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": 0.18273725037333932,
+      "learning_rate": 5.128205128205128e-07,
+      "loss": 0.9609,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.4188407238663635,
+      "learning_rate": 2.564102564102564e-06,
+      "loss": 0.8848,
+      "step": 5
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.19777627866810965,
+      "learning_rate": 5.128205128205128e-06,
+      "loss": 0.7625,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.22192343921645707,
+      "learning_rate": 7.692307692307694e-06,
+      "loss": 0.7876,
+      "step": 15
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.18841857488548644,
+      "learning_rate": 1.0256410256410256e-05,
+      "loss": 0.8592,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.33681419457634554,
+      "learning_rate": 1.282051282051282e-05,
+      "loss": 0.8162,
+      "step": 25
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.33600938003479874,
+      "learning_rate": 1.5384615384615387e-05,
+      "loss": 0.8894,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.3176697356314787,
+      "learning_rate": 1.794871794871795e-05,
+      "loss": 0.7155,
+      "step": 35
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.21204726045838998,
+      "learning_rate": 2.0512820512820512e-05,
+      "loss": 0.7677,
+      "step": 40
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.19741016497576444,
+      "learning_rate": 2.307692307692308e-05,
+      "loss": 0.767,
+      "step": 45
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.2867616395864828,
+      "learning_rate": 2.564102564102564e-05,
+      "loss": 0.7459,
+      "step": 50
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.277593018616144,
+      "learning_rate": 2.8205128205128207e-05,
+      "loss": 0.716,
+      "step": 55
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.20494371343645457,
+      "learning_rate": 3.0769230769230774e-05,
+      "loss": 0.7252,
+      "step": 60
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.2514927327142473,
+      "learning_rate": 3.3333333333333335e-05,
+      "loss": 0.7427,
+      "step": 65
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.3569958514232062,
+      "learning_rate": 3.58974358974359e-05,
+      "loss": 0.766,
+      "step": 70
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.24879302023855776,
+      "learning_rate": 3.846153846153846e-05,
+      "loss": 0.7423,
+      "step": 75
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.25452303418558414,
+      "learning_rate": 4.1025641025641023e-05,
+      "loss": 0.6759,
+      "step": 80
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.2839971702285102,
+      "learning_rate": 4.358974358974359e-05,
+      "loss": 0.7479,
+      "step": 85
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.30384992603439903,
+      "learning_rate": 4.615384615384616e-05,
+      "loss": 0.6552,
+      "step": 90
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.3852050595096914,
+      "learning_rate": 4.871794871794872e-05,
+      "loss": 0.6522,
+      "step": 95
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.22298687548607224,
+      "learning_rate": 5.128205128205128e-05,
+      "loss": 0.7471,
+      "step": 100
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.2048934487949906,
+      "learning_rate": 5.384615384615385e-05,
+      "loss": 0.7084,
+      "step": 105
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.2069689289264241,
+      "learning_rate": 5.6410256410256414e-05,
+      "loss": 0.6885,
+      "step": 110
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.24817118260446214,
+      "learning_rate": 5.897435897435898e-05,
+      "loss": 0.7736,
+      "step": 115
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.40895048260028827,
+      "learning_rate": 6.153846153846155e-05,
+      "loss": 0.8098,
+      "step": 120
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.35594857119791506,
+      "learning_rate": 6.410256410256412e-05,
+      "loss": 0.7543,
+      "step": 125
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.21471650478613163,
+      "learning_rate": 6.666666666666667e-05,
+      "loss": 0.6914,
+      "step": 130
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.3610348756628403,
+      "learning_rate": 6.923076923076924e-05,
+      "loss": 0.7253,
+      "step": 135
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.26836543734793383,
+      "learning_rate": 7.17948717948718e-05,
+      "loss": 0.6539,
+      "step": 140
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.2835751411318648,
+      "learning_rate": 7.435897435897436e-05,
+      "loss": 0.8097,
+      "step": 145
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.20105648692084657,
+      "learning_rate": 7.692307692307693e-05,
+      "loss": 0.7385,
+      "step": 150
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.16414727315119573,
+      "learning_rate": 7.948717948717948e-05,
+      "loss": 0.6527,
+      "step": 155
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.24741062361895544,
+      "learning_rate": 8.205128205128205e-05,
+      "loss": 0.6857,
+      "step": 160
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.20873983107979052,
+      "learning_rate": 8.461538461538461e-05,
+      "loss": 0.7498,
+      "step": 165
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.22699678420428857,
+      "learning_rate": 8.717948717948718e-05,
+      "loss": 0.7112,
+      "step": 170
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.19007788693963085,
+      "learning_rate": 8.974358974358975e-05,
+      "loss": 0.751,
+      "step": 175
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.18866388272365925,
+      "learning_rate": 9.230769230769232e-05,
+      "loss": 0.6833,
+      "step": 180
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.19818865870224087,
+      "learning_rate": 9.487179487179487e-05,
+      "loss": 0.6851,
+      "step": 185
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.23302753556690228,
+      "learning_rate": 9.743589743589744e-05,
+      "loss": 0.7745,
+      "step": 190
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.18540393542694006,
+      "learning_rate": 0.0001,
+      "loss": 0.7403,
+      "step": 195
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.1839673690602772,
+      "learning_rate": 0.00010256410256410256,
+      "loss": 0.8231,
+      "step": 200
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.22798420463327215,
+      "learning_rate": 0.00010512820512820514,
+      "loss": 0.6674,
+      "step": 205
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.2464229066359474,
+      "learning_rate": 0.0001076923076923077,
+      "loss": 0.7846,
+      "step": 210
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.19433765535813324,
+      "learning_rate": 0.00011025641025641027,
+      "loss": 0.7632,
+      "step": 215
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.21304653327041365,
+      "learning_rate": 0.00011282051282051283,
+      "loss": 0.7541,
+      "step": 220
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.1770514082218146,
+      "learning_rate": 0.00011538461538461538,
+      "loss": 0.6967,
+      "step": 225
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.22467577773491165,
+      "learning_rate": 0.00011794871794871796,
+      "loss": 0.7131,
+      "step": 230
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.1765044110348834,
+      "learning_rate": 0.00012051282051282052,
+      "loss": 0.7442,
+      "step": 235
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.29795271013699604,
+      "learning_rate": 0.0001230769230769231,
+      "loss": 0.8024,
+      "step": 240
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.14389624633185258,
+      "learning_rate": 0.00012564102564102564,
+      "loss": 0.7101,
+      "step": 245
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.21079245502502064,
+      "learning_rate": 0.00012820512820512823,
+      "loss": 0.709,
+      "step": 250
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.17829547036873233,
+      "learning_rate": 0.00013076923076923077,
+      "loss": 0.7443,
+      "step": 255
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.2173226668602675,
+      "learning_rate": 0.00013333333333333334,
+      "loss": 0.6832,
+      "step": 260
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.26876676153975865,
+      "learning_rate": 0.0001358974358974359,
+      "loss": 0.7284,
+      "step": 265
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.15830628083598494,
+      "learning_rate": 0.00013846153846153847,
+      "loss": 0.7194,
+      "step": 270
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.19342822085523578,
+      "learning_rate": 0.00014102564102564104,
+      "loss": 0.7217,
+      "step": 275
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.18871672612378254,
+      "learning_rate": 0.0001435897435897436,
+      "loss": 0.6593,
+      "step": 280
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.21272446268662773,
+      "learning_rate": 0.00014615384615384615,
+      "loss": 0.6667,
+      "step": 285
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.2155663432457548,
+      "learning_rate": 0.00014871794871794872,
+      "loss": 0.6597,
+      "step": 290
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.44468877213339963,
+      "learning_rate": 0.00015128205128205128,
+      "loss": 0.7204,
+      "step": 295
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.2164584239122828,
+      "learning_rate": 0.00015384615384615385,
+      "loss": 0.8244,
+      "step": 300
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.1938136691775617,
+      "learning_rate": 0.00015641025641025642,
+      "loss": 0.6991,
+      "step": 305
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.1823678116608096,
+      "learning_rate": 0.00015897435897435896,
+      "loss": 0.6541,
+      "step": 310
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.1629985226358383,
+      "learning_rate": 0.00016153846153846155,
+      "loss": 0.7331,
+      "step": 315
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.19599880954946838,
+      "learning_rate": 0.0001641025641025641,
+      "loss": 0.6446,
+      "step": 320
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.23282506824657653,
+      "learning_rate": 0.0001666666666666667,
+      "loss": 0.6723,
+      "step": 325
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.20570313944345442,
+      "learning_rate": 0.00016923076923076923,
+      "loss": 0.7924,
+      "step": 330
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.18124135373904873,
+      "learning_rate": 0.0001717948717948718,
+      "loss": 0.8191,
+      "step": 335
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.18929176365178993,
+      "learning_rate": 0.00017435897435897436,
+      "loss": 0.7289,
+      "step": 340
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.19977748525194552,
+      "learning_rate": 0.00017692307692307693,
+      "loss": 0.7362,
+      "step": 345
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.1990441990377106,
+      "learning_rate": 0.0001794871794871795,
+      "loss": 0.853,
+      "step": 350
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.3087674819281357,
+      "learning_rate": 0.00018205128205128207,
+      "loss": 0.6986,
+      "step": 355
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.25912532033253527,
+      "learning_rate": 0.00018461538461538463,
+      "loss": 0.6727,
+      "step": 360
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.30170168244659684,
+      "learning_rate": 0.0001871794871794872,
+      "loss": 0.7641,
+      "step": 365
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.23159332627603768,
+      "learning_rate": 0.00018974358974358974,
+      "loss": 0.7467,
+      "step": 370
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.21437363907196583,
+      "learning_rate": 0.00019230769230769233,
+      "loss": 0.689,
+      "step": 375
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.18673889188699844,
+      "learning_rate": 0.00019487179487179487,
+      "loss": 0.625,
+      "step": 380
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.24453425122203856,
+      "learning_rate": 0.00019743589743589744,
+      "loss": 0.6991,
+      "step": 385
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.19852275666733094,
+      "learning_rate": 0.0002,
+      "loss": 0.8059,
+      "step": 390
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.183724259914749,
+      "learning_rate": 0.00019999899748734544,
+      "loss": 0.7681,
+      "step": 395
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.2127551390821211,
+      "learning_rate": 0.00019999598996948235,
+      "loss": 0.6547,
+      "step": 400
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.2684683587075289,
+      "learning_rate": 0.00019999097750671223,
+      "loss": 0.7287,
+      "step": 405
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.2283900397579642,
+      "learning_rate": 0.00019998396019953624,
+      "loss": 0.6885,
+      "step": 410
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.24702860686198994,
+      "learning_rate": 0.00019997493818865318,
+      "loss": 0.7212,
+      "step": 415
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.241475192547802,
+      "learning_rate": 0.0001999639116549566,
+      "loss": 0.5949,
+      "step": 420
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.23533430796243504,
+      "learning_rate": 0.00019995088081953136,
+      "loss": 0.7203,
+      "step": 425
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.24956850711444106,
+      "learning_rate": 0.00019993584594364894,
+      "loss": 0.6447,
+      "step": 430
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.20131118173728133,
+      "learning_rate": 0.00019991880732876246,
+      "loss": 0.699,
+      "step": 435
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.2067287885265624,
+      "learning_rate": 0.0001998997653165004,
+      "loss": 0.74,
+      "step": 440
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.20357844001623,
+      "learning_rate": 0.00019987872028866003,
+      "loss": 0.8027,
+      "step": 445
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.23984712872623487,
+      "learning_rate": 0.00019985567266719934,
+      "loss": 0.7839,
+      "step": 450
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.1897862238822482,
+      "learning_rate": 0.00019983062291422908,
+      "loss": 0.7433,
+      "step": 455
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.24095134724957823,
+      "learning_rate": 0.00019980357153200315,
+      "loss": 0.7096,
+      "step": 460
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.2852750359187981,
+      "learning_rate": 0.00019977451906290854,
+      "loss": 0.7163,
+      "step": 465
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.20980919665053135,
+      "learning_rate": 0.00019974346608945466,
+      "loss": 0.7439,
+      "step": 470
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.2295278218977822,
+      "learning_rate": 0.0001997104132342614,
+      "loss": 0.6644,
+      "step": 475
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.19640395085838386,
+      "learning_rate": 0.00019967536116004698,
+      "loss": 0.8113,
+      "step": 480
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.23742262528964145,
+      "learning_rate": 0.00019963831056961433,
+      "loss": 0.6803,
+      "step": 485
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.2209130794031639,
+      "learning_rate": 0.00019959926220583713,
+      "loss": 0.7189,
+      "step": 490
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.4158388376129303,
+      "learning_rate": 0.000199558216851645,
+      "loss": 0.7337,
+      "step": 495
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.21568480709969076,
+      "learning_rate": 0.00019951517533000764,
+      "loss": 0.6068,
+      "step": 500
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.23552741626776288,
+      "learning_rate": 0.00019947013850391847,
+      "loss": 0.7098,
+      "step": 505
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.2277845662969724,
+      "learning_rate": 0.00019942310727637724,
+      "loss": 0.7695,
+      "step": 510
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.24111051524751667,
+      "learning_rate": 0.000199374082590372,
+      "loss": 0.7586,
+      "step": 515
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.19134678351380507,
+      "learning_rate": 0.00019932306542886009,
+      "loss": 0.7393,
+      "step": 520
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.23589856188985894,
+      "learning_rate": 0.0001992700568147485,
+      "loss": 0.7205,
+      "step": 525
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.252896098878244,
+      "learning_rate": 0.00019921505781087334,
+      "loss": 0.7336,
+      "step": 530
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.25248409296825886,
+      "learning_rate": 0.00019915806951997862,
+      "loss": 0.7949,
+      "step": 535
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.20137551343484256,
+      "learning_rate": 0.00019909909308469398,
+      "loss": 0.7558,
+      "step": 540
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.2111631505276513,
+      "learning_rate": 0.0001990381296875118,
+      "loss": 0.6837,
+      "step": 545
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.29184203037550105,
+      "learning_rate": 0.0001989751805507637,
+      "loss": 0.8117,
+      "step": 550
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.2206173102779168,
+      "learning_rate": 0.0001989102469365958,
+      "loss": 0.8164,
+      "step": 555
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.21507931251783377,
+      "learning_rate": 0.00019884333014694345,
+      "loss": 0.7133,
+      "step": 560
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.22239957224386717,
+      "learning_rate": 0.00019877443152350527,
+      "loss": 0.7375,
+      "step": 565
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.21591708491473222,
+      "learning_rate": 0.00019870355244771607,
+      "loss": 0.6708,
+      "step": 570
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.25918641174746915,
+      "learning_rate": 0.0001986306943407193,
+      "loss": 0.7389,
+      "step": 575
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.22262329838643527,
+      "learning_rate": 0.00019855585866333835,
+      "loss": 0.7479,
+      "step": 580
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.2449638774409369,
+      "learning_rate": 0.00019847904691604757,
+      "loss": 0.7486,
+      "step": 585
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.2485277159299432,
+      "learning_rate": 0.00019840026063894193,
+      "loss": 0.6979,
+      "step": 590
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.28798789472500747,
+      "learning_rate": 0.0001983195014117062,
+      "loss": 0.7392,
+      "step": 595
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.2560345551346601,
+      "learning_rate": 0.00019823677085358335,
+      "loss": 0.6557,
+      "step": 600
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.2503029916293632,
+      "learning_rate": 0.00019815207062334197,
+      "loss": 0.684,
+      "step": 605
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.22343460564706522,
+      "learning_rate": 0.00019806540241924317,
+      "loss": 0.7469,
+      "step": 610
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.23478030369638528,
+      "learning_rate": 0.00019797676797900633,
+      "loss": 0.7651,
+      "step": 615
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.23393987623099682,
+      "learning_rate": 0.00019788616907977441,
+      "loss": 0.7356,
+      "step": 620
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.1835277686003351,
+      "learning_rate": 0.0001977936075380783,
+      "loss": 0.6319,
+      "step": 625
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.22084991905168924,
+      "learning_rate": 0.00019769908520980034,
+      "loss": 0.7451,
+      "step": 630
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.2532435897912756,
+      "learning_rate": 0.00019760260399013708,
+      "loss": 0.7217,
+      "step": 635
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.23471511654283184,
+      "learning_rate": 0.00019750416581356146,
+      "loss": 0.7114,
+      "step": 640
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.21710442836493216,
+      "learning_rate": 0.0001974037726537838,
+      "loss": 0.628,
+      "step": 645
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.267924875976266,
+      "learning_rate": 0.00019730142652371236,
+      "loss": 0.6469,
+      "step": 650
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.268474985537624,
+      "learning_rate": 0.00019719712947541295,
+      "loss": 0.6829,
+      "step": 655
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.16700886295844172,
+      "learning_rate": 0.0001970908836000678,
+      "loss": 0.6877,
+      "step": 660
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.22055090509747258,
+      "learning_rate": 0.00019698269102793358,
+      "loss": 0.6525,
+      "step": 665
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.27128839715879177,
+      "learning_rate": 0.00019687255392829877,
+      "loss": 0.6582,
+      "step": 670
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.21159901567551387,
+      "learning_rate": 0.00019676047450944008,
+      "loss": 0.5732,
+      "step": 675
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.24374137032809187,
+      "learning_rate": 0.0001966464550185782,
+      "loss": 0.7634,
+      "step": 680
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.24672679998063676,
+      "learning_rate": 0.00019653049774183282,
+      "loss": 0.5845,
+      "step": 685
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.2482930627863346,
+      "learning_rate": 0.0001964126050041767,
+      "loss": 0.7651,
+      "step": 690
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.2553369576381814,
+      "learning_rate": 0.000196292779169389,
+      "loss": 0.6712,
+      "step": 695
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.30911752230079176,
+      "learning_rate": 0.0001961710226400081,
+      "loss": 0.7166,
+      "step": 700
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.31535946743812887,
+      "learning_rate": 0.00019604733785728317,
+      "loss": 0.7117,
+      "step": 705
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.28460415510467557,
+      "learning_rate": 0.00019592172730112544,
+      "loss": 0.709,
+      "step": 710
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.26288174322385877,
+      "learning_rate": 0.00019579419349005837,
+      "loss": 0.7949,
+      "step": 715
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.24918287719559648,
+      "learning_rate": 0.00019566473898116713,
+      "loss": 0.7045,
+      "step": 720
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.26861534205040266,
+      "learning_rate": 0.00019553336637004735,
+      "loss": 0.7401,
+      "step": 725
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.20350229137993933,
+      "learning_rate": 0.0001954000782907532,
+      "loss": 0.7521,
+      "step": 730
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.253527308804326,
+      "learning_rate": 0.00019526487741574437,
+      "loss": 0.7833,
+      "step": 735
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.27106014622439245,
+      "learning_rate": 0.00019512776645583263,
+      "loss": 0.6898,
+      "step": 740
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.24534894717688374,
+      "learning_rate": 0.0001949887481601274,
+      "loss": 0.8102,
+      "step": 745
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.37673623038318094,
+      "learning_rate": 0.00019484782531598073,
+      "loss": 0.6929,
+      "step": 750
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.28044355813302363,
+      "learning_rate": 0.0001947050007489313,
+      "loss": 0.8396,
+      "step": 755
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.25883167182505523,
+      "learning_rate": 0.00019456027732264784,
+      "loss": 0.7169,
+      "step": 760
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.33212091144862455,
+      "learning_rate": 0.00019441365793887162,
+      "loss": 0.7562,
+      "step": 765
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.2265256995053789,
+      "learning_rate": 0.00019426514553735848,
+      "loss": 0.733,
+      "step": 770
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.1952769250523588,
+      "learning_rate": 0.00019411474309581958,
+      "loss": 0.6875,
+      "step": 775
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.23715930446049704,
+      "learning_rate": 0.00019396245362986197,
+      "loss": 0.7183,
+      "step": 780
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.26737163458762153,
+      "learning_rate": 0.00019380828019292798,
+      "loss": 0.6693,
+      "step": 785
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.3093664057575012,
+      "learning_rate": 0.00019365222587623405,
+      "loss": 0.6971,
+      "step": 790
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.22025157297022468,
+      "learning_rate": 0.00019349429380870873,
+      "loss": 0.7437,
+      "step": 795
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.23171097491730766,
+      "learning_rate": 0.00019333448715692995,
+      "loss": 0.7528,
+      "step": 800
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.23951739750544596,
+      "learning_rate": 0.0001931728091250615,
+      "loss": 0.7849,
+      "step": 805
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.24639640582380729,
+      "learning_rate": 0.00019300926295478884,
+      "loss": 0.7688,
+      "step": 810
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.24023205632785313,
+      "learning_rate": 0.00019284385192525405,
+      "loss": 0.6915,
+      "step": 815
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.2548147235517417,
+      "learning_rate": 0.0001926765793529902,
+      "loss": 0.7137,
+      "step": 820
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.24890950536791295,
+      "learning_rate": 0.00019250744859185468,
+      "loss": 0.7904,
+      "step": 825
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.31444435610399357,
+      "learning_rate": 0.00019233646303296205,
+      "loss": 0.708,
+      "step": 830
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.25866961126213517,
+      "learning_rate": 0.000192163626104616,
+      "loss": 0.6808,
+      "step": 835
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.2605403676294042,
+      "learning_rate": 0.00019198894127224074,
+      "loss": 0.7549,
+      "step": 840
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.36813656529214944,
+      "learning_rate": 0.00019181241203831137,
+      "loss": 0.7715,
+      "step": 845
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.30866872454744043,
+      "learning_rate": 0.0001916340419422837,
+      "loss": 0.7488,
+      "step": 850
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.26715773589614006,
+      "learning_rate": 0.00019145383456052327,
+      "loss": 0.6933,
+      "step": 855
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.2405309690369464,
+      "learning_rate": 0.00019127179350623372,
+      "loss": 0.6964,
+      "step": 860
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.34837890747063266,
+      "learning_rate": 0.00019108792242938425,
+      "loss": 0.6903,
+      "step": 865
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.28304812900111975,
+      "learning_rate": 0.0001909022250166365,
+      "loss": 0.6982,
+      "step": 870
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.2123369848990679,
+      "learning_rate": 0.00019071470499127058,
+      "loss": 0.6929,
+      "step": 875
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.2836396726336964,
+      "learning_rate": 0.00019052536611311046,
+      "loss": 0.7434,
+      "step": 880
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.2247103209214737,
+      "learning_rate": 0.0001903342121784486,
+      "loss": 0.6816,
+      "step": 885
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.3242567099018861,
+      "learning_rate": 0.00019014124701996973,
+      "loss": 0.72,
+      "step": 890
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.2339879402133849,
+      "learning_rate": 0.00018994647450667413,
+      "loss": 0.7041,
+      "step": 895
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.22751659387372616,
+      "learning_rate": 0.00018974989854379996,
+      "loss": 0.7511,
+      "step": 900
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.26549639698173727,
+      "learning_rate": 0.00018955152307274507,
+      "loss": 0.6897,
+      "step": 905
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.21860533117362732,
+      "learning_rate": 0.00018935135207098785,
+      "loss": 0.6648,
+      "step": 910
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.2533868012784127,
+      "learning_rate": 0.00018914938955200754,
+      "loss": 0.6798,
+      "step": 915
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.24616813985140945,
+      "learning_rate": 0.00018894563956520374,
+      "loss": 0.7683,
+      "step": 920
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.2668291016708883,
+      "learning_rate": 0.0001887401061958153,
+      "loss": 0.7791,
+      "step": 925
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.2556952354899014,
+      "learning_rate": 0.00018853279356483826,
+      "loss": 0.7733,
+      "step": 930
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.2824190700646869,
+      "learning_rate": 0.00018832370582894334,
+      "loss": 0.6439,
+      "step": 935
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.3193420311338137,
+      "learning_rate": 0.00018811284718039256,
+      "loss": 0.6791,
+      "step": 940
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.25742868616486125,
+      "learning_rate": 0.00018790022184695523,
+      "loss": 0.6835,
+      "step": 945
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.31407801057679396,
+      "learning_rate": 0.00018768583409182305,
+      "loss": 0.7338,
+      "step": 950
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.2268676531535162,
+      "learning_rate": 0.00018746968821352483,
+      "loss": 0.74,
+      "step": 955
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.2706082597185793,
+      "learning_rate": 0.00018725178854584007,
+      "loss": 0.7372,
+      "step": 960
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.23942119358366734,
+      "learning_rate": 0.00018703213945771229,
+      "loss": 0.7486,
+      "step": 965
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.23415626163520642,
+      "learning_rate": 0.00018681074535316125,
+      "loss": 0.8149,
+      "step": 970
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.23625273115645273,
+      "learning_rate": 0.00018658761067119484,
+      "loss": 0.6954,
+      "step": 975
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.29616413603025993,
+      "learning_rate": 0.00018636273988571991,
+      "loss": 0.6825,
+      "step": 980
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.24204431321093076,
+      "learning_rate": 0.00018613613750545255,
+      "loss": 0.6402,
+      "step": 985
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.23843536272675503,
+      "learning_rate": 0.0001859078080738279,
+      "loss": 0.7034,
+      "step": 990
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.2422607510213604,
+      "learning_rate": 0.00018567775616890888,
+      "loss": 0.7258,
+      "step": 995
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.2367697702106968,
+      "learning_rate": 0.00018544598640329432,
+      "loss": 0.7125,
+      "step": 1000
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.21770302111684567,
+      "learning_rate": 0.00018521250342402672,
+      "loss": 0.7651,
+      "step": 1005
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.19822970981192417,
+      "learning_rate": 0.00018497731191249894,
+      "loss": 0.7705,
+      "step": 1010
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.23080773465076526,
+      "learning_rate": 0.00018474041658436027,
+      "loss": 0.7721,
+      "step": 1015
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.22821163739265107,
+      "learning_rate": 0.000184501822189422,
+      "loss": 0.7324,
+      "step": 1020
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.22804940526863002,
+      "learning_rate": 0.0001842615335115621,
+      "loss": 0.7121,
+      "step": 1025
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.2703519880608756,
+      "learning_rate": 0.00018401955536862948,
+      "loss": 0.8063,
+      "step": 1030
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.27550361128710155,
+      "learning_rate": 0.00018377589261234705,
+      "loss": 0.7164,
+      "step": 1035
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.19381814425434957,
+      "learning_rate": 0.0001835305501282148,
+      "loss": 0.8273,
+      "step": 1040
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.26340620883694543,
+      "learning_rate": 0.00018328353283541158,
+      "loss": 0.7239,
+      "step": 1045
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.2620607615738064,
+      "learning_rate": 0.00018303484568669667,
+      "loss": 0.6904,
+      "step": 1050
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.15029608182232604,
+      "learning_rate": 0.00018278449366831035,
+      "loss": 0.6003,
+      "step": 1055
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.2187384668280594,
+      "learning_rate": 0.00018253248179987388,
+      "loss": 0.707,
+      "step": 1060
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.2560638257516859,
+      "learning_rate": 0.00018227881513428908,
+      "loss": 0.7375,
+      "step": 1065
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.26865842108524934,
+      "learning_rate": 0.0001820234987576368,
+      "loss": 0.7077,
+      "step": 1070
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.24178826048580734,
+      "learning_rate": 0.00018176653778907492,
+      "loss": 0.7354,
+      "step": 1075
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.27854735242372025,
+      "learning_rate": 0.00018150793738073602,
+      "loss": 0.6523,
+      "step": 1080
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.28792834941972884,
+      "learning_rate": 0.00018124770271762364,
+      "loss": 0.6969,
+      "step": 1085
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.30573100271052395,
+      "learning_rate": 0.00018098583901750867,
+      "loss": 0.7093,
+      "step": 1090
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.35336463231783916,
+      "learning_rate": 0.00018072235153082455,
+      "loss": 0.6441,
+      "step": 1095
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.23753636938939662,
+      "learning_rate": 0.00018045724554056214,
+      "loss": 0.7617,
+      "step": 1100
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.2830346076575104,
+      "learning_rate": 0.0001801905263621636,
+      "loss": 0.7642,
+      "step": 1105
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.2431998404354548,
+      "learning_rate": 0.0001799221993434159,
+      "loss": 0.7183,
+      "step": 1110
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.19371637012307855,
+      "learning_rate": 0.00017965226986434377,
+      "loss": 0.6834,
+      "step": 1115
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.371529742091536,
+      "learning_rate": 0.00017938074333710157,
+      "loss": 0.7232,
+      "step": 1120
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.26377679420092087,
+      "learning_rate": 0.00017910762520586485,
+      "loss": 0.7247,
+      "step": 1125
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.2285136389359427,
+      "learning_rate": 0.00017883292094672128,
+      "loss": 0.7205,
+      "step": 1130
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.29796654170976006,
+      "learning_rate": 0.00017855663606756078,
+      "loss": 0.6848,
+      "step": 1135
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.2046721683597502,
+      "learning_rate": 0.00017827877610796514,
+      "loss": 0.6759,
+      "step": 1140
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.3268765805116052,
+      "learning_rate": 0.00017799934663909682,
+      "loss": 0.716,
+      "step": 1145
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.285963694250622,
+      "learning_rate": 0.00017771835326358743,
+      "loss": 0.6486,
+      "step": 1150
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.2520717807255334,
+      "learning_rate": 0.00017743580161542525,
+      "loss": 0.6561,
+      "step": 1155
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.1981002914068182,
+      "learning_rate": 0.00017715169735984233,
+      "loss": 0.6599,
+      "step": 1160
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.26598319836843726,
+      "learning_rate": 0.00017686604619320093,
+      "loss": 0.7375,
+      "step": 1165
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.26235482801170973,
+      "learning_rate": 0.0001765788538428792,
+      "loss": 0.8063,
+      "step": 1170
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.3377576354866203,
+      "learning_rate": 0.00017629012606715648,
+      "loss": 0.7493,
+      "step": 1175
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.3678264436119463,
+      "learning_rate": 0.00017599986865509767,
+      "loss": 0.6989,
+      "step": 1180
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.32747709902543076,
+      "learning_rate": 0.00017570808742643746,
+      "loss": 0.7948,
+      "step": 1185
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.24912874308295901,
+      "learning_rate": 0.00017541478823146327,
+      "loss": 0.7187,
+      "step": 1190
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.2353274982649047,
+      "learning_rate": 0.00017511997695089822,
+      "loss": 0.7232,
+      "step": 1195
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.2537606981534416,
+      "learning_rate": 0.00017482365949578302,
+      "loss": 0.6607,
+      "step": 1200
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.2372984581891708,
+      "learning_rate": 0.00017452584180735774,
+      "loss": 0.732,
+      "step": 1205
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.23094752394219156,
+      "learning_rate": 0.00017422652985694237,
+      "loss": 0.7065,
+      "step": 1210
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.21971428715056102,
+      "learning_rate": 0.00017392572964581725,
+      "loss": 0.7012,
+      "step": 1215
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.21591121328899532,
+      "learning_rate": 0.00017362344720510278,
+      "loss": 0.7083,
+      "step": 1220
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.2779472292134201,
+      "learning_rate": 0.00017331968859563834,
+      "loss": 0.7777,
+      "step": 1225
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.37145422903296105,
+      "learning_rate": 0.00017301445990786102,
+      "loss": 0.7592,
+      "step": 1230
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.2838257377376314,
+      "learning_rate": 0.00017270776726168317,
+      "loss": 0.7147,
+      "step": 1235
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.24195971773139494,
+      "learning_rate": 0.00017239961680637,
+      "loss": 0.7442,
+      "step": 1240
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.2531071173762594,
+      "learning_rate": 0.00017209001472041617,
+      "loss": 0.7964,
+      "step": 1245
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.2392782350831628,
+      "learning_rate": 0.0001717789672114218,
+      "loss": 0.7626,
+      "step": 1250
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.24505477687731997,
+      "learning_rate": 0.00017146648051596822,
+      "loss": 0.7101,
+      "step": 1255
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.27488228336035475,
+      "learning_rate": 0.0001711525608994927,
+      "loss": 0.6718,
+      "step": 1260
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.1916616192257934,
+      "learning_rate": 0.00017083721465616306,
+      "loss": 0.6653,
+      "step": 1265
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.2960082493036361,
+      "learning_rate": 0.00017052044810875126,
+      "loss": 0.8172,
+      "step": 1270
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.29672233161501615,
+      "learning_rate": 0.00017020226760850677,
+      "loss": 0.695,
+      "step": 1275
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.29971905663286563,
+      "learning_rate": 0.00016988267953502913,
+      "loss": 0.6921,
+      "step": 1280
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.1972856751409119,
+      "learning_rate": 0.0001695616902961401,
+      "loss": 0.6707,
+      "step": 1285
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.24592016529147168,
+      "learning_rate": 0.00016923930632775516,
+      "loss": 0.7049,
+      "step": 1290
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.2362085501756166,
+      "learning_rate": 0.00016891553409375444,
+      "loss": 0.7222,
+      "step": 1295
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.2720995150566844,
+      "learning_rate": 0.00016859038008585326,
+      "loss": 0.7355,
+      "step": 1300
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.31383124999534473,
+      "learning_rate": 0.0001682638508234717,
+      "loss": 0.6735,
+      "step": 1305
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.19810610550474098,
+      "learning_rate": 0.0001679359528536041,
+      "loss": 0.6522,
+      "step": 1310
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.31481804318721185,
+      "learning_rate": 0.0001676066927506878,
+      "loss": 0.7391,
+      "step": 1315
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.27086869426476023,
+      "learning_rate": 0.00016727607711647114,
+      "loss": 0.6897,
+      "step": 1320
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.22378657478817104,
+      "learning_rate": 0.00016694411257988135,
+      "loss": 0.6582,
+      "step": 1325
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.2277896285249334,
+      "learning_rate": 0.00016661080579689132,
+      "loss": 0.7067,
+      "step": 1330
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.24128076526967204,
+      "learning_rate": 0.00016627616345038642,
+      "loss": 0.6632,
+      "step": 1335
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.425397202711577,
+      "learning_rate": 0.0001659401922500304,
+      "loss": 0.7114,
+      "step": 1340
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.2401567805733719,
+      "learning_rate": 0.0001656028989321309,
+      "loss": 0.6438,
+      "step": 1345
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.2153720205033084,
+      "learning_rate": 0.00016526429025950424,
+      "loss": 0.6838,
+      "step": 1350
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.22736402630025615,
+      "learning_rate": 0.00016492437302134008,
+      "loss": 0.7221,
+      "step": 1355
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.23982260160339397,
+      "learning_rate": 0.00016458315403306502,
+      "loss": 0.7908,
+      "step": 1360
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.27541369809262867,
+      "learning_rate": 0.00016424064013620615,
+      "loss": 0.6968,
+      "step": 1365
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.2658031068926381,
+      "learning_rate": 0.0001638968381982538,
+      "loss": 0.6988,
+      "step": 1370
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.22486989496520882,
+      "learning_rate": 0.0001635517551125238,
+      "loss": 0.81,
+      "step": 1375
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.30674338452959843,
+      "learning_rate": 0.0001632053977980194,
+      "loss": 0.7256,
+      "step": 1380
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.30285588055303475,
+      "learning_rate": 0.0001628577731992924,
+      "loss": 0.7188,
+      "step": 1385
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.36884764888792976,
+      "learning_rate": 0.000162508888286304,
+      "loss": 0.7081,
+      "step": 1390
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.2400495764148457,
+      "learning_rate": 0.00016215875005428499,
+      "loss": 0.7678,
+      "step": 1395
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.20873038269420108,
+      "learning_rate": 0.00016180736552359553,
+      "loss": 0.7497,
+      "step": 1400
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.2457699326637798,
+      "learning_rate": 0.0001614547417395844,
+      "loss": 0.7734,
+      "step": 1405
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.31218574475793187,
+      "learning_rate": 0.00016110088577244773,
+      "loss": 0.7162,
+      "step": 1410
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.27454082866628093,
+      "learning_rate": 0.0001607458047170872,
+      "loss": 0.745,
+      "step": 1415
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.28758887844401904,
+      "learning_rate": 0.00016038950569296785,
+      "loss": 0.6231,
+      "step": 1420
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.21604728751635988,
+      "learning_rate": 0.00016003199584397528,
+      "loss": 0.8008,
+      "step": 1425
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.23310840527224116,
+      "learning_rate": 0.00015967328233827249,
+      "loss": 0.6987,
+      "step": 1430
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.2568649534565718,
+      "learning_rate": 0.000159313372368156,
+      "loss": 0.6466,
+      "step": 1435
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.32113279474719314,
+      "learning_rate": 0.00015895227314991178,
+      "loss": 0.6838,
+      "step": 1440
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.296258785686713,
+      "learning_rate": 0.0001585899919236706,
+      "loss": 0.6935,
+      "step": 1445
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.23501470975504843,
+      "learning_rate": 0.00015822653595326275,
+      "loss": 0.7261,
+      "step": 1450
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.2783256304635712,
+      "learning_rate": 0.0001578619125260724,
+      "loss": 0.7512,
+      "step": 1455
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.268618761910228,
+      "learning_rate": 0.00015749612895289152,
+      "loss": 0.6482,
+      "step": 1460
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.21978849504732878,
+      "learning_rate": 0.00015712919256777335,
+      "loss": 0.7327,
+      "step": 1465
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.23357753091278965,
+      "learning_rate": 0.00015676111072788527,
+      "loss": 0.681,
+      "step": 1470
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.271489123229419,
+      "learning_rate": 0.0001563918908133614,
+      "loss": 0.7193,
+      "step": 1475
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.31414437765976133,
+      "learning_rate": 0.00015602154022715435,
+      "loss": 0.7501,
+      "step": 1480
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.27016203325071253,
+      "learning_rate": 0.00015565006639488722,
+      "loss": 0.6972,
+      "step": 1485
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.22532821311440346,
+      "learning_rate": 0.0001552774767647043,
+      "loss": 0.6922,
+      "step": 1490
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.27403456862921993,
+      "learning_rate": 0.000154903778807122,
+      "loss": 0.6949,
+      "step": 1495
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.27739513562892554,
+      "learning_rate": 0.0001545289800148789,
+      "loss": 0.7243,
+      "step": 1500
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.2238572112501968,
+      "learning_rate": 0.00015415308790278572,
+      "loss": 0.7353,
+      "step": 1505
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.24964811413280275,
+      "learning_rate": 0.0001537761100075744,
+      "loss": 0.7345,
+      "step": 1510
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.31756535201065944,
+      "learning_rate": 0.00015339805388774714,
+      "loss": 0.7856,
+      "step": 1515
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.23296901823074231,
+      "learning_rate": 0.00015301892712342482,
+      "loss": 0.729,
+      "step": 1520
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.2352427696844548,
+      "learning_rate": 0.00015263873731619508,
+      "loss": 0.7662,
+      "step": 1525
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.2761737699751625,
+      "learning_rate": 0.00015225749208895968,
+      "loss": 0.7204,
+      "step": 1530
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.3417520503815876,
+      "learning_rate": 0.000151875199085782,
+      "loss": 0.7737,
+      "step": 1535
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.2668391012505777,
+      "learning_rate": 0.0001514918659717335,
+      "loss": 0.7105,
+      "step": 1540
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.30187672878563326,
+      "learning_rate": 0.00015110750043274008,
+      "loss": 0.7326,
+      "step": 1545
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.2242593810622975,
+      "learning_rate": 0.00015072211017542813,
+      "loss": 0.7472,
+      "step": 1550
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.29972249141011964,
+      "learning_rate": 0.0001503357029269698,
+      "loss": 0.6271,
+      "step": 1555
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.26475447856915474,
+      "learning_rate": 0.00014994828643492827,
+      "loss": 0.7354,
+      "step": 1560
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.26450457112724596,
+      "learning_rate": 0.00014955986846710222,
+      "loss": 0.7264,
+      "step": 1565
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.3219325804352173,
+      "learning_rate": 0.00014917045681137026,
+      "loss": 0.689,
+      "step": 1570
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.26984976027685303,
+      "learning_rate": 0.00014878005927553456,
+      "loss": 0.6566,
+      "step": 1575
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.31512358131320956,
+      "learning_rate": 0.0001483886836871646,
+      "loss": 0.7099,
+      "step": 1580
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.27177868856475285,
+      "learning_rate": 0.00014799633789343994,
+      "loss": 0.7864,
+      "step": 1585
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.2905196107184957,
+      "learning_rate": 0.00014760302976099304,
+      "loss": 0.6429,
+      "step": 1590
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.2429940903609081,
+      "learning_rate": 0.00014720876717575155,
+      "loss": 0.6899,
+      "step": 1595
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.29122627651097244,
+      "learning_rate": 0.00014681355804278001,
+      "loss": 0.6956,
+      "step": 1600
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.214312815706962,
+      "learning_rate": 0.00014641741028612162,
+      "loss": 0.641,
+      "step": 1605
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.1858785757380131,
+      "learning_rate": 0.00014602033184863913,
+      "loss": 0.7175,
+      "step": 1610
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.2755921853014141,
+      "learning_rate": 0.00014562233069185572,
+      "loss": 0.8379,
+      "step": 1615
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.25002568814104187,
+      "learning_rate": 0.00014522341479579533,
+      "loss": 0.7584,
+      "step": 1620
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.257559638932229,
+      "learning_rate": 0.0001448235921588226,
+      "loss": 0.7739,
+      "step": 1625
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.20960264146716834,
+      "learning_rate": 0.00014442287079748263,
+      "loss": 0.7014,
+      "step": 1630
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.2521461373853711,
+      "learning_rate": 0.00014402125874634012,
+      "loss": 0.6859,
+      "step": 1635
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.249737216302013,
+      "learning_rate": 0.00014361876405781832,
+      "loss": 0.7109,
+      "step": 1640
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.2126565997638845,
+      "learning_rate": 0.00014321539480203764,
+      "loss": 0.7759,
+      "step": 1645
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.32895767720771935,
+      "learning_rate": 0.00014281115906665374,
+      "loss": 0.7,
+      "step": 1650
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.24271085601266817,
+      "learning_rate": 0.00014240606495669538,
+      "loss": 0.6983,
+      "step": 1655
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.3030757588062575,
+      "learning_rate": 0.00014200012059440207,
+      "loss": 0.6991,
+      "step": 1660
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.37464984671186286,
+      "learning_rate": 0.00014159333411906095,
+      "loss": 0.7526,
+      "step": 1665
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.23935406631920386,
+      "learning_rate": 0.00014118571368684383,
+      "loss": 0.7923,
+      "step": 1670
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.296146472400592,
+      "learning_rate": 0.00014077726747064353,
+      "loss": 0.6202,
+      "step": 1675
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.23565535010314678,
+      "learning_rate": 0.00014036800365991008,
+      "loss": 0.6796,
+      "step": 1680
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.2660771376040614,
+      "learning_rate": 0.00013995793046048643,
+      "loss": 0.6981,
+      "step": 1685
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.264009781126536,
+      "learning_rate": 0.00013954705609444404,
+      "loss": 0.6901,
+      "step": 1690
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.28853297225085933,
+      "learning_rate": 0.0001391353887999179,
+      "loss": 0.7664,
+      "step": 1695
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.24731691813618162,
+      "learning_rate": 0.00013872293683094152,
+      "loss": 0.6689,
+      "step": 1700
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.2594833661555708,
+      "learning_rate": 0.00013830970845728126,
+      "loss": 0.7701,
+      "step": 1705
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.20644246333852043,
+      "learning_rate": 0.00013789571196427055,
+      "loss": 0.6513,
+      "step": 1710
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.2339458014501443,
+      "learning_rate": 0.00013748095565264384,
+      "loss": 0.6494,
+      "step": 1715
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.2961333190555473,
+      "learning_rate": 0.00013706544783837022,
+      "loss": 0.7054,
+      "step": 1720
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.27743496626408143,
+      "learning_rate": 0.00013664919685248649,
+      "loss": 0.7295,
+      "step": 1725
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.27257332981945664,
+      "learning_rate": 0.00013623221104093025,
+      "loss": 0.8299,
+      "step": 1730
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.2592802724543621,
+      "learning_rate": 0.0001358144987643726,
+      "loss": 0.7277,
+      "step": 1735
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.2238450594222949,
+      "learning_rate": 0.00013539606839805036,
+      "loss": 0.7238,
+      "step": 1740
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.2735424814559735,
+      "learning_rate": 0.00013497692833159828,
+      "loss": 0.7015,
+      "step": 1745
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.2332151330743781,
+      "learning_rate": 0.00013455708696888085,
+      "loss": 0.6969,
+      "step": 1750
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.23230692211371998,
+      "learning_rate": 0.00013413655272782356,
+      "loss": 0.7266,
+      "step": 1755
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.23959034038420193,
+      "learning_rate": 0.00013371533404024438,
+      "loss": 0.7636,
+      "step": 1760
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.25479290452971226,
+      "learning_rate": 0.00013329343935168466,
+      "loss": 0.7075,
+      "step": 1765
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.23415173017919827,
+      "learning_rate": 0.00013287087712123962,
+      "loss": 0.7747,
+      "step": 1770
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.2286511684566787,
+      "learning_rate": 0.000132447655821389,
+      "loss": 0.6613,
+      "step": 1775
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.2976728303873654,
+      "learning_rate": 0.00013202378393782692,
+      "loss": 0.8001,
+      "step": 1780
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.26302357314609043,
+      "learning_rate": 0.00013159926996929192,
+      "loss": 0.6637,
+      "step": 1785
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.25180945871071514,
+      "learning_rate": 0.00013117412242739655,
+      "loss": 0.6515,
+      "step": 1790
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.22843461927288725,
+      "learning_rate": 0.00013074834983645657,
+      "loss": 0.7534,
+      "step": 1795
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.2746800608584564,
+      "learning_rate": 0.00013032196073332027,
+      "loss": 0.7448,
+      "step": 1800
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.23358914777160025,
+      "learning_rate": 0.000129894963667197,
+      "loss": 0.7462,
+      "step": 1805
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.21563694009001097,
+      "learning_rate": 0.00012946736719948607,
+      "loss": 0.7174,
+      "step": 1810
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.25466538845306436,
+      "learning_rate": 0.00012903917990360485,
+      "loss": 0.6958,
+      "step": 1815
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.23650763866157107,
+      "learning_rate": 0.000128610410364817,
+      "loss": 0.6948,
+      "step": 1820
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.32664640496556885,
+      "learning_rate": 0.00012818106718006023,
+      "loss": 0.6132,
+      "step": 1825
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.2195897657124799,
+      "learning_rate": 0.00012775115895777417,
+      "loss": 0.7536,
+      "step": 1830
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.23682325415599154,
+      "learning_rate": 0.0001273206943177274,
+      "loss": 0.6292,
+      "step": 1835
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.2708829289481542,
+      "learning_rate": 0.00012688968189084493,
+      "loss": 0.7273,
+      "step": 1840
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.26405392878632394,
+      "learning_rate": 0.000126458130319035,
+      "loss": 0.7254,
+      "step": 1845
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.2710846037228565,
+      "learning_rate": 0.00012602604825501587,
+      "loss": 0.7051,
+      "step": 1850
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.18625479643676604,
+      "learning_rate": 0.00012559344436214223,
+      "loss": 0.634,
+      "step": 1855
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.23174335299007431,
+      "learning_rate": 0.00012516032731423165,
+      "loss": 0.721,
+      "step": 1860
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.2674954048671644,
+      "learning_rate": 0.00012472670579539055,
+      "loss": 0.7039,
+      "step": 1865
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.23350249784430052,
+      "learning_rate": 0.00012429258849984014,
+      "loss": 0.7119,
+      "step": 1870
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.2307966491739631,
+      "learning_rate": 0.00012385798413174206,
+      "loss": 0.7267,
+      "step": 1875
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.2587108210616084,
+      "learning_rate": 0.00012342290140502388,
+      "loss": 0.7738,
+      "step": 1880
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.22704947603538148,
+      "learning_rate": 0.00012298734904320438,
+      "loss": 0.6771,
+      "step": 1885
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.24005242145344702,
+      "learning_rate": 0.00012255133577921868,
+      "loss": 0.7774,
+      "step": 1890
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.21236521326266408,
+      "learning_rate": 0.00012211487035524305,
+      "loss": 0.689,
+      "step": 1895
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.26548226590256707,
+      "learning_rate": 0.0001216779615225197,
+      "loss": 0.6958,
+      "step": 1900
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.28092405712871943,
+      "learning_rate": 0.00012124061804118137,
+      "loss": 0.7061,
+      "step": 1905
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.24730560435280138,
+      "learning_rate": 0.00012080284868007541,
+      "loss": 0.6964,
+      "step": 1910
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.1991516776169903,
+      "learning_rate": 0.00012036466221658847,
+      "loss": 0.72,
+      "step": 1915
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.29645767275685886,
+      "learning_rate": 0.0001199260674364699,
+      "loss": 0.7778,
+      "step": 1920
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.25070174095345915,
+      "learning_rate": 0.00011948707313365614,
+      "loss": 0.8114,
+      "step": 1925
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.27676560069954154,
+      "learning_rate": 0.00011904768811009405,
+      "loss": 0.7192,
+      "step": 1930
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.3078376640692026,
+      "learning_rate": 0.00011860792117556454,
+      "loss": 0.709,
+      "step": 1935
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.2049781937326678,
+      "learning_rate": 0.00011816778114750593,
+      "loss": 0.7672,
+      "step": 1940
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.21305453207892233,
+      "learning_rate": 0.00011772727685083724,
+      "loss": 0.7744,
+      "step": 1945
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.24046462080108547,
+      "learning_rate": 0.00011728641711778103,
+      "loss": 0.7229,
+      "step": 1950
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.21072752361715988,
+      "learning_rate": 0.00011684521078768658,
+      "loss": 0.6853,
+      "step": 1955
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.17170386164857987,
+      "learning_rate": 0.00011640366670685248,
+      "loss": 0.6348,
+      "step": 1960
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.27104402407445916,
+      "learning_rate": 0.00011596179372834924,
+      "loss": 0.8441,
+      "step": 1965
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.24600997987590698,
+      "learning_rate": 0.00011551960071184195,
+      "loss": 0.8251,
+      "step": 1970
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.22967416303582927,
+      "learning_rate": 0.00011507709652341256,
+      "loss": 0.716,
+      "step": 1975
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.24788639781264693,
+      "learning_rate": 0.00011463429003538196,
+      "loss": 0.7307,
+      "step": 1980
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.25315558172885294,
+      "learning_rate": 0.00011419119012613233,
+      "loss": 0.7196,
+      "step": 1985
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.24182617769996623,
+      "learning_rate": 0.000113747805679929,
+      "loss": 0.8012,
+      "step": 1990
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.2500833847981834,
+      "learning_rate": 0.00011330414558674234,
+      "loss": 0.7113,
+      "step": 1995
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.3651534404891642,
+      "learning_rate": 0.00011286021874206952,
+      "loss": 0.6693,
+      "step": 2000
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.3146361713006211,
+      "learning_rate": 0.00011241603404675609,
+      "loss": 0.7867,
+      "step": 2005
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.2994271238230683,
+      "learning_rate": 0.00011197160040681762,
+      "loss": 0.8188,
+      "step": 2010
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.2287392648112236,
+      "learning_rate": 0.00011152692673326111,
+      "loss": 0.7414,
+      "step": 2015
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.21561987456802226,
+      "learning_rate": 0.0001110820219419062,
+      "loss": 0.743,
+      "step": 2020
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.26286058284503494,
+      "learning_rate": 0.00011063689495320661,
+      "loss": 0.709,
+      "step": 2025
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.23831373981952758,
+      "learning_rate": 0.0001101915546920711,
+      "loss": 0.7487,
+      "step": 2030
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.4169631396437136,
+      "learning_rate": 0.00010974601008768461,
+      "loss": 0.6595,
+      "step": 2035
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.23258437702962542,
+      "learning_rate": 0.00010930027007332923,
+      "loss": 0.6659,
+      "step": 2040
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.23047360031102437,
+      "learning_rate": 0.00010885434358620507,
+      "loss": 0.7374,
+      "step": 2045
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.2586140902144001,
+      "learning_rate": 0.00010840823956725103,
+      "loss": 0.7431,
+      "step": 2050
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.2700901236204115,
+      "learning_rate": 0.00010796196696096564,
+      "loss": 0.7076,
+      "step": 2055
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.20102755645586845,
+      "learning_rate": 0.00010751553471522757,
+      "loss": 0.7332,
+      "step": 2060
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.25286502232358293,
+      "learning_rate": 0.00010706895178111634,
+      "loss": 0.7297,
+      "step": 2065
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.2664571423670369,
+      "learning_rate": 0.00010662222711273279,
+      "loss": 0.6364,
+      "step": 2070
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.22487420863698854,
+      "learning_rate": 0.00010617536966701962,
+      "loss": 0.7288,
+      "step": 2075
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.24184706211280177,
+      "learning_rate": 0.00010572838840358168,
+      "loss": 0.6334,
+      "step": 2080
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.26180250015553935,
+      "learning_rate": 0.00010528129228450645,
+      "loss": 0.6579,
+      "step": 2085
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.24611552837710818,
+      "learning_rate": 0.00010483409027418425,
+      "loss": 0.7506,
+      "step": 2090
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.22771528096688773,
+      "learning_rate": 0.00010438679133912858,
+      "loss": 0.6517,
+      "step": 2095
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.25425984413622493,
+      "learning_rate": 0.00010393940444779635,
+      "loss": 0.6997,
+      "step": 2100
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.25948687177354873,
+      "learning_rate": 0.00010349193857040787,
+      "loss": 0.7736,
+      "step": 2105
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.29503492872852033,
+      "learning_rate": 0.00010304440267876727,
+      "loss": 0.7855,
+      "step": 2110
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.21574528692226247,
+      "learning_rate": 0.00010259680574608248,
+      "loss": 0.765,
+      "step": 2115
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.2530425727116828,
+      "learning_rate": 0.00010214915674678523,
+      "loss": 0.7705,
+      "step": 2120
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.2355690254358619,
+      "learning_rate": 0.00010170146465635126,
+      "loss": 0.7785,
+      "step": 2125
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.2393846506779741,
+      "learning_rate": 0.00010125373845112034,
+      "loss": 0.7422,
+      "step": 2130
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.2262608510161545,
+      "learning_rate": 0.00010080598710811615,
+      "loss": 0.6922,
+      "step": 2135
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.3256276878966541,
+      "learning_rate": 0.00010035821960486643,
+      "loss": 0.7713,
+      "step": 2140
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.2496398007237336,
+      "learning_rate": 9.9910444919223e-05,
+      "loss": 0.7058,
+      "step": 2145
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.29531137204475016,
+      "learning_rate": 9.946267202918157e-05,
+      "loss": 0.6994,
+      "step": 2150
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.2596775947550526,
+      "learning_rate": 9.901490991270201e-05,
+      "loss": 0.6993,
+      "step": 2155
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.5102446180121213,
+      "learning_rate": 9.856716754752796e-05,
+      "loss": 0.6543,
+      "step": 2160
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.2754247927387255,
+      "learning_rate": 9.811945391100724e-05,
+      "loss": 0.7366,
+      "step": 2165
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.23454396466279584,
+      "learning_rate": 9.767177797991155e-05,
+      "loss": 0.7441,
+      "step": 2170
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.22647788058437757,
+      "learning_rate": 9.722414873025664e-05,
+      "loss": 0.6849,
+      "step": 2175
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.2210584391132073,
+      "learning_rate": 9.677657513712221e-05,
+      "loss": 0.727,
+      "step": 2180
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.24884640250686244,
+      "learning_rate": 9.632906617447212e-05,
+      "loss": 0.6939,
+      "step": 2185
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.2189510126025836,
+      "learning_rate": 9.588163081497427e-05,
+      "loss": 0.7871,
+      "step": 2190
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.25847268142498553,
+      "learning_rate": 9.543427802982095e-05,
+      "loss": 0.765,
+      "step": 2195
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.2887583084675781,
+      "learning_rate": 9.498701678854865e-05,
+      "loss": 0.6855,
+      "step": 2200
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.21832636439823108,
+      "learning_rate": 9.453985605885851e-05,
+      "loss": 0.732,
+      "step": 2205
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.25875988987504445,
+      "learning_rate": 9.409280480643628e-05,
+      "loss": 0.776,
+      "step": 2210
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.2937585368281696,
+      "learning_rate": 9.364587199477276e-05,
+      "loss": 0.7234,
+      "step": 2215
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.21327899166987224,
+      "learning_rate": 9.319906658498389e-05,
+      "loss": 0.7519,
+      "step": 2220
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.24464572895636677,
+      "learning_rate": 9.275239753563128e-05,
+      "loss": 0.7196,
+      "step": 2225
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.22386709524202114,
+      "learning_rate": 9.230587380254237e-05,
+      "loss": 0.6901,
+      "step": 2230
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.23753591042587965,
+      "learning_rate": 9.185950433863107e-05,
+      "loss": 0.6676,
+      "step": 2235
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.17549702993152236,
+      "learning_rate": 9.141329809371803e-05,
+      "loss": 0.7646,
+      "step": 2240
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.23044355517418907,
+      "learning_rate": 9.096726401435146e-05,
+      "loss": 0.7121,
+      "step": 2245
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.2037402971541335,
+      "learning_rate": 9.052141104362748e-05,
+      "loss": 0.6541,
+      "step": 2250
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.2173788729813436,
+      "learning_rate": 9.007574812101107e-05,
+      "loss": 0.7522,
+      "step": 2255
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.24289241576799026,
+      "learning_rate": 8.963028418215653e-05,
+      "loss": 0.8154,
+      "step": 2260
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.24110107744789394,
+      "learning_rate": 8.918502815872865e-05,
+      "loss": 0.7512,
+      "step": 2265
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.2505142119875894,
+      "learning_rate": 8.873998897822336e-05,
+      "loss": 0.7519,
+      "step": 2270
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.2603773293471231,
+      "learning_rate": 8.829517556378888e-05,
+      "loss": 0.6786,
+      "step": 2275
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.30552669526483317,
+      "learning_rate": 8.785059683404672e-05,
+      "loss": 0.7016,
+      "step": 2280
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.30878912939747605,
+      "learning_rate": 8.740626170291297e-05,
+      "loss": 0.7535,
+      "step": 2285
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.28441949312000736,
+      "learning_rate": 8.696217907941941e-05,
+      "loss": 0.7209,
+      "step": 2290
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.25705307392892374,
+      "learning_rate": 8.65183578675351e-05,
+      "loss": 0.7042,
+      "step": 2295
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.2575869631501832,
+      "learning_rate": 8.607480696598762e-05,
+      "loss": 0.6118,
+      "step": 2300
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.3327254376840343,
+      "learning_rate": 8.563153526808484e-05,
+      "loss": 0.7738,
+      "step": 2305
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.24348419071615154,
+      "learning_rate": 8.518855166153644e-05,
+      "loss": 0.6924,
+      "step": 2310
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.3469772616735954,
+      "learning_rate": 8.474586502827591e-05,
+      "loss": 0.7001,
+      "step": 2315
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.19770150428519379,
+      "learning_rate": 8.43034842442822e-05,
+      "loss": 0.6706,
+      "step": 2320
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.4391168084334309,
+      "learning_rate": 8.386141817940213e-05,
+      "loss": 0.701,
+      "step": 2325
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.24046204644423808,
+      "learning_rate": 8.341967569717202e-05,
+      "loss": 0.6511,
+      "step": 2330
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.24121857059524954,
+      "learning_rate": 8.297826565464057e-05,
+      "loss": 0.637,
+      "step": 2335
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.289807786292161,
+      "learning_rate": 8.253719690219079e-05,
+      "loss": 0.7298,
+      "step": 2340
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.2675093976169691,
+      "learning_rate": 8.20964782833628e-05,
+      "loss": 0.6229,
+      "step": 2345
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.32873199218989857,
+      "learning_rate": 8.165611863467644e-05,
+      "loss": 0.7254,
+      "step": 2350
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.21463882575401755,
+      "learning_rate": 8.12161267854542e-05,
+      "loss": 0.6896,
+      "step": 2355
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.20392542858991738,
+      "learning_rate": 8.077651155764387e-05,
+      "loss": 0.6524,
+      "step": 2360
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.21818270130807613,
+      "learning_rate": 8.033728176564216e-05,
+      "loss": 0.7482,
+      "step": 2365
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.3236950015797695,
+      "learning_rate": 7.98984462161175e-05,
+      "loss": 0.7463,
+      "step": 2370
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.23726714785703115,
+      "learning_rate": 7.946001370783375e-05,
+      "loss": 0.7916,
+      "step": 2375
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.2087887063125164,
+      "learning_rate": 7.902199303147363e-05,
+      "loss": 0.6875,
+      "step": 2380
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.27106485088624854,
+      "learning_rate": 7.858439296946263e-05,
+      "loss": 0.7802,
+      "step": 2385
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.2851539067095835,
+      "learning_rate": 7.814722229579264e-05,
+      "loss": 0.6769,
+      "step": 2390
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.24800513231309682,
+      "learning_rate": 7.771048977584641e-05,
+      "loss": 0.6148,
+      "step": 2395
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.30123103203277307,
+      "learning_rate": 7.727420416622144e-05,
+      "loss": 0.8299,
+      "step": 2400
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.28336701564838623,
+      "learning_rate": 7.683837421455466e-05,
+      "loss": 0.6799,
+      "step": 2405
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.24418965636330248,
+      "learning_rate": 7.640300865934687e-05,
+      "loss": 0.6924,
+      "step": 2410
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.23926492236991967,
+      "learning_rate": 7.596811622978765e-05,
+      "loss": 0.7157,
+      "step": 2415
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.2584954114418789,
+      "learning_rate": 7.553370564558032e-05,
+      "loss": 0.7652,
+      "step": 2420
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.29773986167748034,
+      "learning_rate": 7.509978561676699e-05,
+      "loss": 0.7581,
+      "step": 2425
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.2368205960863638,
+      "learning_rate": 7.46663648435541e-05,
+      "loss": 0.6941,
+      "step": 2430
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.2839556861514916,
+      "learning_rate": 7.423345201613778e-05,
+      "loss": 0.677,
+      "step": 2435
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.32721343862085367,
+      "learning_rate": 7.380105581452987e-05,
+      "loss": 0.7381,
+      "step": 2440
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.2620811211111583,
+      "learning_rate": 7.336918490838356e-05,
+      "loss": 0.7808,
+      "step": 2445
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.23444393543960068,
+      "learning_rate": 7.293784795681994e-05,
+      "loss": 0.7252,
+      "step": 2450
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.2457300140925708,
+      "learning_rate": 7.250705360825392e-05,
+      "loss": 0.6721,
+      "step": 2455
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.27505166245514767,
+      "learning_rate": 7.207681050022132e-05,
+      "loss": 0.7478,
+      "step": 2460
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.2910558326018566,
+      "learning_rate": 7.16471272592053e-05,
+      "loss": 0.7808,
+      "step": 2465
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.24862527602067405,
+      "learning_rate": 7.121801250046363e-05,
+      "loss": 0.7691,
+      "step": 2470
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.1902191441389871,
+      "learning_rate": 7.078947482785576e-05,
+      "loss": 0.7097,
+      "step": 2475
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.23472131108538954,
+      "learning_rate": 7.036152283367056e-05,
+      "loss": 0.6912,
+      "step": 2480
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.2189915658804591,
+      "learning_rate": 6.993416509845376e-05,
+      "loss": 0.7211,
+      "step": 2485
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.2603896298420689,
+      "learning_rate": 6.950741019083617e-05,
+      "loss": 0.7188,
+      "step": 2490
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.2559070868163805,
+      "learning_rate": 6.908126666736165e-05,
+      "loss": 0.6925,
+      "step": 2495
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.29850542028796345,
+      "learning_rate": 6.865574307231575e-05,
+      "loss": 0.7119,
+      "step": 2500
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.19364511158581588,
+      "learning_rate": 6.823084793755417e-05,
+      "loss": 0.685,
+      "step": 2505
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.3367238354963921,
+      "learning_rate": 6.780658978233199e-05,
+      "loss": 0.7398,
+      "step": 2510
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.19047235797749595,
+      "learning_rate": 6.73829771131325e-05,
+      "loss": 0.7311,
+      "step": 2515
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.3005689166877733,
+      "learning_rate": 6.696001842349702e-05,
+      "loss": 0.6747,
+      "step": 2520
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.31211848702866324,
+      "learning_rate": 6.653772219385427e-05,
+      "loss": 0.759,
+      "step": 2525
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.25625855783899115,
+      "learning_rate": 6.611609689135056e-05,
+      "loss": 0.7853,
+      "step": 2530
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.23106616213334621,
+      "learning_rate": 6.569515096967991e-05,
+      "loss": 0.6437,
+      "step": 2535
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.18916281943913288,
+      "learning_rate": 6.527489286891459e-05,
+      "loss": 0.5797,
+      "step": 2540
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.21350268536847394,
+      "learning_rate": 6.485533101533583e-05,
+      "loss": 0.7013,
+      "step": 2545
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.28763484217532065,
+      "learning_rate": 6.443647382126509e-05,
+      "loss": 0.7229,
+      "step": 2550
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.2488901500729856,
+      "learning_rate": 6.401832968489501e-05,
+      "loss": 0.686,
+      "step": 2555
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.20556797301856877,
+      "learning_rate": 6.360090699012145e-05,
+      "loss": 0.7435,
+      "step": 2560
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.25971455100175256,
+      "learning_rate": 6.3184214106375e-05,
+      "loss": 0.6772,
+      "step": 2565
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.2438737947426562,
+      "learning_rate": 6.27682593884535e-05,
+      "loss": 0.6351,
+      "step": 2570
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.2239080056853126,
+      "learning_rate": 6.235305117635428e-05,
+      "loss": 0.7386,
+      "step": 2575
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.2284401860369392,
+      "learning_rate": 6.193859779510712e-05,
+      "loss": 0.6939,
+      "step": 2580
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.24456271598752877,
+      "learning_rate": 6.152490755460715e-05,
+      "loss": 0.6706,
+      "step": 2585
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.22235388160369995,
+      "learning_rate": 6.111198874944845e-05,
+      "loss": 0.6903,
+      "step": 2590
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.2144202608438619,
+      "learning_rate": 6.0699849658757545e-05,
+      "loss": 0.7528,
+      "step": 2595
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.19956268282116124,
+      "learning_rate": 6.0288498546027536e-05,
+      "loss": 0.775,
+      "step": 2600
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.22191365952846825,
+      "learning_rate": 5.987794365895229e-05,
+      "loss": 0.6759,
+      "step": 2605
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.31000384496295275,
+      "learning_rate": 5.946819322926127e-05,
+      "loss": 0.6828,
+      "step": 2610
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.27838526207253456,
+      "learning_rate": 5.9059255472554195e-05,
+      "loss": 0.6906,
+      "step": 2615
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.2320736604730188,
+      "learning_rate": 5.865113858813673e-05,
+      "loss": 0.6665,
+      "step": 2620
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.282228906687919,
+      "learning_rate": 5.824385075885557e-05,
+      "loss": 0.6861,
+      "step": 2625
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.21066270814730526,
+      "learning_rate": 5.783740015093484e-05,
+      "loss": 0.7175,
+      "step": 2630
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.2665172474983474,
+      "learning_rate": 5.743179491381204e-05,
+      "loss": 0.706,
+      "step": 2635
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.2338869396395331,
+      "learning_rate": 5.702704317997492e-05,
+      "loss": 0.6729,
+      "step": 2640
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.21891639712761243,
+      "learning_rate": 5.662315306479806e-05,
+      "loss": 0.7053,
+      "step": 2645
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.22963997129503227,
+      "learning_rate": 5.6220132666380635e-05,
+      "loss": 0.624,
+      "step": 2650
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.28440467945947445,
+      "learning_rate": 5.581799006538354e-05,
+      "loss": 0.7612,
+      "step": 2655
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.29609689514958054,
+      "learning_rate": 5.541673332486773e-05,
+      "loss": 0.7444,
+      "step": 2660
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.2650460281999568,
+      "learning_rate": 5.5016370490132364e-05,
+      "loss": 0.8484,
+      "step": 2665
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.28809355156027594,
+      "learning_rate": 5.4616909588553674e-05,
+      "loss": 0.7871,
+      "step": 2670
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.24812243233821338,
+      "learning_rate": 5.4218358629423794e-05,
+      "loss": 0.7977,
+      "step": 2675
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.2920348572251461,
+      "learning_rate": 5.3820725603790346e-05,
+      "loss": 0.7394,
+      "step": 2680
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.29200632111876645,
+      "learning_rate": 5.342401848429615e-05,
+      "loss": 0.778,
+      "step": 2685
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.2432759115984862,
+      "learning_rate": 5.30282452250193e-05,
+      "loss": 0.7885,
+      "step": 2690
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.2838656924088362,
+      "learning_rate": 5.263341376131382e-05,
+      "loss": 0.7023,
+      "step": 2695
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.28056038476141837,
+      "learning_rate": 5.223953200965055e-05,
+      "loss": 0.7367,
+      "step": 2700
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.22746358276548237,
+      "learning_rate": 5.1846607867458196e-05,
+      "loss": 0.8169,
+      "step": 2705
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.2754771874933674,
+      "learning_rate": 5.145464921296537e-05,
+      "loss": 0.7762,
+      "step": 2710
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.275520392192173,
+      "learning_rate": 5.1063663905042225e-05,
+      "loss": 0.6907,
+      "step": 2715
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.2742206750054377,
+      "learning_rate": 5.067365978304315e-05,
+      "loss": 0.6962,
+      "step": 2720
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.3043303805027072,
+      "learning_rate": 5.0284644666649485e-05,
+      "loss": 0.6863,
+      "step": 2725
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.26287170160778367,
+      "learning_rate": 4.9896626355712805e-05,
+      "loss": 0.6837,
+      "step": 2730
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.23968527436607373,
+      "learning_rate": 4.95096126300984e-05,
+      "loss": 0.6932,
+      "step": 2735
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.19323389498485988,
+      "learning_rate": 4.912361124952948e-05,
+      "loss": 0.7342,
+      "step": 2740
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.2216783500769926,
+      "learning_rate": 4.873862995343139e-05,
+      "loss": 0.6647,
+      "step": 2745
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.26023716724912155,
+      "learning_rate": 4.835467646077656e-05,
+      "loss": 0.7357,
+      "step": 2750
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.25899624407357125,
+      "learning_rate": 4.797175846992964e-05,
+      "loss": 0.6544,
+      "step": 2755
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.18644116956534368,
+      "learning_rate": 4.7589883658493296e-05,
+      "loss": 0.7359,
+      "step": 2760
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.24394521839077538,
+      "learning_rate": 4.72090596831542e-05,
+      "loss": 0.7304,
+      "step": 2765
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.3406491454673553,
+      "learning_rate": 4.682929417952939e-05,
+      "loss": 0.6508,
+      "step": 2770
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.2508477121094201,
+      "learning_rate": 4.645059476201333e-05,
+      "loss": 0.6859,
+      "step": 2775
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.20268926784051644,
+      "learning_rate": 4.6072969023625165e-05,
+      "loss": 0.7103,
+      "step": 2780
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.24062467799728976,
+      "learning_rate": 4.5696424535856574e-05,
+      "loss": 0.705,
+      "step": 2785
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.26693143779686646,
+      "learning_rate": 4.532096884851978e-05,
+      "loss": 0.7319,
+      "step": 2790
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.23409445770349235,
+      "learning_rate": 4.494660948959645e-05,
+      "loss": 0.791,
+      "step": 2795
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.2845933689629394,
+      "learning_rate": 4.457335396508631e-05,
+      "loss": 0.6062,
+      "step": 2800
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.24119287628680353,
+      "learning_rate": 4.420120975885723e-05,
+      "loss": 0.7077,
+      "step": 2805
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.21991354406483815,
+      "learning_rate": 4.383018433249464e-05,
+      "loss": 0.6805,
+      "step": 2810
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.2475445630279962,
+      "learning_rate": 4.346028512515232e-05,
+      "loss": 0.593,
+      "step": 2815
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.24157860899387687,
+      "learning_rate": 4.309151955340297e-05,
+      "loss": 0.7195,
+      "step": 2820
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.2123078817060314,
+      "learning_rate": 4.272389501108962e-05,
+      "loss": 0.7566,
+      "step": 2825
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.33254744474875086,
+      "learning_rate": 4.2357418869177354e-05,
+      "loss": 0.7517,
+      "step": 2830
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.2450146085609281,
+      "learning_rate": 4.1992098475605654e-05,
+      "loss": 0.7792,
+      "step": 2835
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.2704781574806764,
+      "learning_rate": 4.162794115514078e-05,
+      "loss": 0.7238,
+      "step": 2840
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.37954117128340364,
+      "learning_rate": 4.1264954209229254e-05,
+      "loss": 0.6556,
+      "step": 2845
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.2860982586980259,
+      "learning_rate": 4.0903144915851174e-05,
+      "loss": 0.6506,
+      "step": 2850
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.20667223525821882,
+      "learning_rate": 4.054252052937444e-05,
+      "loss": 0.6182,
+      "step": 2855
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.2634554788563136,
+      "learning_rate": 4.018308828040924e-05,
+      "loss": 0.727,
+      "step": 2860
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.2786956847780988,
+      "learning_rate": 3.982485537566321e-05,
+      "loss": 0.7262,
+      "step": 2865
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.18437896768984685,
+      "learning_rate": 3.946782899779667e-05,
+      "loss": 0.6169,
+      "step": 2870
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.2622931367541885,
+      "learning_rate": 3.911201630527894e-05,
+      "loss": 0.7385,
+      "step": 2875
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.2748261196950285,
+      "learning_rate": 3.875742443224451e-05,
+      "loss": 0.6136,
+      "step": 2880
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.28744844039364625,
+      "learning_rate": 3.84040604883502e-05,
+      "loss": 0.7284,
+      "step": 2885
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.21857313698448572,
+      "learning_rate": 3.805193155863247e-05,
+      "loss": 0.6383,
+      "step": 2890
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.2128180862906637,
+      "learning_rate": 3.770104470336555e-05,
+      "loss": 0.7173,
+      "step": 2895
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.2130466930744276,
+      "learning_rate": 3.7351406957919636e-05,
+      "loss": 0.7266,
+      "step": 2900
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.30988412023187845,
+      "learning_rate": 3.700302533262007e-05,
+      "loss": 0.717,
+      "step": 2905
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.29577601576252954,
+      "learning_rate": 3.665590681260658e-05,
+      "loss": 0.6991,
+      "step": 2910
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.29759259924638304,
+      "learning_rate": 3.631005835769334e-05,
+      "loss": 0.8029,
+      "step": 2915
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.22770941704354283,
+      "learning_rate": 3.59654869022294e-05,
+      "loss": 0.7255,
+      "step": 2920
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.2245089532543722,
+      "learning_rate": 3.5622199354959677e-05,
+      "loss": 0.8236,
+      "step": 2925
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.209100399963671,
+      "learning_rate": 3.5280202598886324e-05,
+      "loss": 0.6472,
+      "step": 2930
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.22771150718592803,
+      "learning_rate": 3.4939503491130934e-05,
+      "loss": 0.6684,
+      "step": 2935
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.2617878485718964,
+      "learning_rate": 3.4600108862796796e-05,
+      "loss": 0.7271,
+      "step": 2940
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.2674851467564555,
+      "learning_rate": 3.426202551883213e-05,
+      "loss": 0.7587,
+      "step": 2945
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.2200031558966552,
+      "learning_rate": 3.392526023789349e-05,
+      "loss": 0.7258,
+      "step": 2950
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.21417253713351433,
+      "learning_rate": 3.358981977221006e-05,
+      "loss": 0.6981,
+      "step": 2955
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.2217852576193653,
+      "learning_rate": 3.325571084744803e-05,
+      "loss": 0.6972,
+      "step": 2960
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.2984053822842527,
+      "learning_rate": 3.292294016257598e-05,
+      "loss": 0.6919,
+      "step": 2965
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.25032718638928686,
+      "learning_rate": 3.259151438973024e-05,
+      "loss": 0.7997,
+      "step": 2970
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.26398149994753983,
+      "learning_rate": 3.2261440174081525e-05,
+      "loss": 0.7703,
+      "step": 2975
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.2088495888581316,
+      "learning_rate": 3.1932724133701344e-05,
+      "loss": 0.5416,
+      "step": 2980
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.2886491177936703,
+      "learning_rate": 3.160537285942956e-05,
+      "loss": 0.6853,
+      "step": 2985
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.2594733197020413,
+      "learning_rate": 3.1279392914742046e-05,
+      "loss": 0.7261,
+      "step": 2990
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.3017152993780796,
+      "learning_rate": 3.0954790835619195e-05,
+      "loss": 0.6988,
+      "step": 2995
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.2613281888272547,
+      "learning_rate": 3.06315731304148e-05,
+      "loss": 0.7953,
+      "step": 3000
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.3030649141190901,
+      "learning_rate": 3.0309746279725748e-05,
+      "loss": 0.6728,
+      "step": 3005
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.2049675658528854,
+      "learning_rate": 2.998931673626175e-05,
+      "loss": 0.6939,
+      "step": 3010
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.23405886262502706,
+      "learning_rate": 2.9670290924716337e-05,
+      "loss": 0.7633,
+      "step": 3015
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.16410838860875954,
+      "learning_rate": 2.935267524163774e-05,
+      "loss": 0.7258,
+      "step": 3020
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.24028216791671228,
+      "learning_rate": 2.9036476055300765e-05,
+      "loss": 0.759,
+      "step": 3025
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.2512891253559569,
+      "learning_rate": 2.872169970557913e-05,
+      "loss": 0.7622,
+      "step": 3030
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.24326580414626328,
+      "learning_rate": 2.8408352503818357e-05,
+      "loss": 0.7471,
+      "step": 3035
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.30684996416868315,
+      "learning_rate": 2.8096440732709083e-05,
+      "loss": 0.7234,
+      "step": 3040
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.2610371672143647,
+      "learning_rate": 2.778597064616133e-05,
+      "loss": 0.7114,
+      "step": 3045
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.28649543075043066,
+      "learning_rate": 2.7476948469178887e-05,
+      "loss": 0.6914,
+      "step": 3050
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.24940516209657004,
+      "learning_rate": 2.716938039773459e-05,
+      "loss": 0.754,
+      "step": 3055
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.31813491613101524,
+      "learning_rate": 2.6863272598646106e-05,
+      "loss": 0.6058,
+      "step": 3060
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.21911314508163085,
+      "learning_rate": 2.6558631209452323e-05,
+      "loss": 0.7083,
+      "step": 3065
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.24723577107070233,
+      "learning_rate": 2.625546233829016e-05,
+      "loss": 0.6785,
+      "step": 3070
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.26215344220728737,
+      "learning_rate": 2.5953772063772252e-05,
+      "loss": 0.7434,
+      "step": 3075
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.2182580130319897,
+      "learning_rate": 2.5653566434864928e-05,
+      "loss": 0.7029,
+      "step": 3080
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.2589107810453318,
+      "learning_rate": 2.5354851470767037e-05,
+      "loss": 0.702,
+      "step": 3085
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.25236658924095434,
+      "learning_rate": 2.5057633160789184e-05,
+      "loss": 0.7096,
+      "step": 3090
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.2751653091880903,
+      "learning_rate": 2.476191746423373e-05,
+      "loss": 0.7791,
+      "step": 3095
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.27791939731022475,
+      "learning_rate": 2.446771031027527e-05,
+      "loss": 0.7008,
+      "step": 3100
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.2337931562400226,
+      "learning_rate": 2.4175017597841666e-05,
+      "loss": 0.7951,
+      "step": 3105
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.23060409799544296,
+      "learning_rate": 2.3883845195495878e-05,
+      "loss": 0.7475,
+      "step": 3110
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.19515481997034956,
+      "learning_rate": 2.3594198941318248e-05,
+      "loss": 0.677,
+      "step": 3115
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.21013466271107667,
+      "learning_rate": 2.330608464278953e-05,
+      "loss": 0.7343,
+      "step": 3120
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.2655018550539882,
+      "learning_rate": 2.3019508076674267e-05,
+      "loss": 0.6745,
+      "step": 3125
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.22554453658559034,
+      "learning_rate": 2.273447498890521e-05,
+      "loss": 0.8374,
+      "step": 3130
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.290174375117781,
+      "learning_rate": 2.2450991094467865e-05,
+      "loss": 0.668,
+      "step": 3135
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.22779449704015578,
+      "learning_rate": 2.2169062077286075e-05,
+      "loss": 0.6899,
+      "step": 3140
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.25494436011601035,
+      "learning_rate": 2.1888693590107946e-05,
+      "loss": 0.7013,
+      "step": 3145
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.14266544623444502,
+      "learning_rate": 2.1609891254392678e-05,
+      "loss": 0.6311,
+      "step": 3150
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.23298437086921395,
+      "learning_rate": 2.1332660660197602e-05,
+      "loss": 0.6729,
+      "step": 3155
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.30998975291365666,
+      "learning_rate": 2.1057007366066373e-05,
+      "loss": 0.777,
+      "step": 3160
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.24570574953745566,
+      "learning_rate": 2.0782936898917204e-05,
+      "loss": 0.6747,
+      "step": 3165
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.20435326171988946,
+      "learning_rate": 2.0510454753932395e-05,
+      "loss": 0.6928,
+      "step": 3170
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.24135761451835352,
+      "learning_rate": 2.0239566394447874e-05,
+      "loss": 0.7221,
+      "step": 3175
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.21829756764053368,
+      "learning_rate": 1.9970277251843862e-05,
+      "loss": 0.7179,
+      "step": 3180
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.19424951966772022,
+      "learning_rate": 1.9702592725435786e-05,
+      "loss": 0.7273,
+      "step": 3185
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.234887739233565,
+      "learning_rate": 1.9436518182366158e-05,
+      "loss": 0.7391,
+      "step": 3190
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.26397014493832816,
+      "learning_rate": 1.9172058957496876e-05,
+      "loss": 0.7456,
+      "step": 3195
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.19484126439669286,
+      "learning_rate": 1.8909220353302392e-05,
+      "loss": 0.6979,
+      "step": 3200
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.3263727199356017,
+      "learning_rate": 1.8648007639763176e-05,
+      "loss": 0.7104,
+      "step": 3205
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.2490196028039222,
+      "learning_rate": 1.838842605426031e-05,
+      "loss": 0.6245,
+      "step": 3210
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.22590753983736134,
+      "learning_rate": 1.813048080147025e-05,
+      "loss": 0.6477,
+      "step": 3215
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.1934318577669605,
+      "learning_rate": 1.7874177053260598e-05,
+      "loss": 0.6484,
+      "step": 3220
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.19824954761719257,
+      "learning_rate": 1.7619519948586348e-05,
+      "loss": 0.6,
+      "step": 3225
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.29596182070597254,
+      "learning_rate": 1.736651459338695e-05,
+      "loss": 0.6995,
+      "step": 3230
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.2552348994159813,
+      "learning_rate": 1.711516606048377e-05,
+      "loss": 0.7123,
+      "step": 3235
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.23821982266920058,
+      "learning_rate": 1.6865479389478545e-05,
+      "loss": 0.7231,
+      "step": 3240
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.20223828918906173,
+      "learning_rate": 1.6617459586652196e-05,
+      "loss": 0.673,
+      "step": 3245
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.21541619069004347,
+      "learning_rate": 1.6371111624864543e-05,
+      "loss": 0.6788,
+      "step": 3250
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.23132279393395067,
+      "learning_rate": 1.612644044345456e-05,
+      "loss": 0.7001,
+      "step": 3255
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.3104102493541718,
+      "learning_rate": 1.5883450948141377e-05,
+      "loss": 0.6592,
+      "step": 3260
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.19498096199332454,
+      "learning_rate": 1.564214801092585e-05,
+      "loss": 0.764,
+      "step": 3265
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.22350807889337018,
+      "learning_rate": 1.540253646999299e-05,
+      "loss": 0.7285,
+      "step": 3270
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.2530056423195899,
+      "learning_rate": 1.5164621129614787e-05,
+      "loss": 0.6813,
+      "step": 3275
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.24407289762340453,
+      "learning_rate": 1.4928406760054059e-05,
+      "loss": 0.702,
+      "step": 3280
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.27667414480460156,
+      "learning_rate": 1.4693898097468662e-05,
+      "loss": 0.7285,
+      "step": 3285
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.24681628743295808,
+      "learning_rate": 1.4461099843816684e-05,
+      "loss": 0.6954,
+      "step": 3290
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.24873679256421594,
+      "learning_rate": 1.4230016666761981e-05,
+      "loss": 0.7655,
+      "step": 3295
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.22862446264265862,
+      "learning_rate": 1.4000653199580782e-05,
+      "loss": 0.7687,
+      "step": 3300
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.3046659102738475,
+      "learning_rate": 1.3773014041068633e-05,
+      "loss": 0.7156,
+      "step": 3305
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.22407782938978063,
+      "learning_rate": 1.3547103755448287e-05,
+      "loss": 0.7858,
+      "step": 3310
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.19036374903254158,
+      "learning_rate": 1.3322926872278097e-05,
+      "loss": 0.7766,
+      "step": 3315
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.26699406565117034,
+      "learning_rate": 1.3100487886361379e-05,
+      "loss": 0.7457,
+      "step": 3320
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.22344849327126515,
+      "learning_rate": 1.2879791257656049e-05,
+      "loss": 0.7227,
+      "step": 3325
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.1919000639308728,
+      "learning_rate": 1.266084141118542e-05,
+      "loss": 0.683,
+      "step": 3330
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.2126312569959486,
+      "learning_rate": 1.2443642736949258e-05,
+      "loss": 0.702,
+      "step": 3335
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.21328716751976873,
+      "learning_rate": 1.2228199589835999e-05,
+      "loss": 0.7592,
+      "step": 3340
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.25817417734261355,
+      "learning_rate": 1.2014516289535249e-05,
+      "loss": 0.7101,
+      "step": 3345
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.24654221507857982,
+      "learning_rate": 1.1802597120451286e-05,
+      "loss": 0.7651,
+      "step": 3350
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.2729024564998873,
+      "learning_rate": 1.1592446331617068e-05,
+      "loss": 0.6934,
+      "step": 3355
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.27603586653087275,
+      "learning_rate": 1.1384068136609105e-05,
+      "loss": 0.7848,
+      "step": 3360
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.3079236794395131,
+      "learning_rate": 1.1177466713462926e-05,
+      "loss": 0.6581,
+      "step": 3365
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.31502255340495816,
+      "learning_rate": 1.0972646204589377e-05,
+      "loss": 0.6583,
+      "step": 3370
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.22084726894868312,
+      "learning_rate": 1.076961071669148e-05,
+      "loss": 0.7294,
+      "step": 3375
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.2472008055830039,
+      "learning_rate": 1.0568364320682178e-05,
+      "loss": 0.7834,
+      "step": 3380
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.2277343730480719,
+      "learning_rate": 1.0368911051602636e-05,
+      "loss": 0.66,
+      "step": 3385
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.2415285274184652,
+      "learning_rate": 1.0171254908541372e-05,
+      "loss": 0.6748,
+      "step": 3390
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.2861895095719399,
+      "learning_rate": 9.975399854554068e-06,
+      "loss": 0.6721,
+      "step": 3395
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.25395148971387527,
+      "learning_rate": 9.781349816584162e-06,
+      "loss": 0.7267,
+      "step": 3400
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.21070182161070172,
+      "learning_rate": 9.589108685383975e-06,
+      "loss": 0.7106,
+      "step": 3405
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.25593439329555395,
+      "learning_rate": 9.398680315436903e-06,
+      "loss": 0.6957,
+      "step": 3410
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.21752295344145742,
+      "learning_rate": 9.210068524879923e-06,
+      "loss": 0.685,
+      "step": 3415
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.2446368196142467,
+      "learning_rate": 9.023277095427173e-06,
+      "loss": 0.7026,
+      "step": 3420
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.23687207505265248,
+      "learning_rate": 8.838309772294085e-06,
+      "loss": 0.744,
+      "step": 3425
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.25129247742502786,
+      "learning_rate": 8.655170264122303e-06,
+      "loss": 0.731,
+      "step": 3430
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.22617739846256776,
+      "learning_rate": 8.473862242905339e-06,
+      "loss": 0.7936,
+      "step": 3435
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.19703347689106585,
+      "learning_rate": 8.294389343914899e-06,
+      "loss": 0.6508,
+      "step": 3440
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.2290192280255233,
+      "learning_rate": 8.11675516562802e-06,
+      "loss": 0.7213,
+      "step": 3445
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.22631388422212603,
+      "learning_rate": 7.940963269654922e-06,
+      "loss": 0.8072,
+      "step": 3450
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.23898504344416246,
+      "learning_rate": 7.767017180667645e-06,
+      "loss": 0.7186,
+      "step": 3455
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.1771416341491615,
+      "learning_rate": 7.594920386329252e-06,
+      "loss": 0.6843,
+      "step": 3460
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.32232301037200284,
+      "learning_rate": 7.4246763372240795e-06,
+      "loss": 0.6092,
+      "step": 3465
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.20983698372327894,
+      "learning_rate": 7.256288446788362e-06,
+      "loss": 0.6873,
+      "step": 3470
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.23642454930535492,
+      "learning_rate": 7.089760091241937e-06,
+      "loss": 0.7668,
+      "step": 3475
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.19342866826333685,
+      "learning_rate": 6.925094609520455e-06,
+      "loss": 0.6985,
+      "step": 3480
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.1659312022620069,
+      "learning_rate": 6.762295303208532e-06,
+      "loss": 0.6082,
+      "step": 3485
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.23314155105131054,
+      "learning_rate": 6.601365436473439e-06,
+      "loss": 0.7625,
+      "step": 3490
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.2266096665634237,
+      "learning_rate": 6.44230823599975e-06,
+      "loss": 0.7175,
+      "step": 3495
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.285619796115372,
+      "learning_rate": 6.2851268909245865e-06,
+      "loss": 0.6727,
+      "step": 3500
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.2499473433608083,
+      "learning_rate": 6.1298245527737e-06,
+      "loss": 0.7572,
+      "step": 3505
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.23560223011085285,
+      "learning_rate": 5.976404335398256e-06,
+      "loss": 0.6292,
+      "step": 3510
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.33344906781754363,
+      "learning_rate": 5.824869314912473e-06,
+      "loss": 0.7019,
+      "step": 3515
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.19032010411212455,
+      "learning_rate": 5.675222529631841e-06,
+      "loss": 0.7159,
+      "step": 3520
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.33161459892073875,
+      "learning_rate": 5.527466980012297e-06,
+      "loss": 0.7218,
+      "step": 3525
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.2327886817133746,
+      "learning_rate": 5.381605628590003e-06,
+      "loss": 0.7328,
+      "step": 3530
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.22800725916467116,
+      "learning_rate": 5.237641399921955e-06,
+      "loss": 0.6795,
+      "step": 3535
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.23265993183122352,
+      "learning_rate": 5.095577180527378e-06,
+      "loss": 0.752,
+      "step": 3540
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.21740715842870284,
+      "learning_rate": 4.9554158188298445e-06,
+      "loss": 0.7713,
+      "step": 3545
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.2575708746269016,
+      "learning_rate": 4.817160125100106e-06,
+      "loss": 0.7631,
+      "step": 3550
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.26658430974820374,
+      "learning_rate": 4.680812871399854e-06,
+      "loss": 0.7095,
+      "step": 3555
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.22923585645582018,
+      "learning_rate": 4.546376791525975e-06,
+      "loss": 0.7197,
+      "step": 3560
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.20839075381158542,
+      "learning_rate": 4.413854580955945e-06,
+      "loss": 0.717,
+      "step": 3565
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.2126973684335578,
+      "learning_rate": 4.2832488967935795e-06,
+      "loss": 0.7439,
+      "step": 3570
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.22046801778519598,
+      "learning_rate": 4.154562357715952e-06,
+      "loss": 0.7287,
+      "step": 3575
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.22198088834095264,
+      "learning_rate": 4.02779754392072e-06,
+      "loss": 0.6862,
+      "step": 3580
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.2562953238617015,
+      "learning_rate": 3.90295699707447e-06,
+      "loss": 0.705,
+      "step": 3585
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.2832586699922064,
+      "learning_rate": 3.780043220261764e-06,
+      "loss": 0.7525,
+      "step": 3590
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.23818590199709755,
+      "learning_rate": 3.659058677934957e-06,
+      "loss": 0.7018,
+      "step": 3595
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.3095559851055113,
+      "learning_rate": 3.540005795864709e-06,
+      "loss": 0.6073,
+      "step": 3600
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.1861043245337588,
+      "learning_rate": 3.422886961091476e-06,
+      "loss": 0.6862,
+      "step": 3605
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.2807414585458434,
+      "learning_rate": 3.3077045218775192e-06,
+      "loss": 0.7819,
+      "step": 3610
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.23756038931466225,
+      "learning_rate": 3.194460787659892e-06,
+      "loss": 0.7211,
+      "step": 3615
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.2375925655429517,
+      "learning_rate": 3.0831580290041184e-06,
+      "loss": 0.7611,
+      "step": 3620
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.35878418981357185,
+      "learning_rate": 2.973798477558709e-06,
+      "loss": 0.7034,
+      "step": 3625
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.18014791198441923,
+      "learning_rate": 2.8663843260103074e-06,
+      "loss": 0.6713,
+      "step": 3630
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.2066769742143976,
+      "learning_rate": 2.76091772803988e-06,
+      "loss": 0.7165,
+      "step": 3635
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.29548750089547526,
+      "learning_rate": 2.6574007982793857e-06,
+      "loss": 0.7431,
+      "step": 3640
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.276827175378835,
+      "learning_rate": 2.555835612269475e-06,
+      "loss": 0.665,
+      "step": 3645
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.21719297673422058,
+      "learning_rate": 2.456224206417812e-06,
+      "loss": 0.697,
+      "step": 3650
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.24117198209251042,
+      "learning_rate": 2.35856857795832e-06,
+      "loss": 0.7668,
+      "step": 3655
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.4419996258477192,
+      "learning_rate": 2.262870684911045e-06,
+      "loss": 0.6807,
+      "step": 3660
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.2922290003913108,
+      "learning_rate": 2.169132446042976e-06,
+      "loss": 0.6772,
+      "step": 3665
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.21578507137937,
+      "learning_rate": 2.0773557408295343e-06,
+      "loss": 0.7143,
+      "step": 3670
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.22699656079084274,
+      "learning_rate": 1.9875424094168647e-06,
+      "loss": 0.6936,
+      "step": 3675
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.24001514084329242,
+      "learning_rate": 1.8996942525850047e-06,
+      "loss": 0.6977,
+      "step": 3680
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.26514502359960324,
+      "learning_rate": 1.813813031711742e-06,
+      "loss": 0.701,
+      "step": 3685
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.23244532178437396,
+      "learning_rate": 1.7299004687372665e-06,
+      "loss": 0.6792,
+      "step": 3690
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.2517442477838529,
+      "learning_rate": 1.6479582461297217e-06,
+      "loss": 0.6781,
+      "step": 3695
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.22695769814956426,
+      "learning_rate": 1.5679880068514174e-06,
+      "loss": 0.6858,
+      "step": 3700
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.21710844280108782,
+      "learning_rate": 1.4899913543258814e-06,
+      "loss": 0.7148,
+      "step": 3705
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.2967274268521252,
+      "learning_rate": 1.4139698524057165e-06,
+      "loss": 0.781,
+      "step": 3710
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.30352069911343654,
+      "learning_rate": 1.3399250253413154e-06,
+      "loss": 0.7702,
+      "step": 3715
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.2929106725336177,
+      "learning_rate": 1.2678583577501624e-06,
+      "loss": 0.6759,
+      "step": 3720
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.29851324638490223,
+      "learning_rate": 1.1977712945872243e-06,
+      "loss": 0.7205,
+      "step": 3725
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.26915452333889833,
+      "learning_rate": 1.1296652411158182e-06,
+      "loss": 0.7057,
+      "step": 3730
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.2245310916127946,
+      "learning_rate": 1.0635415628795665e-06,
+      "loss": 0.6926,
+      "step": 3735
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.2178134240444238,
+      "learning_rate": 9.994015856749527e-07,
+      "loss": 0.7104,
+      "step": 3740
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.18742301371018372,
+      "learning_rate": 9.372465955247544e-07,
+      "loss": 0.6046,
+      "step": 3745
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.2282365317937403,
+      "learning_rate": 8.770778386522627e-07,
+      "loss": 0.7643,
+      "step": 3750
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.23411248148329084,
+      "learning_rate": 8.188965214562804e-07,
+      "loss": 0.7397,
+      "step": 3755
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.25053318378159295,
+      "learning_rate": 7.627038104869199e-07,
+      "loss": 0.6724,
+      "step": 3760
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.27403052281802975,
+      "learning_rate": 7.085008324222763e-07,
+      "loss": 0.6412,
+      "step": 3765
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.2635701878097459,
+      "learning_rate": 6.562886740457797e-07,
+      "loss": 0.7377,
+      "step": 3770
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.21017227480655776,
+      "learning_rate": 6.060683822244117e-07,
+      "loss": 0.6977,
+      "step": 3775
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.3453902794038148,
+      "learning_rate": 5.578409638877457e-07,
+      "loss": 0.5952,
+      "step": 3780
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.24555784325804006,
+      "learning_rate": 5.116073860077064e-07,
+      "loss": 0.6711,
+      "step": 3785
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.29390073833065394,
+      "learning_rate": 4.6736857557925227e-07,
+      "loss": 0.691,
+      "step": 3790
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.2782701779800116,
+      "learning_rate": 4.2512541960171294e-07,
+      "loss": 0.7768,
+      "step": 3795
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.37197636711074117,
+      "learning_rate": 3.8487876506106966e-07,
+      "loss": 0.7539,
+      "step": 3800
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.23773581142305195,
+      "learning_rate": 3.466294189129249e-07,
+      "loss": 0.6583,
+      "step": 3805
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.2623487880042545,
+      "learning_rate": 3.1037814806634815e-07,
+      "loss": 0.7085,
+      "step": 3810
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.21088747078936257,
+      "learning_rate": 2.7612567936849964e-07,
+      "loss": 0.6627,
+      "step": 3815
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.19741270913916117,
+      "learning_rate": 2.43872699590042e-07,
+      "loss": 0.7,
+      "step": 3820
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.2518577876286286,
+      "learning_rate": 2.136198554113844e-07,
+      "loss": 0.7888,
+      "step": 3825
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.25499260076898056,
+      "learning_rate": 1.8536775340970425e-07,
+      "loss": 0.7371,
+      "step": 3830
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.22301016570777385,
+      "learning_rate": 1.591169600468123e-07,
+      "loss": 0.7354,
+      "step": 3835
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.24054514645205372,
+      "learning_rate": 1.348680016577397e-07,
+      "loss": 0.7868,
+      "step": 3840
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.24969353601595792,
+      "learning_rate": 1.126213644402463e-07,
+      "loss": 0.7333,
+      "step": 3845
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.22761401425681646,
+      "learning_rate": 9.237749444505062e-08,
+      "loss": 0.6615,
+      "step": 3850
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.24426703862468954,
+      "learning_rate": 7.413679756684832e-08,
+      "loss": 0.7324,
+      "step": 3855
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.2867096795033602,
+      "learning_rate": 5.7899639536251883e-08,
+      "loss": 0.7088,
+      "step": 3860
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.26451528734107205,
+      "learning_rate": 4.366634591237428e-08,
+      "loss": 0.7354,
+      "step": 3865
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.2064960678251034,
+      "learning_rate": 3.143720207635648e-08,
+      "loss": 0.674,
+      "step": 3870
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.32104009231527136,
+      "learning_rate": 2.1212453225627482e-08,
+      "loss": 0.7268,
+      "step": 3875
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.28886214510321495,
+      "learning_rate": 1.299230436898613e-08,
+      "loss": 0.782,
+      "step": 3880
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.25342207402007316,
+      "learning_rate": 6.776920322515423e-09,
+      "loss": 0.6615,
+      "step": 3885
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.2610844053652662,
+      "learning_rate": 2.566425706218567e-09,
+      "loss": 0.6916,
+      "step": 3890
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.21290106643942017,
+      "learning_rate": 3.609049415764787e-10,
+      "loss": 0.7345,
+      "step": 3895
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": NaN,
+      "eval_runtime": 2056.4641,
+      "eval_samples_per_second": 3.371,
+      "eval_steps_per_second": 0.843,
+      "step": 3898
+    },
+    {
+      "epoch": 1.0,
+      "step": 3898,
+      "total_flos": 1.0023612803710976e+16,
+      "train_loss": 0.7175084921933248,
+      "train_runtime": 21380.3358,
+      "train_samples_per_second": 2.917,
+      "train_steps_per_second": 0.182
+    }
+  ],
+  "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
+}