diff --git "a/checkpoint-7500/trainer_state.json" "b/checkpoint-7500/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-7500/trainer_state.json"
@@ -0,0 +1,5391 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.2976190476190476,
+  "eval_steps": 500,
+  "global_step": 7500,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 23.495065689086914,
+      "learning_rate": 4.0000000000000003e-07,
+      "loss": 8.6232,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 24.87381935119629,
+      "learning_rate": 9e-07,
+      "loss": 8.4514,
+      "step": 20
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 22.63846778869629,
+      "learning_rate": 1.4000000000000001e-06,
+      "loss": 8.3152,
+      "step": 30
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 20.65507698059082,
+      "learning_rate": 1.9e-06,
+      "loss": 7.9732,
+      "step": 40
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 16.147356033325195,
+      "learning_rate": 2.4000000000000003e-06,
+      "loss": 7.6879,
+      "step": 50
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 14.252687454223633,
+      "learning_rate": 2.9e-06,
+      "loss": 7.4152,
+      "step": 60
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 12.011382102966309,
+      "learning_rate": 3.4000000000000005e-06,
+      "loss": 7.2472,
+      "step": 70
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 10.081536293029785,
+      "learning_rate": 3.9e-06,
+      "loss": 7.2433,
+      "step": 80
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 9.65556526184082,
+      "learning_rate": 4.4e-06,
+      "loss": 7.2298,
+      "step": 90
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 8.830799102783203,
+      "learning_rate": 4.9000000000000005e-06,
+      "loss": 7.19,
+      "step": 100
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 9.019427299499512,
+      "learning_rate": 5.4e-06,
+      "loss": 7.2362,
+      "step": 110
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 9.13768482208252,
+      "learning_rate": 5.9e-06,
+      "loss": 7.2804,
+      "step": 120
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 9.02840805053711,
+      "learning_rate": 6.4000000000000006e-06,
+      "loss": 7.2345,
+      "step": 130
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 8.178731918334961,
+      "learning_rate": 6.900000000000001e-06,
+      "loss": 7.3014,
+      "step": 140
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 8.83882999420166,
+      "learning_rate": 7.4e-06,
+      "loss": 7.3875,
+      "step": 150
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 8.3687162399292,
+      "learning_rate": 7.9e-06,
+      "loss": 7.4546,
+      "step": 160
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 8.09121036529541,
+      "learning_rate": 8.400000000000001e-06,
+      "loss": 7.407,
+      "step": 170
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 12.731450080871582,
+      "learning_rate": 8.9e-06,
+      "loss": 7.4321,
+      "step": 180
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 7.601162910461426,
+      "learning_rate": 9.4e-06,
+      "loss": 7.4155,
+      "step": 190
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 8.006213188171387,
+      "learning_rate": 9.900000000000002e-06,
+      "loss": 7.3705,
+      "step": 200
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 7.79794454574585,
+      "learning_rate": 1.04e-05,
+      "loss": 7.3759,
+      "step": 210
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 7.281346321105957,
+      "learning_rate": 1.09e-05,
+      "loss": 7.3798,
+      "step": 220
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 7.605266094207764,
+      "learning_rate": 1.1400000000000001e-05,
+      "loss": 7.3878,
+      "step": 230
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 7.446412563323975,
+      "learning_rate": 1.19e-05,
+      "loss": 7.3098,
+      "step": 240
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 7.6324334144592285,
+      "learning_rate": 1.24e-05,
+      "loss": 7.2995,
+      "step": 250
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 8.330737113952637,
+      "learning_rate": 1.29e-05,
+      "loss": 7.1887,
+      "step": 260
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 7.283708572387695,
+      "learning_rate": 1.3400000000000002e-05,
+      "loss": 7.1184,
+      "step": 270
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.950069904327393,
+      "learning_rate": 1.3900000000000002e-05,
+      "loss": 6.9876,
+      "step": 280
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.844958782196045,
+      "learning_rate": 1.44e-05,
+      "loss": 6.9319,
+      "step": 290
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 7.499362468719482,
+      "learning_rate": 1.49e-05,
+      "loss": 6.9253,
+      "step": 300
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 8.45572280883789,
+      "learning_rate": 1.54e-05,
+      "loss": 6.7999,
+      "step": 310
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 8.108625411987305,
+      "learning_rate": 1.59e-05,
+      "loss": 6.7667,
+      "step": 320
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 8.131596565246582,
+      "learning_rate": 1.6400000000000002e-05,
+      "loss": 6.7599,
+      "step": 330
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 7.2159247398376465,
+      "learning_rate": 1.69e-05,
+      "loss": 6.7379,
+      "step": 340
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 7.335048675537109,
+      "learning_rate": 1.74e-05,
+      "loss": 6.6553,
+      "step": 350
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 7.361081600189209,
+      "learning_rate": 1.79e-05,
+      "loss": 6.6905,
+      "step": 360
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 7.59442138671875,
+      "learning_rate": 1.84e-05,
+      "loss": 6.6356,
+      "step": 370
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.31695032119751,
+      "learning_rate": 1.8900000000000002e-05,
+      "loss": 6.6538,
+      "step": 380
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.541380882263184,
+      "learning_rate": 1.94e-05,
+      "loss": 6.6873,
+      "step": 390
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 8.056180000305176,
+      "learning_rate": 1.9900000000000003e-05,
+      "loss": 6.6332,
+      "step": 400
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.74859094619751,
+      "learning_rate": 2.04e-05,
+      "loss": 6.5488,
+      "step": 410
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.700525760650635,
+      "learning_rate": 2.09e-05,
+      "loss": 6.611,
+      "step": 420
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.347009181976318,
+      "learning_rate": 2.1400000000000002e-05,
+      "loss": 6.6621,
+      "step": 430
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.607337951660156,
+      "learning_rate": 2.19e-05,
+      "loss": 6.568,
+      "step": 440
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.816825866699219,
+      "learning_rate": 2.2400000000000002e-05,
+      "loss": 6.6293,
+      "step": 450
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.984443664550781,
+      "learning_rate": 2.29e-05,
+      "loss": 6.5667,
+      "step": 460
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.995021820068359,
+      "learning_rate": 2.3400000000000003e-05,
+      "loss": 6.5062,
+      "step": 470
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.876286029815674,
+      "learning_rate": 2.39e-05,
+      "loss": 6.5034,
+      "step": 480
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.953365802764893,
+      "learning_rate": 2.44e-05,
+      "loss": 6.459,
+      "step": 490
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.246520519256592,
+      "learning_rate": 2.4900000000000002e-05,
+      "loss": 6.4359,
+      "step": 500
+    },
+    {
+      "epoch": 0.02,
+      "eval_loss": 6.133457660675049,
+      "eval_runtime": 422.8798,
+      "eval_samples_per_second": 50.182,
+      "eval_steps_per_second": 0.393,
+      "step": 500
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.1657819747924805,
+      "learning_rate": 2.54e-05,
+      "loss": 6.4372,
+      "step": 510
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.203813552856445,
+      "learning_rate": 2.5900000000000003e-05,
+      "loss": 6.3684,
+      "step": 520
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.829228401184082,
+      "learning_rate": 2.64e-05,
+      "loss": 6.3275,
+      "step": 530
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.641738414764404,
+      "learning_rate": 2.6900000000000003e-05,
+      "loss": 6.2923,
+      "step": 540
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.824960708618164,
+      "learning_rate": 2.7400000000000002e-05,
+      "loss": 6.3685,
+      "step": 550
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.7754364013671875,
+      "learning_rate": 2.7900000000000004e-05,
+      "loss": 6.3069,
+      "step": 560
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.3608903884887695,
+      "learning_rate": 2.84e-05,
+      "loss": 6.2915,
+      "step": 570
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.551064968109131,
+      "learning_rate": 2.8899999999999998e-05,
+      "loss": 6.1676,
+      "step": 580
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.588657379150391,
+      "learning_rate": 2.94e-05,
+      "loss": 6.2311,
+      "step": 590
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.631275177001953,
+      "learning_rate": 2.9900000000000002e-05,
+      "loss": 6.1824,
+      "step": 600
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.2990288734436035,
+      "learning_rate": 3.04e-05,
+      "loss": 6.1641,
+      "step": 610
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.6439714431762695,
+      "learning_rate": 3.09e-05,
+      "loss": 6.0125,
+      "step": 620
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.402431964874268,
+      "learning_rate": 3.1400000000000004e-05,
+      "loss": 6.1314,
+      "step": 630
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.883937358856201,
+      "learning_rate": 3.19e-05,
+      "loss": 5.9858,
+      "step": 640
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.988973140716553,
+      "learning_rate": 3.24e-05,
+      "loss": 5.9886,
+      "step": 650
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.186301231384277,
+      "learning_rate": 3.29e-05,
+      "loss": 5.9732,
+      "step": 660
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.2982354164123535,
+      "learning_rate": 3.3400000000000005e-05,
+      "loss": 6.0391,
+      "step": 670
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.037730693817139,
+      "learning_rate": 3.3900000000000004e-05,
+      "loss": 6.0547,
+      "step": 680
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.280949592590332,
+      "learning_rate": 3.4399999999999996e-05,
+      "loss": 5.8734,
+      "step": 690
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.861902236938477,
+      "learning_rate": 3.49e-05,
+      "loss": 5.8997,
+      "step": 700
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.4033589363098145,
+      "learning_rate": 3.54e-05,
+      "loss": 5.8111,
+      "step": 710
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.376819133758545,
+      "learning_rate": 3.59e-05,
+      "loss": 5.9027,
+      "step": 720
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.777069568634033,
+      "learning_rate": 3.6400000000000004e-05,
+      "loss": 5.727,
+      "step": 730
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.183162689208984,
+      "learning_rate": 3.69e-05,
+      "loss": 5.733,
+      "step": 740
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.658604621887207,
+      "learning_rate": 3.74e-05,
+      "loss": 5.7018,
+      "step": 750
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.263491153717041,
+      "learning_rate": 3.79e-05,
+      "loss": 5.7315,
+      "step": 760
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.86756706237793,
+      "learning_rate": 3.8400000000000005e-05,
+      "loss": 5.6637,
+      "step": 770
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.122657299041748,
+      "learning_rate": 3.8900000000000004e-05,
+      "loss": 5.6849,
+      "step": 780
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.6024580001831055,
+      "learning_rate": 3.94e-05,
+      "loss": 5.5835,
+      "step": 790
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.619604587554932,
+      "learning_rate": 3.99e-05,
+      "loss": 5.7,
+      "step": 800
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.7342352867126465,
+      "learning_rate": 4.0400000000000006e-05,
+      "loss": 5.4825,
+      "step": 810
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.796237468719482,
+      "learning_rate": 4.09e-05,
+      "loss": 5.6496,
+      "step": 820
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.601135730743408,
+      "learning_rate": 4.14e-05,
+      "loss": 5.3526,
+      "step": 830
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.576679229736328,
+      "learning_rate": 4.19e-05,
+      "loss": 5.3757,
+      "step": 840
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.774380683898926,
+      "learning_rate": 4.24e-05,
+      "loss": 5.5164,
+      "step": 850
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.140038967132568,
+      "learning_rate": 4.29e-05,
+      "loss": 5.3124,
+      "step": 860
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.971868515014648,
+      "learning_rate": 4.3400000000000005e-05,
+      "loss": 5.3387,
+      "step": 870
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.431824207305908,
+      "learning_rate": 4.39e-05,
+      "loss": 5.4363,
+      "step": 880
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.174604892730713,
+      "learning_rate": 4.44e-05,
+      "loss": 5.282,
+      "step": 890
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.313474178314209,
+      "learning_rate": 4.49e-05,
+      "loss": 5.3754,
+      "step": 900
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.066281318664551,
+      "learning_rate": 4.5400000000000006e-05,
+      "loss": 5.3197,
+      "step": 910
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 7.4433817863464355,
+      "learning_rate": 4.5900000000000004e-05,
+      "loss": 5.3277,
+      "step": 920
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.884599208831787,
+      "learning_rate": 4.64e-05,
+      "loss": 5.0681,
+      "step": 930
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.17655611038208,
+      "learning_rate": 4.69e-05,
+      "loss": 5.1159,
+      "step": 940
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 7.090456485748291,
+      "learning_rate": 4.74e-05,
+      "loss": 5.2264,
+      "step": 950
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.508240699768066,
+      "learning_rate": 4.79e-05,
+      "loss": 5.2113,
+      "step": 960
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.5788445472717285,
+      "learning_rate": 4.8400000000000004e-05,
+      "loss": 4.9996,
+      "step": 970
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.722221374511719,
+      "learning_rate": 4.89e-05,
+      "loss": 4.9688,
+      "step": 980
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.813933849334717,
+      "learning_rate": 4.94e-05,
+      "loss": 4.8691,
+      "step": 990
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.425748348236084,
+      "learning_rate": 4.99e-05,
+      "loss": 4.9883,
+      "step": 1000
+    },
+    {
+      "epoch": 0.04,
+      "eval_loss": 4.425854682922363,
+      "eval_runtime": 422.6592,
+      "eval_samples_per_second": 50.208,
+      "eval_steps_per_second": 0.393,
+      "step": 1000
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.170848369598389,
+      "learning_rate": 4.9999996764541565e-05,
+      "loss": 4.8794,
+      "step": 1010
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.247808933258057,
+      "learning_rate": 4.99999836204931e-05,
+      "loss": 4.9664,
+      "step": 1020
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.27077054977417,
+      "learning_rate": 4.9999960365643774e-05,
+      "loss": 4.7753,
+      "step": 1030
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.284353733062744,
+      "learning_rate": 4.999992700000298e-05,
+      "loss": 4.8909,
+      "step": 1040
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.716318607330322,
+      "learning_rate": 4.999988352358421e-05,
+      "loss": 4.861,
+      "step": 1050
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.448729991912842,
+      "learning_rate": 4.999982993640506e-05,
+      "loss": 4.8776,
+      "step": 1060
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.401464939117432,
+      "learning_rate": 4.9999766238487204e-05,
+      "loss": 4.8507,
+      "step": 1070
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.219631195068359,
+      "learning_rate": 4.999969242985639e-05,
+      "loss": 4.8234,
+      "step": 1080
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.225848197937012,
+      "learning_rate": 4.9999608510542485e-05,
+      "loss": 4.6093,
+      "step": 1090
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.117152690887451,
+      "learning_rate": 4.999951448057941e-05,
+      "loss": 4.786,
+      "step": 1100
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.788968086242676,
+      "learning_rate": 4.999941034000521e-05,
+      "loss": 4.8389,
+      "step": 1110
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.037156581878662,
+      "learning_rate": 4.999929608886198e-05,
+      "loss": 4.7674,
+      "step": 1120
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.567563056945801,
+      "learning_rate": 4.999917172719596e-05,
+      "loss": 4.8247,
+      "step": 1130
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.557987213134766,
+      "learning_rate": 4.999903725505742e-05,
+      "loss": 4.7517,
+      "step": 1140
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.490599632263184,
+      "learning_rate": 4.999889267250075e-05,
+      "loss": 4.6037,
+      "step": 1150
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.000855922698975,
+      "learning_rate": 4.999873797958443e-05,
+      "loss": 4.6077,
+      "step": 1160
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.378462314605713,
+      "learning_rate": 4.999857317637102e-05,
+      "loss": 4.3662,
+      "step": 1170
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 7.4290924072265625,
+      "learning_rate": 4.999839826292718e-05,
+      "loss": 4.5339,
+      "step": 1180
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.727224826812744,
+      "learning_rate": 4.999821323932363e-05,
+      "loss": 4.5744,
+      "step": 1190
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.893082618713379,
+      "learning_rate": 4.9998018105635214e-05,
+      "loss": 4.5247,
+      "step": 1200
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.759618759155273,
+      "learning_rate": 4.999781286194085e-05,
+      "loss": 4.5744,
+      "step": 1210
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.082376003265381,
+      "learning_rate": 4.999759750832355e-05,
+      "loss": 4.6011,
+      "step": 1220
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.305749416351318,
+      "learning_rate": 4.999737204487039e-05,
+      "loss": 4.4587,
+      "step": 1230
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.710574626922607,
+      "learning_rate": 4.999713647167258e-05,
+      "loss": 4.46,
+      "step": 1240
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.753530502319336,
+      "learning_rate": 4.999689078882537e-05,
+      "loss": 4.5674,
+      "step": 1250
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.949398040771484,
+      "learning_rate": 4.999663499642814e-05,
+      "loss": 4.3999,
+      "step": 1260
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.530541896820068,
+      "learning_rate": 4.999636909458434e-05,
+      "loss": 4.533,
+      "step": 1270
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.9646453857421875,
+      "learning_rate": 4.9996093083401495e-05,
+      "loss": 4.374,
+      "step": 1280
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.346386432647705,
+      "learning_rate": 4.9995806962991246e-05,
+      "loss": 4.4758,
+      "step": 1290
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.797152519226074,
+      "learning_rate": 4.9995510733469295e-05,
+      "loss": 4.6275,
+      "step": 1300
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.426751136779785,
+      "learning_rate": 4.999520439495547e-05,
+      "loss": 4.3561,
+      "step": 1310
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.3839030265808105,
+      "learning_rate": 4.9994887947573646e-05,
+      "loss": 4.3134,
+      "step": 1320
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.01127290725708,
+      "learning_rate": 4.9994594501953555e-05,
+      "loss": 4.4336,
+      "step": 1330
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.228607654571533,
+      "learning_rate": 4.999425884807851e-05,
+      "loss": 4.1543,
+      "step": 1340
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.958486080169678,
+      "learning_rate": 4.999391308571788e-05,
+      "loss": 4.3806,
+      "step": 1350
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.816386699676514,
+      "learning_rate": 4.99935572150115e-05,
+      "loss": 4.3759,
+      "step": 1360
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.345638751983643,
+      "learning_rate": 4.9993191236103295e-05,
+      "loss": 4.5759,
+      "step": 1370
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.738880634307861,
+      "learning_rate": 4.9992815149141294e-05,
+      "loss": 4.3274,
+      "step": 1380
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.694160461425781,
+      "learning_rate": 4.9992428954277573e-05,
+      "loss": 4.376,
+      "step": 1390
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.050902843475342,
+      "learning_rate": 4.999203265166834e-05,
+      "loss": 4.1629,
+      "step": 1400
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 7.09380578994751,
+      "learning_rate": 4.9991626241473876e-05,
+      "loss": 4.3638,
+      "step": 1410
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.538166046142578,
+      "learning_rate": 4.999120972385853e-05,
+      "loss": 4.1318,
+      "step": 1420
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.052664279937744,
+      "learning_rate": 4.9990783098990765e-05,
+      "loss": 4.1856,
+      "step": 1430
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.707901477813721,
+      "learning_rate": 4.9990346367043114e-05,
+      "loss": 4.3361,
+      "step": 1440
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.3822150230407715,
+      "learning_rate": 4.998989952819223e-05,
+      "loss": 4.301,
+      "step": 1450
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.586170196533203,
+      "learning_rate": 4.998944258261879e-05,
+      "loss": 4.3131,
+      "step": 1460
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.62904691696167,
+      "learning_rate": 4.998897553050763e-05,
+      "loss": 4.3357,
+      "step": 1470
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.669075965881348,
+      "learning_rate": 4.9988498372047633e-05,
+      "loss": 4.2434,
+      "step": 1480
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.776549339294434,
+      "learning_rate": 4.998801110743177e-05,
+      "loss": 4.0441,
+      "step": 1490
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.442574977874756,
+      "learning_rate": 4.9987513736857106e-05,
+      "loss": 4.0931,
+      "step": 1500
+    },
+    {
+      "epoch": 0.06,
+      "eval_loss": 3.676598310470581,
+      "eval_runtime": 423.4536,
+      "eval_samples_per_second": 50.114,
+      "eval_steps_per_second": 0.392,
+      "step": 1500
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.529382705688477,
+      "learning_rate": 4.998700626052481e-05,
+      "loss": 4.1739,
+      "step": 1510
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.5605387687683105,
+      "learning_rate": 4.99864886786401e-05,
+      "loss": 4.1861,
+      "step": 1520
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.168206214904785,
+      "learning_rate": 4.998596099141231e-05,
+      "loss": 4.0245,
+      "step": 1530
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.634309768676758,
+      "learning_rate": 4.998542319905486e-05,
+      "loss": 4.2131,
+      "step": 1540
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.191194534301758,
+      "learning_rate": 4.998487530178525e-05,
+      "loss": 4.163,
+      "step": 1550
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.52882719039917,
+      "learning_rate": 4.998431729982506e-05,
+      "loss": 4.0668,
+      "step": 1560
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.614863395690918,
+      "learning_rate": 4.998374919339998e-05,
+      "loss": 4.2501,
+      "step": 1570
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.7852373123168945,
+      "learning_rate": 4.9983170982739745e-05,
+      "loss": 4.207,
+      "step": 1580
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 7.439071178436279,
+      "learning_rate": 4.998258266807822e-05,
+      "loss": 4.0462,
+      "step": 1590
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.510477542877197,
+      "learning_rate": 4.998198424965333e-05,
+      "loss": 4.277,
+      "step": 1600
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.421988487243652,
+      "learning_rate": 4.9981375727707114e-05,
+      "loss": 4.0289,
+      "step": 1610
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.632848739624023,
+      "learning_rate": 4.998075710248565e-05,
+      "loss": 4.2922,
+      "step": 1620
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.031602382659912,
+      "learning_rate": 4.9980128374239156e-05,
+      "loss": 3.9845,
+      "step": 1630
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.3867669105529785,
+      "learning_rate": 4.997948954322189e-05,
+      "loss": 4.1095,
+      "step": 1640
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.150663375854492,
+      "learning_rate": 4.997884060969222e-05,
+      "loss": 3.9247,
+      "step": 1650
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.717851638793945,
+      "learning_rate": 4.99781815739126e-05,
+      "loss": 4.1911,
+      "step": 1660
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 6.766101837158203,
+      "learning_rate": 4.9977512436149563e-05,
+      "loss": 3.9194,
+      "step": 1670
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 6.316621780395508,
+      "learning_rate": 4.9976833196673735e-05,
+      "loss": 4.101,
+      "step": 1680
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.107331275939941,
+      "learning_rate": 4.9976143855759814e-05,
+      "loss": 4.0706,
+      "step": 1690
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.533346652984619,
+      "learning_rate": 4.99754444136866e-05,
+      "loss": 4.0164,
+      "step": 1700
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.967229843139648,
+      "learning_rate": 4.9974734870736975e-05,
+      "loss": 4.1229,
+      "step": 1710
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.502392768859863,
+      "learning_rate": 4.997401522719788e-05,
+      "loss": 4.0456,
+      "step": 1720
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.6358962059021,
+      "learning_rate": 4.997328548336038e-05,
+      "loss": 3.9432,
+      "step": 1730
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 6.525423049926758,
+      "learning_rate": 4.99725456395196e-05,
+      "loss": 4.05,
+      "step": 1740
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.675175666809082,
+      "learning_rate": 4.997179569597475e-05,
+      "loss": 3.8145,
+      "step": 1750
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.409067630767822,
+      "learning_rate": 4.9971035653029153e-05,
+      "loss": 3.914,
+      "step": 1760
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.812373161315918,
+      "learning_rate": 4.997026551099017e-05,
+      "loss": 3.9817,
+      "step": 1770
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.165591239929199,
+      "learning_rate": 4.996948527016929e-05,
+      "loss": 3.9414,
+      "step": 1780
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.0605034828186035,
+      "learning_rate": 4.996869493088205e-05,
+      "loss": 3.8139,
+      "step": 1790
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 6.348527908325195,
+      "learning_rate": 4.996789449344812e-05,
+      "loss": 3.7905,
+      "step": 1800
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.135634422302246,
+      "learning_rate": 4.9967083958191187e-05,
+      "loss": 3.7834,
+      "step": 1810
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 8.48004150390625,
+      "learning_rate": 4.996626332543907e-05,
+      "loss": 3.7551,
+      "step": 1820
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.518482208251953,
+      "learning_rate": 4.996543259552367e-05,
+      "loss": 3.8619,
+      "step": 1830
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.7253737449646,
+      "learning_rate": 4.996459176878095e-05,
+      "loss": 3.7978,
+      "step": 1840
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 6.030267715454102,
+      "learning_rate": 4.9963740845550965e-05,
+      "loss": 3.9309,
+      "step": 1850
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.561445713043213,
+      "learning_rate": 4.996287982617787e-05,
+      "loss": 3.9761,
+      "step": 1860
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.555076599121094,
+      "learning_rate": 4.996200871100988e-05,
+      "loss": 3.8587,
+      "step": 1870
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.502940654754639,
+      "learning_rate": 4.9961127500399295e-05,
+      "loss": 3.847,
+      "step": 1880
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.557964324951172,
+      "learning_rate": 4.996023619470252e-05,
+      "loss": 3.6117,
+      "step": 1890
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.37682580947876,
+      "learning_rate": 4.9959334794280014e-05,
+      "loss": 3.7581,
+      "step": 1900
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.152722358703613,
+      "learning_rate": 4.9958423299496344e-05,
+      "loss": 3.8454,
+      "step": 1910
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.4067301750183105,
+      "learning_rate": 4.995750171072014e-05,
+      "loss": 3.9353,
+      "step": 1920
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.564638614654541,
+      "learning_rate": 4.9956570028324124e-05,
+      "loss": 3.738,
+      "step": 1930
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.320713520050049,
+      "learning_rate": 4.99556282526851e-05,
+      "loss": 3.5365,
+      "step": 1940
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.384575843811035,
+      "learning_rate": 4.995467638418395e-05,
+      "loss": 3.8153,
+      "step": 1950
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.424959659576416,
+      "learning_rate": 4.995371442320565e-05,
+      "loss": 3.779,
+      "step": 1960
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.928106784820557,
+      "learning_rate": 4.9952742370139224e-05,
+      "loss": 3.7395,
+      "step": 1970
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.09404182434082,
+      "learning_rate": 4.995176022537784e-05,
+      "loss": 3.757,
+      "step": 1980
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 8.332275390625,
+      "learning_rate": 4.995076798931867e-05,
+      "loss": 3.8866,
+      "step": 1990
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.44631814956665,
+      "learning_rate": 4.994976566236303e-05,
+      "loss": 3.8395,
+      "step": 2000
+    },
+    {
+      "epoch": 0.08,
+      "eval_loss": 3.2655727863311768,
+      "eval_runtime": 423.1622,
+      "eval_samples_per_second": 50.149,
+      "eval_steps_per_second": 0.392,
+      "step": 2000
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.428518295288086,
+      "learning_rate": 4.994875324491629e-05,
+      "loss": 3.7103,
+      "step": 2010
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.11751651763916,
+      "learning_rate": 4.9947730737387885e-05,
+      "loss": 3.8224,
+      "step": 2020
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.404431343078613,
+      "learning_rate": 4.994669814019138e-05,
+      "loss": 3.9208,
+      "step": 2030
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.487029075622559,
+      "learning_rate": 4.994565545374438e-05,
+      "loss": 3.7412,
+      "step": 2040
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.979548454284668,
+      "learning_rate": 4.9944602678468575e-05,
+      "loss": 3.6008,
+      "step": 2050
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.379942417144775,
+      "learning_rate": 4.994353981478974e-05,
+      "loss": 3.6077,
+      "step": 2060
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.666346549987793,
+      "learning_rate": 4.994246686313774e-05,
+      "loss": 3.8157,
+      "step": 2070
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.252011299133301,
+      "learning_rate": 4.99413838239465e-05,
+      "loss": 3.6929,
+      "step": 2080
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.579212188720703,
+      "learning_rate": 4.9940290697654045e-05,
+      "loss": 3.9075,
+      "step": 2090
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.747458457946777,
+      "learning_rate": 4.993918748470246e-05,
+      "loss": 3.5528,
+      "step": 2100
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.356452465057373,
+      "learning_rate": 4.993818596932118e-05,
+      "loss": 3.4987,
+      "step": 2110
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.761812686920166,
+      "learning_rate": 4.993706359294986e-05,
+      "loss": 3.7627,
+      "step": 2120
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.732122898101807,
+      "learning_rate": 4.993593113122456e-05,
+      "loss": 3.6106,
+      "step": 2130
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 11.450267791748047,
+      "learning_rate": 4.993478858460329e-05,
+      "loss": 3.4716,
+      "step": 2140
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.774083614349365,
+      "learning_rate": 4.993363595354813e-05,
+      "loss": 3.5741,
+      "step": 2150
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.652318000793457,
+      "learning_rate": 4.9932473238525247e-05,
+      "loss": 3.676,
+      "step": 2160
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.903106689453125,
+      "learning_rate": 4.993130044000487e-05,
+      "loss": 3.6334,
+      "step": 2170
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.358557224273682,
+      "learning_rate": 4.993011755846132e-05,
+      "loss": 3.5592,
+      "step": 2180
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 6.509594440460205,
+      "learning_rate": 4.9928924594372995e-05,
+      "loss": 3.6512,
+      "step": 2190
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 6.318192958831787,
+      "learning_rate": 4.9927721548222374e-05,
+      "loss": 3.3691,
+      "step": 2200
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.848221778869629,
+      "learning_rate": 4.9926508420496e-05,
+      "loss": 3.3721,
+      "step": 2210
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.9782490730285645,
+      "learning_rate": 4.992528521168449e-05,
+      "loss": 3.5859,
+      "step": 2220
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 6.406031608581543,
+      "learning_rate": 4.992405192228258e-05,
+      "loss": 3.6206,
+      "step": 2230
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 6.102032661437988,
+      "learning_rate": 4.992280855278903e-05,
+      "loss": 3.4615,
+      "step": 2240
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.5005035400390625,
+      "learning_rate": 4.992155510370669e-05,
+      "loss": 3.507,
+      "step": 2250
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 8.336762428283691,
+      "learning_rate": 4.9920291575542515e-05,
+      "loss": 3.3897,
+      "step": 2260
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 7.851747989654541,
+      "learning_rate": 4.9919017968807514e-05,
+      "loss": 3.5232,
+      "step": 2270
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.816250324249268,
+      "learning_rate": 4.991773428401676e-05,
+      "loss": 3.5638,
+      "step": 2280
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.746279716491699,
+      "learning_rate": 4.9916440521689433e-05,
+      "loss": 3.5446,
+      "step": 2290
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.8899922370910645,
+      "learning_rate": 4.9915136682348774e-05,
+      "loss": 3.3533,
+      "step": 2300
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 6.160922050476074,
+      "learning_rate": 4.9913822766522076e-05,
+      "loss": 3.4222,
+      "step": 2310
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 6.936460494995117,
+      "learning_rate": 4.991249877474075e-05,
+      "loss": 3.5763,
+      "step": 2320
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 16.517559051513672,
+      "learning_rate": 4.991116470754025e-05,
+      "loss": 3.5324,
+      "step": 2330
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.840503692626953,
+      "learning_rate": 4.9909820565460116e-05,
+      "loss": 3.5764,
+      "step": 2340
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.952908515930176,
+      "learning_rate": 4.990846634904398e-05,
+      "loss": 3.4369,
+      "step": 2350
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.108431339263916,
+      "learning_rate": 4.9907102058839504e-05,
+      "loss": 3.4518,
+      "step": 2360
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.068904399871826,
+      "learning_rate": 4.990572769539848e-05,
+      "loss": 3.5594,
+      "step": 2370
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.512117385864258,
+      "learning_rate": 4.990434325927671e-05,
+      "loss": 3.6366,
+      "step": 2380
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.916763782501221,
+      "learning_rate": 4.990294875103413e-05,
+      "loss": 3.4344,
+      "step": 2390
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 6.244757175445557,
+      "learning_rate": 4.990154417123472e-05,
+      "loss": 3.5858,
+      "step": 2400
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 6.071867942810059,
+      "learning_rate": 4.990012952044654e-05,
+      "loss": 3.5888,
+      "step": 2410
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.984373092651367,
+      "learning_rate": 4.98987047992417e-05,
+      "loss": 3.5084,
+      "step": 2420
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 7.062913417816162,
+      "learning_rate": 4.989727000819644e-05,
+      "loss": 3.4169,
+      "step": 2430
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.671230792999268,
+      "learning_rate": 4.9895825147891e-05,
+      "loss": 3.3354,
+      "step": 2440
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.235222339630127,
+      "learning_rate": 4.989437021890975e-05,
+      "loss": 3.4756,
+      "step": 2450
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.888786792755127,
+      "learning_rate": 4.98929052218411e-05,
+      "loss": 3.3409,
+      "step": 2460
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 6.014183044433594,
+      "learning_rate": 4.989143015727754e-05,
+      "loss": 3.4246,
+      "step": 2470
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 8.75373649597168,
+      "learning_rate": 4.988994502581565e-05,
+      "loss": 3.2431,
+      "step": 2480
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.800814628601074,
+      "learning_rate": 4.988844982805605e-05,
+      "loss": 3.4702,
+      "step": 2490
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.22312068939209,
+      "learning_rate": 4.9886944564603444e-05,
+      "loss": 3.2994,
+      "step": 2500
+    },
+    {
+      "epoch": 0.1,
+      "eval_loss": 2.9097418785095215,
+      "eval_runtime": 423.0547,
+      "eval_samples_per_second": 50.161,
+      "eval_steps_per_second": 0.392,
+      "step": 2500
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.41413688659668,
+      "learning_rate": 4.988542923606663e-05,
+      "loss": 3.5138,
+      "step": 2510
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.5836968421936035,
+      "learning_rate": 4.988390384305843e-05,
+      "loss": 3.4161,
+      "step": 2520
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.957060813903809,
+      "learning_rate": 4.9882368386195776e-05,
+      "loss": 3.4227,
+      "step": 2530
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.718446731567383,
+      "learning_rate": 4.988082286609965e-05,
+      "loss": 3.3632,
+      "step": 2540
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 6.273780822753906,
+      "learning_rate": 4.9879267283395125e-05,
+      "loss": 3.4018,
+      "step": 2550
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.000144004821777,
+      "learning_rate": 4.9877701638711314e-05,
+      "loss": 3.2865,
+      "step": 2560
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 6.369386672973633,
+      "learning_rate": 4.987612593268142e-05,
+      "loss": 3.2916,
+      "step": 2570
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.275691032409668,
+      "learning_rate": 4.98745401659427e-05,
+      "loss": 3.3678,
+      "step": 2580
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 6.77783727645874,
+      "learning_rate": 4.9872944339136503e-05,
+      "loss": 3.2567,
+      "step": 2590
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 6.3765435218811035,
+      "learning_rate": 4.987133845290822e-05,
+      "loss": 3.3649,
+      "step": 2600
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 6.956243991851807,
+      "learning_rate": 4.9869722507907343e-05,
+      "loss": 3.428,
+      "step": 2610
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.933529376983643,
+      "learning_rate": 4.9868096504787395e-05,
+      "loss": 3.0448,
+      "step": 2620
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 6.319052696228027,
+      "learning_rate": 4.9866460444205983e-05,
+      "loss": 3.3518,
+      "step": 2630
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.592201232910156,
+      "learning_rate": 4.986481432682479e-05,
+      "loss": 3.1236,
+      "step": 2640
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.243892192840576,
+      "learning_rate": 4.986315815330956e-05,
+      "loss": 3.2447,
+      "step": 2650
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 6.197099208831787,
+      "learning_rate": 4.98614919243301e-05,
+      "loss": 3.3031,
+      "step": 2660
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 6.438327312469482,
+      "learning_rate": 4.98598156405603e-05,
+      "loss": 3.5433,
+      "step": 2670
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.134957313537598,
+      "learning_rate": 4.985812930267808e-05,
+      "loss": 3.3124,
+      "step": 2680
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.727309703826904,
+      "learning_rate": 4.985643291136546e-05,
+      "loss": 3.2169,
+      "step": 2690
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.680230617523193,
+      "learning_rate": 4.985472646730852e-05,
+      "loss": 3.0445,
+      "step": 2700
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.9644598960876465,
+      "learning_rate": 4.985300997119739e-05,
+      "loss": 3.1588,
+      "step": 2710
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.430914402008057,
+      "learning_rate": 4.985128342372629e-05,
+      "loss": 3.2385,
+      "step": 2720
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.565753936767578,
+      "learning_rate": 4.9849546825593476e-05,
+      "loss": 3.2654,
+      "step": 2730
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 6.126991271972656,
+      "learning_rate": 4.9847800177501305e-05,
+      "loss": 3.3024,
+      "step": 2740
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.406001567840576,
+      "learning_rate": 4.984604348015616e-05,
+      "loss": 3.219,
+      "step": 2750
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.55892276763916,
+      "learning_rate": 4.984427673426851e-05,
+      "loss": 3.3882,
+      "step": 2760
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.547699451446533,
+      "learning_rate": 4.9842499940552886e-05,
+      "loss": 3.2334,
+      "step": 2770
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 6.514155864715576,
+      "learning_rate": 4.9840713099727885e-05,
+      "loss": 3.2041,
+      "step": 2780
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.4210991859436035,
+      "learning_rate": 4.983891621251616e-05,
+      "loss": 3.1746,
+      "step": 2790
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.854736328125,
+      "learning_rate": 4.983710927964442e-05,
+      "loss": 3.1961,
+      "step": 2800
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.277487754821777,
+      "learning_rate": 4.9835292301843476e-05,
+      "loss": 2.9688,
+      "step": 2810
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.925424098968506,
+      "learning_rate": 4.983346527984813e-05,
+      "loss": 3.2315,
+      "step": 2820
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 7.818734169006348,
+      "learning_rate": 4.983162821439733e-05,
+      "loss": 3.325,
+      "step": 2830
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 6.312294006347656,
+      "learning_rate": 4.982978110623402e-05,
+      "loss": 3.2141,
+      "step": 2840
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 6.804452419281006,
+      "learning_rate": 4.982792395610524e-05,
+      "loss": 3.1214,
+      "step": 2850
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 6.256209850311279,
+      "learning_rate": 4.9826056764762074e-05,
+      "loss": 3.2357,
+      "step": 2860
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.713292121887207,
+      "learning_rate": 4.982417953295968e-05,
+      "loss": 3.0448,
+      "step": 2870
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.762032985687256,
+      "learning_rate": 4.982229226145727e-05,
+      "loss": 3.1855,
+      "step": 2880
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.245543956756592,
+      "learning_rate": 4.982039495101812e-05,
+      "loss": 3.1024,
+      "step": 2890
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.456753730773926,
+      "learning_rate": 4.9818487602409556e-05,
+      "loss": 3.2703,
+      "step": 2900
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.831927299499512,
+      "learning_rate": 4.981657021640298e-05,
+      "loss": 3.2172,
+      "step": 2910
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.4401350021362305,
+      "learning_rate": 4.981464279377384e-05,
+      "loss": 3.2972,
+      "step": 2920
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.1732892990112305,
+      "learning_rate": 4.981270533530164e-05,
+      "loss": 3.1305,
+      "step": 2930
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.929070472717285,
+      "learning_rate": 4.9810757841769976e-05,
+      "loss": 3.0932,
+      "step": 2940
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.50985860824585,
+      "learning_rate": 4.980880031396645e-05,
+      "loss": 3.1417,
+      "step": 2950
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.071355819702148,
+      "learning_rate": 4.980683275268277e-05,
+      "loss": 3.1796,
+      "step": 2960
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.182791709899902,
+      "learning_rate": 4.980485515871467e-05,
+      "loss": 3.268,
+      "step": 2970
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.4410786628723145,
+      "learning_rate": 4.980286753286195e-05,
+      "loss": 3.2922,
+      "step": 2980
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 22.151193618774414,
+      "learning_rate": 4.9800869875928475e-05,
+      "loss": 3.0482,
+      "step": 2990
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.808634281158447,
+      "learning_rate": 4.9798862188722164e-05,
+      "loss": 3.1515,
+      "step": 3000
+    },
+    {
+      "epoch": 0.12,
+      "eval_loss": 2.6660187244415283,
+      "eval_runtime": 423.2518,
+      "eval_samples_per_second": 50.138,
+      "eval_steps_per_second": 0.392,
+      "step": 3000
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 9.806035995483398,
+      "learning_rate": 4.979684447205499e-05,
+      "loss": 3.0872,
+      "step": 3010
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.456057071685791,
+      "learning_rate": 4.9794816726742976e-05,
+      "loss": 2.882,
+      "step": 3020
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.348060607910156,
+      "learning_rate": 4.979277895360622e-05,
+      "loss": 3.1984,
+      "step": 3030
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.867720603942871,
+      "learning_rate": 4.979073115346885e-05,
+      "loss": 3.0328,
+      "step": 3040
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.720703601837158,
+      "learning_rate": 4.9788673327159076e-05,
+      "loss": 3.1852,
+      "step": 3050
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.027235984802246,
+      "learning_rate": 4.978660547550914e-05,
+      "loss": 3.1082,
+      "step": 3060
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.131529808044434,
+      "learning_rate": 4.978452759935535e-05,
+      "loss": 3.0318,
+      "step": 3070
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.773056983947754,
+      "learning_rate": 4.978243969953806e-05,
+      "loss": 3.3087,
+      "step": 3080
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.788400650024414,
+      "learning_rate": 4.9780341776901696e-05,
+      "loss": 2.9589,
+      "step": 3090
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.815254211425781,
+      "learning_rate": 4.9778233832294724e-05,
+      "loss": 3.1083,
+      "step": 3100
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.483798027038574,
+      "learning_rate": 4.9776115866569654e-05,
+      "loss": 3.1023,
+      "step": 3110
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.394237995147705,
+      "learning_rate": 4.9773987880583075e-05,
+      "loss": 3.0322,
+      "step": 3120
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.961243152618408,
+      "learning_rate": 4.97718498751956e-05,
+      "loss": 3.0204,
+      "step": 3130
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.827451229095459,
+      "learning_rate": 4.976970185127191e-05,
+      "loss": 3.1435,
+      "step": 3140
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.503066539764404,
+      "learning_rate": 4.9767543809680744e-05,
+      "loss": 3.0807,
+      "step": 3150
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.9387288093566895,
+      "learning_rate": 4.9765375751294874e-05,
+      "loss": 2.7987,
+      "step": 3160
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.066097259521484,
+      "learning_rate": 4.9763197676991134e-05,
+      "loss": 3.0099,
+      "step": 3170
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 7.227752685546875,
+      "learning_rate": 4.976100958765042e-05,
+      "loss": 2.9368,
+      "step": 3180
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.03649377822876,
+      "learning_rate": 4.975881148415765e-05,
+      "loss": 2.9237,
+      "step": 3190
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.1522979736328125,
+      "learning_rate": 4.9756603367401814e-05,
+      "loss": 2.96,
+      "step": 3200
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.068181037902832,
+      "learning_rate": 4.975438523827594e-05,
+      "loss": 2.9398,
+      "step": 3210
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.104649066925049,
+      "learning_rate": 4.975215709767712e-05,
+      "loss": 2.781,
+      "step": 3220
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 7.2377471923828125,
+      "learning_rate": 4.974991894650649e-05,
+      "loss": 3.2773,
+      "step": 3230
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.730634689331055,
+      "learning_rate": 4.974767078566922e-05,
+      "loss": 3.0184,
+      "step": 3240
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.900280952453613,
+      "learning_rate": 4.974541261607454e-05,
+      "loss": 2.9432,
+      "step": 3250
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.755638122558594,
+      "learning_rate": 4.974314443863573e-05,
+      "loss": 2.9988,
+      "step": 3260
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.039461612701416,
+      "learning_rate": 4.974086625427011e-05,
+      "loss": 3.0655,
+      "step": 3270
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.955098628997803,
+      "learning_rate": 4.9738578063899066e-05,
+      "loss": 2.822,
+      "step": 3280
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.892226696014404,
+      "learning_rate": 4.9736279868447996e-05,
+      "loss": 2.9688,
+      "step": 3290
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.9840497970581055,
+      "learning_rate": 4.9733971668846376e-05,
+      "loss": 2.9008,
+      "step": 3300
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.418862819671631,
+      "learning_rate": 4.973165346602771e-05,
+      "loss": 2.9117,
+      "step": 3310
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.666804790496826,
+      "learning_rate": 4.972932526092956e-05,
+      "loss": 3.0392,
+      "step": 3320
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.902047634124756,
+      "learning_rate": 4.972698705449353e-05,
+      "loss": 3.0601,
+      "step": 3330
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.236441612243652,
+      "learning_rate": 4.972463884766525e-05,
+      "loss": 2.961,
+      "step": 3340
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.548509120941162,
+      "learning_rate": 4.972228064139444e-05,
+      "loss": 2.8966,
+      "step": 3350
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.943577766418457,
+      "learning_rate": 4.9719912436634796e-05,
+      "loss": 3.0408,
+      "step": 3360
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.531796455383301,
+      "learning_rate": 4.971753423434413e-05,
+      "loss": 2.8598,
+      "step": 3370
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.425514221191406,
+      "learning_rate": 4.971514603548425e-05,
+      "loss": 3.0003,
+      "step": 3380
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.9626593589782715,
+      "learning_rate": 4.9712747841021024e-05,
+      "loss": 3.0137,
+      "step": 3390
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.935604572296143,
+      "learning_rate": 4.971033965192435e-05,
+      "loss": 2.9509,
+      "step": 3400
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 6.610740661621094,
+      "learning_rate": 4.9708163737130644e-05,
+      "loss": 3.0231,
+      "step": 3410
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 6.248142242431641,
+      "learning_rate": 4.970573656091701e-05,
+      "loss": 2.7345,
+      "step": 3420
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.92423677444458,
+      "learning_rate": 4.9703299392905516e-05,
+      "loss": 2.9733,
+      "step": 3430
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.833972454071045,
+      "learning_rate": 4.9700852234081843e-05,
+      "loss": 2.8882,
+      "step": 3440
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.712262153625488,
+      "learning_rate": 4.96983950854357e-05,
+      "loss": 2.9849,
+      "step": 3450
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.70693826675415,
+      "learning_rate": 4.969592794796083e-05,
+      "loss": 2.7034,
+      "step": 3460
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 6.603397369384766,
+      "learning_rate": 4.969345082265503e-05,
+      "loss": 2.9441,
+      "step": 3470
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.455355644226074,
+      "learning_rate": 4.9690963710520125e-05,
+      "loss": 3.0818,
+      "step": 3480
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.410728454589844,
+      "learning_rate": 4.968846661256199e-05,
+      "loss": 2.7837,
+      "step": 3490
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.591252326965332,
+      "learning_rate": 4.9685959529790526e-05,
+      "loss": 2.8103,
+      "step": 3500
+    },
+    {
+      "epoch": 0.14,
+      "eval_loss": 2.493670701980591,
+      "eval_runtime": 423.4603,
+      "eval_samples_per_second": 50.113,
+      "eval_steps_per_second": 0.392,
+      "step": 3500
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 6.1827216148376465,
+      "learning_rate": 4.9683442463219674e-05,
+      "loss": 2.9208,
+      "step": 3510
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.817008018493652,
+      "learning_rate": 4.968091541386742e-05,
+      "loss": 2.9904,
+      "step": 3520
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.470141887664795,
+      "learning_rate": 4.96783783827558e-05,
+      "loss": 2.7532,
+      "step": 3530
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 6.129313945770264,
+      "learning_rate": 4.967583137091085e-05,
+      "loss": 3.1164,
+      "step": 3540
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.951664924621582,
+      "learning_rate": 4.967327437936268e-05,
+      "loss": 3.0038,
+      "step": 3550
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.184429168701172,
+      "learning_rate": 4.96707074091454e-05,
+      "loss": 2.8789,
+      "step": 3560
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.133409023284912,
+      "learning_rate": 4.9668130461297194e-05,
+      "loss": 2.8137,
+      "step": 3570
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.30552339553833,
+      "learning_rate": 4.9665543536860255e-05,
+      "loss": 2.764,
+      "step": 3580
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.916990280151367,
+      "learning_rate": 4.966294663688083e-05,
+      "loss": 2.8359,
+      "step": 3590
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.573614597320557,
+      "learning_rate": 4.966033976240916e-05,
+      "loss": 2.9216,
+      "step": 3600
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.318258285522461,
+      "learning_rate": 4.965772291449958e-05,
+      "loss": 2.8818,
+      "step": 3610
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 6.087459564208984,
+      "learning_rate": 4.965509609421042e-05,
+      "loss": 2.6802,
+      "step": 3620
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.408172607421875,
+      "learning_rate": 4.9652459302604046e-05,
+      "loss": 2.7713,
+      "step": 3630
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.441077709197998,
+      "learning_rate": 4.964981254074686e-05,
+      "loss": 2.8727,
+      "step": 3640
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 6.799238681793213,
+      "learning_rate": 4.9647155809709306e-05,
+      "loss": 2.9911,
+      "step": 3650
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.362485885620117,
+      "learning_rate": 4.9644489110565845e-05,
+      "loss": 2.8088,
+      "step": 3660
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.419576168060303,
+      "learning_rate": 4.964181244439498e-05,
+      "loss": 2.8133,
+      "step": 3670
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.19252347946167,
+      "learning_rate": 4.963912581227924e-05,
+      "loss": 2.8429,
+      "step": 3680
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.425354480743408,
+      "learning_rate": 4.963642921530518e-05,
+      "loss": 2.8272,
+      "step": 3690
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.866875171661377,
+      "learning_rate": 4.9633722654563405e-05,
+      "loss": 2.816,
+      "step": 3700
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.425843238830566,
+      "learning_rate": 4.963100613114852e-05,
+      "loss": 2.8552,
+      "step": 3710
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.268444538116455,
+      "learning_rate": 4.962827964615919e-05,
+      "loss": 2.8642,
+      "step": 3720
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.511022090911865,
+      "learning_rate": 4.962554320069808e-05,
+      "loss": 2.8388,
+      "step": 3730
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 8.639039993286133,
+      "learning_rate": 4.9622796795871904e-05,
+      "loss": 2.809,
+      "step": 3740
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.79989767074585,
+      "learning_rate": 4.9620040432791395e-05,
+      "loss": 3.0316,
+      "step": 3750
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.568778991699219,
+      "learning_rate": 4.961727411257131e-05,
+      "loss": 2.7867,
+      "step": 3760
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.364297389984131,
+      "learning_rate": 4.961449783633045e-05,
+      "loss": 3.0213,
+      "step": 3770
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.523791790008545,
+      "learning_rate": 4.9611711605191625e-05,
+      "loss": 2.7875,
+      "step": 3780
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.39138650894165,
+      "learning_rate": 4.9608915420281675e-05,
+      "loss": 2.7144,
+      "step": 3790
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.800009250640869,
+      "learning_rate": 4.9606109282731463e-05,
+      "loss": 2.8091,
+      "step": 3800
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.470329284667969,
+      "learning_rate": 4.96032931936759e-05,
+      "loss": 2.8055,
+      "step": 3810
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.461717128753662,
+      "learning_rate": 4.9600467154253885e-05,
+      "loss": 2.8633,
+      "step": 3820
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.095701217651367,
+      "learning_rate": 4.959763116560836e-05,
+      "loss": 2.9004,
+      "step": 3830
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.429464340209961,
+      "learning_rate": 4.95947852288863e-05,
+      "loss": 2.6428,
+      "step": 3840
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.361551761627197,
+      "learning_rate": 4.95919293452387e-05,
+      "loss": 2.867,
+      "step": 3850
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.917137622833252,
+      "learning_rate": 4.958906351582054e-05,
+      "loss": 2.7266,
+      "step": 3860
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.074246406555176,
+      "learning_rate": 4.958618774179089e-05,
+      "loss": 2.8333,
+      "step": 3870
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.167778968811035,
+      "learning_rate": 4.9583302024312796e-05,
+      "loss": 2.7009,
+      "step": 3880
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.73693323135376,
+      "learning_rate": 4.9580406364553325e-05,
+      "loss": 2.5967,
+      "step": 3890
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.558803558349609,
+      "learning_rate": 4.957750076368357e-05,
+      "loss": 2.6763,
+      "step": 3900
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 7.1173481941223145,
+      "learning_rate": 4.957458522287868e-05,
+      "loss": 2.8593,
+      "step": 3910
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 7.378798484802246,
+      "learning_rate": 4.9571659743317754e-05,
+      "loss": 2.7876,
+      "step": 3920
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.375428199768066,
+      "learning_rate": 4.956872432618399e-05,
+      "loss": 2.7443,
+      "step": 3930
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.509357452392578,
+      "learning_rate": 4.9565778972664534e-05,
+      "loss": 2.8787,
+      "step": 3940
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.430800437927246,
+      "learning_rate": 4.95628236839506e-05,
+      "loss": 2.609,
+      "step": 3950
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.8563618659973145,
+      "learning_rate": 4.95598584612374e-05,
+      "loss": 2.7912,
+      "step": 3960
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.738502502441406,
+      "learning_rate": 4.9556883305724155e-05,
+      "loss": 2.8222,
+      "step": 3970
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.658450603485107,
+      "learning_rate": 4.955389821861412e-05,
+      "loss": 2.6731,
+      "step": 3980
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.250921726226807,
+      "learning_rate": 4.955090320111457e-05,
+      "loss": 2.6865,
+      "step": 3990
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.725744247436523,
+      "learning_rate": 4.954789825443678e-05,
+      "loss": 2.6249,
+      "step": 4000
+    },
+    {
+      "epoch": 0.16,
+      "eval_loss": 2.342557668685913,
+      "eval_runtime": 423.1443,
+      "eval_samples_per_second": 50.151,
+      "eval_steps_per_second": 0.392,
+      "step": 4000
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.008103847503662,
+      "learning_rate": 4.954488337979604e-05,
+      "loss": 2.6645,
+      "step": 4010
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.027274131774902,
+      "learning_rate": 4.9541858578411674e-05,
+      "loss": 2.8588,
+      "step": 4020
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.368979454040527,
+      "learning_rate": 4.9538823851507006e-05,
+      "loss": 2.7103,
+      "step": 4030
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.687747001647949,
+      "learning_rate": 4.953577920030937e-05,
+      "loss": 2.5958,
+      "step": 4040
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.669676780700684,
+      "learning_rate": 4.953272462605013e-05,
+      "loss": 2.7199,
+      "step": 4050
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.0332112312316895,
+      "learning_rate": 4.952966012996466e-05,
+      "loss": 2.62,
+      "step": 4060
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.005265235900879,
+      "learning_rate": 4.952658571329233e-05,
+      "loss": 2.5974,
+      "step": 4070
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.806674003601074,
+      "learning_rate": 4.952350137727653e-05,
+      "loss": 2.9317,
+      "step": 4080
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.117702484130859,
+      "learning_rate": 4.9520407123164676e-05,
+      "loss": 2.6551,
+      "step": 4090
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.5031023025512695,
+      "learning_rate": 4.9517302952208174e-05,
+      "loss": 2.7362,
+      "step": 4100
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.63228702545166,
+      "learning_rate": 4.951418886566247e-05,
+      "loss": 2.6743,
+      "step": 4110
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.421300888061523,
+      "learning_rate": 4.951106486478698e-05,
+      "loss": 2.6827,
+      "step": 4120
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.456374168395996,
+      "learning_rate": 4.950793095084516e-05,
+      "loss": 2.7266,
+      "step": 4130
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.39730978012085,
+      "learning_rate": 4.950478712510446e-05,
+      "loss": 2.8165,
+      "step": 4140
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.817816734313965,
+      "learning_rate": 4.9501633388836354e-05,
+      "loss": 2.6275,
+      "step": 4150
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 7.556663990020752,
+      "learning_rate": 4.949846974331631e-05,
+      "loss": 2.6368,
+      "step": 4160
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 6.171590805053711,
+      "learning_rate": 4.949529618982381e-05,
+      "loss": 2.8336,
+      "step": 4170
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 6.013977527618408,
+      "learning_rate": 4.9492112729642346e-05,
+      "loss": 2.6988,
+      "step": 4180
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.861149311065674,
+      "learning_rate": 4.948891936405941e-05,
+      "loss": 2.8883,
+      "step": 4190
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.240298271179199,
+      "learning_rate": 4.948571609436649e-05,
+      "loss": 2.8353,
+      "step": 4200
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.673245429992676,
+      "learning_rate": 4.94825029218591e-05,
+      "loss": 2.6835,
+      "step": 4210
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.680504322052002,
+      "learning_rate": 4.947927984783674e-05,
+      "loss": 2.594,
+      "step": 4220
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.967173099517822,
+      "learning_rate": 4.947604687360295e-05,
+      "loss": 2.5797,
+      "step": 4230
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 6.29856538772583,
+      "learning_rate": 4.947280400046525e-05,
+      "loss": 2.6197,
+      "step": 4240
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.518858432769775,
+      "learning_rate": 4.9469551229735126e-05,
+      "loss": 2.6727,
+      "step": 4250
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.493609428405762,
+      "learning_rate": 4.946628856272814e-05,
+      "loss": 2.521,
+      "step": 4260
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.581683158874512,
+      "learning_rate": 4.9463016000763794e-05,
+      "loss": 2.4878,
+      "step": 4270
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.769138336181641,
+      "learning_rate": 4.945973354516564e-05,
+      "loss": 2.6396,
+      "step": 4280
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.217662811279297,
+      "learning_rate": 4.9456441197261195e-05,
+      "loss": 2.3943,
+      "step": 4290
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.986959457397461,
+      "learning_rate": 4.9453138958381994e-05,
+      "loss": 2.9015,
+      "step": 4300
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.872792720794678,
+      "learning_rate": 4.944982682986357e-05,
+      "loss": 2.7326,
+      "step": 4310
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.723078727722168,
+      "learning_rate": 4.944650481304545e-05,
+      "loss": 2.7212,
+      "step": 4320
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 6.189296722412109,
+      "learning_rate": 4.944317290927117e-05,
+      "loss": 2.8079,
+      "step": 4330
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.358461856842041,
+      "learning_rate": 4.943983111988827e-05,
+      "loss": 2.6931,
+      "step": 4340
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.47238826751709,
+      "learning_rate": 4.943647944624826e-05,
+      "loss": 2.5772,
+      "step": 4350
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.701056957244873,
+      "learning_rate": 4.943311788970667e-05,
+      "loss": 2.5407,
+      "step": 4360
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 6.032464504241943,
+      "learning_rate": 4.9429746451623024e-05,
+      "loss": 2.6685,
+      "step": 4370
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.141815185546875,
+      "learning_rate": 4.9426365133360834e-05,
+      "loss": 2.7409,
+      "step": 4380
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 6.483497619628906,
+      "learning_rate": 4.942297393628763e-05,
+      "loss": 2.5305,
+      "step": 4390
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 6.223423957824707,
+      "learning_rate": 4.941957286177491e-05,
+      "loss": 2.8228,
+      "step": 4400
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.8881402015686035,
+      "learning_rate": 4.9416161911198176e-05,
+      "loss": 2.5842,
+      "step": 4410
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.5558929443359375,
+      "learning_rate": 4.941274108593693e-05,
+      "loss": 2.5049,
+      "step": 4420
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 7.3317694664001465,
+      "learning_rate": 4.940931038737468e-05,
+      "loss": 2.5361,
+      "step": 4430
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 6.913518905639648,
+      "learning_rate": 4.940586981689888e-05,
+      "loss": 2.7733,
+      "step": 4440
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 8.628571510314941,
+      "learning_rate": 4.940241937590102e-05,
+      "loss": 2.642,
+      "step": 4450
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 6.870291233062744,
+      "learning_rate": 4.939895906577658e-05,
+      "loss": 2.9244,
+      "step": 4460
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.8803253173828125,
+      "learning_rate": 4.939548888792502e-05,
+      "loss": 2.6324,
+      "step": 4470
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 6.114381790161133,
+      "learning_rate": 4.939200884374979e-05,
+      "loss": 2.674,
+      "step": 4480
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 7.145431041717529,
+      "learning_rate": 4.938851893465832e-05,
+      "loss": 2.6255,
+      "step": 4490
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 6.35526704788208,
+      "learning_rate": 4.938501916206206e-05,
+      "loss": 2.8467,
+      "step": 4500
+    },
+    {
+      "epoch": 0.18,
+      "eval_loss": 2.259162664413452,
+      "eval_runtime": 421.9557,
+      "eval_samples_per_second": 50.292,
+      "eval_steps_per_second": 0.393,
+      "step": 4500
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 6.096010208129883,
+      "learning_rate": 4.938150952737642e-05,
+      "loss": 2.6732,
+      "step": 4510
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 4.902897357940674,
+      "learning_rate": 4.937799003202081e-05,
+      "loss": 2.5127,
+      "step": 4520
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.106553077697754,
+      "learning_rate": 4.9374460677418635e-05,
+      "loss": 2.5394,
+      "step": 4530
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.985318183898926,
+      "learning_rate": 4.937092146499727e-05,
+      "loss": 2.6342,
+      "step": 4540
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.647487163543701,
+      "learning_rate": 4.9367372396188095e-05,
+      "loss": 2.7253,
+      "step": 4550
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.041297435760498,
+      "learning_rate": 4.9363813472426465e-05,
+      "loss": 2.7264,
+      "step": 4560
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 6.295587062835693,
+      "learning_rate": 4.9360244695151716e-05,
+      "loss": 2.6379,
+      "step": 4570
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.751054286956787,
+      "learning_rate": 4.935666606580719e-05,
+      "loss": 2.6497,
+      "step": 4580
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.367273330688477,
+      "learning_rate": 4.935307758584019e-05,
+      "loss": 2.3892,
+      "step": 4590
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.491433143615723,
+      "learning_rate": 4.9349479256702023e-05,
+      "loss": 2.4803,
+      "step": 4600
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.676270484924316,
+      "learning_rate": 4.934587107984796e-05,
+      "loss": 2.4972,
+      "step": 4610
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.060196876525879,
+      "learning_rate": 4.934225305673728e-05,
+      "loss": 2.4878,
+      "step": 4620
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.352468967437744,
+      "learning_rate": 4.93386251888332e-05,
+      "loss": 2.4713,
+      "step": 4630
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 8.105636596679688,
+      "learning_rate": 4.9334987477602975e-05,
+      "loss": 2.7637,
+      "step": 4640
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.352108478546143,
+      "learning_rate": 4.933133992451779e-05,
+      "loss": 2.5642,
+      "step": 4650
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.43104362487793,
+      "learning_rate": 4.932768253105286e-05,
+      "loss": 2.6106,
+      "step": 4660
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.641717433929443,
+      "learning_rate": 4.932401529868733e-05,
+      "loss": 2.6049,
+      "step": 4670
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.041056156158447,
+      "learning_rate": 4.932033822890436e-05,
+      "loss": 2.512,
+      "step": 4680
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.870728015899658,
+      "learning_rate": 4.9316651323191067e-05,
+      "loss": 2.4649,
+      "step": 4690
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 6.347024917602539,
+      "learning_rate": 4.931295458303856e-05,
+      "loss": 2.5996,
+      "step": 4700
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.286810398101807,
+      "learning_rate": 4.9309248009941914e-05,
+      "loss": 2.5899,
+      "step": 4710
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 6.839949131011963,
+      "learning_rate": 4.93055316054002e-05,
+      "loss": 2.5646,
+      "step": 4720
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 6.071991920471191,
+      "learning_rate": 4.930180537091644e-05,
+      "loss": 2.7168,
+      "step": 4730
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.042398452758789,
+      "learning_rate": 4.929806930799765e-05,
+      "loss": 2.7169,
+      "step": 4740
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.924988269805908,
+      "learning_rate": 4.9294323418154805e-05,
+      "loss": 2.4587,
+      "step": 4750
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 6.021846294403076,
+      "learning_rate": 4.929056770290288e-05,
+      "loss": 2.5327,
+      "step": 4760
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.043169975280762,
+      "learning_rate": 4.928680216376079e-05,
+      "loss": 2.4268,
+      "step": 4770
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 7.106345176696777,
+      "learning_rate": 4.9283026802251454e-05,
+      "loss": 2.5116,
+      "step": 4780
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.8474650382995605,
+      "learning_rate": 4.927924161990175e-05,
+      "loss": 2.5163,
+      "step": 4790
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.736364841461182,
+      "learning_rate": 4.9275446618242516e-05,
+      "loss": 2.4918,
+      "step": 4800
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.512024879455566,
+      "learning_rate": 4.927164179880858e-05,
+      "loss": 2.5756,
+      "step": 4810
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.911081790924072,
+      "learning_rate": 4.926782716313874e-05,
+      "loss": 2.4661,
+      "step": 4820
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.932026386260986,
+      "learning_rate": 4.9264002712775745e-05,
+      "loss": 2.5059,
+      "step": 4830
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 6.116228103637695,
+      "learning_rate": 4.9260168449266335e-05,
+      "loss": 2.4183,
+      "step": 4840
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.023027420043945,
+      "learning_rate": 4.925632437416121e-05,
+      "loss": 2.4144,
+      "step": 4850
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.4858832359313965,
+      "learning_rate": 4.925247048901504e-05,
+      "loss": 2.4708,
+      "step": 4860
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 6.690423965454102,
+      "learning_rate": 4.924860679538645e-05,
+      "loss": 2.7061,
+      "step": 4870
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 6.818970203399658,
+      "learning_rate": 4.924473329483806e-05,
+      "loss": 2.6152,
+      "step": 4880
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.238198280334473,
+      "learning_rate": 4.924084998893642e-05,
+      "loss": 2.3184,
+      "step": 4890
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.587952613830566,
+      "learning_rate": 4.923695687925207e-05,
+      "loss": 2.6381,
+      "step": 4900
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.2837066650390625,
+      "learning_rate": 4.923305396735952e-05,
+      "loss": 2.52,
+      "step": 4910
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.908483028411865,
+      "learning_rate": 4.9229141254837226e-05,
+      "loss": 2.675,
+      "step": 4920
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.803688049316406,
+      "learning_rate": 4.922521874326761e-05,
+      "loss": 2.6054,
+      "step": 4930
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.124865531921387,
+      "learning_rate": 4.922128643423707e-05,
+      "loss": 2.6076,
+      "step": 4940
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.534984111785889,
+      "learning_rate": 4.921734432933596e-05,
+      "loss": 2.4958,
+      "step": 4950
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 7.013035297393799,
+      "learning_rate": 4.921339243015858e-05,
+      "loss": 2.6099,
+      "step": 4960
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.598790645599365,
+      "learning_rate": 4.920943073830322e-05,
+      "loss": 2.493,
+      "step": 4970
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.958088397979736,
+      "learning_rate": 4.920545925537212e-05,
+      "loss": 2.5689,
+      "step": 4980
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.180896282196045,
+      "learning_rate": 4.9201477982971466e-05,
+      "loss": 2.6138,
+      "step": 4990
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.9859209060668945,
+      "learning_rate": 4.919748692271141e-05,
+      "loss": 2.5003,
+      "step": 5000
+    },
+    {
+      "epoch": 0.2,
+      "eval_loss": 2.1625306606292725,
+      "eval_runtime": 422.5397,
+      "eval_samples_per_second": 50.223,
+      "eval_steps_per_second": 0.393,
+      "step": 5000
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.54530143737793,
+      "learning_rate": 4.9193486076206075e-05,
+      "loss": 2.446,
+      "step": 5010
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.814847469329834,
+      "learning_rate": 4.918947544507353e-05,
+      "loss": 2.3595,
+      "step": 5020
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.315622806549072,
+      "learning_rate": 4.91854550309358e-05,
+      "loss": 2.6138,
+      "step": 5030
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.739955425262451,
+      "learning_rate": 4.9181424835418874e-05,
+      "loss": 2.5347,
+      "step": 5040
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.838977336883545,
+      "learning_rate": 4.91773848601527e-05,
+      "loss": 2.3811,
+      "step": 5050
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.2752766609191895,
+      "learning_rate": 4.9173335106771154e-05,
+      "loss": 2.3222,
+      "step": 5060
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 7.31117057800293,
+      "learning_rate": 4.916927557691211e-05,
+      "loss": 2.3912,
+      "step": 5070
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.800614833831787,
+      "learning_rate": 4.916520627221736e-05,
+      "loss": 2.5183,
+      "step": 5080
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.574156284332275,
+      "learning_rate": 4.9161127194332655e-05,
+      "loss": 2.4722,
+      "step": 5090
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.964874744415283,
+      "learning_rate": 4.915703834490773e-05,
+      "loss": 2.6562,
+      "step": 5100
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.67961311340332,
+      "learning_rate": 4.915293972559623e-05,
+      "loss": 2.6234,
+      "step": 5110
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.129438877105713,
+      "learning_rate": 4.9148831338055766e-05,
+      "loss": 2.3889,
+      "step": 5120
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.661647796630859,
+      "learning_rate": 4.9144713183947915e-05,
+      "loss": 2.337,
+      "step": 5130
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.4586567878723145,
+      "learning_rate": 4.914058526493819e-05,
+      "loss": 2.525,
+      "step": 5140
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.5740532875061035,
+      "learning_rate": 4.913644758269604e-05,
+      "loss": 2.4817,
+      "step": 5150
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.7009124755859375,
+      "learning_rate": 4.9132300138894895e-05,
+      "loss": 2.3586,
+      "step": 5160
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.846322536468506,
+      "learning_rate": 4.912814293521211e-05,
+      "loss": 2.4756,
+      "step": 5170
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.497846603393555,
+      "learning_rate": 4.9123975973328986e-05,
+      "loss": 2.3421,
+      "step": 5180
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.7938971519470215,
+      "learning_rate": 4.911979925493079e-05,
+      "loss": 2.609,
+      "step": 5190
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.914577960968018,
+      "learning_rate": 4.91156127817067e-05,
+      "loss": 2.5344,
+      "step": 5200
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.262577533721924,
+      "learning_rate": 4.911141655534989e-05,
+      "loss": 2.4138,
+      "step": 5210
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.629760265350342,
+      "learning_rate": 4.9107210577557415e-05,
+      "loss": 2.4292,
+      "step": 5220
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 18.421449661254883,
+      "learning_rate": 4.9102994850030344e-05,
+      "loss": 2.594,
+      "step": 5230
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.440621376037598,
+      "learning_rate": 4.909876937447363e-05,
+      "loss": 2.338,
+      "step": 5240
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.139665603637695,
+      "learning_rate": 4.90945341525962e-05,
+      "loss": 2.4571,
+      "step": 5250
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.897266864776611,
+      "learning_rate": 4.909028918611091e-05,
+      "loss": 2.4774,
+      "step": 5260
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.971827507019043,
+      "learning_rate": 4.908603447673457e-05,
+      "loss": 2.5588,
+      "step": 5270
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.420186996459961,
+      "learning_rate": 4.9081770026187914e-05,
+      "loss": 2.4299,
+      "step": 5280
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.321455478668213,
+      "learning_rate": 4.907749583619562e-05,
+      "loss": 2.4522,
+      "step": 5290
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.8206400871276855,
+      "learning_rate": 4.907321190848632e-05,
+      "loss": 2.5081,
+      "step": 5300
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.017814636230469,
+      "learning_rate": 4.9068918244792565e-05,
+      "loss": 2.2797,
+      "step": 5310
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.493027210235596,
+      "learning_rate": 4.906461484685085e-05,
+      "loss": 2.4124,
+      "step": 5320
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.415233612060547,
+      "learning_rate": 4.906030171640163e-05,
+      "loss": 2.4818,
+      "step": 5330
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.612936496734619,
+      "learning_rate": 4.905597885518923e-05,
+      "loss": 2.5769,
+      "step": 5340
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.570766448974609,
+      "learning_rate": 4.9051646264962005e-05,
+      "loss": 2.3302,
+      "step": 5350
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.15294885635376,
+      "learning_rate": 4.904730394747216e-05,
+      "loss": 2.3181,
+      "step": 5360
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.568897247314453,
+      "learning_rate": 4.904295190447587e-05,
+      "loss": 2.4157,
+      "step": 5370
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.8973236083984375,
+      "learning_rate": 4.903859013773325e-05,
+      "loss": 2.5564,
+      "step": 5380
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.484323024749756,
+      "learning_rate": 4.903421864900835e-05,
+      "loss": 2.3377,
+      "step": 5390
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.30608606338501,
+      "learning_rate": 4.902983744006913e-05,
+      "loss": 2.3119,
+      "step": 5400
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.372623920440674,
+      "learning_rate": 4.902544651268749e-05,
+      "loss": 2.3764,
+      "step": 5410
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.923567771911621,
+      "learning_rate": 4.902104586863927e-05,
+      "loss": 2.452,
+      "step": 5420
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 6.1613664627075195,
+      "learning_rate": 4.901663550970422e-05,
+      "loss": 2.5677,
+      "step": 5430
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.322874069213867,
+      "learning_rate": 4.901221543766604e-05,
+      "loss": 2.3442,
+      "step": 5440
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.805524826049805,
+      "learning_rate": 4.900778565431236e-05,
+      "loss": 2.377,
+      "step": 5450
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.61544942855835,
+      "learning_rate": 4.900334616143471e-05,
+      "loss": 2.2881,
+      "step": 5460
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 9.518655776977539,
+      "learning_rate": 4.8998896960828576e-05,
+      "loss": 2.5395,
+      "step": 5470
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.4181108474731445,
+      "learning_rate": 4.899443805429336e-05,
+      "loss": 2.5815,
+      "step": 5480
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 7.65767765045166,
+      "learning_rate": 4.8989969443632366e-05,
+      "loss": 2.383,
+      "step": 5490
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 6.252597332000732,
+      "learning_rate": 4.898549113065287e-05,
+      "loss": 2.3579,
+      "step": 5500
+    },
+    {
+      "epoch": 0.22,
+      "eval_loss": 2.0875422954559326,
+      "eval_runtime": 423.4806,
+      "eval_samples_per_second": 50.111,
+      "eval_steps_per_second": 0.392,
+      "step": 5500
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.8088860511779785,
+      "learning_rate": 4.898100311716604e-05,
+      "loss": 2.4613,
+      "step": 5510
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 6.02274227142334,
+      "learning_rate": 4.897650540498697e-05,
+      "loss": 2.2701,
+      "step": 5520
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.915895938873291,
+      "learning_rate": 4.8971997995934675e-05,
+      "loss": 2.3393,
+      "step": 5530
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.784215450286865,
+      "learning_rate": 4.896748089183211e-05,
+      "loss": 2.5954,
+      "step": 5540
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.802945613861084,
+      "learning_rate": 4.896295409450612e-05,
+      "loss": 2.2532,
+      "step": 5550
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.0984649658203125,
+      "learning_rate": 4.895841760578751e-05,
+      "loss": 2.3553,
+      "step": 5560
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.783207893371582,
+      "learning_rate": 4.8953871427510965e-05,
+      "loss": 2.445,
+      "step": 5570
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 6.820375919342041,
+      "learning_rate": 4.894931556151512e-05,
+      "loss": 2.4465,
+      "step": 5580
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 54.05234146118164,
+      "learning_rate": 4.894475000964249e-05,
+      "loss": 2.4557,
+      "step": 5590
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.997950077056885,
+      "learning_rate": 4.8940174773739564e-05,
+      "loss": 2.2733,
+      "step": 5600
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.4290642738342285,
+      "learning_rate": 4.8935589855656694e-05,
+      "loss": 2.3262,
+      "step": 5610
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.6273064613342285,
+      "learning_rate": 4.893099525724818e-05,
+      "loss": 2.4704,
+      "step": 5620
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 6.293818473815918,
+      "learning_rate": 4.892639098037221e-05,
+      "loss": 2.525,
+      "step": 5630
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 6.767632484436035,
+      "learning_rate": 4.892177702689091e-05,
+      "loss": 2.2745,
+      "step": 5640
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.506222248077393,
+      "learning_rate": 4.891715339867033e-05,
+      "loss": 2.4635,
+      "step": 5650
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.971251010894775,
+      "learning_rate": 4.8912520097580395e-05,
+      "loss": 2.5934,
+      "step": 5660
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.757556438446045,
+      "learning_rate": 4.890787712549496e-05,
+      "loss": 2.4497,
+      "step": 5670
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 8.545730590820312,
+      "learning_rate": 4.890369018346882e-05,
+      "loss": 2.3384,
+      "step": 5680
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.717365264892578,
+      "learning_rate": 4.889902884166843e-05,
+      "loss": 2.3127,
+      "step": 5690
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.706165790557861,
+      "learning_rate": 4.889435783432884e-05,
+      "loss": 2.1936,
+      "step": 5700
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.551875591278076,
+      "learning_rate": 4.888967716333917e-05,
+      "loss": 2.5815,
+      "step": 5710
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.260705471038818,
+      "learning_rate": 4.888498683059243e-05,
+      "loss": 2.3263,
+      "step": 5720
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.164261341094971,
+      "learning_rate": 4.8880286837985526e-05,
+      "loss": 2.276,
+      "step": 5730
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.247165679931641,
+      "learning_rate": 4.887557718741931e-05,
+      "loss": 2.4281,
+      "step": 5740
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.835640907287598,
+      "learning_rate": 4.88708578807985e-05,
+      "loss": 2.2763,
+      "step": 5750
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.9643354415893555,
+      "learning_rate": 4.886612892003174e-05,
+      "loss": 2.4745,
+      "step": 5760
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 7.18971061706543,
+      "learning_rate": 4.886139030703159e-05,
+      "loss": 2.4685,
+      "step": 5770
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.121529579162598,
+      "learning_rate": 4.885664204371446e-05,
+      "loss": 2.5183,
+      "step": 5780
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.581070423126221,
+      "learning_rate": 4.885188413200075e-05,
+      "loss": 2.5167,
+      "step": 5790
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.491427898406982,
+      "learning_rate": 4.8847116573814676e-05,
+      "loss": 2.383,
+      "step": 5800
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.595372676849365,
+      "learning_rate": 4.8842339371084414e-05,
+      "loss": 2.0989,
+      "step": 5810
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.8517165184021,
+      "learning_rate": 4.8837552525742004e-05,
+      "loss": 2.4912,
+      "step": 5820
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.275095462799072,
+      "learning_rate": 4.88327560397234e-05,
+      "loss": 2.3524,
+      "step": 5830
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.616918563842773,
+      "learning_rate": 4.8827949914968474e-05,
+      "loss": 1.9738,
+      "step": 5840
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 7.1793999671936035,
+      "learning_rate": 4.882313415342097e-05,
+      "loss": 2.2815,
+      "step": 5850
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.783013343811035,
+      "learning_rate": 4.881830875702852e-05,
+      "loss": 2.4371,
+      "step": 5860
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 18.31147003173828,
+      "learning_rate": 4.88134737277427e-05,
+      "loss": 2.5314,
+      "step": 5870
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.92976713180542,
+      "learning_rate": 4.880862906751893e-05,
+      "loss": 2.2746,
+      "step": 5880
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.150814056396484,
+      "learning_rate": 4.8803774778316557e-05,
+      "loss": 2.277,
+      "step": 5890
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.579446792602539,
+      "learning_rate": 4.87989108620988e-05,
+      "loss": 2.2136,
+      "step": 5900
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.303000450134277,
+      "learning_rate": 4.8794037320832806e-05,
+      "loss": 2.258,
+      "step": 5910
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.469798564910889,
+      "learning_rate": 4.878915415648957e-05,
+      "loss": 2.4185,
+      "step": 5920
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.8365960121154785,
+      "learning_rate": 4.8784261371044016e-05,
+      "loss": 2.3902,
+      "step": 5930
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.7992987632751465,
+      "learning_rate": 4.877984963973593e-05,
+      "loss": 2.3468,
+      "step": 5940
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.936422348022461,
+      "learning_rate": 4.877493857965077e-05,
+      "loss": 2.3446,
+      "step": 5950
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.544583797454834,
+      "learning_rate": 4.8770017904212525e-05,
+      "loss": 2.2751,
+      "step": 5960
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.835214138031006,
+      "learning_rate": 4.876508761541129e-05,
+      "loss": 2.3919,
+      "step": 5970
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.438685417175293,
+      "learning_rate": 4.8760147715241e-05,
+      "loss": 2.3636,
+      "step": 5980
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.63602876663208,
+      "learning_rate": 4.875519820569954e-05,
+      "loss": 2.4277,
+      "step": 5990
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.563483238220215,
+      "learning_rate": 4.8750239088788636e-05,
+      "loss": 2.2586,
+      "step": 6000
+    },
+    {
+      "epoch": 0.24,
+      "eval_loss": 2.014089584350586,
+      "eval_runtime": 422.9188,
+      "eval_samples_per_second": 50.177,
+      "eval_steps_per_second": 0.393,
+      "step": 6000
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.133357524871826,
+      "learning_rate": 4.8745270366513914e-05,
+      "loss": 2.2622,
+      "step": 6010
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 7.025454521179199,
+      "learning_rate": 4.87402920408849e-05,
+      "loss": 2.2786,
+      "step": 6020
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.6243181228637695,
+      "learning_rate": 4.873530411391498e-05,
+      "loss": 2.3901,
+      "step": 6030
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.4480109214782715,
+      "learning_rate": 4.873030658762143e-05,
+      "loss": 2.1692,
+      "step": 6040
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.633693695068359,
+      "learning_rate": 4.8725299464025414e-05,
+      "loss": 2.198,
+      "step": 6050
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.928077220916748,
+      "learning_rate": 4.872028274515198e-05,
+      "loss": 2.0226,
+      "step": 6060
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.940571308135986,
+      "learning_rate": 4.871525643303004e-05,
+      "loss": 2.5249,
+      "step": 6070
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.45697021484375,
+      "learning_rate": 4.871022052969241e-05,
+      "loss": 2.184,
+      "step": 6080
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.87355899810791,
+      "learning_rate": 4.870517503717576e-05,
+      "loss": 2.4252,
+      "step": 6090
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.142642021179199,
+      "learning_rate": 4.8700119957520653e-05,
+      "loss": 2.2134,
+      "step": 6100
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.03863000869751,
+      "learning_rate": 4.869505529277154e-05,
+      "loss": 2.4,
+      "step": 6110
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.4546003341674805,
+      "learning_rate": 4.8689981044976715e-05,
+      "loss": 2.2927,
+      "step": 6120
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 10.309746742248535,
+      "learning_rate": 4.868489721618838e-05,
+      "loss": 2.3127,
+      "step": 6130
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.640440464019775,
+      "learning_rate": 4.867980380846259e-05,
+      "loss": 2.1445,
+      "step": 6140
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 7.428249835968018,
+      "learning_rate": 4.86747008238593e-05,
+      "loss": 2.4585,
+      "step": 6150
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.020784854888916,
+      "learning_rate": 4.86695882644423e-05,
+      "loss": 2.2498,
+      "step": 6160
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.098503112792969,
+      "learning_rate": 4.8664466132279294e-05,
+      "loss": 2.1231,
+      "step": 6170
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.907447814941406,
+      "learning_rate": 4.8659334429441825e-05,
+      "loss": 2.2518,
+      "step": 6180
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 38.992706298828125,
+      "learning_rate": 4.8654193158005336e-05,
+      "loss": 2.4063,
+      "step": 6190
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.094198226928711,
+      "learning_rate": 4.86490423200491e-05,
+      "loss": 2.1258,
+      "step": 6200
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 6.0221686363220215,
+      "learning_rate": 4.864388191765629e-05,
+      "loss": 2.5631,
+      "step": 6210
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.385717391967773,
+      "learning_rate": 4.863871195291395e-05,
+      "loss": 2.4827,
+      "step": 6220
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 6.672618389129639,
+      "learning_rate": 4.863353242791297e-05,
+      "loss": 2.2295,
+      "step": 6230
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 6.568421840667725,
+      "learning_rate": 4.8628343344748116e-05,
+      "loss": 2.2235,
+      "step": 6240
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.92976713180542,
+      "learning_rate": 4.8623144705518034e-05,
+      "loss": 2.2561,
+      "step": 6250
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 7.64150857925415,
+      "learning_rate": 4.8617936512325214e-05,
+      "loss": 2.5747,
+      "step": 6260
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.804096221923828,
+      "learning_rate": 4.861271876727601e-05,
+      "loss": 2.1542,
+      "step": 6270
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.928376197814941,
+      "learning_rate": 4.860749147248066e-05,
+      "loss": 2.2101,
+      "step": 6280
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 6.408993244171143,
+      "learning_rate": 4.8602254630053235e-05,
+      "loss": 2.3248,
+      "step": 6290
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.5954813957214355,
+      "learning_rate": 4.8597008242111694e-05,
+      "loss": 2.3062,
+      "step": 6300
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 6.591384410858154,
+      "learning_rate": 4.859175231077785e-05,
+      "loss": 2.2668,
+      "step": 6310
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 6.448415279388428,
+      "learning_rate": 4.858648683817736e-05,
+      "loss": 2.2404,
+      "step": 6320
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.797817707061768,
+      "learning_rate": 4.8581211826439765e-05,
+      "loss": 2.3924,
+      "step": 6330
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.555346965789795,
+      "learning_rate": 4.8575927277698435e-05,
+      "loss": 2.2588,
+      "step": 6340
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 6.0208821296691895,
+      "learning_rate": 4.857063319409062e-05,
+      "loss": 2.2423,
+      "step": 6350
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.1246466636657715,
+      "learning_rate": 4.8565329577757426e-05,
+      "loss": 2.1666,
+      "step": 6360
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 12.975321769714355,
+      "learning_rate": 4.856001643084378e-05,
+      "loss": 2.253,
+      "step": 6370
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.758956432342529,
+      "learning_rate": 4.855469375549853e-05,
+      "loss": 2.061,
+      "step": 6380
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.913463115692139,
+      "learning_rate": 4.8549361553874295e-05,
+      "loss": 2.2273,
+      "step": 6390
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 8.23082160949707,
+      "learning_rate": 4.854401982812762e-05,
+      "loss": 2.599,
+      "step": 6400
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 8.17214298248291,
+      "learning_rate": 4.853866858041887e-05,
+      "loss": 2.5361,
+      "step": 6410
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 9.37897777557373,
+      "learning_rate": 4.853330781291224e-05,
+      "loss": 2.3414,
+      "step": 6420
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.9388427734375,
+      "learning_rate": 4.852793752777582e-05,
+      "loss": 2.3078,
+      "step": 6430
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.1688055992126465,
+      "learning_rate": 4.8522557727181517e-05,
+      "loss": 2.2146,
+      "step": 6440
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 13.751556396484375,
+      "learning_rate": 4.85171684133051e-05,
+      "loss": 2.3457,
+      "step": 6450
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.383643627166748,
+      "learning_rate": 4.851176958832618e-05,
+      "loss": 2.4478,
+      "step": 6460
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.072232723236084,
+      "learning_rate": 4.850636125442821e-05,
+      "loss": 2.0503,
+      "step": 6470
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.105084419250488,
+      "learning_rate": 4.850094341379851e-05,
+      "loss": 2.4092,
+      "step": 6480
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.889256477355957,
+      "learning_rate": 4.8495516068628216e-05,
+      "loss": 2.1762,
+      "step": 6490
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 10.305624008178711,
+      "learning_rate": 4.849007922111233e-05,
+      "loss": 2.2757,
+      "step": 6500
+    },
+    {
+      "epoch": 0.26,
+      "eval_loss": 1.962647795677185,
+      "eval_runtime": 422.8287,
+      "eval_samples_per_second": 50.188,
+      "eval_steps_per_second": 0.393,
+      "step": 6500
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 12.288636207580566,
+      "learning_rate": 4.8484632873449676e-05,
+      "loss": 2.3152,
+      "step": 6510
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 18.38760757446289,
+      "learning_rate": 4.847917702784295e-05,
+      "loss": 2.3602,
+      "step": 6520
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 7.56703519821167,
+      "learning_rate": 4.8473711686498655e-05,
+      "loss": 2.3064,
+      "step": 6530
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.525967121124268,
+      "learning_rate": 4.8468236851627176e-05,
+      "loss": 2.2774,
+      "step": 6540
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.542328357696533,
+      "learning_rate": 4.846275252544269e-05,
+      "loss": 2.2453,
+      "step": 6550
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 9.50683879852295,
+      "learning_rate": 4.845725871016324e-05,
+      "loss": 2.43,
+      "step": 6560
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 7.898688793182373,
+      "learning_rate": 4.84517554080107e-05,
+      "loss": 2.1673,
+      "step": 6570
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.77755069732666,
+      "learning_rate": 4.8446242621210804e-05,
+      "loss": 2.2338,
+      "step": 6580
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.6925225257873535,
+      "learning_rate": 4.844072035199307e-05,
+      "loss": 2.2819,
+      "step": 6590
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.21223258972168,
+      "learning_rate": 4.843518860259091e-05,
+      "loss": 2.3132,
+      "step": 6600
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.520556926727295,
+      "learning_rate": 4.842964737524153e-05,
+      "loss": 2.3449,
+      "step": 6610
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 21.28665542602539,
+      "learning_rate": 4.842409667218598e-05,
+      "loss": 2.315,
+      "step": 6620
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.1759033203125,
+      "learning_rate": 4.841853649566914e-05,
+      "loss": 2.3547,
+      "step": 6630
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 19.148984909057617,
+      "learning_rate": 4.841296684793973e-05,
+      "loss": 2.5135,
+      "step": 6640
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 9.910473823547363,
+      "learning_rate": 4.84073877312503e-05,
+      "loss": 2.204,
+      "step": 6650
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 27.07410430908203,
+      "learning_rate": 4.8401799147857216e-05,
+      "loss": 2.4857,
+      "step": 6660
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 62.342281341552734,
+      "learning_rate": 4.839620110002069e-05,
+      "loss": 2.3909,
+      "step": 6670
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 103.58372497558594,
+      "learning_rate": 4.8391154766739814e-05,
+      "loss": 2.5908,
+      "step": 6680
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 22.150035858154297,
+      "learning_rate": 4.83855387427013e-05,
+      "loss": 2.3127,
+      "step": 6690
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 14.04951000213623,
+      "learning_rate": 4.8379913260795586e-05,
+      "loss": 2.4046,
+      "step": 6700
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 22.72818946838379,
+      "learning_rate": 4.837427832329779e-05,
+      "loss": 2.8313,
+      "step": 6710
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 13.296052932739258,
+      "learning_rate": 4.836863393248684e-05,
+      "loss": 2.4321,
+      "step": 6720
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.5498833656311035,
+      "learning_rate": 4.836298009064554e-05,
+      "loss": 2.3137,
+      "step": 6730
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 13.326327323913574,
+      "learning_rate": 4.835731680006047e-05,
+      "loss": 2.291,
+      "step": 6740
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 32.48208999633789,
+      "learning_rate": 4.835164406302205e-05,
+      "loss": 2.6136,
+      "step": 6750
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 11.908808708190918,
+      "learning_rate": 4.834596188182453e-05,
+      "loss": 2.1887,
+      "step": 6760
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 10.075331687927246,
+      "learning_rate": 4.834027025876595e-05,
+      "loss": 2.3643,
+      "step": 6770
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.72551155090332,
+      "learning_rate": 4.83345691961482e-05,
+      "loss": 2.2839,
+      "step": 6780
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.926360130310059,
+      "learning_rate": 4.8328858696276955e-05,
+      "loss": 2.3022,
+      "step": 6790
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 9.422292709350586,
+      "learning_rate": 4.832313876146175e-05,
+      "loss": 2.4454,
+      "step": 6800
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 13.58336353302002,
+      "learning_rate": 4.83174093940159e-05,
+      "loss": 2.176,
+      "step": 6810
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 8.902631759643555,
+      "learning_rate": 4.831167059625654e-05,
+      "loss": 2.3952,
+      "step": 6820
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 16.641265869140625,
+      "learning_rate": 4.830592237050464e-05,
+      "loss": 2.2495,
+      "step": 6830
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 11.762859344482422,
+      "learning_rate": 4.830016471908496e-05,
+      "loss": 2.4345,
+      "step": 6840
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.909867763519287,
+      "learning_rate": 4.829439764432607e-05,
+      "loss": 2.207,
+      "step": 6850
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 13.684906959533691,
+      "learning_rate": 4.828862114856038e-05,
+      "loss": 2.3951,
+      "step": 6860
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.779490947723389,
+      "learning_rate": 4.8282835234124074e-05,
+      "loss": 2.3883,
+      "step": 6870
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 7.6955366134643555,
+      "learning_rate": 4.827703990335718e-05,
+      "loss": 2.3341,
+      "step": 6880
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 7.92070198059082,
+      "learning_rate": 4.82712351586035e-05,
+      "loss": 2.4001,
+      "step": 6890
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 12.462532997131348,
+      "learning_rate": 4.826542100221067e-05,
+      "loss": 2.2261,
+      "step": 6900
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 18.708791732788086,
+      "learning_rate": 4.8259597436530125e-05,
+      "loss": 2.1191,
+      "step": 6910
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.56512975692749,
+      "learning_rate": 4.8253764463917096e-05,
+      "loss": 2.2219,
+      "step": 6920
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.9286699295043945,
+      "learning_rate": 4.8247922086730634e-05,
+      "loss": 2.3153,
+      "step": 6930
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 7.228487491607666,
+      "learning_rate": 4.8242070307333574e-05,
+      "loss": 2.4929,
+      "step": 6940
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 11.886446952819824,
+      "learning_rate": 4.8236209128092566e-05,
+      "loss": 2.3487,
+      "step": 6950
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.152067184448242,
+      "learning_rate": 4.823033855137807e-05,
+      "loss": 2.2251,
+      "step": 6960
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 14.04540729522705,
+      "learning_rate": 4.8224458579564334e-05,
+      "loss": 2.3503,
+      "step": 6970
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 9.333481788635254,
+      "learning_rate": 4.8218569215029405e-05,
+      "loss": 2.3221,
+      "step": 6980
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 35.72920227050781,
+      "learning_rate": 4.821267046015513e-05,
+      "loss": 2.3805,
+      "step": 6990
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 18.05036163330078,
+      "learning_rate": 4.820676231732716e-05,
+      "loss": 2.2986,
+      "step": 7000
+    },
+    {
+      "epoch": 0.28,
+      "eval_loss": 1.956457495689392,
+      "eval_runtime": 423.6428,
+      "eval_samples_per_second": 50.092,
+      "eval_steps_per_second": 0.392,
+      "step": 7000
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 18.98715591430664,
+      "learning_rate": 4.820084478893494e-05,
+      "loss": 2.3742,
+      "step": 7010
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 7.488858699798584,
+      "learning_rate": 4.819491787737171e-05,
+      "loss": 2.3052,
+      "step": 7020
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 7.39491605758667,
+      "learning_rate": 4.81889815850345e-05,
+      "loss": 2.54,
+      "step": 7030
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 7.2730865478515625,
+      "learning_rate": 4.8183035914324136e-05,
+      "loss": 2.2487,
+      "step": 7040
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 7.395474433898926,
+      "learning_rate": 4.8177080867645245e-05,
+      "loss": 2.3872,
+      "step": 7050
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 13.659703254699707,
+      "learning_rate": 4.817111644740624e-05,
+      "loss": 2.3713,
+      "step": 7060
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 21.210573196411133,
+      "learning_rate": 4.8165142656019325e-05,
+      "loss": 2.2776,
+      "step": 7070
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 35.836185455322266,
+      "learning_rate": 4.81591594959005e-05,
+      "loss": 2.4872,
+      "step": 7080
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 13.00560188293457,
+      "learning_rate": 4.815316696946953e-05,
+      "loss": 2.2952,
+      "step": 7090
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 7.423733234405518,
+      "learning_rate": 4.814716507915e-05,
+      "loss": 2.3123,
+      "step": 7100
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 29.948333740234375,
+      "learning_rate": 4.8141153827369264e-05,
+      "loss": 2.0931,
+      "step": 7110
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 7.296632289886475,
+      "learning_rate": 4.813513321655848e-05,
+      "loss": 2.2626,
+      "step": 7120
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.078789710998535,
+      "learning_rate": 4.8129103249152554e-05,
+      "loss": 2.193,
+      "step": 7130
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.776204586029053,
+      "learning_rate": 4.812306392759022e-05,
+      "loss": 2.2646,
+      "step": 7140
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.253006458282471,
+      "learning_rate": 4.811701525431396e-05,
+      "loss": 2.2343,
+      "step": 7150
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 41.79158401489258,
+      "learning_rate": 4.811095723177006e-05,
+      "loss": 2.232,
+      "step": 7160
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 8.95724105834961,
+      "learning_rate": 4.810488986240858e-05,
+      "loss": 2.2008,
+      "step": 7170
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 9.038932800292969,
+      "learning_rate": 4.8098813148683356e-05,
+      "loss": 2.1931,
+      "step": 7180
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.18593692779541,
+      "learning_rate": 4.809272709305201e-05,
+      "loss": 2.2143,
+      "step": 7190
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 7.01367712020874,
+      "learning_rate": 4.808663169797594e-05,
+      "loss": 2.1562,
+      "step": 7200
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.437080383300781,
+      "learning_rate": 4.808052696592032e-05,
+      "loss": 2.1143,
+      "step": 7210
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 7.393974304199219,
+      "learning_rate": 4.807441289935411e-05,
+      "loss": 2.4122,
+      "step": 7220
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.011038780212402,
+      "learning_rate": 4.806828950075002e-05,
+      "loss": 2.1467,
+      "step": 7230
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 9.512205123901367,
+      "learning_rate": 4.806215677258456e-05,
+      "loss": 2.2061,
+      "step": 7240
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.728775978088379,
+      "learning_rate": 4.805601471733801e-05,
+      "loss": 2.234,
+      "step": 7250
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 10.132803916931152,
+      "learning_rate": 4.8049863337494383e-05,
+      "loss": 2.249,
+      "step": 7260
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 11.986092567443848,
+      "learning_rate": 4.8043702635541534e-05,
+      "loss": 2.2069,
+      "step": 7270
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.403532981872559,
+      "learning_rate": 4.8037532613971035e-05,
+      "loss": 2.1335,
+      "step": 7280
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 28.95196533203125,
+      "learning_rate": 4.803135327527825e-05,
+      "loss": 2.3068,
+      "step": 7290
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 7.692062854766846,
+      "learning_rate": 4.8025164621962284e-05,
+      "loss": 2.4062,
+      "step": 7300
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.148292541503906,
+      "learning_rate": 4.801896665652603e-05,
+      "loss": 2.2943,
+      "step": 7310
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 7.389751434326172,
+      "learning_rate": 4.801275938147617e-05,
+      "loss": 2.2432,
+      "step": 7320
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 9.887344360351562,
+      "learning_rate": 4.800654279932311e-05,
+      "loss": 2.3278,
+      "step": 7330
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.3506927490234375,
+      "learning_rate": 4.800031691258103e-05,
+      "loss": 2.1808,
+      "step": 7340
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 7.447039604187012,
+      "learning_rate": 4.79940817237679e-05,
+      "loss": 2.3918,
+      "step": 7350
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 7.131879806518555,
+      "learning_rate": 4.798783723540541e-05,
+      "loss": 2.3571,
+      "step": 7360
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 9.371427536010742,
+      "learning_rate": 4.7981583450019045e-05,
+      "loss": 2.3176,
+      "step": 7370
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.0981903076171875,
+      "learning_rate": 4.797532037013803e-05,
+      "loss": 2.2672,
+      "step": 7380
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.475356578826904,
+      "learning_rate": 4.796904799829537e-05,
+      "loss": 2.1954,
+      "step": 7390
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.3766937255859375,
+      "learning_rate": 4.796276633702781e-05,
+      "loss": 2.0606,
+      "step": 7400
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.111029624938965,
+      "learning_rate": 4.795647538887584e-05,
+      "loss": 2.148,
+      "step": 7410
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.0170817375183105,
+      "learning_rate": 4.795017515638375e-05,
+      "loss": 1.9902,
+      "step": 7420
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.29054594039917,
+      "learning_rate": 4.794386564209953e-05,
+      "loss": 2.0952,
+      "step": 7430
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.003581523895264,
+      "learning_rate": 4.793754684857497e-05,
+      "loss": 1.987,
+      "step": 7440
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.239002227783203,
+      "learning_rate": 4.7931218778365594e-05,
+      "loss": 2.1665,
+      "step": 7450
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.7367377281188965,
+      "learning_rate": 4.792488143403067e-05,
+      "loss": 2.1673,
+      "step": 7460
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.604746341705322,
+      "learning_rate": 4.791853481813322e-05,
+      "loss": 2.1562,
+      "step": 7470
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.525786876678467,
+      "learning_rate": 4.7912178933240035e-05,
+      "loss": 2.1059,
+      "step": 7480
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.215798377990723,
+      "learning_rate": 4.790581378192164e-05,
+      "loss": 2.1964,
+      "step": 7490
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 21.55533790588379,
+      "learning_rate": 4.7899439366752294e-05,
+      "loss": 2.2484,
+      "step": 7500
+    },
+    {
+      "epoch": 0.3,
+      "eval_loss": 1.8438102006912231,
+      "eval_runtime": 423.8305,
+      "eval_samples_per_second": 50.07,
+      "eval_steps_per_second": 0.392,
+      "step": 7500
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 50400,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 500,
+  "total_flos": 0.0,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}