0.0001990381296875118,
+      "loss": 0.6869,
+      "step": 545
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.35582198947398647,
+      "learning_rate": 0.0001989751805507637,
+      "loss": 0.814,
+      "step": 550
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.2711433452256402,
+      "learning_rate": 0.0001989102469365958,
+      "loss": 0.8162,
+      "step": 555
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.23300078896983795,
+      "learning_rate": 0.00019884333014694345,
+      "loss": 0.7118,
+      "step": 560
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.2640371108479178,
+      "learning_rate": 0.00019877443152350527,
+      "loss": 0.7364,
+      "step": 565
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.24281805118776953,
+      "learning_rate": 0.00019870355244771607,
+      "loss": 0.6726,
+      "step": 570
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.2937389542412289,
+      "learning_rate": 0.0001986306943407193,
+      "loss": 0.7418,
+      "step": 575
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.244991044922138,
+      "learning_rate": 0.00019855585866333835,
+      "loss": 0.7476,
+      "step": 580
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.2799677386092935,
+      "learning_rate": 0.00019847904691604757,
+      "loss": 0.751,
+      "step": 585
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.28203702245279344,
+      "learning_rate": 0.00019840026063894193,
+      "loss": 0.6952,
+      "step": 590
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.32320414101450645,
+      "learning_rate": 0.0001983195014117062,
+      "loss": 0.7425,
+      "step": 595
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.2991698548751533,
+      "learning_rate": 0.00019823677085358335,
+      "loss": 0.6539,
+      "step": 600
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.279534327063564,
+      "learning_rate": 0.00019815207062334197,
+      "loss": 0.6844,
+      "step": 605
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.23855364154241135,
+      "learning_rate": 0.00019806540241924317,
+      "loss": 0.7484,
+      "step": 610
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.2750191162103587,
+      "learning_rate": 0.00019797676797900633,
+      "loss": 0.7655,
+      "step": 615
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.2592922229570989,
+      "learning_rate": 0.00019788616907977441,
+      "loss": 0.7368,
+      "step": 620
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.21539765246350112,
+      "learning_rate": 0.0001977936075380783,
+      "loss": 0.6314,
+      "step": 625
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.24593391755509492,
+      "learning_rate": 0.00019769908520980034,
+      "loss": 0.7463,
+      "step": 630
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.277301291124461,
+      "learning_rate": 0.00019760260399013708,
+      "loss": 0.721,
+      "step": 635
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.2676695353061291,
+      "learning_rate": 0.00019750416581356146,
+      "loss": 0.7112,
+      "step": 640
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.24096399922364115,
+      "learning_rate": 0.0001974037726537838,
+      "loss": 0.6304,
+      "step": 645
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.2973476162334585,
+      "learning_rate": 0.00019730142652371236,
+      "loss": 0.6467,
+      "step": 650
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.2766324556390413,
+      "learning_rate": 0.00019719712947541295,
+      "loss": 0.6812,
+      "step": 655
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.18389408950636116,
+      "learning_rate": 0.0001970908836000678,
+      "loss": 0.6861,
+      "step": 660
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.2410351531016274,
+      "learning_rate": 0.00019698269102793358,
+      "loss": 0.6532,
+      "step": 665
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.30102039950416376,
+      "learning_rate": 0.00019687255392829877,
+      "loss": 0.6568,
+      "step": 670
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.23183235360102647,
+      "learning_rate": 0.00019676047450944008,
+      "loss": 0.5751,
+      "step": 675
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.2933000436636493,
+      "learning_rate": 0.0001966464550185782,
+      "loss": 0.7648,
+      "step": 680
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.277707855647548,
+      "learning_rate": 0.00019653049774183282,
+      "loss": 0.5829,
+      "step": 685
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.2734745539147106,
+      "learning_rate": 0.0001964126050041767,
+      "loss": 0.7626,
+      "step": 690
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.27801067049837663,
+      "learning_rate": 0.000196292779169389,
+      "loss": 0.6745,
+      "step": 695
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.35804549859433416,
+      "learning_rate": 0.0001961710226400081,
+      "loss": 0.7154,
+      "step": 700
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.33687341140848004,
+      "learning_rate": 0.00019604733785728317,
+      "loss": 0.7147,
+      "step": 705
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.3002177976890752,
+      "learning_rate": 0.00019592172730112544,
+      "loss": 0.708,
+      "step": 710
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.29390663775324355,
+      "learning_rate": 0.00019579419349005837,
+      "loss": 0.7968,
+      "step": 715
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.2746665644289368,
+      "learning_rate": 0.00019566473898116713,
+      "loss": 0.7023,
+      "step": 720
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.30268203726113624,
+      "learning_rate": 0.00019553336637004735,
+      "loss": 0.742,
+      "step": 725
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.2203190362696549,
+      "learning_rate": 0.0001954000782907532,
+      "loss": 0.7494,
+      "step": 730
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.27657841960508833,
+      "learning_rate": 0.00019526487741574437,
+      "loss": 0.7819,
+      "step": 735
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.30505267558261334,
+      "learning_rate": 0.00019512776645583263,
+      "loss": 0.6908,
+      "step": 740
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.275012136618905,
+      "learning_rate": 0.0001949887481601274,
+      "loss": 0.812,
+      "step": 745
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.4270284365039847,
+      "learning_rate": 0.00019484782531598073,
+      "loss": 0.6919,
+      "step": 750
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.33434183829428676,
+      "learning_rate": 0.0001947050007489313,
+      "loss": 0.8373,
+      "step": 755
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.30427892181618366,
+      "learning_rate": 0.00019456027732264784,
+      "loss": 0.7148,
+      "step": 760
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.3415786735415135,
+      "learning_rate": 0.00019441365793887162,
+      "loss": 0.7583,
+      "step": 765
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.2507569242816778,
+      "learning_rate": 0.00019426514553735848,
+      "loss": 0.7327,
+      "step": 770
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.21839556494141277,
+      "learning_rate": 0.00019411474309581958,
+      "loss": 0.6872,
+      "step": 775
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.27502259439590165,
+      "learning_rate": 0.00019396245362986197,
+      "loss": 0.7212,
+      "step": 780
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.31223346554540926,
+      "learning_rate": 0.00019380828019292798,
+      "loss": 0.6689,
+      "step": 785
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.36757959748248226,
+      "learning_rate": 0.00019365222587623405,
+      "loss": 0.6952,
+      "step": 790
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.22435035120320537,
+      "learning_rate": 0.00019349429380870873,
+      "loss": 0.7388,
+      "step": 795
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.2829985423238805,
+      "learning_rate": 0.00019333448715692995,
+      "loss": 0.7512,
+      "step": 800
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.2496925408376058,
+      "learning_rate": 0.0001931728091250615,
+      "loss": 0.785,
+      "step": 805
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.25741128000037666,
+      "learning_rate": 0.00019300926295478884,
+      "loss": 0.769,
+      "step": 810
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.27594077518066623,
+      "learning_rate": 0.00019284385192525405,
+      "loss": 0.692,
+      "step": 815
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.2853537905685808,
+      "learning_rate": 0.0001926765793529902,
+      "loss": 0.7143,
+      "step": 820
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.2651467635782436,
+      "learning_rate": 0.00019250744859185468,
+      "loss": 0.7936,
+      "step": 825
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.3346954952733162,
+      "learning_rate": 0.00019233646303296205,
+      "loss": 0.7081,
+      "step": 830
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.26867358358733323,
+      "learning_rate": 0.000192163626104616,
+      "loss": 0.6809,
+      "step": 835
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.28504787352387484,
+      "learning_rate": 0.00019198894127224074,
+      "loss": 0.7529,
+      "step": 840
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.3899181097549748,
+      "learning_rate": 0.00019181241203831137,
+      "loss": 0.7733,
+      "step": 845
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.3386788353875776,
+      "learning_rate": 0.0001916340419422837,
+      "loss": 0.7467,
+      "step": 850
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.27850551310804966,
+      "learning_rate": 0.00019145383456052327,
+      "loss": 0.6916,
+      "step": 855
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.2671031051978256,
+      "learning_rate": 0.00019127179350623372,
+      "loss": 0.6951,
+      "step": 860
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.3758548589898052,
+      "learning_rate": 0.00019108792242938425,
+      "loss": 0.6892,
+      "step": 865
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.301048604600236,
+      "learning_rate": 0.0001909022250166365,
+      "loss": 0.6979,
+      "step": 870
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.23229573279853194,
+      "learning_rate": 0.00019071470499127058,
+      "loss": 0.6935,
+      "step": 875
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.30646168773183613,
+      "learning_rate": 0.00019052536611311046,
+      "loss": 0.7406,
+      "step": 880
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.23840081090154241,
+      "learning_rate": 0.0001903342121784486,
+      "loss": 0.6804,
+      "step": 885
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.3415488557571022,
+      "learning_rate": 0.00019014124701996973,
+      "loss": 0.7208,
+      "step": 890
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.2437048041319469,
+      "learning_rate": 0.00018994647450667413,
+      "loss": 0.7033,
+      "step": 895
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.27194596378878544,
+      "learning_rate": 0.00018974989854379996,
+      "loss": 0.7523,
+      "step": 900
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.29587923128889887,
+      "learning_rate": 0.00018955152307274507,
+      "loss": 0.687,
+      "step": 905
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.2461081590732809,
+      "learning_rate": 0.00018935135207098785,
+      "loss": 0.6655,
+      "step": 910
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.2924444087440772,
+      "learning_rate": 0.00018914938955200754,
+      "loss": 0.6819,
+      "step": 915
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.2553944811817433,
+      "learning_rate": 0.00018894563956520374,
+      "loss": 0.7684,
+      "step": 920
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.28957181182369396,
+      "learning_rate": 0.0001887401061958153,
+      "loss": 0.7766,
+      "step": 925
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.2785032642860862,
+      "learning_rate": 0.00018853279356483826,
+      "loss": 0.7712,
+      "step": 930
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.28527351035777204,
+      "learning_rate": 0.00018832370582894334,
+      "loss": 0.6425,
+      "step": 935
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.3575295869608585,
+      "learning_rate": 0.00018811284718039256,
+      "loss": 0.6777,
+      "step": 940
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.27729585332023104,
+      "learning_rate": 0.00018790022184695523,
+      "loss": 0.6822,
+      "step": 945
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.35797548004437607,
+      "learning_rate": 0.00018768583409182305,
+      "loss": 0.7349,
+      "step": 950
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.23438312320664967,
+      "learning_rate": 0.00018746968821352483,
+      "loss": 0.7424,
+      "step": 955
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.30004449608881456,
+      "learning_rate": 0.00018725178854584007,
+      "loss": 0.7366,
+      "step": 960
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.25679879736213285,
+      "learning_rate": 0.00018703213945771229,
+      "loss": 0.7473,
+      "step": 965
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.28382114645109796,
+      "learning_rate": 0.00018681074535316125,
+      "loss": 0.8144,
+      "step": 970
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.24972781946965186,
+      "learning_rate": 0.00018658761067119484,
+      "loss": 0.6927,
+      "step": 975
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.30834044136784544,
+      "learning_rate": 0.00018636273988571991,
+      "loss": 0.6803,
+      "step": 980
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.2649505204628836,
+      "learning_rate": 0.00018613613750545255,
+      "loss": 0.6411,
+      "step": 985
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.27400500427151636,
+      "learning_rate": 0.0001859078080738279,
+      "loss": 0.7043,
+      "step": 990
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.2723459029072118,
+      "learning_rate": 0.00018567775616890888,
+      "loss": 0.7234,
+      "step": 995
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.2635191155504064,
+      "learning_rate": 0.00018544598640329432,
+      "loss": 0.7131,
+      "step": 1000
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.24356420411141796,
+      "learning_rate": 0.00018521250342402672,
+      "loss": 0.7638,
+      "step": 1005
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.21340372473955896,
+      "learning_rate": 0.00018497731191249894,
+      "loss": 0.7671,
+      "step": 1010
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.24554004517500114,
+      "learning_rate": 0.00018474041658436027,
+      "loss": 0.7679,
+      "step": 1015
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.2556998456537391,
+      "learning_rate": 0.000184501822189422,
+      "loss": 0.7323,
+      "step": 1020
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.25911000225159375,
+      "learning_rate": 0.0001842615335115621,
+      "loss": 0.7126,
+      "step": 1025
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.29529555659956847,
+      "learning_rate": 0.00018401955536862948,
+      "loss": 0.8061,
+      "step": 1030
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.2923465430082093,
+      "learning_rate": 0.00018377589261234705,
+      "loss": 0.7151,
+      "step": 1035
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.2186889691059259,
+      "learning_rate": 0.0001835305501282148,
+      "loss": 0.8274,
+      "step": 1040
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.2841289282189221,
+      "learning_rate": 0.00018328353283541158,
+      "loss": 0.724,
+      "step": 1045
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.2951337244107648,
+      "learning_rate": 0.00018303484568669667,
+      "loss": 0.6899,
+      "step": 1050
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.15330703509842647,
+      "learning_rate": 0.00018278449366831035,
+      "loss": 0.5995,
+      "step": 1055
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.24116956612307694,
+      "learning_rate": 0.00018253248179987388,
+      "loss": 0.7076,
+      "step": 1060
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.2871300642732821,
+      "learning_rate": 0.00018227881513428908,
+      "loss": 0.7362,
+      "step": 1065
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.2979215272527789,
+      "learning_rate": 0.0001820234987576368,
+      "loss": 0.7023,
+      "step": 1070
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.2663915798849715,
+      "learning_rate": 0.00018176653778907492,
+      "loss": 0.7347,
+      "step": 1075
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.29904674261298314,
+      "learning_rate": 0.00018150793738073602,
+      "loss": 0.6495,
+      "step": 1080
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.3165528665189004,
+      "learning_rate": 0.00018124770271762364,
+      "loss": 0.6942,
+      "step": 1085
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.3378794686368825,
+      "learning_rate": 0.00018098583901750867,
+      "loss": 0.7095,
+      "step": 1090
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.1302992690531124,
+      "learning_rate": 0.00018072235153082455,
+      "loss": 0.6416,
+      "step": 1095
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.254510249703615,
+      "learning_rate": 0.00018045724554056214,
+      "loss": 0.7611,
+      "step": 1100
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.32229289030294994,
+      "learning_rate": 0.0001801905263621636,
+      "loss": 0.7632,
+      "step": 1105
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.2671186610260709,
+      "learning_rate": 0.0001799221993434159,
+      "loss": 0.7229,
+      "step": 1110
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.21433416331211266,
+      "learning_rate": 0.00017965226986434377,
+      "loss": 0.6888,
+      "step": 1115
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.39600504574420103,
+      "learning_rate": 0.00017938074333710157,
+      "loss": 0.7202,
+      "step": 1120
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.3134397942298124,
+      "learning_rate": 0.00017910762520586485,
+      "loss": 0.7254,
+      "step": 1125
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.24898380650617621,
+      "learning_rate": 0.00017883292094672128,
+      "loss": 0.7194,
+      "step": 1130
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.32622878287643925,
+      "learning_rate": 0.00017855663606756078,
+      "loss": 0.6838,
+      "step": 1135
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.2301148369047203,
+      "learning_rate": 0.00017827877610796514,
+      "loss": 0.6755,
+      "step": 1140
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.35924771537726957,
+      "learning_rate": 0.00017799934663909682,
+      "loss": 0.7154,
+      "step": 1145
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.3190069733998683,
+      "learning_rate": 0.00017771835326358743,
+      "loss": 0.6468,
+      "step": 1150
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.26523634849673267,
+      "learning_rate": 0.00017743580161542525,
+      "loss": 0.656,
+      "step": 1155
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.2073552819004321,
+      "learning_rate": 0.00017715169735984233,
+      "loss": 0.662,
+      "step": 1160
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.29144492490212454,
+      "learning_rate": 0.00017686604619320093,
+      "loss": 0.7378,
+      "step": 1165
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.2826085444372445,
+      "learning_rate": 0.0001765788538428792,
+      "loss": 0.8061,
+      "step": 1170
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.3503084495094886,
+      "learning_rate": 0.00017629012606715648,
+      "loss": 0.7476,
+      "step": 1175
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.40190773179750555,
+      "learning_rate": 0.00017599986865509767,
+      "loss": 0.7002,
+      "step": 1180
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.3512537138147811,
+      "learning_rate": 0.00017570808742643746,
+      "loss": 0.7955,
+      "step": 1185
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.2686040145129207,
+      "learning_rate": 0.00017541478823146327,
+      "loss": 0.7205,
+      "step": 1190
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.2572903388079848,
+      "learning_rate": 0.00017511997695089822,
+      "loss": 0.7258,
+      "step": 1195
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.2746446577869629,
+      "learning_rate": 0.00017482365949578302,
+      "loss": 0.6587,
+      "step": 1200
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.2656858612857696,
+      "learning_rate": 0.00017452584180735774,
+      "loss": 0.7334,
+      "step": 1205
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.25317024575228997,
+      "learning_rate": 0.00017422652985694237,
+      "loss": 0.7051,
+      "step": 1210
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.2309813519294251,
+      "learning_rate": 0.00017392572964581725,
+      "loss": 0.7013,
+      "step": 1215
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.23745896843674344,
+      "learning_rate": 0.00017362344720510278,
+      "loss": 0.706,
+      "step": 1220
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.2965283402007185,
+      "learning_rate": 0.00017331968859563834,
+      "loss": 0.7749,
+      "step": 1225
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.26530828241172705,
+      "learning_rate": 0.00017301445990786102,
+      "loss": 0.7566,
+      "step": 1230
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.3188702889740104,
+      "learning_rate": 0.00017270776726168317,
+      "loss": 0.7122,
+      "step": 1235
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.269581496964367,
+      "learning_rate": 0.00017239961680637,
+      "loss": 0.7457,
+      "step": 1240
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.2706727342486633,
+      "learning_rate": 0.00017209001472041617,
+      "loss": 0.7937,
+      "step": 1245
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.25517714775768086,
+      "learning_rate": 0.0001717789672114218,
+      "loss": 0.7613,
+      "step": 1250
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.2636311888681917,
+      "learning_rate": 0.00017146648051596822,
+      "loss": 0.7079,
+      "step": 1255
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.2989402934427053,
+      "learning_rate": 0.0001711525608994927,
+      "loss": 0.6716,
+      "step": 1260
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.22104379693485937,
+      "learning_rate": 0.00017083721465616306,
+      "loss": 0.6682,
+      "step": 1265
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.330168955232277,
+      "learning_rate": 0.00017052044810875126,
+      "loss": 0.8137,
+      "step": 1270
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.31150819405882774,
+      "learning_rate": 0.00017020226760850677,
+      "loss": 0.6959,
+      "step": 1275
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.32633550990131655,
+      "learning_rate": 0.00016988267953502913,
+      "loss": 0.6908,
+      "step": 1280
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.2247055690731712,
+      "learning_rate": 0.0001695616902961401,
+      "loss": 0.6666,
+      "step": 1285
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.2810455775039889,
+      "learning_rate": 0.00016923930632775516,
+      "loss": 0.7025,
+      "step": 1290
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.25483406919968543,
+      "learning_rate": 0.00016891553409375444,
+      "loss": 0.7234,
+      "step": 1295
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.28464930239979624,
+      "learning_rate": 0.00016859038008585326,
+      "loss": 0.7341,
+      "step": 1300
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.32647501669318885,
+      "learning_rate": 0.0001682638508234717,
+      "loss": 0.6706,
+      "step": 1305
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.2092534288356594,
+      "learning_rate": 0.0001679359528536041,
+      "loss": 0.6493,
+      "step": 1310
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.3422837500211391,
+      "learning_rate": 0.0001676066927506878,
+      "loss": 0.7394,
+      "step": 1315
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.29359271678876697,
+      "learning_rate": 0.00016727607711647114,
+      "loss": 0.6866,
+      "step": 1320
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.24439616326346505,
+      "learning_rate": 0.00016694411257988135,
+      "loss": 0.6546,
+      "step": 1325
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.25431678595144386,
+      "learning_rate": 0.00016661080579689132,
+      "loss": 0.7069,
+      "step": 1330
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.2643304489008311,
+      "learning_rate": 0.00016627616345038642,
+      "loss": 0.6616,
+      "step": 1335
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.3025866711501466,
+      "learning_rate": 0.0001659401922500304,
+      "loss": 0.7082,
+      "step": 1340
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.2576690605267584,
+      "learning_rate": 0.0001656028989321309,
+      "loss": 0.6431,
+      "step": 1345
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.23465616583898175,
+      "learning_rate": 0.00016526429025950424,
+      "loss": 0.6826,
+      "step": 1350
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.265312873732132,
+      "learning_rate": 0.00016492437302134008,
+      "loss": 0.7189,
+      "step": 1355
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.24774713265915735,
+      "learning_rate": 0.00016458315403306502,
+      "loss": 0.7884,
+      "step": 1360
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.30882736567653557,
+      "learning_rate": 0.00016424064013620615,
+      "loss": 0.6989,
+      "step": 1365
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.29435047655097046,
+      "learning_rate": 0.0001638968381982538,
+      "loss": 0.6983,
+      "step": 1370
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.24898245263549598,
+      "learning_rate": 0.0001635517551125238,
+      "loss": 0.8087,
+      "step": 1375
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.3361644987929908,
+      "learning_rate": 0.0001632053977980194,
+      "loss": 0.7269,
+      "step": 1380
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.3313137963733492,
+      "learning_rate": 0.0001628577731992924,
+      "loss": 0.7212,
+      "step": 1385
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.40222894480073607,
+      "learning_rate": 0.000162508888286304,
+      "loss": 0.7047,
+      "step": 1390
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.28365087568805586,
+      "learning_rate": 0.00016215875005428499,
+      "loss": 0.7654,
+      "step": 1395
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.22437081577512294,
+      "learning_rate": 0.00016180736552359553,
+      "loss": 0.7497,
+      "step": 1400
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.2713094825399269,
+      "learning_rate": 0.0001614547417395844,
+      "loss": 0.772,
+      "step": 1405
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.31175625126445866,
+      "learning_rate": 0.00016110088577244773,
+      "loss": 0.7136,
+      "step": 1410
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.295775151866562,
+      "learning_rate": 0.0001607458047170872,
+      "loss": 0.7448,
+      "step": 1415
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.29743619662352133,
+      "learning_rate": 0.00016038950569296785,
+      "loss": 0.6254,
+      "step": 1420
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.23830246086659249,
+      "learning_rate": 0.00016003199584397528,
+      "loss": 0.7976,
+      "step": 1425
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.24738045684528134,
+      "learning_rate": 0.00015967328233827249,
+      "loss": 0.6983,
+      "step": 1430
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.27326756366363736,
+      "learning_rate": 0.000159313372368156,
+      "loss": 0.6441,
+      "step": 1435
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.3475531782988717,
+      "learning_rate": 0.00015895227314991178,
+      "loss": 0.6814,
+      "step": 1440
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.3148530822785684,
+      "learning_rate": 0.0001585899919236706,
+      "loss": 0.6948,
+      "step": 1445
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.25377488886498006,
+      "learning_rate": 0.00015822653595326275,
+      "loss": 0.7238,
+      "step": 1450
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.3013200425652317,
+      "learning_rate": 0.0001578619125260724,
+      "loss": 0.75,
+      "step": 1455
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.2801858745291455,
+      "learning_rate": 0.00015749612895289152,
+      "loss": 0.6473,
+      "step": 1460
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.2643097355421632,
+      "learning_rate": 0.00015712919256777335,
+      "loss": 0.732,
+      "step": 1465
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.24859906355405817,
+      "learning_rate": 0.00015676111072788527,
+      "loss": 0.684,
+      "step": 1470
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.2978599865430522,
+      "learning_rate": 0.0001563918908133614,
+      "loss": 0.718,
+      "step": 1475
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.3175233807927027,
+      "learning_rate": 0.00015602154022715435,
+      "loss": 0.7452,
+      "step": 1480
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.29411432367231416,
+      "learning_rate": 0.00015565006639488722,
+      "loss": 0.6951,
+      "step": 1485
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.24712225993420803,
+      "learning_rate": 0.0001552774767647043,
+      "loss": 0.6921,
+      "step": 1490
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.296954935837503,
+      "learning_rate": 0.000154903778807122,
+      "loss": 0.695,
+      "step": 1495
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.3090817515220463,
+      "learning_rate": 0.0001545289800148789,
+      "loss": 0.725,
+      "step": 1500
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.25680112732997773,
+      "learning_rate": 0.00015415308790278572,
+      "loss": 0.7326,
+      "step": 1505
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.2713659444254739,
+      "learning_rate": 0.0001537761100075744,
+      "loss": 0.7369,
+      "step": 1510
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.3498089133387774,
+      "learning_rate": 0.00015339805388774714,
+      "loss": 0.7819,
+      "step": 1515
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.2657852286056311,
+      "learning_rate": 0.00015301892712342482,
+      "loss": 0.7273,
+      "step": 1520
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.25285783621953456,
+      "learning_rate": 0.00015263873731619508,
+      "loss": 0.766,
+      "step": 1525
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.3072970711277045,
+      "learning_rate": 0.00015225749208895968,
+      "loss": 0.7195,
+      "step": 1530
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.3660032567021492,
+      "learning_rate": 0.000151875199085782,
+      "loss": 0.7699,
+      "step": 1535
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.2924264482609829,
+      "learning_rate": 0.0001514918659717335,
+      "loss": 0.7105,
+      "step": 1540
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.3309598652416674,
+      "learning_rate": 0.00015110750043274008,
+      "loss": 0.732,
+      "step": 1545
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.25120067042745076,
+      "learning_rate": 0.00015072211017542813,
+      "loss": 0.7447,
+      "step": 1550
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.29548371062233497,
+      "learning_rate": 0.0001503357029269698,
+      "loss": 0.6245,
+      "step": 1555
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.2882495017189674,
+      "learning_rate": 0.00014994828643492827,
+      "loss": 0.7354,
+      "step": 1560
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.2926891269519082,
+      "learning_rate": 0.00014955986846710222,
+      "loss": 0.7234,
+      "step": 1565
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.3208205900575705,
+      "learning_rate": 0.00014917045681137026,
+      "loss": 0.6859,
+      "step": 1570
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.28730436475223414,
+      "learning_rate": 0.00014878005927553456,
+      "loss": 0.655,
+      "step": 1575
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.31375609878578203,
+      "learning_rate": 0.0001483886836871646,
+      "loss": 0.7068,
+      "step": 1580
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.28634685391130577,
+      "learning_rate": 0.00014799633789343994,
+      "loss": 0.7873,
+      "step": 1585
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.29219569906271287,
+      "learning_rate": 0.00014760302976099304,
+      "loss": 0.6414,
+      "step": 1590
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.2809961899361123,
+      "learning_rate": 0.00014720876717575155,
+      "loss": 0.689,
+      "step": 1595
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.2966023851758411,
+      "learning_rate": 0.00014681355804278001,
+      "loss": 0.6939,
+      "step": 1600
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.22135705676789802,
+      "learning_rate": 0.00014641741028612162,
+      "loss": 0.6388,
+      "step": 1605
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.2017382590189762,
+      "learning_rate": 0.00014602033184863913,
+      "loss": 0.7172,
+      "step": 1610
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.29520895754468235,
+      "learning_rate": 0.00014562233069185572,
+      "loss": 0.8331,
+      "step": 1615
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.25912176225264727,
+      "learning_rate": 0.00014522341479579533,
+      "loss": 0.7583,
+      "step": 1620
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.27070806912751083,
+      "learning_rate": 0.0001448235921588226,
+      "loss": 0.7706,
+      "step": 1625
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.22470926163816077,
+      "learning_rate": 0.00014442287079748263,
+      "loss": 0.701,
+      "step": 1630
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.26012437368269403,
+      "learning_rate": 0.00014402125874634012,
+      "loss": 0.6856,
+      "step": 1635
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.27655512567748813,
+      "learning_rate": 0.00014361876405781832,
+      "loss": 0.7092,
+      "step": 1640
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.22547503156029328,
+      "learning_rate": 0.00014321539480203764,
+      "loss": 0.7723,
+      "step": 1645
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.35807881396099706,
+      "learning_rate": 0.00014281115906665374,
+      "loss": 0.699,
+      "step": 1650
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.25312726731980933,
+      "learning_rate": 0.00014240606495669538,
+      "loss": 0.694,
+      "step": 1655
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.3021591327476025,
+      "learning_rate": 0.00014200012059440207,
+      "loss": 0.6978,
+      "step": 1660
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.3776238352797525,
+      "learning_rate": 0.00014159333411906095,
+      "loss": 0.7518,
+      "step": 1665
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.24540414704743882,
+      "learning_rate": 0.00014118571368684383,
+      "loss": 0.7891,
+      "step": 1670
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.3138641993671755,
+      "learning_rate": 0.00014077726747064353,
+      "loss": 0.6213,
+      "step": 1675
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.24104497594679253,
+      "learning_rate": 0.00014036800365991008,
+      "loss": 0.6773,
+      "step": 1680
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.2712106009342219,
+      "learning_rate": 0.00013995793046048643,
+      "loss": 0.6977,
+      "step": 1685
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.2847114290804009,
+      "learning_rate": 0.00013954705609444404,
+      "loss": 0.6903,
+      "step": 1690
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.31132954968304133,
+      "learning_rate": 0.0001391353887999179,
+      "loss": 0.7666,
+      "step": 1695
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.2648924104898787,
+      "learning_rate": 0.00013872293683094152,
+      "loss": 0.6656,
+      "step": 1700
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.2628528269004981,
+      "learning_rate": 0.00013830970845728126,
+      "loss": 0.7704,
+      "step": 1705
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.21693389371822336,
+      "learning_rate": 0.00013789571196427055,
+      "loss": 0.6493,
+      "step": 1710
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.2447657421175684,
+      "learning_rate": 0.00013748095565264384,
+      "loss": 0.6459,
+      "step": 1715
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.3180367525541911,
+      "learning_rate": 0.00013706544783837022,
+      "loss": 0.7049,
+      "step": 1720
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.2972710048973474,
+      "learning_rate": 0.00013664919685248649,
+      "loss": 0.7238,
+      "step": 1725
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.29131857236762787,
+      "learning_rate": 0.00013623221104093025,
+      "loss": 0.8293,
+      "step": 1730
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.27510973199735633,
+      "learning_rate": 0.0001358144987643726,
+      "loss": 0.7285,
+      "step": 1735
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.23692910598713893,
+      "learning_rate": 0.00013539606839805036,
+      "loss": 0.7199,
+      "step": 1740
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.29598035037077913,
+      "learning_rate": 0.00013497692833159828,
+      "loss": 0.7042,
+      "step": 1745
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.25149104979862424,
+      "learning_rate": 0.00013455708696888085,
+      "loss": 0.6942,
+      "step": 1750
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.25496421670247854,
+      "learning_rate": 0.00013413655272782356,
+      "loss": 0.7256,
+      "step": 1755
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.2554748374618532,
+      "learning_rate": 0.00013371533404024438,
+      "loss": 0.7607,
+      "step": 1760
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.2627306476634245,
+      "learning_rate": 0.00013329343935168466,
+      "loss": 0.7101,
+      "step": 1765
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.24657317727636707,
+      "learning_rate": 0.00013287087712123962,
+      "loss": 0.7726,
+      "step": 1770
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.2559430395751914,
+      "learning_rate": 0.000132447655821389,
+      "loss": 0.6601,
+      "step": 1775
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.3049767818100991,
+      "learning_rate": 0.00013202378393782692,
+      "loss": 0.7961,
+      "step": 1780
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.28360752211200474,
+      "learning_rate": 0.00013159926996929192,
+      "loss": 0.6639,
+      "step": 1785
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.26804633258042826,
+      "learning_rate": 0.00013117412242739655,
+      "loss": 0.6503,
+      "step": 1790
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.2411104477481089,
+      "learning_rate": 0.00013074834983645657,
+      "loss": 0.7532,
+      "step": 1795
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.2900355787539025,
+      "learning_rate": 0.00013032196073332027,
+      "loss": 0.7479,
+      "step": 1800
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.2568819906683512,
+      "learning_rate": 0.000129894963667197,
+      "loss": 0.7433,
+      "step": 1805
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.22709405983327266,
+      "learning_rate": 0.00012946736719948607,
+      "loss": 0.7163,
+      "step": 1810
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.28704276569845133,
+      "learning_rate": 0.00012903917990360485,
+      "loss": 0.6965,
+      "step": 1815
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.2573742688385154,
+      "learning_rate": 0.000128610410364817,
+      "loss": 0.6919,
+      "step": 1820
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.3570766420018621,
+      "learning_rate": 0.00012818106718006023,
+      "loss": 0.6073,
+      "step": 1825
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.2471571809588589,
+      "learning_rate": 0.00012775115895777417,
+      "loss": 0.753,
+      "step": 1830
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.25821147959621743,
+      "learning_rate": 0.0001273206943177274,
+      "loss": 0.6319,
+      "step": 1835
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.30013480040772167,
+      "learning_rate": 0.00012688968189084493,
+      "loss": 0.7263,
+      "step": 1840
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.28654821268777825,
+      "learning_rate": 0.000126458130319035,
+      "loss": 0.7259,
+      "step": 1845
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.29336529390659893,
+      "learning_rate": 0.00012602604825501587,
+      "loss": 0.7032,
+      "step": 1850
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.20273978059145287,
+      "learning_rate": 0.00012559344436214223,
+      "loss": 0.6322,
+      "step": 1855
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.2483147925181707,
+      "learning_rate": 0.00012516032731423165,
+      "loss": 0.7185,
+      "step": 1860
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.28968579800640504,
+      "learning_rate": 0.00012472670579539055,
+      "loss": 0.7028,
+      "step": 1865
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.24871377665254096,
+      "learning_rate": 0.00012429258849984014,
+      "loss": 0.7093,
+      "step": 1870
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.23758441118744317,
+      "learning_rate": 0.00012385798413174206,
+      "loss": 0.7267,
+      "step": 1875
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.29112304576955467,
+      "learning_rate": 0.00012342290140502388,
+      "loss": 0.7714,
+      "step": 1880
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.246124461049581,
+      "learning_rate": 0.00012298734904320438,
+      "loss": 0.6727,
+      "step": 1885
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.25856322825399286,
+      "learning_rate": 0.00012255133577921868,
+      "loss": 0.7769,
+      "step": 1890
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.22410369058426746,
+      "learning_rate": 0.00012211487035524305,
+      "loss": 0.6866,
+      "step": 1895
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.28079722454610034,
+      "learning_rate": 0.0001216779615225197,
+      "loss": 0.6963,
+      "step": 1900
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.30696044399665856,
+      "learning_rate": 0.00012124061804118137,
+      "loss": 0.7051,
+      "step": 1905
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.247967331116475,
+      "learning_rate": 0.00012080284868007541,
+      "loss": 0.6944,
+      "step": 1910
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.21128171750447838,
+      "learning_rate": 0.00012036466221658847,
+      "loss": 0.7199,
+      "step": 1915
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.31901890228430696,
+      "learning_rate": 0.0001199260674364699,
+      "loss": 0.7786,
+      "step": 1920
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.2755016732959131,
+      "learning_rate": 0.00011948707313365614,
+      "loss": 0.8069,
+      "step": 1925
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.28668018656021443,
+      "learning_rate": 0.00011904768811009405,
+      "loss": 0.7157,
+      "step": 1930
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.3328138328405363,
+      "learning_rate": 0.00011860792117556454,
+      "loss": 0.7099,
+      "step": 1935
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.21788073287798893,
+      "learning_rate": 0.00011816778114750593,
+      "loss": 0.7644,
+      "step": 1940
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.23062823471771338,
+      "learning_rate": 0.00011772727685083724,
+      "loss": 0.7716,
+      "step": 1945
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.26093096066749466,
+      "learning_rate": 0.00011728641711778103,
+      "loss": 0.7229,
+      "step": 1950
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.23801319602695262,
+      "learning_rate": 0.00011684521078768658,
+      "loss": 0.6845,
+      "step": 1955
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.18154354473533865,
+      "learning_rate": 0.00011640366670685248,
+      "loss": 0.6315,
+      "step": 1960
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.29572089884772323,
+      "learning_rate": 0.00011596179372834924,
+      "loss": 0.8437,
+      "step": 1965
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.27149388513396633,
+      "learning_rate": 0.00011551960071184195,
+      "loss": 0.8232,
+      "step": 1970
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.24540172557967463,
+      "learning_rate": 0.00011507709652341256,
+      "loss": 0.7132,
+      "step": 1975
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.26964683600534267,
+      "learning_rate": 0.00011463429003538196,
+      "loss": 0.7275,
+      "step": 1980
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.287051761373442,
+      "learning_rate": 0.00011419119012613233,
+      "loss": 0.7184,
+      "step": 1985
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.2663459541020728,
+      "learning_rate": 0.000113747805679929,
+      "loss": 0.7964,
+      "step": 1990
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.28638851869565946,
+      "learning_rate": 0.00011330414558674234,
+      "loss": 0.7129,
+      "step": 1995
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.3786862886137478,
+      "learning_rate": 0.00011286021874206952,
+      "loss": 0.6706,
+      "step": 2000
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.34276302477862775,
+      "learning_rate": 0.00011241603404675609,
+      "loss": 0.7856,
+      "step": 2005
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.3366461922626376,
+      "learning_rate": 0.00011197160040681762,
+      "loss": 0.8178,
+      "step": 2010
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.25846779036672146,
+      "learning_rate": 0.00011152692673326111,
+      "loss": 0.7387,
+      "step": 2015
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.2364492132114722,
+      "learning_rate": 0.0001110820219419062,
+      "loss": 0.7434,
+      "step": 2020
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.2699184257070918,
+      "learning_rate": 0.00011063689495320661,
+      "loss": 0.7102,
+      "step": 2025
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.2506412498240298,
+      "learning_rate": 0.0001101915546920711,
+      "loss": 0.7451,
+      "step": 2030
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.4833851120529976,
+      "learning_rate": 0.00010974601008768461,
+      "loss": 0.6558,
+      "step": 2035
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.25503131460937667,
+      "learning_rate": 0.00010930027007332923,
+      "loss": 0.6642,
+      "step": 2040
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.24985640167049433,
+      "learning_rate": 0.00010885434358620507,
+      "loss": 0.737,
+      "step": 2045
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.28451521701088694,
+      "learning_rate": 0.00010840823956725103,
+      "loss": 0.7406,
+      "step": 2050
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.29244380187254954,
+      "learning_rate": 0.00010796196696096564,
+      "loss": 0.7047,
+      "step": 2055
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.2236043726332343,
+      "learning_rate": 0.00010751553471522757,
+      "loss": 0.7312,
+      "step": 2060
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.27232955976439666,
+      "learning_rate": 0.00010706895178111634,
+      "loss": 0.7284,
+      "step": 2065
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.30610542432506455,
+      "learning_rate": 0.00010662222711273279,
+      "loss": 0.6335,
+      "step": 2070
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.23424590700585504,
+      "learning_rate": 0.00010617536966701962,
+      "loss": 0.7269,
+      "step": 2075
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.25989610387137024,
+      "learning_rate": 0.00010572838840358168,
+      "loss": 0.6304,
+      "step": 2080
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.2830081164838504,
+      "learning_rate": 0.00010528129228450645,
+      "loss": 0.6597,
+      "step": 2085
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.2603374987935435,
+      "learning_rate": 0.00010483409027418425,
+      "loss": 0.7489,
+      "step": 2090
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.24810278191222906,
+      "learning_rate": 0.00010438679133912858,
+      "loss": 0.6502,
+      "step": 2095
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.2528579675694808,
+      "learning_rate": 0.00010393940444779635,
+      "loss": 0.6969,
+      "step": 2100
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.28413657506453693,
+      "learning_rate": 0.00010349193857040787,
+      "loss": 0.7726,
+      "step": 2105
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.3144890009802897,
+      "learning_rate": 0.00010304440267876727,
+      "loss": 0.7856,
+      "step": 2110
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.2397459238431048,
+      "learning_rate": 0.00010259680574608248,
+      "loss": 0.764,
+      "step": 2115
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.2868615633571584,
+      "learning_rate": 0.00010214915674678523,
+      "loss": 0.7677,
+      "step": 2120
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.2497448183867612,
+      "learning_rate": 0.00010170146465635126,
+      "loss": 0.7783,
+      "step": 2125
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.253908792089182,
+      "learning_rate": 0.00010125373845112034,
+      "loss": 0.7461,
+      "step": 2130
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.23687933586461904,
+      "learning_rate": 0.00010080598710811615,
+      "loss": 0.69,
+      "step": 2135
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.3734210753745544,
+      "learning_rate": 0.00010035821960486643,
+      "loss": 0.77,
+      "step": 2140
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.2788998422640762,
+      "learning_rate": 9.9910444919223e-05,
+      "loss": 0.7058,
+      "step": 2145
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.3178188169556896,
+      "learning_rate": 9.946267202918157e-05,
+      "loss": 0.6965,
+      "step": 2150
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.26945241065973596,
+      "learning_rate": 9.901490991270201e-05,
+      "loss": 0.702,
+      "step": 2155
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.5207203316817794,
+      "learning_rate": 9.856716754752796e-05,
+      "loss": 0.6478,
+      "step": 2160
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.29256712809053254,
+      "learning_rate": 9.811945391100724e-05,
+      "loss": 0.7372,
+      "step": 2165
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.25601912122707865,
+      "learning_rate": 9.767177797991155e-05,
+      "loss": 0.7444,
+      "step": 2170
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.23247510741792463,
+      "learning_rate": 9.722414873025664e-05,
+      "loss": 0.6836,
+      "step": 2175
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.2460904591621441,
+      "learning_rate": 9.677657513712221e-05,
+      "loss": 0.7268,
+      "step": 2180
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.26208399329605536,
+      "learning_rate": 9.632906617447212e-05,
+      "loss": 0.6924,
+      "step": 2185
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.22915215031643407,
+      "learning_rate": 9.588163081497427e-05,
+      "loss": 0.7864,
+      "step": 2190
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.3119119948293986,
+      "learning_rate": 9.543427802982095e-05,
+      "loss": 0.768,
+      "step": 2195
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.2973539056568546,
+      "learning_rate": 9.498701678854865e-05,
+      "loss": 0.6836,
+      "step": 2200
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.24915238391515174,
+      "learning_rate": 9.453985605885851e-05,
+      "loss": 0.7307,
+      "step": 2205
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.29466985701358944,
+      "learning_rate": 9.409280480643628e-05,
+      "loss": 0.7751,
+      "step": 2210
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.31502684795729613,
+      "learning_rate": 9.364587199477276e-05,
+      "loss": 0.7183,
+      "step": 2215
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.23712422669700867,
+      "learning_rate": 9.319906658498389e-05,
+      "loss": 0.7522,
+      "step": 2220
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.2570700308864798,
+      "learning_rate": 9.275239753563128e-05,
+      "loss": 0.7171,
+      "step": 2225
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.23813307578157708,
+      "learning_rate": 9.230587380254237e-05,
+      "loss": 0.689,
+      "step": 2230
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.2656059785729245,
+      "learning_rate": 9.185950433863107e-05,
+      "loss": 0.6694,
+      "step": 2235
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.1857069050131154,
+      "learning_rate": 9.141329809371803e-05,
+      "loss": 0.7626,
+      "step": 2240
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.25034208426960974,
+      "learning_rate": 9.096726401435146e-05,
+      "loss": 0.7116,
+      "step": 2245
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.22067489301839913,
+      "learning_rate": 9.052141104362748e-05,
+      "loss": 0.6499,
+      "step": 2250
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.23199516425281616,
+      "learning_rate": 9.007574812101107e-05,
+      "loss": 0.7501,
+      "step": 2255
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.25421299632271965,
+      "learning_rate": 8.963028418215653e-05,
+      "loss": 0.8147,
+      "step": 2260
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.260651537286194,
+      "learning_rate": 8.918502815872865e-05,
+      "loss": 0.7485,
+      "step": 2265
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.2655797468541603,
+      "learning_rate": 8.873998897822336e-05,
+      "loss": 0.7505,
+      "step": 2270
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.2917164025171058,
+      "learning_rate": 8.829517556378888e-05,
+      "loss": 0.6755,
+      "step": 2275
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.3350035349023889,
+      "learning_rate": 8.785059683404672e-05,
+      "loss": 0.7015,
+      "step": 2280
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.3253925397932177,
+      "learning_rate": 8.740626170291297e-05,
+      "loss": 0.7551,
+      "step": 2285
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.2899593792802913,
+      "learning_rate": 8.696217907941941e-05,
+      "loss": 0.7185,
+      "step": 2290
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.274238966245006,
+      "learning_rate": 8.65183578675351e-05,
+      "loss": 0.7031,
+      "step": 2295
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.27886346870120726,
+      "learning_rate": 8.607480696598762e-05,
+      "loss": 0.61,
+      "step": 2300
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.3552726057492985,
+      "learning_rate": 8.563153526808484e-05,
+      "loss": 0.7706,
+      "step": 2305
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.26275345892575847,
+      "learning_rate": 8.518855166153644e-05,
+      "loss": 0.6925,
+      "step": 2310
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.3537795284485823,
+      "learning_rate": 8.474586502827591e-05,
+      "loss": 0.7005,
+      "step": 2315
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.2113496203404715,
+      "learning_rate": 8.43034842442822e-05,
+      "loss": 0.6709,
+      "step": 2320
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.46932602124140005,
+      "learning_rate": 8.386141817940213e-05,
+      "loss": 0.7019,
+      "step": 2325
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.2647970131377573,
+      "learning_rate": 8.341967569717202e-05,
+      "loss": 0.6522,
+      "step": 2330
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.2495289774517795,
+      "learning_rate": 8.297826565464057e-05,
+      "loss": 0.6333,
+      "step": 2335
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.317401195281625,
+      "learning_rate": 8.253719690219079e-05,
+      "loss": 0.7297,
+      "step": 2340
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.2722505836316711,
+      "learning_rate": 8.20964782833628e-05,
+      "loss": 0.6206,
+      "step": 2345
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.3347413066633305,
+      "learning_rate": 8.165611863467644e-05,
+      "loss": 0.7258,
+      "step": 2350
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.23149723477405734,
+      "learning_rate": 8.12161267854542e-05,
+      "loss": 0.6872,
+      "step": 2355
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.21852198583337765,
+      "learning_rate": 8.077651155764387e-05,
+      "loss": 0.6529,
+      "step": 2360
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.23663927702457488,
+      "learning_rate": 8.033728176564216e-05,
+      "loss": 0.7468,
+      "step": 2365
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.28901081689808944,
+      "learning_rate": 7.98984462161175e-05,
+      "loss": 0.7436,
+      "step": 2370
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.25802965477339335,
+      "learning_rate": 7.946001370783375e-05,
+      "loss": 0.7899,
+      "step": 2375
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.22951991679268438,
+      "learning_rate": 7.902199303147363e-05,
+      "loss": 0.6861,
+      "step": 2380
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.3057181147340439,
+      "learning_rate": 7.858439296946263e-05,
+      "loss": 0.7769,
+      "step": 2385
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.3159742724328365,
+      "learning_rate": 7.814722229579264e-05,
+      "loss": 0.6756,
+      "step": 2390
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.27245093189237785,
+      "learning_rate": 7.771048977584641e-05,
+      "loss": 0.6128,
+      "step": 2395
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.324925192823721,
+      "learning_rate": 7.727420416622144e-05,
+      "loss": 0.8306,
+      "step": 2400
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.2999192623424217,
+      "learning_rate": 7.683837421455466e-05,
+      "loss": 0.6782,
+      "step": 2405
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.25980740031389876,
+      "learning_rate": 7.640300865934687e-05,
+      "loss": 0.6913,
+      "step": 2410
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.25802354488510615,
+      "learning_rate": 7.596811622978765e-05,
+      "loss": 0.7178,
+      "step": 2415
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.26975633006496086,
+      "learning_rate": 7.553370564558032e-05,
+      "loss": 0.7667,
+      "step": 2420
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.3248726081338585,
+      "learning_rate": 7.509978561676699e-05,
+      "loss": 0.7559,
+      "step": 2425
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.27315448139602455,
+      "learning_rate": 7.46663648435541e-05,
+      "loss": 0.6903,
+      "step": 2430
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.2950802099290335,
+      "learning_rate": 7.423345201613778e-05,
+      "loss": 0.6731,
+      "step": 2435
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.35343151310449955,
+      "learning_rate": 7.380105581452987e-05,
+      "loss": 0.735,
+      "step": 2440
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.2766967860281418,
+      "learning_rate": 7.336918490838356e-05,
+      "loss": 0.7779,
+      "step": 2445
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.255190126277391,
+      "learning_rate": 7.293784795681994e-05,
+      "loss": 0.7213,
+      "step": 2450
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.23722517953499153,
+      "learning_rate": 7.250705360825392e-05,
+      "loss": 0.6734,
+      "step": 2455
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.30683380454221565,
+      "learning_rate": 7.207681050022132e-05,
+      "loss": 0.7439,
+      "step": 2460
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.3072464317121022,
+      "learning_rate": 7.16471272592053e-05,
+      "loss": 0.7804,
+      "step": 2465
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.2565247543734341,
+      "learning_rate": 7.121801250046363e-05,
+      "loss": 0.7678,
+      "step": 2470
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.2106442564735487,
+      "learning_rate": 7.078947482785576e-05,
+      "loss": 0.7063,
+      "step": 2475
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.2476449919758559,
+      "learning_rate": 7.036152283367056e-05,
+      "loss": 0.6918,
+      "step": 2480
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.247411584313298,
+      "learning_rate": 6.993416509845376e-05,
+      "loss": 0.7182,
+      "step": 2485
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.280770588683194,
+      "learning_rate": 6.950741019083617e-05,
+      "loss": 0.7192,
+      "step": 2490
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.28453079995263136,
+      "learning_rate": 6.908126666736165e-05,
+      "loss": 0.6912,
+      "step": 2495
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.34116367134660075,
+      "learning_rate": 6.865574307231575e-05,
+      "loss": 0.7064,
+      "step": 2500
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.20407453115123925,
+      "learning_rate": 6.823084793755417e-05,
+      "loss": 0.6835,
+      "step": 2505
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.34227090827399664,
+      "learning_rate": 6.780658978233199e-05,
+      "loss": 0.7364,
+      "step": 2510
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.20382515840687232,
+      "learning_rate": 6.73829771131325e-05,
+      "loss": 0.7278,
+      "step": 2515
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.32066035575970564,
+      "learning_rate": 6.696001842349702e-05,
+      "loss": 0.6735,
+      "step": 2520
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.30087509915880645,
+      "learning_rate": 6.653772219385427e-05,
+      "loss": 0.7563,
+      "step": 2525
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.2609826902639363,
+      "learning_rate": 6.611609689135056e-05,
+      "loss": 0.7865,
+      "step": 2530
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.2541688409079705,
+      "learning_rate": 6.569515096967991e-05,
+      "loss": 0.645,
+      "step": 2535
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.18975728074739295,
+      "learning_rate": 6.527489286891459e-05,
+      "loss": 0.5808,
+      "step": 2540
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.2395876243767242,
+      "learning_rate": 6.485533101533583e-05,
+      "loss": 0.6973,
+      "step": 2545
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.3063762562351843,
+      "learning_rate": 6.443647382126509e-05,
+      "loss": 0.7221,
+      "step": 2550
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.2674321599248453,
+      "learning_rate": 6.401832968489501e-05,
+      "loss": 0.6818,
+      "step": 2555
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.2231267411467715,
+      "learning_rate": 6.360090699012145e-05,
+      "loss": 0.7404,
+      "step": 2560
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.27582828393948144,
+      "learning_rate": 6.3184214106375e-05,
+      "loss": 0.6741,
+      "step": 2565
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.27355370388789374,
+      "learning_rate": 6.27682593884535e-05,
+      "loss": 0.6353,
+      "step": 2570
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.24340603859531101,
+      "learning_rate": 6.235305117635428e-05,
+      "loss": 0.7348,
+      "step": 2575
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.24374512948837185,
+      "learning_rate": 6.193859779510712e-05,
+      "loss": 0.6906,
+      "step": 2580
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.2617782514018871,
+      "learning_rate": 6.152490755460715e-05,
+      "loss": 0.6707,
+      "step": 2585
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.25140263838285776,
+      "learning_rate": 6.111198874944845e-05,
+      "loss": 0.6892,
+      "step": 2590
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.2382806536370909,
+      "learning_rate": 6.0699849658757545e-05,
+      "loss": 0.7549,
+      "step": 2595
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.2176694237385903,
+      "learning_rate": 6.0288498546027536e-05,
+      "loss": 0.7753,
+      "step": 2600
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.23853235406621046,
+      "learning_rate": 5.987794365895229e-05,
+      "loss": 0.6735,
+      "step": 2605
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.33567228876946426,
+      "learning_rate": 5.946819322926127e-05,
+      "loss": 0.6817,
+      "step": 2610
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.2947916364685711,
+      "learning_rate": 5.9059255472554195e-05,
+      "loss": 0.6897,
+      "step": 2615
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.25054528848541685,
+      "learning_rate": 5.865113858813673e-05,
+      "loss": 0.6654,
+      "step": 2620
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.295602257498029,
+      "learning_rate": 5.824385075885557e-05,
+      "loss": 0.6837,
+      "step": 2625
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.22560608641117053,
+      "learning_rate": 5.783740015093484e-05,
+      "loss": 0.7137,
+      "step": 2630
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.2710250240814673,
+      "learning_rate": 5.743179491381204e-05,
+      "loss": 0.7015,
+      "step": 2635
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.25266821650481586,
+      "learning_rate": 5.702704317997492e-05,
+      "loss": 0.6724,
+      "step": 2640
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.23287193866674372,
+      "learning_rate": 5.662315306479806e-05,
+      "loss": 0.7023,
+      "step": 2645
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.3278896346775186,
+      "learning_rate": 5.6220132666380635e-05,
+      "loss": 0.6209,
+      "step": 2650
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.3090011108241083,
+      "learning_rate": 5.581799006538354e-05,
+      "loss": 0.7575,
+      "step": 2655
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.30486291365985196,
+      "learning_rate": 5.541673332486773e-05,
+      "loss": 0.741,
+      "step": 2660
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.29222447612374014,
+      "learning_rate": 5.5016370490132364e-05,
+      "loss": 0.8456,
+      "step": 2665
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.30928681296284377,
+      "learning_rate": 5.4616909588553674e-05,
+      "loss": 0.785,
+      "step": 2670
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.2751241257699663,
+      "learning_rate": 5.4218358629423794e-05,
+      "loss": 0.797,
+      "step": 2675
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.3020155622331884,
+      "learning_rate": 5.3820725603790346e-05,
+      "loss": 0.7335,
+      "step": 2680
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.3132294366433861,
+      "learning_rate": 5.342401848429615e-05,
+      "loss": 0.7751,
+      "step": 2685
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.2617050345460146,
+      "learning_rate": 5.30282452250193e-05,
+      "loss": 0.7865,
+      "step": 2690
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.3088811287968099,
+      "learning_rate": 5.263341376131382e-05,
+      "loss": 0.7012,
+      "step": 2695
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.3053801166515793,
+      "learning_rate": 5.223953200965055e-05,
+      "loss": 0.733,
+      "step": 2700
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.23354984138597276,
+      "learning_rate": 5.1846607867458196e-05,
+      "loss": 0.8136,
+      "step": 2705
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.2958514046814973,
+      "learning_rate": 5.145464921296537e-05,
+      "loss": 0.7738,
+      "step": 2710
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.30080922548213934,
+      "learning_rate": 5.1063663905042225e-05,
+      "loss": 0.6893,
+      "step": 2715
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.3034194981171166,
+      "learning_rate": 5.067365978304315e-05,
+      "loss": 0.6922,
+      "step": 2720
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.33312809082970085,
+      "learning_rate": 5.0284644666649485e-05,
+      "loss": 0.6818,
+      "step": 2725
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.2615876279834696,
+      "learning_rate": 4.9896626355712805e-05,
+      "loss": 0.6837,
+      "step": 2730
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.23784349313669656,
+      "learning_rate": 4.95096126300984e-05,
+      "loss": 0.6899,
+      "step": 2735
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.2103897994837222,
+      "learning_rate": 4.912361124952948e-05,
+      "loss": 0.7308,
+      "step": 2740
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.23734181510531857,
+      "learning_rate": 4.873862995343139e-05,
+      "loss": 0.6635,
+      "step": 2745
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.28754328918934463,
+      "learning_rate": 4.835467646077656e-05,
+      "loss": 0.7335,
+      "step": 2750
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.2801177269384902,
+      "learning_rate": 4.797175846992964e-05,
+      "loss": 0.6538,
+      "step": 2755
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.20377437169807378,
+      "learning_rate": 4.7589883658493296e-05,
+      "loss": 0.7343,
+      "step": 2760
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.2633015839389786,
+      "learning_rate": 4.72090596831542e-05,
+      "loss": 0.7279,
+      "step": 2765
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.3365363313211822,
+      "learning_rate": 4.682929417952939e-05,
+      "loss": 0.6493,
+      "step": 2770
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.26197423166598477,
+      "learning_rate": 4.645059476201333e-05,
+      "loss": 0.6823,
+      "step": 2775
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.22148899156479446,
+      "learning_rate": 4.6072969023625165e-05,
+      "loss": 0.7062,
+      "step": 2780
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.25093425817830756,
+      "learning_rate": 4.5696424535856574e-05,
+      "loss": 0.7058,
+      "step": 2785
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.29374695702116843,
+      "learning_rate": 4.532096884851978e-05,
+      "loss": 0.7307,
+      "step": 2790
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.24148367178848454,
+      "learning_rate": 4.494660948959645e-05,
+      "loss": 0.7908,
+      "step": 2795
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.29661735702318875,
+      "learning_rate": 4.457335396508631e-05,
+      "loss": 0.6051,
+      "step": 2800
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.2587416887380861,
+      "learning_rate": 4.420120975885723e-05,
+      "loss": 0.7074,
+      "step": 2805
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.25077785676951697,
+      "learning_rate": 4.383018433249464e-05,
+      "loss": 0.6776,
+      "step": 2810
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.26766730021230084,
+      "learning_rate": 4.346028512515232e-05,
+      "loss": 0.5915,
+      "step": 2815
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.26255599946956093,
+      "learning_rate": 4.309151955340297e-05,
+      "loss": 0.7202,
+      "step": 2820
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.23590120420333469,
+      "learning_rate": 4.272389501108962e-05,
+      "loss": 0.754,
+      "step": 2825
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.34961647734551377,
+      "learning_rate": 4.2357418869177354e-05,
+      "loss": 0.7484,
+      "step": 2830
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.2567914752516631,
+      "learning_rate": 4.1992098475605654e-05,
+      "loss": 0.7758,
+      "step": 2835
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.2859760124915357,
+      "learning_rate": 4.162794115514078e-05,
+      "loss": 0.7259,
+      "step": 2840
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.40354694842899336,
+      "learning_rate": 4.1264954209229254e-05,
+      "loss": 0.6554,
+      "step": 2845
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.31410918890445977,
+      "learning_rate": 4.0903144915851174e-05,
+      "loss": 0.6495,
+      "step": 2850
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.21920834371525927,
+      "learning_rate": 4.054252052937444e-05,
+      "loss": 0.6159,
+      "step": 2855
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.2919386673597319,
+      "learning_rate": 4.018308828040924e-05,
+      "loss": 0.7266,
+      "step": 2860
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.2907127546681081,
+      "learning_rate": 3.982485537566321e-05,
+      "loss": 0.7245,
+      "step": 2865
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.19919891868305908,
+      "learning_rate": 3.946782899779667e-05,
+      "loss": 0.6152,
+      "step": 2870
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.2874798645438607,
+      "learning_rate": 3.911201630527894e-05,
+      "loss": 0.7391,
+      "step": 2875
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.29325850253665986,
+      "learning_rate": 3.875742443224451e-05,
+      "loss": 0.6121,
+      "step": 2880
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.3084377618826001,
+      "learning_rate": 3.84040604883502e-05,
+      "loss": 0.7255,
+      "step": 2885
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.2536420954931237,
+      "learning_rate": 3.805193155863247e-05,
+      "loss": 0.636,
+      "step": 2890
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.22759460676784352,
+      "learning_rate": 3.770104470336555e-05,
+      "loss": 0.717,
+      "step": 2895
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.22465050243662746,
+      "learning_rate": 3.7351406957919636e-05,
+      "loss": 0.726,
+      "step": 2900
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.30700513035899746,
+      "learning_rate": 3.700302533262007e-05,
+      "loss": 0.7157,
+      "step": 2905
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.3126415326969147,
+      "learning_rate": 3.665590681260658e-05,
+      "loss": 0.6963,
+      "step": 2910
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.3014647601543792,
+      "learning_rate": 3.631005835769334e-05,
+      "loss": 0.7982,
+      "step": 2915
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.2519716019821263,
+      "learning_rate": 3.59654869022294e-05,
+      "loss": 0.7267,
+      "step": 2920
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.2454160383073075,
+      "learning_rate": 3.5622199354959677e-05,
+      "loss": 0.8258,
+      "step": 2925
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.22670327761402992,
+      "learning_rate": 3.5280202598886324e-05,
+      "loss": 0.6447,
+      "step": 2930
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.24579997279476928,
+      "learning_rate": 3.4939503491130934e-05,
+      "loss": 0.6665,
+      "step": 2935
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.2849552092195358,
+      "learning_rate": 3.4600108862796796e-05,
+      "loss": 0.7252,
+      "step": 2940
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.3060590068774777,
+      "learning_rate": 3.426202551883213e-05,
+      "loss": 0.7553,
+      "step": 2945
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.25467435210378453,
+      "learning_rate": 3.392526023789349e-05,
+      "loss": 0.7237,
+      "step": 2950
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.2276342572822446,
+      "learning_rate": 3.358981977221006e-05,
+      "loss": 0.6985,
+      "step": 2955
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.24334295151794627,
+      "learning_rate": 3.325571084744803e-05,
+      "loss": 0.6964,
+      "step": 2960
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.3063546229020857,
+      "learning_rate": 3.292294016257598e-05,
+      "loss": 0.69,
+      "step": 2965
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.26843353730082814,
+      "learning_rate": 3.259151438973024e-05,
+      "loss": 0.7961,
+      "step": 2970
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.28440009360240165,
+      "learning_rate": 3.2261440174081525e-05,
+      "loss": 0.7703,
+      "step": 2975
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.2307084225718035,
+      "learning_rate": 3.1932724133701344e-05,
+      "loss": 0.5392,
+      "step": 2980
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.29912287559451467,
+      "learning_rate": 3.160537285942956e-05,
+      "loss": 0.6836,
+      "step": 2985
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.278878639560016,
+      "learning_rate": 3.1279392914742046e-05,
+      "loss": 0.7267,
+      "step": 2990
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.32874234679202635,
+      "learning_rate": 3.0954790835619195e-05,
+      "loss": 0.6985,
+      "step": 2995
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.27469910090881833,
+      "learning_rate": 3.06315731304148e-05,
+      "loss": 0.7915,
+      "step": 3000
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.33032565430804056,
+      "learning_rate": 3.0309746279725748e-05,
+      "loss": 0.672,
+      "step": 3005
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.22509955106392437,
+      "learning_rate": 2.998931673626175e-05,
+      "loss": 0.6924,
+      "step": 3010
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.24954960719448557,
+      "learning_rate": 2.9670290924716337e-05,
+      "loss": 0.7648,
+      "step": 3015
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.18097577821872354,
+      "learning_rate": 2.935267524163774e-05,
+      "loss": 0.724,
+      "step": 3020
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.2531814760979886,
+      "learning_rate": 2.9036476055300765e-05,
+      "loss": 0.7565,
+      "step": 3025
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.26537867964143863,
+      "learning_rate": 2.872169970557913e-05,
+      "loss": 0.7587,
+      "step": 3030
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.2503354683898301,
+      "learning_rate": 2.8408352503818357e-05,
+      "loss": 0.7464,
+      "step": 3035
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.31309433063885866,
+      "learning_rate": 2.8096440732709083e-05,
+      "loss": 0.7215,
+      "step": 3040
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.273515150138006,
+      "learning_rate": 2.778597064616133e-05,
+      "loss": 0.7107,
+      "step": 3045
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.29261711206548846,
+      "learning_rate": 2.7476948469178887e-05,
+      "loss": 0.6895,
+      "step": 3050
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.2613608515124986,
+      "learning_rate": 2.716938039773459e-05,
+      "loss": 0.751,
+      "step": 3055
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.3275410166575591,
+      "learning_rate": 2.6863272598646106e-05,
+      "loss": 0.6056,
+      "step": 3060
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.243877381080221,
+      "learning_rate": 2.6558631209452323e-05,
+      "loss": 0.7083,
+      "step": 3065
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.2670921124404005,
+      "learning_rate": 2.625546233829016e-05,
+      "loss": 0.6769,
+      "step": 3070
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.26772171237531533,
+      "learning_rate": 2.5953772063772252e-05,
+      "loss": 0.7421,
+      "step": 3075
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.24929713338883455,
+      "learning_rate": 2.5653566434864928e-05,
+      "loss": 0.7006,
+      "step": 3080
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.27784311313283716,
+      "learning_rate": 2.5354851470767037e-05,
+      "loss": 0.6997,
+      "step": 3085
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.27642936475366325,
+      "learning_rate": 2.5057633160789184e-05,
+      "loss": 0.7078,
+      "step": 3090
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.2916050170142823,
+      "learning_rate": 2.476191746423373e-05,
+      "loss": 0.7769,
+      "step": 3095
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.29742464103075256,
+      "learning_rate": 2.446771031027527e-05,
+      "loss": 0.6983,
+      "step": 3100
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.2520881480340573,
+      "learning_rate": 2.4175017597841666e-05,
+      "loss": 0.7945,
+      "step": 3105
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.2456996066425862,
+      "learning_rate": 2.3883845195495878e-05,
+      "loss": 0.7445,
+      "step": 3110
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.20892213586106406,
+      "learning_rate": 2.3594198941318248e-05,
+      "loss": 0.6771,
+      "step": 3115
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.22864359852007016,
+      "learning_rate": 2.330608464278953e-05,
+      "loss": 0.7326,
+      "step": 3120
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.2630175043544034,
+      "learning_rate": 2.3019508076674267e-05,
+      "loss": 0.6733,
+      "step": 3125
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.25018036673355504,
+      "learning_rate": 2.273447498890521e-05,
+      "loss": 0.8375,
+      "step": 3130
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.2970455279248964,
+      "learning_rate": 2.2450991094467865e-05,
+      "loss": 0.665,
+      "step": 3135
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.24388424842231599,
+      "learning_rate": 2.2169062077286075e-05,
+      "loss": 0.6899,
+      "step": 3140
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.2620919423557573,
+      "learning_rate": 2.1888693590107946e-05,
+      "loss": 0.7008,
+      "step": 3145
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.1597077931299718,
+      "learning_rate": 2.1609891254392678e-05,
+      "loss": 0.6285,
+      "step": 3150
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.24978397247092607,
+      "learning_rate": 2.1332660660197602e-05,
+      "loss": 0.6707,
+      "step": 3155
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.32226574046620643,
+      "learning_rate": 2.1057007366066373e-05,
+      "loss": 0.7732,
+      "step": 3160
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.2633118389195218,
+      "learning_rate": 2.0782936898917204e-05,
+      "loss": 0.6722,
+      "step": 3165
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.2065809642257561,
+      "learning_rate": 2.0510454753932395e-05,
+      "loss": 0.6913,
+      "step": 3170
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.2502396270636488,
+      "learning_rate": 2.0239566394447874e-05,
+      "loss": 0.7196,
+      "step": 3175
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.2553706993409484,
+      "learning_rate": 1.9970277251843862e-05,
+      "loss": 0.7153,
+      "step": 3180
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.20930260480653445,
+      "learning_rate": 1.9702592725435786e-05,
+      "loss": 0.7252,
+      "step": 3185
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.2624209534259566,
+      "learning_rate": 1.9436518182366158e-05,
+      "loss": 0.7354,
+      "step": 3190
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.2835973619438766,
+      "learning_rate": 1.9172058957496876e-05,
+      "loss": 0.7456,
+      "step": 3195
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.20758735208627346,
+      "learning_rate": 1.8909220353302392e-05,
+      "loss": 0.6972,
+      "step": 3200
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.34020958211549873,
+      "learning_rate": 1.8648007639763176e-05,
+      "loss": 0.7108,
+      "step": 3205
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.25980592525235724,
+      "learning_rate": 1.838842605426031e-05,
+      "loss": 0.6254,
+      "step": 3210
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.23905673738763042,
+      "learning_rate": 1.813048080147025e-05,
+      "loss": 0.6463,
+      "step": 3215
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.21218256948907488,
+      "learning_rate": 1.7874177053260598e-05,
+      "loss": 0.6492,
+      "step": 3220
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.2000868258423336,
+      "learning_rate": 1.7619519948586348e-05,
+      "loss": 0.6001,
+      "step": 3225
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.103753197876994,
+      "learning_rate": 1.736651459338695e-05,
+      "loss": 0.6952,
+      "step": 3230
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.2682682663167215,
+      "learning_rate": 1.711516606048377e-05,
+      "loss": 0.7109,
+      "step": 3235
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.25063878481734475,
+      "learning_rate": 1.6865479389478545e-05,
+      "loss": 0.7225,
+      "step": 3240
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.2149654094754697,
+      "learning_rate": 1.6617459586652196e-05,
+      "loss": 0.6709,
+      "step": 3245
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.24092454032089566,
+      "learning_rate": 1.6371111624864543e-05,
+      "loss": 0.6764,
+      "step": 3250
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.2454121794760406,
+      "learning_rate": 1.612644044345456e-05,
+      "loss": 0.6982,
+      "step": 3255
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.3041665993053294,
+      "learning_rate": 1.5883450948141377e-05,
+      "loss": 0.658,
+      "step": 3260
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.2143063192080754,
+      "learning_rate": 1.564214801092585e-05,
+      "loss": 0.7606,
+      "step": 3265
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.23715373616102758,
+      "learning_rate": 1.540253646999299e-05,
+      "loss": 0.7262,
+      "step": 3270
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.266422273258976,
+      "learning_rate": 1.5164621129614787e-05,
+      "loss": 0.6795,
+      "step": 3275
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.2303123926465437,
+      "learning_rate": 1.4928406760054059e-05,
+      "loss": 0.6995,
+      "step": 3280
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.30575921163679864,
+      "learning_rate": 1.4693898097468662e-05,
+      "loss": 0.7238,
+      "step": 3285
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.2723263223045365,
+      "learning_rate": 1.4461099843816684e-05,
+      "loss": 0.694,
+      "step": 3290
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.2704567901239254,
+      "learning_rate": 1.4230016666761981e-05,
+      "loss": 0.7654,
+      "step": 3295
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.24548577793970916,
+      "learning_rate": 1.4000653199580782e-05,
+      "loss": 0.766,
+      "step": 3300
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.31075109788590083,
+      "learning_rate": 1.3773014041068633e-05,
+      "loss": 0.7156,
+      "step": 3305
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.23774900014129338,
+      "learning_rate": 1.3547103755448287e-05,
+      "loss": 0.7829,
+      "step": 3310
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.20200953268790334,
+      "learning_rate": 1.3322926872278097e-05,
+      "loss": 0.7726,
+      "step": 3315
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.26897850573503673,
+      "learning_rate": 1.3100487886361379e-05,
+      "loss": 0.7458,
+      "step": 3320
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.2397184808414706,
+      "learning_rate": 1.2879791257656049e-05,
+      "loss": 0.7193,
+      "step": 3325
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.20474419583059367,
+      "learning_rate": 1.266084141118542e-05,
+      "loss": 0.6816,
+      "step": 3330
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.22050960020614993,
+      "learning_rate": 1.2443642736949258e-05,
+      "loss": 0.7013,
+      "step": 3335
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.22350828170159398,
+      "learning_rate": 1.2228199589835999e-05,
+      "loss": 0.7583,
+      "step": 3340
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.27678043547723374,
+      "learning_rate": 1.2014516289535249e-05,
+      "loss": 0.7093,
+      "step": 3345
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.2641609501988626,
+      "learning_rate": 1.1802597120451286e-05,
+      "loss": 0.7644,
+      "step": 3350
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.29287525021588734,
+      "learning_rate": 1.1592446331617068e-05,
+      "loss": 0.6913,
+      "step": 3355
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.3132569389457329,
+      "learning_rate": 1.1384068136609105e-05,
+      "loss": 0.7836,
+      "step": 3360
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.3257533444384463,
+      "learning_rate": 1.1177466713462926e-05,
+      "loss": 0.6575,
+      "step": 3365
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.3658464833825346,
+      "learning_rate": 1.0972646204589377e-05,
+      "loss": 0.6571,
+      "step": 3370
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.24393335900845253,
+      "learning_rate": 1.076961071669148e-05,
+      "loss": 0.7282,
+      "step": 3375
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.2673318448109565,
+      "learning_rate": 1.0568364320682178e-05,
+      "loss": 0.7788,
+      "step": 3380
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.24802808118552172,
+      "learning_rate": 1.0368911051602636e-05,
+      "loss": 0.6588,
+      "step": 3385
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.2725189792906911,
+      "learning_rate": 1.0171254908541372e-05,
+      "loss": 0.6729,
+      "step": 3390
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.3108637636142955,
+      "learning_rate": 9.975399854554068e-06,
+      "loss": 0.6717,
+      "step": 3395
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.2770989947603813,
+      "learning_rate": 9.781349816584162e-06,
+      "loss": 0.7259,
+      "step": 3400
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.23614090997589118,
+      "learning_rate": 9.589108685383975e-06,
+      "loss": 0.7101,
+      "step": 3405
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.2808653787629317,
+      "learning_rate": 9.398680315436903e-06,
+      "loss": 0.6933,
+      "step": 3410
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.24876373568373047,
+      "learning_rate": 9.210068524879923e-06,
+      "loss": 0.6842,
+      "step": 3415
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.2678931806909137,
+      "learning_rate": 9.023277095427173e-06,
+      "loss": 0.7012,
+      "step": 3420
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.25424385975928604,
+      "learning_rate": 8.838309772294085e-06,
+      "loss": 0.7443,
+      "step": 3425
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.2402611926537631,
+      "learning_rate": 8.655170264122303e-06,
+      "loss": 0.7288,
+      "step": 3430
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.24815108653858006,
+      "learning_rate": 8.473862242905339e-06,
+      "loss": 0.7894,
+      "step": 3435
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.21869677597116047,
+      "learning_rate": 8.294389343914899e-06,
+      "loss": 0.6481,
+      "step": 3440
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.2515057283571701,
+      "learning_rate": 8.11675516562802e-06,
+      "loss": 0.7195,
+      "step": 3445
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.24394419541003282,
+      "learning_rate": 7.940963269654922e-06,
+      "loss": 0.8022,
+      "step": 3450
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.2745733416076033,
+      "learning_rate": 7.767017180667645e-06,
+      "loss": 0.7189,
+      "step": 3455
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.19526287456709943,
+      "learning_rate": 7.594920386329252e-06,
+      "loss": 0.6829,
+      "step": 3460
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.3408386827932112,
+      "learning_rate": 7.4246763372240795e-06,
+      "loss": 0.6084,
+      "step": 3465
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.22382037366933571,
+      "learning_rate": 7.256288446788362e-06,
+      "loss": 0.6839,
+      "step": 3470
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.25700615722112985,
+      "learning_rate": 7.089760091241937e-06,
+      "loss": 0.766,
+      "step": 3475
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.20934034960089873,
+      "learning_rate": 6.925094609520455e-06,
+      "loss": 0.6959,
+      "step": 3480
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.1887454649423391,
+      "learning_rate": 6.762295303208532e-06,
+      "loss": 0.6084,
+      "step": 3485
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.24866922377491815,
+      "learning_rate": 6.601365436473439e-06,
+      "loss": 0.76,
+      "step": 3490
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.26188268702284273,
+      "learning_rate": 6.44230823599975e-06,
+      "loss": 0.716,
+      "step": 3495
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.29822606608650276,
+      "learning_rate": 6.2851268909245865e-06,
+      "loss": 0.6727,
+      "step": 3500
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.2732588418921964,
+      "learning_rate": 6.1298245527737e-06,
+      "loss": 0.7553,
+      "step": 3505
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.24961501560552232,
+      "learning_rate": 5.976404335398256e-06,
+      "loss": 0.6262,
+      "step": 3510
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.36125761108834664,
+      "learning_rate": 5.824869314912473e-06,
+      "loss": 0.702,
+      "step": 3515
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.20211776842690213,
+      "learning_rate": 5.675222529631841e-06,
+      "loss": 0.7146,
+      "step": 3520
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.34319105583503823,
+      "learning_rate": 5.527466980012297e-06,
+      "loss": 0.7202,
+      "step": 3525
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.2551086799124065,
+      "learning_rate": 5.381605628590003e-06,
+      "loss": 0.7343,
+      "step": 3530
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.24191784730605104,
+      "learning_rate": 5.237641399921955e-06,
+      "loss": 0.676,
+      "step": 3535
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.23642128893948972,
+      "learning_rate": 5.095577180527378e-06,
+      "loss": 0.75,
+      "step": 3540
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.24384299779088495,
+      "learning_rate": 4.9554158188298445e-06,
+      "loss": 0.7689,
+      "step": 3545
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.2622807786449119,
+      "learning_rate": 4.817160125100106e-06,
+      "loss": 0.7586,
+      "step": 3550
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.2849253620392849,
+      "learning_rate": 4.680812871399854e-06,
+      "loss": 0.7066,
+      "step": 3555
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.24191852621283047,
+      "learning_rate": 4.546376791525975e-06,
+      "loss": 0.7219,
+      "step": 3560
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.2212643288267702,
+      "learning_rate": 4.413854580955945e-06,
+      "loss": 0.7128,
+      "step": 3565
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.23498346434116985,
+      "learning_rate": 4.2832488967935795e-06,
+      "loss": 0.7419,
+      "step": 3570
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.24375104729509095,
+      "learning_rate": 4.154562357715952e-06,
+      "loss": 0.7244,
+      "step": 3575
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.2349397771494649,
+      "learning_rate": 4.02779754392072e-06,
+      "loss": 0.686,
+      "step": 3580
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.2657929830383169,
+      "learning_rate": 3.90295699707447e-06,
+      "loss": 0.7006,
+      "step": 3585
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.313209415878294,
+      "learning_rate": 3.780043220261764e-06,
+      "loss": 0.7512,
+      "step": 3590
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.2555739248275014,
+      "learning_rate": 3.659058677934957e-06,
+      "loss": 0.7011,
+      "step": 3595
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.33802960858245146,
+      "learning_rate": 3.540005795864709e-06,
+      "loss": 0.6056,
+      "step": 3600
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.1922803471445042,
+      "learning_rate": 3.422886961091476e-06,
+      "loss": 0.6843,
+      "step": 3605
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.29061663111511976,
+      "learning_rate": 3.3077045218775192e-06,
+      "loss": 0.7806,
+      "step": 3610
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.247673796937873,
+      "learning_rate": 3.194460787659892e-06,
+      "loss": 0.7209,
+      "step": 3615
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.2560647317152861,
+      "learning_rate": 3.0831580290041184e-06,
+      "loss": 0.7605,
+      "step": 3620
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.3634088101447885,
+      "learning_rate": 2.973798477558709e-06,
+      "loss": 0.7043,
+      "step": 3625
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.18733318063360602,
+      "learning_rate": 2.8663843260103074e-06,
+      "loss": 0.6693,
+      "step": 3630
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.21584750971265176,
+      "learning_rate": 2.76091772803988e-06,
+      "loss": 0.7167,
+      "step": 3635
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.3177595541385448,
+      "learning_rate": 2.6574007982793857e-06,
+      "loss": 0.7402,
+      "step": 3640
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.28268884396101174,
+      "learning_rate": 2.555835612269475e-06,
+      "loss": 0.6604,
+      "step": 3645
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.23788632053961686,
+      "learning_rate": 2.456224206417812e-06,
+      "loss": 0.6957,
+      "step": 3650
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.26249888785251196,
+      "learning_rate": 2.35856857795832e-06,
+      "loss": 0.7676,
+      "step": 3655
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.4060765673216887,
+      "learning_rate": 2.262870684911045e-06,
+      "loss": 0.6793,
+      "step": 3660
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.3094366288959559,
+      "learning_rate": 2.169132446042976e-06,
+      "loss": 0.6742,
+      "step": 3665
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.23238542996653788,
+      "learning_rate": 2.0773557408295343e-06,
+      "loss": 0.7142,
+      "step": 3670
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.24241260570013387,
+      "learning_rate": 1.9875424094168647e-06,
+      "loss": 0.694,
+      "step": 3675
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.24945986800032627,
+      "learning_rate": 1.8996942525850047e-06,
+      "loss": 0.6959,
+      "step": 3680
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.2943157341191807,
+      "learning_rate": 1.813813031711742e-06,
+      "loss": 0.6987,
+      "step": 3685
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.2501518225722179,
+      "learning_rate": 1.7299004687372665e-06,
+      "loss": 0.6767,
+      "step": 3690
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.2773238988005913,
+      "learning_rate": 1.6479582461297217e-06,
+      "loss": 0.6753,
+      "step": 3695
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.24642550651253037,
+      "learning_rate": 1.5679880068514174e-06,
+      "loss": 0.6836,
+      "step": 3700
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.23944679732673346,
+      "learning_rate": 1.4899913543258814e-06,
+      "loss": 0.7099,
+      "step": 3705
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.321999340037235,
+      "learning_rate": 1.4139698524057165e-06,
+      "loss": 0.7798,
+      "step": 3710
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.3325659255065644,
+      "learning_rate": 1.3399250253413154e-06,
+      "loss": 0.767,
+      "step": 3715
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.30837054124513025,
+      "learning_rate": 1.2678583577501624e-06,
+      "loss": 0.6706,
+      "step": 3720
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.31468665328396034,
+      "learning_rate": 1.1977712945872243e-06,
+      "loss": 0.7187,
+      "step": 3725
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.2858601801052754,
+      "learning_rate": 1.1296652411158182e-06,
+      "loss": 0.7022,
+      "step": 3730
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.24729340626419855,
+      "learning_rate": 1.0635415628795665e-06,
+      "loss": 0.691,
+      "step": 3735
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.23960085243766105,
+      "learning_rate": 9.994015856749527e-07,
+      "loss": 0.7091,
+      "step": 3740
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.20680354782081353,
+      "learning_rate": 9.372465955247544e-07,
+      "loss": 0.6042,
+      "step": 3745
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.2644774077674383,
+      "learning_rate": 8.770778386522627e-07,
+      "loss": 0.7627,
+      "step": 3750
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.2557259933660033,
+      "learning_rate": 8.188965214562804e-07,
+      "loss": 0.7372,
+      "step": 3755
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.2637639760583473,
+      "learning_rate": 7.627038104869199e-07,
+      "loss": 0.6702,
+      "step": 3760
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.28884424646282286,
+      "learning_rate": 7.085008324222763e-07,
+      "loss": 0.6375,
+      "step": 3765
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.27748515379743166,
+      "learning_rate": 6.562886740457797e-07,
+      "loss": 0.7361,
+      "step": 3770
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.2323425377751082,
+      "learning_rate": 6.060683822244117e-07,
+      "loss": 0.6933,
+      "step": 3775
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.357455400684556,
+      "learning_rate": 5.578409638877457e-07,
+      "loss": 0.5931,
+      "step": 3780
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.2724599339735716,
+      "learning_rate": 5.116073860077064e-07,
+      "loss": 0.6707,
+      "step": 3785
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.30393724062041544,
+      "learning_rate": 4.6736857557925227e-07,
+      "loss": 0.6907,
+      "step": 3790
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.30371804708625333,
+      "learning_rate": 4.2512541960171294e-07,
+      "loss": 0.7718,
+      "step": 3795
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.35894189655237235,
+      "learning_rate": 3.8487876506106966e-07,
+      "loss": 0.7522,
+      "step": 3800
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.24596908002441525,
+      "learning_rate": 3.466294189129249e-07,
+      "loss": 0.6583,
+      "step": 3805
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.2920176199203356,
+      "learning_rate": 3.1037814806634815e-07,
+      "loss": 0.707,
+      "step": 3810
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.2180144616080907,
+      "learning_rate": 2.7612567936849964e-07,
+      "loss": 0.6582,
+      "step": 3815
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.26484981601847507,
+      "learning_rate": 2.43872699590042e-07,
+      "loss": 0.6977,
+      "step": 3820
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.2626803702380842,
+      "learning_rate": 2.136198554113844e-07,
+      "loss": 0.7858,
+      "step": 3825
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.2744308629524476,
+      "learning_rate": 1.8536775340970425e-07,
+      "loss": 0.7359,
+      "step": 3830
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.2397873340818912,
+      "learning_rate": 1.591169600468123e-07,
+      "loss": 0.7332,
+      "step": 3835
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.261016022377339,
+      "learning_rate": 1.348680016577397e-07,
+      "loss": 0.7853,
+      "step": 3840
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.26956288380911886,
+      "learning_rate": 1.126213644402463e-07,
+      "loss": 0.7287,
+      "step": 3845
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.250467321318094,
+      "learning_rate": 9.237749444505062e-08,
+      "loss": 0.6572,
+      "step": 3850
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.2693162150713498,
+      "learning_rate": 7.413679756684832e-08,
+      "loss": 0.7318,
+      "step": 3855
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.29918573890964684,
+      "learning_rate": 5.7899639536251883e-08,
+      "loss": 0.707,
+      "step": 3860
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.27475327570345426,
+      "learning_rate": 4.366634591237428e-08,
+      "loss": 0.7348,
+      "step": 3865
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.21723287925397622,
+      "learning_rate": 3.143720207635648e-08,
+      "loss": 0.6721,
+      "step": 3870
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.35560080211348427,
+      "learning_rate": 2.1212453225627482e-08,
+      "loss": 0.7261,
+      "step": 3875
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.30524103845820444,
+      "learning_rate": 1.299230436898613e-08,
+      "loss": 0.7778,
+      "step": 3880
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.2662759700171976,
+      "learning_rate": 6.776920322515423e-09,
+      "loss": 0.6589,
+      "step": 3885
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.2828428158719728,
+      "learning_rate": 2.566425706218567e-09,
+      "loss": 0.6907,
+      "step": 3890
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.23136076886647983,
+      "learning_rate": 3.609049415764787e-10,
+      "loss": 0.7324,
+      "step": 3895
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": NaN,
+      "eval_runtime": 1502.4585,
+      "eval_samples_per_second": 1.538,
+      "eval_steps_per_second": 0.385,
+      "step": 3898
+    },
+    {
+      "epoch": 1.0,
+      "step": 3898,
+      "total_flos": 1.0023612803710976e+16,
+      "train_loss": 0.45687119700224843,
+      "train_runtime": 26657.4924,
+      "train_samples_per_second": 2.339,
+      "train_steps_per_second": 0.146
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 3898,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 100,
+  "total_flos": 1.0023612803710976e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}