diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,7738 @@
+{
+  "best_metric": 0.44340217113494873,
+  "best_model_checkpoint": "saved_model/lop_jun2024/checkpoint-10982",
+  "epoch": 2.999931707983337,
+  "eval_steps": 500,
+  "global_step": 10982,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": NaN,
+      "learning_rate": 0.0,
+      "loss": 74.2157,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 13.99621295928955,
+      "learning_rate": 2.5e-06,
+      "loss": 74.5237,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 14.391114234924316,
+      "learning_rate": 6.5000000000000004e-06,
+      "loss": 74.2338,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 15.539421081542969,
+      "learning_rate": 1.1500000000000002e-05,
+      "loss": 72.8847,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 16.487953186035156,
+      "learning_rate": 1.65e-05,
+      "loss": 70.7699,
+      "step": 40
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 21.840412139892578,
+      "learning_rate": 2.15e-05,
+      "loss": 66.4777,
+      "step": 50
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 23.74176025390625,
+      "learning_rate": 2.6000000000000002e-05,
+      "loss": 59.438,
+      "step": 60
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 26.7894287109375,
+      "learning_rate": 3.1e-05,
+      "loss": 45.7214,
+      "step": 70
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 21.767683029174805,
+      "learning_rate": 3.55e-05,
+      "loss": 27.7672,
+      "step": 80
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 15.309523582458496,
+      "learning_rate": 4.05e-05,
+      "loss": 12.9366,
+      "step": 90
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 7.069633960723877,
+      "learning_rate": 4.55e-05,
+      "loss": 4.342,
+      "step": 100
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 10.602043151855469,
+      "learning_rate": 5.05e-05,
+      "loss": 1.7191,
+      "step": 110
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 8.00312614440918,
+      "learning_rate": 5.550000000000001e-05,
+      "loss": 1.3934,
+      "step": 120
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 7.026750087738037,
+      "learning_rate": 6.05e-05,
+      "loss": 1.3728,
+      "step": 130
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 10.686494827270508,
+      "learning_rate": 6.55e-05,
+      "loss": 1.2595,
+      "step": 140
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 7.127823352813721,
+      "learning_rate": 7.05e-05,
+      "loss": 1.2226,
+      "step": 150
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.253384590148926,
+      "learning_rate": 7.55e-05,
+      "loss": 1.2934,
+      "step": 160
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 7.56700325012207,
+      "learning_rate": 8.05e-05,
+      "loss": 1.2571,
+      "step": 170
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 9.671475410461426,
+      "learning_rate": 8.55e-05,
+      "loss": 1.2005,
+      "step": 180
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 7.772515296936035,
+      "learning_rate": 9.05e-05,
+      "loss": 1.253,
+      "step": 190
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.593009948730469,
+      "learning_rate": 9.55e-05,
+      "loss": 1.2269,
+      "step": 200
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 9.796707153320312,
+      "learning_rate": 9.999725274725276e-05,
+      "loss": 1.2425,
+      "step": 210
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.557741641998291,
+      "learning_rate": 9.996978021978023e-05,
+      "loss": 1.2383,
+      "step": 220
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 7.321559906005859,
+      "learning_rate": 9.994230769230771e-05,
+      "loss": 1.2026,
+      "step": 230
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.337256908416748,
+      "learning_rate": 9.991483516483518e-05,
+      "loss": 1.1289,
+      "step": 240
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.283127307891846,
+      "learning_rate": 9.988736263736264e-05,
+      "loss": 1.0357,
+      "step": 250
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.2405242919921875,
+      "learning_rate": 9.985989010989013e-05,
+      "loss": 1.0067,
+      "step": 260
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 6.622239112854004,
+      "learning_rate": 9.98324175824176e-05,
+      "loss": 0.9816,
+      "step": 270
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.521809101104736,
+      "learning_rate": 9.980494505494506e-05,
+      "loss": 0.9578,
+      "step": 280
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.7347898483276367,
+      "learning_rate": 9.977747252747254e-05,
+      "loss": 0.8762,
+      "step": 290
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.452709674835205,
+      "learning_rate": 9.975000000000001e-05,
+      "loss": 0.8188,
+      "step": 300
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.2089035511016846,
+      "learning_rate": 9.972252747252748e-05,
+      "loss": 0.8247,
+      "step": 310
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.455322742462158,
+      "learning_rate": 9.969505494505496e-05,
+      "loss": 0.7599,
+      "step": 320
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.537048816680908,
+      "learning_rate": 9.966758241758242e-05,
+      "loss": 0.7968,
+      "step": 330
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.754340648651123,
+      "learning_rate": 9.964010989010988e-05,
+      "loss": 0.72,
+      "step": 340
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.538580894470215,
+      "learning_rate": 9.961263736263737e-05,
+      "loss": 0.7207,
+      "step": 350
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.3613457679748535,
+      "learning_rate": 9.958516483516483e-05,
+      "loss": 0.7235,
+      "step": 360
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.2669098377227783,
+      "learning_rate": 9.95576923076923e-05,
+      "loss": 0.7018,
+      "step": 370
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.329736232757568,
+      "learning_rate": 9.953021978021978e-05,
+      "loss": 0.7193,
+      "step": 380
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.5049781799316406,
+      "learning_rate": 9.950274725274725e-05,
+      "loss": 0.6948,
+      "step": 390
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.5907931327819824,
+      "learning_rate": 9.947527472527472e-05,
+      "loss": 0.6637,
+      "step": 400
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.264726161956787,
+      "learning_rate": 9.94478021978022e-05,
+      "loss": 0.6588,
+      "step": 410
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.763380765914917,
+      "learning_rate": 9.942032967032967e-05,
+      "loss": 0.6479,
+      "step": 420
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.9857630729675293,
+      "learning_rate": 9.939285714285714e-05,
+      "loss": 0.6481,
+      "step": 430
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.649977922439575,
+      "learning_rate": 9.936538461538462e-05,
+      "loss": 0.6284,
+      "step": 440
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.4232739210128784,
+      "learning_rate": 9.933791208791209e-05,
+      "loss": 0.6214,
+      "step": 450
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.327674865722656,
+      "learning_rate": 9.931043956043956e-05,
+      "loss": 0.633,
+      "step": 460
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.673241376876831,
+      "learning_rate": 9.928296703296704e-05,
+      "loss": 0.6792,
+      "step": 470
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.6698083877563477,
+      "learning_rate": 9.92554945054945e-05,
+      "loss": 0.6297,
+      "step": 480
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.056175470352173,
+      "learning_rate": 9.922802197802197e-05,
+      "loss": 0.6172,
+      "step": 490
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.345395803451538,
+      "learning_rate": 9.920054945054946e-05,
+      "loss": 0.605,
+      "step": 500
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.7540231943130493,
+      "learning_rate": 9.917307692307692e-05,
+      "loss": 0.6288,
+      "step": 510
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.050562858581543,
+      "learning_rate": 9.914560439560439e-05,
+      "loss": 0.6281,
+      "step": 520
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.210700511932373,
+      "learning_rate": 9.911813186813187e-05,
+      "loss": 0.6138,
+      "step": 530
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.6927013397216797,
+      "learning_rate": 9.909065934065934e-05,
+      "loss": 0.6033,
+      "step": 540
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.4765613079071045,
+      "learning_rate": 9.906318681318681e-05,
+      "loss": 0.5991,
+      "step": 550
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.8127527236938477,
+      "learning_rate": 9.903571428571429e-05,
+      "loss": 0.5994,
+      "step": 560
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.6926054954528809,
+      "learning_rate": 9.900824175824176e-05,
+      "loss": 0.5988,
+      "step": 570
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.3790557384490967,
+      "learning_rate": 9.898076923076923e-05,
+      "loss": 0.6147,
+      "step": 580
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.558159589767456,
+      "learning_rate": 9.895329670329671e-05,
+      "loss": 0.5858,
+      "step": 590
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.5203194618225098,
+      "learning_rate": 9.892582417582418e-05,
+      "loss": 0.6036,
+      "step": 600
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.4314634799957275,
+      "learning_rate": 9.889835164835165e-05,
+      "loss": 0.5789,
+      "step": 610
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.956555128097534,
+      "learning_rate": 9.887087912087913e-05,
+      "loss": 0.6021,
+      "step": 620
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.231612205505371,
+      "learning_rate": 9.88434065934066e-05,
+      "loss": 0.5825,
+      "step": 630
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.1588950157165527,
+      "learning_rate": 9.881593406593406e-05,
+      "loss": 0.5924,
+      "step": 640
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.8479809761047363,
+      "learning_rate": 9.878846153846155e-05,
+      "loss": 0.584,
+      "step": 650
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.2029404640197754,
+      "learning_rate": 9.876098901098901e-05,
+      "loss": 0.5855,
+      "step": 660
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.243802070617676,
+      "learning_rate": 9.873351648351648e-05,
+      "loss": 0.584,
+      "step": 670
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 2.8837687969207764,
+      "learning_rate": 9.870604395604396e-05,
+      "loss": 0.5867,
+      "step": 680
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 2.0768215656280518,
+      "learning_rate": 9.867857142857143e-05,
+      "loss": 0.5823,
+      "step": 690
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 2.950639247894287,
+      "learning_rate": 9.86510989010989e-05,
+      "loss": 0.5832,
+      "step": 700
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.0848056077957153,
+      "learning_rate": 9.862362637362638e-05,
+      "loss": 0.5595,
+      "step": 710
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.293261170387268,
+      "learning_rate": 9.859615384615385e-05,
+      "loss": 0.5926,
+      "step": 720
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 2.286271095275879,
+      "learning_rate": 9.856868131868132e-05,
+      "loss": 0.5658,
+      "step": 730
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 2.9983866214752197,
+      "learning_rate": 9.85412087912088e-05,
+      "loss": 0.5749,
+      "step": 740
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 2.712332248687744,
+      "learning_rate": 9.851373626373627e-05,
+      "loss": 0.5757,
+      "step": 750
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.5591819286346436,
+      "learning_rate": 9.848626373626374e-05,
+      "loss": 0.5706,
+      "step": 760
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.00175142288208,
+      "learning_rate": 9.845879120879122e-05,
+      "loss": 0.5654,
+      "step": 770
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.5290532112121582,
+      "learning_rate": 9.843131868131869e-05,
+      "loss": 0.578,
+      "step": 780
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.8737398386001587,
+      "learning_rate": 9.840384615384615e-05,
+      "loss": 0.5598,
+      "step": 790
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.190427541732788,
+      "learning_rate": 9.837637362637364e-05,
+      "loss": 0.5507,
+      "step": 800
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 2.8329577445983887,
+      "learning_rate": 9.83489010989011e-05,
+      "loss": 0.5394,
+      "step": 810
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 2.6253738403320312,
+      "learning_rate": 9.832142857142857e-05,
+      "loss": 0.5404,
+      "step": 820
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 2.1394851207733154,
+      "learning_rate": 9.829395604395605e-05,
+      "loss": 0.5616,
+      "step": 830
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.5708709955215454,
+      "learning_rate": 9.826648351648352e-05,
+      "loss": 0.5606,
+      "step": 840
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.521190881729126,
+      "learning_rate": 9.823901098901099e-05,
+      "loss": 0.5605,
+      "step": 850
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.7570327520370483,
+      "learning_rate": 9.821153846153847e-05,
+      "loss": 0.5475,
+      "step": 860
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 2.039400577545166,
+      "learning_rate": 9.818406593406594e-05,
+      "loss": 0.5451,
+      "step": 870
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.9855557680130005,
+      "learning_rate": 9.815659340659341e-05,
+      "loss": 0.5378,
+      "step": 880
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.3518325090408325,
+      "learning_rate": 9.812912087912089e-05,
+      "loss": 0.5447,
+      "step": 890
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 2.36753511428833,
+      "learning_rate": 9.810164835164836e-05,
+      "loss": 0.5593,
+      "step": 900
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.611568570137024,
+      "learning_rate": 9.807417582417583e-05,
+      "loss": 0.536,
+      "step": 910
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.3011342287063599,
+      "learning_rate": 9.804670329670331e-05,
+      "loss": 0.5508,
+      "step": 920
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.6673119068145752,
+      "learning_rate": 9.801923076923078e-05,
+      "loss": 0.5427,
+      "step": 930
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.4966050386428833,
+      "learning_rate": 9.799175824175824e-05,
+      "loss": 0.5364,
+      "step": 940
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.2000831365585327,
+      "learning_rate": 9.796428571428573e-05,
+      "loss": 0.5452,
+      "step": 950
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.9247843027114868,
+      "learning_rate": 9.79368131868132e-05,
+      "loss": 0.529,
+      "step": 960
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.6240330934524536,
+      "learning_rate": 9.790934065934066e-05,
+      "loss": 0.5523,
+      "step": 970
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.4320908784866333,
+      "learning_rate": 9.788186813186814e-05,
+      "loss": 0.5567,
+      "step": 980
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.7884244918823242,
+      "learning_rate": 9.785439560439561e-05,
+      "loss": 0.5244,
+      "step": 990
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.9070675373077393,
+      "learning_rate": 9.782692307692308e-05,
+      "loss": 0.5354,
+      "step": 1000
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.0304174423217773,
+      "learning_rate": 9.779945054945056e-05,
+      "loss": 0.5443,
+      "step": 1010
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 2.184297800064087,
+      "learning_rate": 9.777197802197803e-05,
+      "loss": 0.5201,
+      "step": 1020
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.2110681533813477,
+      "learning_rate": 9.77445054945055e-05,
+      "loss": 0.5363,
+      "step": 1030
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 2.25685977935791,
+      "learning_rate": 9.771703296703298e-05,
+      "loss": 0.5392,
+      "step": 1040
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 2.558457851409912,
+      "learning_rate": 9.768956043956045e-05,
+      "loss": 0.5466,
+      "step": 1050
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.5898990631103516,
+      "learning_rate": 9.766208791208792e-05,
+      "loss": 0.5386,
+      "step": 1060
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 2.6933228969573975,
+      "learning_rate": 9.76346153846154e-05,
+      "loss": 0.5565,
+      "step": 1070
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.440042495727539,
+      "learning_rate": 9.760714285714287e-05,
+      "loss": 0.5507,
+      "step": 1080
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.9476529359817505,
+      "learning_rate": 9.757967032967033e-05,
+      "loss": 0.5276,
+      "step": 1090
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 3.142998218536377,
+      "learning_rate": 9.755219780219782e-05,
+      "loss": 0.5238,
+      "step": 1100
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 2.13360857963562,
+      "learning_rate": 9.752472527472528e-05,
+      "loss": 0.5326,
+      "step": 1110
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.2551095485687256,
+      "learning_rate": 9.749725274725275e-05,
+      "loss": 0.5569,
+      "step": 1120
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.0858782529830933,
+      "learning_rate": 9.746978021978023e-05,
+      "loss": 0.5425,
+      "step": 1130
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.3724364042282104,
+      "learning_rate": 9.74423076923077e-05,
+      "loss": 0.5381,
+      "step": 1140
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.2211335897445679,
+      "learning_rate": 9.741483516483517e-05,
+      "loss": 0.5147,
+      "step": 1150
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.806498646736145,
+      "learning_rate": 9.738736263736264e-05,
+      "loss": 0.5218,
+      "step": 1160
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.7817394733428955,
+      "learning_rate": 9.735989010989012e-05,
+      "loss": 0.539,
+      "step": 1170
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 2.054344415664673,
+      "learning_rate": 9.733241758241759e-05,
+      "loss": 0.5305,
+      "step": 1180
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.8021188974380493,
+      "learning_rate": 9.730494505494506e-05,
+      "loss": 0.519,
+      "step": 1190
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.833899736404419,
+      "learning_rate": 9.727747252747254e-05,
+      "loss": 0.5589,
+      "step": 1200
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 2.6109519004821777,
+      "learning_rate": 9.725e-05,
+      "loss": 0.5333,
+      "step": 1210
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 2.0866036415100098,
+      "learning_rate": 9.722252747252747e-05,
+      "loss": 0.5436,
+      "step": 1220
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 2.150935649871826,
+      "learning_rate": 9.719505494505496e-05,
+      "loss": 0.5202,
+      "step": 1230
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.280685544013977,
+      "learning_rate": 9.716758241758242e-05,
+      "loss": 0.5146,
+      "step": 1240
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.7753365635871887,
+      "learning_rate": 9.714010989010989e-05,
+      "loss": 0.5106,
+      "step": 1250
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.8706464767456055,
+      "learning_rate": 9.711263736263737e-05,
+      "loss": 0.4952,
+      "step": 1260
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.3467265367507935,
+      "learning_rate": 9.708516483516484e-05,
+      "loss": 0.5105,
+      "step": 1270
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 2.5618174076080322,
+      "learning_rate": 9.705769230769231e-05,
+      "loss": 0.5065,
+      "step": 1280
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.3450369834899902,
+      "learning_rate": 9.703021978021979e-05,
+      "loss": 0.5201,
+      "step": 1290
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.3272931575775146,
+      "learning_rate": 9.700274725274726e-05,
+      "loss": 0.5189,
+      "step": 1300
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 2.324739694595337,
+      "learning_rate": 9.697527472527473e-05,
+      "loss": 0.5298,
+      "step": 1310
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.375444769859314,
+      "learning_rate": 9.694780219780221e-05,
+      "loss": 0.527,
+      "step": 1320
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.1685923337936401,
+      "learning_rate": 9.692032967032968e-05,
+      "loss": 0.5303,
+      "step": 1330
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.6014115810394287,
+      "learning_rate": 9.689285714285715e-05,
+      "loss": 0.5399,
+      "step": 1340
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 2.3533401489257812,
+      "learning_rate": 9.686538461538463e-05,
+      "loss": 0.5185,
+      "step": 1350
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.6813533306121826,
+      "learning_rate": 9.68379120879121e-05,
+      "loss": 0.4986,
+      "step": 1360
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.1102067232131958,
+      "learning_rate": 9.681043956043956e-05,
+      "loss": 0.5313,
+      "step": 1370
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.576412320137024,
+      "learning_rate": 9.678296703296705e-05,
+      "loss": 0.5087,
+      "step": 1380
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.5507354736328125,
+      "learning_rate": 9.675549450549451e-05,
+      "loss": 0.532,
+      "step": 1390
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.4840365648269653,
+      "learning_rate": 9.672802197802198e-05,
+      "loss": 0.5191,
+      "step": 1400
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.8767573833465576,
+      "learning_rate": 9.670054945054946e-05,
+      "loss": 0.5252,
+      "step": 1410
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.0486218929290771,
+      "learning_rate": 9.667307692307693e-05,
+      "loss": 0.5152,
+      "step": 1420
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 2.066174030303955,
+      "learning_rate": 9.66456043956044e-05,
+      "loss": 0.5163,
+      "step": 1430
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.7480829954147339,
+      "learning_rate": 9.661813186813188e-05,
+      "loss": 0.5292,
+      "step": 1440
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.6554828882217407,
+      "learning_rate": 9.659065934065935e-05,
+      "loss": 0.5301,
+      "step": 1450
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.0723321437835693,
+      "learning_rate": 9.656318681318682e-05,
+      "loss": 0.5118,
+      "step": 1460
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.7398673295974731,
+      "learning_rate": 9.653571428571429e-05,
+      "loss": 0.5268,
+      "step": 1470
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.2732940912246704,
+      "learning_rate": 9.650824175824175e-05,
+      "loss": 0.5085,
+      "step": 1480
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.0260119438171387,
+      "learning_rate": 9.648076923076924e-05,
+      "loss": 0.5243,
+      "step": 1490
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.8816061019897461,
+      "learning_rate": 9.64532967032967e-05,
+      "loss": 0.5158,
+      "step": 1500
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 2.7940518856048584,
+      "learning_rate": 9.642582417582417e-05,
+      "loss": 0.5051,
+      "step": 1510
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.1889257431030273,
+      "learning_rate": 9.639835164835165e-05,
+      "loss": 0.5111,
+      "step": 1520
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.8643608093261719,
+      "learning_rate": 9.637087912087912e-05,
+      "loss": 0.5136,
+      "step": 1530
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.658339023590088,
+      "learning_rate": 9.634340659340659e-05,
+      "loss": 0.5132,
+      "step": 1540
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.6262873411178589,
+      "learning_rate": 9.631593406593407e-05,
+      "loss": 0.5047,
+      "step": 1550
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.2276610136032104,
+      "learning_rate": 9.628846153846154e-05,
+      "loss": 0.4997,
+      "step": 1560
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.3661067485809326,
+      "learning_rate": 9.626098901098901e-05,
+      "loss": 0.5113,
+      "step": 1570
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.423560380935669,
+      "learning_rate": 9.623351648351649e-05,
+      "loss": 0.5167,
+      "step": 1580
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.391201376914978,
+      "learning_rate": 9.620604395604396e-05,
+      "loss": 0.5057,
+      "step": 1590
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.5331265926361084,
+      "learning_rate": 9.617857142857143e-05,
+      "loss": 0.5105,
+      "step": 1600
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 2.0285804271698,
+      "learning_rate": 9.615109890109891e-05,
+      "loss": 0.5009,
+      "step": 1610
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.470106840133667,
+      "learning_rate": 9.612362637362638e-05,
+      "loss": 0.5082,
+      "step": 1620
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.8231136798858643,
+      "learning_rate": 9.609615384615384e-05,
+      "loss": 0.4893,
+      "step": 1630
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.4825439453125,
+      "learning_rate": 9.606868131868133e-05,
+      "loss": 0.5132,
+      "step": 1640
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.8054287433624268,
+      "learning_rate": 9.60412087912088e-05,
+      "loss": 0.4878,
+      "step": 1650
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.9411340355873108,
+      "learning_rate": 9.601373626373626e-05,
+      "loss": 0.4908,
+      "step": 1660
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.754782795906067,
+      "learning_rate": 9.598626373626374e-05,
+      "loss": 0.5206,
+      "step": 1670
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.1740814447402954,
+      "learning_rate": 9.595879120879121e-05,
+      "loss": 0.514,
+      "step": 1680
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.0887936353683472,
+      "learning_rate": 9.593131868131868e-05,
+      "loss": 0.5095,
+      "step": 1690
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 2.0167980194091797,
+      "learning_rate": 9.590384615384616e-05,
+      "loss": 0.5066,
+      "step": 1700
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.6176742315292358,
+      "learning_rate": 9.587637362637363e-05,
+      "loss": 0.5016,
+      "step": 1710
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.0261584520339966,
+      "learning_rate": 9.58489010989011e-05,
+      "loss": 0.5032,
+      "step": 1720
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.0735365152359009,
+      "learning_rate": 9.582142857142858e-05,
+      "loss": 0.4894,
+      "step": 1730
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.4293698072433472,
+      "learning_rate": 9.579395604395605e-05,
+      "loss": 0.5111,
+      "step": 1740
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.296971321105957,
+      "learning_rate": 9.576648351648352e-05,
+      "loss": 0.5078,
+      "step": 1750
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.7727321982383728,
+      "learning_rate": 9.5739010989011e-05,
+      "loss": 0.4923,
+      "step": 1760
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 2.6028566360473633,
+      "learning_rate": 9.571153846153847e-05,
+      "loss": 0.4994,
+      "step": 1770
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.2606744766235352,
+      "learning_rate": 9.568406593406593e-05,
+      "loss": 0.4885,
+      "step": 1780
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.812680721282959,
+      "learning_rate": 9.565659340659342e-05,
+      "loss": 0.5101,
+      "step": 1790
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.2432329654693604,
+      "learning_rate": 9.562912087912088e-05,
+      "loss": 0.5012,
+      "step": 1800
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.365795373916626,
+      "learning_rate": 9.560164835164835e-05,
+      "loss": 0.5093,
+      "step": 1810
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.4509589672088623,
+      "learning_rate": 9.557417582417583e-05,
+      "loss": 0.4921,
+      "step": 1820
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.0986443758010864,
+      "learning_rate": 9.55467032967033e-05,
+      "loss": 0.5003,
+      "step": 1830
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.46748685836792,
+      "learning_rate": 9.551923076923077e-05,
+      "loss": 0.4925,
+      "step": 1840
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.3228458166122437,
+      "learning_rate": 9.549175824175825e-05,
+      "loss": 0.5032,
+      "step": 1850
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.8542389273643494,
+      "learning_rate": 9.546428571428572e-05,
+      "loss": 0.4896,
+      "step": 1860
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.7039434909820557,
+      "learning_rate": 9.543681318681319e-05,
+      "loss": 0.4852,
+      "step": 1870
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.1597633361816406,
+      "learning_rate": 9.540934065934067e-05,
+      "loss": 0.5015,
+      "step": 1880
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 2.178933620452881,
+      "learning_rate": 9.538186813186814e-05,
+      "loss": 0.5298,
+      "step": 1890
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.1454273462295532,
+      "learning_rate": 9.53543956043956e-05,
+      "loss": 0.5046,
+      "step": 1900
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.7407316565513611,
+      "learning_rate": 9.532692307692309e-05,
+      "loss": 0.486,
+      "step": 1910
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.7394108772277832,
+      "learning_rate": 9.529945054945056e-05,
+      "loss": 0.4977,
+      "step": 1920
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.9852740168571472,
+      "learning_rate": 9.527197802197802e-05,
+      "loss": 0.4871,
+      "step": 1930
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.4632673263549805,
+      "learning_rate": 9.52445054945055e-05,
+      "loss": 0.5082,
+      "step": 1940
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.4848504066467285,
+      "learning_rate": 9.521703296703297e-05,
+      "loss": 0.5067,
+      "step": 1950
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.5140033960342407,
+      "learning_rate": 9.518956043956044e-05,
+      "loss": 0.5063,
+      "step": 1960
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 2.4344804286956787,
+      "learning_rate": 9.516208791208791e-05,
+      "loss": 0.5099,
+      "step": 1970
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.8636808395385742,
+      "learning_rate": 9.513461538461539e-05,
+      "loss": 0.4902,
+      "step": 1980
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.4363267421722412,
+      "learning_rate": 9.510714285714286e-05,
+      "loss": 0.498,
+      "step": 1990
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.7588983774185181,
+      "learning_rate": 9.507967032967033e-05,
+      "loss": 0.4884,
+      "step": 2000
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.4090018272399902,
+      "learning_rate": 9.505219780219781e-05,
+      "loss": 0.5082,
+      "step": 2010
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 2.2110331058502197,
+      "learning_rate": 9.502472527472528e-05,
+      "loss": 0.5064,
+      "step": 2020
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.2035874128341675,
+      "learning_rate": 9.499725274725275e-05,
+      "loss": 0.4952,
+      "step": 2030
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.597834825515747,
+      "learning_rate": 9.496978021978023e-05,
+      "loss": 0.4956,
+      "step": 2040
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.4878921508789062,
+      "learning_rate": 9.49423076923077e-05,
+      "loss": 0.4865,
+      "step": 2050
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.6747124195098877,
+      "learning_rate": 9.491483516483516e-05,
+      "loss": 0.4935,
+      "step": 2060
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.3186380863189697,
+      "learning_rate": 9.488736263736264e-05,
+      "loss": 0.4938,
+      "step": 2070
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 2.2865145206451416,
+      "learning_rate": 9.485989010989011e-05,
+      "loss": 0.4985,
+      "step": 2080
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 2.515855550765991,
+      "learning_rate": 9.483241758241758e-05,
+      "loss": 0.5049,
+      "step": 2090
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.2123080492019653,
+      "learning_rate": 9.480494505494506e-05,
+      "loss": 0.5004,
+      "step": 2100
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.0919526815414429,
+      "learning_rate": 9.477747252747253e-05,
+      "loss": 0.4854,
+      "step": 2110
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.3105583190917969,
+      "learning_rate": 9.475e-05,
+      "loss": 0.4866,
+      "step": 2120
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.745622158050537,
+      "learning_rate": 9.472252747252748e-05,
+      "loss": 0.4892,
+      "step": 2130
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.6633902192115784,
+      "learning_rate": 9.469505494505495e-05,
+      "loss": 0.507,
+      "step": 2140
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.0530184507369995,
+      "learning_rate": 9.466758241758242e-05,
+      "loss": 0.5035,
+      "step": 2150
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.2395071983337402,
+      "learning_rate": 9.46401098901099e-05,
+      "loss": 0.4926,
+      "step": 2160
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.139224886894226,
+      "learning_rate": 9.461263736263737e-05,
+      "loss": 0.4861,
+      "step": 2170
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.844968318939209,
+      "learning_rate": 9.458516483516484e-05,
+      "loss": 0.4855,
+      "step": 2180
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.2388585805892944,
+      "learning_rate": 9.455769230769232e-05,
+      "loss": 0.4887,
+      "step": 2190
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.4267388582229614,
+      "learning_rate": 9.453021978021978e-05,
+      "loss": 0.5031,
+      "step": 2200
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.7797860503196716,
+      "learning_rate": 9.450274725274725e-05,
+      "loss": 0.5014,
+      "step": 2210
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.3723804950714111,
+      "learning_rate": 9.447527472527473e-05,
+      "loss": 0.4887,
+      "step": 2220
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.396022081375122,
+      "learning_rate": 9.44478021978022e-05,
+      "loss": 0.4869,
+      "step": 2230
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.3555471897125244,
+      "learning_rate": 9.442032967032967e-05,
+      "loss": 0.4879,
+      "step": 2240
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.9702646136283875,
+      "learning_rate": 9.439285714285715e-05,
+      "loss": 0.4894,
+      "step": 2250
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.7753974199295044,
+      "learning_rate": 9.436538461538462e-05,
+      "loss": 0.5092,
+      "step": 2260
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.020336389541626,
+      "learning_rate": 9.433791208791209e-05,
+      "loss": 0.4961,
+      "step": 2270
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.2176239490509033,
+      "learning_rate": 9.431043956043957e-05,
+      "loss": 0.4827,
+      "step": 2280
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.8337396383285522,
+      "learning_rate": 9.428296703296704e-05,
+      "loss": 0.4931,
+      "step": 2290
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.8999798893928528,
+      "learning_rate": 9.425549450549451e-05,
+      "loss": 0.4912,
+      "step": 2300
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.6992242336273193,
+      "learning_rate": 9.422802197802199e-05,
+      "loss": 0.4835,
+      "step": 2310
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.7739867568016052,
+      "learning_rate": 9.420054945054946e-05,
+      "loss": 0.4783,
+      "step": 2320
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.2350486516952515,
+      "learning_rate": 9.417307692307692e-05,
+      "loss": 0.4904,
+      "step": 2330
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.1423838138580322,
+      "learning_rate": 9.41456043956044e-05,
+      "loss": 0.4884,
+      "step": 2340
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.0562171936035156,
+      "learning_rate": 9.411813186813187e-05,
+      "loss": 0.5082,
+      "step": 2350
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.7696706056594849,
+      "learning_rate": 9.409065934065934e-05,
+      "loss": 0.4889,
+      "step": 2360
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.5980774760246277,
+      "learning_rate": 9.406318681318682e-05,
+      "loss": 0.48,
+      "step": 2370
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.413926124572754,
+      "learning_rate": 9.403571428571429e-05,
+      "loss": 0.4971,
+      "step": 2380
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.8020157217979431,
+      "learning_rate": 9.400824175824176e-05,
+      "loss": 0.4926,
+      "step": 2390
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.6620104312896729,
+      "learning_rate": 9.398076923076924e-05,
+      "loss": 0.4941,
+      "step": 2400
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.9766381978988647,
+      "learning_rate": 9.395329670329671e-05,
+      "loss": 0.4867,
+      "step": 2410
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.3235608339309692,
+      "learning_rate": 9.392582417582418e-05,
+      "loss": 0.5061,
+      "step": 2420
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.8266246914863586,
+      "learning_rate": 9.389835164835166e-05,
+      "loss": 0.4842,
+      "step": 2430
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.1611332893371582,
+      "learning_rate": 9.387087912087913e-05,
+      "loss": 0.4795,
+      "step": 2440
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.750346064567566,
+      "learning_rate": 9.38434065934066e-05,
+      "loss": 0.4859,
+      "step": 2450
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.7080681324005127,
+      "learning_rate": 9.381593406593408e-05,
+      "loss": 0.4762,
+      "step": 2460
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.3531988859176636,
+      "learning_rate": 9.378846153846155e-05,
+      "loss": 0.4921,
+      "step": 2470
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.8100531697273254,
+      "learning_rate": 9.376098901098901e-05,
+      "loss": 0.4912,
+      "step": 2480
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.6417909860610962,
+      "learning_rate": 9.37335164835165e-05,
+      "loss": 0.4868,
+      "step": 2490
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 2.7357239723205566,
+      "learning_rate": 9.370604395604396e-05,
+      "loss": 0.4837,
+      "step": 2500
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.7493315935134888,
+      "learning_rate": 9.367857142857143e-05,
+      "loss": 0.494,
+      "step": 2510
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.4594992399215698,
+      "learning_rate": 9.365109890109891e-05,
+      "loss": 0.4967,
+      "step": 2520
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.2568315267562866,
+      "learning_rate": 9.362362637362638e-05,
+      "loss": 0.4865,
+      "step": 2530
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.8711192607879639,
+      "learning_rate": 9.359615384615385e-05,
+      "loss": 0.4915,
+      "step": 2540
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.6470551490783691,
+      "learning_rate": 9.356868131868133e-05,
+      "loss": 0.4736,
+      "step": 2550
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 2.410200357437134,
+      "learning_rate": 9.35412087912088e-05,
+      "loss": 0.5078,
+      "step": 2560
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.0611746311187744,
+      "learning_rate": 9.351373626373627e-05,
+      "loss": 0.4986,
+      "step": 2570
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.820321798324585,
+      "learning_rate": 9.348626373626375e-05,
+      "loss": 0.4788,
+      "step": 2580
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.7887302041053772,
+      "learning_rate": 9.345879120879122e-05,
+      "loss": 0.4878,
+      "step": 2590
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.2939894199371338,
+      "learning_rate": 9.343131868131869e-05,
+      "loss": 0.4896,
+      "step": 2600
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.7468405961990356,
+      "learning_rate": 9.340384615384615e-05,
+      "loss": 0.4932,
+      "step": 2610
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.63861882686615,
+      "learning_rate": 9.337637362637362e-05,
+      "loss": 0.4739,
+      "step": 2620
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.1842321157455444,
+      "learning_rate": 9.33489010989011e-05,
+      "loss": 0.4845,
+      "step": 2630
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.0613666772842407,
+      "learning_rate": 9.332142857142857e-05,
+      "loss": 0.4806,
+      "step": 2640
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.9317752718925476,
+      "learning_rate": 9.329395604395604e-05,
+      "loss": 0.4803,
+      "step": 2650
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.378394603729248,
+      "learning_rate": 9.326648351648352e-05,
+      "loss": 0.4942,
+      "step": 2660
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.748068630695343,
+      "learning_rate": 9.323901098901099e-05,
+      "loss": 0.4763,
+      "step": 2670
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.4142990112304688,
+      "learning_rate": 9.321153846153846e-05,
+      "loss": 0.4913,
+      "step": 2680
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.6746355891227722,
+      "learning_rate": 9.318406593406594e-05,
+      "loss": 0.4781,
+      "step": 2690
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.146001935005188,
+      "learning_rate": 9.315659340659341e-05,
+      "loss": 0.4789,
+      "step": 2700
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.0291472673416138,
+      "learning_rate": 9.312912087912088e-05,
+      "loss": 0.4813,
+      "step": 2710
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.7905545234680176,
+      "learning_rate": 9.310164835164836e-05,
+      "loss": 0.475,
+      "step": 2720
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.8916068077087402,
+      "learning_rate": 9.307417582417583e-05,
+      "loss": 0.4891,
+      "step": 2730
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.0133769512176514,
+      "learning_rate": 9.30467032967033e-05,
+      "loss": 0.4857,
+      "step": 2740
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.6718567609786987,
+      "learning_rate": 9.301923076923078e-05,
+      "loss": 0.4915,
+      "step": 2750
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.2999911308288574,
+      "learning_rate": 9.299175824175824e-05,
+      "loss": 0.4872,
+      "step": 2760
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.4812780618667603,
+      "learning_rate": 9.296428571428571e-05,
+      "loss": 0.4886,
+      "step": 2770
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.3728208541870117,
+      "learning_rate": 9.293681318681318e-05,
+      "loss": 0.5005,
+      "step": 2780
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.2773633003234863,
+      "learning_rate": 9.290934065934066e-05,
+      "loss": 0.4796,
+      "step": 2790
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.8197434544563293,
+      "learning_rate": 9.288186813186813e-05,
+      "loss": 0.4817,
+      "step": 2800
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.830235481262207,
+      "learning_rate": 9.28543956043956e-05,
+      "loss": 0.4956,
+      "step": 2810
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 3.1286449432373047,
+      "learning_rate": 9.282692307692308e-05,
+      "loss": 0.4822,
+      "step": 2820
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.9349818825721741,
+      "learning_rate": 9.279945054945055e-05,
+      "loss": 0.4924,
+      "step": 2830
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.8471746444702148,
+      "learning_rate": 9.277197802197802e-05,
+      "loss": 0.4771,
+      "step": 2840
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.9180948734283447,
+      "learning_rate": 9.27445054945055e-05,
+      "loss": 0.4855,
+      "step": 2850
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.7180835604667664,
+      "learning_rate": 9.271703296703297e-05,
+      "loss": 0.4801,
+      "step": 2860
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.5345978736877441,
+      "learning_rate": 9.268956043956043e-05,
+      "loss": 0.4791,
+      "step": 2870
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.521747350692749,
+      "learning_rate": 9.266208791208792e-05,
+      "loss": 0.4793,
+      "step": 2880
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.2364057302474976,
+      "learning_rate": 9.263461538461538e-05,
+      "loss": 0.4756,
+      "step": 2890
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.5741211771965027,
+      "learning_rate": 9.260714285714285e-05,
+      "loss": 0.4849,
+      "step": 2900
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.7251524925231934,
+      "learning_rate": 9.257967032967033e-05,
+      "loss": 0.4762,
+      "step": 2910
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 2.704491138458252,
+      "learning_rate": 9.25521978021978e-05,
+      "loss": 0.4777,
+      "step": 2920
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.9178886413574219,
+      "learning_rate": 9.252472527472527e-05,
+      "loss": 0.5077,
+      "step": 2930
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.2007173299789429,
+      "learning_rate": 9.249725274725275e-05,
+      "loss": 0.4916,
+      "step": 2940
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.8556911945343018,
+      "learning_rate": 9.246978021978022e-05,
+      "loss": 0.477,
+      "step": 2950
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.7754554152488708,
+      "learning_rate": 9.244230769230769e-05,
+      "loss": 0.4859,
+      "step": 2960
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 2.333521842956543,
+      "learning_rate": 9.241483516483517e-05,
+      "loss": 0.4811,
+      "step": 2970
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.9892757534980774,
+      "learning_rate": 9.238736263736264e-05,
+      "loss": 0.4777,
+      "step": 2980
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.0326083898544312,
+      "learning_rate": 9.23598901098901e-05,
+      "loss": 0.4891,
+      "step": 2990
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.4555460214614868,
+      "learning_rate": 9.233241758241759e-05,
+      "loss": 0.4802,
+      "step": 3000
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.8390610814094543,
+      "learning_rate": 9.230494505494506e-05,
+      "loss": 0.4835,
+      "step": 3010
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.0490790605545044,
+      "learning_rate": 9.227747252747252e-05,
+      "loss": 0.4806,
+      "step": 3020
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.8793150186538696,
+      "learning_rate": 9.225e-05,
+      "loss": 0.4846,
+      "step": 3030
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.2842731475830078,
+      "learning_rate": 9.222252747252747e-05,
+      "loss": 0.4715,
+      "step": 3040
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.7668898701667786,
+      "learning_rate": 9.219505494505494e-05,
+      "loss": 0.4749,
+      "step": 3050
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.8144903182983398,
+      "learning_rate": 9.216758241758242e-05,
+      "loss": 0.4705,
+      "step": 3060
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.6727710962295532,
+      "learning_rate": 9.214010989010989e-05,
+      "loss": 0.4826,
+      "step": 3070
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.8702958822250366,
+      "learning_rate": 9.211263736263736e-05,
+      "loss": 0.4679,
+      "step": 3080
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.9223732352256775,
+      "learning_rate": 9.208516483516484e-05,
+      "loss": 0.4875,
+      "step": 3090
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.7086817622184753,
+      "learning_rate": 9.205769230769231e-05,
+      "loss": 0.4771,
+      "step": 3100
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.000447154045105,
+      "learning_rate": 9.203021978021978e-05,
+      "loss": 0.4674,
+      "step": 3110
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.6982853412628174,
+      "learning_rate": 9.200274725274726e-05,
+      "loss": 0.4853,
+      "step": 3120
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.8975854516029358,
+      "learning_rate": 9.197527472527473e-05,
+      "loss": 0.4757,
+      "step": 3130
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.638816237449646,
+      "learning_rate": 9.19478021978022e-05,
+      "loss": 0.4862,
+      "step": 3140
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.0290194749832153,
+      "learning_rate": 9.192032967032968e-05,
+      "loss": 0.4758,
+      "step": 3150
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.0205775499343872,
+      "learning_rate": 9.189285714285715e-05,
+      "loss": 0.4723,
+      "step": 3160
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.2193140983581543,
+      "learning_rate": 9.186538461538461e-05,
+      "loss": 0.4869,
+      "step": 3170
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.0402013063430786,
+      "learning_rate": 9.18379120879121e-05,
+      "loss": 0.4911,
+      "step": 3180
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.1915889978408813,
+      "learning_rate": 9.181043956043956e-05,
+      "loss": 0.4757,
+      "step": 3190
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.2371594905853271,
+      "learning_rate": 9.178296703296703e-05,
+      "loss": 0.4811,
+      "step": 3200
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.9872990846633911,
+      "learning_rate": 9.175549450549451e-05,
+      "loss": 0.4784,
+      "step": 3210
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.9502705335617065,
+      "learning_rate": 9.172802197802198e-05,
+      "loss": 0.4716,
+      "step": 3220
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.9108206033706665,
+      "learning_rate": 9.170054945054945e-05,
+      "loss": 0.4828,
+      "step": 3230
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.5662224888801575,
+      "learning_rate": 9.167307692307693e-05,
+      "loss": 0.485,
+      "step": 3240
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.9142247438430786,
+      "learning_rate": 9.16456043956044e-05,
+      "loss": 0.4851,
+      "step": 3250
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.817579984664917,
+      "learning_rate": 9.161813186813187e-05,
+      "loss": 0.4807,
+      "step": 3260
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.3611973524093628,
+      "learning_rate": 9.159065934065935e-05,
+      "loss": 0.4815,
+      "step": 3270
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.4995872974395752,
+      "learning_rate": 9.156318681318682e-05,
+      "loss": 0.4779,
+      "step": 3280
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.8315975069999695,
+      "learning_rate": 9.153571428571429e-05,
+      "loss": 0.4732,
+      "step": 3290
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.285447359085083,
+      "learning_rate": 9.150824175824177e-05,
+      "loss": 0.4769,
+      "step": 3300
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.265435814857483,
+      "learning_rate": 9.148076923076924e-05,
+      "loss": 0.4884,
+      "step": 3310
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.7437517046928406,
+      "learning_rate": 9.14532967032967e-05,
+      "loss": 0.4807,
+      "step": 3320
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.8307008743286133,
+      "learning_rate": 9.142582417582419e-05,
+      "loss": 0.4723,
+      "step": 3330
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.2148619890213013,
+      "learning_rate": 9.139835164835165e-05,
+      "loss": 0.461,
+      "step": 3340
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.5682019591331482,
+      "learning_rate": 9.137087912087912e-05,
+      "loss": 0.4673,
+      "step": 3350
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.582708477973938,
+      "learning_rate": 9.13434065934066e-05,
+      "loss": 0.4657,
+      "step": 3360
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.0724083185195923,
+      "learning_rate": 9.131593406593407e-05,
+      "loss": 0.4634,
+      "step": 3370
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.511168360710144,
+      "learning_rate": 9.128846153846154e-05,
+      "loss": 0.4888,
+      "step": 3380
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.5445438623428345,
+      "learning_rate": 9.126098901098902e-05,
+      "loss": 0.4757,
+      "step": 3390
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.9609503746032715,
+      "learning_rate": 9.123351648351649e-05,
+      "loss": 0.4813,
+      "step": 3400
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 2.094918966293335,
+      "learning_rate": 9.120604395604396e-05,
+      "loss": 0.4963,
+      "step": 3410
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.6844788789749146,
+      "learning_rate": 9.117857142857144e-05,
+      "loss": 0.4902,
+      "step": 3420
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.2747464179992676,
+      "learning_rate": 9.115109890109891e-05,
+      "loss": 0.4884,
+      "step": 3430
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.4035594463348389,
+      "learning_rate": 9.112362637362638e-05,
+      "loss": 0.482,
+      "step": 3440
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.6047748923301697,
+      "learning_rate": 9.109615384615386e-05,
+      "loss": 0.4846,
+      "step": 3450
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.6141401529312134,
+      "learning_rate": 9.106868131868133e-05,
+      "loss": 0.4744,
+      "step": 3460
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.8329553008079529,
+      "learning_rate": 9.10412087912088e-05,
+      "loss": 0.4738,
+      "step": 3470
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.7500654458999634,
+      "learning_rate": 9.101373626373628e-05,
+      "loss": 0.4763,
+      "step": 3480
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.8617835640907288,
+      "learning_rate": 9.098626373626374e-05,
+      "loss": 0.4738,
+      "step": 3490
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.0339410305023193,
+      "learning_rate": 9.095879120879121e-05,
+      "loss": 0.4726,
+      "step": 3500
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.13742995262146,
+      "learning_rate": 9.09313186813187e-05,
+      "loss": 0.4827,
+      "step": 3510
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.611344575881958,
+      "learning_rate": 9.090384615384616e-05,
+      "loss": 0.4844,
+      "step": 3520
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.056773066520691,
+      "learning_rate": 9.087637362637363e-05,
+      "loss": 0.4799,
+      "step": 3530
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.2290446758270264,
+      "learning_rate": 9.084890109890111e-05,
+      "loss": 0.4764,
+      "step": 3540
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.7461050748825073,
+      "learning_rate": 9.082142857142858e-05,
+      "loss": 0.4664,
+      "step": 3550
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.5372506380081177,
+      "learning_rate": 9.079395604395605e-05,
+      "loss": 0.4795,
+      "step": 3560
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.658233642578125,
+      "learning_rate": 9.076648351648353e-05,
+      "loss": 0.4655,
+      "step": 3570
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.7918599843978882,
+      "learning_rate": 9.0739010989011e-05,
+      "loss": 0.4777,
+      "step": 3580
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.9504883289337158,
+      "learning_rate": 9.071153846153847e-05,
+      "loss": 0.4791,
+      "step": 3590
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.5631986856460571,
+      "learning_rate": 9.068406593406595e-05,
+      "loss": 0.4807,
+      "step": 3600
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.9991289377212524,
+      "learning_rate": 9.065659340659342e-05,
+      "loss": 0.4679,
+      "step": 3610
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.130434274673462,
+      "learning_rate": 9.062912087912088e-05,
+      "loss": 0.4781,
+      "step": 3620
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.639451265335083,
+      "learning_rate": 9.060164835164837e-05,
+      "loss": 0.4815,
+      "step": 3630
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.1776291131973267,
+      "learning_rate": 9.057417582417583e-05,
+      "loss": 0.4983,
+      "step": 3640
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 1.4073742628097534,
+      "learning_rate": 9.05467032967033e-05,
+      "loss": 0.4711,
+      "step": 3650
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 1.308344841003418,
+      "learning_rate": 9.051923076923078e-05,
+      "loss": 0.4847,
+      "step": 3660
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 0.44810301065444946,
+      "eval_runtime": 1256.2985,
+      "eval_samples_per_second": 232.516,
+      "eval_steps_per_second": 3.634,
+      "step": 3660
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.6774665117263794,
+      "learning_rate": 9.049175824175825e-05,
+      "loss": 0.4791,
+      "step": 3670
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.6826792359352112,
+      "learning_rate": 9.046428571428572e-05,
+      "loss": 0.4775,
+      "step": 3680
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.7845907211303711,
+      "learning_rate": 9.04368131868132e-05,
+      "loss": 0.4765,
+      "step": 3690
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.9099870324134827,
+      "learning_rate": 9.040934065934067e-05,
+      "loss": 0.4689,
+      "step": 3700
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 1.374972939491272,
+      "learning_rate": 9.038186813186814e-05,
+      "loss": 0.469,
+      "step": 3710
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.8698770403862,
+      "learning_rate": 9.035439560439562e-05,
+      "loss": 0.4717,
+      "step": 3720
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.8832442164421082,
+      "learning_rate": 9.032692307692309e-05,
+      "loss": 0.4653,
+      "step": 3730
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.722259521484375,
+      "learning_rate": 9.029945054945054e-05,
+      "loss": 0.4818,
+      "step": 3740
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 1.1362375020980835,
+      "learning_rate": 9.027197802197802e-05,
+      "loss": 0.4674,
+      "step": 3750
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 1.0420037508010864,
+      "learning_rate": 9.024450549450549e-05,
+      "loss": 0.4806,
+      "step": 3760
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.9857313632965088,
+      "learning_rate": 9.021703296703296e-05,
+      "loss": 0.478,
+      "step": 3770
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 1.1428401470184326,
+      "learning_rate": 9.018956043956044e-05,
+      "loss": 0.4742,
+      "step": 3780
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 1.5837888717651367,
+      "learning_rate": 9.016208791208791e-05,
+      "loss": 0.4841,
+      "step": 3790
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 1.9123620986938477,
+      "learning_rate": 9.013461538461538e-05,
+      "loss": 0.4707,
+      "step": 3800
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.9019024968147278,
+      "learning_rate": 9.010714285714286e-05,
+      "loss": 0.4731,
+      "step": 3810
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.9601519107818604,
+      "learning_rate": 9.007967032967033e-05,
+      "loss": 0.4644,
+      "step": 3820
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.4769773483276367,
+      "learning_rate": 9.00521978021978e-05,
+      "loss": 0.4755,
+      "step": 3830
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 1.2578529119491577,
+      "learning_rate": 9.002472527472528e-05,
+      "loss": 0.4703,
+      "step": 3840
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 1.1451194286346436,
+      "learning_rate": 8.999725274725275e-05,
+      "loss": 0.4748,
+      "step": 3850
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.4718180000782013,
+      "learning_rate": 8.996978021978021e-05,
+      "loss": 0.4673,
+      "step": 3860
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 1.0323463678359985,
+      "learning_rate": 8.99423076923077e-05,
+      "loss": 0.4611,
+      "step": 3870
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.905623197555542,
+      "learning_rate": 8.991483516483516e-05,
+      "loss": 0.4651,
+      "step": 3880
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.6803371906280518,
+      "learning_rate": 8.988736263736263e-05,
+      "loss": 0.4693,
+      "step": 3890
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.7653062343597412,
+      "learning_rate": 8.985989010989011e-05,
+      "loss": 0.4676,
+      "step": 3900
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.6493901610374451,
+      "learning_rate": 8.983241758241758e-05,
+      "loss": 0.4796,
+      "step": 3910
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 1.3194290399551392,
+      "learning_rate": 8.980494505494505e-05,
+      "loss": 0.4709,
+      "step": 3920
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 1.3907560110092163,
+      "learning_rate": 8.977747252747253e-05,
+      "loss": 0.4791,
+      "step": 3930
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.9326291084289551,
+      "learning_rate": 8.975e-05,
+      "loss": 0.4733,
+      "step": 3940
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.6711992025375366,
+      "learning_rate": 8.972252747252747e-05,
+      "loss": 0.4607,
+      "step": 3950
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 1.0052268505096436,
+      "learning_rate": 8.969505494505495e-05,
+      "loss": 0.4565,
+      "step": 3960
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 1.893130898475647,
+      "learning_rate": 8.966758241758242e-05,
+      "loss": 0.4678,
+      "step": 3970
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 1.019831657409668,
+      "learning_rate": 8.964010989010989e-05,
+      "loss": 0.4929,
+      "step": 3980
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 1.096341848373413,
+      "learning_rate": 8.961263736263737e-05,
+      "loss": 0.4812,
+      "step": 3990
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 1.3617221117019653,
+      "learning_rate": 8.958516483516484e-05,
+      "loss": 0.479,
+      "step": 4000
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.6594845056533813,
+      "learning_rate": 8.95576923076923e-05,
+      "loss": 0.4784,
+      "step": 4010
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.7108995914459229,
+      "learning_rate": 8.953021978021979e-05,
+      "loss": 0.4624,
+      "step": 4020
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.4914230704307556,
+      "learning_rate": 8.950274725274725e-05,
+      "loss": 0.4573,
+      "step": 4030
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.738325834274292,
+      "learning_rate": 8.947527472527472e-05,
+      "loss": 0.4598,
+      "step": 4040
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 1.5622313022613525,
+      "learning_rate": 8.94478021978022e-05,
+      "loss": 0.4769,
+      "step": 4050
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.84259033203125,
+      "learning_rate": 8.942032967032967e-05,
+      "loss": 0.4587,
+      "step": 4060
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 1.1515220403671265,
+      "learning_rate": 8.939285714285714e-05,
+      "loss": 0.4698,
+      "step": 4070
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 1.2757478952407837,
+      "learning_rate": 8.936538461538462e-05,
+      "loss": 0.4662,
+      "step": 4080
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 1.413041591644287,
+      "learning_rate": 8.933791208791209e-05,
+      "loss": 0.4739,
+      "step": 4090
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 1.3950985670089722,
+      "learning_rate": 8.931043956043956e-05,
+      "loss": 0.4652,
+      "step": 4100
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 1.1247773170471191,
+      "learning_rate": 8.928296703296704e-05,
+      "loss": 0.4704,
+      "step": 4110
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.7197868824005127,
+      "learning_rate": 8.925549450549451e-05,
+      "loss": 0.466,
+      "step": 4120
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 1.606062889099121,
+      "learning_rate": 8.922802197802198e-05,
+      "loss": 0.4697,
+      "step": 4130
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.8834720849990845,
+      "learning_rate": 8.920054945054946e-05,
+      "loss": 0.4787,
+      "step": 4140
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.6989302635192871,
+      "learning_rate": 8.917307692307693e-05,
+      "loss": 0.4553,
+      "step": 4150
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 1.1178115606307983,
+      "learning_rate": 8.91456043956044e-05,
+      "loss": 0.4679,
+      "step": 4160
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.8493313789367676,
+      "learning_rate": 8.911813186813188e-05,
+      "loss": 0.4915,
+      "step": 4170
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.6862457990646362,
+      "learning_rate": 8.909065934065934e-05,
+      "loss": 0.4667,
+      "step": 4180
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.5265287160873413,
+      "learning_rate": 8.906318681318681e-05,
+      "loss": 0.4675,
+      "step": 4190
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 1.6895486116409302,
+      "learning_rate": 8.903571428571429e-05,
+      "loss": 0.4694,
+      "step": 4200
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.9672292470932007,
+      "learning_rate": 8.900824175824176e-05,
+      "loss": 0.4788,
+      "step": 4210
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.9348291158676147,
+      "learning_rate": 8.898076923076923e-05,
+      "loss": 0.461,
+      "step": 4220
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.5047799348831177,
+      "learning_rate": 8.895329670329671e-05,
+      "loss": 0.467,
+      "step": 4230
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 1.6985957622528076,
+      "learning_rate": 8.892582417582418e-05,
+      "loss": 0.4581,
+      "step": 4240
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.8573766946792603,
+      "learning_rate": 8.889835164835165e-05,
+      "loss": 0.4624,
+      "step": 4250
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.6403746008872986,
+      "learning_rate": 8.887087912087913e-05,
+      "loss": 0.4574,
+      "step": 4260
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.9546878933906555,
+      "learning_rate": 8.88434065934066e-05,
+      "loss": 0.4803,
+      "step": 4270
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.6992652416229248,
+      "learning_rate": 8.881593406593407e-05,
+      "loss": 0.4696,
+      "step": 4280
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 1.035543441772461,
+      "learning_rate": 8.878846153846155e-05,
+      "loss": 0.4787,
+      "step": 4290
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 1.5563474893569946,
+      "learning_rate": 8.876098901098902e-05,
+      "loss": 0.4708,
+      "step": 4300
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 1.7203930616378784,
+      "learning_rate": 8.873351648351648e-05,
+      "loss": 0.4725,
+      "step": 4310
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 1.4295490980148315,
+      "learning_rate": 8.870604395604396e-05,
+      "loss": 0.464,
+      "step": 4320
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 1.1832698583602905,
+      "learning_rate": 8.867857142857143e-05,
+      "loss": 0.475,
+      "step": 4330
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.9679028391838074,
+      "learning_rate": 8.86510989010989e-05,
+      "loss": 0.4656,
+      "step": 4340
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.9646287560462952,
+      "learning_rate": 8.862362637362638e-05,
+      "loss": 0.4707,
+      "step": 4350
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 1.2284992933273315,
+      "learning_rate": 8.859615384615385e-05,
+      "loss": 0.4812,
+      "step": 4360
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.7886459231376648,
+      "learning_rate": 8.856868131868132e-05,
+      "loss": 0.4641,
+      "step": 4370
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.8464663028717041,
+      "learning_rate": 8.85412087912088e-05,
+      "loss": 0.4685,
+      "step": 4380
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 1.0549769401550293,
+      "learning_rate": 8.851373626373627e-05,
+      "loss": 0.4573,
+      "step": 4390
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 1.3452584743499756,
+      "learning_rate": 8.848626373626374e-05,
+      "loss": 0.4823,
+      "step": 4400
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 1.5250259637832642,
+      "learning_rate": 8.845879120879122e-05,
+      "loss": 0.4762,
+      "step": 4410
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.7843578457832336,
+      "learning_rate": 8.843131868131869e-05,
+      "loss": 0.4714,
+      "step": 4420
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 1.0666967630386353,
+      "learning_rate": 8.840384615384616e-05,
+      "loss": 0.4743,
+      "step": 4430
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 1.4348440170288086,
+      "learning_rate": 8.837637362637364e-05,
+      "loss": 0.4879,
+      "step": 4440
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 1.036281704902649,
+      "learning_rate": 8.83489010989011e-05,
+      "loss": 0.469,
+      "step": 4450
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.7014042139053345,
+      "learning_rate": 8.832142857142857e-05,
+      "loss": 0.4716,
+      "step": 4460
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 1.593533992767334,
+      "learning_rate": 8.829395604395605e-05,
+      "loss": 0.4745,
+      "step": 4470
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.4499157965183258,
+      "learning_rate": 8.826648351648352e-05,
+      "loss": 0.4764,
+      "step": 4480
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.7330986857414246,
+      "learning_rate": 8.823901098901099e-05,
+      "loss": 0.4733,
+      "step": 4490
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.7915944457054138,
+      "learning_rate": 8.821153846153847e-05,
+      "loss": 0.4605,
+      "step": 4500
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.9147601127624512,
+      "learning_rate": 8.818406593406594e-05,
+      "loss": 0.4729,
+      "step": 4510
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 1.1145246028900146,
+      "learning_rate": 8.815659340659341e-05,
+      "loss": 0.4757,
+      "step": 4520
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 1.0525473356246948,
+      "learning_rate": 8.812912087912089e-05,
+      "loss": 0.4585,
+      "step": 4530
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.7515787482261658,
+      "learning_rate": 8.810164835164836e-05,
+      "loss": 0.4557,
+      "step": 4540
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.9737600088119507,
+      "learning_rate": 8.807417582417583e-05,
+      "loss": 0.4647,
+      "step": 4550
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 1.1443486213684082,
+      "learning_rate": 8.804670329670331e-05,
+      "loss": 0.4685,
+      "step": 4560
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 1.4745419025421143,
+      "learning_rate": 8.801923076923078e-05,
+      "loss": 0.4736,
+      "step": 4570
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 1.1491049528121948,
+      "learning_rate": 8.799175824175824e-05,
+      "loss": 0.4604,
+      "step": 4580
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.742324948310852,
+      "learning_rate": 8.796428571428573e-05,
+      "loss": 0.457,
+      "step": 4590
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 1.1914622783660889,
+      "learning_rate": 8.79368131868132e-05,
+      "loss": 0.4638,
+      "step": 4600
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.6555135846138,
+      "learning_rate": 8.790934065934066e-05,
+      "loss": 0.4806,
+      "step": 4610
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.8627590537071228,
+      "learning_rate": 8.788186813186814e-05,
+      "loss": 0.4803,
+      "step": 4620
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 1.0868836641311646,
+      "learning_rate": 8.785439560439561e-05,
+      "loss": 0.4598,
+      "step": 4630
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 1.291483998298645,
+      "learning_rate": 8.782692307692308e-05,
+      "loss": 0.4707,
+      "step": 4640
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.9766128659248352,
+      "learning_rate": 8.779945054945056e-05,
+      "loss": 0.4707,
+      "step": 4650
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 1.0301162004470825,
+      "learning_rate": 8.777197802197803e-05,
+      "loss": 0.4821,
+      "step": 4660
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 1.0474865436553955,
+      "learning_rate": 8.77445054945055e-05,
+      "loss": 0.4603,
+      "step": 4670
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 1.1218842267990112,
+      "learning_rate": 8.771703296703298e-05,
+      "loss": 0.4827,
+      "step": 4680
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 1.0587611198425293,
+      "learning_rate": 8.768956043956045e-05,
+      "loss": 0.4746,
+      "step": 4690
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.8249474763870239,
+      "learning_rate": 8.766208791208792e-05,
+      "loss": 0.4648,
+      "step": 4700
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.6225512027740479,
+      "learning_rate": 8.76346153846154e-05,
+      "loss": 0.4582,
+      "step": 4710
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 1.2126622200012207,
+      "learning_rate": 8.760714285714287e-05,
+      "loss": 0.4617,
+      "step": 4720
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 1.6850072145462036,
+      "learning_rate": 8.757967032967033e-05,
+      "loss": 0.4793,
+      "step": 4730
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.5053130984306335,
+      "learning_rate": 8.755219780219782e-05,
+      "loss": 0.4648,
+      "step": 4740
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.4924371838569641,
+      "learning_rate": 8.752472527472528e-05,
+      "loss": 0.4694,
+      "step": 4750
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.7182402014732361,
+      "learning_rate": 8.749725274725275e-05,
+      "loss": 0.4611,
+      "step": 4760
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.5569953322410583,
+      "learning_rate": 8.746978021978023e-05,
+      "loss": 0.4632,
+      "step": 4770
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 1.0777251720428467,
+      "learning_rate": 8.74423076923077e-05,
+      "loss": 0.4735,
+      "step": 4780
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.6310242414474487,
+      "learning_rate": 8.741483516483517e-05,
+      "loss": 0.4567,
+      "step": 4790
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.94218510389328,
+      "learning_rate": 8.738736263736265e-05,
+      "loss": 0.4626,
+      "step": 4800
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.4554530084133148,
+      "learning_rate": 8.735989010989012e-05,
+      "loss": 0.4681,
+      "step": 4810
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 1.2381364107131958,
+      "learning_rate": 8.733241758241759e-05,
+      "loss": 0.467,
+      "step": 4820
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 1.1469919681549072,
+      "learning_rate": 8.730494505494507e-05,
+      "loss": 0.4643,
+      "step": 4830
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.7159097194671631,
+      "learning_rate": 8.727747252747254e-05,
+      "loss": 0.4731,
+      "step": 4840
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.8695803880691528,
+      "learning_rate": 8.725e-05,
+      "loss": 0.4646,
+      "step": 4850
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.8535722494125366,
+      "learning_rate": 8.722252747252749e-05,
+      "loss": 0.4612,
+      "step": 4860
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 1.231641173362732,
+      "learning_rate": 8.719505494505496e-05,
+      "loss": 0.48,
+      "step": 4870
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 1.0003236532211304,
+      "learning_rate": 8.716758241758241e-05,
+      "loss": 0.4741,
+      "step": 4880
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 1.8520704507827759,
+      "learning_rate": 8.714010989010989e-05,
+      "loss": 0.4693,
+      "step": 4890
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 1.0661410093307495,
+      "learning_rate": 8.711263736263736e-05,
+      "loss": 0.4848,
+      "step": 4900
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.8280912637710571,
+      "learning_rate": 8.708516483516483e-05,
+      "loss": 0.4703,
+      "step": 4910
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 1.169047474861145,
+      "learning_rate": 8.705769230769231e-05,
+      "loss": 0.4777,
+      "step": 4920
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.7568297982215881,
+      "learning_rate": 8.703021978021978e-05,
+      "loss": 0.4752,
+      "step": 4930
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 1.5538333654403687,
+      "learning_rate": 8.700274725274725e-05,
+      "loss": 0.4569,
+      "step": 4940
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.6673538684844971,
+      "learning_rate": 8.697527472527473e-05,
+      "loss": 0.4659,
+      "step": 4950
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.9370101690292358,
+      "learning_rate": 8.69478021978022e-05,
+      "loss": 0.4644,
+      "step": 4960
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 1.748659610748291,
+      "learning_rate": 8.692032967032966e-05,
+      "loss": 0.458,
+      "step": 4970
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.8026989102363586,
+      "learning_rate": 8.689285714285715e-05,
+      "loss": 0.4696,
+      "step": 4980
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 1.7652183771133423,
+      "learning_rate": 8.686538461538461e-05,
+      "loss": 0.4676,
+      "step": 4990
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.6286242008209229,
+      "learning_rate": 8.683791208791208e-05,
+      "loss": 0.4638,
+      "step": 5000
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 2.5077078342437744,
+      "learning_rate": 8.681043956043956e-05,
+      "loss": 0.4774,
+      "step": 5010
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.6874992251396179,
+      "learning_rate": 8.678296703296703e-05,
+      "loss": 0.4621,
+      "step": 5020
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 1.2562603950500488,
+      "learning_rate": 8.67554945054945e-05,
+      "loss": 0.4745,
+      "step": 5030
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 1.0717673301696777,
+      "learning_rate": 8.672802197802198e-05,
+      "loss": 0.4726,
+      "step": 5040
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.744522750377655,
+      "learning_rate": 8.670054945054945e-05,
+      "loss": 0.4823,
+      "step": 5050
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.879859447479248,
+      "learning_rate": 8.667307692307692e-05,
+      "loss": 0.4651,
+      "step": 5060
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.6319553852081299,
+      "learning_rate": 8.66456043956044e-05,
+      "loss": 0.4714,
+      "step": 5070
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.8513883948326111,
+      "learning_rate": 8.661813186813187e-05,
+      "loss": 0.4644,
+      "step": 5080
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 1.7870076894760132,
+      "learning_rate": 8.659065934065934e-05,
+      "loss": 0.4714,
+      "step": 5090
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.8447625041007996,
+      "learning_rate": 8.656318681318682e-05,
+      "loss": 0.4717,
+      "step": 5100
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.7471101880073547,
+      "learning_rate": 8.653571428571429e-05,
+      "loss": 0.4594,
+      "step": 5110
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.7438103556632996,
+      "learning_rate": 8.650824175824175e-05,
+      "loss": 0.4622,
+      "step": 5120
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.8779565095901489,
+      "learning_rate": 8.648076923076924e-05,
+      "loss": 0.4639,
+      "step": 5130
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 1.4103574752807617,
+      "learning_rate": 8.64532967032967e-05,
+      "loss": 0.4568,
+      "step": 5140
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 1.2644096612930298,
+      "learning_rate": 8.642582417582417e-05,
+      "loss": 0.4696,
+      "step": 5150
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.7385484576225281,
+      "learning_rate": 8.639835164835165e-05,
+      "loss": 0.4638,
+      "step": 5160
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.7698835730552673,
+      "learning_rate": 8.637087912087912e-05,
+      "loss": 0.4762,
+      "step": 5170
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 1.049194574356079,
+      "learning_rate": 8.634340659340659e-05,
+      "loss": 0.4588,
+      "step": 5180
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.7376847267150879,
+      "learning_rate": 8.631593406593407e-05,
+      "loss": 0.46,
+      "step": 5190
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 1.1413201093673706,
+      "learning_rate": 8.628846153846154e-05,
+      "loss": 0.4662,
+      "step": 5200
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.9843841195106506,
+      "learning_rate": 8.626098901098901e-05,
+      "loss": 0.4663,
+      "step": 5210
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.6186099052429199,
+      "learning_rate": 8.623351648351649e-05,
+      "loss": 0.4683,
+      "step": 5220
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 1.4082993268966675,
+      "learning_rate": 8.620604395604396e-05,
+      "loss": 0.4639,
+      "step": 5230
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.9625537395477295,
+      "learning_rate": 8.617857142857143e-05,
+      "loss": 0.4646,
+      "step": 5240
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 1.0915727615356445,
+      "learning_rate": 8.615109890109891e-05,
+      "loss": 0.4669,
+      "step": 5250
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.9645349383354187,
+      "learning_rate": 8.612362637362638e-05,
+      "loss": 0.4572,
+      "step": 5260
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.7993157505989075,
+      "learning_rate": 8.609615384615384e-05,
+      "loss": 0.4623,
+      "step": 5270
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 1.3415577411651611,
+      "learning_rate": 8.606868131868133e-05,
+      "loss": 0.4569,
+      "step": 5280
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 1.1217197179794312,
+      "learning_rate": 8.60412087912088e-05,
+      "loss": 0.4632,
+      "step": 5290
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.6021980047225952,
+      "learning_rate": 8.601373626373626e-05,
+      "loss": 0.4688,
+      "step": 5300
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.9672883749008179,
+      "learning_rate": 8.598626373626374e-05,
+      "loss": 0.4594,
+      "step": 5310
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.8534507155418396,
+      "learning_rate": 8.595879120879121e-05,
+      "loss": 0.4496,
+      "step": 5320
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 1.6172966957092285,
+      "learning_rate": 8.593131868131868e-05,
+      "loss": 0.4751,
+      "step": 5330
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 1.378061056137085,
+      "learning_rate": 8.590384615384616e-05,
+      "loss": 0.4597,
+      "step": 5340
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.8239535093307495,
+      "learning_rate": 8.587637362637363e-05,
+      "loss": 0.4549,
+      "step": 5350
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.6672975420951843,
+      "learning_rate": 8.58489010989011e-05,
+      "loss": 0.4694,
+      "step": 5360
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 1.042051076889038,
+      "learning_rate": 8.582142857142858e-05,
+      "loss": 0.4667,
+      "step": 5370
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 1.6398990154266357,
+      "learning_rate": 8.579395604395605e-05,
+      "loss": 0.4657,
+      "step": 5380
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.5243840217590332,
+      "learning_rate": 8.576648351648352e-05,
+      "loss": 0.4733,
+      "step": 5390
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 1.3996922969818115,
+      "learning_rate": 8.5739010989011e-05,
+      "loss": 0.4747,
+      "step": 5400
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.7381965517997742,
+      "learning_rate": 8.571153846153847e-05,
+      "loss": 0.4694,
+      "step": 5410
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 1.1151881217956543,
+      "learning_rate": 8.568406593406593e-05,
+      "loss": 0.4658,
+      "step": 5420
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 1.0418914556503296,
+      "learning_rate": 8.565659340659342e-05,
+      "loss": 0.4568,
+      "step": 5430
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.7290151119232178,
+      "learning_rate": 8.562912087912088e-05,
+      "loss": 0.4642,
+      "step": 5440
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.8483603000640869,
+      "learning_rate": 8.560164835164835e-05,
+      "loss": 0.4912,
+      "step": 5450
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 1.1545953750610352,
+      "learning_rate": 8.557417582417583e-05,
+      "loss": 0.4738,
+      "step": 5460
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 1.0104776620864868,
+      "learning_rate": 8.55467032967033e-05,
+      "loss": 0.4751,
+      "step": 5470
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.9386792778968811,
+      "learning_rate": 8.551923076923077e-05,
+      "loss": 0.4549,
+      "step": 5480
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.6779844760894775,
+      "learning_rate": 8.549175824175825e-05,
+      "loss": 0.4682,
+      "step": 5490
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.7957237958908081,
+      "learning_rate": 8.546428571428572e-05,
+      "loss": 0.4561,
+      "step": 5500
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.8393499255180359,
+      "learning_rate": 8.543681318681319e-05,
+      "loss": 0.4659,
+      "step": 5510
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.9744943380355835,
+      "learning_rate": 8.540934065934067e-05,
+      "loss": 0.4545,
+      "step": 5520
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 1.1456732749938965,
+      "learning_rate": 8.538186813186814e-05,
+      "loss": 0.4614,
+      "step": 5530
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.8066684603691101,
+      "learning_rate": 8.53543956043956e-05,
+      "loss": 0.475,
+      "step": 5540
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 1.4929931163787842,
+      "learning_rate": 8.532692307692309e-05,
+      "loss": 0.4695,
+      "step": 5550
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.7223564386367798,
+      "learning_rate": 8.529945054945056e-05,
+      "loss": 0.4692,
+      "step": 5560
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 1.3222299814224243,
+      "learning_rate": 8.527197802197802e-05,
+      "loss": 0.4533,
+      "step": 5570
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 1.1262595653533936,
+      "learning_rate": 8.52445054945055e-05,
+      "loss": 0.4637,
+      "step": 5580
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.9097437858581543,
+      "learning_rate": 8.521703296703297e-05,
+      "loss": 0.4692,
+      "step": 5590
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 1.5121077299118042,
+      "learning_rate": 8.518956043956044e-05,
+      "loss": 0.4591,
+      "step": 5600
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 1.1432225704193115,
+      "learning_rate": 8.516208791208792e-05,
+      "loss": 0.4723,
+      "step": 5610
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 1.3745416402816772,
+      "learning_rate": 8.513461538461539e-05,
+      "loss": 0.4725,
+      "step": 5620
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.8508789539337158,
+      "learning_rate": 8.510714285714286e-05,
+      "loss": 0.4607,
+      "step": 5630
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 1.3097485303878784,
+      "learning_rate": 8.507967032967034e-05,
+      "loss": 0.4635,
+      "step": 5640
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 1.29444420337677,
+      "learning_rate": 8.505219780219781e-05,
+      "loss": 0.4628,
+      "step": 5650
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.5336117744445801,
+      "learning_rate": 8.502472527472528e-05,
+      "loss": 0.4568,
+      "step": 5660
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.7139732837677002,
+      "learning_rate": 8.499725274725276e-05,
+      "loss": 0.4574,
+      "step": 5670
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.7436287999153137,
+      "learning_rate": 8.496978021978023e-05,
+      "loss": 0.455,
+      "step": 5680
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.9436907172203064,
+      "learning_rate": 8.49423076923077e-05,
+      "loss": 0.4678,
+      "step": 5690
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.9827747344970703,
+      "learning_rate": 8.491483516483518e-05,
+      "loss": 0.4614,
+      "step": 5700
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.7332632541656494,
+      "learning_rate": 8.488736263736265e-05,
+      "loss": 0.4675,
+      "step": 5710
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 1.1933000087738037,
+      "learning_rate": 8.485989010989011e-05,
+      "loss": 0.4528,
+      "step": 5720
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 1.1396616697311401,
+      "learning_rate": 8.48324175824176e-05,
+      "loss": 0.4663,
+      "step": 5730
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 1.4335391521453857,
+      "learning_rate": 8.480494505494506e-05,
+      "loss": 0.4658,
+      "step": 5740
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.85484778881073,
+      "learning_rate": 8.477747252747253e-05,
+      "loss": 0.4536,
+      "step": 5750
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 1.206037163734436,
+      "learning_rate": 8.475000000000001e-05,
+      "loss": 0.4665,
+      "step": 5760
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 1.0052751302719116,
+      "learning_rate": 8.472252747252748e-05,
+      "loss": 0.4657,
+      "step": 5770
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.6977716088294983,
+      "learning_rate": 8.469505494505495e-05,
+      "loss": 0.4594,
+      "step": 5780
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.975903332233429,
+      "learning_rate": 8.466758241758243e-05,
+      "loss": 0.4739,
+      "step": 5790
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 1.0114597082138062,
+      "learning_rate": 8.46401098901099e-05,
+      "loss": 0.4732,
+      "step": 5800
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.5334266424179077,
+      "learning_rate": 8.461263736263737e-05,
+      "loss": 0.4585,
+      "step": 5810
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 1.6815012693405151,
+      "learning_rate": 8.458516483516485e-05,
+      "loss": 0.4737,
+      "step": 5820
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 1.1055243015289307,
+      "learning_rate": 8.455769230769232e-05,
+      "loss": 0.4608,
+      "step": 5830
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 1.3854832649230957,
+      "learning_rate": 8.453021978021979e-05,
+      "loss": 0.4659,
+      "step": 5840
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 1.0487163066864014,
+      "learning_rate": 8.450274725274727e-05,
+      "loss": 0.4519,
+      "step": 5850
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 1.0301164388656616,
+      "learning_rate": 8.447527472527474e-05,
+      "loss": 0.4598,
+      "step": 5860
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.7901486754417419,
+      "learning_rate": 8.44478021978022e-05,
+      "loss": 0.4727,
+      "step": 5870
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.6557927131652832,
+      "learning_rate": 8.442032967032969e-05,
+      "loss": 0.4523,
+      "step": 5880
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.8954731225967407,
+      "learning_rate": 8.439285714285715e-05,
+      "loss": 0.4672,
+      "step": 5890
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.6038153171539307,
+      "learning_rate": 8.436538461538462e-05,
+      "loss": 0.4604,
+      "step": 5900
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.9118047952651978,
+      "learning_rate": 8.433791208791209e-05,
+      "loss": 0.4592,
+      "step": 5910
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 1.125707745552063,
+      "learning_rate": 8.431043956043957e-05,
+      "loss": 0.4713,
+      "step": 5920
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.7669263482093811,
+      "learning_rate": 8.428296703296704e-05,
+      "loss": 0.465,
+      "step": 5930
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 1.0012739896774292,
+      "learning_rate": 8.425549450549451e-05,
+      "loss": 0.4661,
+      "step": 5940
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.650155782699585,
+      "learning_rate": 8.422802197802199e-05,
+      "loss": 0.4649,
+      "step": 5950
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 1.224027156829834,
+      "learning_rate": 8.420054945054946e-05,
+      "loss": 0.4486,
+      "step": 5960
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 1.350738763809204,
+      "learning_rate": 8.417307692307693e-05,
+      "loss": 0.4714,
+      "step": 5970
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 1.238620400428772,
+      "learning_rate": 8.414560439560441e-05,
+      "loss": 0.4644,
+      "step": 5980
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 1.4456909894943237,
+      "learning_rate": 8.411813186813188e-05,
+      "loss": 0.4674,
+      "step": 5990
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.9819515943527222,
+      "learning_rate": 8.409065934065934e-05,
+      "loss": 0.4681,
+      "step": 6000
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.7437455058097839,
+      "learning_rate": 8.406318681318683e-05,
+      "loss": 0.4701,
+      "step": 6010
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.7628225088119507,
+      "learning_rate": 8.403571428571428e-05,
+      "loss": 0.4608,
+      "step": 6020
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 1.0726768970489502,
+      "learning_rate": 8.400824175824176e-05,
+      "loss": 0.461,
+      "step": 6030
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.799857497215271,
+      "learning_rate": 8.398076923076923e-05,
+      "loss": 0.4587,
+      "step": 6040
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 1.242092251777649,
+      "learning_rate": 8.39532967032967e-05,
+      "loss": 0.4569,
+      "step": 6050
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.971057116985321,
+      "learning_rate": 8.392582417582418e-05,
+      "loss": 0.4537,
+      "step": 6060
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 1.1566013097763062,
+      "learning_rate": 8.389835164835165e-05,
+      "loss": 0.4638,
+      "step": 6070
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 1.2802190780639648,
+      "learning_rate": 8.387087912087912e-05,
+      "loss": 0.4663,
+      "step": 6080
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 1.0269036293029785,
+      "learning_rate": 8.38434065934066e-05,
+      "loss": 0.4628,
+      "step": 6090
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.7012946605682373,
+      "learning_rate": 8.381593406593407e-05,
+      "loss": 0.4513,
+      "step": 6100
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.5895891189575195,
+      "learning_rate": 8.378846153846153e-05,
+      "loss": 0.4584,
+      "step": 6110
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.9521509408950806,
+      "learning_rate": 8.376098901098902e-05,
+      "loss": 0.4587,
+      "step": 6120
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 1.409681797027588,
+      "learning_rate": 8.373351648351648e-05,
+      "loss": 0.4592,
+      "step": 6130
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 1.0833555459976196,
+      "learning_rate": 8.370604395604395e-05,
+      "loss": 0.4562,
+      "step": 6140
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.8994373679161072,
+      "learning_rate": 8.367857142857143e-05,
+      "loss": 0.4593,
+      "step": 6150
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 1.2591134309768677,
+      "learning_rate": 8.36510989010989e-05,
+      "loss": 0.4632,
+      "step": 6160
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.8163917064666748,
+      "learning_rate": 8.362362637362637e-05,
+      "loss": 0.4767,
+      "step": 6170
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.6897978186607361,
+      "learning_rate": 8.359615384615385e-05,
+      "loss": 0.4569,
+      "step": 6180
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 1.0527364015579224,
+      "learning_rate": 8.356868131868132e-05,
+      "loss": 0.4618,
+      "step": 6190
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 1.4118075370788574,
+      "learning_rate": 8.354120879120879e-05,
+      "loss": 0.4542,
+      "step": 6200
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.9567866325378418,
+      "learning_rate": 8.351373626373627e-05,
+      "loss": 0.4673,
+      "step": 6210
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 1.0207067728042603,
+      "learning_rate": 8.348626373626374e-05,
+      "loss": 0.4538,
+      "step": 6220
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.5897815227508545,
+      "learning_rate": 8.34587912087912e-05,
+      "loss": 0.4577,
+      "step": 6230
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.5948817133903503,
+      "learning_rate": 8.343131868131869e-05,
+      "loss": 0.4629,
+      "step": 6240
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 1.6685130596160889,
+      "learning_rate": 8.340384615384616e-05,
+      "loss": 0.461,
+      "step": 6250
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 1.502945899963379,
+      "learning_rate": 8.337637362637362e-05,
+      "loss": 0.4649,
+      "step": 6260
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 1.5422593355178833,
+      "learning_rate": 8.33489010989011e-05,
+      "loss": 0.4494,
+      "step": 6270
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.999934732913971,
+      "learning_rate": 8.332142857142857e-05,
+      "loss": 0.464,
+      "step": 6280
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.6580365896224976,
+      "learning_rate": 8.329395604395604e-05,
+      "loss": 0.4382,
+      "step": 6290
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.9826914668083191,
+      "learning_rate": 8.326648351648352e-05,
+      "loss": 0.4646,
+      "step": 6300
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 1.6006591320037842,
+      "learning_rate": 8.323901098901099e-05,
+      "loss": 0.4709,
+      "step": 6310
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.46454253792762756,
+      "learning_rate": 8.321153846153846e-05,
+      "loss": 0.4572,
+      "step": 6320
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 1.1660053730010986,
+      "learning_rate": 8.318406593406594e-05,
+      "loss": 0.4537,
+      "step": 6330
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.7951260805130005,
+      "learning_rate": 8.315659340659341e-05,
+      "loss": 0.4491,
+      "step": 6340
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.9338926672935486,
+      "learning_rate": 8.312912087912088e-05,
+      "loss": 0.4608,
+      "step": 6350
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 1.2365031242370605,
+      "learning_rate": 8.310164835164836e-05,
+      "loss": 0.4582,
+      "step": 6360
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.7614923119544983,
+      "learning_rate": 8.307417582417583e-05,
+      "loss": 0.4665,
+      "step": 6370
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.9391283392906189,
+      "learning_rate": 8.30467032967033e-05,
+      "loss": 0.4593,
+      "step": 6380
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 1.4633766412734985,
+      "learning_rate": 8.301923076923078e-05,
+      "loss": 0.4611,
+      "step": 6390
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 1.159623146057129,
+      "learning_rate": 8.299175824175825e-05,
+      "loss": 0.4622,
+      "step": 6400
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 1.0221785306930542,
+      "learning_rate": 8.296428571428571e-05,
+      "loss": 0.4483,
+      "step": 6410
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 1.5070372819900513,
+      "learning_rate": 8.29368131868132e-05,
+      "loss": 0.4677,
+      "step": 6420
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.9225922226905823,
+      "learning_rate": 8.290934065934066e-05,
+      "loss": 0.4711,
+      "step": 6430
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.9344289302825928,
+      "learning_rate": 8.288186813186813e-05,
+      "loss": 0.4591,
+      "step": 6440
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.5448782444000244,
+      "learning_rate": 8.285439560439561e-05,
+      "loss": 0.4635,
+      "step": 6450
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.465524822473526,
+      "learning_rate": 8.282692307692308e-05,
+      "loss": 0.4581,
+      "step": 6460
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 1.3583608865737915,
+      "learning_rate": 8.279945054945055e-05,
+      "loss": 0.4718,
+      "step": 6470
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 1.3173762559890747,
+      "learning_rate": 8.277197802197803e-05,
+      "loss": 0.4577,
+      "step": 6480
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 1.3845728635787964,
+      "learning_rate": 8.27445054945055e-05,
+      "loss": 0.4623,
+      "step": 6490
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 1.0992422103881836,
+      "learning_rate": 8.271703296703297e-05,
+      "loss": 0.4543,
+      "step": 6500
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.5556319355964661,
+      "learning_rate": 8.268956043956045e-05,
+      "loss": 0.4613,
+      "step": 6510
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.7095728516578674,
+      "learning_rate": 8.266208791208792e-05,
+      "loss": 0.455,
+      "step": 6520
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 1.1960381269454956,
+      "learning_rate": 8.263461538461539e-05,
+      "loss": 0.4545,
+      "step": 6530
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 1.0696423053741455,
+      "learning_rate": 8.260714285714287e-05,
+      "loss": 0.4666,
+      "step": 6540
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.5211924314498901,
+      "learning_rate": 8.257967032967034e-05,
+      "loss": 0.4535,
+      "step": 6550
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 1.1284767389297485,
+      "learning_rate": 8.25521978021978e-05,
+      "loss": 0.4633,
+      "step": 6560
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 1.3033208847045898,
+      "learning_rate": 8.252472527472529e-05,
+      "loss": 0.4812,
+      "step": 6570
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 1.2627853155136108,
+      "learning_rate": 8.249725274725275e-05,
+      "loss": 0.4676,
+      "step": 6580
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.5073243379592896,
+      "learning_rate": 8.246978021978022e-05,
+      "loss": 0.4638,
+      "step": 6590
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.6492868661880493,
+      "learning_rate": 8.24423076923077e-05,
+      "loss": 0.46,
+      "step": 6600
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.8350780010223389,
+      "learning_rate": 8.241483516483517e-05,
+      "loss": 0.4549,
+      "step": 6610
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 1.3048648834228516,
+      "learning_rate": 8.238736263736264e-05,
+      "loss": 0.47,
+      "step": 6620
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 1.2212404012680054,
+      "learning_rate": 8.235989010989012e-05,
+      "loss": 0.4623,
+      "step": 6630
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.9617910981178284,
+      "learning_rate": 8.233241758241759e-05,
+      "loss": 0.4566,
+      "step": 6640
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.5445905923843384,
+      "learning_rate": 8.230494505494506e-05,
+      "loss": 0.458,
+      "step": 6650
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.6512438654899597,
+      "learning_rate": 8.227747252747254e-05,
+      "loss": 0.4516,
+      "step": 6660
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 1.2203713655471802,
+      "learning_rate": 8.225000000000001e-05,
+      "loss": 0.4566,
+      "step": 6670
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.5803031325340271,
+      "learning_rate": 8.222252747252748e-05,
+      "loss": 0.4622,
+      "step": 6680
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.5457717180252075,
+      "learning_rate": 8.219505494505496e-05,
+      "loss": 0.4509,
+      "step": 6690
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.4855049252510071,
+      "learning_rate": 8.216758241758242e-05,
+      "loss": 0.4517,
+      "step": 6700
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.49980971217155457,
+      "learning_rate": 8.214010989010989e-05,
+      "loss": 0.4592,
+      "step": 6710
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.8323667049407959,
+      "learning_rate": 8.211263736263736e-05,
+      "loss": 0.4515,
+      "step": 6720
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.6058827638626099,
+      "learning_rate": 8.208516483516484e-05,
+      "loss": 0.4437,
+      "step": 6730
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 1.3227874040603638,
+      "learning_rate": 8.205769230769231e-05,
+      "loss": 0.4497,
+      "step": 6740
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 1.14075767993927,
+      "learning_rate": 8.203021978021978e-05,
+      "loss": 0.455,
+      "step": 6750
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 1.0633773803710938,
+      "learning_rate": 8.200274725274726e-05,
+      "loss": 0.4582,
+      "step": 6760
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.671051025390625,
+      "learning_rate": 8.197527472527473e-05,
+      "loss": 0.4584,
+      "step": 6770
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.8227337002754211,
+      "learning_rate": 8.19478021978022e-05,
+      "loss": 0.4564,
+      "step": 6780
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 1.3290925025939941,
+      "learning_rate": 8.192032967032968e-05,
+      "loss": 0.4527,
+      "step": 6790
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 1.160757303237915,
+      "learning_rate": 8.189285714285715e-05,
+      "loss": 0.4417,
+      "step": 6800
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.8233458399772644,
+      "learning_rate": 8.186538461538462e-05,
+      "loss": 0.4459,
+      "step": 6810
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.7575914859771729,
+      "learning_rate": 8.18379120879121e-05,
+      "loss": 0.4636,
+      "step": 6820
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.35609036684036255,
+      "learning_rate": 8.181043956043956e-05,
+      "loss": 0.4479,
+      "step": 6830
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 1.0687757730484009,
+      "learning_rate": 8.178296703296703e-05,
+      "loss": 0.4494,
+      "step": 6840
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.936469316482544,
+      "learning_rate": 8.175549450549451e-05,
+      "loss": 0.4605,
+      "step": 6850
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.6257254481315613,
+      "learning_rate": 8.172802197802198e-05,
+      "loss": 0.4559,
+      "step": 6860
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 1.0161064863204956,
+      "learning_rate": 8.170054945054945e-05,
+      "loss": 0.4677,
+      "step": 6870
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 1.0407649278640747,
+      "learning_rate": 8.167307692307693e-05,
+      "loss": 0.4542,
+      "step": 6880
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.7439064383506775,
+      "learning_rate": 8.16456043956044e-05,
+      "loss": 0.4747,
+      "step": 6890
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.37432214617729187,
+      "learning_rate": 8.161813186813187e-05,
+      "loss": 0.4595,
+      "step": 6900
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.9039118885993958,
+      "learning_rate": 8.159065934065935e-05,
+      "loss": 0.4701,
+      "step": 6910
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.5294177532196045,
+      "learning_rate": 8.156318681318682e-05,
+      "loss": 0.4533,
+      "step": 6920
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.8023412227630615,
+      "learning_rate": 8.153571428571429e-05,
+      "loss": 0.4632,
+      "step": 6930
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 1.1089890003204346,
+      "learning_rate": 8.150824175824177e-05,
+      "loss": 0.4627,
+      "step": 6940
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.5587044954299927,
+      "learning_rate": 8.148076923076924e-05,
+      "loss": 0.4613,
+      "step": 6950
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.7494495511054993,
+      "learning_rate": 8.14532967032967e-05,
+      "loss": 0.4478,
+      "step": 6960
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.7425660490989685,
+      "learning_rate": 8.142582417582419e-05,
+      "loss": 0.4585,
+      "step": 6970
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.9916179180145264,
+      "learning_rate": 8.139835164835165e-05,
+      "loss": 0.464,
+      "step": 6980
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.9130895137786865,
+      "learning_rate": 8.137087912087912e-05,
+      "loss": 0.4628,
+      "step": 6990
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.7386452555656433,
+      "learning_rate": 8.13434065934066e-05,
+      "loss": 0.457,
+      "step": 7000
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 1.5726503133773804,
+      "learning_rate": 8.131593406593407e-05,
+      "loss": 0.461,
+      "step": 7010
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.6793839335441589,
+      "learning_rate": 8.128846153846154e-05,
+      "loss": 0.4521,
+      "step": 7020
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 1.5302358865737915,
+      "learning_rate": 8.126098901098902e-05,
+      "loss": 0.4598,
+      "step": 7030
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.9924741387367249,
+      "learning_rate": 8.123351648351649e-05,
+      "loss": 0.4599,
+      "step": 7040
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.7319537997245789,
+      "learning_rate": 8.120604395604396e-05,
+      "loss": 0.4585,
+      "step": 7050
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 1.067428469657898,
+      "learning_rate": 8.117857142857144e-05,
+      "loss": 0.4509,
+      "step": 7060
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 1.2225853204727173,
+      "learning_rate": 8.115109890109891e-05,
+      "loss": 0.4591,
+      "step": 7070
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.7233381867408752,
+      "learning_rate": 8.112362637362638e-05,
+      "loss": 0.4568,
+      "step": 7080
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 1.3318099975585938,
+      "learning_rate": 8.109615384615386e-05,
+      "loss": 0.4715,
+      "step": 7090
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 1.4362515211105347,
+      "learning_rate": 8.106868131868133e-05,
+      "loss": 0.4539,
+      "step": 7100
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.8498636484146118,
+      "learning_rate": 8.10412087912088e-05,
+      "loss": 0.4531,
+      "step": 7110
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.5961604118347168,
+      "learning_rate": 8.101373626373628e-05,
+      "loss": 0.4561,
+      "step": 7120
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 1.013180136680603,
+      "learning_rate": 8.098626373626374e-05,
+      "loss": 0.4581,
+      "step": 7130
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.9159004092216492,
+      "learning_rate": 8.095879120879121e-05,
+      "loss": 0.4498,
+      "step": 7140
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.9847846031188965,
+      "learning_rate": 8.093131868131868e-05,
+      "loss": 0.4584,
+      "step": 7150
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 1.5007630586624146,
+      "learning_rate": 8.090384615384615e-05,
+      "loss": 0.4524,
+      "step": 7160
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.8937559723854065,
+      "learning_rate": 8.087637362637363e-05,
+      "loss": 0.4701,
+      "step": 7170
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.9551745653152466,
+      "learning_rate": 8.08489010989011e-05,
+      "loss": 0.4412,
+      "step": 7180
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.7207891345024109,
+      "learning_rate": 8.082142857142857e-05,
+      "loss": 0.4577,
+      "step": 7190
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.5330765843391418,
+      "learning_rate": 8.079395604395605e-05,
+      "loss": 0.4531,
+      "step": 7200
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 1.0517854690551758,
+      "learning_rate": 8.076648351648352e-05,
+      "loss": 0.4339,
+      "step": 7210
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 1.2942813634872437,
+      "learning_rate": 8.073901098901098e-05,
+      "loss": 0.4575,
+      "step": 7220
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.9348652362823486,
+      "learning_rate": 8.071153846153847e-05,
+      "loss": 0.4472,
+      "step": 7230
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 1.095439076423645,
+      "learning_rate": 8.068406593406593e-05,
+      "loss": 0.4503,
+      "step": 7240
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.5965198278427124,
+      "learning_rate": 8.06565934065934e-05,
+      "loss": 0.4499,
+      "step": 7250
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.5533403158187866,
+      "learning_rate": 8.062912087912088e-05,
+      "loss": 0.4676,
+      "step": 7260
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 1.1052013635635376,
+      "learning_rate": 8.060164835164835e-05,
+      "loss": 0.4635,
+      "step": 7270
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.9924907088279724,
+      "learning_rate": 8.057417582417582e-05,
+      "loss": 0.4606,
+      "step": 7280
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.5882344245910645,
+      "learning_rate": 8.05467032967033e-05,
+      "loss": 0.4456,
+      "step": 7290
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 1.0022166967391968,
+      "learning_rate": 8.051923076923077e-05,
+      "loss": 0.4505,
+      "step": 7300
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.8352431058883667,
+      "learning_rate": 8.049175824175824e-05,
+      "loss": 0.4567,
+      "step": 7310
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 1.0245552062988281,
+      "learning_rate": 8.046428571428572e-05,
+      "loss": 0.4509,
+      "step": 7320
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 0.44436022639274597,
+      "eval_runtime": 1246.3805,
+      "eval_samples_per_second": 234.366,
+      "eval_steps_per_second": 3.663,
+      "step": 7321
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.6890380382537842,
+      "learning_rate": 8.043681318681319e-05,
+      "loss": 0.4559,
+      "step": 7330
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.8071466088294983,
+      "learning_rate": 8.040934065934066e-05,
+      "loss": 0.4494,
+      "step": 7340
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.5615882873535156,
+      "learning_rate": 8.038186813186814e-05,
+      "loss": 0.444,
+      "step": 7350
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 1.312966227531433,
+      "learning_rate": 8.03543956043956e-05,
+      "loss": 0.4549,
+      "step": 7360
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.7778192162513733,
+      "learning_rate": 8.032692307692307e-05,
+      "loss": 0.4561,
+      "step": 7370
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.9343681335449219,
+      "learning_rate": 8.029945054945056e-05,
+      "loss": 0.4565,
+      "step": 7380
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.585110068321228,
+      "learning_rate": 8.027197802197802e-05,
+      "loss": 0.4502,
+      "step": 7390
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 1.0197051763534546,
+      "learning_rate": 8.024450549450549e-05,
+      "loss": 0.4595,
+      "step": 7400
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 1.0265450477600098,
+      "learning_rate": 8.021703296703297e-05,
+      "loss": 0.4425,
+      "step": 7410
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 1.1836851835250854,
+      "learning_rate": 8.018956043956044e-05,
+      "loss": 0.4606,
+      "step": 7420
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.9758961796760559,
+      "learning_rate": 8.016208791208791e-05,
+      "loss": 0.4494,
+      "step": 7430
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 1.0412979125976562,
+      "learning_rate": 8.013461538461539e-05,
+      "loss": 0.4593,
+      "step": 7440
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 1.13507878780365,
+      "learning_rate": 8.010714285714286e-05,
+      "loss": 0.4598,
+      "step": 7450
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 1.118143081665039,
+      "learning_rate": 8.007967032967033e-05,
+      "loss": 0.4597,
+      "step": 7460
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.6100110411643982,
+      "learning_rate": 8.005219780219781e-05,
+      "loss": 0.46,
+      "step": 7470
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 1.4135487079620361,
+      "learning_rate": 8.002472527472528e-05,
+      "loss": 0.455,
+      "step": 7480
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.7120094299316406,
+      "learning_rate": 7.999725274725275e-05,
+      "loss": 0.4459,
+      "step": 7490
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.7432197332382202,
+      "learning_rate": 7.996978021978023e-05,
+      "loss": 0.4509,
+      "step": 7500
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 1.1492975950241089,
+      "learning_rate": 7.99423076923077e-05,
+      "loss": 0.4638,
+      "step": 7510
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 1.0090208053588867,
+      "learning_rate": 7.991483516483516e-05,
+      "loss": 0.4702,
+      "step": 7520
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 1.09323251247406,
+      "learning_rate": 7.988736263736263e-05,
+      "loss": 0.4518,
+      "step": 7530
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 1.0353306531906128,
+      "learning_rate": 7.985989010989011e-05,
+      "loss": 0.4522,
+      "step": 7540
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.8499657511711121,
+      "learning_rate": 7.983241758241758e-05,
+      "loss": 0.4637,
+      "step": 7550
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 1.0327955484390259,
+      "learning_rate": 7.980494505494505e-05,
+      "loss": 0.4444,
+      "step": 7560
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 1.239608883857727,
+      "learning_rate": 7.977747252747253e-05,
+      "loss": 0.4536,
+      "step": 7570
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.6841428279876709,
+      "learning_rate": 7.975e-05,
+      "loss": 0.4519,
+      "step": 7580
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 1.1058799028396606,
+      "learning_rate": 7.972252747252747e-05,
+      "loss": 0.4511,
+      "step": 7590
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 1.1671322584152222,
+      "learning_rate": 7.969505494505495e-05,
+      "loss": 0.4512,
+      "step": 7600
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.6706550121307373,
+      "learning_rate": 7.966758241758242e-05,
+      "loss": 0.4493,
+      "step": 7610
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.752795398235321,
+      "learning_rate": 7.964010989010989e-05,
+      "loss": 0.4494,
+      "step": 7620
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.6255919933319092,
+      "learning_rate": 7.961263736263737e-05,
+      "loss": 0.4525,
+      "step": 7630
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.4543440341949463,
+      "learning_rate": 7.958516483516484e-05,
+      "loss": 0.4527,
+      "step": 7640
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.6038147807121277,
+      "learning_rate": 7.95576923076923e-05,
+      "loss": 0.4431,
+      "step": 7650
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 1.010972261428833,
+      "learning_rate": 7.953021978021979e-05,
+      "loss": 0.451,
+      "step": 7660
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 1.0627601146697998,
+      "learning_rate": 7.950274725274725e-05,
+      "loss": 0.4441,
+      "step": 7670
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.6361072063446045,
+      "learning_rate": 7.947527472527472e-05,
+      "loss": 0.4592,
+      "step": 7680
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.897415280342102,
+      "learning_rate": 7.94478021978022e-05,
+      "loss": 0.4535,
+      "step": 7690
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.6545829772949219,
+      "learning_rate": 7.942032967032967e-05,
+      "loss": 0.4589,
+      "step": 7700
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.9311442375183105,
+      "learning_rate": 7.939285714285714e-05,
+      "loss": 0.4417,
+      "step": 7710
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 1.5107172727584839,
+      "learning_rate": 7.936538461538462e-05,
+      "loss": 0.4503,
+      "step": 7720
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.9508922100067139,
+      "learning_rate": 7.933791208791209e-05,
+      "loss": 0.4532,
+      "step": 7730
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.8631314039230347,
+      "learning_rate": 7.931043956043956e-05,
+      "loss": 0.4551,
+      "step": 7740
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.639757513999939,
+      "learning_rate": 7.928296703296704e-05,
+      "loss": 0.4531,
+      "step": 7750
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.8901143074035645,
+      "learning_rate": 7.925549450549451e-05,
+      "loss": 0.4563,
+      "step": 7760
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 1.7677491903305054,
+      "learning_rate": 7.922802197802198e-05,
+      "loss": 0.4467,
+      "step": 7770
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.9918643236160278,
+      "learning_rate": 7.920054945054946e-05,
+      "loss": 0.4604,
+      "step": 7780
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.7385686635971069,
+      "learning_rate": 7.917307692307693e-05,
+      "loss": 0.4484,
+      "step": 7790
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.4740804433822632,
+      "learning_rate": 7.91456043956044e-05,
+      "loss": 0.4581,
+      "step": 7800
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.3722800016403198,
+      "learning_rate": 7.911813186813188e-05,
+      "loss": 0.4437,
+      "step": 7810
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.5825007557868958,
+      "learning_rate": 7.909065934065934e-05,
+      "loss": 0.4495,
+      "step": 7820
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.7577399015426636,
+      "learning_rate": 7.906318681318681e-05,
+      "loss": 0.4507,
+      "step": 7830
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.5723931193351746,
+      "learning_rate": 7.90357142857143e-05,
+      "loss": 0.4512,
+      "step": 7840
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 1.0552576780319214,
+      "learning_rate": 7.900824175824176e-05,
+      "loss": 0.4493,
+      "step": 7850
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 1.1965206861495972,
+      "learning_rate": 7.898076923076923e-05,
+      "loss": 0.4502,
+      "step": 7860
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.7538771629333496,
+      "learning_rate": 7.895329670329671e-05,
+      "loss": 0.4576,
+      "step": 7870
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 1.2679303884506226,
+      "learning_rate": 7.892582417582418e-05,
+      "loss": 0.4464,
+      "step": 7880
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.8973550200462341,
+      "learning_rate": 7.889835164835165e-05,
+      "loss": 0.4596,
+      "step": 7890
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.7081739902496338,
+      "learning_rate": 7.887087912087913e-05,
+      "loss": 0.4368,
+      "step": 7900
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.7710354328155518,
+      "learning_rate": 7.88434065934066e-05,
+      "loss": 0.4463,
+      "step": 7910
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.6749390363693237,
+      "learning_rate": 7.881593406593407e-05,
+      "loss": 0.4515,
+      "step": 7920
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.6680644154548645,
+      "learning_rate": 7.878846153846155e-05,
+      "loss": 0.4433,
+      "step": 7930
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 1.5634280443191528,
+      "learning_rate": 7.876098901098902e-05,
+      "loss": 0.4702,
+      "step": 7940
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.6959145665168762,
+      "learning_rate": 7.873351648351648e-05,
+      "loss": 0.4504,
+      "step": 7950
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.7739858627319336,
+      "learning_rate": 7.870604395604397e-05,
+      "loss": 0.4606,
+      "step": 7960
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.5519704818725586,
+      "learning_rate": 7.867857142857143e-05,
+      "loss": 0.4658,
+      "step": 7970
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.5425352454185486,
+      "learning_rate": 7.86510989010989e-05,
+      "loss": 0.4445,
+      "step": 7980
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.8816283941268921,
+      "learning_rate": 7.862362637362638e-05,
+      "loss": 0.4621,
+      "step": 7990
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 1.0816186666488647,
+      "learning_rate": 7.859615384615385e-05,
+      "loss": 0.4622,
+      "step": 8000
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.8483924865722656,
+      "learning_rate": 7.856868131868132e-05,
+      "loss": 0.4533,
+      "step": 8010
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.45229572057724,
+      "learning_rate": 7.85412087912088e-05,
+      "loss": 0.4626,
+      "step": 8020
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.6593546867370605,
+      "learning_rate": 7.851373626373627e-05,
+      "loss": 0.4498,
+      "step": 8030
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 1.214553952217102,
+      "learning_rate": 7.848626373626374e-05,
+      "loss": 0.4497,
+      "step": 8040
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.7151647210121155,
+      "learning_rate": 7.845879120879122e-05,
+      "loss": 0.4447,
+      "step": 8050
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.7126444578170776,
+      "learning_rate": 7.843131868131869e-05,
+      "loss": 0.4622,
+      "step": 8060
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.8902627229690552,
+      "learning_rate": 7.840384615384616e-05,
+      "loss": 0.4439,
+      "step": 8070
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.7599775195121765,
+      "learning_rate": 7.837637362637364e-05,
+      "loss": 0.4537,
+      "step": 8080
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.9020054340362549,
+      "learning_rate": 7.83489010989011e-05,
+      "loss": 0.4658,
+      "step": 8090
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.8659080266952515,
+      "learning_rate": 7.832142857142857e-05,
+      "loss": 0.4553,
+      "step": 8100
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.6685529351234436,
+      "learning_rate": 7.829395604395606e-05,
+      "loss": 0.4604,
+      "step": 8110
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.6612065434455872,
+      "learning_rate": 7.826648351648352e-05,
+      "loss": 0.4522,
+      "step": 8120
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.7674763202667236,
+      "learning_rate": 7.823901098901099e-05,
+      "loss": 0.4523,
+      "step": 8130
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.5680274963378906,
+      "learning_rate": 7.821153846153847e-05,
+      "loss": 0.4649,
+      "step": 8140
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 1.0545960664749146,
+      "learning_rate": 7.818406593406594e-05,
+      "loss": 0.4507,
+      "step": 8150
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.5723569393157959,
+      "learning_rate": 7.815659340659341e-05,
+      "loss": 0.4418,
+      "step": 8160
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.9119620323181152,
+      "learning_rate": 7.812912087912089e-05,
+      "loss": 0.4535,
+      "step": 8170
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 1.0251758098602295,
+      "learning_rate": 7.810164835164836e-05,
+      "loss": 0.4503,
+      "step": 8180
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.6839396357536316,
+      "learning_rate": 7.807417582417583e-05,
+      "loss": 0.452,
+      "step": 8190
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 1.3468687534332275,
+      "learning_rate": 7.804670329670331e-05,
+      "loss": 0.4442,
+      "step": 8200
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.9562901854515076,
+      "learning_rate": 7.801923076923078e-05,
+      "loss": 0.4596,
+      "step": 8210
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 1.092677354812622,
+      "learning_rate": 7.799175824175825e-05,
+      "loss": 0.4454,
+      "step": 8220
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.7029426097869873,
+      "learning_rate": 7.796428571428573e-05,
+      "loss": 0.4463,
+      "step": 8230
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.7543437480926514,
+      "learning_rate": 7.79368131868132e-05,
+      "loss": 0.4588,
+      "step": 8240
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 1.5089532136917114,
+      "learning_rate": 7.790934065934066e-05,
+      "loss": 0.4531,
+      "step": 8250
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.6691949963569641,
+      "learning_rate": 7.788186813186815e-05,
+      "loss": 0.4494,
+      "step": 8260
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 1.0210888385772705,
+      "learning_rate": 7.785439560439561e-05,
+      "loss": 0.4548,
+      "step": 8270
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.9251496195793152,
+      "learning_rate": 7.782692307692308e-05,
+      "loss": 0.4576,
+      "step": 8280
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 1.2618598937988281,
+      "learning_rate": 7.779945054945055e-05,
+      "loss": 0.4478,
+      "step": 8290
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 1.2252585887908936,
+      "learning_rate": 7.777197802197802e-05,
+      "loss": 0.4573,
+      "step": 8300
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 0.6124866604804993,
+      "learning_rate": 7.77445054945055e-05,
+      "loss": 0.4506,
+      "step": 8310
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 1.0547674894332886,
+      "learning_rate": 7.771703296703297e-05,
+      "loss": 0.4418,
+      "step": 8320
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 1.1824958324432373,
+      "learning_rate": 7.768956043956044e-05,
+      "loss": 0.4529,
+      "step": 8330
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.9790034890174866,
+      "learning_rate": 7.76620879120879e-05,
+      "loss": 0.455,
+      "step": 8340
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.9925460815429688,
+      "learning_rate": 7.763461538461539e-05,
+      "loss": 0.4607,
+      "step": 8350
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.8586925864219666,
+      "learning_rate": 7.760714285714285e-05,
+      "loss": 0.4644,
+      "step": 8360
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 1.4445006847381592,
+      "learning_rate": 7.757967032967032e-05,
+      "loss": 0.4513,
+      "step": 8370
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.9150937795639038,
+      "learning_rate": 7.75521978021978e-05,
+      "loss": 0.4534,
+      "step": 8380
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.9148111939430237,
+      "learning_rate": 7.752472527472527e-05,
+      "loss": 0.4424,
+      "step": 8390
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.9973462224006653,
+      "learning_rate": 7.749725274725274e-05,
+      "loss": 0.4523,
+      "step": 8400
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 1.022926926612854,
+      "learning_rate": 7.746978021978022e-05,
+      "loss": 0.444,
+      "step": 8410
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.7249711155891418,
+      "learning_rate": 7.744230769230769e-05,
+      "loss": 0.4473,
+      "step": 8420
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.44306471943855286,
+      "learning_rate": 7.741483516483516e-05,
+      "loss": 0.4452,
+      "step": 8430
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.9625614881515503,
+      "learning_rate": 7.738736263736264e-05,
+      "loss": 0.4373,
+      "step": 8440
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.5116167664527893,
+      "learning_rate": 7.735989010989011e-05,
+      "loss": 0.4505,
+      "step": 8450
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.7496402859687805,
+      "learning_rate": 7.733241758241758e-05,
+      "loss": 0.4451,
+      "step": 8460
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 1.3204615116119385,
+      "learning_rate": 7.730494505494506e-05,
+      "loss": 0.4576,
+      "step": 8470
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.7599005699157715,
+      "learning_rate": 7.727747252747253e-05,
+      "loss": 0.4493,
+      "step": 8480
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.5699270367622375,
+      "learning_rate": 7.725e-05,
+      "loss": 0.4568,
+      "step": 8490
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.6626598238945007,
+      "learning_rate": 7.722252747252748e-05,
+      "loss": 0.4504,
+      "step": 8500
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.6977862119674683,
+      "learning_rate": 7.719505494505494e-05,
+      "loss": 0.4529,
+      "step": 8510
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.883980393409729,
+      "learning_rate": 7.716758241758241e-05,
+      "loss": 0.4527,
+      "step": 8520
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.9930477738380432,
+      "learning_rate": 7.71401098901099e-05,
+      "loss": 0.4524,
+      "step": 8530
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.5075715780258179,
+      "learning_rate": 7.711263736263736e-05,
+      "loss": 0.4436,
+      "step": 8540
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.7683893442153931,
+      "learning_rate": 7.708516483516483e-05,
+      "loss": 0.4608,
+      "step": 8550
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 1.6546239852905273,
+      "learning_rate": 7.705769230769231e-05,
+      "loss": 0.4521,
+      "step": 8560
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.8132153153419495,
+      "learning_rate": 7.703021978021978e-05,
+      "loss": 0.4542,
+      "step": 8570
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.5142685770988464,
+      "learning_rate": 7.700274725274725e-05,
+      "loss": 0.4548,
+      "step": 8580
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 1.100368618965149,
+      "learning_rate": 7.697527472527473e-05,
+      "loss": 0.4602,
+      "step": 8590
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 1.018351674079895,
+      "learning_rate": 7.69478021978022e-05,
+      "loss": 0.4424,
+      "step": 8600
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.557134211063385,
+      "learning_rate": 7.692032967032967e-05,
+      "loss": 0.4486,
+      "step": 8610
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.701932966709137,
+      "learning_rate": 7.689285714285715e-05,
+      "loss": 0.4543,
+      "step": 8620
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 1.0351616144180298,
+      "learning_rate": 7.686538461538462e-05,
+      "loss": 0.4628,
+      "step": 8630
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.5028539896011353,
+      "learning_rate": 7.683791208791208e-05,
+      "loss": 0.4566,
+      "step": 8640
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.7769728302955627,
+      "learning_rate": 7.681043956043957e-05,
+      "loss": 0.4429,
+      "step": 8650
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.7966272830963135,
+      "learning_rate": 7.678296703296703e-05,
+      "loss": 0.4377,
+      "step": 8660
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 1.1576814651489258,
+      "learning_rate": 7.67554945054945e-05,
+      "loss": 0.4507,
+      "step": 8670
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.8654575943946838,
+      "learning_rate": 7.672802197802198e-05,
+      "loss": 0.4598,
+      "step": 8680
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.7794680595397949,
+      "learning_rate": 7.670054945054945e-05,
+      "loss": 0.4424,
+      "step": 8690
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.8271937370300293,
+      "learning_rate": 7.667307692307692e-05,
+      "loss": 0.4586,
+      "step": 8700
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.8499754071235657,
+      "learning_rate": 7.66456043956044e-05,
+      "loss": 0.4476,
+      "step": 8710
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.7553250789642334,
+      "learning_rate": 7.661813186813187e-05,
+      "loss": 0.4435,
+      "step": 8720
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.5660779476165771,
+      "learning_rate": 7.659065934065934e-05,
+      "loss": 0.4518,
+      "step": 8730
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 1.4875767230987549,
+      "learning_rate": 7.656318681318682e-05,
+      "loss": 0.4542,
+      "step": 8740
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 1.1859349012374878,
+      "learning_rate": 7.653846153846153e-05,
+      "loss": 0.4528,
+      "step": 8750
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 0.6259100437164307,
+      "learning_rate": 7.651098901098902e-05,
+      "loss": 0.4629,
+      "step": 8760
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.4763346314430237,
+      "learning_rate": 7.648351648351648e-05,
+      "loss": 0.4479,
+      "step": 8770
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.6478772759437561,
+      "learning_rate": 7.645604395604395e-05,
+      "loss": 0.4313,
+      "step": 8780
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 1.1578466892242432,
+      "learning_rate": 7.642857142857143e-05,
+      "loss": 0.4499,
+      "step": 8790
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.5163312554359436,
+      "learning_rate": 7.64010989010989e-05,
+      "loss": 0.4479,
+      "step": 8800
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 1.183361530303955,
+      "learning_rate": 7.637362637362637e-05,
+      "loss": 0.4524,
+      "step": 8810
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.6872965097427368,
+      "learning_rate": 7.634615384615385e-05,
+      "loss": 0.4445,
+      "step": 8820
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.7339246273040771,
+      "learning_rate": 7.631868131868132e-05,
+      "loss": 0.4446,
+      "step": 8830
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.7101340889930725,
+      "learning_rate": 7.629120879120879e-05,
+      "loss": 0.45,
+      "step": 8840
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.5717099905014038,
+      "learning_rate": 7.626373626373627e-05,
+      "loss": 0.4503,
+      "step": 8850
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.6713473796844482,
+      "learning_rate": 7.623626373626374e-05,
+      "loss": 0.4528,
+      "step": 8860
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.9480137228965759,
+      "learning_rate": 7.62087912087912e-05,
+      "loss": 0.4569,
+      "step": 8870
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.594673752784729,
+      "learning_rate": 7.618131868131869e-05,
+      "loss": 0.4627,
+      "step": 8880
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.8691803216934204,
+      "learning_rate": 7.615384615384616e-05,
+      "loss": 0.4459,
+      "step": 8890
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.8882876634597778,
+      "learning_rate": 7.612637362637362e-05,
+      "loss": 0.4478,
+      "step": 8900
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.48794516921043396,
+      "learning_rate": 7.60989010989011e-05,
+      "loss": 0.4479,
+      "step": 8910
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.4966289699077606,
+      "learning_rate": 7.607142857142857e-05,
+      "loss": 0.4395,
+      "step": 8920
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.792293131351471,
+      "learning_rate": 7.604395604395604e-05,
+      "loss": 0.4353,
+      "step": 8930
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.6615070104598999,
+      "learning_rate": 7.601648351648351e-05,
+      "loss": 0.4478,
+      "step": 8940
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.8686419725418091,
+      "learning_rate": 7.598901098901099e-05,
+      "loss": 0.4516,
+      "step": 8950
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 1.2833001613616943,
+      "learning_rate": 7.596153846153846e-05,
+      "loss": 0.4526,
+      "step": 8960
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.7414630055427551,
+      "learning_rate": 7.593406593406593e-05,
+      "loss": 0.4549,
+      "step": 8970
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.9745138883590698,
+      "learning_rate": 7.590659340659341e-05,
+      "loss": 0.4412,
+      "step": 8980
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.8407401442527771,
+      "learning_rate": 7.587912087912088e-05,
+      "loss": 0.4511,
+      "step": 8990
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 1.1383010149002075,
+      "learning_rate": 7.585164835164835e-05,
+      "loss": 0.4556,
+      "step": 9000
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 1.0419915914535522,
+      "learning_rate": 7.582417582417583e-05,
+      "loss": 0.4504,
+      "step": 9010
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.931141197681427,
+      "learning_rate": 7.57967032967033e-05,
+      "loss": 0.4464,
+      "step": 9020
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.6264098286628723,
+      "learning_rate": 7.576923076923076e-05,
+      "loss": 0.4627,
+      "step": 9030
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.8912045359611511,
+      "learning_rate": 7.574175824175825e-05,
+      "loss": 0.4452,
+      "step": 9040
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 1.3010141849517822,
+      "learning_rate": 7.571428571428571e-05,
+      "loss": 0.4595,
+      "step": 9050
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.8275282979011536,
+      "learning_rate": 7.568681318681318e-05,
+      "loss": 0.4486,
+      "step": 9060
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.8537049293518066,
+      "learning_rate": 7.565934065934066e-05,
+      "loss": 0.4506,
+      "step": 9070
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.5747140645980835,
+      "learning_rate": 7.563186813186813e-05,
+      "loss": 0.4488,
+      "step": 9080
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.8508340716362,
+      "learning_rate": 7.56043956043956e-05,
+      "loss": 0.4436,
+      "step": 9090
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 1.1689362525939941,
+      "learning_rate": 7.557692307692308e-05,
+      "loss": 0.4586,
+      "step": 9100
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 1.1048357486724854,
+      "learning_rate": 7.554945054945055e-05,
+      "loss": 0.4568,
+      "step": 9110
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 1.1694691181182861,
+      "learning_rate": 7.552197802197802e-05,
+      "loss": 0.4571,
+      "step": 9120
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.9257465600967407,
+      "learning_rate": 7.54945054945055e-05,
+      "loss": 0.4544,
+      "step": 9130
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 1.3826684951782227,
+      "learning_rate": 7.546703296703297e-05,
+      "loss": 0.44,
+      "step": 9140
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.8815349340438843,
+      "learning_rate": 7.543956043956044e-05,
+      "loss": 0.4438,
+      "step": 9150
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.661222517490387,
+      "learning_rate": 7.541208791208792e-05,
+      "loss": 0.4479,
+      "step": 9160
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 1.222886085510254,
+      "learning_rate": 7.538461538461539e-05,
+      "loss": 0.4487,
+      "step": 9170
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 1.1454687118530273,
+      "learning_rate": 7.535714285714285e-05,
+      "loss": 0.4625,
+      "step": 9180
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.49560263752937317,
+      "learning_rate": 7.532967032967034e-05,
+      "loss": 0.4518,
+      "step": 9190
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.8997675776481628,
+      "learning_rate": 7.53021978021978e-05,
+      "loss": 0.4426,
+      "step": 9200
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.9074673056602478,
+      "learning_rate": 7.527472527472527e-05,
+      "loss": 0.4524,
+      "step": 9210
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.7328269481658936,
+      "learning_rate": 7.524725274725275e-05,
+      "loss": 0.4512,
+      "step": 9220
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.7607899904251099,
+      "learning_rate": 7.521978021978022e-05,
+      "loss": 0.4491,
+      "step": 9230
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.5627162456512451,
+      "learning_rate": 7.519230769230769e-05,
+      "loss": 0.4511,
+      "step": 9240
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.9358956813812256,
+      "learning_rate": 7.516483516483517e-05,
+      "loss": 0.453,
+      "step": 9250
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 1.2619479894638062,
+      "learning_rate": 7.513736263736264e-05,
+      "loss": 0.4432,
+      "step": 9260
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.9375318884849548,
+      "learning_rate": 7.510989010989011e-05,
+      "loss": 0.4539,
+      "step": 9270
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.8454050421714783,
+      "learning_rate": 7.508241758241759e-05,
+      "loss": 0.4515,
+      "step": 9280
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.9856541156768799,
+      "learning_rate": 7.505494505494506e-05,
+      "loss": 0.445,
+      "step": 9290
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.6597672700881958,
+      "learning_rate": 7.502747252747253e-05,
+      "loss": 0.4505,
+      "step": 9300
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 1.0249258279800415,
+      "learning_rate": 7.500000000000001e-05,
+      "loss": 0.4516,
+      "step": 9310
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 1.003083348274231,
+      "learning_rate": 7.497252747252748e-05,
+      "loss": 0.4457,
+      "step": 9320
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 1.236377477645874,
+      "learning_rate": 7.494505494505494e-05,
+      "loss": 0.4573,
+      "step": 9330
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 1.5360130071640015,
+      "learning_rate": 7.491758241758242e-05,
+      "loss": 0.4478,
+      "step": 9340
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.9379085898399353,
+      "learning_rate": 7.489010989010989e-05,
+      "loss": 0.4575,
+      "step": 9350
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.7734600901603699,
+      "learning_rate": 7.486263736263736e-05,
+      "loss": 0.4571,
+      "step": 9360
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.6792673468589783,
+      "learning_rate": 7.483516483516484e-05,
+      "loss": 0.4522,
+      "step": 9370
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.8936319351196289,
+      "learning_rate": 7.480769230769231e-05,
+      "loss": 0.4576,
+      "step": 9380
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.7541890144348145,
+      "learning_rate": 7.478021978021978e-05,
+      "loss": 0.4456,
+      "step": 9390
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.8661054968833923,
+      "learning_rate": 7.475274725274726e-05,
+      "loss": 0.4428,
+      "step": 9400
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.919201672077179,
+      "learning_rate": 7.472527472527473e-05,
+      "loss": 0.4507,
+      "step": 9410
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 1.0688129663467407,
+      "learning_rate": 7.46978021978022e-05,
+      "loss": 0.4469,
+      "step": 9420
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 7.527632236480713,
+      "learning_rate": 7.467032967032968e-05,
+      "loss": 0.4512,
+      "step": 9430
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.5668841600418091,
+      "learning_rate": 7.464285714285715e-05,
+      "loss": 0.4484,
+      "step": 9440
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.7174908518791199,
+      "learning_rate": 7.461538461538462e-05,
+      "loss": 0.4361,
+      "step": 9450
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.693144679069519,
+      "learning_rate": 7.45879120879121e-05,
+      "loss": 0.4483,
+      "step": 9460
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 1.2530773878097534,
+      "learning_rate": 7.456043956043956e-05,
+      "loss": 0.4495,
+      "step": 9470
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.7528218626976013,
+      "learning_rate": 7.453296703296703e-05,
+      "loss": 0.4429,
+      "step": 9480
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.596477746963501,
+      "learning_rate": 7.450549450549451e-05,
+      "loss": 0.4526,
+      "step": 9490
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.5093569755554199,
+      "learning_rate": 7.447802197802198e-05,
+      "loss": 0.4419,
+      "step": 9500
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 1.0681297779083252,
+      "learning_rate": 7.445054945054945e-05,
+      "loss": 0.4394,
+      "step": 9510
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 1.280721664428711,
+      "learning_rate": 7.442307692307693e-05,
+      "loss": 0.4425,
+      "step": 9520
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.8306493759155273,
+      "learning_rate": 7.43956043956044e-05,
+      "loss": 0.4464,
+      "step": 9530
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.7929884195327759,
+      "learning_rate": 7.436813186813187e-05,
+      "loss": 0.4532,
+      "step": 9540
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.9622000455856323,
+      "learning_rate": 7.434065934065935e-05,
+      "loss": 0.4516,
+      "step": 9550
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.8442659378051758,
+      "learning_rate": 7.431318681318682e-05,
+      "loss": 0.4417,
+      "step": 9560
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.6593378782272339,
+      "learning_rate": 7.428571428571429e-05,
+      "loss": 0.4417,
+      "step": 9570
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.8290765881538391,
+      "learning_rate": 7.425824175824177e-05,
+      "loss": 0.4471,
+      "step": 9580
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.9711737632751465,
+      "learning_rate": 7.423076923076924e-05,
+      "loss": 0.4403,
+      "step": 9590
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.9451545476913452,
+      "learning_rate": 7.42032967032967e-05,
+      "loss": 0.4522,
+      "step": 9600
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 1.1259006261825562,
+      "learning_rate": 7.417582417582419e-05,
+      "loss": 0.4557,
+      "step": 9610
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.6328985095024109,
+      "learning_rate": 7.414835164835165e-05,
+      "loss": 0.4454,
+      "step": 9620
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.9557839035987854,
+      "learning_rate": 7.412087912087912e-05,
+      "loss": 0.4627,
+      "step": 9630
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.6751061081886292,
+      "learning_rate": 7.40934065934066e-05,
+      "loss": 0.4521,
+      "step": 9640
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.49840301275253296,
+      "learning_rate": 7.406593406593407e-05,
+      "loss": 0.4541,
+      "step": 9650
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 1.2455097436904907,
+      "learning_rate": 7.403846153846154e-05,
+      "loss": 0.4466,
+      "step": 9660
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 1.1257638931274414,
+      "learning_rate": 7.401098901098902e-05,
+      "loss": 0.4492,
+      "step": 9670
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.623653769493103,
+      "learning_rate": 7.398351648351649e-05,
+      "loss": 0.4475,
+      "step": 9680
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.6021344661712646,
+      "learning_rate": 7.395604395604396e-05,
+      "loss": 0.4393,
+      "step": 9690
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.7180885672569275,
+      "learning_rate": 7.392857142857144e-05,
+      "loss": 0.4571,
+      "step": 9700
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.5711954832077026,
+      "learning_rate": 7.390109890109891e-05,
+      "loss": 0.4534,
+      "step": 9710
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.8202961683273315,
+      "learning_rate": 7.387362637362638e-05,
+      "loss": 0.4568,
+      "step": 9720
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.5622140765190125,
+      "learning_rate": 7.384615384615386e-05,
+      "loss": 0.4415,
+      "step": 9730
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.9798756837844849,
+      "learning_rate": 7.381868131868133e-05,
+      "loss": 0.4371,
+      "step": 9740
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.6842076778411865,
+      "learning_rate": 7.37912087912088e-05,
+      "loss": 0.4495,
+      "step": 9750
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.6150854229927063,
+      "learning_rate": 7.376373626373628e-05,
+      "loss": 0.4566,
+      "step": 9760
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.5068964958190918,
+      "learning_rate": 7.373626373626374e-05,
+      "loss": 0.4445,
+      "step": 9770
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.8547071218490601,
+      "learning_rate": 7.370879120879121e-05,
+      "loss": 0.4398,
+      "step": 9780
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.9119787216186523,
+      "learning_rate": 7.36813186813187e-05,
+      "loss": 0.4622,
+      "step": 9790
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 0.6831972002983093,
+      "learning_rate": 7.365384615384616e-05,
+      "loss": 0.4493,
+      "step": 9800
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 1.850950002670288,
+      "learning_rate": 7.362637362637363e-05,
+      "loss": 0.4418,
+      "step": 9810
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 1.3177289962768555,
+      "learning_rate": 7.359890109890111e-05,
+      "loss": 0.4514,
+      "step": 9820
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.8253613114356995,
+      "learning_rate": 7.357142857142858e-05,
+      "loss": 0.4536,
+      "step": 9830
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.9882771968841553,
+      "learning_rate": 7.354395604395605e-05,
+      "loss": 0.4388,
+      "step": 9840
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 1.1343753337860107,
+      "learning_rate": 7.351648351648353e-05,
+      "loss": 0.4495,
+      "step": 9850
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.9710454344749451,
+      "learning_rate": 7.3489010989011e-05,
+      "loss": 0.4546,
+      "step": 9860
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 1.0125503540039062,
+      "learning_rate": 7.346153846153847e-05,
+      "loss": 0.4436,
+      "step": 9870
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.972205400466919,
+      "learning_rate": 7.343406593406593e-05,
+      "loss": 0.4603,
+      "step": 9880
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.8217471837997437,
+      "learning_rate": 7.34065934065934e-05,
+      "loss": 0.4551,
+      "step": 9890
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.9497684836387634,
+      "learning_rate": 7.337912087912087e-05,
+      "loss": 0.4398,
+      "step": 9900
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 1.2341935634613037,
+      "learning_rate": 7.335164835164835e-05,
+      "loss": 0.4412,
+      "step": 9910
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 1.0097248554229736,
+      "learning_rate": 7.332417582417582e-05,
+      "loss": 0.4536,
+      "step": 9920
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 1.1135015487670898,
+      "learning_rate": 7.329670329670329e-05,
+      "loss": 0.4482,
+      "step": 9930
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 1.0130164623260498,
+      "learning_rate": 7.326923076923077e-05,
+      "loss": 0.4499,
+      "step": 9940
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.802757740020752,
+      "learning_rate": 7.324175824175824e-05,
+      "loss": 0.4544,
+      "step": 9950
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.5492908954620361,
+      "learning_rate": 7.321428571428571e-05,
+      "loss": 0.4457,
+      "step": 9960
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 1.1192032098770142,
+      "learning_rate": 7.318681318681319e-05,
+      "loss": 0.4464,
+      "step": 9970
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.5543531179428101,
+      "learning_rate": 7.315934065934066e-05,
+      "loss": 0.456,
+      "step": 9980
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.741788387298584,
+      "learning_rate": 7.313186813186812e-05,
+      "loss": 0.4384,
+      "step": 9990
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.6588950157165527,
+      "learning_rate": 7.31043956043956e-05,
+      "loss": 0.4606,
+      "step": 10000
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.5929410457611084,
+      "learning_rate": 7.307692307692307e-05,
+      "loss": 0.4474,
+      "step": 10010
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.624821662902832,
+      "learning_rate": 7.304945054945054e-05,
+      "loss": 0.4486,
+      "step": 10020
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.7971046566963196,
+      "learning_rate": 7.302197802197802e-05,
+      "loss": 0.4433,
+      "step": 10030
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.81767737865448,
+      "learning_rate": 7.299450549450549e-05,
+      "loss": 0.4583,
+      "step": 10040
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.7773674726486206,
+      "learning_rate": 7.296703296703296e-05,
+      "loss": 0.4488,
+      "step": 10050
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.7860414981842041,
+      "learning_rate": 7.293956043956044e-05,
+      "loss": 0.4548,
+      "step": 10060
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.627571702003479,
+      "learning_rate": 7.291208791208791e-05,
+      "loss": 0.4477,
+      "step": 10070
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.7049760818481445,
+      "learning_rate": 7.288461538461538e-05,
+      "loss": 0.4376,
+      "step": 10080
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.5372808575630188,
+      "learning_rate": 7.285714285714286e-05,
+      "loss": 0.4552,
+      "step": 10090
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.6712595224380493,
+      "learning_rate": 7.282967032967033e-05,
+      "loss": 0.4555,
+      "step": 10100
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.4722815752029419,
+      "learning_rate": 7.28021978021978e-05,
+      "loss": 0.4536,
+      "step": 10110
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.48171865940093994,
+      "learning_rate": 7.277472527472528e-05,
+      "loss": 0.4542,
+      "step": 10120
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.8346227407455444,
+      "learning_rate": 7.274725274725275e-05,
+      "loss": 0.454,
+      "step": 10130
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.42196372151374817,
+      "learning_rate": 7.271978021978021e-05,
+      "loss": 0.4514,
+      "step": 10140
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.655756950378418,
+      "learning_rate": 7.26923076923077e-05,
+      "loss": 0.458,
+      "step": 10150
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 1.0113352537155151,
+      "learning_rate": 7.266483516483516e-05,
+      "loss": 0.4537,
+      "step": 10160
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.5985142588615417,
+      "learning_rate": 7.263736263736263e-05,
+      "loss": 0.4408,
+      "step": 10170
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 1.50113844871521,
+      "learning_rate": 7.260989010989011e-05,
+      "loss": 0.4516,
+      "step": 10180
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 1.536289095878601,
+      "learning_rate": 7.258241758241758e-05,
+      "loss": 0.4465,
+      "step": 10190
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.6726506948471069,
+      "learning_rate": 7.255494505494505e-05,
+      "loss": 0.4365,
+      "step": 10200
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.6390563249588013,
+      "learning_rate": 7.252747252747253e-05,
+      "loss": 0.4401,
+      "step": 10210
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.7105517983436584,
+      "learning_rate": 7.25e-05,
+      "loss": 0.445,
+      "step": 10220
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.9596685767173767,
+      "learning_rate": 7.247252747252747e-05,
+      "loss": 0.4367,
+      "step": 10230
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 1.161080002784729,
+      "learning_rate": 7.244505494505495e-05,
+      "loss": 0.4372,
+      "step": 10240
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.9476664662361145,
+      "learning_rate": 7.241758241758242e-05,
+      "loss": 0.4416,
+      "step": 10250
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.7320071458816528,
+      "learning_rate": 7.239010989010989e-05,
+      "loss": 0.4549,
+      "step": 10260
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 1.2132612466812134,
+      "learning_rate": 7.236263736263737e-05,
+      "loss": 0.4573,
+      "step": 10270
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.5772107839584351,
+      "learning_rate": 7.233516483516484e-05,
+      "loss": 0.4502,
+      "step": 10280
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 1.0478837490081787,
+      "learning_rate": 7.23076923076923e-05,
+      "loss": 0.452,
+      "step": 10290
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.6411312222480774,
+      "learning_rate": 7.228021978021979e-05,
+      "loss": 0.4424,
+      "step": 10300
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.7831696271896362,
+      "learning_rate": 7.225274725274725e-05,
+      "loss": 0.4525,
+      "step": 10310
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.45390376448631287,
+      "learning_rate": 7.222527472527472e-05,
+      "loss": 0.4401,
+      "step": 10320
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.8081623911857605,
+      "learning_rate": 7.21978021978022e-05,
+      "loss": 0.4396,
+      "step": 10330
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 1.091840147972107,
+      "learning_rate": 7.217032967032967e-05,
+      "loss": 0.4481,
+      "step": 10340
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 1.3704334497451782,
+      "learning_rate": 7.214285714285714e-05,
+      "loss": 0.4563,
+      "step": 10350
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.8897873163223267,
+      "learning_rate": 7.211538461538462e-05,
+      "loss": 0.4537,
+      "step": 10360
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 1.2123398780822754,
+      "learning_rate": 7.208791208791209e-05,
+      "loss": 0.4421,
+      "step": 10370
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.7756679058074951,
+      "learning_rate": 7.206043956043956e-05,
+      "loss": 0.4455,
+      "step": 10380
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.5030421614646912,
+      "learning_rate": 7.203296703296704e-05,
+      "loss": 0.4482,
+      "step": 10390
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.5804952383041382,
+      "learning_rate": 7.200549450549451e-05,
+      "loss": 0.4468,
+      "step": 10400
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.5755320191383362,
+      "learning_rate": 7.197802197802198e-05,
+      "loss": 0.4473,
+      "step": 10410
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.7598744034767151,
+      "learning_rate": 7.195054945054946e-05,
+      "loss": 0.4609,
+      "step": 10420
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.5830199122428894,
+      "learning_rate": 7.192307692307693e-05,
+      "loss": 0.4574,
+      "step": 10430
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.6908131837844849,
+      "learning_rate": 7.18956043956044e-05,
+      "loss": 0.4419,
+      "step": 10440
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 1.064139485359192,
+      "learning_rate": 7.186813186813188e-05,
+      "loss": 0.4467,
+      "step": 10450
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 1.5694611072540283,
+      "learning_rate": 7.184065934065934e-05,
+      "loss": 0.4638,
+      "step": 10460
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.6071065664291382,
+      "learning_rate": 7.181318681318681e-05,
+      "loss": 0.4449,
+      "step": 10470
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.7417973279953003,
+      "learning_rate": 7.17857142857143e-05,
+      "loss": 0.4452,
+      "step": 10480
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 1.0433728694915771,
+      "learning_rate": 7.175824175824176e-05,
+      "loss": 0.4413,
+      "step": 10490
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.5444348454475403,
+      "learning_rate": 7.173076923076923e-05,
+      "loss": 0.453,
+      "step": 10500
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 1.0602647066116333,
+      "learning_rate": 7.170329670329671e-05,
+      "loss": 0.4489,
+      "step": 10510
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.7549176216125488,
+      "learning_rate": 7.167582417582418e-05,
+      "loss": 0.4322,
+      "step": 10520
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.7759855389595032,
+      "learning_rate": 7.164835164835165e-05,
+      "loss": 0.4459,
+      "step": 10530
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.6525812149047852,
+      "learning_rate": 7.162087912087913e-05,
+      "loss": 0.4533,
+      "step": 10540
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.7948151230812073,
+      "learning_rate": 7.15934065934066e-05,
+      "loss": 0.4412,
+      "step": 10550
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.8408129215240479,
+      "learning_rate": 7.156593406593407e-05,
+      "loss": 0.4441,
+      "step": 10560
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.8035346865653992,
+      "learning_rate": 7.153846153846155e-05,
+      "loss": 0.463,
+      "step": 10570
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.9992803335189819,
+      "learning_rate": 7.151098901098902e-05,
+      "loss": 0.4426,
+      "step": 10580
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.7550234794616699,
+      "learning_rate": 7.148351648351648e-05,
+      "loss": 0.4455,
+      "step": 10590
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.47757139801979065,
+      "learning_rate": 7.145604395604397e-05,
+      "loss": 0.4487,
+      "step": 10600
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.6475192308425903,
+      "learning_rate": 7.142857142857143e-05,
+      "loss": 0.4493,
+      "step": 10610
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 1.0151840448379517,
+      "learning_rate": 7.14010989010989e-05,
+      "loss": 0.432,
+      "step": 10620
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.586617648601532,
+      "learning_rate": 7.137362637362638e-05,
+      "loss": 0.44,
+      "step": 10630
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.49713996052742004,
+      "learning_rate": 7.134615384615385e-05,
+      "loss": 0.4411,
+      "step": 10640
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.8117252588272095,
+      "learning_rate": 7.131868131868132e-05,
+      "loss": 0.4505,
+      "step": 10650
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.9269289374351501,
+      "learning_rate": 7.12912087912088e-05,
+      "loss": 0.4481,
+      "step": 10660
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 1.0942814350128174,
+      "learning_rate": 7.126373626373627e-05,
+      "loss": 0.4407,
+      "step": 10670
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.6903828382492065,
+      "learning_rate": 7.123626373626374e-05,
+      "loss": 0.439,
+      "step": 10680
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.8993271589279175,
+      "learning_rate": 7.120879120879122e-05,
+      "loss": 0.4421,
+      "step": 10690
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.8367154002189636,
+      "learning_rate": 7.118131868131869e-05,
+      "loss": 0.4492,
+      "step": 10700
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.7720317244529724,
+      "learning_rate": 7.115384615384616e-05,
+      "loss": 0.4521,
+      "step": 10710
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 1.0028263330459595,
+      "learning_rate": 7.112637362637364e-05,
+      "loss": 0.4537,
+      "step": 10720
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.663088321685791,
+      "learning_rate": 7.10989010989011e-05,
+      "loss": 0.4484,
+      "step": 10730
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 1.360903263092041,
+      "learning_rate": 7.107142857142857e-05,
+      "loss": 0.4663,
+      "step": 10740
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 1.5924625396728516,
+      "learning_rate": 7.104395604395606e-05,
+      "loss": 0.4599,
+      "step": 10750
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.7223337888717651,
+      "learning_rate": 7.101648351648352e-05,
+      "loss": 0.4528,
+      "step": 10760
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.8362312316894531,
+      "learning_rate": 7.098901098901099e-05,
+      "loss": 0.4469,
+      "step": 10770
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.5594087839126587,
+      "learning_rate": 7.096153846153847e-05,
+      "loss": 0.4484,
+      "step": 10780
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.8571014404296875,
+      "learning_rate": 7.093406593406594e-05,
+      "loss": 0.4414,
+      "step": 10790
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.9150867462158203,
+      "learning_rate": 7.090659340659341e-05,
+      "loss": 0.453,
+      "step": 10800
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.6607587337493896,
+      "learning_rate": 7.087912087912089e-05,
+      "loss": 0.4507,
+      "step": 10810
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.7428970336914062,
+      "learning_rate": 7.085164835164836e-05,
+      "loss": 0.4498,
+      "step": 10820
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.9267810583114624,
+      "learning_rate": 7.082417582417583e-05,
+      "loss": 0.4536,
+      "step": 10830
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.4540146291255951,
+      "learning_rate": 7.079670329670331e-05,
+      "loss": 0.4396,
+      "step": 10840
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 1.3146427869796753,
+      "learning_rate": 7.076923076923078e-05,
+      "loss": 0.4439,
+      "step": 10850
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.7954951524734497,
+      "learning_rate": 7.074175824175825e-05,
+      "loss": 0.4452,
+      "step": 10860
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.9023503661155701,
+      "learning_rate": 7.071428571428573e-05,
+      "loss": 0.4248,
+      "step": 10870
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.7970243692398071,
+      "learning_rate": 7.06868131868132e-05,
+      "loss": 0.4546,
+      "step": 10880
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.6860815286636353,
+      "learning_rate": 7.065934065934066e-05,
+      "loss": 0.4393,
+      "step": 10890
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.6077113747596741,
+      "learning_rate": 7.063186813186815e-05,
+      "loss": 0.4438,
+      "step": 10900
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.5479728579521179,
+      "learning_rate": 7.060439560439561e-05,
+      "loss": 0.4474,
+      "step": 10910
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.8500062823295593,
+      "learning_rate": 7.057692307692308e-05,
+      "loss": 0.4403,
+      "step": 10920
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.8599127531051636,
+      "learning_rate": 7.054945054945056e-05,
+      "loss": 0.4367,
+      "step": 10930
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 1.3294146060943604,
+      "learning_rate": 7.052197802197803e-05,
+      "loss": 0.4352,
+      "step": 10940
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.7617499828338623,
+      "learning_rate": 7.04945054945055e-05,
+      "loss": 0.4454,
+      "step": 10950
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 1.2405683994293213,
+      "learning_rate": 7.046703296703298e-05,
+      "loss": 0.4513,
+      "step": 10960
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.768068253993988,
+      "learning_rate": 7.043956043956045e-05,
+      "loss": 0.4448,
+      "step": 10970
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 1.0528051853179932,
+      "learning_rate": 7.041208791208792e-05,
+      "loss": 0.4436,
+      "step": 10980
+    },
+    {
+      "epoch": 3.0,
+      "eval_loss": 0.44340217113494873,
+      "eval_runtime": 1239.9169,
+      "eval_samples_per_second": 235.588,
+      "eval_steps_per_second": 3.682,
+      "step": 10982
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 36600,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "total_flos": 4.683801644117262e+18,
+  "train_batch_size": 6,
+  "trial_name": null,
+  "trial_params": null
+}