diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4921 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 28.571428571428573,
+  "eval_steps": 500,
+  "global_step": 7000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.530505657196045,
+      "learning_rate": 0.0002,
+      "loss": 2.7431,
+      "step": 10
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.857336163520813,
+      "learning_rate": 0.0002,
+      "loss": 1.8606,
+      "step": 20
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.33056318759918213,
+      "learning_rate": 0.0002,
+      "loss": 1.6893,
+      "step": 30
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.2524290680885315,
+      "learning_rate": 0.0002,
+      "loss": 1.6122,
+      "step": 40
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.2743304669857025,
+      "learning_rate": 0.0002,
+      "loss": 1.586,
+      "step": 50
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.23376622796058655,
+      "learning_rate": 0.0002,
+      "loss": 1.5535,
+      "step": 60
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.25336822867393494,
+      "learning_rate": 0.0002,
+      "loss": 1.5735,
+      "step": 70
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.2602108418941498,
+      "learning_rate": 0.0002,
+      "loss": 1.5429,
+      "step": 80
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.26768583059310913,
+      "learning_rate": 0.0002,
+      "loss": 1.5375,
+      "step": 90
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.25630825757980347,
+      "learning_rate": 0.0002,
+      "loss": 1.526,
+      "step": 100
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.25623977184295654,
+      "learning_rate": 0.0002,
+      "loss": 1.5298,
+      "step": 110
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.25586453080177307,
+      "learning_rate": 0.0002,
+      "loss": 1.5183,
+      "step": 120
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.23581761121749878,
+      "learning_rate": 0.0002,
+      "loss": 1.5248,
+      "step": 130
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.23642699420452118,
+      "learning_rate": 0.0002,
+      "loss": 1.5168,
+      "step": 140
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.24895624816417694,
+      "learning_rate": 0.0002,
+      "loss": 1.4975,
+      "step": 150
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.2535015940666199,
+      "learning_rate": 0.0002,
+      "loss": 1.5084,
+      "step": 160
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.23402313888072968,
+      "learning_rate": 0.0002,
+      "loss": 1.5068,
+      "step": 170
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.2852034568786621,
+      "learning_rate": 0.0002,
+      "loss": 1.5113,
+      "step": 180
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.2674921154975891,
+      "learning_rate": 0.0002,
+      "loss": 1.498,
+      "step": 190
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.23999463021755219,
+      "learning_rate": 0.0002,
+      "loss": 1.4894,
+      "step": 200
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.23633356392383575,
+      "learning_rate": 0.0002,
+      "loss": 1.4992,
+      "step": 210
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.24229061603546143,
+      "learning_rate": 0.0002,
+      "loss": 1.4827,
+      "step": 220
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.2810875177383423,
+      "learning_rate": 0.0002,
+      "loss": 1.4954,
+      "step": 230
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.2547338902950287,
+      "learning_rate": 0.0002,
+      "loss": 1.4881,
+      "step": 240
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.2462625801563263,
+      "learning_rate": 0.0002,
+      "loss": 1.4704,
+      "step": 250
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.2700079679489136,
+      "learning_rate": 0.0002,
+      "loss": 1.4527,
+      "step": 260
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.26295939087867737,
+      "learning_rate": 0.0002,
+      "loss": 1.4482,
+      "step": 270
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.26326537132263184,
+      "learning_rate": 0.0002,
+      "loss": 1.4453,
+      "step": 280
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.28642284870147705,
+      "learning_rate": 0.0002,
+      "loss": 1.4439,
+      "step": 290
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.26639145612716675,
+      "learning_rate": 0.0002,
+      "loss": 1.4398,
+      "step": 300
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.2638935446739197,
+      "learning_rate": 0.0002,
+      "loss": 1.4545,
+      "step": 310
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.2797674238681793,
+      "learning_rate": 0.0002,
+      "loss": 1.4487,
+      "step": 320
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.2942632734775543,
+      "learning_rate": 0.0002,
+      "loss": 1.4441,
+      "step": 330
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.2636421024799347,
+      "learning_rate": 0.0002,
+      "loss": 1.4401,
+      "step": 340
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.2771311402320862,
+      "learning_rate": 0.0002,
+      "loss": 1.4477,
+      "step": 350
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.26533156633377075,
+      "learning_rate": 0.0002,
+      "loss": 1.4404,
+      "step": 360
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.27516719698905945,
+      "learning_rate": 0.0002,
+      "loss": 1.4409,
+      "step": 370
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.2677710950374603,
+      "learning_rate": 0.0002,
+      "loss": 1.4414,
+      "step": 380
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.30368202924728394,
+      "learning_rate": 0.0002,
+      "loss": 1.4334,
+      "step": 390
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.2915685176849365,
+      "learning_rate": 0.0002,
+      "loss": 1.4434,
+      "step": 400
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.2796754539012909,
+      "learning_rate": 0.0002,
+      "loss": 1.4368,
+      "step": 410
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.2571493983268738,
+      "learning_rate": 0.0002,
+      "loss": 1.4388,
+      "step": 420
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.2667161226272583,
+      "learning_rate": 0.0002,
+      "loss": 1.4286,
+      "step": 430
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.26809781789779663,
+      "learning_rate": 0.0002,
+      "loss": 1.435,
+      "step": 440
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.27343207597732544,
+      "learning_rate": 0.0002,
+      "loss": 1.4568,
+      "step": 450
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.30309197306632996,
+      "learning_rate": 0.0002,
+      "loss": 1.4386,
+      "step": 460
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.2669704556465149,
+      "learning_rate": 0.0002,
+      "loss": 1.4459,
+      "step": 470
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.2722471058368683,
+      "learning_rate": 0.0002,
+      "loss": 1.4372,
+      "step": 480
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.2800472676753998,
+      "learning_rate": 0.0002,
+      "loss": 1.4334,
+      "step": 490
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.2805831730365753,
+      "learning_rate": 0.0002,
+      "loss": 1.3786,
+      "step": 500
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.2959790527820587,
+      "learning_rate": 0.0002,
+      "loss": 1.387,
+      "step": 510
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.2905080318450928,
+      "learning_rate": 0.0002,
+      "loss": 1.3776,
+      "step": 520
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.29701876640319824,
+      "learning_rate": 0.0002,
+      "loss": 1.3894,
+      "step": 530
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.28363561630249023,
+      "learning_rate": 0.0002,
+      "loss": 1.388,
+      "step": 540
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.30612465739250183,
+      "learning_rate": 0.0002,
+      "loss": 1.3889,
+      "step": 550
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.3003120422363281,
+      "learning_rate": 0.0002,
+      "loss": 1.383,
+      "step": 560
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.3208465874195099,
+      "learning_rate": 0.0002,
+      "loss": 1.3983,
+      "step": 570
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.29617467522621155,
+      "learning_rate": 0.0002,
+      "loss": 1.3922,
+      "step": 580
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.3164435029029846,
+      "learning_rate": 0.0002,
+      "loss": 1.3845,
+      "step": 590
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.30847829580307007,
+      "learning_rate": 0.0002,
+      "loss": 1.3897,
+      "step": 600
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.29763272404670715,
+      "learning_rate": 0.0002,
+      "loss": 1.4008,
+      "step": 610
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.29194703698158264,
+      "learning_rate": 0.0002,
+      "loss": 1.3868,
+      "step": 620
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.300700843334198,
+      "learning_rate": 0.0002,
+      "loss": 1.3937,
+      "step": 630
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.307488352060318,
+      "learning_rate": 0.0002,
+      "loss": 1.3837,
+      "step": 640
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.29698067903518677,
+      "learning_rate": 0.0002,
+      "loss": 1.3958,
+      "step": 650
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.3050895035266876,
+      "learning_rate": 0.0002,
+      "loss": 1.3819,
+      "step": 660
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.3051309585571289,
+      "learning_rate": 0.0002,
+      "loss": 1.3798,
+      "step": 670
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.31160038709640503,
+      "learning_rate": 0.0002,
+      "loss": 1.3905,
+      "step": 680
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.28441959619522095,
+      "learning_rate": 0.0002,
+      "loss": 1.3931,
+      "step": 690
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.30637431144714355,
+      "learning_rate": 0.0002,
+      "loss": 1.3834,
+      "step": 700
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.3127947449684143,
+      "learning_rate": 0.0002,
+      "loss": 1.39,
+      "step": 710
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.29069823026657104,
+      "learning_rate": 0.0002,
+      "loss": 1.3918,
+      "step": 720
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.2803474962711334,
+      "learning_rate": 0.0002,
+      "loss": 1.3854,
+      "step": 730
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 0.30701911449432373,
+      "learning_rate": 0.0002,
+      "loss": 1.3639,
+      "step": 740
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 0.32369452714920044,
+      "learning_rate": 0.0002,
+      "loss": 1.3234,
+      "step": 750
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 0.32104864716529846,
+      "learning_rate": 0.0002,
+      "loss": 1.3304,
+      "step": 760
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 0.34017834067344666,
+      "learning_rate": 0.0002,
+      "loss": 1.3288,
+      "step": 770
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 0.3346630036830902,
+      "learning_rate": 0.0002,
+      "loss": 1.3336,
+      "step": 780
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 0.32203203439712524,
+      "learning_rate": 0.0002,
+      "loss": 1.3414,
+      "step": 790
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 0.32335567474365234,
+      "learning_rate": 0.0002,
+      "loss": 1.337,
+      "step": 800
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 0.3451857566833496,
+      "learning_rate": 0.0002,
+      "loss": 1.341,
+      "step": 810
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 0.3403126895427704,
+      "learning_rate": 0.0002,
+      "loss": 1.3458,
+      "step": 820
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 0.33177173137664795,
+      "learning_rate": 0.0002,
+      "loss": 1.3205,
+      "step": 830
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 0.3264249861240387,
+      "learning_rate": 0.0002,
+      "loss": 1.3406,
+      "step": 840
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 0.33892783522605896,
+      "learning_rate": 0.0002,
+      "loss": 1.3425,
+      "step": 850
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 0.33876103162765503,
+      "learning_rate": 0.0002,
+      "loss": 1.3456,
+      "step": 860
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 0.3388749957084656,
+      "learning_rate": 0.0002,
+      "loss": 1.3377,
+      "step": 870
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 0.33248665928840637,
+      "learning_rate": 0.0002,
+      "loss": 1.3423,
+      "step": 880
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 0.32318586111068726,
+      "learning_rate": 0.0002,
+      "loss": 1.342,
+      "step": 890
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 0.3268294334411621,
+      "learning_rate": 0.0002,
+      "loss": 1.3366,
+      "step": 900
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 0.3288852870464325,
+      "learning_rate": 0.0002,
+      "loss": 1.3469,
+      "step": 910
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 0.3296649158000946,
+      "learning_rate": 0.0002,
+      "loss": 1.3364,
+      "step": 920
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 0.34164395928382874,
+      "learning_rate": 0.0002,
+      "loss": 1.3491,
+      "step": 930
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 0.32904016971588135,
+      "learning_rate": 0.0002,
+      "loss": 1.3442,
+      "step": 940
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 0.3377957344055176,
+      "learning_rate": 0.0002,
+      "loss": 1.3502,
+      "step": 950
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 0.3297536075115204,
+      "learning_rate": 0.0002,
+      "loss": 1.3442,
+      "step": 960
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 0.34833455085754395,
+      "learning_rate": 0.0002,
+      "loss": 1.339,
+      "step": 970
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 0.32754483819007874,
+      "learning_rate": 0.0002,
+      "loss": 1.3565,
+      "step": 980
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 0.3560408651828766,
+      "learning_rate": 0.0002,
+      "loss": 1.2724,
+      "step": 990
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 0.3614589273929596,
+      "learning_rate": 0.0002,
+      "loss": 1.2782,
+      "step": 1000
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 0.34403491020202637,
+      "learning_rate": 0.0002,
+      "loss": 1.2794,
+      "step": 1010
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 0.35711953043937683,
+      "learning_rate": 0.0002,
+      "loss": 1.2898,
+      "step": 1020
+    },
+    {
+      "epoch": 4.2,
+      "grad_norm": 0.3667463958263397,
+      "learning_rate": 0.0002,
+      "loss": 1.2783,
+      "step": 1030
+    },
+    {
+      "epoch": 4.24,
+      "grad_norm": 0.35901930928230286,
+      "learning_rate": 0.0002,
+      "loss": 1.2905,
+      "step": 1040
+    },
+    {
+      "epoch": 4.29,
+      "grad_norm": 0.3650436997413635,
+      "learning_rate": 0.0002,
+      "loss": 1.2947,
+      "step": 1050
+    },
+    {
+      "epoch": 4.33,
+      "grad_norm": 0.36757028102874756,
+      "learning_rate": 0.0002,
+      "loss": 1.2811,
+      "step": 1060
+    },
+    {
+      "epoch": 4.37,
+      "grad_norm": 0.3632780611515045,
+      "learning_rate": 0.0002,
+      "loss": 1.2944,
+      "step": 1070
+    },
+    {
+      "epoch": 4.41,
+      "grad_norm": 0.3927798271179199,
+      "learning_rate": 0.0002,
+      "loss": 1.2957,
+      "step": 1080
+    },
+    {
+      "epoch": 4.45,
+      "grad_norm": 0.366366982460022,
+      "learning_rate": 0.0002,
+      "loss": 1.3086,
+      "step": 1090
+    },
+    {
+      "epoch": 4.49,
+      "grad_norm": 0.37549176812171936,
+      "learning_rate": 0.0002,
+      "loss": 1.2879,
+      "step": 1100
+    },
+    {
+      "epoch": 4.53,
+      "grad_norm": 0.3678297698497772,
+      "learning_rate": 0.0002,
+      "loss": 1.289,
+      "step": 1110
+    },
+    {
+      "epoch": 4.57,
+      "grad_norm": 0.3672493100166321,
+      "learning_rate": 0.0002,
+      "loss": 1.29,
+      "step": 1120
+    },
+    {
+      "epoch": 4.61,
+      "grad_norm": 0.3762926161289215,
+      "learning_rate": 0.0002,
+      "loss": 1.3001,
+      "step": 1130
+    },
+    {
+      "epoch": 4.65,
+      "grad_norm": 0.36861684918403625,
+      "learning_rate": 0.0002,
+      "loss": 1.2947,
+      "step": 1140
+    },
+    {
+      "epoch": 4.69,
+      "grad_norm": 0.3841796815395355,
+      "learning_rate": 0.0002,
+      "loss": 1.2968,
+      "step": 1150
+    },
+    {
+      "epoch": 4.73,
+      "grad_norm": 0.37184152007102966,
+      "learning_rate": 0.0002,
+      "loss": 1.297,
+      "step": 1160
+    },
+    {
+      "epoch": 4.78,
+      "grad_norm": 0.3747577369213104,
+      "learning_rate": 0.0002,
+      "loss": 1.2922,
+      "step": 1170
+    },
+    {
+      "epoch": 4.82,
+      "grad_norm": 0.36549532413482666,
+      "learning_rate": 0.0002,
+      "loss": 1.2963,
+      "step": 1180
+    },
+    {
+      "epoch": 4.86,
+      "grad_norm": 0.3813631534576416,
+      "learning_rate": 0.0002,
+      "loss": 1.313,
+      "step": 1190
+    },
+    {
+      "epoch": 4.9,
+      "grad_norm": 0.3664815127849579,
+      "learning_rate": 0.0002,
+      "loss": 1.2999,
+      "step": 1200
+    },
+    {
+      "epoch": 4.94,
+      "grad_norm": 0.38972264528274536,
+      "learning_rate": 0.0002,
+      "loss": 1.3066,
+      "step": 1210
+    },
+    {
+      "epoch": 4.98,
+      "grad_norm": 0.3710123300552368,
+      "learning_rate": 0.0002,
+      "loss": 1.3133,
+      "step": 1220
+    },
+    {
+      "epoch": 5.02,
+      "grad_norm": 0.3784914016723633,
+      "learning_rate": 0.0002,
+      "loss": 1.2625,
+      "step": 1230
+    },
+    {
+      "epoch": 5.06,
+      "grad_norm": 0.416645348072052,
+      "learning_rate": 0.0002,
+      "loss": 1.236,
+      "step": 1240
+    },
+    {
+      "epoch": 5.1,
+      "grad_norm": 0.3829210102558136,
+      "learning_rate": 0.0002,
+      "loss": 1.2314,
+      "step": 1250
+    },
+    {
+      "epoch": 5.14,
+      "grad_norm": 0.38960739970207214,
+      "learning_rate": 0.0002,
+      "loss": 1.2239,
+      "step": 1260
+    },
+    {
+      "epoch": 5.18,
+      "grad_norm": 0.3890588581562042,
+      "learning_rate": 0.0002,
+      "loss": 1.2399,
+      "step": 1270
+    },
+    {
+      "epoch": 5.22,
+      "grad_norm": 0.4003150165081024,
+      "learning_rate": 0.0002,
+      "loss": 1.2365,
+      "step": 1280
+    },
+    {
+      "epoch": 5.27,
+      "grad_norm": 0.4123793840408325,
+      "learning_rate": 0.0002,
+      "loss": 1.2429,
+      "step": 1290
+    },
+    {
+      "epoch": 5.31,
+      "grad_norm": 0.43428850173950195,
+      "learning_rate": 0.0002,
+      "loss": 1.2407,
+      "step": 1300
+    },
+    {
+      "epoch": 5.35,
+      "grad_norm": 0.4154336154460907,
+      "learning_rate": 0.0002,
+      "loss": 1.2417,
+      "step": 1310
+    },
+    {
+      "epoch": 5.39,
+      "grad_norm": 0.40663963556289673,
+      "learning_rate": 0.0002,
+      "loss": 1.2517,
+      "step": 1320
+    },
+    {
+      "epoch": 5.43,
+      "grad_norm": 0.40631529688835144,
+      "learning_rate": 0.0002,
+      "loss": 1.2617,
+      "step": 1330
+    },
+    {
+      "epoch": 5.47,
+      "grad_norm": 0.4252465069293976,
+      "learning_rate": 0.0002,
+      "loss": 1.2564,
+      "step": 1340
+    },
+    {
+      "epoch": 5.51,
+      "grad_norm": 0.4079826772212982,
+      "learning_rate": 0.0002,
+      "loss": 1.2554,
+      "step": 1350
+    },
+    {
+      "epoch": 5.55,
+      "grad_norm": 0.41097384691238403,
+      "learning_rate": 0.0002,
+      "loss": 1.2503,
+      "step": 1360
+    },
+    {
+      "epoch": 5.59,
+      "grad_norm": 0.42485907673835754,
+      "learning_rate": 0.0002,
+      "loss": 1.249,
+      "step": 1370
+    },
+    {
+      "epoch": 5.63,
+      "grad_norm": 0.4294757544994354,
+      "learning_rate": 0.0002,
+      "loss": 1.2423,
+      "step": 1380
+    },
+    {
+      "epoch": 5.67,
+      "grad_norm": 0.4221188426017761,
+      "learning_rate": 0.0002,
+      "loss": 1.25,
+      "step": 1390
+    },
+    {
+      "epoch": 5.71,
+      "grad_norm": 0.3959333002567291,
+      "learning_rate": 0.0002,
+      "loss": 1.2428,
+      "step": 1400
+    },
+    {
+      "epoch": 5.76,
+      "grad_norm": 0.3953990340232849,
+      "learning_rate": 0.0002,
+      "loss": 1.2706,
+      "step": 1410
+    },
+    {
+      "epoch": 5.8,
+      "grad_norm": 0.3927258849143982,
+      "learning_rate": 0.0002,
+      "loss": 1.2535,
+      "step": 1420
+    },
+    {
+      "epoch": 5.84,
+      "grad_norm": 0.4078606069087982,
+      "learning_rate": 0.0002,
+      "loss": 1.2764,
+      "step": 1430
+    },
+    {
+      "epoch": 5.88,
+      "grad_norm": 0.3999166786670685,
+      "learning_rate": 0.0002,
+      "loss": 1.2641,
+      "step": 1440
+    },
+    {
+      "epoch": 5.92,
+      "grad_norm": 0.39490771293640137,
+      "learning_rate": 0.0002,
+      "loss": 1.2693,
+      "step": 1450
+    },
+    {
+      "epoch": 5.96,
+      "grad_norm": 0.4049830138683319,
+      "learning_rate": 0.0002,
+      "loss": 1.2672,
+      "step": 1460
+    },
+    {
+      "epoch": 6.0,
+      "grad_norm": 0.40505218505859375,
+      "learning_rate": 0.0002,
+      "loss": 1.276,
+      "step": 1470
+    },
+    {
+      "epoch": 6.04,
+      "grad_norm": 0.45407840609550476,
+      "learning_rate": 0.0002,
+      "loss": 1.1899,
+      "step": 1480
+    },
+    {
+      "epoch": 6.08,
+      "grad_norm": 0.43418335914611816,
+      "learning_rate": 0.0002,
+      "loss": 1.185,
+      "step": 1490
+    },
+    {
+      "epoch": 6.12,
+      "grad_norm": 0.44777223467826843,
+      "learning_rate": 0.0002,
+      "loss": 1.1904,
+      "step": 1500
+    },
+    {
+      "epoch": 6.16,
+      "grad_norm": 0.45681995153427124,
+      "learning_rate": 0.0002,
+      "loss": 1.1749,
+      "step": 1510
+    },
+    {
+      "epoch": 6.2,
+      "grad_norm": 0.4477551281452179,
+      "learning_rate": 0.0002,
+      "loss": 1.1987,
+      "step": 1520
+    },
+    {
+      "epoch": 6.24,
+      "grad_norm": 0.4677373766899109,
+      "learning_rate": 0.0002,
+      "loss": 1.2036,
+      "step": 1530
+    },
+    {
+      "epoch": 6.29,
+      "grad_norm": 0.4448174238204956,
+      "learning_rate": 0.0002,
+      "loss": 1.1921,
+      "step": 1540
+    },
+    {
+      "epoch": 6.33,
+      "grad_norm": 0.47542881965637207,
+      "learning_rate": 0.0002,
+      "loss": 1.2062,
+      "step": 1550
+    },
+    {
+      "epoch": 6.37,
+      "grad_norm": 0.4679628312587738,
+      "learning_rate": 0.0002,
+      "loss": 1.1987,
+      "step": 1560
+    },
+    {
+      "epoch": 6.41,
+      "grad_norm": 0.45168885588645935,
+      "learning_rate": 0.0002,
+      "loss": 1.2101,
+      "step": 1570
+    },
+    {
+      "epoch": 6.45,
+      "grad_norm": 0.4579172730445862,
+      "learning_rate": 0.0002,
+      "loss": 1.2143,
+      "step": 1580
+    },
+    {
+      "epoch": 6.49,
+      "grad_norm": 0.44909214973449707,
+      "learning_rate": 0.0002,
+      "loss": 1.2047,
+      "step": 1590
+    },
+    {
+      "epoch": 6.53,
+      "grad_norm": 0.4445750415325165,
+      "learning_rate": 0.0002,
+      "loss": 1.2098,
+      "step": 1600
+    },
+    {
+      "epoch": 6.57,
+      "grad_norm": 0.455679327249527,
+      "learning_rate": 0.0002,
+      "loss": 1.2149,
+      "step": 1610
+    },
+    {
+      "epoch": 6.61,
+      "grad_norm": 0.44062715768814087,
+      "learning_rate": 0.0002,
+      "loss": 1.2209,
+      "step": 1620
+    },
+    {
+      "epoch": 6.65,
+      "grad_norm": 0.4541163444519043,
+      "learning_rate": 0.0002,
+      "loss": 1.2164,
+      "step": 1630
+    },
+    {
+      "epoch": 6.69,
+      "grad_norm": 0.43290677666664124,
+      "learning_rate": 0.0002,
+      "loss": 1.2092,
+      "step": 1640
+    },
+    {
+      "epoch": 6.73,
+      "grad_norm": 0.44760000705718994,
+      "learning_rate": 0.0002,
+      "loss": 1.2211,
+      "step": 1650
+    },
+    {
+      "epoch": 6.78,
+      "grad_norm": 0.4385235011577606,
+      "learning_rate": 0.0002,
+      "loss": 1.2205,
+      "step": 1660
+    },
+    {
+      "epoch": 6.82,
+      "grad_norm": 0.44804102182388306,
+      "learning_rate": 0.0002,
+      "loss": 1.2244,
+      "step": 1670
+    },
+    {
+      "epoch": 6.86,
+      "grad_norm": 0.4405447244644165,
+      "learning_rate": 0.0002,
+      "loss": 1.2388,
+      "step": 1680
+    },
+    {
+      "epoch": 6.9,
+      "grad_norm": 0.4315509796142578,
+      "learning_rate": 0.0002,
+      "loss": 1.2322,
+      "step": 1690
+    },
+    {
+      "epoch": 6.94,
+      "grad_norm": 0.44136691093444824,
+      "learning_rate": 0.0002,
+      "loss": 1.2252,
+      "step": 1700
+    },
+    {
+      "epoch": 6.98,
+      "grad_norm": 0.44583484530448914,
+      "learning_rate": 0.0002,
+      "loss": 1.2238,
+      "step": 1710
+    },
+    {
+      "epoch": 7.02,
+      "grad_norm": 0.471617728471756,
+      "learning_rate": 0.0002,
+      "loss": 1.1937,
+      "step": 1720
+    },
+    {
+      "epoch": 7.06,
+      "grad_norm": 0.477583110332489,
+      "learning_rate": 0.0002,
+      "loss": 1.1456,
+      "step": 1730
+    },
+    {
+      "epoch": 7.1,
+      "grad_norm": 0.4998668432235718,
+      "learning_rate": 0.0002,
+      "loss": 1.1457,
+      "step": 1740
+    },
+    {
+      "epoch": 7.14,
+      "grad_norm": 0.4928749203681946,
+      "learning_rate": 0.0002,
+      "loss": 1.1546,
+      "step": 1750
+    },
+    {
+      "epoch": 7.18,
+      "grad_norm": 0.49452856183052063,
+      "learning_rate": 0.0002,
+      "loss": 1.1453,
+      "step": 1760
+    },
+    {
+      "epoch": 7.22,
+      "grad_norm": 0.5097588896751404,
+      "learning_rate": 0.0002,
+      "loss": 1.1501,
+      "step": 1770
+    },
+    {
+      "epoch": 7.27,
+      "grad_norm": 0.4926673471927643,
+      "learning_rate": 0.0002,
+      "loss": 1.1606,
+      "step": 1780
+    },
+    {
+      "epoch": 7.31,
+      "grad_norm": 0.5037740468978882,
+      "learning_rate": 0.0002,
+      "loss": 1.1675,
+      "step": 1790
+    },
+    {
+      "epoch": 7.35,
+      "grad_norm": 0.4997263550758362,
+      "learning_rate": 0.0002,
+      "loss": 1.1653,
+      "step": 1800
+    },
+    {
+      "epoch": 7.39,
+      "grad_norm": 0.4845424294471741,
+      "learning_rate": 0.0002,
+      "loss": 1.1544,
+      "step": 1810
+    },
+    {
+      "epoch": 7.43,
+      "grad_norm": 0.49940812587738037,
+      "learning_rate": 0.0002,
+      "loss": 1.1812,
+      "step": 1820
+    },
+    {
+      "epoch": 7.47,
+      "grad_norm": 0.4817129373550415,
+      "learning_rate": 0.0002,
+      "loss": 1.1714,
+      "step": 1830
+    },
+    {
+      "epoch": 7.51,
+      "grad_norm": 0.4989680051803589,
+      "learning_rate": 0.0002,
+      "loss": 1.1741,
+      "step": 1840
+    },
+    {
+      "epoch": 7.55,
+      "grad_norm": 0.5034418106079102,
+      "learning_rate": 0.0002,
+      "loss": 1.1808,
+      "step": 1850
+    },
+    {
+      "epoch": 7.59,
+      "grad_norm": 0.5104469060897827,
+      "learning_rate": 0.0002,
+      "loss": 1.1773,
+      "step": 1860
+    },
+    {
+      "epoch": 7.63,
+      "grad_norm": 0.4944339394569397,
+      "learning_rate": 0.0002,
+      "loss": 1.1758,
+      "step": 1870
+    },
+    {
+      "epoch": 7.67,
+      "grad_norm": 0.4833206236362457,
+      "learning_rate": 0.0002,
+      "loss": 1.1786,
+      "step": 1880
+    },
+    {
+      "epoch": 7.71,
+      "grad_norm": 0.48878568410873413,
+      "learning_rate": 0.0002,
+      "loss": 1.176,
+      "step": 1890
+    },
+    {
+      "epoch": 7.76,
+      "grad_norm": 0.4889214038848877,
+      "learning_rate": 0.0002,
+      "loss": 1.1796,
+      "step": 1900
+    },
+    {
+      "epoch": 7.8,
+      "grad_norm": 0.49351948499679565,
+      "learning_rate": 0.0002,
+      "loss": 1.1847,
+      "step": 1910
+    },
+    {
+      "epoch": 7.84,
+      "grad_norm": 0.4790671169757843,
+      "learning_rate": 0.0002,
+      "loss": 1.184,
+      "step": 1920
+    },
+    {
+      "epoch": 7.88,
+      "grad_norm": 0.49084314703941345,
+      "learning_rate": 0.0002,
+      "loss": 1.1905,
+      "step": 1930
+    },
+    {
+      "epoch": 7.92,
+      "grad_norm": 0.49543461203575134,
+      "learning_rate": 0.0002,
+      "loss": 1.1998,
+      "step": 1940
+    },
+    {
+      "epoch": 7.96,
+      "grad_norm": 0.4767250716686249,
+      "learning_rate": 0.0002,
+      "loss": 1.1921,
+      "step": 1950
+    },
+    {
+      "epoch": 8.0,
+      "grad_norm": 0.4650532901287079,
+      "learning_rate": 0.0002,
+      "loss": 1.1992,
+      "step": 1960
+    },
+    {
+      "epoch": 8.04,
+      "grad_norm": 0.5073546767234802,
+      "learning_rate": 0.0002,
+      "loss": 1.1067,
+      "step": 1970
+    },
+    {
+      "epoch": 8.08,
+      "grad_norm": 0.5141373872756958,
+      "learning_rate": 0.0002,
+      "loss": 1.1067,
+      "step": 1980
+    },
+    {
+      "epoch": 8.12,
+      "grad_norm": 0.5466707348823547,
+      "learning_rate": 0.0002,
+      "loss": 1.106,
+      "step": 1990
+    },
+    {
+      "epoch": 8.16,
+      "grad_norm": 0.5313953161239624,
+      "learning_rate": 0.0002,
+      "loss": 1.118,
+      "step": 2000
+    },
+    {
+      "epoch": 8.2,
+      "grad_norm": 0.5327037572860718,
+      "learning_rate": 0.0002,
+      "loss": 1.1202,
+      "step": 2010
+    },
+    {
+      "epoch": 8.24,
+      "grad_norm": 0.5393937230110168,
+      "learning_rate": 0.0002,
+      "loss": 1.1195,
+      "step": 2020
+    },
+    {
+      "epoch": 8.29,
+      "grad_norm": 0.5502837896347046,
+      "learning_rate": 0.0002,
+      "loss": 1.1163,
+      "step": 2030
+    },
+    {
+      "epoch": 8.33,
+      "grad_norm": 0.5364841818809509,
+      "learning_rate": 0.0002,
+      "loss": 1.131,
+      "step": 2040
+    },
+    {
+      "epoch": 8.37,
+      "grad_norm": 0.5162956118583679,
+      "learning_rate": 0.0002,
+      "loss": 1.1349,
+      "step": 2050
+    },
+    {
+      "epoch": 8.41,
+      "grad_norm": 0.5319573283195496,
+      "learning_rate": 0.0002,
+      "loss": 1.1331,
+      "step": 2060
+    },
+    {
+      "epoch": 8.45,
+      "grad_norm": 0.5466243028640747,
+      "learning_rate": 0.0002,
+      "loss": 1.138,
+      "step": 2070
+    },
+    {
+      "epoch": 8.49,
+      "grad_norm": 0.5623811483383179,
+      "learning_rate": 0.0002,
+      "loss": 1.1358,
+      "step": 2080
+    },
+    {
+      "epoch": 8.53,
+      "grad_norm": 0.5489851832389832,
+      "learning_rate": 0.0002,
+      "loss": 1.1368,
+      "step": 2090
+    },
+    {
+      "epoch": 8.57,
+      "grad_norm": 0.5441839694976807,
+      "learning_rate": 0.0002,
+      "loss": 1.1433,
+      "step": 2100
+    },
+    {
+      "epoch": 8.61,
+      "grad_norm": 0.5348924398422241,
+      "learning_rate": 0.0002,
+      "loss": 1.1457,
+      "step": 2110
+    },
+    {
+      "epoch": 8.65,
+      "grad_norm": 0.5622116327285767,
+      "learning_rate": 0.0002,
+      "loss": 1.1498,
+      "step": 2120
+    },
+    {
+      "epoch": 8.69,
+      "grad_norm": 0.530335545539856,
+      "learning_rate": 0.0002,
+      "loss": 1.1316,
+      "step": 2130
+    },
+    {
+      "epoch": 8.73,
+      "grad_norm": 0.529975175857544,
+      "learning_rate": 0.0002,
+      "loss": 1.1485,
+      "step": 2140
+    },
+    {
+      "epoch": 8.78,
+      "grad_norm": 0.5368651151657104,
+      "learning_rate": 0.0002,
+      "loss": 1.1551,
+      "step": 2150
+    },
+    {
+      "epoch": 8.82,
+      "grad_norm": 0.534313976764679,
+      "learning_rate": 0.0002,
+      "loss": 1.1513,
+      "step": 2160
+    },
+    {
+      "epoch": 8.86,
+      "grad_norm": 0.5263510346412659,
+      "learning_rate": 0.0002,
+      "loss": 1.1538,
+      "step": 2170
+    },
+    {
+      "epoch": 8.9,
+      "grad_norm": 0.5384341478347778,
+      "learning_rate": 0.0002,
+      "loss": 1.1493,
+      "step": 2180
+    },
+    {
+      "epoch": 8.94,
+      "grad_norm": 0.5207712054252625,
+      "learning_rate": 0.0002,
+      "loss": 1.1633,
+      "step": 2190
+    },
+    {
+      "epoch": 8.98,
+      "grad_norm": 0.5430459976196289,
+      "learning_rate": 0.0002,
+      "loss": 1.159,
+      "step": 2200
+    },
+    {
+      "epoch": 9.02,
+      "grad_norm": 0.5784776210784912,
+      "learning_rate": 0.0002,
+      "loss": 1.1247,
+      "step": 2210
+    },
+    {
+      "epoch": 9.06,
+      "grad_norm": 0.544442892074585,
+      "learning_rate": 0.0002,
+      "loss": 1.0716,
+      "step": 2220
+    },
+    {
+      "epoch": 9.1,
+      "grad_norm": 0.559004545211792,
+      "learning_rate": 0.0002,
+      "loss": 1.0747,
+      "step": 2230
+    },
+    {
+      "epoch": 9.14,
+      "grad_norm": 0.5693913698196411,
+      "learning_rate": 0.0002,
+      "loss": 1.0856,
+      "step": 2240
+    },
+    {
+      "epoch": 9.18,
+      "grad_norm": 0.5680646896362305,
+      "learning_rate": 0.0002,
+      "loss": 1.071,
+      "step": 2250
+    },
+    {
+      "epoch": 9.22,
+      "grad_norm": 0.5713136196136475,
+      "learning_rate": 0.0002,
+      "loss": 1.077,
+      "step": 2260
+    },
+    {
+      "epoch": 9.27,
+      "grad_norm": 0.5708666443824768,
+      "learning_rate": 0.0002,
+      "loss": 1.09,
+      "step": 2270
+    },
+    {
+      "epoch": 9.31,
+      "grad_norm": 0.5802373290061951,
+      "learning_rate": 0.0002,
+      "loss": 1.0949,
+      "step": 2280
+    },
+    {
+      "epoch": 9.35,
+      "grad_norm": 0.5722843408584595,
+      "learning_rate": 0.0002,
+      "loss": 1.0973,
+      "step": 2290
+    },
+    {
+      "epoch": 9.39,
+      "grad_norm": 0.5759919881820679,
+      "learning_rate": 0.0002,
+      "loss": 1.0966,
+      "step": 2300
+    },
+    {
+      "epoch": 9.43,
+      "grad_norm": 0.609473466873169,
+      "learning_rate": 0.0002,
+      "loss": 1.107,
+      "step": 2310
+    },
+    {
+      "epoch": 9.47,
+      "grad_norm": 0.5732268691062927,
+      "learning_rate": 0.0002,
+      "loss": 1.1016,
+      "step": 2320
+    },
+    {
+      "epoch": 9.51,
+      "grad_norm": 0.5877196192741394,
+      "learning_rate": 0.0002,
+      "loss": 1.0994,
+      "step": 2330
+    },
+    {
+      "epoch": 9.55,
+      "grad_norm": 0.5665601491928101,
+      "learning_rate": 0.0002,
+      "loss": 1.1022,
+      "step": 2340
+    },
+    {
+      "epoch": 9.59,
+      "grad_norm": 0.5648427605628967,
+      "learning_rate": 0.0002,
+      "loss": 1.1135,
+      "step": 2350
+    },
+    {
+      "epoch": 9.63,
+      "grad_norm": 0.5821457505226135,
+      "learning_rate": 0.0002,
+      "loss": 1.1129,
+      "step": 2360
+    },
+    {
+      "epoch": 9.67,
+      "grad_norm": 0.5789903402328491,
+      "learning_rate": 0.0002,
+      "loss": 1.1108,
+      "step": 2370
+    },
+    {
+      "epoch": 9.71,
+      "grad_norm": 0.5680549144744873,
+      "learning_rate": 0.0002,
+      "loss": 1.1149,
+      "step": 2380
+    },
+    {
+      "epoch": 9.76,
+      "grad_norm": 0.5805366039276123,
+      "learning_rate": 0.0002,
+      "loss": 1.1176,
+      "step": 2390
+    },
+    {
+      "epoch": 9.8,
+      "grad_norm": 0.575652003288269,
+      "learning_rate": 0.0002,
+      "loss": 1.1175,
+      "step": 2400
+    },
+    {
+      "epoch": 9.84,
+      "grad_norm": 0.5799399614334106,
+      "learning_rate": 0.0002,
+      "loss": 1.1145,
+      "step": 2410
+    },
+    {
+      "epoch": 9.88,
+      "grad_norm": 0.5476598143577576,
+      "learning_rate": 0.0002,
+      "loss": 1.1241,
+      "step": 2420
+    },
+    {
+      "epoch": 9.92,
+      "grad_norm": 0.5581275224685669,
+      "learning_rate": 0.0002,
+      "loss": 1.1253,
+      "step": 2430
+    },
+    {
+      "epoch": 9.96,
+      "grad_norm": 0.5669559836387634,
+      "learning_rate": 0.0002,
+      "loss": 1.1261,
+      "step": 2440
+    },
+    {
+      "epoch": 10.0,
+      "grad_norm": 0.5465613603591919,
+      "learning_rate": 0.0002,
+      "loss": 1.1384,
+      "step": 2450
+    },
+    {
+      "epoch": 10.04,
+      "grad_norm": 0.6073935031890869,
+      "learning_rate": 0.0002,
+      "loss": 1.0359,
+      "step": 2460
+    },
+    {
+      "epoch": 10.08,
+      "grad_norm": 0.6463605165481567,
+      "learning_rate": 0.0002,
+      "loss": 1.0479,
+      "step": 2470
+    },
+    {
+      "epoch": 10.12,
+      "grad_norm": 0.6026530861854553,
+      "learning_rate": 0.0002,
+      "loss": 1.0404,
+      "step": 2480
+    },
+    {
+      "epoch": 10.16,
+      "grad_norm": 0.6082301139831543,
+      "learning_rate": 0.0002,
+      "loss": 1.0428,
+      "step": 2490
+    },
+    {
+      "epoch": 10.2,
+      "grad_norm": 0.612219512462616,
+      "learning_rate": 0.0002,
+      "loss": 1.0477,
+      "step": 2500
+    },
+    {
+      "epoch": 10.24,
+      "grad_norm": 0.6004548668861389,
+      "learning_rate": 0.0002,
+      "loss": 1.052,
+      "step": 2510
+    },
+    {
+      "epoch": 10.29,
+      "grad_norm": 0.6319270133972168,
+      "learning_rate": 0.0002,
+      "loss": 1.0643,
+      "step": 2520
+    },
+    {
+      "epoch": 10.33,
+      "grad_norm": 0.644594132900238,
+      "learning_rate": 0.0002,
+      "loss": 1.0536,
+      "step": 2530
+    },
+    {
+      "epoch": 10.37,
+      "grad_norm": 0.6086618304252625,
+      "learning_rate": 0.0002,
+      "loss": 1.0633,
+      "step": 2540
+    },
+    {
+      "epoch": 10.41,
+      "grad_norm": 0.6085098385810852,
+      "learning_rate": 0.0002,
+      "loss": 1.0688,
+      "step": 2550
+    },
+    {
+      "epoch": 10.45,
+      "grad_norm": 0.6334928870201111,
+      "learning_rate": 0.0002,
+      "loss": 1.0797,
+      "step": 2560
+    },
+    {
+      "epoch": 10.49,
+      "grad_norm": 0.5865157246589661,
+      "learning_rate": 0.0002,
+      "loss": 1.0689,
+      "step": 2570
+    },
+    {
+      "epoch": 10.53,
+      "grad_norm": 0.6139411926269531,
+      "learning_rate": 0.0002,
+      "loss": 1.0754,
+      "step": 2580
+    },
+    {
+      "epoch": 10.57,
+      "grad_norm": 0.6306671500205994,
+      "learning_rate": 0.0002,
+      "loss": 1.0831,
+      "step": 2590
+    },
+    {
+      "epoch": 10.61,
+      "grad_norm": 0.5879333019256592,
+      "learning_rate": 0.0002,
+      "loss": 1.0838,
+      "step": 2600
+    },
+    {
+      "epoch": 10.65,
+      "grad_norm": 0.6128151416778564,
+      "learning_rate": 0.0002,
+      "loss": 1.0862,
+      "step": 2610
+    },
+    {
+      "epoch": 10.69,
+      "grad_norm": 0.6163910031318665,
+      "learning_rate": 0.0002,
+      "loss": 1.0784,
+      "step": 2620
+    },
+    {
+      "epoch": 10.73,
+      "grad_norm": 0.6311454772949219,
+      "learning_rate": 0.0002,
+      "loss": 1.0858,
+      "step": 2630
+    },
+    {
+      "epoch": 10.78,
+      "grad_norm": 0.5947273373603821,
+      "learning_rate": 0.0002,
+      "loss": 1.0885,
+      "step": 2640
+    },
+    {
+      "epoch": 10.82,
+      "grad_norm": 0.622706949710846,
+      "learning_rate": 0.0002,
+      "loss": 1.0953,
+      "step": 2650
+    },
+    {
+      "epoch": 10.86,
+      "grad_norm": 0.606569230556488,
+      "learning_rate": 0.0002,
+      "loss": 1.092,
+      "step": 2660
+    },
+    {
+      "epoch": 10.9,
+      "grad_norm": 0.6014845371246338,
+      "learning_rate": 0.0002,
+      "loss": 1.0998,
+      "step": 2670
+    },
+    {
+      "epoch": 10.94,
+      "grad_norm": 0.6037503480911255,
+      "learning_rate": 0.0002,
+      "loss": 1.0997,
+      "step": 2680
+    },
+    {
+      "epoch": 10.98,
+      "grad_norm": 0.5877541899681091,
+      "learning_rate": 0.0002,
+      "loss": 1.1033,
+      "step": 2690
+    },
+    {
+      "epoch": 11.02,
+      "grad_norm": 0.6770003437995911,
+      "learning_rate": 0.0002,
+      "loss": 1.0473,
+      "step": 2700
+    },
+    {
+      "epoch": 11.06,
+      "grad_norm": 0.6640401482582092,
+      "learning_rate": 0.0002,
+      "loss": 1.0122,
+      "step": 2710
+    },
+    {
+      "epoch": 11.1,
+      "grad_norm": 0.6727384924888611,
+      "learning_rate": 0.0002,
+      "loss": 1.0112,
+      "step": 2720
+    },
+    {
+      "epoch": 11.14,
+      "grad_norm": 0.6241724491119385,
+      "learning_rate": 0.0002,
+      "loss": 1.0163,
+      "step": 2730
+    },
+    {
+      "epoch": 11.18,
+      "grad_norm": 0.6085687875747681,
+      "learning_rate": 0.0002,
+      "loss": 1.0273,
+      "step": 2740
+    },
+    {
+      "epoch": 11.22,
+      "grad_norm": 0.6746591329574585,
+      "learning_rate": 0.0002,
+      "loss": 1.0187,
+      "step": 2750
+    },
+    {
+      "epoch": 11.27,
+      "grad_norm": 0.6451332569122314,
+      "learning_rate": 0.0002,
+      "loss": 1.0229,
+      "step": 2760
+    },
+    {
+      "epoch": 11.31,
+      "grad_norm": 0.6338198781013489,
+      "learning_rate": 0.0002,
+      "loss": 1.0256,
+      "step": 2770
+    },
+    {
+      "epoch": 11.35,
+      "grad_norm": 0.6467145085334778,
+      "learning_rate": 0.0002,
+      "loss": 1.0301,
+      "step": 2780
+    },
+    {
+      "epoch": 11.39,
+      "grad_norm": 0.6614335179328918,
+      "learning_rate": 0.0002,
+      "loss": 1.044,
+      "step": 2790
+    },
+    {
+      "epoch": 11.43,
+      "grad_norm": 0.6373239755630493,
+      "learning_rate": 0.0002,
+      "loss": 1.0453,
+      "step": 2800
+    },
+    {
+      "epoch": 11.47,
+      "grad_norm": 0.6863211393356323,
+      "learning_rate": 0.0002,
+      "loss": 1.0461,
+      "step": 2810
+    },
+    {
+      "epoch": 11.51,
+      "grad_norm": 0.608102023601532,
+      "learning_rate": 0.0002,
+      "loss": 1.0356,
+      "step": 2820
+    },
+    {
+      "epoch": 11.55,
+      "grad_norm": 0.6649318933486938,
+      "learning_rate": 0.0002,
+      "loss": 1.0521,
+      "step": 2830
+    },
+    {
+      "epoch": 11.59,
+      "grad_norm": 0.6344706416130066,
+      "learning_rate": 0.0002,
+      "loss": 1.0556,
+      "step": 2840
+    },
+    {
+      "epoch": 11.63,
+      "grad_norm": 0.6278392672538757,
+      "learning_rate": 0.0002,
+      "loss": 1.0476,
+      "step": 2850
+    },
+    {
+      "epoch": 11.67,
+      "grad_norm": 0.6885977983474731,
+      "learning_rate": 0.0002,
+      "loss": 1.0565,
+      "step": 2860
+    },
+    {
+      "epoch": 11.71,
+      "grad_norm": 0.6405135989189148,
+      "learning_rate": 0.0002,
+      "loss": 1.064,
+      "step": 2870
+    },
+    {
+      "epoch": 11.76,
+      "grad_norm": 0.639030396938324,
+      "learning_rate": 0.0002,
+      "loss": 1.069,
+      "step": 2880
+    },
+    {
+      "epoch": 11.8,
+      "grad_norm": 0.6438204050064087,
+      "learning_rate": 0.0002,
+      "loss": 1.0726,
+      "step": 2890
+    },
+    {
+      "epoch": 11.84,
+      "grad_norm": 0.654549777507782,
+      "learning_rate": 0.0002,
+      "loss": 1.0697,
+      "step": 2900
+    },
+    {
+      "epoch": 11.88,
+      "grad_norm": 0.626146674156189,
+      "learning_rate": 0.0002,
+      "loss": 1.0682,
+      "step": 2910
+    },
+    {
+      "epoch": 11.92,
+      "grad_norm": 0.6238541007041931,
+      "learning_rate": 0.0002,
+      "loss": 1.0669,
+      "step": 2920
+    },
+    {
+      "epoch": 11.96,
+      "grad_norm": 0.6308706402778625,
+      "learning_rate": 0.0002,
+      "loss": 1.07,
+      "step": 2930
+    },
+    {
+      "epoch": 12.0,
+      "grad_norm": 0.6104647517204285,
+      "learning_rate": 0.0002,
+      "loss": 1.0726,
+      "step": 2940
+    },
+    {
+      "epoch": 12.04,
+      "grad_norm": 0.6474860906600952,
+      "learning_rate": 0.0002,
+      "loss": 0.9839,
+      "step": 2950
+    },
+    {
+      "epoch": 12.08,
+      "grad_norm": 0.647906482219696,
+      "learning_rate": 0.0002,
+      "loss": 0.982,
+      "step": 2960
+    },
+    {
+      "epoch": 12.12,
+      "grad_norm": 0.6834690570831299,
+      "learning_rate": 0.0002,
+      "loss": 0.9793,
+      "step": 2970
+    },
+    {
+      "epoch": 12.16,
+      "grad_norm": 0.6847568154335022,
+      "learning_rate": 0.0002,
+      "loss": 0.9796,
+      "step": 2980
+    },
+    {
+      "epoch": 12.2,
+      "grad_norm": 0.6866670846939087,
+      "learning_rate": 0.0002,
+      "loss": 0.9938,
+      "step": 2990
+    },
+    {
+      "epoch": 12.24,
+      "grad_norm": 0.6928325891494751,
+      "learning_rate": 0.0002,
+      "loss": 0.9911,
+      "step": 3000
+    },
+    {
+      "epoch": 12.29,
+      "grad_norm": 0.680347204208374,
+      "learning_rate": 0.0002,
+      "loss": 1.0056,
+      "step": 3010
+    },
+    {
+      "epoch": 12.33,
+      "grad_norm": 0.656384289264679,
+      "learning_rate": 0.0002,
+      "loss": 1.0116,
+      "step": 3020
+    },
+    {
+      "epoch": 12.37,
+      "grad_norm": 0.6849060654640198,
+      "learning_rate": 0.0002,
+      "loss": 1.0191,
+      "step": 3030
+    },
+    {
+      "epoch": 12.41,
+      "grad_norm": 0.6677911877632141,
+      "learning_rate": 0.0002,
+      "loss": 1.0152,
+      "step": 3040
+    },
+    {
+      "epoch": 12.45,
+      "grad_norm": 0.697119414806366,
+      "learning_rate": 0.0002,
+      "loss": 1.0144,
+      "step": 3050
+    },
+    {
+      "epoch": 12.49,
+      "grad_norm": 0.6748687028884888,
+      "learning_rate": 0.0002,
+      "loss": 1.0099,
+      "step": 3060
+    },
+    {
+      "epoch": 12.53,
+      "grad_norm": 0.6613055467605591,
+      "learning_rate": 0.0002,
+      "loss": 1.0213,
+      "step": 3070
+    },
+    {
+      "epoch": 12.57,
+      "grad_norm": 0.6857287287712097,
+      "learning_rate": 0.0002,
+      "loss": 1.0309,
+      "step": 3080
+    },
+    {
+      "epoch": 12.61,
+      "grad_norm": 0.6758785843849182,
+      "learning_rate": 0.0002,
+      "loss": 1.0447,
+      "step": 3090
+    },
+    {
+      "epoch": 12.65,
+      "grad_norm": 0.6673085689544678,
+      "learning_rate": 0.0002,
+      "loss": 1.0445,
+      "step": 3100
+    },
+    {
+      "epoch": 12.69,
+      "grad_norm": 0.6487051844596863,
+      "learning_rate": 0.0002,
+      "loss": 1.0271,
+      "step": 3110
+    },
+    {
+      "epoch": 12.73,
+      "grad_norm": 0.7024916410446167,
+      "learning_rate": 0.0002,
+      "loss": 1.0357,
+      "step": 3120
+    },
+    {
+      "epoch": 12.78,
+      "grad_norm": 0.6643483638763428,
+      "learning_rate": 0.0002,
+      "loss": 1.0296,
+      "step": 3130
+    },
+    {
+      "epoch": 12.82,
+      "grad_norm": 0.6508316993713379,
+      "learning_rate": 0.0002,
+      "loss": 1.05,
+      "step": 3140
+    },
+    {
+      "epoch": 12.86,
+      "grad_norm": 0.6692200303077698,
+      "learning_rate": 0.0002,
+      "loss": 1.0419,
+      "step": 3150
+    },
+    {
+      "epoch": 12.9,
+      "grad_norm": 0.6588943004608154,
+      "learning_rate": 0.0002,
+      "loss": 1.0413,
+      "step": 3160
+    },
+    {
+      "epoch": 12.94,
+      "grad_norm": 0.7068329453468323,
+      "learning_rate": 0.0002,
+      "loss": 1.0469,
+      "step": 3170
+    },
+    {
+      "epoch": 12.98,
+      "grad_norm": 0.638953685760498,
+      "learning_rate": 0.0002,
+      "loss": 1.0491,
+      "step": 3180
+    },
+    {
+      "epoch": 13.02,
+      "grad_norm": 0.7413187026977539,
+      "learning_rate": 0.0002,
+      "loss": 1.0124,
+      "step": 3190
+    },
+    {
+      "epoch": 13.06,
+      "grad_norm": 0.7384817600250244,
+      "learning_rate": 0.0002,
+      "loss": 0.9625,
+      "step": 3200
+    },
+    {
+      "epoch": 13.1,
+      "grad_norm": 0.7144999504089355,
+      "learning_rate": 0.0002,
+      "loss": 0.9614,
+      "step": 3210
+    },
+    {
+      "epoch": 13.14,
+      "grad_norm": 0.7315141558647156,
+      "learning_rate": 0.0002,
+      "loss": 0.9597,
+      "step": 3220
+    },
+    {
+      "epoch": 13.18,
+      "grad_norm": 0.6805746555328369,
+      "learning_rate": 0.0002,
+      "loss": 0.9732,
+      "step": 3230
+    },
+    {
+      "epoch": 13.22,
+      "grad_norm": 0.6961046457290649,
+      "learning_rate": 0.0002,
+      "loss": 0.9827,
+      "step": 3240
+    },
+    {
+      "epoch": 13.27,
+      "grad_norm": 0.7174761891365051,
+      "learning_rate": 0.0002,
+      "loss": 0.9706,
+      "step": 3250
+    },
+    {
+      "epoch": 13.31,
+      "grad_norm": 0.7133183479309082,
+      "learning_rate": 0.0002,
+      "loss": 0.9731,
+      "step": 3260
+    },
+    {
+      "epoch": 13.35,
+      "grad_norm": 0.7056016325950623,
+      "learning_rate": 0.0002,
+      "loss": 0.9816,
+      "step": 3270
+    },
+    {
+      "epoch": 13.39,
+      "grad_norm": 0.7086601853370667,
+      "learning_rate": 0.0002,
+      "loss": 0.9971,
+      "step": 3280
+    },
+    {
+      "epoch": 13.43,
+      "grad_norm": 0.6886954307556152,
+      "learning_rate": 0.0002,
+      "loss": 0.9845,
+      "step": 3290
+    },
+    {
+      "epoch": 13.47,
+      "grad_norm": 0.7162395119667053,
+      "learning_rate": 0.0002,
+      "loss": 0.9968,
+      "step": 3300
+    },
+    {
+      "epoch": 13.51,
+      "grad_norm": 0.711045503616333,
+      "learning_rate": 0.0002,
+      "loss": 0.999,
+      "step": 3310
+    },
+    {
+      "epoch": 13.55,
+      "grad_norm": 0.6801750659942627,
+      "learning_rate": 0.0002,
+      "loss": 0.9953,
+      "step": 3320
+    },
+    {
+      "epoch": 13.59,
+      "grad_norm": 0.721878707408905,
+      "learning_rate": 0.0002,
+      "loss": 1.0077,
+      "step": 3330
+    },
+    {
+      "epoch": 13.63,
+      "grad_norm": 0.6965034008026123,
+      "learning_rate": 0.0002,
+      "loss": 1.0135,
+      "step": 3340
+    },
+    {
+      "epoch": 13.67,
+      "grad_norm": 0.7133123278617859,
+      "learning_rate": 0.0002,
+      "loss": 1.0047,
+      "step": 3350
+    },
+    {
+      "epoch": 13.71,
+      "grad_norm": 0.7012418508529663,
+      "learning_rate": 0.0002,
+      "loss": 1.0086,
+      "step": 3360
+    },
+    {
+      "epoch": 13.76,
+      "grad_norm": 0.6652898788452148,
+      "learning_rate": 0.0002,
+      "loss": 1.0069,
+      "step": 3370
+    },
+    {
+      "epoch": 13.8,
+      "grad_norm": 0.6638497114181519,
+      "learning_rate": 0.0002,
+      "loss": 1.0126,
+      "step": 3380
+    },
+    {
+      "epoch": 13.84,
+      "grad_norm": 0.6930794715881348,
+      "learning_rate": 0.0002,
+      "loss": 1.009,
+      "step": 3390
+    },
+    {
+      "epoch": 13.88,
+      "grad_norm": 0.7075713276863098,
+      "learning_rate": 0.0002,
+      "loss": 1.0263,
+      "step": 3400
+    },
+    {
+      "epoch": 13.92,
+      "grad_norm": 0.7108717560768127,
+      "learning_rate": 0.0002,
+      "loss": 1.0321,
+      "step": 3410
+    },
+    {
+      "epoch": 13.96,
+      "grad_norm": 0.6672636866569519,
+      "learning_rate": 0.0002,
+      "loss": 1.0277,
+      "step": 3420
+    },
+    {
+      "epoch": 14.0,
+      "grad_norm": 0.667779803276062,
+      "learning_rate": 0.0002,
+      "loss": 1.0282,
+      "step": 3430
+    },
+    {
+      "epoch": 14.04,
+      "grad_norm": 0.7257822751998901,
+      "learning_rate": 0.0002,
+      "loss": 0.9243,
+      "step": 3440
+    },
+    {
+      "epoch": 14.08,
+      "grad_norm": 0.7209460735321045,
+      "learning_rate": 0.0002,
+      "loss": 0.9293,
+      "step": 3450
+    },
+    {
+      "epoch": 14.12,
+      "grad_norm": 0.773958146572113,
+      "learning_rate": 0.0002,
+      "loss": 0.9408,
+      "step": 3460
+    },
+    {
+      "epoch": 14.16,
+      "grad_norm": 0.7353633642196655,
+      "learning_rate": 0.0002,
+      "loss": 0.939,
+      "step": 3470
+    },
+    {
+      "epoch": 14.2,
+      "grad_norm": 0.7394313812255859,
+      "learning_rate": 0.0002,
+      "loss": 0.9425,
+      "step": 3480
+    },
+    {
+      "epoch": 14.24,
+      "grad_norm": 0.7675743699073792,
+      "learning_rate": 0.0002,
+      "loss": 0.953,
+      "step": 3490
+    },
+    {
+      "epoch": 14.29,
+      "grad_norm": 0.7433834075927734,
+      "learning_rate": 0.0002,
+      "loss": 0.9611,
+      "step": 3500
+    },
+    {
+      "epoch": 14.33,
+      "grad_norm": 0.7248120903968811,
+      "learning_rate": 0.0002,
+      "loss": 0.9671,
+      "step": 3510
+    },
+    {
+      "epoch": 14.37,
+      "grad_norm": 0.7551497220993042,
+      "learning_rate": 0.0002,
+      "loss": 0.9766,
+      "step": 3520
+    },
+    {
+      "epoch": 14.41,
+      "grad_norm": 0.7119362354278564,
+      "learning_rate": 0.0002,
+      "loss": 0.9735,
+      "step": 3530
+    },
+    {
+      "epoch": 14.45,
+      "grad_norm": 0.7423197031021118,
+      "learning_rate": 0.0002,
+      "loss": 0.97,
+      "step": 3540
+    },
+    {
+      "epoch": 14.49,
+      "grad_norm": 0.7277734279632568,
+      "learning_rate": 0.0002,
+      "loss": 0.9675,
+      "step": 3550
+    },
+    {
+      "epoch": 14.53,
+      "grad_norm": 0.7504876852035522,
+      "learning_rate": 0.0002,
+      "loss": 0.9821,
+      "step": 3560
+    },
+    {
+      "epoch": 14.57,
+      "grad_norm": 0.7577353119850159,
+      "learning_rate": 0.0002,
+      "loss": 0.9813,
+      "step": 3570
+    },
+    {
+      "epoch": 14.61,
+      "grad_norm": 0.7336677312850952,
+      "learning_rate": 0.0002,
+      "loss": 0.9858,
+      "step": 3580
+    },
+    {
+      "epoch": 14.65,
+      "grad_norm": 0.7539398670196533,
+      "learning_rate": 0.0002,
+      "loss": 0.9973,
+      "step": 3590
+    },
+    {
+      "epoch": 14.69,
+      "grad_norm": 0.7318299412727356,
+      "learning_rate": 0.0002,
+      "loss": 0.9886,
+      "step": 3600
+    },
+    {
+      "epoch": 14.73,
+      "grad_norm": 0.6988388299942017,
+      "learning_rate": 0.0002,
+      "loss": 0.9927,
+      "step": 3610
+    },
+    {
+      "epoch": 14.78,
+      "grad_norm": 0.7509441375732422,
+      "learning_rate": 0.0002,
+      "loss": 0.9999,
+      "step": 3620
+    },
+    {
+      "epoch": 14.82,
+      "grad_norm": 0.7021549344062805,
+      "learning_rate": 0.0002,
+      "loss": 1.0011,
+      "step": 3630
+    },
+    {
+      "epoch": 14.86,
+      "grad_norm": 0.7372128367424011,
+      "learning_rate": 0.0002,
+      "loss": 1.0,
+      "step": 3640
+    },
+    {
+      "epoch": 14.9,
+      "grad_norm": 0.7391555905342102,
+      "learning_rate": 0.0002,
+      "loss": 0.9976,
+      "step": 3650
+    },
+    {
+      "epoch": 14.94,
+      "grad_norm": 0.7135056853294373,
+      "learning_rate": 0.0002,
+      "loss": 1.0077,
+      "step": 3660
+    },
+    {
+      "epoch": 14.98,
+      "grad_norm": 0.7650583982467651,
+      "learning_rate": 0.0002,
+      "loss": 0.9963,
+      "step": 3670
+    },
+    {
+      "epoch": 15.02,
+      "grad_norm": 0.8550894856452942,
+      "learning_rate": 0.0002,
+      "loss": 0.9701,
+      "step": 3680
+    },
+    {
+      "epoch": 15.06,
+      "grad_norm": 0.8099115490913391,
+      "learning_rate": 0.0002,
+      "loss": 0.9164,
+      "step": 3690
+    },
+    {
+      "epoch": 15.1,
+      "grad_norm": 0.7715261578559875,
+      "learning_rate": 0.0002,
+      "loss": 0.9122,
+      "step": 3700
+    },
+    {
+      "epoch": 15.14,
+      "grad_norm": 0.7662339806556702,
+      "learning_rate": 0.0002,
+      "loss": 0.9292,
+      "step": 3710
+    },
+    {
+      "epoch": 15.18,
+      "grad_norm": 0.7850679755210876,
+      "learning_rate": 0.0002,
+      "loss": 0.927,
+      "step": 3720
+    },
+    {
+      "epoch": 15.22,
+      "grad_norm": 0.775543212890625,
+      "learning_rate": 0.0002,
+      "loss": 0.9323,
+      "step": 3730
+    },
+    {
+      "epoch": 15.27,
+      "grad_norm": 0.75054931640625,
+      "learning_rate": 0.0002,
+      "loss": 0.9387,
+      "step": 3740
+    },
+    {
+      "epoch": 15.31,
+      "grad_norm": 0.758657693862915,
+      "learning_rate": 0.0002,
+      "loss": 0.9404,
+      "step": 3750
+    },
+    {
+      "epoch": 15.35,
+      "grad_norm": 0.7817949056625366,
+      "learning_rate": 0.0002,
+      "loss": 0.9367,
+      "step": 3760
+    },
+    {
+      "epoch": 15.39,
+      "grad_norm": 0.7654836773872375,
+      "learning_rate": 0.0002,
+      "loss": 0.9466,
+      "step": 3770
+    },
+    {
+      "epoch": 15.43,
+      "grad_norm": 0.7418621778488159,
+      "learning_rate": 0.0002,
+      "loss": 0.9446,
+      "step": 3780
+    },
+    {
+      "epoch": 15.47,
+      "grad_norm": 0.7546373009681702,
+      "learning_rate": 0.0002,
+      "loss": 0.9456,
+      "step": 3790
+    },
+    {
+      "epoch": 15.51,
+      "grad_norm": 0.7481057047843933,
+      "learning_rate": 0.0002,
+      "loss": 0.9532,
+      "step": 3800
+    },
+    {
+      "epoch": 15.55,
+      "grad_norm": 0.7622807025909424,
+      "learning_rate": 0.0002,
+      "loss": 0.9618,
+      "step": 3810
+    },
+    {
+      "epoch": 15.59,
+      "grad_norm": 0.7714546322822571,
+      "learning_rate": 0.0002,
+      "loss": 0.9596,
+      "step": 3820
+    },
+    {
+      "epoch": 15.63,
+      "grad_norm": 0.7699543237686157,
+      "learning_rate": 0.0002,
+      "loss": 0.9655,
+      "step": 3830
+    },
+    {
+      "epoch": 15.67,
+      "grad_norm": 0.7812892198562622,
+      "learning_rate": 0.0002,
+      "loss": 0.9691,
+      "step": 3840
+    },
+    {
+      "epoch": 15.71,
+      "grad_norm": 0.7776221036911011,
+      "learning_rate": 0.0002,
+      "loss": 0.9659,
+      "step": 3850
+    },
+    {
+      "epoch": 15.76,
+      "grad_norm": 0.7372971773147583,
+      "learning_rate": 0.0002,
+      "loss": 0.9709,
+      "step": 3860
+    },
+    {
+      "epoch": 15.8,
+      "grad_norm": 0.7554543018341064,
+      "learning_rate": 0.0002,
+      "loss": 0.9763,
+      "step": 3870
+    },
+    {
+      "epoch": 15.84,
+      "grad_norm": 0.77300626039505,
+      "learning_rate": 0.0002,
+      "loss": 0.9856,
+      "step": 3880
+    },
+    {
+      "epoch": 15.88,
+      "grad_norm": 0.7235614061355591,
+      "learning_rate": 0.0002,
+      "loss": 0.9782,
+      "step": 3890
+    },
+    {
+      "epoch": 15.92,
+      "grad_norm": 0.7121375799179077,
+      "learning_rate": 0.0002,
+      "loss": 0.9886,
+      "step": 3900
+    },
+    {
+      "epoch": 15.96,
+      "grad_norm": 0.7238226532936096,
+      "learning_rate": 0.0002,
+      "loss": 0.9896,
+      "step": 3910
+    },
+    {
+      "epoch": 16.0,
+      "grad_norm": 0.7624945640563965,
+      "learning_rate": 0.0002,
+      "loss": 0.9885,
+      "step": 3920
+    },
+    {
+      "epoch": 16.04,
+      "grad_norm": 0.7497121691703796,
+      "learning_rate": 0.0002,
+      "loss": 0.8911,
+      "step": 3930
+    },
+    {
+      "epoch": 16.08,
+      "grad_norm": 0.7473620176315308,
+      "learning_rate": 0.0002,
+      "loss": 0.8938,
+      "step": 3940
+    },
+    {
+      "epoch": 16.12,
+      "grad_norm": 0.75458824634552,
+      "learning_rate": 0.0002,
+      "loss": 0.8986,
+      "step": 3950
+    },
+    {
+      "epoch": 16.16,
+      "grad_norm": 0.788559079170227,
+      "learning_rate": 0.0002,
+      "loss": 0.9065,
+      "step": 3960
+    },
+    {
+      "epoch": 16.2,
+      "grad_norm": 0.8157529830932617,
+      "learning_rate": 0.0002,
+      "loss": 0.91,
+      "step": 3970
+    },
+    {
+      "epoch": 16.24,
+      "grad_norm": 0.8045533895492554,
+      "learning_rate": 0.0002,
+      "loss": 0.9108,
+      "step": 3980
+    },
+    {
+      "epoch": 16.29,
+      "grad_norm": 0.8106245398521423,
+      "learning_rate": 0.0002,
+      "loss": 0.9243,
+      "step": 3990
+    },
+    {
+      "epoch": 16.33,
+      "grad_norm": 0.7821611762046814,
+      "learning_rate": 0.0002,
+      "loss": 0.9207,
+      "step": 4000
+    },
+    {
+      "epoch": 16.37,
+      "grad_norm": 0.7940720915794373,
+      "learning_rate": 0.0002,
+      "loss": 0.9216,
+      "step": 4010
+    },
+    {
+      "epoch": 16.41,
+      "grad_norm": 0.8178911209106445,
+      "learning_rate": 0.0002,
+      "loss": 0.9305,
+      "step": 4020
+    },
+    {
+      "epoch": 16.45,
+      "grad_norm": 0.768060564994812,
+      "learning_rate": 0.0002,
+      "loss": 0.931,
+      "step": 4030
+    },
+    {
+      "epoch": 16.49,
+      "grad_norm": 0.7514263987541199,
+      "learning_rate": 0.0002,
+      "loss": 0.9358,
+      "step": 4040
+    },
+    {
+      "epoch": 16.53,
+      "grad_norm": 0.7860275506973267,
+      "learning_rate": 0.0002,
+      "loss": 0.9395,
+      "step": 4050
+    },
+    {
+      "epoch": 16.57,
+      "grad_norm": 0.8064606785774231,
+      "learning_rate": 0.0002,
+      "loss": 0.947,
+      "step": 4060
+    },
+    {
+      "epoch": 16.61,
+      "grad_norm": 0.7772034406661987,
+      "learning_rate": 0.0002,
+      "loss": 0.9334,
+      "step": 4070
+    },
+    {
+      "epoch": 16.65,
+      "grad_norm": 0.7714650630950928,
+      "learning_rate": 0.0002,
+      "loss": 0.9493,
+      "step": 4080
+    },
+    {
+      "epoch": 16.69,
+      "grad_norm": 0.7864323258399963,
+      "learning_rate": 0.0002,
+      "loss": 0.9594,
+      "step": 4090
+    },
+    {
+      "epoch": 16.73,
+      "grad_norm": 0.7611095905303955,
+      "learning_rate": 0.0002,
+      "loss": 0.9526,
+      "step": 4100
+    },
+    {
+      "epoch": 16.78,
+      "grad_norm": 0.7637573480606079,
+      "learning_rate": 0.0002,
+      "loss": 0.9504,
+      "step": 4110
+    },
+    {
+      "epoch": 16.82,
+      "grad_norm": 0.7966247200965881,
+      "learning_rate": 0.0002,
+      "loss": 0.9617,
+      "step": 4120
+    },
+    {
+      "epoch": 16.86,
+      "grad_norm": 0.7575141191482544,
+      "learning_rate": 0.0002,
+      "loss": 0.9726,
+      "step": 4130
+    },
+    {
+      "epoch": 16.9,
+      "grad_norm": 0.7849740982055664,
+      "learning_rate": 0.0002,
+      "loss": 0.9674,
+      "step": 4140
+    },
+    {
+      "epoch": 16.94,
+      "grad_norm": 0.7860226631164551,
+      "learning_rate": 0.0002,
+      "loss": 0.9645,
+      "step": 4150
+    },
+    {
+      "epoch": 16.98,
+      "grad_norm": 0.7693030834197998,
+      "learning_rate": 0.0002,
+      "loss": 0.9714,
+      "step": 4160
+    },
+    {
+      "epoch": 17.02,
+      "grad_norm": 0.8382769227027893,
+      "learning_rate": 0.0002,
+      "loss": 0.9205,
+      "step": 4170
+    },
+    {
+      "epoch": 17.06,
+      "grad_norm": 0.8301939368247986,
+      "learning_rate": 0.0002,
+      "loss": 0.8819,
+      "step": 4180
+    },
+    {
+      "epoch": 17.1,
+      "grad_norm": 0.820909321308136,
+      "learning_rate": 0.0002,
+      "loss": 0.8783,
+      "step": 4190
+    },
+    {
+      "epoch": 17.14,
+      "grad_norm": 0.8272302150726318,
+      "learning_rate": 0.0002,
+      "loss": 0.8888,
+      "step": 4200
+    },
+    {
+      "epoch": 17.18,
+      "grad_norm": 0.8278018236160278,
+      "learning_rate": 0.0002,
+      "loss": 0.8841,
+      "step": 4210
+    },
+    {
+      "epoch": 17.22,
+      "grad_norm": 0.8393099308013916,
+      "learning_rate": 0.0002,
+      "loss": 0.8974,
+      "step": 4220
+    },
+    {
+      "epoch": 17.27,
+      "grad_norm": 0.8167082071304321,
+      "learning_rate": 0.0002,
+      "loss": 0.8938,
+      "step": 4230
+    },
+    {
+      "epoch": 17.31,
+      "grad_norm": 0.7832567691802979,
+      "learning_rate": 0.0002,
+      "loss": 0.905,
+      "step": 4240
+    },
+    {
+      "epoch": 17.35,
+      "grad_norm": 0.794981837272644,
+      "learning_rate": 0.0002,
+      "loss": 0.9109,
+      "step": 4250
+    },
+    {
+      "epoch": 17.39,
+      "grad_norm": 0.828493058681488,
+      "learning_rate": 0.0002,
+      "loss": 0.9087,
+      "step": 4260
+    },
+    {
+      "epoch": 17.43,
+      "grad_norm": 0.7956497669219971,
+      "learning_rate": 0.0002,
+      "loss": 0.9181,
+      "step": 4270
+    },
+    {
+      "epoch": 17.47,
+      "grad_norm": 0.8334782123565674,
+      "learning_rate": 0.0002,
+      "loss": 0.9177,
+      "step": 4280
+    },
+    {
+      "epoch": 17.51,
+      "grad_norm": 0.820360004901886,
+      "learning_rate": 0.0002,
+      "loss": 0.9268,
+      "step": 4290
+    },
+    {
+      "epoch": 17.55,
+      "grad_norm": 0.8249423503875732,
+      "learning_rate": 0.0002,
+      "loss": 0.9238,
+      "step": 4300
+    },
+    {
+      "epoch": 17.59,
+      "grad_norm": 0.8076381683349609,
+      "learning_rate": 0.0002,
+      "loss": 0.9256,
+      "step": 4310
+    },
+    {
+      "epoch": 17.63,
+      "grad_norm": 0.792510986328125,
+      "learning_rate": 0.0002,
+      "loss": 0.9404,
+      "step": 4320
+    },
+    {
+      "epoch": 17.67,
+      "grad_norm": 0.835329532623291,
+      "learning_rate": 0.0002,
+      "loss": 0.9316,
+      "step": 4330
+    },
+    {
+      "epoch": 17.71,
+      "grad_norm": 0.8437380790710449,
+      "learning_rate": 0.0002,
+      "loss": 0.9416,
+      "step": 4340
+    },
+    {
+      "epoch": 17.76,
+      "grad_norm": 0.7976727485656738,
+      "learning_rate": 0.0002,
+      "loss": 0.9326,
+      "step": 4350
+    },
+    {
+      "epoch": 17.8,
+      "grad_norm": 0.7968472242355347,
+      "learning_rate": 0.0002,
+      "loss": 0.9409,
+      "step": 4360
+    },
+    {
+      "epoch": 17.84,
+      "grad_norm": 0.7687345147132874,
+      "learning_rate": 0.0002,
+      "loss": 0.9419,
+      "step": 4370
+    },
+    {
+      "epoch": 17.88,
+      "grad_norm": 0.7815808653831482,
+      "learning_rate": 0.0002,
+      "loss": 0.9522,
+      "step": 4380
+    },
+    {
+      "epoch": 17.92,
+      "grad_norm": 0.816125214099884,
+      "learning_rate": 0.0002,
+      "loss": 0.9427,
+      "step": 4390
+    },
+    {
+      "epoch": 17.96,
+      "grad_norm": 0.7884161472320557,
+      "learning_rate": 0.0002,
+      "loss": 0.9494,
+      "step": 4400
+    },
+    {
+      "epoch": 18.0,
+      "grad_norm": 0.7613294124603271,
+      "learning_rate": 0.0002,
+      "loss": 0.9592,
+      "step": 4410
+    },
+    {
+      "epoch": 18.04,
+      "grad_norm": 0.7931154370307922,
+      "learning_rate": 0.0002,
+      "loss": 0.8652,
+      "step": 4420
+    },
+    {
+      "epoch": 18.08,
+      "grad_norm": 0.855189323425293,
+      "learning_rate": 0.0002,
+      "loss": 0.8679,
+      "step": 4430
+    },
+    {
+      "epoch": 18.12,
+      "grad_norm": 0.8522574305534363,
+      "learning_rate": 0.0002,
+      "loss": 0.8644,
+      "step": 4440
+    },
+    {
+      "epoch": 18.16,
+      "grad_norm": 0.8371242880821228,
+      "learning_rate": 0.0002,
+      "loss": 0.8635,
+      "step": 4450
+    },
+    {
+      "epoch": 18.2,
+      "grad_norm": 0.8409127593040466,
+      "learning_rate": 0.0002,
+      "loss": 0.8723,
+      "step": 4460
+    },
+    {
+      "epoch": 18.24,
+      "grad_norm": 0.8761357665061951,
+      "learning_rate": 0.0002,
+      "loss": 0.8783,
+      "step": 4470
+    },
+    {
+      "epoch": 18.29,
+      "grad_norm": 0.8433621525764465,
+      "learning_rate": 0.0002,
+      "loss": 0.8876,
+      "step": 4480
+    },
+    {
+      "epoch": 18.33,
+      "grad_norm": 0.8117638230323792,
+      "learning_rate": 0.0002,
+      "loss": 0.892,
+      "step": 4490
+    },
+    {
+      "epoch": 18.37,
+      "grad_norm": 0.8237631916999817,
+      "learning_rate": 0.0002,
+      "loss": 0.8943,
+      "step": 4500
+    },
+    {
+      "epoch": 18.41,
+      "grad_norm": 0.8152850270271301,
+      "learning_rate": 0.0002,
+      "loss": 0.8909,
+      "step": 4510
+    },
+    {
+      "epoch": 18.45,
+      "grad_norm": 0.9371042251586914,
+      "learning_rate": 0.0002,
+      "loss": 0.8989,
+      "step": 4520
+    },
+    {
+      "epoch": 18.49,
+      "grad_norm": 0.8668140172958374,
+      "learning_rate": 0.0002,
+      "loss": 0.9023,
+      "step": 4530
+    },
+    {
+      "epoch": 18.53,
+      "grad_norm": 0.8652714490890503,
+      "learning_rate": 0.0002,
+      "loss": 0.9007,
+      "step": 4540
+    },
+    {
+      "epoch": 18.57,
+      "grad_norm": 0.8392309546470642,
+      "learning_rate": 0.0002,
+      "loss": 0.919,
+      "step": 4550
+    },
+    {
+      "epoch": 18.61,
+      "grad_norm": 0.8538224101066589,
+      "learning_rate": 0.0002,
+      "loss": 0.9169,
+      "step": 4560
+    },
+    {
+      "epoch": 18.65,
+      "grad_norm": 0.7933239340782166,
+      "learning_rate": 0.0002,
+      "loss": 0.9183,
+      "step": 4570
+    },
+    {
+      "epoch": 18.69,
+      "grad_norm": 0.8141274452209473,
+      "learning_rate": 0.0002,
+      "loss": 0.9179,
+      "step": 4580
+    },
+    {
+      "epoch": 18.73,
+      "grad_norm": 0.8790974617004395,
+      "learning_rate": 0.0002,
+      "loss": 0.9297,
+      "step": 4590
+    },
+    {
+      "epoch": 18.78,
+      "grad_norm": 0.8103710412979126,
+      "learning_rate": 0.0002,
+      "loss": 0.9331,
+      "step": 4600
+    },
+    {
+      "epoch": 18.82,
+      "grad_norm": 0.7885395288467407,
+      "learning_rate": 0.0002,
+      "loss": 0.931,
+      "step": 4610
+    },
+    {
+      "epoch": 18.86,
+      "grad_norm": 0.80265212059021,
+      "learning_rate": 0.0002,
+      "loss": 0.9323,
+      "step": 4620
+    },
+    {
+      "epoch": 18.9,
+      "grad_norm": 0.8537465929985046,
+      "learning_rate": 0.0002,
+      "loss": 0.932,
+      "step": 4630
+    },
+    {
+      "epoch": 18.94,
+      "grad_norm": 0.7968863844871521,
+      "learning_rate": 0.0002,
+      "loss": 0.9288,
+      "step": 4640
+    },
+    {
+      "epoch": 18.98,
+      "grad_norm": 0.8105510473251343,
+      "learning_rate": 0.0002,
+      "loss": 0.9361,
+      "step": 4650
+    },
+    {
+      "epoch": 19.02,
+      "grad_norm": 0.8854714632034302,
+      "learning_rate": 0.0002,
+      "loss": 0.8973,
+      "step": 4660
+    },
+    {
+      "epoch": 19.06,
+      "grad_norm": 0.8548041582107544,
+      "learning_rate": 0.0002,
+      "loss": 0.8499,
+      "step": 4670
+    },
+    {
+      "epoch": 19.1,
+      "grad_norm": 0.870018482208252,
+      "learning_rate": 0.0002,
+      "loss": 0.8518,
+      "step": 4680
+    },
+    {
+      "epoch": 19.14,
+      "grad_norm": 0.8623254895210266,
+      "learning_rate": 0.0002,
+      "loss": 0.8517,
+      "step": 4690
+    },
+    {
+      "epoch": 19.18,
+      "grad_norm": 0.863254725933075,
+      "learning_rate": 0.0002,
+      "loss": 0.8652,
+      "step": 4700
+    },
+    {
+      "epoch": 19.22,
+      "grad_norm": 0.867749810218811,
+      "learning_rate": 0.0002,
+      "loss": 0.8659,
+      "step": 4710
+    },
+    {
+      "epoch": 19.27,
+      "grad_norm": 0.8433037996292114,
+      "learning_rate": 0.0002,
+      "loss": 0.8728,
+      "step": 4720
+    },
+    {
+      "epoch": 19.31,
+      "grad_norm": 0.8570507764816284,
+      "learning_rate": 0.0002,
+      "loss": 0.8647,
+      "step": 4730
+    },
+    {
+      "epoch": 19.35,
+      "grad_norm": 0.830174446105957,
+      "learning_rate": 0.0002,
+      "loss": 0.8837,
+      "step": 4740
+    },
+    {
+      "epoch": 19.39,
+      "grad_norm": 0.8436498641967773,
+      "learning_rate": 0.0002,
+      "loss": 0.8792,
+      "step": 4750
+    },
+    {
+      "epoch": 19.43,
+      "grad_norm": 0.8207152485847473,
+      "learning_rate": 0.0002,
+      "loss": 0.8945,
+      "step": 4760
+    },
+    {
+      "epoch": 19.47,
+      "grad_norm": 0.8116417527198792,
+      "learning_rate": 0.0002,
+      "loss": 0.8813,
+      "step": 4770
+    },
+    {
+      "epoch": 19.51,
+      "grad_norm": 0.8591784834861755,
+      "learning_rate": 0.0002,
+      "loss": 0.885,
+      "step": 4780
+    },
+    {
+      "epoch": 19.55,
+      "grad_norm": 0.8506168127059937,
+      "learning_rate": 0.0002,
+      "loss": 0.8994,
+      "step": 4790
+    },
+    {
+      "epoch": 19.59,
+      "grad_norm": 0.816745400428772,
+      "learning_rate": 0.0002,
+      "loss": 0.8994,
+      "step": 4800
+    },
+    {
+      "epoch": 19.63,
+      "grad_norm": 0.8365273475646973,
+      "learning_rate": 0.0002,
+      "loss": 0.8909,
+      "step": 4810
+    },
+    {
+      "epoch": 19.67,
+      "grad_norm": 0.8306399583816528,
+      "learning_rate": 0.0002,
+      "loss": 0.9127,
+      "step": 4820
+    },
+    {
+      "epoch": 19.71,
+      "grad_norm": 0.8440532088279724,
+      "learning_rate": 0.0002,
+      "loss": 0.9106,
+      "step": 4830
+    },
+    {
+      "epoch": 19.76,
+      "grad_norm": 0.8330156803131104,
+      "learning_rate": 0.0002,
+      "loss": 0.9093,
+      "step": 4840
+    },
+    {
+      "epoch": 19.8,
+      "grad_norm": 0.8263149857521057,
+      "learning_rate": 0.0002,
+      "loss": 0.9057,
+      "step": 4850
+    },
+    {
+      "epoch": 19.84,
+      "grad_norm": 0.8197040557861328,
+      "learning_rate": 0.0002,
+      "loss": 0.9145,
+      "step": 4860
+    },
+    {
+      "epoch": 19.88,
+      "grad_norm": 0.8366541266441345,
+      "learning_rate": 0.0002,
+      "loss": 0.9107,
+      "step": 4870
+    },
+    {
+      "epoch": 19.92,
+      "grad_norm": 0.8039381504058838,
+      "learning_rate": 0.0002,
+      "loss": 0.9162,
+      "step": 4880
+    },
+    {
+      "epoch": 19.96,
+      "grad_norm": 0.8434823751449585,
+      "learning_rate": 0.0002,
+      "loss": 0.9264,
+      "step": 4890
+    },
+    {
+      "epoch": 20.0,
+      "grad_norm": 0.8329556584358215,
+      "learning_rate": 0.0002,
+      "loss": 0.923,
+      "step": 4900
+    },
+    {
+      "epoch": 20.04,
+      "grad_norm": 0.8620508313179016,
+      "learning_rate": 0.0002,
+      "loss": 0.8292,
+      "step": 4910
+    },
+    {
+      "epoch": 20.08,
+      "grad_norm": 0.8627603650093079,
+      "learning_rate": 0.0002,
+      "loss": 0.8284,
+      "step": 4920
+    },
+    {
+      "epoch": 20.12,
+      "grad_norm": 0.8428844213485718,
+      "learning_rate": 0.0002,
+      "loss": 0.8452,
+      "step": 4930
+    },
+    {
+      "epoch": 20.16,
+      "grad_norm": 0.8328269720077515,
+      "learning_rate": 0.0002,
+      "loss": 0.8424,
+      "step": 4940
+    },
+    {
+      "epoch": 20.2,
+      "grad_norm": 0.8306934237480164,
+      "learning_rate": 0.0002,
+      "loss": 0.8471,
+      "step": 4950
+    },
+    {
+      "epoch": 20.24,
+      "grad_norm": 0.8320675492286682,
+      "learning_rate": 0.0002,
+      "loss": 0.8498,
+      "step": 4960
+    },
+    {
+      "epoch": 20.29,
+      "grad_norm": 0.8537613153457642,
+      "learning_rate": 0.0002,
+      "loss": 0.8675,
+      "step": 4970
+    },
+    {
+      "epoch": 20.33,
+      "grad_norm": 0.8733486533164978,
+      "learning_rate": 0.0002,
+      "loss": 0.8682,
+      "step": 4980
+    },
+    {
+      "epoch": 20.37,
+      "grad_norm": 0.8653712868690491,
+      "learning_rate": 0.0002,
+      "loss": 0.8717,
+      "step": 4990
+    },
+    {
+      "epoch": 20.41,
+      "grad_norm": 0.8138574957847595,
+      "learning_rate": 0.0002,
+      "loss": 0.8655,
+      "step": 5000
+    },
+    {
+      "epoch": 20.45,
+      "grad_norm": 0.8490427732467651,
+      "learning_rate": 0.0002,
+      "loss": 0.8761,
+      "step": 5010
+    },
+    {
+      "epoch": 20.49,
+      "grad_norm": 0.8796558380126953,
+      "learning_rate": 0.0002,
+      "loss": 0.8751,
+      "step": 5020
+    },
+    {
+      "epoch": 20.53,
+      "grad_norm": 0.8779765963554382,
+      "learning_rate": 0.0002,
+      "loss": 0.8755,
+      "step": 5030
+    },
+    {
+      "epoch": 20.57,
+      "grad_norm": 0.8556944131851196,
+      "learning_rate": 0.0002,
+      "loss": 0.8799,
+      "step": 5040
+    },
+    {
+      "epoch": 20.61,
+      "grad_norm": 0.8503381609916687,
+      "learning_rate": 0.0002,
+      "loss": 0.879,
+      "step": 5050
+    },
+    {
+      "epoch": 20.65,
+      "grad_norm": 0.837138831615448,
+      "learning_rate": 0.0002,
+      "loss": 0.8818,
+      "step": 5060
+    },
+    {
+      "epoch": 20.69,
+      "grad_norm": 0.8664600849151611,
+      "learning_rate": 0.0002,
+      "loss": 0.8952,
+      "step": 5070
+    },
+    {
+      "epoch": 20.73,
+      "grad_norm": 0.8737181425094604,
+      "learning_rate": 0.0002,
+      "loss": 0.8997,
+      "step": 5080
+    },
+    {
+      "epoch": 20.78,
+      "grad_norm": 0.8698042631149292,
+      "learning_rate": 0.0002,
+      "loss": 0.9058,
+      "step": 5090
+    },
+    {
+      "epoch": 20.82,
+      "grad_norm": 0.8353410959243774,
+      "learning_rate": 0.0002,
+      "loss": 0.905,
+      "step": 5100
+    },
+    {
+      "epoch": 20.86,
+      "grad_norm": 0.8429268002510071,
+      "learning_rate": 0.0002,
+      "loss": 0.9037,
+      "step": 5110
+    },
+    {
+      "epoch": 20.9,
+      "grad_norm": 0.8391215205192566,
+      "learning_rate": 0.0002,
+      "loss": 0.9017,
+      "step": 5120
+    },
+    {
+      "epoch": 20.94,
+      "grad_norm": 0.8510826826095581,
+      "learning_rate": 0.0002,
+      "loss": 0.9075,
+      "step": 5130
+    },
+    {
+      "epoch": 20.98,
+      "grad_norm": 0.8314224481582642,
+      "learning_rate": 0.0002,
+      "loss": 0.9158,
+      "step": 5140
+    },
+    {
+      "epoch": 21.02,
+      "grad_norm": 0.9201707243919373,
+      "learning_rate": 0.0002,
+      "loss": 0.859,
+      "step": 5150
+    },
+    {
+      "epoch": 21.06,
+      "grad_norm": 0.8647860884666443,
+      "learning_rate": 0.0002,
+      "loss": 0.8171,
+      "step": 5160
+    },
+    {
+      "epoch": 21.1,
+      "grad_norm": 0.8973507881164551,
+      "learning_rate": 0.0002,
+      "loss": 0.8298,
+      "step": 5170
+    },
+    {
+      "epoch": 21.14,
+      "grad_norm": 0.9124497175216675,
+      "learning_rate": 0.0002,
+      "loss": 0.8338,
+      "step": 5180
+    },
+    {
+      "epoch": 21.18,
+      "grad_norm": 0.8619136214256287,
+      "learning_rate": 0.0002,
+      "loss": 0.8246,
+      "step": 5190
+    },
+    {
+      "epoch": 21.22,
+      "grad_norm": 0.8383534550666809,
+      "learning_rate": 0.0002,
+      "loss": 0.8379,
+      "step": 5200
+    },
+    {
+      "epoch": 21.27,
+      "grad_norm": 0.8382851481437683,
+      "learning_rate": 0.0002,
+      "loss": 0.8409,
+      "step": 5210
+    },
+    {
+      "epoch": 21.31,
+      "grad_norm": 0.8646281957626343,
+      "learning_rate": 0.0002,
+      "loss": 0.8475,
+      "step": 5220
+    },
+    {
+      "epoch": 21.35,
+      "grad_norm": 0.9145320057868958,
+      "learning_rate": 0.0002,
+      "loss": 0.8591,
+      "step": 5230
+    },
+    {
+      "epoch": 21.39,
+      "grad_norm": 0.8727798461914062,
+      "learning_rate": 0.0002,
+      "loss": 0.8564,
+      "step": 5240
+    },
+    {
+      "epoch": 21.43,
+      "grad_norm": 0.8649429082870483,
+      "learning_rate": 0.0002,
+      "loss": 0.8569,
+      "step": 5250
+    },
+    {
+      "epoch": 21.47,
+      "grad_norm": 0.8714181184768677,
+      "learning_rate": 0.0002,
+      "loss": 0.8596,
+      "step": 5260
+    },
+    {
+      "epoch": 21.51,
+      "grad_norm": 0.8774384260177612,
+      "learning_rate": 0.0002,
+      "loss": 0.8713,
+      "step": 5270
+    },
+    {
+      "epoch": 21.55,
+      "grad_norm": 0.8977541327476501,
+      "learning_rate": 0.0002,
+      "loss": 0.8626,
+      "step": 5280
+    },
+    {
+      "epoch": 21.59,
+      "grad_norm": 0.8677563667297363,
+      "learning_rate": 0.0002,
+      "loss": 0.8784,
+      "step": 5290
+    },
+    {
+      "epoch": 21.63,
+      "grad_norm": 0.8705365657806396,
+      "learning_rate": 0.0002,
+      "loss": 0.8722,
+      "step": 5300
+    },
+    {
+      "epoch": 21.67,
+      "grad_norm": 0.8717777132987976,
+      "learning_rate": 0.0002,
+      "loss": 0.8833,
+      "step": 5310
+    },
+    {
+      "epoch": 21.71,
+      "grad_norm": 0.8530117273330688,
+      "learning_rate": 0.0002,
+      "loss": 0.8829,
+      "step": 5320
+    },
+    {
+      "epoch": 21.76,
+      "grad_norm": 0.8550623655319214,
+      "learning_rate": 0.0002,
+      "loss": 0.8797,
+      "step": 5330
+    },
+    {
+      "epoch": 21.8,
+      "grad_norm": 0.8520897030830383,
+      "learning_rate": 0.0002,
+      "loss": 0.893,
+      "step": 5340
+    },
+    {
+      "epoch": 21.84,
+      "grad_norm": 0.8366932272911072,
+      "learning_rate": 0.0002,
+      "loss": 0.8916,
+      "step": 5350
+    },
+    {
+      "epoch": 21.88,
+      "grad_norm": 0.8556586503982544,
+      "learning_rate": 0.0002,
+      "loss": 0.8992,
+      "step": 5360
+    },
+    {
+      "epoch": 21.92,
+      "grad_norm": 0.869047224521637,
+      "learning_rate": 0.0002,
+      "loss": 0.8932,
+      "step": 5370
+    },
+    {
+      "epoch": 21.96,
+      "grad_norm": 0.8768820762634277,
+      "learning_rate": 0.0002,
+      "loss": 0.8959,
+      "step": 5380
+    },
+    {
+      "epoch": 22.0,
+      "grad_norm": 0.849677562713623,
+      "learning_rate": 0.0002,
+      "loss": 0.8959,
+      "step": 5390
+    },
+    {
+      "epoch": 22.04,
+      "grad_norm": 0.8397349119186401,
+      "learning_rate": 0.0002,
+      "loss": 0.7987,
+      "step": 5400
+    },
+    {
+      "epoch": 22.08,
+      "grad_norm": 0.8741442561149597,
+      "learning_rate": 0.0002,
+      "loss": 0.8045,
+      "step": 5410
+    },
+    {
+      "epoch": 22.12,
+      "grad_norm": 0.9024834036827087,
+      "learning_rate": 0.0002,
+      "loss": 0.8124,
+      "step": 5420
+    },
+    {
+      "epoch": 22.16,
+      "grad_norm": 0.8942772746086121,
+      "learning_rate": 0.0002,
+      "loss": 0.8169,
+      "step": 5430
+    },
+    {
+      "epoch": 22.2,
+      "grad_norm": 0.8915479183197021,
+      "learning_rate": 0.0002,
+      "loss": 0.8271,
+      "step": 5440
+    },
+    {
+      "epoch": 22.24,
+      "grad_norm": 0.9181793332099915,
+      "learning_rate": 0.0002,
+      "loss": 0.8283,
+      "step": 5450
+    },
+    {
+      "epoch": 22.29,
+      "grad_norm": 0.8821321725845337,
+      "learning_rate": 0.0002,
+      "loss": 0.827,
+      "step": 5460
+    },
+    {
+      "epoch": 22.33,
+      "grad_norm": 0.8514798879623413,
+      "learning_rate": 0.0002,
+      "loss": 0.8345,
+      "step": 5470
+    },
+    {
+      "epoch": 22.37,
+      "grad_norm": 0.8772318363189697,
+      "learning_rate": 0.0002,
+      "loss": 0.8325,
+      "step": 5480
+    },
+    {
+      "epoch": 22.41,
+      "grad_norm": 0.8917196989059448,
+      "learning_rate": 0.0002,
+      "loss": 0.854,
+      "step": 5490
+    },
+    {
+      "epoch": 22.45,
+      "grad_norm": 0.8567050695419312,
+      "learning_rate": 0.0002,
+      "loss": 0.8576,
+      "step": 5500
+    },
+    {
+      "epoch": 22.49,
+      "grad_norm": 0.8994786143302917,
+      "learning_rate": 0.0002,
+      "loss": 0.8482,
+      "step": 5510
+    },
+    {
+      "epoch": 22.53,
+      "grad_norm": 0.8770323395729065,
+      "learning_rate": 0.0002,
+      "loss": 0.8568,
+      "step": 5520
+    },
+    {
+      "epoch": 22.57,
+      "grad_norm": 0.870095431804657,
+      "learning_rate": 0.0002,
+      "loss": 0.8617,
+      "step": 5530
+    },
+    {
+      "epoch": 22.61,
+      "grad_norm": 0.8750589489936829,
+      "learning_rate": 0.0002,
+      "loss": 0.8668,
+      "step": 5540
+    },
+    {
+      "epoch": 22.65,
+      "grad_norm": 0.8779079914093018,
+      "learning_rate": 0.0002,
+      "loss": 0.868,
+      "step": 5550
+    },
+    {
+      "epoch": 22.69,
+      "grad_norm": 0.868596613407135,
+      "learning_rate": 0.0002,
+      "loss": 0.8788,
+      "step": 5560
+    },
+    {
+      "epoch": 22.73,
+      "grad_norm": 0.8535884022712708,
+      "learning_rate": 0.0002,
+      "loss": 0.8783,
+      "step": 5570
+    },
+    {
+      "epoch": 22.78,
+      "grad_norm": 0.8387210965156555,
+      "learning_rate": 0.0002,
+      "loss": 0.8717,
+      "step": 5580
+    },
+    {
+      "epoch": 22.82,
+      "grad_norm": 0.8729193806648254,
+      "learning_rate": 0.0002,
+      "loss": 0.8789,
+      "step": 5590
+    },
+    {
+      "epoch": 22.86,
+      "grad_norm": 0.8541831970214844,
+      "learning_rate": 0.0002,
+      "loss": 0.8804,
+      "step": 5600
+    },
+    {
+      "epoch": 22.9,
+      "grad_norm": 0.8559602499008179,
+      "learning_rate": 0.0002,
+      "loss": 0.8777,
+      "step": 5610
+    },
+    {
+      "epoch": 22.94,
+      "grad_norm": 0.8594103455543518,
+      "learning_rate": 0.0002,
+      "loss": 0.8841,
+      "step": 5620
+    },
+    {
+      "epoch": 22.98,
+      "grad_norm": 0.884156346321106,
+      "learning_rate": 0.0002,
+      "loss": 0.8936,
+      "step": 5630
+    },
+    {
+      "epoch": 23.02,
+      "grad_norm": 0.9736906290054321,
+      "learning_rate": 0.0002,
+      "loss": 0.8374,
+      "step": 5640
+    },
+    {
+      "epoch": 23.06,
+      "grad_norm": 0.9304066300392151,
+      "learning_rate": 0.0002,
+      "loss": 0.8011,
+      "step": 5650
+    },
+    {
+      "epoch": 23.1,
+      "grad_norm": 0.8798370361328125,
+      "learning_rate": 0.0002,
+      "loss": 0.7971,
+      "step": 5660
+    },
+    {
+      "epoch": 23.14,
+      "grad_norm": 0.8726293444633484,
+      "learning_rate": 0.0002,
+      "loss": 0.8068,
+      "step": 5670
+    },
+    {
+      "epoch": 23.18,
+      "grad_norm": 0.918418824672699,
+      "learning_rate": 0.0002,
+      "loss": 0.8051,
+      "step": 5680
+    },
+    {
+      "epoch": 23.22,
+      "grad_norm": 0.8974969387054443,
+      "learning_rate": 0.0002,
+      "loss": 0.8089,
+      "step": 5690
+    },
+    {
+      "epoch": 23.27,
+      "grad_norm": 0.9124098420143127,
+      "learning_rate": 0.0002,
+      "loss": 0.821,
+      "step": 5700
+    },
+    {
+      "epoch": 23.31,
+      "grad_norm": 0.8955199122428894,
+      "learning_rate": 0.0002,
+      "loss": 0.8203,
+      "step": 5710
+    },
+    {
+      "epoch": 23.35,
+      "grad_norm": 0.8901395201683044,
+      "learning_rate": 0.0002,
+      "loss": 0.8324,
+      "step": 5720
+    },
+    {
+      "epoch": 23.39,
+      "grad_norm": 0.9032876491546631,
+      "learning_rate": 0.0002,
+      "loss": 0.8348,
+      "step": 5730
+    },
+    {
+      "epoch": 23.43,
+      "grad_norm": 0.9339714646339417,
+      "learning_rate": 0.0002,
+      "loss": 0.8424,
+      "step": 5740
+    },
+    {
+      "epoch": 23.47,
+      "grad_norm": 0.8993861675262451,
+      "learning_rate": 0.0002,
+      "loss": 0.8364,
+      "step": 5750
+    },
+    {
+      "epoch": 23.51,
+      "grad_norm": 0.8764722347259521,
+      "learning_rate": 0.0002,
+      "loss": 0.8457,
+      "step": 5760
+    },
+    {
+      "epoch": 23.55,
+      "grad_norm": 0.8839731812477112,
+      "learning_rate": 0.0002,
+      "loss": 0.8459,
+      "step": 5770
+    },
+    {
+      "epoch": 23.59,
+      "grad_norm": 0.8804398775100708,
+      "learning_rate": 0.0002,
+      "loss": 0.8564,
+      "step": 5780
+    },
+    {
+      "epoch": 23.63,
+      "grad_norm": 0.922806978225708,
+      "learning_rate": 0.0002,
+      "loss": 0.8613,
+      "step": 5790
+    },
+    {
+      "epoch": 23.67,
+      "grad_norm": 0.8472769260406494,
+      "learning_rate": 0.0002,
+      "loss": 0.8597,
+      "step": 5800
+    },
+    {
+      "epoch": 23.71,
+      "grad_norm": 0.8768579959869385,
+      "learning_rate": 0.0002,
+      "loss": 0.8738,
+      "step": 5810
+    },
+    {
+      "epoch": 23.76,
+      "grad_norm": 0.895548403263092,
+      "learning_rate": 0.0002,
+      "loss": 0.8569,
+      "step": 5820
+    },
+    {
+      "epoch": 23.8,
+      "grad_norm": 0.9207329154014587,
+      "learning_rate": 0.0002,
+      "loss": 0.8677,
+      "step": 5830
+    },
+    {
+      "epoch": 23.84,
+      "grad_norm": 0.8925573825836182,
+      "learning_rate": 0.0002,
+      "loss": 0.8711,
+      "step": 5840
+    },
+    {
+      "epoch": 23.88,
+      "grad_norm": 0.8939485549926758,
+      "learning_rate": 0.0002,
+      "loss": 0.8735,
+      "step": 5850
+    },
+    {
+      "epoch": 23.92,
+      "grad_norm": 0.8766481280326843,
+      "learning_rate": 0.0002,
+      "loss": 0.8805,
+      "step": 5860
+    },
+    {
+      "epoch": 23.96,
+      "grad_norm": 0.8925796747207642,
+      "learning_rate": 0.0002,
+      "loss": 0.8695,
+      "step": 5870
+    },
+    {
+      "epoch": 24.0,
+      "grad_norm": 0.8501390814781189,
+      "learning_rate": 0.0002,
+      "loss": 0.8646,
+      "step": 5880
+    },
+    {
+      "epoch": 24.04,
+      "grad_norm": 0.8877448439598083,
+      "learning_rate": 0.0002,
+      "loss": 0.7849,
+      "step": 5890
+    },
+    {
+      "epoch": 24.08,
+      "grad_norm": 0.860008955001831,
+      "learning_rate": 0.0002,
+      "loss": 0.7923,
+      "step": 5900
+    },
+    {
+      "epoch": 24.12,
+      "grad_norm": 0.9319033622741699,
+      "learning_rate": 0.0002,
+      "loss": 0.7965,
+      "step": 5910
+    },
+    {
+      "epoch": 24.16,
+      "grad_norm": 0.928855299949646,
+      "learning_rate": 0.0002,
+      "loss": 0.7961,
+      "step": 5920
+    },
+    {
+      "epoch": 24.2,
+      "grad_norm": 0.9242786765098572,
+      "learning_rate": 0.0002,
+      "loss": 0.7964,
+      "step": 5930
+    },
+    {
+      "epoch": 24.24,
+      "grad_norm": 0.8834139108657837,
+      "learning_rate": 0.0002,
+      "loss": 0.8113,
+      "step": 5940
+    },
+    {
+      "epoch": 24.29,
+      "grad_norm": 0.8913507461547852,
+      "learning_rate": 0.0002,
+      "loss": 0.8051,
+      "step": 5950
+    },
+    {
+      "epoch": 24.33,
+      "grad_norm": 0.8687690496444702,
+      "learning_rate": 0.0002,
+      "loss": 0.8158,
+      "step": 5960
+    },
+    {
+      "epoch": 24.37,
+      "grad_norm": 0.9044011831283569,
+      "learning_rate": 0.0002,
+      "loss": 0.8232,
+      "step": 5970
+    },
+    {
+      "epoch": 24.41,
+      "grad_norm": 0.8891428709030151,
+      "learning_rate": 0.0002,
+      "loss": 0.8234,
+      "step": 5980
+    },
+    {
+      "epoch": 24.45,
+      "grad_norm": 0.910785436630249,
+      "learning_rate": 0.0002,
+      "loss": 0.8343,
+      "step": 5990
+    },
+    {
+      "epoch": 24.49,
+      "grad_norm": 0.8859900832176208,
+      "learning_rate": 0.0002,
+      "loss": 0.8271,
+      "step": 6000
+    },
+    {
+      "epoch": 24.53,
+      "grad_norm": 0.8689603209495544,
+      "learning_rate": 0.0002,
+      "loss": 0.8257,
+      "step": 6010
+    },
+    {
+      "epoch": 24.57,
+      "grad_norm": 0.908275306224823,
+      "learning_rate": 0.0002,
+      "loss": 0.8429,
+      "step": 6020
+    },
+    {
+      "epoch": 24.61,
+      "grad_norm": 0.9454265832901001,
+      "learning_rate": 0.0002,
+      "loss": 0.8423,
+      "step": 6030
+    },
+    {
+      "epoch": 24.65,
+      "grad_norm": 0.874933123588562,
+      "learning_rate": 0.0002,
+      "loss": 0.8464,
+      "step": 6040
+    },
+    {
+      "epoch": 24.69,
+      "grad_norm": 0.9061406254768372,
+      "learning_rate": 0.0002,
+      "loss": 0.8523,
+      "step": 6050
+    },
+    {
+      "epoch": 24.73,
+      "grad_norm": 0.9278773665428162,
+      "learning_rate": 0.0002,
+      "loss": 0.8487,
+      "step": 6060
+    },
+    {
+      "epoch": 24.78,
+      "grad_norm": 0.924838125705719,
+      "learning_rate": 0.0002,
+      "loss": 0.8463,
+      "step": 6070
+    },
+    {
+      "epoch": 24.82,
+      "grad_norm": 0.9214856624603271,
+      "learning_rate": 0.0002,
+      "loss": 0.8565,
+      "step": 6080
+    },
+    {
+      "epoch": 24.86,
+      "grad_norm": 0.8918023705482483,
+      "learning_rate": 0.0002,
+      "loss": 0.8645,
+      "step": 6090
+    },
+    {
+      "epoch": 24.9,
+      "grad_norm": 0.9123814702033997,
+      "learning_rate": 0.0002,
+      "loss": 0.8645,
+      "step": 6100
+    },
+    {
+      "epoch": 24.94,
+      "grad_norm": 0.8688812851905823,
+      "learning_rate": 0.0002,
+      "loss": 0.8679,
+      "step": 6110
+    },
+    {
+      "epoch": 24.98,
+      "grad_norm": 0.9204630851745605,
+      "learning_rate": 0.0002,
+      "loss": 0.8685,
+      "step": 6120
+    },
+    {
+      "epoch": 25.02,
+      "grad_norm": 0.9178334474563599,
+      "learning_rate": 0.0002,
+      "loss": 0.8285,
+      "step": 6130
+    },
+    {
+      "epoch": 25.06,
+      "grad_norm": 0.915169894695282,
+      "learning_rate": 0.0002,
+      "loss": 0.7686,
+      "step": 6140
+    },
+    {
+      "epoch": 25.1,
+      "grad_norm": 0.8733026385307312,
+      "learning_rate": 0.0002,
+      "loss": 0.7739,
+      "step": 6150
+    },
+    {
+      "epoch": 25.14,
+      "grad_norm": 0.9435804486274719,
+      "learning_rate": 0.0002,
+      "loss": 0.7821,
+      "step": 6160
+    },
+    {
+      "epoch": 25.18,
+      "grad_norm": 0.9279885292053223,
+      "learning_rate": 0.0002,
+      "loss": 0.7951,
+      "step": 6170
+    },
+    {
+      "epoch": 25.22,
+      "grad_norm": 0.8908799886703491,
+      "learning_rate": 0.0002,
+      "loss": 0.796,
+      "step": 6180
+    },
+    {
+      "epoch": 25.27,
+      "grad_norm": 0.9093460440635681,
+      "learning_rate": 0.0002,
+      "loss": 0.8075,
+      "step": 6190
+    },
+    {
+      "epoch": 25.31,
+      "grad_norm": 0.9308035373687744,
+      "learning_rate": 0.0002,
+      "loss": 0.808,
+      "step": 6200
+    },
+    {
+      "epoch": 25.35,
+      "grad_norm": 0.9349119663238525,
+      "learning_rate": 0.0002,
+      "loss": 0.8114,
+      "step": 6210
+    },
+    {
+      "epoch": 25.39,
+      "grad_norm": 0.9189342856407166,
+      "learning_rate": 0.0002,
+      "loss": 0.8171,
+      "step": 6220
+    },
+    {
+      "epoch": 25.43,
+      "grad_norm": 0.9681046605110168,
+      "learning_rate": 0.0002,
+      "loss": 0.8207,
+      "step": 6230
+    },
+    {
+      "epoch": 25.47,
+      "grad_norm": 0.9576332569122314,
+      "learning_rate": 0.0002,
+      "loss": 0.827,
+      "step": 6240
+    },
+    {
+      "epoch": 25.51,
+      "grad_norm": 0.9052371382713318,
+      "learning_rate": 0.0002,
+      "loss": 0.8129,
+      "step": 6250
+    },
+    {
+      "epoch": 25.55,
+      "grad_norm": 0.9296489357948303,
+      "learning_rate": 0.0002,
+      "loss": 0.8267,
+      "step": 6260
+    },
+    {
+      "epoch": 25.59,
+      "grad_norm": 0.9182922840118408,
+      "learning_rate": 0.0002,
+      "loss": 0.831,
+      "step": 6270
+    },
+    {
+      "epoch": 25.63,
+      "grad_norm": 0.9139277338981628,
+      "learning_rate": 0.0002,
+      "loss": 0.833,
+      "step": 6280
+    },
+    {
+      "epoch": 25.67,
+      "grad_norm": 0.9042106866836548,
+      "learning_rate": 0.0002,
+      "loss": 0.8393,
+      "step": 6290
+    },
+    {
+      "epoch": 25.71,
+      "grad_norm": 0.9238719344139099,
+      "learning_rate": 0.0002,
+      "loss": 0.8394,
+      "step": 6300
+    },
+    {
+      "epoch": 25.76,
+      "grad_norm": 0.8891541957855225,
+      "learning_rate": 0.0002,
+      "loss": 0.8513,
+      "step": 6310
+    },
+    {
+      "epoch": 25.8,
+      "grad_norm": 0.8962535858154297,
+      "learning_rate": 0.0002,
+      "loss": 0.8407,
+      "step": 6320
+    },
+    {
+      "epoch": 25.84,
+      "grad_norm": 0.9288328886032104,
+      "learning_rate": 0.0002,
+      "loss": 0.8504,
+      "step": 6330
+    },
+    {
+      "epoch": 25.88,
+      "grad_norm": 0.8737248778343201,
+      "learning_rate": 0.0002,
+      "loss": 0.8483,
+      "step": 6340
+    },
+    {
+      "epoch": 25.92,
+      "grad_norm": 0.8971844911575317,
+      "learning_rate": 0.0002,
+      "loss": 0.8455,
+      "step": 6350
+    },
+    {
+      "epoch": 25.96,
+      "grad_norm": 0.9223408102989197,
+      "learning_rate": 0.0002,
+      "loss": 0.8489,
+      "step": 6360
+    },
+    {
+      "epoch": 26.0,
+      "grad_norm": 0.8855815529823303,
+      "learning_rate": 0.0002,
+      "loss": 0.8668,
+      "step": 6370
+    },
+    {
+      "epoch": 26.04,
+      "grad_norm": 0.9498929977416992,
+      "learning_rate": 0.0002,
+      "loss": 0.7654,
+      "step": 6380
+    },
+    {
+      "epoch": 26.08,
+      "grad_norm": 0.9300951957702637,
+      "learning_rate": 0.0002,
+      "loss": 0.773,
+      "step": 6390
+    },
+    {
+      "epoch": 26.12,
+      "grad_norm": 0.9423607587814331,
+      "learning_rate": 0.0002,
+      "loss": 0.7736,
+      "step": 6400
+    },
+    {
+      "epoch": 26.16,
+      "grad_norm": 0.9024314284324646,
+      "learning_rate": 0.0002,
+      "loss": 0.7751,
+      "step": 6410
+    },
+    {
+      "epoch": 26.2,
+      "grad_norm": 0.9432561993598938,
+      "learning_rate": 0.0002,
+      "loss": 0.7874,
+      "step": 6420
+    },
+    {
+      "epoch": 26.24,
+      "grad_norm": 0.9214005470275879,
+      "learning_rate": 0.0002,
+      "loss": 0.792,
+      "step": 6430
+    },
+    {
+      "epoch": 26.29,
+      "grad_norm": 0.911280632019043,
+      "learning_rate": 0.0002,
+      "loss": 0.7925,
+      "step": 6440
+    },
+    {
+      "epoch": 26.33,
+      "grad_norm": 0.9321004748344421,
+      "learning_rate": 0.0002,
+      "loss": 0.8028,
+      "step": 6450
+    },
+    {
+      "epoch": 26.37,
+      "grad_norm": 0.9027372598648071,
+      "learning_rate": 0.0002,
+      "loss": 0.8058,
+      "step": 6460
+    },
+    {
+      "epoch": 26.41,
+      "grad_norm": 0.9106818437576294,
+      "learning_rate": 0.0002,
+      "loss": 0.8035,
+      "step": 6470
+    },
+    {
+      "epoch": 26.45,
+      "grad_norm": 0.9135387539863586,
+      "learning_rate": 0.0002,
+      "loss": 0.8075,
+      "step": 6480
+    },
+    {
+      "epoch": 26.49,
+      "grad_norm": 0.9544934034347534,
+      "learning_rate": 0.0002,
+      "loss": 0.8114,
+      "step": 6490
+    },
+    {
+      "epoch": 26.53,
+      "grad_norm": 0.9143434762954712,
+      "learning_rate": 0.0002,
+      "loss": 0.814,
+      "step": 6500
+    },
+    {
+      "epoch": 26.57,
+      "grad_norm": 0.9120767712593079,
+      "learning_rate": 0.0002,
+      "loss": 0.823,
+      "step": 6510
+    },
+    {
+      "epoch": 26.61,
+      "grad_norm": 0.9214150905609131,
+      "learning_rate": 0.0002,
+      "loss": 0.8263,
+      "step": 6520
+    },
+    {
+      "epoch": 26.65,
+      "grad_norm": 0.909246563911438,
+      "learning_rate": 0.0002,
+      "loss": 0.8313,
+      "step": 6530
+    },
+    {
+      "epoch": 26.69,
+      "grad_norm": 0.8854065537452698,
+      "learning_rate": 0.0002,
+      "loss": 0.8309,
+      "step": 6540
+    },
+    {
+      "epoch": 26.73,
+      "grad_norm": 0.8996104001998901,
+      "learning_rate": 0.0002,
+      "loss": 0.8276,
+      "step": 6550
+    },
+    {
+      "epoch": 26.78,
+      "grad_norm": 0.9389463067054749,
+      "learning_rate": 0.0002,
+      "loss": 0.8276,
+      "step": 6560
+    },
+    {
+      "epoch": 26.82,
+      "grad_norm": 0.9044690132141113,
+      "learning_rate": 0.0002,
+      "loss": 0.8384,
+      "step": 6570
+    },
+    {
+      "epoch": 26.86,
+      "grad_norm": 0.9313196539878845,
+      "learning_rate": 0.0002,
+      "loss": 0.8362,
+      "step": 6580
+    },
+    {
+      "epoch": 26.9,
+      "grad_norm": 0.9324984550476074,
+      "learning_rate": 0.0002,
+      "loss": 0.8483,
+      "step": 6590
+    },
+    {
+      "epoch": 26.94,
+      "grad_norm": 0.9040074348449707,
+      "learning_rate": 0.0002,
+      "loss": 0.8502,
+      "step": 6600
+    },
+    {
+      "epoch": 26.98,
+      "grad_norm": 0.9061781764030457,
+      "learning_rate": 0.0002,
+      "loss": 0.839,
+      "step": 6610
+    },
+    {
+      "epoch": 27.02,
+      "grad_norm": 0.9643654823303223,
+      "learning_rate": 0.0002,
+      "loss": 0.8103,
+      "step": 6620
+    },
+    {
+      "epoch": 27.06,
+      "grad_norm": 0.9526975154876709,
+      "learning_rate": 0.0002,
+      "loss": 0.7568,
+      "step": 6630
+    },
+    {
+      "epoch": 27.1,
+      "grad_norm": 0.9363268613815308,
+      "learning_rate": 0.0002,
+      "loss": 0.768,
+      "step": 6640
+    },
+    {
+      "epoch": 27.14,
+      "grad_norm": 0.9408143162727356,
+      "learning_rate": 0.0002,
+      "loss": 0.7657,
+      "step": 6650
+    },
+    {
+      "epoch": 27.18,
+      "grad_norm": 0.9114474654197693,
+      "learning_rate": 0.0002,
+      "loss": 0.7704,
+      "step": 6660
+    },
+    {
+      "epoch": 27.22,
+      "grad_norm": 0.9404338002204895,
+      "learning_rate": 0.0002,
+      "loss": 0.7752,
+      "step": 6670
+    },
+    {
+      "epoch": 27.27,
+      "grad_norm": 0.924858570098877,
+      "learning_rate": 0.0002,
+      "loss": 0.7791,
+      "step": 6680
+    },
+    {
+      "epoch": 27.31,
+      "grad_norm": 0.8972033262252808,
+      "learning_rate": 0.0002,
+      "loss": 0.7875,
+      "step": 6690
+    },
+    {
+      "epoch": 27.35,
+      "grad_norm": 0.9168384075164795,
+      "learning_rate": 0.0002,
+      "loss": 0.7833,
+      "step": 6700
+    },
+    {
+      "epoch": 27.39,
+      "grad_norm": 0.9331495761871338,
+      "learning_rate": 0.0002,
+      "loss": 0.7934,
+      "step": 6710
+    },
+    {
+      "epoch": 27.43,
+      "grad_norm": 0.9218735694885254,
+      "learning_rate": 0.0002,
+      "loss": 0.8008,
+      "step": 6720
+    },
+    {
+      "epoch": 27.47,
+      "grad_norm": 0.9129732847213745,
+      "learning_rate": 0.0002,
+      "loss": 0.8017,
+      "step": 6730
+    },
+    {
+      "epoch": 27.51,
+      "grad_norm": 0.927643895149231,
+      "learning_rate": 0.0002,
+      "loss": 0.806,
+      "step": 6740
+    },
+    {
+      "epoch": 27.55,
+      "grad_norm": 0.9381563067436218,
+      "learning_rate": 0.0002,
+      "loss": 0.8121,
+      "step": 6750
+    },
+    {
+      "epoch": 27.59,
+      "grad_norm": 0.9063361287117004,
+      "learning_rate": 0.0002,
+      "loss": 0.8076,
+      "step": 6760
+    },
+    {
+      "epoch": 27.63,
+      "grad_norm": 0.9507086873054504,
+      "learning_rate": 0.0002,
+      "loss": 0.8201,
+      "step": 6770
+    },
+    {
+      "epoch": 27.67,
+      "grad_norm": 0.9145575165748596,
+      "learning_rate": 0.0002,
+      "loss": 0.8215,
+      "step": 6780
+    },
+    {
+      "epoch": 27.71,
+      "grad_norm": 0.9236624836921692,
+      "learning_rate": 0.0002,
+      "loss": 0.8217,
+      "step": 6790
+    },
+    {
+      "epoch": 27.76,
+      "grad_norm": 0.9380799531936646,
+      "learning_rate": 0.0002,
+      "loss": 0.8268,
+      "step": 6800
+    },
+    {
+      "epoch": 27.8,
+      "grad_norm": 0.9496703743934631,
+      "learning_rate": 0.0002,
+      "loss": 0.8285,
+      "step": 6810
+    },
+    {
+      "epoch": 27.84,
+      "grad_norm": 0.9168816208839417,
+      "learning_rate": 0.0002,
+      "loss": 0.8323,
+      "step": 6820
+    },
+    {
+      "epoch": 27.88,
+      "grad_norm": 0.9149271845817566,
+      "learning_rate": 0.0002,
+      "loss": 0.835,
+      "step": 6830
+    },
+    {
+      "epoch": 27.92,
+      "grad_norm": 0.996411144733429,
+      "learning_rate": 0.0002,
+      "loss": 0.837,
+      "step": 6840
+    },
+    {
+      "epoch": 27.96,
+      "grad_norm": 0.9247832894325256,
+      "learning_rate": 0.0002,
+      "loss": 0.8409,
+      "step": 6850
+    },
+    {
+      "epoch": 28.0,
+      "grad_norm": 0.9518089294433594,
+      "learning_rate": 0.0002,
+      "loss": 0.8404,
+      "step": 6860
+    },
+    {
+      "epoch": 28.04,
+      "grad_norm": 0.8640215396881104,
+      "learning_rate": 0.0002,
+      "loss": 0.7456,
+      "step": 6870
+    },
+    {
+      "epoch": 28.08,
+      "grad_norm": 0.9442835450172424,
+      "learning_rate": 0.0002,
+      "loss": 0.7483,
+      "step": 6880
+    },
+    {
+      "epoch": 28.12,
+      "grad_norm": 0.9471011757850647,
+      "learning_rate": 0.0002,
+      "loss": 0.7511,
+      "step": 6890
+    },
+    {
+      "epoch": 28.16,
+      "grad_norm": 0.9230690002441406,
+      "learning_rate": 0.0002,
+      "loss": 0.7554,
+      "step": 6900
+    },
+    {
+      "epoch": 28.2,
+      "grad_norm": 0.917255699634552,
+      "learning_rate": 0.0002,
+      "loss": 0.7686,
+      "step": 6910
+    },
+    {
+      "epoch": 28.24,
+      "grad_norm": 0.9495767951011658,
+      "learning_rate": 0.0002,
+      "loss": 0.7692,
+      "step": 6920
+    },
+    {
+      "epoch": 28.29,
+      "grad_norm": 0.9603391289710999,
+      "learning_rate": 0.0002,
+      "loss": 0.784,
+      "step": 6930
+    },
+    {
+      "epoch": 28.33,
+      "grad_norm": 0.9270275235176086,
+      "learning_rate": 0.0002,
+      "loss": 0.7817,
+      "step": 6940
+    },
+    {
+      "epoch": 28.37,
+      "grad_norm": 0.966315507888794,
+      "learning_rate": 0.0002,
+      "loss": 0.7857,
+      "step": 6950
+    },
+    {
+      "epoch": 28.41,
+      "grad_norm": 0.9359502196311951,
+      "learning_rate": 0.0002,
+      "loss": 0.7963,
+      "step": 6960
+    },
+    {
+      "epoch": 28.45,
+      "grad_norm": 0.9358251094818115,
+      "learning_rate": 0.0002,
+      "loss": 0.7962,
+      "step": 6970
+    },
+    {
+      "epoch": 28.49,
+      "grad_norm": 0.8844741582870483,
+      "learning_rate": 0.0002,
+      "loss": 0.8093,
+      "step": 6980
+    },
+    {
+      "epoch": 28.53,
+      "grad_norm": 0.9237121939659119,
+      "learning_rate": 0.0002,
+      "loss": 0.7941,
+      "step": 6990
+    },
+    {
+      "epoch": 28.57,
+      "grad_norm": 0.9162352681159973,
+      "learning_rate": 0.0002,
+      "loss": 0.8014,
+      "step": 7000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 10000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 41,
+  "save_steps": 100,
+  "total_flos": 2.74107564294144e+18,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}