diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5254 @@
+{
+  "best_global_step": 7400,
+  "best_metric": 0.02321171946823597,
+  "best_model_checkpoint": "./t5-med-ner/checkpoint-7400",
+  "epoch": 5.0,
+  "eval_steps": 500,
+  "global_step": 7400,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.006756756756756757,
+      "grad_norm": 16.129451751708984,
+      "learning_rate": 4.993918918918919e-05,
+      "loss": 15.3156,
+      "step": 10
+    },
+    {
+      "epoch": 0.013513513513513514,
+      "grad_norm": 18.336261749267578,
+      "learning_rate": 4.987162162162162e-05,
+      "loss": 12.4855,
+      "step": 20
+    },
+    {
+      "epoch": 0.02027027027027027,
+      "grad_norm": 12.824548721313477,
+      "learning_rate": 4.980405405405406e-05,
+      "loss": 9.9164,
+      "step": 30
+    },
+    {
+      "epoch": 0.02702702702702703,
+      "grad_norm": 12.065224647521973,
+      "learning_rate": 4.973648648648649e-05,
+      "loss": 8.696,
+      "step": 40
+    },
+    {
+      "epoch": 0.033783783783783786,
+      "grad_norm": 11.090178489685059,
+      "learning_rate": 4.966891891891892e-05,
+      "loss": 7.5063,
+      "step": 50
+    },
+    {
+      "epoch": 0.04054054054054054,
+      "grad_norm": 15.085448265075684,
+      "learning_rate": 4.9601351351351355e-05,
+      "loss": 6.2429,
+      "step": 60
+    },
+    {
+      "epoch": 0.0472972972972973,
+      "grad_norm": 13.112349510192871,
+      "learning_rate": 4.953378378378379e-05,
+      "loss": 5.2701,
+      "step": 70
+    },
+    {
+      "epoch": 0.05405405405405406,
+      "grad_norm": 18.061439514160156,
+      "learning_rate": 4.946621621621622e-05,
+      "loss": 4.3682,
+      "step": 80
+    },
+    {
+      "epoch": 0.060810810810810814,
+      "grad_norm": 10.440735816955566,
+      "learning_rate": 4.939864864864865e-05,
+      "loss": 3.6778,
+      "step": 90
+    },
+    {
+      "epoch": 0.06756756756756757,
+      "grad_norm": 11.351167678833008,
+      "learning_rate": 4.9331081081081084e-05,
+      "loss": 3.2569,
+      "step": 100
+    },
+    {
+      "epoch": 0.07432432432432433,
+      "grad_norm": 12.01860237121582,
+      "learning_rate": 4.926351351351352e-05,
+      "loss": 2.8868,
+      "step": 110
+    },
+    {
+      "epoch": 0.08108108108108109,
+      "grad_norm": 12.096404075622559,
+      "learning_rate": 4.919594594594595e-05,
+      "loss": 2.4611,
+      "step": 120
+    },
+    {
+      "epoch": 0.08783783783783784,
+      "grad_norm": 72.67072296142578,
+      "learning_rate": 4.912837837837838e-05,
+      "loss": 2.0179,
+      "step": 130
+    },
+    {
+      "epoch": 0.0945945945945946,
+      "grad_norm": 10.841756820678711,
+      "learning_rate": 4.9060810810810813e-05,
+      "loss": 1.6158,
+      "step": 140
+    },
+    {
+      "epoch": 0.10135135135135136,
+      "grad_norm": 8.544488906860352,
+      "learning_rate": 4.8993243243243246e-05,
+      "loss": 1.3312,
+      "step": 150
+    },
+    {
+      "epoch": 0.10810810810810811,
+      "grad_norm": 13.064513206481934,
+      "learning_rate": 4.892567567567568e-05,
+      "loss": 1.0426,
+      "step": 160
+    },
+    {
+      "epoch": 0.11486486486486487,
+      "grad_norm": 7.233090877532959,
+      "learning_rate": 4.885810810810811e-05,
+      "loss": 0.8803,
+      "step": 170
+    },
+    {
+      "epoch": 0.12162162162162163,
+      "grad_norm": 2.897141456604004,
+      "learning_rate": 4.879054054054054e-05,
+      "loss": 0.7543,
+      "step": 180
+    },
+    {
+      "epoch": 0.12837837837837837,
+      "grad_norm": 1.864625096321106,
+      "learning_rate": 4.8722972972972975e-05,
+      "loss": 0.6558,
+      "step": 190
+    },
+    {
+      "epoch": 0.13513513513513514,
+      "grad_norm": 1.1638946533203125,
+      "learning_rate": 4.865540540540541e-05,
+      "loss": 0.5462,
+      "step": 200
+    },
+    {
+      "epoch": 0.14189189189189189,
+      "grad_norm": 1.4550234079360962,
+      "learning_rate": 4.858783783783784e-05,
+      "loss": 0.4295,
+      "step": 210
+    },
+    {
+      "epoch": 0.14864864864864866,
+      "grad_norm": 1.272861361503601,
+      "learning_rate": 4.852027027027027e-05,
+      "loss": 0.4206,
+      "step": 220
+    },
+    {
+      "epoch": 0.1554054054054054,
+      "grad_norm": 1.448193907737732,
+      "learning_rate": 4.8452702702702704e-05,
+      "loss": 0.3532,
+      "step": 230
+    },
+    {
+      "epoch": 0.16216216216216217,
+      "grad_norm": 0.8406581878662109,
+      "learning_rate": 4.838513513513514e-05,
+      "loss": 0.3442,
+      "step": 240
+    },
+    {
+      "epoch": 0.16891891891891891,
+      "grad_norm": 1.0974594354629517,
+      "learning_rate": 4.831756756756757e-05,
+      "loss": 0.3011,
+      "step": 250
+    },
+    {
+      "epoch": 0.17567567567567569,
+      "grad_norm": 10.254583358764648,
+      "learning_rate": 4.825e-05,
+      "loss": 0.2916,
+      "step": 260
+    },
+    {
+      "epoch": 0.18243243243243243,
+      "grad_norm": 1.023024320602417,
+      "learning_rate": 4.818243243243243e-05,
+      "loss": 0.301,
+      "step": 270
+    },
+    {
+      "epoch": 0.1891891891891892,
+      "grad_norm": 0.8870915770530701,
+      "learning_rate": 4.8114864864864865e-05,
+      "loss": 0.2768,
+      "step": 280
+    },
+    {
+      "epoch": 0.19594594594594594,
+      "grad_norm": 1.4651856422424316,
+      "learning_rate": 4.80472972972973e-05,
+      "loss": 0.2543,
+      "step": 290
+    },
+    {
+      "epoch": 0.20270270270270271,
+      "grad_norm": 0.8050569891929626,
+      "learning_rate": 4.7979729729729736e-05,
+      "loss": 0.2513,
+      "step": 300
+    },
+    {
+      "epoch": 0.20945945945945946,
+      "grad_norm": 0.7741556763648987,
+      "learning_rate": 4.791216216216217e-05,
+      "loss": 0.2343,
+      "step": 310
+    },
+    {
+      "epoch": 0.21621621621621623,
+      "grad_norm": 0.7951841354370117,
+      "learning_rate": 4.78445945945946e-05,
+      "loss": 0.2058,
+      "step": 320
+    },
+    {
+      "epoch": 0.22297297297297297,
+      "grad_norm": 4.641398906707764,
+      "learning_rate": 4.7777027027027026e-05,
+      "loss": 0.2249,
+      "step": 330
+    },
+    {
+      "epoch": 0.22972972972972974,
+      "grad_norm": 0.795923113822937,
+      "learning_rate": 4.770945945945946e-05,
+      "loss": 0.1895,
+      "step": 340
+    },
+    {
+      "epoch": 0.23648648648648649,
+      "grad_norm": 1.0727757215499878,
+      "learning_rate": 4.764189189189189e-05,
+      "loss": 0.1918,
+      "step": 350
+    },
+    {
+      "epoch": 0.24324324324324326,
+      "grad_norm": 3.4822981357574463,
+      "learning_rate": 4.757432432432432e-05,
+      "loss": 0.2068,
+      "step": 360
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.637052059173584,
+      "learning_rate": 4.750675675675676e-05,
+      "loss": 0.1923,
+      "step": 370
+    },
+    {
+      "epoch": 0.25675675675675674,
+      "grad_norm": 0.893659770488739,
+      "learning_rate": 4.7439189189189194e-05,
+      "loss": 0.176,
+      "step": 380
+    },
+    {
+      "epoch": 0.2635135135135135,
+      "grad_norm": 0.8992719650268555,
+      "learning_rate": 4.7371621621621627e-05,
+      "loss": 0.185,
+      "step": 390
+    },
+    {
+      "epoch": 0.2702702702702703,
+      "grad_norm": 1.35341477394104,
+      "learning_rate": 4.730405405405406e-05,
+      "loss": 0.2073,
+      "step": 400
+    },
+    {
+      "epoch": 0.27702702702702703,
+      "grad_norm": 0.8321266770362854,
+      "learning_rate": 4.7236486486486484e-05,
+      "loss": 0.1682,
+      "step": 410
+    },
+    {
+      "epoch": 0.28378378378378377,
+      "grad_norm": 0.7957965731620789,
+      "learning_rate": 4.7168918918918917e-05,
+      "loss": 0.1377,
+      "step": 420
+    },
+    {
+      "epoch": 0.2905405405405405,
+      "grad_norm": 1.0086628198623657,
+      "learning_rate": 4.7101351351351356e-05,
+      "loss": 0.1468,
+      "step": 430
+    },
+    {
+      "epoch": 0.2972972972972973,
+      "grad_norm": 1.8587403297424316,
+      "learning_rate": 4.703378378378379e-05,
+      "loss": 0.1559,
+      "step": 440
+    },
+    {
+      "epoch": 0.30405405405405406,
+      "grad_norm": 0.9978078603744507,
+      "learning_rate": 4.696621621621622e-05,
+      "loss": 0.1582,
+      "step": 450
+    },
+    {
+      "epoch": 0.3108108108108108,
+      "grad_norm": 0.7275466918945312,
+      "learning_rate": 4.689864864864865e-05,
+      "loss": 0.1515,
+      "step": 460
+    },
+    {
+      "epoch": 0.31756756756756754,
+      "grad_norm": 0.6467192769050598,
+      "learning_rate": 4.6831081081081085e-05,
+      "loss": 0.1469,
+      "step": 470
+    },
+    {
+      "epoch": 0.32432432432432434,
+      "grad_norm": 0.5906888246536255,
+      "learning_rate": 4.676351351351352e-05,
+      "loss": 0.1566,
+      "step": 480
+    },
+    {
+      "epoch": 0.3310810810810811,
+      "grad_norm": 0.6035364866256714,
+      "learning_rate": 4.669594594594595e-05,
+      "loss": 0.1376,
+      "step": 490
+    },
+    {
+      "epoch": 0.33783783783783783,
+      "grad_norm": 0.7139237523078918,
+      "learning_rate": 4.662837837837838e-05,
+      "loss": 0.1404,
+      "step": 500
+    },
+    {
+      "epoch": 0.34459459459459457,
+      "grad_norm": 0.818231999874115,
+      "learning_rate": 4.6560810810810814e-05,
+      "loss": 0.1406,
+      "step": 510
+    },
+    {
+      "epoch": 0.35135135135135137,
+      "grad_norm": 0.4550321102142334,
+      "learning_rate": 4.6493243243243246e-05,
+      "loss": 0.1283,
+      "step": 520
+    },
+    {
+      "epoch": 0.3581081081081081,
+      "grad_norm": 0.7775529623031616,
+      "learning_rate": 4.642567567567568e-05,
+      "loss": 0.1203,
+      "step": 530
+    },
+    {
+      "epoch": 0.36486486486486486,
+      "grad_norm": 0.6731743812561035,
+      "learning_rate": 4.635810810810811e-05,
+      "loss": 0.1187,
+      "step": 540
+    },
+    {
+      "epoch": 0.3716216216216216,
+      "grad_norm": 0.5695653557777405,
+      "learning_rate": 4.629054054054054e-05,
+      "loss": 0.1206,
+      "step": 550
+    },
+    {
+      "epoch": 0.3783783783783784,
+      "grad_norm": 1.0783650875091553,
+      "learning_rate": 4.6222972972972975e-05,
+      "loss": 0.1233,
+      "step": 560
+    },
+    {
+      "epoch": 0.38513513513513514,
+      "grad_norm": 0.4626968801021576,
+      "learning_rate": 4.615540540540541e-05,
+      "loss": 0.1104,
+      "step": 570
+    },
+    {
+      "epoch": 0.3918918918918919,
+      "grad_norm": 0.6568089723587036,
+      "learning_rate": 4.608783783783784e-05,
+      "loss": 0.1509,
+      "step": 580
+    },
+    {
+      "epoch": 0.39864864864864863,
+      "grad_norm": 0.5651082396507263,
+      "learning_rate": 4.602027027027027e-05,
+      "loss": 0.1167,
+      "step": 590
+    },
+    {
+      "epoch": 0.40540540540540543,
+      "grad_norm": 0.8560924530029297,
+      "learning_rate": 4.5952702702702704e-05,
+      "loss": 0.1245,
+      "step": 600
+    },
+    {
+      "epoch": 0.41216216216216217,
+      "grad_norm": 0.8468086123466492,
+      "learning_rate": 4.5885135135135136e-05,
+      "loss": 0.1198,
+      "step": 610
+    },
+    {
+      "epoch": 0.4189189189189189,
+      "grad_norm": 8.868446350097656,
+      "learning_rate": 4.581756756756757e-05,
+      "loss": 0.1022,
+      "step": 620
+    },
+    {
+      "epoch": 0.42567567567567566,
+      "grad_norm": 0.7043650150299072,
+      "learning_rate": 4.575e-05,
+      "loss": 0.1227,
+      "step": 630
+    },
+    {
+      "epoch": 0.43243243243243246,
+      "grad_norm": 0.6890026926994324,
+      "learning_rate": 4.568243243243244e-05,
+      "loss": 0.1158,
+      "step": 640
+    },
+    {
+      "epoch": 0.4391891891891892,
+      "grad_norm": 0.48028209805488586,
+      "learning_rate": 4.5614864864864865e-05,
+      "loss": 0.0918,
+      "step": 650
+    },
+    {
+      "epoch": 0.44594594594594594,
+      "grad_norm": 0.5559847354888916,
+      "learning_rate": 4.55472972972973e-05,
+      "loss": 0.0968,
+      "step": 660
+    },
+    {
+      "epoch": 0.4527027027027027,
+      "grad_norm": 2.659057855606079,
+      "learning_rate": 4.547972972972973e-05,
+      "loss": 0.1308,
+      "step": 670
+    },
+    {
+      "epoch": 0.4594594594594595,
+      "grad_norm": 0.3449746370315552,
+      "learning_rate": 4.541216216216216e-05,
+      "loss": 0.0965,
+      "step": 680
+    },
+    {
+      "epoch": 0.46621621621621623,
+      "grad_norm": 0.4225215017795563,
+      "learning_rate": 4.5344594594594594e-05,
+      "loss": 0.0989,
+      "step": 690
+    },
+    {
+      "epoch": 0.47297297297297297,
+      "grad_norm": 0.5477343201637268,
+      "learning_rate": 4.527702702702703e-05,
+      "loss": 0.1088,
+      "step": 700
+    },
+    {
+      "epoch": 0.4797297297297297,
+      "grad_norm": 0.552731990814209,
+      "learning_rate": 4.5209459459459466e-05,
+      "loss": 0.0935,
+      "step": 710
+    },
+    {
+      "epoch": 0.4864864864864865,
+      "grad_norm": 0.4366927444934845,
+      "learning_rate": 4.51418918918919e-05,
+      "loss": 0.101,
+      "step": 720
+    },
+    {
+      "epoch": 0.49324324324324326,
+      "grad_norm": 0.49442631006240845,
+      "learning_rate": 4.507432432432432e-05,
+      "loss": 0.1172,
+      "step": 730
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.43195462226867676,
+      "learning_rate": 4.5006756756756756e-05,
+      "loss": 0.1102,
+      "step": 740
+    },
+    {
+      "epoch": 0.5067567567567568,
+      "grad_norm": 0.9043081998825073,
+      "learning_rate": 4.493918918918919e-05,
+      "loss": 0.1347,
+      "step": 750
+    },
+    {
+      "epoch": 0.5135135135135135,
+      "grad_norm": 0.5600760579109192,
+      "learning_rate": 4.487162162162162e-05,
+      "loss": 0.0974,
+      "step": 760
+    },
+    {
+      "epoch": 0.5202702702702703,
+      "grad_norm": 0.7427008152008057,
+      "learning_rate": 4.480405405405406e-05,
+      "loss": 0.0772,
+      "step": 770
+    },
+    {
+      "epoch": 0.527027027027027,
+      "grad_norm": 1.1261482238769531,
+      "learning_rate": 4.473648648648649e-05,
+      "loss": 0.104,
+      "step": 780
+    },
+    {
+      "epoch": 0.5337837837837838,
+      "grad_norm": 0.8397054672241211,
+      "learning_rate": 4.4668918918918924e-05,
+      "loss": 0.1021,
+      "step": 790
+    },
+    {
+      "epoch": 0.5405405405405406,
+      "grad_norm": 0.4203931391239166,
+      "learning_rate": 4.4601351351351356e-05,
+      "loss": 0.1103,
+      "step": 800
+    },
+    {
+      "epoch": 0.5472972972972973,
+      "grad_norm": 0.6614381670951843,
+      "learning_rate": 4.453378378378378e-05,
+      "loss": 0.1014,
+      "step": 810
+    },
+    {
+      "epoch": 0.5540540540540541,
+      "grad_norm": 0.3424564301967621,
+      "learning_rate": 4.4466216216216214e-05,
+      "loss": 0.0834,
+      "step": 820
+    },
+    {
+      "epoch": 0.5608108108108109,
+      "grad_norm": 0.6263936758041382,
+      "learning_rate": 4.439864864864865e-05,
+      "loss": 0.0942,
+      "step": 830
+    },
+    {
+      "epoch": 0.5675675675675675,
+      "grad_norm": 0.5943464636802673,
+      "learning_rate": 4.4331081081081085e-05,
+      "loss": 0.085,
+      "step": 840
+    },
+    {
+      "epoch": 0.5743243243243243,
+      "grad_norm": 0.4014379680156708,
+      "learning_rate": 4.426351351351352e-05,
+      "loss": 0.0879,
+      "step": 850
+    },
+    {
+      "epoch": 0.581081081081081,
+      "grad_norm": 0.5546994209289551,
+      "learning_rate": 4.419594594594595e-05,
+      "loss": 0.0751,
+      "step": 860
+    },
+    {
+      "epoch": 0.5878378378378378,
+      "grad_norm": 0.5836130380630493,
+      "learning_rate": 4.412837837837838e-05,
+      "loss": 0.1219,
+      "step": 870
+    },
+    {
+      "epoch": 0.5945945945945946,
+      "grad_norm": 0.6041303873062134,
+      "learning_rate": 4.4060810810810814e-05,
+      "loss": 0.0982,
+      "step": 880
+    },
+    {
+      "epoch": 0.6013513513513513,
+      "grad_norm": 1.5622942447662354,
+      "learning_rate": 4.3993243243243246e-05,
+      "loss": 0.0923,
+      "step": 890
+    },
+    {
+      "epoch": 0.6081081081081081,
+      "grad_norm": 0.4068175256252289,
+      "learning_rate": 4.392567567567568e-05,
+      "loss": 0.075,
+      "step": 900
+    },
+    {
+      "epoch": 0.6148648648648649,
+      "grad_norm": 0.4515470862388611,
+      "learning_rate": 4.385810810810811e-05,
+      "loss": 0.0845,
+      "step": 910
+    },
+    {
+      "epoch": 0.6216216216216216,
+      "grad_norm": 0.500089168548584,
+      "learning_rate": 4.379054054054054e-05,
+      "loss": 0.0939,
+      "step": 920
+    },
+    {
+      "epoch": 0.6283783783783784,
+      "grad_norm": 0.8668215274810791,
+      "learning_rate": 4.3722972972972975e-05,
+      "loss": 0.094,
+      "step": 930
+    },
+    {
+      "epoch": 0.6351351351351351,
+      "grad_norm": 0.5503737926483154,
+      "learning_rate": 4.365540540540541e-05,
+      "loss": 0.0936,
+      "step": 940
+    },
+    {
+      "epoch": 0.6418918918918919,
+      "grad_norm": 0.6960216760635376,
+      "learning_rate": 4.358783783783784e-05,
+      "loss": 0.0847,
+      "step": 950
+    },
+    {
+      "epoch": 0.6486486486486487,
+      "grad_norm": 0.5142303705215454,
+      "learning_rate": 4.352027027027027e-05,
+      "loss": 0.0906,
+      "step": 960
+    },
+    {
+      "epoch": 0.6554054054054054,
+      "grad_norm": 0.775428056716919,
+      "learning_rate": 4.3452702702702704e-05,
+      "loss": 0.0881,
+      "step": 970
+    },
+    {
+      "epoch": 0.6621621621621622,
+      "grad_norm": 0.43797916173934937,
+      "learning_rate": 4.3385135135135136e-05,
+      "loss": 0.0898,
+      "step": 980
+    },
+    {
+      "epoch": 0.668918918918919,
+      "grad_norm": 0.743198812007904,
+      "learning_rate": 4.331756756756757e-05,
+      "loss": 0.0751,
+      "step": 990
+    },
+    {
+      "epoch": 0.6756756756756757,
+      "grad_norm": 1.0289562940597534,
+      "learning_rate": 4.325e-05,
+      "loss": 0.1113,
+      "step": 1000
+    },
+    {
+      "epoch": 0.6824324324324325,
+      "grad_norm": 0.5582059621810913,
+      "learning_rate": 4.318243243243243e-05,
+      "loss": 0.0817,
+      "step": 1010
+    },
+    {
+      "epoch": 0.6891891891891891,
+      "grad_norm": 0.42011767625808716,
+      "learning_rate": 4.3114864864864865e-05,
+      "loss": 0.0843,
+      "step": 1020
+    },
+    {
+      "epoch": 0.6959459459459459,
+      "grad_norm": 0.6274346709251404,
+      "learning_rate": 4.30472972972973e-05,
+      "loss": 0.0869,
+      "step": 1030
+    },
+    {
+      "epoch": 0.7027027027027027,
+      "grad_norm": 0.3823902904987335,
+      "learning_rate": 4.297972972972974e-05,
+      "loss": 0.0739,
+      "step": 1040
+    },
+    {
+      "epoch": 0.7094594594594594,
+      "grad_norm": 0.3813062608242035,
+      "learning_rate": 4.291216216216216e-05,
+      "loss": 0.0972,
+      "step": 1050
+    },
+    {
+      "epoch": 0.7162162162162162,
+      "grad_norm": 0.5466370582580566,
+      "learning_rate": 4.2844594594594594e-05,
+      "loss": 0.0955,
+      "step": 1060
+    },
+    {
+      "epoch": 0.722972972972973,
+      "grad_norm": 0.3452337980270386,
+      "learning_rate": 4.277702702702703e-05,
+      "loss": 0.0755,
+      "step": 1070
+    },
+    {
+      "epoch": 0.7297297297297297,
+      "grad_norm": 0.7460477948188782,
+      "learning_rate": 4.270945945945946e-05,
+      "loss": 0.0776,
+      "step": 1080
+    },
+    {
+      "epoch": 0.7364864864864865,
+      "grad_norm": 0.6906008124351501,
+      "learning_rate": 4.264189189189189e-05,
+      "loss": 0.0749,
+      "step": 1090
+    },
+    {
+      "epoch": 0.7432432432432432,
+      "grad_norm": 0.5057837963104248,
+      "learning_rate": 4.257432432432433e-05,
+      "loss": 0.0701,
+      "step": 1100
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.6320849061012268,
+      "learning_rate": 4.250675675675676e-05,
+      "loss": 0.0785,
+      "step": 1110
+    },
+    {
+      "epoch": 0.7567567567567568,
+      "grad_norm": 1.015070915222168,
+      "learning_rate": 4.2439189189189195e-05,
+      "loss": 0.0727,
+      "step": 1120
+    },
+    {
+      "epoch": 0.7635135135135135,
+      "grad_norm": 0.5887625217437744,
+      "learning_rate": 4.237162162162162e-05,
+      "loss": 0.0666,
+      "step": 1130
+    },
+    {
+      "epoch": 0.7702702702702703,
+      "grad_norm": 0.5994888544082642,
+      "learning_rate": 4.230405405405405e-05,
+      "loss": 0.071,
+      "step": 1140
+    },
+    {
+      "epoch": 0.777027027027027,
+      "grad_norm": 0.6155841946601868,
+      "learning_rate": 4.2236486486486485e-05,
+      "loss": 0.0735,
+      "step": 1150
+    },
+    {
+      "epoch": 0.7837837837837838,
+      "grad_norm": 0.8194956183433533,
+      "learning_rate": 4.2168918918918924e-05,
+      "loss": 0.0699,
+      "step": 1160
+    },
+    {
+      "epoch": 0.7905405405405406,
+      "grad_norm": 0.3856644034385681,
+      "learning_rate": 4.2101351351351356e-05,
+      "loss": 0.089,
+      "step": 1170
+    },
+    {
+      "epoch": 0.7972972972972973,
+      "grad_norm": 0.7099316120147705,
+      "learning_rate": 4.203378378378379e-05,
+      "loss": 0.0572,
+      "step": 1180
+    },
+    {
+      "epoch": 0.8040540540540541,
+      "grad_norm": 0.4837072789669037,
+      "learning_rate": 4.196621621621622e-05,
+      "loss": 0.0598,
+      "step": 1190
+    },
+    {
+      "epoch": 0.8108108108108109,
+      "grad_norm": 1.0372607707977295,
+      "learning_rate": 4.189864864864865e-05,
+      "loss": 0.0896,
+      "step": 1200
+    },
+    {
+      "epoch": 0.8175675675675675,
+      "grad_norm": 0.36857545375823975,
+      "learning_rate": 4.183108108108108e-05,
+      "loss": 0.0766,
+      "step": 1210
+    },
+    {
+      "epoch": 0.8243243243243243,
+      "grad_norm": 0.6140791177749634,
+      "learning_rate": 4.176351351351351e-05,
+      "loss": 0.0651,
+      "step": 1220
+    },
+    {
+      "epoch": 0.831081081081081,
+      "grad_norm": 0.4973204433917999,
+      "learning_rate": 4.169594594594595e-05,
+      "loss": 0.0771,
+      "step": 1230
+    },
+    {
+      "epoch": 0.8378378378378378,
+      "grad_norm": 0.8992588520050049,
+      "learning_rate": 4.162837837837838e-05,
+      "loss": 0.0803,
+      "step": 1240
+    },
+    {
+      "epoch": 0.8445945945945946,
+      "grad_norm": 0.4852398633956909,
+      "learning_rate": 4.1560810810810814e-05,
+      "loss": 0.0864,
+      "step": 1250
+    },
+    {
+      "epoch": 0.8513513513513513,
+      "grad_norm": 0.7607585787773132,
+      "learning_rate": 4.1493243243243246e-05,
+      "loss": 0.0726,
+      "step": 1260
+    },
+    {
+      "epoch": 0.8581081081081081,
+      "grad_norm": 0.7050380706787109,
+      "learning_rate": 4.142567567567568e-05,
+      "loss": 0.0736,
+      "step": 1270
+    },
+    {
+      "epoch": 0.8648648648648649,
+      "grad_norm": 0.6202526688575745,
+      "learning_rate": 4.135810810810811e-05,
+      "loss": 0.0632,
+      "step": 1280
+    },
+    {
+      "epoch": 0.8716216216216216,
+      "grad_norm": 1.4994258880615234,
+      "learning_rate": 4.129054054054054e-05,
+      "loss": 0.0745,
+      "step": 1290
+    },
+    {
+      "epoch": 0.8783783783783784,
+      "grad_norm": 0.4370647668838501,
+      "learning_rate": 4.1222972972972975e-05,
+      "loss": 0.0782,
+      "step": 1300
+    },
+    {
+      "epoch": 0.8851351351351351,
+      "grad_norm": 0.28213265538215637,
+      "learning_rate": 4.115540540540541e-05,
+      "loss": 0.0735,
+      "step": 1310
+    },
+    {
+      "epoch": 0.8918918918918919,
+      "grad_norm": 0.31794193387031555,
+      "learning_rate": 4.108783783783784e-05,
+      "loss": 0.0543,
+      "step": 1320
+    },
+    {
+      "epoch": 0.8986486486486487,
+      "grad_norm": 0.5807836055755615,
+      "learning_rate": 4.102027027027027e-05,
+      "loss": 0.0688,
+      "step": 1330
+    },
+    {
+      "epoch": 0.9054054054054054,
+      "grad_norm": 0.5014552474021912,
+      "learning_rate": 4.0952702702702704e-05,
+      "loss": 0.0659,
+      "step": 1340
+    },
+    {
+      "epoch": 0.9121621621621622,
+      "grad_norm": 0.3839617371559143,
+      "learning_rate": 4.088513513513514e-05,
+      "loss": 0.0655,
+      "step": 1350
+    },
+    {
+      "epoch": 0.918918918918919,
+      "grad_norm": 0.6149204969406128,
+      "learning_rate": 4.081756756756757e-05,
+      "loss": 0.0693,
+      "step": 1360
+    },
+    {
+      "epoch": 0.9256756756756757,
+      "grad_norm": 0.49168112874031067,
+      "learning_rate": 4.075e-05,
+      "loss": 0.06,
+      "step": 1370
+    },
+    {
+      "epoch": 0.9324324324324325,
+      "grad_norm": 0.6225050687789917,
+      "learning_rate": 4.0682432432432433e-05,
+      "loss": 0.0762,
+      "step": 1380
+    },
+    {
+      "epoch": 0.9391891891891891,
+      "grad_norm": 0.5382749438285828,
+      "learning_rate": 4.0614864864864866e-05,
+      "loss": 0.0695,
+      "step": 1390
+    },
+    {
+      "epoch": 0.9459459459459459,
+      "grad_norm": 0.8806264400482178,
+      "learning_rate": 4.05472972972973e-05,
+      "loss": 0.0847,
+      "step": 1400
+    },
+    {
+      "epoch": 0.9527027027027027,
+      "grad_norm": 0.6600827574729919,
+      "learning_rate": 4.047972972972973e-05,
+      "loss": 0.072,
+      "step": 1410
+    },
+    {
+      "epoch": 0.9594594594594594,
+      "grad_norm": 0.2923494279384613,
+      "learning_rate": 4.041216216216216e-05,
+      "loss": 0.0647,
+      "step": 1420
+    },
+    {
+      "epoch": 0.9662162162162162,
+      "grad_norm": 0.40617331862449646,
+      "learning_rate": 4.0344594594594595e-05,
+      "loss": 0.058,
+      "step": 1430
+    },
+    {
+      "epoch": 0.972972972972973,
+      "grad_norm": 0.5751014947891235,
+      "learning_rate": 4.0277027027027034e-05,
+      "loss": 0.0685,
+      "step": 1440
+    },
+    {
+      "epoch": 0.9797297297297297,
+      "grad_norm": 0.4234154224395752,
+      "learning_rate": 4.020945945945946e-05,
+      "loss": 0.0793,
+      "step": 1450
+    },
+    {
+      "epoch": 0.9864864864864865,
+      "grad_norm": 0.6290509700775146,
+      "learning_rate": 4.014189189189189e-05,
+      "loss": 0.0705,
+      "step": 1460
+    },
+    {
+      "epoch": 0.9932432432432432,
+      "grad_norm": 0.5429525375366211,
+      "learning_rate": 4.0074324324324324e-05,
+      "loss": 0.0694,
+      "step": 1470
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.3089545965194702,
+      "learning_rate": 4.0006756756756756e-05,
+      "loss": 0.0579,
+      "step": 1480
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 0.03719908744096756,
+      "eval_runtime": 267.8762,
+      "eval_samples_per_second": 3.901,
+      "eval_steps_per_second": 0.978,
+      "step": 1480
+    },
+    {
+      "epoch": 1.0067567567567568,
+      "grad_norm": 0.429411381483078,
+      "learning_rate": 3.993918918918919e-05,
+      "loss": 0.0615,
+      "step": 1490
+    },
+    {
+      "epoch": 1.0135135135135136,
+      "grad_norm": 0.4686361253261566,
+      "learning_rate": 3.987162162162163e-05,
+      "loss": 0.0774,
+      "step": 1500
+    },
+    {
+      "epoch": 1.0202702702702702,
+      "grad_norm": 0.38840433955192566,
+      "learning_rate": 3.980405405405406e-05,
+      "loss": 0.0499,
+      "step": 1510
+    },
+    {
+      "epoch": 1.027027027027027,
+      "grad_norm": 0.6909990906715393,
+      "learning_rate": 3.973648648648649e-05,
+      "loss": 0.0698,
+      "step": 1520
+    },
+    {
+      "epoch": 1.0337837837837838,
+      "grad_norm": 1.1144709587097168,
+      "learning_rate": 3.966891891891892e-05,
+      "loss": 0.0735,
+      "step": 1530
+    },
+    {
+      "epoch": 1.0405405405405406,
+      "grad_norm": 0.37668439745903015,
+      "learning_rate": 3.960135135135135e-05,
+      "loss": 0.0529,
+      "step": 1540
+    },
+    {
+      "epoch": 1.0472972972972974,
+      "grad_norm": 0.5073342323303223,
+      "learning_rate": 3.953378378378378e-05,
+      "loss": 0.0667,
+      "step": 1550
+    },
+    {
+      "epoch": 1.054054054054054,
+      "grad_norm": 0.4015592038631439,
+      "learning_rate": 3.946621621621622e-05,
+      "loss": 0.0625,
+      "step": 1560
+    },
+    {
+      "epoch": 1.0608108108108107,
+      "grad_norm": 0.7579006552696228,
+      "learning_rate": 3.939864864864865e-05,
+      "loss": 0.06,
+      "step": 1570
+    },
+    {
+      "epoch": 1.0675675675675675,
+      "grad_norm": 0.7990680932998657,
+      "learning_rate": 3.9331081081081085e-05,
+      "loss": 0.0593,
+      "step": 1580
+    },
+    {
+      "epoch": 1.0743243243243243,
+      "grad_norm": 0.6557222604751587,
+      "learning_rate": 3.926351351351352e-05,
+      "loss": 0.0726,
+      "step": 1590
+    },
+    {
+      "epoch": 1.0810810810810811,
+      "grad_norm": 0.5203543305397034,
+      "learning_rate": 3.919594594594595e-05,
+      "loss": 0.0918,
+      "step": 1600
+    },
+    {
+      "epoch": 1.087837837837838,
+      "grad_norm": 0.5405670404434204,
+      "learning_rate": 3.9128378378378375e-05,
+      "loss": 0.0712,
+      "step": 1610
+    },
+    {
+      "epoch": 1.0945945945945945,
+      "grad_norm": 0.36374953389167786,
+      "learning_rate": 3.906081081081081e-05,
+      "loss": 0.0608,
+      "step": 1620
+    },
+    {
+      "epoch": 1.1013513513513513,
+      "grad_norm": 0.3525446653366089,
+      "learning_rate": 3.8993243243243247e-05,
+      "loss": 0.051,
+      "step": 1630
+    },
+    {
+      "epoch": 1.1081081081081081,
+      "grad_norm": 0.38730859756469727,
+      "learning_rate": 3.892567567567568e-05,
+      "loss": 0.0652,
+      "step": 1640
+    },
+    {
+      "epoch": 1.114864864864865,
+      "grad_norm": 0.3817211389541626,
+      "learning_rate": 3.885810810810811e-05,
+      "loss": 0.0702,
+      "step": 1650
+    },
+    {
+      "epoch": 1.1216216216216217,
+      "grad_norm": 0.29431015253067017,
+      "learning_rate": 3.879054054054054e-05,
+      "loss": 0.0607,
+      "step": 1660
+    },
+    {
+      "epoch": 1.1283783783783783,
+      "grad_norm": 0.5619912147521973,
+      "learning_rate": 3.8722972972972976e-05,
+      "loss": 0.0564,
+      "step": 1670
+    },
+    {
+      "epoch": 1.135135135135135,
+      "grad_norm": 0.5539348125457764,
+      "learning_rate": 3.865540540540541e-05,
+      "loss": 0.0683,
+      "step": 1680
+    },
+    {
+      "epoch": 1.1418918918918919,
+      "grad_norm": 0.7165750861167908,
+      "learning_rate": 3.858783783783784e-05,
+      "loss": 0.0701,
+      "step": 1690
+    },
+    {
+      "epoch": 1.1486486486486487,
+      "grad_norm": 0.3067423403263092,
+      "learning_rate": 3.852027027027027e-05,
+      "loss": 0.0594,
+      "step": 1700
+    },
+    {
+      "epoch": 1.1554054054054055,
+      "grad_norm": 0.33841240406036377,
+      "learning_rate": 3.8452702702702705e-05,
+      "loss": 0.0446,
+      "step": 1710
+    },
+    {
+      "epoch": 1.1621621621621623,
+      "grad_norm": 0.3610815405845642,
+      "learning_rate": 3.838513513513514e-05,
+      "loss": 0.0607,
+      "step": 1720
+    },
+    {
+      "epoch": 1.1689189189189189,
+      "grad_norm": 0.5924748778343201,
+      "learning_rate": 3.831756756756757e-05,
+      "loss": 0.0641,
+      "step": 1730
+    },
+    {
+      "epoch": 1.1756756756756757,
+      "grad_norm": 0.39871370792388916,
+      "learning_rate": 3.825e-05,
+      "loss": 0.0558,
+      "step": 1740
+    },
+    {
+      "epoch": 1.1824324324324325,
+      "grad_norm": 0.5780407190322876,
+      "learning_rate": 3.8182432432432434e-05,
+      "loss": 0.0603,
+      "step": 1750
+    },
+    {
+      "epoch": 1.1891891891891893,
+      "grad_norm": 0.6301029920578003,
+      "learning_rate": 3.8114864864864866e-05,
+      "loss": 0.0553,
+      "step": 1760
+    },
+    {
+      "epoch": 1.195945945945946,
+      "grad_norm": 0.4417901039123535,
+      "learning_rate": 3.8047297297297305e-05,
+      "loss": 0.0706,
+      "step": 1770
+    },
+    {
+      "epoch": 1.2027027027027026,
+      "grad_norm": 0.5784780979156494,
+      "learning_rate": 3.797972972972973e-05,
+      "loss": 0.0552,
+      "step": 1780
+    },
+    {
+      "epoch": 1.2094594594594594,
+      "grad_norm": 0.3300984799861908,
+      "learning_rate": 3.791216216216216e-05,
+      "loss": 0.0539,
+      "step": 1790
+    },
+    {
+      "epoch": 1.2162162162162162,
+      "grad_norm": 0.28047096729278564,
+      "learning_rate": 3.7844594594594595e-05,
+      "loss": 0.0599,
+      "step": 1800
+    },
+    {
+      "epoch": 1.222972972972973,
+      "grad_norm": 0.3282848000526428,
+      "learning_rate": 3.777702702702703e-05,
+      "loss": 0.0757,
+      "step": 1810
+    },
+    {
+      "epoch": 1.2297297297297298,
+      "grad_norm": 0.5447771549224854,
+      "learning_rate": 3.770945945945946e-05,
+      "loss": 0.054,
+      "step": 1820
+    },
+    {
+      "epoch": 1.2364864864864864,
+      "grad_norm": 0.38858866691589355,
+      "learning_rate": 3.764189189189189e-05,
+      "loss": 0.0568,
+      "step": 1830
+    },
+    {
+      "epoch": 1.2432432432432432,
+      "grad_norm": 0.553166925907135,
+      "learning_rate": 3.757432432432433e-05,
+      "loss": 0.0417,
+      "step": 1840
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.3401406705379486,
+      "learning_rate": 3.750675675675676e-05,
+      "loss": 0.081,
+      "step": 1850
+    },
+    {
+      "epoch": 1.2567567567567568,
+      "grad_norm": 0.43225792050361633,
+      "learning_rate": 3.743918918918919e-05,
+      "loss": 0.0487,
+      "step": 1860
+    },
+    {
+      "epoch": 1.2635135135135136,
+      "grad_norm": 0.33819809556007385,
+      "learning_rate": 3.737162162162162e-05,
+      "loss": 0.0567,
+      "step": 1870
+    },
+    {
+      "epoch": 1.2702702702702702,
+      "grad_norm": 0.3383878469467163,
+      "learning_rate": 3.730405405405405e-05,
+      "loss": 0.0474,
+      "step": 1880
+    },
+    {
+      "epoch": 1.277027027027027,
+      "grad_norm": 0.5455739498138428,
+      "learning_rate": 3.7236486486486485e-05,
+      "loss": 0.0508,
+      "step": 1890
+    },
+    {
+      "epoch": 1.2837837837837838,
+      "grad_norm": 0.6346222758293152,
+      "learning_rate": 3.7168918918918924e-05,
+      "loss": 0.0531,
+      "step": 1900
+    },
+    {
+      "epoch": 1.2905405405405406,
+      "grad_norm": 0.3384900689125061,
+      "learning_rate": 3.7101351351351357e-05,
+      "loss": 0.0604,
+      "step": 1910
+    },
+    {
+      "epoch": 1.2972972972972974,
+      "grad_norm": 0.44982558488845825,
+      "learning_rate": 3.703378378378379e-05,
+      "loss": 0.0663,
+      "step": 1920
+    },
+    {
+      "epoch": 1.304054054054054,
+      "grad_norm": 0.49966755509376526,
+      "learning_rate": 3.696621621621622e-05,
+      "loss": 0.0506,
+      "step": 1930
+    },
+    {
+      "epoch": 1.3108108108108107,
+      "grad_norm": 0.38622167706489563,
+      "learning_rate": 3.6898648648648646e-05,
+      "loss": 0.053,
+      "step": 1940
+    },
+    {
+      "epoch": 1.3175675675675675,
+      "grad_norm": 0.37733376026153564,
+      "learning_rate": 3.683108108108108e-05,
+      "loss": 0.0548,
+      "step": 1950
+    },
+    {
+      "epoch": 1.3243243243243243,
+      "grad_norm": 0.525426983833313,
+      "learning_rate": 3.676351351351352e-05,
+      "loss": 0.0621,
+      "step": 1960
+    },
+    {
+      "epoch": 1.3310810810810811,
+      "grad_norm": 0.6133748292922974,
+      "learning_rate": 3.669594594594595e-05,
+      "loss": 0.0556,
+      "step": 1970
+    },
+    {
+      "epoch": 1.3378378378378377,
+      "grad_norm": 0.5005325078964233,
+      "learning_rate": 3.662837837837838e-05,
+      "loss": 0.0633,
+      "step": 1980
+    },
+    {
+      "epoch": 1.3445945945945945,
+      "grad_norm": 0.45884352922439575,
+      "learning_rate": 3.6560810810810815e-05,
+      "loss": 0.0584,
+      "step": 1990
+    },
+    {
+      "epoch": 1.3513513513513513,
+      "grad_norm": 0.3934214115142822,
+      "learning_rate": 3.649324324324325e-05,
+      "loss": 0.0522,
+      "step": 2000
+    },
+    {
+      "epoch": 1.3581081081081081,
+      "grad_norm": 0.3990820050239563,
+      "learning_rate": 3.642567567567568e-05,
+      "loss": 0.0596,
+      "step": 2010
+    },
+    {
+      "epoch": 1.364864864864865,
+      "grad_norm": 0.43622463941574097,
+      "learning_rate": 3.6358108108108105e-05,
+      "loss": 0.069,
+      "step": 2020
+    },
+    {
+      "epoch": 1.3716216216216215,
+      "grad_norm": 0.4469757378101349,
+      "learning_rate": 3.6290540540540544e-05,
+      "loss": 0.0591,
+      "step": 2030
+    },
+    {
+      "epoch": 1.3783783783783785,
+      "grad_norm": 0.3687591552734375,
+      "learning_rate": 3.6222972972972976e-05,
+      "loss": 0.0517,
+      "step": 2040
+    },
+    {
+      "epoch": 1.385135135135135,
+      "grad_norm": 0.24021026492118835,
+      "learning_rate": 3.615540540540541e-05,
+      "loss": 0.0429,
+      "step": 2050
+    },
+    {
+      "epoch": 1.3918918918918919,
+      "grad_norm": 0.38636326789855957,
+      "learning_rate": 3.608783783783784e-05,
+      "loss": 0.0481,
+      "step": 2060
+    },
+    {
+      "epoch": 1.3986486486486487,
+      "grad_norm": 0.7737833857536316,
+      "learning_rate": 3.602027027027027e-05,
+      "loss": 0.0482,
+      "step": 2070
+    },
+    {
+      "epoch": 1.4054054054054055,
+      "grad_norm": 0.5627581477165222,
+      "learning_rate": 3.5952702702702705e-05,
+      "loss": 0.0627,
+      "step": 2080
+    },
+    {
+      "epoch": 1.4121621621621623,
+      "grad_norm": 0.33393439650535583,
+      "learning_rate": 3.588513513513514e-05,
+      "loss": 0.0556,
+      "step": 2090
+    },
+    {
+      "epoch": 1.4189189189189189,
+      "grad_norm": 0.47681981325149536,
+      "learning_rate": 3.581756756756757e-05,
+      "loss": 0.0621,
+      "step": 2100
+    },
+    {
+      "epoch": 1.4256756756756757,
+      "grad_norm": 0.4113961160182953,
+      "learning_rate": 3.575e-05,
+      "loss": 0.0559,
+      "step": 2110
+    },
+    {
+      "epoch": 1.4324324324324325,
+      "grad_norm": 0.3725287616252899,
+      "learning_rate": 3.5682432432432434e-05,
+      "loss": 0.0487,
+      "step": 2120
+    },
+    {
+      "epoch": 1.4391891891891893,
+      "grad_norm": 0.36960911750793457,
+      "learning_rate": 3.5614864864864866e-05,
+      "loss": 0.0496,
+      "step": 2130
+    },
+    {
+      "epoch": 1.445945945945946,
+      "grad_norm": 0.3826832175254822,
+      "learning_rate": 3.55472972972973e-05,
+      "loss": 0.0499,
+      "step": 2140
+    },
+    {
+      "epoch": 1.4527027027027026,
+      "grad_norm": 0.5639827847480774,
+      "learning_rate": 3.547972972972973e-05,
+      "loss": 0.0647,
+      "step": 2150
+    },
+    {
+      "epoch": 1.4594594594594594,
+      "grad_norm": 0.7879467606544495,
+      "learning_rate": 3.541216216216216e-05,
+      "loss": 0.0631,
+      "step": 2160
+    },
+    {
+      "epoch": 1.4662162162162162,
+      "grad_norm": 0.27783432602882385,
+      "learning_rate": 3.53445945945946e-05,
+      "loss": 0.0534,
+      "step": 2170
+    },
+    {
+      "epoch": 1.472972972972973,
+      "grad_norm": 0.4450521171092987,
+      "learning_rate": 3.527702702702703e-05,
+      "loss": 0.0578,
+      "step": 2180
+    },
+    {
+      "epoch": 1.4797297297297298,
+      "grad_norm": 0.27280521392822266,
+      "learning_rate": 3.520945945945946e-05,
+      "loss": 0.0508,
+      "step": 2190
+    },
+    {
+      "epoch": 1.4864864864864864,
+      "grad_norm": 0.2709004878997803,
+      "learning_rate": 3.514189189189189e-05,
+      "loss": 0.0466,
+      "step": 2200
+    },
+    {
+      "epoch": 1.4932432432432432,
+      "grad_norm": 0.3932722806930542,
+      "learning_rate": 3.5074324324324324e-05,
+      "loss": 0.0531,
+      "step": 2210
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.4892805516719818,
+      "learning_rate": 3.5006756756756756e-05,
+      "loss": 0.0457,
+      "step": 2220
+    },
+    {
+      "epoch": 1.5067567567567568,
+      "grad_norm": 0.5069631934165955,
+      "learning_rate": 3.493918918918919e-05,
+      "loss": 0.0593,
+      "step": 2230
+    },
+    {
+      "epoch": 1.5135135135135136,
+      "grad_norm": 0.4157819449901581,
+      "learning_rate": 3.487162162162163e-05,
+      "loss": 0.0562,
+      "step": 2240
+    },
+    {
+      "epoch": 1.5202702702702702,
+      "grad_norm": 0.31272444128990173,
+      "learning_rate": 3.480405405405406e-05,
+      "loss": 0.0493,
+      "step": 2250
+    },
+    {
+      "epoch": 1.527027027027027,
+      "grad_norm": 0.5022543668746948,
+      "learning_rate": 3.4736486486486485e-05,
+      "loss": 0.0518,
+      "step": 2260
+    },
+    {
+      "epoch": 1.5337837837837838,
+      "grad_norm": 0.6875999569892883,
+      "learning_rate": 3.466891891891892e-05,
+      "loss": 0.0403,
+      "step": 2270
+    },
+    {
+      "epoch": 1.5405405405405406,
+      "grad_norm": 0.4183546006679535,
+      "learning_rate": 3.460135135135135e-05,
+      "loss": 0.0505,
+      "step": 2280
+    },
+    {
+      "epoch": 1.5472972972972974,
+      "grad_norm": 0.4362141788005829,
+      "learning_rate": 3.453378378378378e-05,
+      "loss": 0.0584,
+      "step": 2290
+    },
+    {
+      "epoch": 1.554054054054054,
+      "grad_norm": 0.5941928029060364,
+      "learning_rate": 3.446621621621622e-05,
+      "loss": 0.069,
+      "step": 2300
+    },
+    {
+      "epoch": 1.560810810810811,
+      "grad_norm": 0.9973124265670776,
+      "learning_rate": 3.4398648648648653e-05,
+      "loss": 0.0676,
+      "step": 2310
+    },
+    {
+      "epoch": 1.5675675675675675,
+      "grad_norm": 0.40244096517562866,
+      "learning_rate": 3.4331081081081086e-05,
+      "loss": 0.061,
+      "step": 2320
+    },
+    {
+      "epoch": 1.5743243243243243,
+      "grad_norm": 0.6185030937194824,
+      "learning_rate": 3.426351351351352e-05,
+      "loss": 0.0512,
+      "step": 2330
+    },
+    {
+      "epoch": 1.5810810810810811,
+      "grad_norm": 0.41134995222091675,
+      "learning_rate": 3.4195945945945943e-05,
+      "loss": 0.0557,
+      "step": 2340
+    },
+    {
+      "epoch": 1.5878378378378377,
+      "grad_norm": 0.34551721811294556,
+      "learning_rate": 3.4128378378378376e-05,
+      "loss": 0.0396,
+      "step": 2350
+    },
+    {
+      "epoch": 1.5945945945945947,
+      "grad_norm": 0.6904975175857544,
+      "learning_rate": 3.4060810810810815e-05,
+      "loss": 0.0503,
+      "step": 2360
+    },
+    {
+      "epoch": 1.6013513513513513,
+      "grad_norm": 0.35074079036712646,
+      "learning_rate": 3.399324324324325e-05,
+      "loss": 0.047,
+      "step": 2370
+    },
+    {
+      "epoch": 1.6081081081081081,
+      "grad_norm": 0.4111129641532898,
+      "learning_rate": 3.392567567567568e-05,
+      "loss": 0.0485,
+      "step": 2380
+    },
+    {
+      "epoch": 1.614864864864865,
+      "grad_norm": 0.597917914390564,
+      "learning_rate": 3.385810810810811e-05,
+      "loss": 0.0473,
+      "step": 2390
+    },
+    {
+      "epoch": 1.6216216216216215,
+      "grad_norm": 0.3226557970046997,
+      "learning_rate": 3.3790540540540544e-05,
+      "loss": 0.0438,
+      "step": 2400
+    },
+    {
+      "epoch": 1.6283783783783785,
+      "grad_norm": 0.31680727005004883,
+      "learning_rate": 3.3722972972972976e-05,
+      "loss": 0.0442,
+      "step": 2410
+    },
+    {
+      "epoch": 1.635135135135135,
+      "grad_norm": 0.5353063941001892,
+      "learning_rate": 3.36554054054054e-05,
+      "loss": 0.0566,
+      "step": 2420
+    },
+    {
+      "epoch": 1.6418918918918919,
+      "grad_norm": 0.29757750034332275,
+      "learning_rate": 3.358783783783784e-05,
+      "loss": 0.0372,
+      "step": 2430
+    },
+    {
+      "epoch": 1.6486486486486487,
+      "grad_norm": 0.32976236939430237,
+      "learning_rate": 3.352027027027027e-05,
+      "loss": 0.0499,
+      "step": 2440
+    },
+    {
+      "epoch": 1.6554054054054053,
+      "grad_norm": 0.45679962635040283,
+      "learning_rate": 3.3452702702702705e-05,
+      "loss": 0.0494,
+      "step": 2450
+    },
+    {
+      "epoch": 1.6621621621621623,
+      "grad_norm": 0.37637847661972046,
+      "learning_rate": 3.338513513513514e-05,
+      "loss": 0.0511,
+      "step": 2460
+    },
+    {
+      "epoch": 1.6689189189189189,
+      "grad_norm": 0.24155497550964355,
+      "learning_rate": 3.331756756756757e-05,
+      "loss": 0.0488,
+      "step": 2470
+    },
+    {
+      "epoch": 1.6756756756756757,
+      "grad_norm": 0.24483831226825714,
+      "learning_rate": 3.325e-05,
+      "loss": 0.0621,
+      "step": 2480
+    },
+    {
+      "epoch": 1.6824324324324325,
+      "grad_norm": 0.3654753863811493,
+      "learning_rate": 3.3182432432432434e-05,
+      "loss": 0.0478,
+      "step": 2490
+    },
+    {
+      "epoch": 1.689189189189189,
+      "grad_norm": 0.8125627040863037,
+      "learning_rate": 3.3114864864864866e-05,
+      "loss": 0.061,
+      "step": 2500
+    },
+    {
+      "epoch": 1.695945945945946,
+      "grad_norm": 0.2318248152732849,
+      "learning_rate": 3.30472972972973e-05,
+      "loss": 0.0485,
+      "step": 2510
+    },
+    {
+      "epoch": 1.7027027027027026,
+      "grad_norm": 0.5518670082092285,
+      "learning_rate": 3.297972972972973e-05,
+      "loss": 0.0535,
+      "step": 2520
+    },
+    {
+      "epoch": 1.7094594594594594,
+      "grad_norm": 0.3407860994338989,
+      "learning_rate": 3.291216216216216e-05,
+      "loss": 0.0423,
+      "step": 2530
+    },
+    {
+      "epoch": 1.7162162162162162,
+      "grad_norm": 0.17148450016975403,
+      "learning_rate": 3.2844594594594595e-05,
+      "loss": 0.0419,
+      "step": 2540
+    },
+    {
+      "epoch": 1.722972972972973,
+      "grad_norm": 0.34441882371902466,
+      "learning_rate": 3.277702702702703e-05,
+      "loss": 0.0443,
+      "step": 2550
+    },
+    {
+      "epoch": 1.7297297297297298,
+      "grad_norm": 0.516750693321228,
+      "learning_rate": 3.270945945945946e-05,
+      "loss": 0.0643,
+      "step": 2560
+    },
+    {
+      "epoch": 1.7364864864864864,
+      "grad_norm": 0.5338383316993713,
+      "learning_rate": 3.26418918918919e-05,
+      "loss": 0.0937,
+      "step": 2570
+    },
+    {
+      "epoch": 1.7432432432432432,
+      "grad_norm": 0.44282281398773193,
+      "learning_rate": 3.2574324324324324e-05,
+      "loss": 0.0434,
+      "step": 2580
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.20837600529193878,
+      "learning_rate": 3.250675675675676e-05,
+      "loss": 0.0518,
+      "step": 2590
+    },
+    {
+      "epoch": 1.7567567567567568,
+      "grad_norm": 0.28318697214126587,
+      "learning_rate": 3.243918918918919e-05,
+      "loss": 0.0521,
+      "step": 2600
+    },
+    {
+      "epoch": 1.7635135135135136,
+      "grad_norm": 0.3730570077896118,
+      "learning_rate": 3.237162162162162e-05,
+      "loss": 0.0557,
+      "step": 2610
+    },
+    {
+      "epoch": 1.7702702702702702,
+      "grad_norm": 0.3967922329902649,
+      "learning_rate": 3.2304054054054053e-05,
+      "loss": 0.0645,
+      "step": 2620
+    },
+    {
+      "epoch": 1.777027027027027,
+      "grad_norm": 0.43898531794548035,
+      "learning_rate": 3.223648648648649e-05,
+      "loss": 0.0524,
+      "step": 2630
+    },
+    {
+      "epoch": 1.7837837837837838,
+      "grad_norm": 0.13809113204479218,
+      "learning_rate": 3.2168918918918925e-05,
+      "loss": 0.0416,
+      "step": 2640
+    },
+    {
+      "epoch": 1.7905405405405406,
+      "grad_norm": 0.24318750202655792,
+      "learning_rate": 3.210135135135136e-05,
+      "loss": 0.0543,
+      "step": 2650
+    },
+    {
+      "epoch": 1.7972972972972974,
+      "grad_norm": 0.3348706066608429,
+      "learning_rate": 3.203378378378378e-05,
+      "loss": 0.0364,
+      "step": 2660
+    },
+    {
+      "epoch": 1.804054054054054,
+      "grad_norm": 0.29943642020225525,
+      "learning_rate": 3.1966216216216215e-05,
+      "loss": 0.0586,
+      "step": 2670
+    },
+    {
+      "epoch": 1.810810810810811,
+      "grad_norm": 0.40303003787994385,
+      "learning_rate": 3.189864864864865e-05,
+      "loss": 0.0574,
+      "step": 2680
+    },
+    {
+      "epoch": 1.8175675675675675,
+      "grad_norm": 29.34443473815918,
+      "learning_rate": 3.183108108108108e-05,
+      "loss": 0.0561,
+      "step": 2690
+    },
+    {
+      "epoch": 1.8243243243243243,
+      "grad_norm": 0.5370333790779114,
+      "learning_rate": 3.176351351351352e-05,
+      "loss": 0.0378,
+      "step": 2700
+    },
+    {
+      "epoch": 1.8310810810810811,
+      "grad_norm": 0.30435851216316223,
+      "learning_rate": 3.169594594594595e-05,
+      "loss": 0.0485,
+      "step": 2710
+    },
+    {
+      "epoch": 1.8378378378378377,
+      "grad_norm": 0.3103540539741516,
+      "learning_rate": 3.162837837837838e-05,
+      "loss": 0.0467,
+      "step": 2720
+    },
+    {
+      "epoch": 1.8445945945945947,
+      "grad_norm": 0.2695455551147461,
+      "learning_rate": 3.1560810810810815e-05,
+      "loss": 0.0437,
+      "step": 2730
+    },
+    {
+      "epoch": 1.8513513513513513,
+      "grad_norm": 0.19030825793743134,
+      "learning_rate": 3.149324324324324e-05,
+      "loss": 0.0349,
+      "step": 2740
+    },
+    {
+      "epoch": 1.8581081081081081,
+      "grad_norm": 0.49268749356269836,
+      "learning_rate": 3.142567567567567e-05,
+      "loss": 0.0425,
+      "step": 2750
+    },
+    {
+      "epoch": 1.864864864864865,
+      "grad_norm": 0.22729387879371643,
+      "learning_rate": 3.135810810810811e-05,
+      "loss": 0.0374,
+      "step": 2760
+    },
+    {
+      "epoch": 1.8716216216216215,
+      "grad_norm": 0.4349585473537445,
+      "learning_rate": 3.1290540540540544e-05,
+      "loss": 0.0554,
+      "step": 2770
+    },
+    {
+      "epoch": 1.8783783783783785,
+      "grad_norm": 0.6437013149261475,
+      "learning_rate": 3.1222972972972976e-05,
+      "loss": 0.0479,
+      "step": 2780
+    },
+    {
+      "epoch": 1.885135135135135,
+      "grad_norm": 0.377573698759079,
+      "learning_rate": 3.115540540540541e-05,
+      "loss": 0.0434,
+      "step": 2790
+    },
+    {
+      "epoch": 1.8918918918918919,
+      "grad_norm": 0.4241003692150116,
+      "learning_rate": 3.108783783783784e-05,
+      "loss": 0.0523,
+      "step": 2800
+    },
+    {
+      "epoch": 1.8986486486486487,
+      "grad_norm": 0.5633030533790588,
+      "learning_rate": 3.102027027027027e-05,
+      "loss": 0.0496,
+      "step": 2810
+    },
+    {
+      "epoch": 1.9054054054054053,
+      "grad_norm": 0.4770616292953491,
+      "learning_rate": 3.09527027027027e-05,
+      "loss": 0.0411,
+      "step": 2820
+    },
+    {
+      "epoch": 1.9121621621621623,
+      "grad_norm": 2.8804574012756348,
+      "learning_rate": 3.088513513513514e-05,
+      "loss": 0.0462,
+      "step": 2830
+    },
+    {
+      "epoch": 1.9189189189189189,
+      "grad_norm": 0.4047267436981201,
+      "learning_rate": 3.081756756756757e-05,
+      "loss": 0.0407,
+      "step": 2840
+    },
+    {
+      "epoch": 1.9256756756756757,
+      "grad_norm": 0.20034578442573547,
+      "learning_rate": 3.075e-05,
+      "loss": 0.0553,
+      "step": 2850
+    },
+    {
+      "epoch": 1.9324324324324325,
+      "grad_norm": 0.5225114226341248,
+      "learning_rate": 3.0682432432432434e-05,
+      "loss": 0.0451,
+      "step": 2860
+    },
+    {
+      "epoch": 1.939189189189189,
+      "grad_norm": 0.5536220669746399,
+      "learning_rate": 3.0614864864864867e-05,
+      "loss": 0.0499,
+      "step": 2870
+    },
+    {
+      "epoch": 1.945945945945946,
+      "grad_norm": 0.29939669370651245,
+      "learning_rate": 3.05472972972973e-05,
+      "loss": 0.039,
+      "step": 2880
+    },
+    {
+      "epoch": 1.9527027027027026,
+      "grad_norm": 0.40010547637939453,
+      "learning_rate": 3.0479729729729734e-05,
+      "loss": 0.049,
+      "step": 2890
+    },
+    {
+      "epoch": 1.9594594594594594,
+      "grad_norm": 0.24146144092082977,
+      "learning_rate": 3.041216216216216e-05,
+      "loss": 0.0555,
+      "step": 2900
+    },
+    {
+      "epoch": 1.9662162162162162,
+      "grad_norm": 0.3182704746723175,
+      "learning_rate": 3.0344594594594596e-05,
+      "loss": 0.0425,
+      "step": 2910
+    },
+    {
+      "epoch": 1.972972972972973,
+      "grad_norm": 0.5397040247917175,
+      "learning_rate": 3.0277027027027028e-05,
+      "loss": 0.0479,
+      "step": 2920
+    },
+    {
+      "epoch": 1.9797297297297298,
+      "grad_norm": 0.31318461894989014,
+      "learning_rate": 3.020945945945946e-05,
+      "loss": 0.0528,
+      "step": 2930
+    },
+    {
+      "epoch": 1.9864864864864864,
+      "grad_norm": 0.2979433238506317,
+      "learning_rate": 3.0141891891891892e-05,
+      "loss": 0.0396,
+      "step": 2940
+    },
+    {
+      "epoch": 1.9932432432432432,
+      "grad_norm": 0.329349160194397,
+      "learning_rate": 3.0074324324324328e-05,
+      "loss": 0.0644,
+      "step": 2950
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 1.2042832374572754,
+      "learning_rate": 3.000675675675676e-05,
+      "loss": 0.0446,
+      "step": 2960
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 0.02900381200015545,
+      "eval_runtime": 232.4586,
+      "eval_samples_per_second": 4.495,
+      "eval_steps_per_second": 1.127,
+      "step": 2960
+    },
+    {
+      "epoch": 2.0067567567567566,
+      "grad_norm": 0.41488581895828247,
+      "learning_rate": 2.9939189189189193e-05,
+      "loss": 0.0478,
+      "step": 2970
+    },
+    {
+      "epoch": 2.0135135135135136,
+      "grad_norm": 0.46764683723449707,
+      "learning_rate": 2.987162162162162e-05,
+      "loss": 0.0565,
+      "step": 2980
+    },
+    {
+      "epoch": 2.02027027027027,
+      "grad_norm": 0.2321554720401764,
+      "learning_rate": 2.9804054054054054e-05,
+      "loss": 0.0286,
+      "step": 2990
+    },
+    {
+      "epoch": 2.027027027027027,
+      "grad_norm": 0.10871370881795883,
+      "learning_rate": 2.9736486486486486e-05,
+      "loss": 0.0461,
+      "step": 3000
+    },
+    {
+      "epoch": 2.0337837837837838,
+      "grad_norm": 0.42073723673820496,
+      "learning_rate": 2.966891891891892e-05,
+      "loss": 0.0396,
+      "step": 3010
+    },
+    {
+      "epoch": 2.0405405405405403,
+      "grad_norm": 0.684948205947876,
+      "learning_rate": 2.9601351351351354e-05,
+      "loss": 0.0789,
+      "step": 3020
+    },
+    {
+      "epoch": 2.0472972972972974,
+      "grad_norm": 2.56610107421875,
+      "learning_rate": 2.9533783783783786e-05,
+      "loss": 0.0561,
+      "step": 3030
+    },
+    {
+      "epoch": 2.054054054054054,
+      "grad_norm": 0.28919854760169983,
+      "learning_rate": 2.9466216216216218e-05,
+      "loss": 0.0486,
+      "step": 3040
+    },
+    {
+      "epoch": 2.060810810810811,
+      "grad_norm": 0.28242143988609314,
+      "learning_rate": 2.9398648648648654e-05,
+      "loss": 0.0511,
+      "step": 3050
+    },
+    {
+      "epoch": 2.0675675675675675,
+      "grad_norm": 0.5599525570869446,
+      "learning_rate": 2.933108108108108e-05,
+      "loss": 0.0501,
+      "step": 3060
+    },
+    {
+      "epoch": 2.074324324324324,
+      "grad_norm": 0.314981609582901,
+      "learning_rate": 2.926351351351351e-05,
+      "loss": 0.036,
+      "step": 3070
+    },
+    {
+      "epoch": 2.081081081081081,
+      "grad_norm": 0.38009947538375854,
+      "learning_rate": 2.9195945945945947e-05,
+      "loss": 0.0449,
+      "step": 3080
+    },
+    {
+      "epoch": 2.0878378378378377,
+      "grad_norm": 0.46030741930007935,
+      "learning_rate": 2.912837837837838e-05,
+      "loss": 0.0485,
+      "step": 3090
+    },
+    {
+      "epoch": 2.0945945945945947,
+      "grad_norm": 0.2718926966190338,
+      "learning_rate": 2.9060810810810812e-05,
+      "loss": 0.0429,
+      "step": 3100
+    },
+    {
+      "epoch": 2.1013513513513513,
+      "grad_norm": 0.41588521003723145,
+      "learning_rate": 2.8993243243243244e-05,
+      "loss": 0.0492,
+      "step": 3110
+    },
+    {
+      "epoch": 2.108108108108108,
+      "grad_norm": 0.503211259841919,
+      "learning_rate": 2.892567567567568e-05,
+      "loss": 0.0399,
+      "step": 3120
+    },
+    {
+      "epoch": 2.114864864864865,
+      "grad_norm": 0.39988529682159424,
+      "learning_rate": 2.8858108108108112e-05,
+      "loss": 0.0528,
+      "step": 3130
+    },
+    {
+      "epoch": 2.1216216216216215,
+      "grad_norm": 0.2753487229347229,
+      "learning_rate": 2.879054054054054e-05,
+      "loss": 0.0554,
+      "step": 3140
+    },
+    {
+      "epoch": 2.1283783783783785,
+      "grad_norm": 0.5936213135719299,
+      "learning_rate": 2.8722972972972973e-05,
+      "loss": 0.0486,
+      "step": 3150
+    },
+    {
+      "epoch": 2.135135135135135,
+      "grad_norm": 4.782855033874512,
+      "learning_rate": 2.8655405405405405e-05,
+      "loss": 0.0544,
+      "step": 3160
+    },
+    {
+      "epoch": 2.141891891891892,
+      "grad_norm": 0.9043190479278564,
+      "learning_rate": 2.8587837837837838e-05,
+      "loss": 0.0422,
+      "step": 3170
+    },
+    {
+      "epoch": 2.1486486486486487,
+      "grad_norm": 0.28013184666633606,
+      "learning_rate": 2.8520270270270273e-05,
+      "loss": 0.0385,
+      "step": 3180
+    },
+    {
+      "epoch": 2.1554054054054053,
+      "grad_norm": 0.5391634106636047,
+      "learning_rate": 2.8452702702702706e-05,
+      "loss": 0.0432,
+      "step": 3190
+    },
+    {
+      "epoch": 2.1621621621621623,
+      "grad_norm": 0.28761211037635803,
+      "learning_rate": 2.8385135135135138e-05,
+      "loss": 0.045,
+      "step": 3200
+    },
+    {
+      "epoch": 2.168918918918919,
+      "grad_norm": 0.43844741582870483,
+      "learning_rate": 2.831756756756757e-05,
+      "loss": 0.051,
+      "step": 3210
+    },
+    {
+      "epoch": 2.175675675675676,
+      "grad_norm": 0.9262340664863586,
+      "learning_rate": 2.825e-05,
+      "loss": 0.0618,
+      "step": 3220
+    },
+    {
+      "epoch": 2.1824324324324325,
+      "grad_norm": 0.2561081349849701,
+      "learning_rate": 2.818243243243243e-05,
+      "loss": 0.0514,
+      "step": 3230
+    },
+    {
+      "epoch": 2.189189189189189,
+      "grad_norm": 0.25118082761764526,
+      "learning_rate": 2.8114864864864863e-05,
+      "loss": 0.0487,
+      "step": 3240
+    },
+    {
+      "epoch": 2.195945945945946,
+      "grad_norm": 0.2230692356824875,
+      "learning_rate": 2.80472972972973e-05,
+      "loss": 0.0324,
+      "step": 3250
+    },
+    {
+      "epoch": 2.2027027027027026,
+      "grad_norm": 0.27720406651496887,
+      "learning_rate": 2.797972972972973e-05,
+      "loss": 0.0473,
+      "step": 3260
+    },
+    {
+      "epoch": 2.2094594594594597,
+      "grad_norm": 0.4395008087158203,
+      "learning_rate": 2.7912162162162164e-05,
+      "loss": 0.0451,
+      "step": 3270
+    },
+    {
+      "epoch": 2.2162162162162162,
+      "grad_norm": 0.34843841195106506,
+      "learning_rate": 2.7844594594594596e-05,
+      "loss": 0.0433,
+      "step": 3280
+    },
+    {
+      "epoch": 2.222972972972973,
+      "grad_norm": 0.3215220868587494,
+      "learning_rate": 2.777702702702703e-05,
+      "loss": 0.0423,
+      "step": 3290
+    },
+    {
+      "epoch": 2.22972972972973,
+      "grad_norm": 0.44971078634262085,
+      "learning_rate": 2.7709459459459457e-05,
+      "loss": 0.0474,
+      "step": 3300
+    },
+    {
+      "epoch": 2.2364864864864864,
+      "grad_norm": 0.4757413864135742,
+      "learning_rate": 2.7641891891891893e-05,
+      "loss": 0.0416,
+      "step": 3310
+    },
+    {
+      "epoch": 2.2432432432432434,
+      "grad_norm": 0.5553408861160278,
+      "learning_rate": 2.7574324324324325e-05,
+      "loss": 0.0491,
+      "step": 3320
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.3604501187801361,
+      "learning_rate": 2.7506756756756757e-05,
+      "loss": 0.041,
+      "step": 3330
+    },
+    {
+      "epoch": 2.2567567567567566,
+      "grad_norm": 0.23111610114574432,
+      "learning_rate": 2.743918918918919e-05,
+      "loss": 0.0437,
+      "step": 3340
+    },
+    {
+      "epoch": 2.2635135135135136,
+      "grad_norm": 0.288889080286026,
+      "learning_rate": 2.7371621621621625e-05,
+      "loss": 0.0375,
+      "step": 3350
+    },
+    {
+      "epoch": 2.27027027027027,
+      "grad_norm": 0.5279484987258911,
+      "learning_rate": 2.7304054054054057e-05,
+      "loss": 0.04,
+      "step": 3360
+    },
+    {
+      "epoch": 2.277027027027027,
+      "grad_norm": 0.259581983089447,
+      "learning_rate": 2.723648648648649e-05,
+      "loss": 0.0337,
+      "step": 3370
+    },
+    {
+      "epoch": 2.2837837837837838,
+      "grad_norm": 0.37831178307533264,
+      "learning_rate": 2.716891891891892e-05,
+      "loss": 0.0427,
+      "step": 3380
+    },
+    {
+      "epoch": 2.2905405405405403,
+      "grad_norm": 0.3985559940338135,
+      "learning_rate": 2.710135135135135e-05,
+      "loss": 0.0394,
+      "step": 3390
+    },
+    {
+      "epoch": 2.2972972972972974,
+      "grad_norm": 0.32648199796676636,
+      "learning_rate": 2.7033783783783783e-05,
+      "loss": 0.035,
+      "step": 3400
+    },
+    {
+      "epoch": 2.304054054054054,
+      "grad_norm": 0.20723649859428406,
+      "learning_rate": 2.696621621621622e-05,
+      "loss": 0.046,
+      "step": 3410
+    },
+    {
+      "epoch": 2.310810810810811,
+      "grad_norm": 0.5824639201164246,
+      "learning_rate": 2.689864864864865e-05,
+      "loss": 0.0445,
+      "step": 3420
+    },
+    {
+      "epoch": 2.3175675675675675,
+      "grad_norm": 0.3186579644680023,
+      "learning_rate": 2.6831081081081083e-05,
+      "loss": 0.0475,
+      "step": 3430
+    },
+    {
+      "epoch": 2.3243243243243246,
+      "grad_norm": 0.40933048725128174,
+      "learning_rate": 2.6763513513513515e-05,
+      "loss": 0.0587,
+      "step": 3440
+    },
+    {
+      "epoch": 2.331081081081081,
+      "grad_norm": 0.48833781480789185,
+      "learning_rate": 2.669594594594595e-05,
+      "loss": 0.0481,
+      "step": 3450
+    },
+    {
+      "epoch": 2.3378378378378377,
+      "grad_norm": 0.2573770582675934,
+      "learning_rate": 2.6628378378378376e-05,
+      "loss": 0.0453,
+      "step": 3460
+    },
+    {
+      "epoch": 2.3445945945945947,
+      "grad_norm": 0.2869914174079895,
+      "learning_rate": 2.656081081081081e-05,
+      "loss": 0.0407,
+      "step": 3470
+    },
+    {
+      "epoch": 2.3513513513513513,
+      "grad_norm": 0.35150277614593506,
+      "learning_rate": 2.6493243243243244e-05,
+      "loss": 0.0326,
+      "step": 3480
+    },
+    {
+      "epoch": 2.358108108108108,
+      "grad_norm": 0.3215469419956207,
+      "learning_rate": 2.6425675675675677e-05,
+      "loss": 0.0612,
+      "step": 3490
+    },
+    {
+      "epoch": 2.364864864864865,
+      "grad_norm": 0.17056582868099213,
+      "learning_rate": 2.635810810810811e-05,
+      "loss": 0.0521,
+      "step": 3500
+    },
+    {
+      "epoch": 2.3716216216216215,
+      "grad_norm": 0.24536234140396118,
+      "learning_rate": 2.629054054054054e-05,
+      "loss": 0.0522,
+      "step": 3510
+    },
+    {
+      "epoch": 2.3783783783783785,
+      "grad_norm": 0.49538880586624146,
+      "learning_rate": 2.6222972972972977e-05,
+      "loss": 0.0328,
+      "step": 3520
+    },
+    {
+      "epoch": 2.385135135135135,
+      "grad_norm": 0.3088550567626953,
+      "learning_rate": 2.615540540540541e-05,
+      "loss": 0.0673,
+      "step": 3530
+    },
+    {
+      "epoch": 2.391891891891892,
+      "grad_norm": 0.29182663559913635,
+      "learning_rate": 2.608783783783784e-05,
+      "loss": 0.0359,
+      "step": 3540
+    },
+    {
+      "epoch": 2.3986486486486487,
+      "grad_norm": 0.3735537528991699,
+      "learning_rate": 2.602027027027027e-05,
+      "loss": 0.0396,
+      "step": 3550
+    },
+    {
+      "epoch": 2.4054054054054053,
+      "grad_norm": 0.9158993363380432,
+      "learning_rate": 2.5952702702702702e-05,
+      "loss": 0.0419,
+      "step": 3560
+    },
+    {
+      "epoch": 2.4121621621621623,
+      "grad_norm": 0.47085681557655334,
+      "learning_rate": 2.5885135135135135e-05,
+      "loss": 0.0474,
+      "step": 3570
+    },
+    {
+      "epoch": 2.418918918918919,
+      "grad_norm": 0.5768277049064636,
+      "learning_rate": 2.581756756756757e-05,
+      "loss": 0.0516,
+      "step": 3580
+    },
+    {
+      "epoch": 2.4256756756756754,
+      "grad_norm": 0.21016757190227509,
+      "learning_rate": 2.5750000000000002e-05,
+      "loss": 0.0513,
+      "step": 3590
+    },
+    {
+      "epoch": 2.4324324324324325,
+      "grad_norm": 0.5895877480506897,
+      "learning_rate": 2.5682432432432435e-05,
+      "loss": 0.0424,
+      "step": 3600
+    },
+    {
+      "epoch": 2.439189189189189,
+      "grad_norm": 0.37401655316352844,
+      "learning_rate": 2.5614864864864867e-05,
+      "loss": 0.0467,
+      "step": 3610
+    },
+    {
+      "epoch": 2.445945945945946,
+      "grad_norm": 0.2630750238895416,
+      "learning_rate": 2.5547297297297303e-05,
+      "loss": 0.0525,
+      "step": 3620
+    },
+    {
+      "epoch": 2.4527027027027026,
+      "grad_norm": 0.25770536065101624,
+      "learning_rate": 2.5479729729729728e-05,
+      "loss": 0.041,
+      "step": 3630
+    },
+    {
+      "epoch": 2.4594594594594597,
+      "grad_norm": 0.5026077032089233,
+      "learning_rate": 2.541216216216216e-05,
+      "loss": 0.0497,
+      "step": 3640
+    },
+    {
+      "epoch": 2.4662162162162162,
+      "grad_norm": 0.3355426788330078,
+      "learning_rate": 2.5344594594594596e-05,
+      "loss": 0.0338,
+      "step": 3650
+    },
+    {
+      "epoch": 2.472972972972973,
+      "grad_norm": 0.1499488651752472,
+      "learning_rate": 2.5277027027027028e-05,
+      "loss": 0.0517,
+      "step": 3660
+    },
+    {
+      "epoch": 2.47972972972973,
+      "grad_norm": 0.5145780444145203,
+      "learning_rate": 2.520945945945946e-05,
+      "loss": 0.0565,
+      "step": 3670
+    },
+    {
+      "epoch": 2.4864864864864864,
+      "grad_norm": 0.503230631351471,
+      "learning_rate": 2.5141891891891893e-05,
+      "loss": 0.053,
+      "step": 3680
+    },
+    {
+      "epoch": 2.4932432432432434,
+      "grad_norm": 0.3355911374092102,
+      "learning_rate": 2.507432432432433e-05,
+      "loss": 0.0375,
+      "step": 3690
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.17536339163780212,
+      "learning_rate": 2.500675675675676e-05,
+      "loss": 0.0367,
+      "step": 3700
+    },
+    {
+      "epoch": 2.506756756756757,
+      "grad_norm": 0.4582735598087311,
+      "learning_rate": 2.493918918918919e-05,
+      "loss": 0.0433,
+      "step": 3710
+    },
+    {
+      "epoch": 2.5135135135135136,
+      "grad_norm": 0.2658029794692993,
+      "learning_rate": 2.4871621621621625e-05,
+      "loss": 0.0446,
+      "step": 3720
+    },
+    {
+      "epoch": 2.52027027027027,
+      "grad_norm": 0.22042320668697357,
+      "learning_rate": 2.4804054054054054e-05,
+      "loss": 0.0409,
+      "step": 3730
+    },
+    {
+      "epoch": 2.527027027027027,
+      "grad_norm": 0.42648428678512573,
+      "learning_rate": 2.4736486486486486e-05,
+      "loss": 0.0503,
+      "step": 3740
+    },
+    {
+      "epoch": 2.5337837837837838,
+      "grad_norm": 0.49982187151908875,
+      "learning_rate": 2.4668918918918922e-05,
+      "loss": 0.0401,
+      "step": 3750
+    },
+    {
+      "epoch": 2.5405405405405403,
+      "grad_norm": 0.3434715270996094,
+      "learning_rate": 2.4601351351351354e-05,
+      "loss": 0.0383,
+      "step": 3760
+    },
+    {
+      "epoch": 2.5472972972972974,
+      "grad_norm": 0.2308432161808014,
+      "learning_rate": 2.4533783783783783e-05,
+      "loss": 0.0399,
+      "step": 3770
+    },
+    {
+      "epoch": 2.554054054054054,
+      "grad_norm": 0.5601356029510498,
+      "learning_rate": 2.446621621621622e-05,
+      "loss": 0.0377,
+      "step": 3780
+    },
+    {
+      "epoch": 2.560810810810811,
+      "grad_norm": 0.31023895740509033,
+      "learning_rate": 2.439864864864865e-05,
+      "loss": 0.054,
+      "step": 3790
+    },
+    {
+      "epoch": 2.5675675675675675,
+      "grad_norm": 0.39844200015068054,
+      "learning_rate": 2.4331081081081083e-05,
+      "loss": 0.0409,
+      "step": 3800
+    },
+    {
+      "epoch": 2.5743243243243246,
+      "grad_norm": 0.44077757000923157,
+      "learning_rate": 2.4263513513513515e-05,
+      "loss": 0.0404,
+      "step": 3810
+    },
+    {
+      "epoch": 2.581081081081081,
+      "grad_norm": 0.3423317074775696,
+      "learning_rate": 2.4195945945945948e-05,
+      "loss": 0.0403,
+      "step": 3820
+    },
+    {
+      "epoch": 2.5878378378378377,
+      "grad_norm": 0.25777336955070496,
+      "learning_rate": 2.412837837837838e-05,
+      "loss": 0.0343,
+      "step": 3830
+    },
+    {
+      "epoch": 2.5945945945945947,
+      "grad_norm": 0.4902068078517914,
+      "learning_rate": 2.4060810810810812e-05,
+      "loss": 0.044,
+      "step": 3840
+    },
+    {
+      "epoch": 2.6013513513513513,
+      "grad_norm": 0.27901652455329895,
+      "learning_rate": 2.3993243243243245e-05,
+      "loss": 0.0516,
+      "step": 3850
+    },
+    {
+      "epoch": 2.608108108108108,
+      "grad_norm": 0.2561982274055481,
+      "learning_rate": 2.3925675675675677e-05,
+      "loss": 0.032,
+      "step": 3860
+    },
+    {
+      "epoch": 2.614864864864865,
+      "grad_norm": 0.6690441370010376,
+      "learning_rate": 2.385810810810811e-05,
+      "loss": 0.0438,
+      "step": 3870
+    },
+    {
+      "epoch": 2.6216216216216215,
+      "grad_norm": 0.8403464555740356,
+      "learning_rate": 2.379054054054054e-05,
+      "loss": 0.0524,
+      "step": 3880
+    },
+    {
+      "epoch": 2.6283783783783785,
+      "grad_norm": 0.24565739929676056,
+      "learning_rate": 2.3722972972972974e-05,
+      "loss": 0.0305,
+      "step": 3890
+    },
+    {
+      "epoch": 2.635135135135135,
+      "grad_norm": 0.39929670095443726,
+      "learning_rate": 2.3655405405405406e-05,
+      "loss": 0.0382,
+      "step": 3900
+    },
+    {
+      "epoch": 2.641891891891892,
+      "grad_norm": 0.48382842540740967,
+      "learning_rate": 2.3587837837837838e-05,
+      "loss": 0.0498,
+      "step": 3910
+    },
+    {
+      "epoch": 2.6486486486486487,
+      "grad_norm": 0.27263057231903076,
+      "learning_rate": 2.3520270270270274e-05,
+      "loss": 0.0337,
+      "step": 3920
+    },
+    {
+      "epoch": 2.6554054054054053,
+      "grad_norm": 0.29816973209381104,
+      "learning_rate": 2.3452702702702703e-05,
+      "loss": 0.0322,
+      "step": 3930
+    },
+    {
+      "epoch": 2.6621621621621623,
+      "grad_norm": 0.35650044679641724,
+      "learning_rate": 2.3385135135135135e-05,
+      "loss": 0.0408,
+      "step": 3940
+    },
+    {
+      "epoch": 2.668918918918919,
+      "grad_norm": 0.22148212790489197,
+      "learning_rate": 2.331756756756757e-05,
+      "loss": 0.0506,
+      "step": 3950
+    },
+    {
+      "epoch": 2.6756756756756754,
+      "grad_norm": 0.330817312002182,
+      "learning_rate": 2.3250000000000003e-05,
+      "loss": 0.0458,
+      "step": 3960
+    },
+    {
+      "epoch": 2.6824324324324325,
+      "grad_norm": 0.3396255373954773,
+      "learning_rate": 2.318243243243243e-05,
+      "loss": 0.0349,
+      "step": 3970
+    },
+    {
+      "epoch": 2.689189189189189,
+      "grad_norm": 0.27983909845352173,
+      "learning_rate": 2.3114864864864867e-05,
+      "loss": 0.0555,
+      "step": 3980
+    },
+    {
+      "epoch": 2.695945945945946,
+      "grad_norm": 0.21532095968723297,
+      "learning_rate": 2.30472972972973e-05,
+      "loss": 0.0414,
+      "step": 3990
+    },
+    {
+      "epoch": 2.7027027027027026,
+      "grad_norm": 0.4602966904640198,
+      "learning_rate": 2.2979729729729732e-05,
+      "loss": 0.0416,
+      "step": 4000
+    },
+    {
+      "epoch": 2.7094594594594597,
+      "grad_norm": 0.37232860922813416,
+      "learning_rate": 2.2912162162162164e-05,
+      "loss": 0.0376,
+      "step": 4010
+    },
+    {
+      "epoch": 2.7162162162162162,
+      "grad_norm": 0.27139970660209656,
+      "learning_rate": 2.2844594594594596e-05,
+      "loss": 0.0457,
+      "step": 4020
+    },
+    {
+      "epoch": 2.722972972972973,
+      "grad_norm": 0.3222922682762146,
+      "learning_rate": 2.277702702702703e-05,
+      "loss": 0.0433,
+      "step": 4030
+    },
+    {
+      "epoch": 2.72972972972973,
+      "grad_norm": 0.5624486207962036,
+      "learning_rate": 2.270945945945946e-05,
+      "loss": 0.0417,
+      "step": 4040
+    },
+    {
+      "epoch": 2.7364864864864864,
+      "grad_norm": 0.7472530603408813,
+      "learning_rate": 2.2641891891891893e-05,
+      "loss": 0.0394,
+      "step": 4050
+    },
+    {
+      "epoch": 2.743243243243243,
+      "grad_norm": 0.38589775562286377,
+      "learning_rate": 2.2574324324324325e-05,
+      "loss": 0.0316,
+      "step": 4060
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.34411191940307617,
+      "learning_rate": 2.2506756756756758e-05,
+      "loss": 0.0361,
+      "step": 4070
+    },
+    {
+      "epoch": 2.756756756756757,
+      "grad_norm": 0.7759385108947754,
+      "learning_rate": 2.243918918918919e-05,
+      "loss": 0.0385,
+      "step": 4080
+    },
+    {
+      "epoch": 2.7635135135135136,
+      "grad_norm": 0.25599023699760437,
+      "learning_rate": 2.2371621621621622e-05,
+      "loss": 0.042,
+      "step": 4090
+    },
+    {
+      "epoch": 2.77027027027027,
+      "grad_norm": 0.44792553782463074,
+      "learning_rate": 2.2304054054054054e-05,
+      "loss": 0.0431,
+      "step": 4100
+    },
+    {
+      "epoch": 2.777027027027027,
+      "grad_norm": 0.32037609815597534,
+      "learning_rate": 2.2236486486486487e-05,
+      "loss": 0.0354,
+      "step": 4110
+    },
+    {
+      "epoch": 2.7837837837837838,
+      "grad_norm": 0.3700566291809082,
+      "learning_rate": 2.2168918918918922e-05,
+      "loss": 0.0371,
+      "step": 4120
+    },
+    {
+      "epoch": 2.7905405405405403,
+      "grad_norm": 0.3106760084629059,
+      "learning_rate": 2.210135135135135e-05,
+      "loss": 0.0294,
+      "step": 4130
+    },
+    {
+      "epoch": 2.7972972972972974,
+      "grad_norm": 0.29399368166923523,
+      "learning_rate": 2.2033783783783783e-05,
+      "loss": 0.045,
+      "step": 4140
+    },
+    {
+      "epoch": 2.804054054054054,
+      "grad_norm": 0.2866055965423584,
+      "learning_rate": 2.196621621621622e-05,
+      "loss": 0.0351,
+      "step": 4150
+    },
+    {
+      "epoch": 2.810810810810811,
+      "grad_norm": 0.18367105722427368,
+      "learning_rate": 2.189864864864865e-05,
+      "loss": 0.035,
+      "step": 4160
+    },
+    {
+      "epoch": 2.8175675675675675,
+      "grad_norm": 0.2915230393409729,
+      "learning_rate": 2.183108108108108e-05,
+      "loss": 0.0362,
+      "step": 4170
+    },
+    {
+      "epoch": 2.8243243243243246,
+      "grad_norm": 0.4409267008304596,
+      "learning_rate": 2.1763513513513516e-05,
+      "loss": 0.0536,
+      "step": 4180
+    },
+    {
+      "epoch": 2.831081081081081,
+      "grad_norm": 0.3458040952682495,
+      "learning_rate": 2.1695945945945948e-05,
+      "loss": 0.0356,
+      "step": 4190
+    },
+    {
+      "epoch": 2.8378378378378377,
+      "grad_norm": 0.2696792781352997,
+      "learning_rate": 2.162837837837838e-05,
+      "loss": 0.0448,
+      "step": 4200
+    },
+    {
+      "epoch": 2.8445945945945947,
+      "grad_norm": 0.3584458827972412,
+      "learning_rate": 2.1560810810810812e-05,
+      "loss": 0.0424,
+      "step": 4210
+    },
+    {
+      "epoch": 2.8513513513513513,
+      "grad_norm": 0.23099841177463531,
+      "learning_rate": 2.1493243243243245e-05,
+      "loss": 0.0378,
+      "step": 4220
+    },
+    {
+      "epoch": 2.858108108108108,
+      "grad_norm": 0.7650930285453796,
+      "learning_rate": 2.1425675675675677e-05,
+      "loss": 0.044,
+      "step": 4230
+    },
+    {
+      "epoch": 2.864864864864865,
+      "grad_norm": 0.49164295196533203,
+      "learning_rate": 2.135810810810811e-05,
+      "loss": 0.0535,
+      "step": 4240
+    },
+    {
+      "epoch": 2.8716216216216215,
+      "grad_norm": 0.4650588631629944,
+      "learning_rate": 2.129054054054054e-05,
+      "loss": 0.0358,
+      "step": 4250
+    },
+    {
+      "epoch": 2.8783783783783785,
+      "grad_norm": 0.4031845033168793,
+      "learning_rate": 2.1222972972972974e-05,
+      "loss": 0.0372,
+      "step": 4260
+    },
+    {
+      "epoch": 2.885135135135135,
+      "grad_norm": 0.2923171818256378,
+      "learning_rate": 2.1155405405405406e-05,
+      "loss": 0.0428,
+      "step": 4270
+    },
+    {
+      "epoch": 2.891891891891892,
+      "grad_norm": 0.16636265814304352,
+      "learning_rate": 2.1087837837837838e-05,
+      "loss": 0.0487,
+      "step": 4280
+    },
+    {
+      "epoch": 2.8986486486486487,
+      "grad_norm": 0.7599406838417053,
+      "learning_rate": 2.102027027027027e-05,
+      "loss": 0.0401,
+      "step": 4290
+    },
+    {
+      "epoch": 2.9054054054054053,
+      "grad_norm": 0.25301793217658997,
+      "learning_rate": 2.0952702702702703e-05,
+      "loss": 0.0338,
+      "step": 4300
+    },
+    {
+      "epoch": 2.9121621621621623,
+      "grad_norm": 0.1861271858215332,
+      "learning_rate": 2.0885135135135135e-05,
+      "loss": 0.0334,
+      "step": 4310
+    },
+    {
+      "epoch": 2.918918918918919,
+      "grad_norm": 0.25895753502845764,
+      "learning_rate": 2.081756756756757e-05,
+      "loss": 0.0311,
+      "step": 4320
+    },
+    {
+      "epoch": 2.9256756756756754,
+      "grad_norm": 0.42530789971351624,
+      "learning_rate": 2.075e-05,
+      "loss": 0.0398,
+      "step": 4330
+    },
+    {
+      "epoch": 2.9324324324324325,
+      "grad_norm": 0.463356614112854,
+      "learning_rate": 2.0682432432432432e-05,
+      "loss": 0.0366,
+      "step": 4340
+    },
+    {
+      "epoch": 2.939189189189189,
+      "grad_norm": 0.4635908007621765,
+      "learning_rate": 2.0614864864864867e-05,
+      "loss": 0.0457,
+      "step": 4350
+    },
+    {
+      "epoch": 2.945945945945946,
+      "grad_norm": 0.5640028715133667,
+      "learning_rate": 2.05472972972973e-05,
+      "loss": 0.0374,
+      "step": 4360
+    },
+    {
+      "epoch": 2.9527027027027026,
+      "grad_norm": 0.5728152394294739,
+      "learning_rate": 2.047972972972973e-05,
+      "loss": 0.0388,
+      "step": 4370
+    },
+    {
+      "epoch": 2.9594594594594597,
+      "grad_norm": 0.25366121530532837,
+      "learning_rate": 2.0412162162162164e-05,
+      "loss": 0.0491,
+      "step": 4380
+    },
+    {
+      "epoch": 2.9662162162162162,
+      "grad_norm": 0.415521502494812,
+      "learning_rate": 2.0344594594594596e-05,
+      "loss": 0.0365,
+      "step": 4390
+    },
+    {
+      "epoch": 2.972972972972973,
+      "grad_norm": 0.24194630980491638,
+      "learning_rate": 2.027702702702703e-05,
+      "loss": 0.0356,
+      "step": 4400
+    },
+    {
+      "epoch": 2.97972972972973,
+      "grad_norm": 0.39360618591308594,
+      "learning_rate": 2.020945945945946e-05,
+      "loss": 0.0359,
+      "step": 4410
+    },
+    {
+      "epoch": 2.9864864864864864,
+      "grad_norm": 0.21369539201259613,
+      "learning_rate": 2.0141891891891893e-05,
+      "loss": 0.0439,
+      "step": 4420
+    },
+    {
+      "epoch": 2.993243243243243,
+      "grad_norm": 0.22118353843688965,
+      "learning_rate": 2.0074324324324325e-05,
+      "loss": 0.0315,
+      "step": 4430
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.870568037033081,
+      "learning_rate": 2.0006756756756758e-05,
+      "loss": 0.04,
+      "step": 4440
+    },
+    {
+      "epoch": 3.0,
+      "eval_loss": 0.02539176121354103,
+      "eval_runtime": 375.2111,
+      "eval_samples_per_second": 2.785,
+      "eval_steps_per_second": 0.698,
+      "step": 4440
+    },
+    {
+      "epoch": 3.0067567567567566,
+      "grad_norm": 0.40895864367485046,
+      "learning_rate": 1.993918918918919e-05,
+      "loss": 0.0406,
+      "step": 4450
+    },
+    {
+      "epoch": 3.0135135135135136,
+      "grad_norm": 0.26677393913269043,
+      "learning_rate": 1.9871621621621622e-05,
+      "loss": 0.0296,
+      "step": 4460
+    },
+    {
+      "epoch": 3.02027027027027,
+      "grad_norm": 0.29075056314468384,
+      "learning_rate": 1.9804054054054054e-05,
+      "loss": 0.0343,
+      "step": 4470
+    },
+    {
+      "epoch": 3.027027027027027,
+      "grad_norm": 0.31686338782310486,
+      "learning_rate": 1.973648648648649e-05,
+      "loss": 0.0341,
+      "step": 4480
+    },
+    {
+      "epoch": 3.0337837837837838,
+      "grad_norm": 0.3891538083553314,
+      "learning_rate": 1.966891891891892e-05,
+      "loss": 0.0383,
+      "step": 4490
+    },
+    {
+      "epoch": 3.0405405405405403,
+      "grad_norm": 0.29185813665390015,
+      "learning_rate": 1.960135135135135e-05,
+      "loss": 0.0642,
+      "step": 4500
+    },
+    {
+      "epoch": 3.0472972972972974,
+      "grad_norm": 0.4038207530975342,
+      "learning_rate": 1.9533783783783784e-05,
+      "loss": 0.0353,
+      "step": 4510
+    },
+    {
+      "epoch": 3.054054054054054,
+      "grad_norm": 0.12403042614459991,
+      "learning_rate": 1.946621621621622e-05,
+      "loss": 0.033,
+      "step": 4520
+    },
+    {
+      "epoch": 3.060810810810811,
+      "grad_norm": 0.32900774478912354,
+      "learning_rate": 1.9398648648648648e-05,
+      "loss": 0.0375,
+      "step": 4530
+    },
+    {
+      "epoch": 3.0675675675675675,
+      "grad_norm": 0.5344650149345398,
+      "learning_rate": 1.933108108108108e-05,
+      "loss": 0.0512,
+      "step": 4540
+    },
+    {
+      "epoch": 3.074324324324324,
+      "grad_norm": 0.2389213591814041,
+      "learning_rate": 1.9263513513513516e-05,
+      "loss": 0.0318,
+      "step": 4550
+    },
+    {
+      "epoch": 3.081081081081081,
+      "grad_norm": 0.30284884572029114,
+      "learning_rate": 1.9195945945945948e-05,
+      "loss": 0.0264,
+      "step": 4560
+    },
+    {
+      "epoch": 3.0878378378378377,
+      "grad_norm": 0.32080718874931335,
+      "learning_rate": 1.9128378378378377e-05,
+      "loss": 0.0413,
+      "step": 4570
+    },
+    {
+      "epoch": 3.0945945945945947,
+      "grad_norm": 0.3493003845214844,
+      "learning_rate": 1.9060810810810813e-05,
+      "loss": 0.0407,
+      "step": 4580
+    },
+    {
+      "epoch": 3.1013513513513513,
+      "grad_norm": 0.17225845158100128,
+      "learning_rate": 1.8993243243243245e-05,
+      "loss": 0.028,
+      "step": 4590
+    },
+    {
+      "epoch": 3.108108108108108,
+      "grad_norm": 0.7874749302864075,
+      "learning_rate": 1.8925675675675677e-05,
+      "loss": 0.037,
+      "step": 4600
+    },
+    {
+      "epoch": 3.114864864864865,
+      "grad_norm": 0.5345508456230164,
+      "learning_rate": 1.885810810810811e-05,
+      "loss": 0.0372,
+      "step": 4610
+    },
+    {
+      "epoch": 3.1216216216216215,
+      "grad_norm": 0.5025699734687805,
+      "learning_rate": 1.8790540540540542e-05,
+      "loss": 0.0374,
+      "step": 4620
+    },
+    {
+      "epoch": 3.1283783783783785,
+      "grad_norm": 0.3688373565673828,
+      "learning_rate": 1.8722972972972974e-05,
+      "loss": 0.0648,
+      "step": 4630
+    },
+    {
+      "epoch": 3.135135135135135,
+      "grad_norm": 0.29858043789863586,
+      "learning_rate": 1.8655405405405406e-05,
+      "loss": 0.0375,
+      "step": 4640
+    },
+    {
+      "epoch": 3.141891891891892,
+      "grad_norm": 0.5293399691581726,
+      "learning_rate": 1.858783783783784e-05,
+      "loss": 0.0411,
+      "step": 4650
+    },
+    {
+      "epoch": 3.1486486486486487,
+      "grad_norm": 0.3941536545753479,
+      "learning_rate": 1.852027027027027e-05,
+      "loss": 0.0379,
+      "step": 4660
+    },
+    {
+      "epoch": 3.1554054054054053,
+      "grad_norm": 0.2352730631828308,
+      "learning_rate": 1.8452702702702703e-05,
+      "loss": 0.0453,
+      "step": 4670
+    },
+    {
+      "epoch": 3.1621621621621623,
+      "grad_norm": 0.418706476688385,
+      "learning_rate": 1.838513513513514e-05,
+      "loss": 0.0363,
+      "step": 4680
+    },
+    {
+      "epoch": 3.168918918918919,
+      "grad_norm": 0.29120874404907227,
+      "learning_rate": 1.8317567567567568e-05,
+      "loss": 0.0305,
+      "step": 4690
+    },
+    {
+      "epoch": 3.175675675675676,
+      "grad_norm": 0.22650307416915894,
+      "learning_rate": 1.825e-05,
+      "loss": 0.0324,
+      "step": 4700
+    },
+    {
+      "epoch": 3.1824324324324325,
+      "grad_norm": 0.26235389709472656,
+      "learning_rate": 1.8182432432432432e-05,
+      "loss": 0.0455,
+      "step": 4710
+    },
+    {
+      "epoch": 3.189189189189189,
+      "grad_norm": 0.1589524745941162,
+      "learning_rate": 1.8114864864864868e-05,
+      "loss": 0.0413,
+      "step": 4720
+    },
+    {
+      "epoch": 3.195945945945946,
+      "grad_norm": 0.43323051929473877,
+      "learning_rate": 1.8047297297297297e-05,
+      "loss": 0.0482,
+      "step": 4730
+    },
+    {
+      "epoch": 3.2027027027027026,
+      "grad_norm": 0.14209793508052826,
+      "learning_rate": 1.797972972972973e-05,
+      "loss": 0.0304,
+      "step": 4740
+    },
+    {
+      "epoch": 3.2094594594594597,
+      "grad_norm": 0.601881742477417,
+      "learning_rate": 1.7912162162162164e-05,
+      "loss": 0.0445,
+      "step": 4750
+    },
+    {
+      "epoch": 3.2162162162162162,
+      "grad_norm": 0.36841222643852234,
+      "learning_rate": 1.7844594594594597e-05,
+      "loss": 0.0409,
+      "step": 4760
+    },
+    {
+      "epoch": 3.222972972972973,
+      "grad_norm": 0.6512994170188904,
+      "learning_rate": 1.7777027027027026e-05,
+      "loss": 0.0325,
+      "step": 4770
+    },
+    {
+      "epoch": 3.22972972972973,
+      "grad_norm": 0.27603307366371155,
+      "learning_rate": 1.770945945945946e-05,
+      "loss": 0.0398,
+      "step": 4780
+    },
+    {
+      "epoch": 3.2364864864864864,
+      "grad_norm": 0.49570050835609436,
+      "learning_rate": 1.7641891891891893e-05,
+      "loss": 0.049,
+      "step": 4790
+    },
+    {
+      "epoch": 3.2432432432432434,
+      "grad_norm": 0.453434020280838,
+      "learning_rate": 1.7574324324324326e-05,
+      "loss": 0.0397,
+      "step": 4800
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 0.7412545084953308,
+      "learning_rate": 1.7506756756756758e-05,
+      "loss": 0.0369,
+      "step": 4810
+    },
+    {
+      "epoch": 3.2567567567567566,
+      "grad_norm": 0.13515311479568481,
+      "learning_rate": 1.743918918918919e-05,
+      "loss": 0.0329,
+      "step": 4820
+    },
+    {
+      "epoch": 3.2635135135135136,
+      "grad_norm": 0.26775646209716797,
+      "learning_rate": 1.7371621621621622e-05,
+      "loss": 0.0321,
+      "step": 4830
+    },
+    {
+      "epoch": 3.27027027027027,
+      "grad_norm": 0.4981396496295929,
+      "learning_rate": 1.7304054054054055e-05,
+      "loss": 0.0443,
+      "step": 4840
+    },
+    {
+      "epoch": 3.277027027027027,
+      "grad_norm": 0.33926641941070557,
+      "learning_rate": 1.7236486486486487e-05,
+      "loss": 0.0548,
+      "step": 4850
+    },
+    {
+      "epoch": 3.2837837837837838,
+      "grad_norm": 0.23301415145397186,
+      "learning_rate": 1.716891891891892e-05,
+      "loss": 0.0425,
+      "step": 4860
+    },
+    {
+      "epoch": 3.2905405405405403,
+      "grad_norm": 0.2623809576034546,
+      "learning_rate": 1.710135135135135e-05,
+      "loss": 0.0319,
+      "step": 4870
+    },
+    {
+      "epoch": 3.2972972972972974,
+      "grad_norm": 0.33261558413505554,
+      "learning_rate": 1.7033783783783787e-05,
+      "loss": 0.0268,
+      "step": 4880
+    },
+    {
+      "epoch": 3.304054054054054,
+      "grad_norm": 0.24392476677894592,
+      "learning_rate": 1.6966216216216216e-05,
+      "loss": 0.0399,
+      "step": 4890
+    },
+    {
+      "epoch": 3.310810810810811,
+      "grad_norm": 0.5819364190101624,
+      "learning_rate": 1.6898648648648648e-05,
+      "loss": 0.0471,
+      "step": 4900
+    },
+    {
+      "epoch": 3.3175675675675675,
+      "grad_norm": 0.5962876677513123,
+      "learning_rate": 1.683108108108108e-05,
+      "loss": 0.0438,
+      "step": 4910
+    },
+    {
+      "epoch": 3.3243243243243246,
+      "grad_norm": 0.226814866065979,
+      "learning_rate": 1.6763513513513516e-05,
+      "loss": 0.029,
+      "step": 4920
+    },
+    {
+      "epoch": 3.331081081081081,
+      "grad_norm": 0.5372828841209412,
+      "learning_rate": 1.6695945945945945e-05,
+      "loss": 0.0413,
+      "step": 4930
+    },
+    {
+      "epoch": 3.3378378378378377,
+      "grad_norm": 0.11127971112728119,
+      "learning_rate": 1.6628378378378377e-05,
+      "loss": 0.0561,
+      "step": 4940
+    },
+    {
+      "epoch": 3.3445945945945947,
+      "grad_norm": 0.24800080060958862,
+      "learning_rate": 1.6560810810810813e-05,
+      "loss": 0.0315,
+      "step": 4950
+    },
+    {
+      "epoch": 3.3513513513513513,
+      "grad_norm": 0.624832272529602,
+      "learning_rate": 1.6493243243243245e-05,
+      "loss": 0.0344,
+      "step": 4960
+    },
+    {
+      "epoch": 3.358108108108108,
+      "grad_norm": 0.37135404348373413,
+      "learning_rate": 1.6425675675675674e-05,
+      "loss": 0.0389,
+      "step": 4970
+    },
+    {
+      "epoch": 3.364864864864865,
+      "grad_norm": 0.5197146534919739,
+      "learning_rate": 1.635810810810811e-05,
+      "loss": 0.0359,
+      "step": 4980
+    },
+    {
+      "epoch": 3.3716216216216215,
+      "grad_norm": 0.34491223096847534,
+      "learning_rate": 1.6290540540540542e-05,
+      "loss": 0.0319,
+      "step": 4990
+    },
+    {
+      "epoch": 3.3783783783783785,
+      "grad_norm": 0.3033651113510132,
+      "learning_rate": 1.6222972972972974e-05,
+      "loss": 0.0584,
+      "step": 5000
+    },
+    {
+      "epoch": 3.385135135135135,
+      "grad_norm": 0.3707546591758728,
+      "learning_rate": 1.6155405405405406e-05,
+      "loss": 0.0361,
+      "step": 5010
+    },
+    {
+      "epoch": 3.391891891891892,
+      "grad_norm": 0.32084357738494873,
+      "learning_rate": 1.608783783783784e-05,
+      "loss": 0.0447,
+      "step": 5020
+    },
+    {
+      "epoch": 3.3986486486486487,
+      "grad_norm": 0.24217896163463593,
+      "learning_rate": 1.602027027027027e-05,
+      "loss": 0.0351,
+      "step": 5030
+    },
+    {
+      "epoch": 3.4054054054054053,
+      "grad_norm": 0.24498729407787323,
+      "learning_rate": 1.5952702702702703e-05,
+      "loss": 0.0511,
+      "step": 5040
+    },
+    {
+      "epoch": 3.4121621621621623,
+      "grad_norm": 0.3093695342540741,
+      "learning_rate": 1.5885135135135135e-05,
+      "loss": 0.0358,
+      "step": 5050
+    },
+    {
+      "epoch": 3.418918918918919,
+      "grad_norm": 0.23047859966754913,
+      "learning_rate": 1.5817567567567568e-05,
+      "loss": 0.035,
+      "step": 5060
+    },
+    {
+      "epoch": 3.4256756756756754,
+      "grad_norm": 0.2470131516456604,
+      "learning_rate": 1.575e-05,
+      "loss": 0.036,
+      "step": 5070
+    },
+    {
+      "epoch": 3.4324324324324325,
+      "grad_norm": 0.22567608952522278,
+      "learning_rate": 1.5682432432432436e-05,
+      "loss": 0.042,
+      "step": 5080
+    },
+    {
+      "epoch": 3.439189189189189,
+      "grad_norm": 0.4833430051803589,
+      "learning_rate": 1.5614864864864864e-05,
+      "loss": 0.0328,
+      "step": 5090
+    },
+    {
+      "epoch": 3.445945945945946,
+      "grad_norm": 0.2073460817337036,
+      "learning_rate": 1.5547297297297297e-05,
+      "loss": 0.0324,
+      "step": 5100
+    },
+    {
+      "epoch": 3.4527027027027026,
+      "grad_norm": 0.4520673453807831,
+      "learning_rate": 1.547972972972973e-05,
+      "loss": 0.0425,
+      "step": 5110
+    },
+    {
+      "epoch": 3.4594594594594597,
+      "grad_norm": 0.387984961271286,
+      "learning_rate": 1.5412162162162165e-05,
+      "loss": 0.0377,
+      "step": 5120
+    },
+    {
+      "epoch": 3.4662162162162162,
+      "grad_norm": 0.1993335336446762,
+      "learning_rate": 1.5344594594594594e-05,
+      "loss": 0.0418,
+      "step": 5130
+    },
+    {
+      "epoch": 3.472972972972973,
+      "grad_norm": 0.17682744562625885,
+      "learning_rate": 1.5277027027027026e-05,
+      "loss": 0.0458,
+      "step": 5140
+    },
+    {
+      "epoch": 3.47972972972973,
+      "grad_norm": 0.2657202184200287,
+      "learning_rate": 1.520945945945946e-05,
+      "loss": 0.0377,
+      "step": 5150
+    },
+    {
+      "epoch": 3.4864864864864864,
+      "grad_norm": 0.21221712231636047,
+      "learning_rate": 1.5141891891891894e-05,
+      "loss": 0.0418,
+      "step": 5160
+    },
+    {
+      "epoch": 3.4932432432432434,
+      "grad_norm": 0.2672747075557709,
+      "learning_rate": 1.5074324324324324e-05,
+      "loss": 0.0336,
+      "step": 5170
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 0.22585569322109222,
+      "learning_rate": 1.5006756756756756e-05,
+      "loss": 0.0353,
+      "step": 5180
+    },
+    {
+      "epoch": 3.506756756756757,
+      "grad_norm": 0.4884142279624939,
+      "learning_rate": 1.493918918918919e-05,
+      "loss": 0.031,
+      "step": 5190
+    },
+    {
+      "epoch": 3.5135135135135136,
+      "grad_norm": 0.2212933450937271,
+      "learning_rate": 1.4871621621621623e-05,
+      "loss": 0.0333,
+      "step": 5200
+    },
+    {
+      "epoch": 3.52027027027027,
+      "grad_norm": 0.22906561195850372,
+      "learning_rate": 1.4804054054054053e-05,
+      "loss": 0.0353,
+      "step": 5210
+    },
+    {
+      "epoch": 3.527027027027027,
+      "grad_norm": 0.2564111053943634,
+      "learning_rate": 1.4736486486486487e-05,
+      "loss": 0.0334,
+      "step": 5220
+    },
+    {
+      "epoch": 3.5337837837837838,
+      "grad_norm": 0.25793004035949707,
+      "learning_rate": 1.466891891891892e-05,
+      "loss": 0.0296,
+      "step": 5230
+    },
+    {
+      "epoch": 3.5405405405405403,
+      "grad_norm": 0.27988219261169434,
+      "learning_rate": 1.4601351351351353e-05,
+      "loss": 0.0323,
+      "step": 5240
+    },
+    {
+      "epoch": 3.5472972972972974,
+      "grad_norm": 0.2663809657096863,
+      "learning_rate": 1.4533783783783784e-05,
+      "loss": 0.027,
+      "step": 5250
+    },
+    {
+      "epoch": 3.554054054054054,
+      "grad_norm": 0.30205437541007996,
+      "learning_rate": 1.4466216216216216e-05,
+      "loss": 0.0441,
+      "step": 5260
+    },
+    {
+      "epoch": 3.560810810810811,
+      "grad_norm": 0.47929883003234863,
+      "learning_rate": 1.439864864864865e-05,
+      "loss": 0.0389,
+      "step": 5270
+    },
+    {
+      "epoch": 3.5675675675675675,
+      "grad_norm": 0.20964276790618896,
+      "learning_rate": 1.4331081081081082e-05,
+      "loss": 0.0393,
+      "step": 5280
+    },
+    {
+      "epoch": 3.5743243243243246,
+      "grad_norm": 0.4340321123600006,
+      "learning_rate": 1.4263513513513513e-05,
+      "loss": 0.0411,
+      "step": 5290
+    },
+    {
+      "epoch": 3.581081081081081,
+      "grad_norm": 0.338561087846756,
+      "learning_rate": 1.4195945945945945e-05,
+      "loss": 0.0364,
+      "step": 5300
+    },
+    {
+      "epoch": 3.5878378378378377,
+      "grad_norm": 0.4030533730983734,
+      "learning_rate": 1.412837837837838e-05,
+      "loss": 0.0371,
+      "step": 5310
+    },
+    {
+      "epoch": 3.5945945945945947,
+      "grad_norm": 0.33323097229003906,
+      "learning_rate": 1.4060810810810811e-05,
+      "loss": 0.0457,
+      "step": 5320
+    },
+    {
+      "epoch": 3.6013513513513513,
+      "grad_norm": 0.22286130487918854,
+      "learning_rate": 1.3993243243243242e-05,
+      "loss": 0.0351,
+      "step": 5330
+    },
+    {
+      "epoch": 3.608108108108108,
+      "grad_norm": 0.16391167044639587,
+      "learning_rate": 1.3925675675675676e-05,
+      "loss": 0.0287,
+      "step": 5340
+    },
+    {
+      "epoch": 3.614864864864865,
+      "grad_norm": 0.2558617889881134,
+      "learning_rate": 1.3858108108108108e-05,
+      "loss": 0.0401,
+      "step": 5350
+    },
+    {
+      "epoch": 3.6216216216216215,
+      "grad_norm": 0.17443585395812988,
+      "learning_rate": 1.3790540540540542e-05,
+      "loss": 0.0383,
+      "step": 5360
+    },
+    {
+      "epoch": 3.6283783783783785,
+      "grad_norm": 0.4235523045063019,
+      "learning_rate": 1.3722972972972974e-05,
+      "loss": 0.0459,
+      "step": 5370
+    },
+    {
+      "epoch": 3.635135135135135,
+      "grad_norm": 0.2450476437807083,
+      "learning_rate": 1.3655405405405405e-05,
+      "loss": 0.0442,
+      "step": 5380
+    },
+    {
+      "epoch": 3.641891891891892,
+      "grad_norm": 0.4506691098213196,
+      "learning_rate": 1.3587837837837839e-05,
+      "loss": 0.0397,
+      "step": 5390
+    },
+    {
+      "epoch": 3.6486486486486487,
+      "grad_norm": 0.29354047775268555,
+      "learning_rate": 1.3520270270270271e-05,
+      "loss": 0.0364,
+      "step": 5400
+    },
+    {
+      "epoch": 3.6554054054054053,
+      "grad_norm": 0.27163824439048767,
+      "learning_rate": 1.3452702702702705e-05,
+      "loss": 0.0354,
+      "step": 5410
+    },
+    {
+      "epoch": 3.6621621621621623,
+      "grad_norm": 0.19688227772712708,
+      "learning_rate": 1.3385135135135136e-05,
+      "loss": 0.0312,
+      "step": 5420
+    },
+    {
+      "epoch": 3.668918918918919,
+      "grad_norm": 0.4131430983543396,
+      "learning_rate": 1.3317567567567568e-05,
+      "loss": 0.0321,
+      "step": 5430
+    },
+    {
+      "epoch": 3.6756756756756754,
+      "grad_norm": 0.31735169887542725,
+      "learning_rate": 1.3250000000000002e-05,
+      "loss": 0.0349,
+      "step": 5440
+    },
+    {
+      "epoch": 3.6824324324324325,
+      "grad_norm": 0.29376840591430664,
+      "learning_rate": 1.3182432432432434e-05,
+      "loss": 0.0337,
+      "step": 5450
+    },
+    {
+      "epoch": 3.689189189189189,
+      "grad_norm": 0.22491705417633057,
+      "learning_rate": 1.3114864864864865e-05,
+      "loss": 0.0422,
+      "step": 5460
+    },
+    {
+      "epoch": 3.695945945945946,
+      "grad_norm": 0.3273302912712097,
+      "learning_rate": 1.3047297297297299e-05,
+      "loss": 0.0334,
+      "step": 5470
+    },
+    {
+      "epoch": 3.7027027027027026,
+      "grad_norm": 0.2812623977661133,
+      "learning_rate": 1.2979729729729731e-05,
+      "loss": 0.0418,
+      "step": 5480
+    },
+    {
+      "epoch": 3.7094594594594597,
+      "grad_norm": 0.34500834345817566,
+      "learning_rate": 1.2912162162162165e-05,
+      "loss": 0.0345,
+      "step": 5490
+    },
+    {
+      "epoch": 3.7162162162162162,
+      "grad_norm": 0.40858641266822815,
+      "learning_rate": 1.2844594594594594e-05,
+      "loss": 0.0445,
+      "step": 5500
+    },
+    {
+      "epoch": 3.722972972972973,
+      "grad_norm": 0.0810331404209137,
+      "learning_rate": 1.2777027027027028e-05,
+      "loss": 0.0505,
+      "step": 5510
+    },
+    {
+      "epoch": 3.72972972972973,
+      "grad_norm": 0.35677656531333923,
+      "learning_rate": 1.270945945945946e-05,
+      "loss": 0.0398,
+      "step": 5520
+    },
+    {
+      "epoch": 3.7364864864864864,
+      "grad_norm": 0.2926190495491028,
+      "learning_rate": 1.2641891891891894e-05,
+      "loss": 0.0385,
+      "step": 5530
+    },
+    {
+      "epoch": 3.743243243243243,
+      "grad_norm": 0.26496851444244385,
+      "learning_rate": 1.2574324324324324e-05,
+      "loss": 0.0334,
+      "step": 5540
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 0.27788522839546204,
+      "learning_rate": 1.2506756756756757e-05,
+      "loss": 0.046,
+      "step": 5550
+    },
+    {
+      "epoch": 3.756756756756757,
+      "grad_norm": 0.2423742562532425,
+      "learning_rate": 1.243918918918919e-05,
+      "loss": 0.0305,
+      "step": 5560
+    },
+    {
+      "epoch": 3.7635135135135136,
+      "grad_norm": 0.19838376343250275,
+      "learning_rate": 1.2371621621621623e-05,
+      "loss": 0.0257,
+      "step": 5570
+    },
+    {
+      "epoch": 3.77027027027027,
+      "grad_norm": 0.4191855490207672,
+      "learning_rate": 1.2304054054054055e-05,
+      "loss": 0.043,
+      "step": 5580
+    },
+    {
+      "epoch": 3.777027027027027,
+      "grad_norm": 0.19260361790657043,
+      "learning_rate": 1.2236486486486487e-05,
+      "loss": 0.0349,
+      "step": 5590
+    },
+    {
+      "epoch": 3.7837837837837838,
+      "grad_norm": 0.14565953612327576,
+      "learning_rate": 1.216891891891892e-05,
+      "loss": 0.0398,
+      "step": 5600
+    },
+    {
+      "epoch": 3.7905405405405403,
+      "grad_norm": 0.37692561745643616,
+      "learning_rate": 1.2101351351351352e-05,
+      "loss": 0.0408,
+      "step": 5610
+    },
+    {
+      "epoch": 3.7972972972972974,
+      "grad_norm": 0.18849320709705353,
+      "learning_rate": 1.2033783783783784e-05,
+      "loss": 0.0325,
+      "step": 5620
+    },
+    {
+      "epoch": 3.804054054054054,
+      "grad_norm": 0.3735746145248413,
+      "learning_rate": 1.1966216216216216e-05,
+      "loss": 0.044,
+      "step": 5630
+    },
+    {
+      "epoch": 3.810810810810811,
+      "grad_norm": 0.1767602264881134,
+      "learning_rate": 1.189864864864865e-05,
+      "loss": 0.0348,
+      "step": 5640
+    },
+    {
+      "epoch": 3.8175675675675675,
+      "grad_norm": 0.30753883719444275,
+      "learning_rate": 1.1831081081081081e-05,
+      "loss": 0.0326,
+      "step": 5650
+    },
+    {
+      "epoch": 3.8243243243243246,
+      "grad_norm": 0.2788727879524231,
+      "learning_rate": 1.1763513513513515e-05,
+      "loss": 0.0366,
+      "step": 5660
+    },
+    {
+      "epoch": 3.831081081081081,
+      "grad_norm": 0.27036169171333313,
+      "learning_rate": 1.1695945945945947e-05,
+      "loss": 0.0291,
+      "step": 5670
+    },
+    {
+      "epoch": 3.8378378378378377,
+      "grad_norm": 0.23913073539733887,
+      "learning_rate": 1.162837837837838e-05,
+      "loss": 0.0297,
+      "step": 5680
+    },
+    {
+      "epoch": 3.8445945945945947,
+      "grad_norm": 0.296947181224823,
+      "learning_rate": 1.1560810810810812e-05,
+      "loss": 0.029,
+      "step": 5690
+    },
+    {
+      "epoch": 3.8513513513513513,
+      "grad_norm": 0.16844850778579712,
+      "learning_rate": 1.1493243243243244e-05,
+      "loss": 0.0383,
+      "step": 5700
+    },
+    {
+      "epoch": 3.858108108108108,
+      "grad_norm": 0.275686115026474,
+      "learning_rate": 1.1425675675675676e-05,
+      "loss": 0.033,
+      "step": 5710
+    },
+    {
+      "epoch": 3.864864864864865,
+      "grad_norm": 0.19295139610767365,
+      "learning_rate": 1.1358108108108108e-05,
+      "loss": 0.0401,
+      "step": 5720
+    },
+    {
+      "epoch": 3.8716216216216215,
+      "grad_norm": 0.4507526159286499,
+      "learning_rate": 1.129054054054054e-05,
+      "loss": 0.0423,
+      "step": 5730
+    },
+    {
+      "epoch": 3.8783783783783785,
+      "grad_norm": 0.16554881632328033,
+      "learning_rate": 1.1222972972972975e-05,
+      "loss": 0.0468,
+      "step": 5740
+    },
+    {
+      "epoch": 3.885135135135135,
+      "grad_norm": 0.22985632717609406,
+      "learning_rate": 1.1155405405405405e-05,
+      "loss": 0.0275,
+      "step": 5750
+    },
+    {
+      "epoch": 3.891891891891892,
+      "grad_norm": 0.13783544301986694,
+      "learning_rate": 1.1087837837837839e-05,
+      "loss": 0.037,
+      "step": 5760
+    },
+    {
+      "epoch": 3.8986486486486487,
+      "grad_norm": 0.4291518032550812,
+      "learning_rate": 1.1020270270270271e-05,
+      "loss": 0.0549,
+      "step": 5770
+    },
+    {
+      "epoch": 3.9054054054054053,
+      "grad_norm": 0.21507710218429565,
+      "learning_rate": 1.0952702702702704e-05,
+      "loss": 0.0376,
+      "step": 5780
+    },
+    {
+      "epoch": 3.9121621621621623,
+      "grad_norm": 0.2827785015106201,
+      "learning_rate": 1.0885135135135136e-05,
+      "loss": 0.0366,
+      "step": 5790
+    },
+    {
+      "epoch": 3.918918918918919,
+      "grad_norm": 0.3014868199825287,
+      "learning_rate": 1.0817567567567568e-05,
+      "loss": 0.0289,
+      "step": 5800
+    },
+    {
+      "epoch": 3.9256756756756754,
+      "grad_norm": 0.14509421586990356,
+      "learning_rate": 1.075e-05,
+      "loss": 0.0293,
+      "step": 5810
+    },
+    {
+      "epoch": 3.9324324324324325,
+      "grad_norm": 0.20983301103115082,
+      "learning_rate": 1.0682432432432433e-05,
+      "loss": 0.0318,
+      "step": 5820
+    },
+    {
+      "epoch": 3.939189189189189,
+      "grad_norm": 0.1960182636976242,
+      "learning_rate": 1.0614864864864865e-05,
+      "loss": 0.0411,
+      "step": 5830
+    },
+    {
+      "epoch": 3.945945945945946,
+      "grad_norm": 0.41651207208633423,
+      "learning_rate": 1.0547297297297299e-05,
+      "loss": 0.0293,
+      "step": 5840
+    },
+    {
+      "epoch": 3.9527027027027026,
+      "grad_norm": 0.3032841980457306,
+      "learning_rate": 1.047972972972973e-05,
+      "loss": 0.0413,
+      "step": 5850
+    },
+    {
+      "epoch": 3.9594594594594597,
+      "grad_norm": 0.2482951283454895,
+      "learning_rate": 1.0412162162162163e-05,
+      "loss": 0.0281,
+      "step": 5860
+    },
+    {
+      "epoch": 3.9662162162162162,
+      "grad_norm": 0.1527014523744583,
+      "learning_rate": 1.0344594594594596e-05,
+      "loss": 0.0407,
+      "step": 5870
+    },
+    {
+      "epoch": 3.972972972972973,
+      "grad_norm": 0.19820819795131683,
+      "learning_rate": 1.0277027027027028e-05,
+      "loss": 0.0312,
+      "step": 5880
+    },
+    {
+      "epoch": 3.97972972972973,
+      "grad_norm": 0.2963590621948242,
+      "learning_rate": 1.020945945945946e-05,
+      "loss": 0.0359,
+      "step": 5890
+    },
+    {
+      "epoch": 3.9864864864864864,
+      "grad_norm": 0.4077160358428955,
+      "learning_rate": 1.0141891891891892e-05,
+      "loss": 0.0439,
+      "step": 5900
+    },
+    {
+      "epoch": 3.993243243243243,
+      "grad_norm": 0.2682552635669708,
+      "learning_rate": 1.0074324324324325e-05,
+      "loss": 0.026,
+      "step": 5910
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 0.41804584860801697,
+      "learning_rate": 1.0006756756756757e-05,
+      "loss": 0.0525,
+      "step": 5920
+    },
+    {
+      "epoch": 4.0,
+      "eval_loss": 0.02366216853260994,
+      "eval_runtime": 246.5939,
+      "eval_samples_per_second": 4.238,
+      "eval_steps_per_second": 1.062,
+      "step": 5920
+    },
+    {
+      "epoch": 4.006756756756757,
+      "grad_norm": 0.3444773554801941,
+      "learning_rate": 9.93918918918919e-06,
+      "loss": 0.0437,
+      "step": 5930
+    },
+    {
+      "epoch": 4.013513513513513,
+      "grad_norm": 0.3024560213088989,
+      "learning_rate": 9.871621621621623e-06,
+      "loss": 0.0465,
+      "step": 5940
+    },
+    {
+      "epoch": 4.02027027027027,
+      "grad_norm": 0.25490209460258484,
+      "learning_rate": 9.804054054054054e-06,
+      "loss": 0.0451,
+      "step": 5950
+    },
+    {
+      "epoch": 4.027027027027027,
+      "grad_norm": 0.23857267200946808,
+      "learning_rate": 9.736486486486488e-06,
+      "loss": 0.0338,
+      "step": 5960
+    },
+    {
+      "epoch": 4.033783783783784,
+      "grad_norm": 0.22253508865833282,
+      "learning_rate": 9.66891891891892e-06,
+      "loss": 0.0307,
+      "step": 5970
+    },
+    {
+      "epoch": 4.04054054054054,
+      "grad_norm": 0.3308720588684082,
+      "learning_rate": 9.601351351351352e-06,
+      "loss": 0.0363,
+      "step": 5980
+    },
+    {
+      "epoch": 4.047297297297297,
+      "grad_norm": 0.4275621473789215,
+      "learning_rate": 9.533783783783784e-06,
+      "loss": 0.0433,
+      "step": 5990
+    },
+    {
+      "epoch": 4.054054054054054,
+      "grad_norm": 0.2730522155761719,
+      "learning_rate": 9.466216216216217e-06,
+      "loss": 0.0395,
+      "step": 6000
+    },
+    {
+      "epoch": 4.0608108108108105,
+      "grad_norm": 0.5394574999809265,
+      "learning_rate": 9.398648648648649e-06,
+      "loss": 0.0308,
+      "step": 6010
+    },
+    {
+      "epoch": 4.0675675675675675,
+      "grad_norm": 0.4886311888694763,
+      "learning_rate": 9.331081081081081e-06,
+      "loss": 0.0304,
+      "step": 6020
+    },
+    {
+      "epoch": 4.074324324324325,
+      "grad_norm": 0.34788379073143005,
+      "learning_rate": 9.263513513513513e-06,
+      "loss": 0.0463,
+      "step": 6030
+    },
+    {
+      "epoch": 4.081081081081081,
+      "grad_norm": 0.6314213275909424,
+      "learning_rate": 9.195945945945947e-06,
+      "loss": 0.0339,
+      "step": 6040
+    },
+    {
+      "epoch": 4.087837837837838,
+      "grad_norm": 0.11922051012516022,
+      "learning_rate": 9.128378378378378e-06,
+      "loss": 0.0275,
+      "step": 6050
+    },
+    {
+      "epoch": 4.094594594594595,
+      "grad_norm": 0.321501225233078,
+      "learning_rate": 9.060810810810812e-06,
+      "loss": 0.0306,
+      "step": 6060
+    },
+    {
+      "epoch": 4.101351351351352,
+      "grad_norm": 0.6931231617927551,
+      "learning_rate": 8.993243243243244e-06,
+      "loss": 0.0373,
+      "step": 6070
+    },
+    {
+      "epoch": 4.108108108108108,
+      "grad_norm": 0.32703468203544617,
+      "learning_rate": 8.925675675675676e-06,
+      "loss": 0.0279,
+      "step": 6080
+    },
+    {
+      "epoch": 4.114864864864865,
+      "grad_norm": 0.5395735502243042,
+      "learning_rate": 8.858108108108109e-06,
+      "loss": 0.0384,
+      "step": 6090
+    },
+    {
+      "epoch": 4.121621621621622,
+      "grad_norm": 0.32894161343574524,
+      "learning_rate": 8.790540540540541e-06,
+      "loss": 0.0372,
+      "step": 6100
+    },
+    {
+      "epoch": 4.128378378378378,
+      "grad_norm": 0.24975889921188354,
+      "learning_rate": 8.722972972972973e-06,
+      "loss": 0.0346,
+      "step": 6110
+    },
+    {
+      "epoch": 4.135135135135135,
+      "grad_norm": 0.36998119950294495,
+      "learning_rate": 8.655405405405405e-06,
+      "loss": 0.0354,
+      "step": 6120
+    },
+    {
+      "epoch": 4.141891891891892,
+      "grad_norm": 0.4094553589820862,
+      "learning_rate": 8.587837837837838e-06,
+      "loss": 0.0367,
+      "step": 6130
+    },
+    {
+      "epoch": 4.148648648648648,
+      "grad_norm": 0.2307623326778412,
+      "learning_rate": 8.520270270270272e-06,
+      "loss": 0.0341,
+      "step": 6140
+    },
+    {
+      "epoch": 4.155405405405405,
+      "grad_norm": 0.3480986952781677,
+      "learning_rate": 8.452702702702702e-06,
+      "loss": 0.0456,
+      "step": 6150
+    },
+    {
+      "epoch": 4.162162162162162,
+      "grad_norm": 0.22716574370861053,
+      "learning_rate": 8.385135135135136e-06,
+      "loss": 0.0391,
+      "step": 6160
+    },
+    {
+      "epoch": 4.168918918918919,
+      "grad_norm": 0.4268530607223511,
+      "learning_rate": 8.317567567567568e-06,
+      "loss": 0.0425,
+      "step": 6170
+    },
+    {
+      "epoch": 4.175675675675675,
+      "grad_norm": 0.29732945561408997,
+      "learning_rate": 8.25e-06,
+      "loss": 0.0363,
+      "step": 6180
+    },
+    {
+      "epoch": 4.1824324324324325,
+      "grad_norm": 0.49700286984443665,
+      "learning_rate": 8.182432432432433e-06,
+      "loss": 0.0432,
+      "step": 6190
+    },
+    {
+      "epoch": 4.1891891891891895,
+      "grad_norm": 0.20685577392578125,
+      "learning_rate": 8.114864864864865e-06,
+      "loss": 0.0288,
+      "step": 6200
+    },
+    {
+      "epoch": 4.195945945945946,
+      "grad_norm": 0.15559187531471252,
+      "learning_rate": 8.047297297297297e-06,
+      "loss": 0.0307,
+      "step": 6210
+    },
+    {
+      "epoch": 4.202702702702703,
+      "grad_norm": 0.3090328872203827,
+      "learning_rate": 7.97972972972973e-06,
+      "loss": 0.0371,
+      "step": 6220
+    },
+    {
+      "epoch": 4.20945945945946,
+      "grad_norm": 1.0721955299377441,
+      "learning_rate": 7.912162162162162e-06,
+      "loss": 0.0471,
+      "step": 6230
+    },
+    {
+      "epoch": 4.216216216216216,
+      "grad_norm": 0.24389462172985077,
+      "learning_rate": 7.844594594594596e-06,
+      "loss": 0.0362,
+      "step": 6240
+    },
+    {
+      "epoch": 4.222972972972973,
+      "grad_norm": 0.22328566014766693,
+      "learning_rate": 7.777027027027026e-06,
+      "loss": 0.0416,
+      "step": 6250
+    },
+    {
+      "epoch": 4.22972972972973,
+      "grad_norm": 0.17206421494483948,
+      "learning_rate": 7.70945945945946e-06,
+      "loss": 0.0267,
+      "step": 6260
+    },
+    {
+      "epoch": 4.236486486486487,
+      "grad_norm": 0.2622784674167633,
+      "learning_rate": 7.641891891891893e-06,
+      "loss": 0.0317,
+      "step": 6270
+    },
+    {
+      "epoch": 4.243243243243243,
+      "grad_norm": 0.561418354511261,
+      "learning_rate": 7.574324324324325e-06,
+      "loss": 0.0487,
+      "step": 6280
+    },
+    {
+      "epoch": 4.25,
+      "grad_norm": 0.39344513416290283,
+      "learning_rate": 7.506756756756757e-06,
+      "loss": 0.0491,
+      "step": 6290
+    },
+    {
+      "epoch": 4.256756756756757,
+      "grad_norm": 0.24725763499736786,
+      "learning_rate": 7.43918918918919e-06,
+      "loss": 0.0302,
+      "step": 6300
+    },
+    {
+      "epoch": 4.263513513513513,
+      "grad_norm": 0.23941102623939514,
+      "learning_rate": 7.371621621621622e-06,
+      "loss": 0.0282,
+      "step": 6310
+    },
+    {
+      "epoch": 4.27027027027027,
+      "grad_norm": 0.2501400113105774,
+      "learning_rate": 7.304054054054055e-06,
+      "loss": 0.0366,
+      "step": 6320
+    },
+    {
+      "epoch": 4.277027027027027,
+      "grad_norm": 0.3194361627101898,
+      "learning_rate": 7.236486486486486e-06,
+      "loss": 0.0336,
+      "step": 6330
+    },
+    {
+      "epoch": 4.283783783783784,
+      "grad_norm": 0.5751128792762756,
+      "learning_rate": 7.168918918918919e-06,
+      "loss": 0.043,
+      "step": 6340
+    },
+    {
+      "epoch": 4.29054054054054,
+      "grad_norm": 0.5228712558746338,
+      "learning_rate": 7.1013513513513515e-06,
+      "loss": 0.0327,
+      "step": 6350
+    },
+    {
+      "epoch": 4.297297297297297,
+      "grad_norm": 0.2010771632194519,
+      "learning_rate": 7.033783783783785e-06,
+      "loss": 0.0314,
+      "step": 6360
+    },
+    {
+      "epoch": 4.304054054054054,
+      "grad_norm": 0.2496619075536728,
+      "learning_rate": 6.966216216216216e-06,
+      "loss": 0.0305,
+      "step": 6370
+    },
+    {
+      "epoch": 4.3108108108108105,
+      "grad_norm": 0.21699196100234985,
+      "learning_rate": 6.898648648648649e-06,
+      "loss": 0.0306,
+      "step": 6380
+    },
+    {
+      "epoch": 4.3175675675675675,
+      "grad_norm": 0.30333423614501953,
+      "learning_rate": 6.831081081081081e-06,
+      "loss": 0.0343,
+      "step": 6390
+    },
+    {
+      "epoch": 4.324324324324325,
+      "grad_norm": 0.39921465516090393,
+      "learning_rate": 6.7635135135135145e-06,
+      "loss": 0.0339,
+      "step": 6400
+    },
+    {
+      "epoch": 4.331081081081081,
+      "grad_norm": 0.5513906478881836,
+      "learning_rate": 6.695945945945946e-06,
+      "loss": 0.0311,
+      "step": 6410
+    },
+    {
+      "epoch": 4.337837837837838,
+      "grad_norm": 0.11431396752595901,
+      "learning_rate": 6.628378378378379e-06,
+      "loss": 0.0284,
+      "step": 6420
+    },
+    {
+      "epoch": 4.344594594594595,
+      "grad_norm": 0.2621834874153137,
+      "learning_rate": 6.56081081081081e-06,
+      "loss": 0.0299,
+      "step": 6430
+    },
+    {
+      "epoch": 4.351351351351352,
+      "grad_norm": 0.6052500605583191,
+      "learning_rate": 6.4932432432432435e-06,
+      "loss": 0.0357,
+      "step": 6440
+    },
+    {
+      "epoch": 4.358108108108108,
+      "grad_norm": 0.19118449091911316,
+      "learning_rate": 6.425675675675676e-06,
+      "loss": 0.0331,
+      "step": 6450
+    },
+    {
+      "epoch": 4.364864864864865,
+      "grad_norm": 0.1641031801700592,
+      "learning_rate": 6.358108108108109e-06,
+      "loss": 0.0409,
+      "step": 6460
+    },
+    {
+      "epoch": 4.371621621621622,
+      "grad_norm": 0.30924248695373535,
+      "learning_rate": 6.29054054054054e-06,
+      "loss": 0.0381,
+      "step": 6470
+    },
+    {
+      "epoch": 4.378378378378378,
+      "grad_norm": 0.33302533626556396,
+      "learning_rate": 6.222972972972973e-06,
+      "loss": 0.0425,
+      "step": 6480
+    },
+    {
+      "epoch": 4.385135135135135,
+      "grad_norm": 0.2112291157245636,
+      "learning_rate": 6.155405405405406e-06,
+      "loss": 0.035,
+      "step": 6490
+    },
+    {
+      "epoch": 4.391891891891892,
+      "grad_norm": 0.5250951051712036,
+      "learning_rate": 6.087837837837839e-06,
+      "loss": 0.0307,
+      "step": 6500
+    },
+    {
+      "epoch": 4.398648648648648,
+      "grad_norm": 0.30093303322792053,
+      "learning_rate": 6.020270270270271e-06,
+      "loss": 0.0362,
+      "step": 6510
+    },
+    {
+      "epoch": 4.405405405405405,
+      "grad_norm": 0.18295438587665558,
+      "learning_rate": 5.952702702702703e-06,
+      "loss": 0.0395,
+      "step": 6520
+    },
+    {
+      "epoch": 4.412162162162162,
+      "grad_norm": 0.2848648130893707,
+      "learning_rate": 5.8851351351351355e-06,
+      "loss": 0.027,
+      "step": 6530
+    },
+    {
+      "epoch": 4.418918918918919,
+      "grad_norm": 0.5116801261901855,
+      "learning_rate": 5.817567567567568e-06,
+      "loss": 0.0349,
+      "step": 6540
+    },
+    {
+      "epoch": 4.425675675675675,
+      "grad_norm": 0.36234357953071594,
+      "learning_rate": 5.750000000000001e-06,
+      "loss": 0.0341,
+      "step": 6550
+    },
+    {
+      "epoch": 4.4324324324324325,
+      "grad_norm": 0.29395073652267456,
+      "learning_rate": 5.682432432432433e-06,
+      "loss": 0.0398,
+      "step": 6560
+    },
+    {
+      "epoch": 4.4391891891891895,
+      "grad_norm": 0.2595879137516022,
+      "learning_rate": 5.614864864864865e-06,
+      "loss": 0.0374,
+      "step": 6570
+    },
+    {
+      "epoch": 4.445945945945946,
+      "grad_norm": 0.18848879635334015,
+      "learning_rate": 5.547297297297298e-06,
+      "loss": 0.0286,
+      "step": 6580
+    },
+    {
+      "epoch": 4.452702702702703,
+      "grad_norm": 0.34463363885879517,
+      "learning_rate": 5.47972972972973e-06,
+      "loss": 0.0275,
+      "step": 6590
+    },
+    {
+      "epoch": 4.45945945945946,
+      "grad_norm": 0.7043229937553406,
+      "learning_rate": 5.412162162162163e-06,
+      "loss": 0.0393,
+      "step": 6600
+    },
+    {
+      "epoch": 4.466216216216216,
+      "grad_norm": 0.33575713634490967,
+      "learning_rate": 5.344594594594595e-06,
+      "loss": 0.0401,
+      "step": 6610
+    },
+    {
+      "epoch": 4.472972972972973,
+      "grad_norm": 0.2630554437637329,
+      "learning_rate": 5.2770270270270275e-06,
+      "loss": 0.0293,
+      "step": 6620
+    },
+    {
+      "epoch": 4.47972972972973,
+      "grad_norm": 0.20551495254039764,
+      "learning_rate": 5.20945945945946e-06,
+      "loss": 0.0269,
+      "step": 6630
+    },
+    {
+      "epoch": 4.486486486486487,
+      "grad_norm": 0.13341334462165833,
+      "learning_rate": 5.141891891891892e-06,
+      "loss": 0.0398,
+      "step": 6640
+    },
+    {
+      "epoch": 4.493243243243243,
+      "grad_norm": 0.2936372458934784,
+      "learning_rate": 5.074324324324325e-06,
+      "loss": 0.0265,
+      "step": 6650
+    },
+    {
+      "epoch": 4.5,
+      "grad_norm": 0.2781808376312256,
+      "learning_rate": 5.006756756756757e-06,
+      "loss": 0.0456,
+      "step": 6660
+    },
+    {
+      "epoch": 4.506756756756757,
+      "grad_norm": 0.17368100583553314,
+      "learning_rate": 4.93918918918919e-06,
+      "loss": 0.0369,
+      "step": 6670
+    },
+    {
+      "epoch": 4.513513513513513,
+      "grad_norm": 0.31108617782592773,
+      "learning_rate": 4.871621621621622e-06,
+      "loss": 0.0527,
+      "step": 6680
+    },
+    {
+      "epoch": 4.52027027027027,
+      "grad_norm": 0.13732381165027618,
+      "learning_rate": 4.804054054054054e-06,
+      "loss": 0.0273,
+      "step": 6690
+    },
+    {
+      "epoch": 4.527027027027027,
+      "grad_norm": 0.32313647866249084,
+      "learning_rate": 4.736486486486487e-06,
+      "loss": 0.0276,
+      "step": 6700
+    },
+    {
+      "epoch": 4.533783783783784,
+      "grad_norm": 0.23118528723716736,
+      "learning_rate": 4.6689189189189195e-06,
+      "loss": 0.0314,
+      "step": 6710
+    },
+    {
+      "epoch": 4.54054054054054,
+      "grad_norm": 0.1537824273109436,
+      "learning_rate": 4.601351351351352e-06,
+      "loss": 0.0313,
+      "step": 6720
+    },
+    {
+      "epoch": 4.547297297297297,
+      "grad_norm": 0.21601910889148712,
+      "learning_rate": 4.533783783783784e-06,
+      "loss": 0.0311,
+      "step": 6730
+    },
+    {
+      "epoch": 4.554054054054054,
+      "grad_norm": 0.2066461145877838,
+      "learning_rate": 4.466216216216216e-06,
+      "loss": 0.0377,
+      "step": 6740
+    },
+    {
+      "epoch": 4.5608108108108105,
+      "grad_norm": 10.0748929977417,
+      "learning_rate": 4.398648648648649e-06,
+      "loss": 0.0614,
+      "step": 6750
+    },
+    {
+      "epoch": 4.5675675675675675,
+      "grad_norm": 0.4590441584587097,
+      "learning_rate": 4.331081081081082e-06,
+      "loss": 0.0416,
+      "step": 6760
+    },
+    {
+      "epoch": 4.574324324324325,
+      "grad_norm": 0.2324705421924591,
+      "learning_rate": 4.263513513513514e-06,
+      "loss": 0.0352,
+      "step": 6770
+    },
+    {
+      "epoch": 4.581081081081081,
+      "grad_norm": 0.16836494207382202,
+      "learning_rate": 4.195945945945946e-06,
+      "loss": 0.0269,
+      "step": 6780
+    },
+    {
+      "epoch": 4.587837837837838,
+      "grad_norm": 0.12347856163978577,
+      "learning_rate": 4.128378378378378e-06,
+      "loss": 0.0306,
+      "step": 6790
+    },
+    {
+      "epoch": 4.594594594594595,
+      "grad_norm": 0.3169582486152649,
+      "learning_rate": 4.0608108108108115e-06,
+      "loss": 0.0412,
+      "step": 6800
+    },
+    {
+      "epoch": 4.601351351351351,
+      "grad_norm": 0.42407071590423584,
+      "learning_rate": 3.993243243243244e-06,
+      "loss": 0.0386,
+      "step": 6810
+    },
+    {
+      "epoch": 4.608108108108108,
+      "grad_norm": 0.3871265649795532,
+      "learning_rate": 3.925675675675676e-06,
+      "loss": 0.0337,
+      "step": 6820
+    },
+    {
+      "epoch": 4.614864864864865,
+      "grad_norm": 0.38576745986938477,
+      "learning_rate": 3.858108108108108e-06,
+      "loss": 0.039,
+      "step": 6830
+    },
+    {
+      "epoch": 4.621621621621622,
+      "grad_norm": 0.23177988827228546,
+      "learning_rate": 3.790540540540541e-06,
+      "loss": 0.0324,
+      "step": 6840
+    },
+    {
+      "epoch": 4.628378378378378,
+      "grad_norm": 0.22654855251312256,
+      "learning_rate": 3.722972972972973e-06,
+      "loss": 0.0583,
+      "step": 6850
+    },
+    {
+      "epoch": 4.635135135135135,
+      "grad_norm": 0.4358038008213043,
+      "learning_rate": 3.655405405405406e-06,
+      "loss": 0.0376,
+      "step": 6860
+    },
+    {
+      "epoch": 4.641891891891892,
+      "grad_norm": 0.11105912178754807,
+      "learning_rate": 3.587837837837838e-06,
+      "loss": 0.0294,
+      "step": 6870
+    },
+    {
+      "epoch": 4.648648648648649,
+      "grad_norm": 0.36878806352615356,
+      "learning_rate": 3.5202702702702704e-06,
+      "loss": 0.0384,
+      "step": 6880
+    },
+    {
+      "epoch": 4.655405405405405,
+      "grad_norm": 0.1898292452096939,
+      "learning_rate": 3.452702702702703e-06,
+      "loss": 0.0299,
+      "step": 6890
+    },
+    {
+      "epoch": 4.662162162162162,
+      "grad_norm": 0.21792760491371155,
+      "learning_rate": 3.3851351351351353e-06,
+      "loss": 0.0396,
+      "step": 6900
+    },
+    {
+      "epoch": 4.668918918918919,
+      "grad_norm": 0.5358079075813293,
+      "learning_rate": 3.317567567567568e-06,
+      "loss": 0.0329,
+      "step": 6910
+    },
+    {
+      "epoch": 4.675675675675675,
+      "grad_norm": 0.19601544737815857,
+      "learning_rate": 3.2500000000000002e-06,
+      "loss": 0.0434,
+      "step": 6920
+    },
+    {
+      "epoch": 4.6824324324324325,
+      "grad_norm": 0.24563315510749817,
+      "learning_rate": 3.182432432432433e-06,
+      "loss": 0.0292,
+      "step": 6930
+    },
+    {
+      "epoch": 4.6891891891891895,
+      "grad_norm": 0.3209024667739868,
+      "learning_rate": 3.114864864864865e-06,
+      "loss": 0.0529,
+      "step": 6940
+    },
+    {
+      "epoch": 4.695945945945946,
+      "grad_norm": 0.18662333488464355,
+      "learning_rate": 3.0472972972972974e-06,
+      "loss": 0.0305,
+      "step": 6950
+    },
+    {
+      "epoch": 4.702702702702703,
+      "grad_norm": 0.18789201974868774,
+      "learning_rate": 2.97972972972973e-06,
+      "loss": 0.0296,
+      "step": 6960
+    },
+    {
+      "epoch": 4.70945945945946,
+      "grad_norm": 0.18934795260429382,
+      "learning_rate": 2.9121621621621623e-06,
+      "loss": 0.0368,
+      "step": 6970
+    },
+    {
+      "epoch": 4.716216216216216,
+      "grad_norm": 0.2662386894226074,
+      "learning_rate": 2.844594594594595e-06,
+      "loss": 0.0335,
+      "step": 6980
+    },
+    {
+      "epoch": 4.722972972972973,
+      "grad_norm": 0.4388953447341919,
+      "learning_rate": 2.7770270270270273e-06,
+      "loss": 0.0316,
+      "step": 6990
+    },
+    {
+      "epoch": 4.72972972972973,
+      "grad_norm": 0.26200464367866516,
+      "learning_rate": 2.7094594594594595e-06,
+      "loss": 0.031,
+      "step": 7000
+    },
+    {
+      "epoch": 4.736486486486487,
+      "grad_norm": 0.27926555275917053,
+      "learning_rate": 2.6418918918918922e-06,
+      "loss": 0.0387,
+      "step": 7010
+    },
+    {
+      "epoch": 4.743243243243243,
+      "grad_norm": 0.28896600008010864,
+      "learning_rate": 2.5743243243243245e-06,
+      "loss": 0.0347,
+      "step": 7020
+    },
+    {
+      "epoch": 4.75,
+      "grad_norm": 0.27116715908050537,
+      "learning_rate": 2.506756756756757e-06,
+      "loss": 0.0307,
+      "step": 7030
+    },
+    {
+      "epoch": 4.756756756756757,
+      "grad_norm": 0.29698431491851807,
+      "learning_rate": 2.4391891891891894e-06,
+      "loss": 0.0331,
+      "step": 7040
+    },
+    {
+      "epoch": 4.763513513513513,
+      "grad_norm": 0.42871206998825073,
+      "learning_rate": 2.3716216216216217e-06,
+      "loss": 0.0337,
+      "step": 7050
+    },
+    {
+      "epoch": 4.77027027027027,
+      "grad_norm": 0.45836636424064636,
+      "learning_rate": 2.3040540540540543e-06,
+      "loss": 0.0314,
+      "step": 7060
+    },
+    {
+      "epoch": 4.777027027027027,
+      "grad_norm": 0.31179630756378174,
+      "learning_rate": 2.2364864864864866e-06,
+      "loss": 0.0404,
+      "step": 7070
+    },
+    {
+      "epoch": 4.783783783783784,
+      "grad_norm": 0.261393278837204,
+      "learning_rate": 2.1689189189189193e-06,
+      "loss": 0.0371,
+      "step": 7080
+    },
+    {
+      "epoch": 4.79054054054054,
+      "grad_norm": 0.5424347519874573,
+      "learning_rate": 2.1013513513513515e-06,
+      "loss": 0.0324,
+      "step": 7090
+    },
+    {
+      "epoch": 4.797297297297297,
+      "grad_norm": 0.44092580676078796,
+      "learning_rate": 2.0337837837837838e-06,
+      "loss": 0.0371,
+      "step": 7100
+    },
+    {
+      "epoch": 4.804054054054054,
+      "grad_norm": 0.2997783422470093,
+      "learning_rate": 1.9662162162162165e-06,
+      "loss": 0.038,
+      "step": 7110
+    },
+    {
+      "epoch": 4.8108108108108105,
+      "grad_norm": 0.270183801651001,
+      "learning_rate": 1.8986486486486487e-06,
+      "loss": 0.0353,
+      "step": 7120
+    },
+    {
+      "epoch": 4.8175675675675675,
+      "grad_norm": 0.5547716021537781,
+      "learning_rate": 1.8310810810810812e-06,
+      "loss": 0.0388,
+      "step": 7130
+    },
+    {
+      "epoch": 4.824324324324325,
+      "grad_norm": 0.3656560182571411,
+      "learning_rate": 1.7635135135135136e-06,
+      "loss": 0.0338,
+      "step": 7140
+    },
+    {
+      "epoch": 4.831081081081081,
+      "grad_norm": 0.17846781015396118,
+      "learning_rate": 1.6959459459459461e-06,
+      "loss": 0.0394,
+      "step": 7150
+    },
+    {
+      "epoch": 4.837837837837838,
+      "grad_norm": 0.4710524380207062,
+      "learning_rate": 1.6283783783783784e-06,
+      "loss": 0.0315,
+      "step": 7160
+    },
+    {
+      "epoch": 4.844594594594595,
+      "grad_norm": 0.3607494831085205,
+      "learning_rate": 1.5608108108108108e-06,
+      "loss": 0.0313,
+      "step": 7170
+    },
+    {
+      "epoch": 4.851351351351351,
+      "grad_norm": 0.3811328709125519,
+      "learning_rate": 1.4932432432432433e-06,
+      "loss": 0.0383,
+      "step": 7180
+    },
+    {
+      "epoch": 4.858108108108108,
+      "grad_norm": 0.23322060704231262,
+      "learning_rate": 1.4256756756756758e-06,
+      "loss": 0.0339,
+      "step": 7190
+    },
+    {
+      "epoch": 4.864864864864865,
+      "grad_norm": 0.2853488028049469,
+      "learning_rate": 1.3581081081081082e-06,
+      "loss": 0.0327,
+      "step": 7200
+    },
+    {
+      "epoch": 4.871621621621622,
+      "grad_norm": 0.3856007158756256,
+      "learning_rate": 1.2905405405405407e-06,
+      "loss": 0.0328,
+      "step": 7210
+    },
+    {
+      "epoch": 4.878378378378378,
+      "grad_norm": 0.3511975109577179,
+      "learning_rate": 1.222972972972973e-06,
+      "loss": 0.0525,
+      "step": 7220
+    },
+    {
+      "epoch": 4.885135135135135,
+      "grad_norm": 0.3545476198196411,
+      "learning_rate": 1.1554054054054054e-06,
+      "loss": 0.0379,
+      "step": 7230
+    },
+    {
+      "epoch": 4.891891891891892,
+      "grad_norm": 0.1747957319021225,
+      "learning_rate": 1.087837837837838e-06,
+      "loss": 0.0246,
+      "step": 7240
+    },
+    {
+      "epoch": 4.898648648648649,
+      "grad_norm": 0.3207055926322937,
+      "learning_rate": 1.0202702702702704e-06,
+      "loss": 0.03,
+      "step": 7250
+    },
+    {
+      "epoch": 4.905405405405405,
+      "grad_norm": 0.1513839215040207,
+      "learning_rate": 9.527027027027027e-07,
+      "loss": 0.0366,
+      "step": 7260
+    },
+    {
+      "epoch": 4.912162162162162,
+      "grad_norm": 0.5031633377075195,
+      "learning_rate": 8.851351351351352e-07,
+      "loss": 0.0289,
+      "step": 7270
+    },
+    {
+      "epoch": 4.918918918918919,
+      "grad_norm": 0.27607399225234985,
+      "learning_rate": 8.175675675675676e-07,
+      "loss": 0.0387,
+      "step": 7280
+    },
+    {
+      "epoch": 4.925675675675675,
+      "grad_norm": 0.2901621460914612,
+      "learning_rate": 7.5e-07,
+      "loss": 0.039,
+      "step": 7290
+    },
+    {
+      "epoch": 4.9324324324324325,
+      "grad_norm": 0.22750946879386902,
+      "learning_rate": 6.824324324324325e-07,
+      "loss": 0.0443,
+      "step": 7300
+    },
+    {
+      "epoch": 4.9391891891891895,
+      "grad_norm": 0.3207789361476898,
+      "learning_rate": 6.148648648648648e-07,
+      "loss": 0.0367,
+      "step": 7310
+    },
+    {
+      "epoch": 4.945945945945946,
+      "grad_norm": 0.28913548588752747,
+      "learning_rate": 5.472972972972973e-07,
+      "loss": 0.0281,
+      "step": 7320
+    },
+    {
+      "epoch": 4.952702702702703,
+      "grad_norm": 0.19479107856750488,
+      "learning_rate": 4.797297297297297e-07,
+      "loss": 0.0303,
+      "step": 7330
+    },
+    {
+      "epoch": 4.95945945945946,
+      "grad_norm": 0.4693133234977722,
+      "learning_rate": 4.1216216216216225e-07,
+      "loss": 0.035,
+      "step": 7340
+    },
+    {
+      "epoch": 4.966216216216216,
+      "grad_norm": 0.1580437868833542,
+      "learning_rate": 3.445945945945946e-07,
+      "loss": 0.0396,
+      "step": 7350
+    },
+    {
+      "epoch": 4.972972972972973,
+      "grad_norm": 0.2763003706932068,
+      "learning_rate": 2.77027027027027e-07,
+      "loss": 0.0332,
+      "step": 7360
+    },
+    {
+      "epoch": 4.97972972972973,
+      "grad_norm": 0.26670750975608826,
+      "learning_rate": 2.0945945945945946e-07,
+      "loss": 0.0388,
+      "step": 7370
+    },
+    {
+      "epoch": 4.986486486486487,
+      "grad_norm": 0.7256395816802979,
+      "learning_rate": 1.418918918918919e-07,
+      "loss": 0.0325,
+      "step": 7380
+    },
+    {
+      "epoch": 4.993243243243243,
+      "grad_norm": 0.47905421257019043,
+      "learning_rate": 7.432432432432433e-08,
+      "loss": 0.0383,
+      "step": 7390
+    },
+    {
+      "epoch": 5.0,
+      "grad_norm": 0.29960232973098755,
+      "learning_rate": 6.756756756756757e-09,
+      "loss": 0.0319,
+      "step": 7400
+    },
+    {
+      "epoch": 5.0,
+      "eval_loss": 0.02321171946823597,
+      "eval_runtime": 241.491,
+      "eval_samples_per_second": 4.327,
+      "eval_steps_per_second": 1.085,
+      "step": 7400
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 7400,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 5,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1235384799068160.0,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}