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": 1.8124697921701305,
+  "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": NaN,
+      "learning_rate": 2.0000000000000002e-07,
+      "loss": 2.0414,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 53.515953063964844,
+      "learning_rate": 6.5e-07,
+      "loss": 1.8355,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 27.920907974243164,
+      "learning_rate": 1.15e-06,
+      "loss": 1.5172,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 20.47172737121582,
+      "learning_rate": 1.65e-06,
+      "loss": 1.2654,
+      "step": 40
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 8.903529167175293,
+      "learning_rate": 2.1499999999999997e-06,
+      "loss": 1.1095,
+      "step": 50
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.902841567993164,
+      "learning_rate": 2.65e-06,
+      "loss": 1.0017,
+      "step": 60
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.9340908527374268,
+      "learning_rate": 3.1500000000000003e-06,
+      "loss": 0.8846,
+      "step": 70
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 8.701367378234863,
+      "learning_rate": 3.6499999999999998e-06,
+      "loss": 0.8572,
+      "step": 80
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.3205362558364868,
+      "learning_rate": 4.15e-06,
+      "loss": 0.839,
+      "step": 90
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.0334558486938477,
+      "learning_rate": 4.65e-06,
+      "loss": 0.7838,
+      "step": 100
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.8086520433425903,
+      "learning_rate": 5.15e-06,
+      "loss": 0.7782,
+      "step": 110
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 29.50135040283203,
+      "learning_rate": 5.65e-06,
+      "loss": 0.7883,
+      "step": 120
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.2708683013916016,
+      "learning_rate": 6.15e-06,
+      "loss": 0.7961,
+      "step": 130
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 33.43790817260742,
+      "learning_rate": 6.650000000000001e-06,
+      "loss": 0.7626,
+      "step": 140
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.2334959506988525,
+      "learning_rate": 7.15e-06,
+      "loss": 0.7732,
+      "step": 150
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.7489686012268066,
+      "learning_rate": 7.65e-06,
+      "loss": 0.7715,
+      "step": 160
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.358823776245117,
+      "learning_rate": 8.15e-06,
+      "loss": 0.7778,
+      "step": 170
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.348245620727539,
+      "learning_rate": 8.65e-06,
+      "loss": 0.7619,
+      "step": 180
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.9023125767707825,
+      "learning_rate": 9.15e-06,
+      "loss": 0.7746,
+      "step": 190
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.9158114790916443,
+      "learning_rate": 9.65e-06,
+      "loss": 0.7592,
+      "step": 200
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.6256226301193237,
+      "learning_rate": 1.0150000000000001e-05,
+      "loss": 0.7787,
+      "step": 210
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.3805778920650482,
+      "learning_rate": 1.065e-05,
+      "loss": 0.7811,
+      "step": 220
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.6968041062355042,
+      "learning_rate": 1.115e-05,
+      "loss": 0.73,
+      "step": 230
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.9035410284996033,
+      "learning_rate": 1.1650000000000002e-05,
+      "loss": 0.7518,
+      "step": 240
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.6727488040924072,
+      "learning_rate": 1.215e-05,
+      "loss": 0.7805,
+      "step": 250
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.4153461158275604,
+      "learning_rate": 1.2650000000000001e-05,
+      "loss": 0.7706,
+      "step": 260
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.4846651554107666,
+      "learning_rate": 1.3150000000000001e-05,
+      "loss": 0.7492,
+      "step": 270
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.6631388664245605,
+      "learning_rate": 1.3650000000000001e-05,
+      "loss": 0.7681,
+      "step": 280
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.6325013041496277,
+      "learning_rate": 1.415e-05,
+      "loss": 0.7753,
+      "step": 290
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.877907395362854,
+      "learning_rate": 1.465e-05,
+      "loss": 0.7188,
+      "step": 300
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.37142279744148254,
+      "learning_rate": 1.515e-05,
+      "loss": 0.7204,
+      "step": 310
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.7337246537208557,
+      "learning_rate": 1.565e-05,
+      "loss": 0.778,
+      "step": 320
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.5847220420837402,
+      "learning_rate": 1.6150000000000003e-05,
+      "loss": 0.7288,
+      "step": 330
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.745180606842041,
+      "learning_rate": 1.665e-05,
+      "loss": 0.7531,
+      "step": 340
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.357301265001297,
+      "learning_rate": 1.7150000000000004e-05,
+      "loss": 0.7448,
+      "step": 350
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.9032486081123352,
+      "learning_rate": 1.765e-05,
+      "loss": 0.7651,
+      "step": 360
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.2864232361316681,
+      "learning_rate": 1.815e-05,
+      "loss": 0.7193,
+      "step": 370
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.8560261726379395,
+      "learning_rate": 1.865e-05,
+      "loss": 0.7421,
+      "step": 380
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.2971792221069336,
+      "learning_rate": 1.915e-05,
+      "loss": 0.7171,
+      "step": 390
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.40850627422332764,
+      "learning_rate": 1.9650000000000003e-05,
+      "loss": 0.7459,
+      "step": 400
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.3934139311313629,
+      "learning_rate": 2.0150000000000002e-05,
+      "loss": 0.7205,
+      "step": 410
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.8674131631851196,
+      "learning_rate": 2.065e-05,
+      "loss": 0.752,
+      "step": 420
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.36707818508148193,
+      "learning_rate": 2.115e-05,
+      "loss": 0.7494,
+      "step": 430
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.5044310092926025,
+      "learning_rate": 2.165e-05,
+      "loss": 0.7595,
+      "step": 440
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.9025150537490845,
+      "learning_rate": 2.215e-05,
+      "loss": 0.7379,
+      "step": 450
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.5004140138626099,
+      "learning_rate": 2.265e-05,
+      "loss": 0.7886,
+      "step": 460
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.544482707977295,
+      "learning_rate": 2.3150000000000004e-05,
+      "loss": 0.7259,
+      "step": 470
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.33023321628570557,
+      "learning_rate": 2.365e-05,
+      "loss": 0.7333,
+      "step": 480
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.3548080325126648,
+      "learning_rate": 2.415e-05,
+      "loss": 0.7527,
+      "step": 490
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.42450150847435,
+      "learning_rate": 2.465e-05,
+      "loss": 0.7443,
+      "step": 500
+    },
+    {
+      "epoch": 0.12,
+      "eval_loss": 0.7428915500640869,
+      "eval_runtime": 68.3143,
+      "eval_samples_per_second": 29.276,
+      "eval_steps_per_second": 0.922,
+      "step": 500
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.3299656808376312,
+      "learning_rate": 2.515e-05,
+      "loss": 0.7447,
+      "step": 510
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.27258285880088806,
+      "learning_rate": 2.5650000000000003e-05,
+      "loss": 0.7278,
+      "step": 520
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.23753009736537933,
+      "learning_rate": 2.6150000000000002e-05,
+      "loss": 0.7721,
+      "step": 530
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.1074901819229126,
+      "learning_rate": 2.6650000000000004e-05,
+      "loss": 0.7383,
+      "step": 540
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.428067147731781,
+      "learning_rate": 2.7150000000000003e-05,
+      "loss": 0.7508,
+      "step": 550
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.35956478118896484,
+      "learning_rate": 2.7650000000000005e-05,
+      "loss": 0.7461,
+      "step": 560
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.23017314076423645,
+      "learning_rate": 2.815e-05,
+      "loss": 0.7463,
+      "step": 570
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.20370900630950928,
+      "learning_rate": 2.865e-05,
+      "loss": 0.7405,
+      "step": 580
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.24586541950702667,
+      "learning_rate": 2.915e-05,
+      "loss": 0.6964,
+      "step": 590
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.27857884764671326,
+      "learning_rate": 2.965e-05,
+      "loss": 0.7376,
+      "step": 600
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.2655833065509796,
+      "learning_rate": 3.015e-05,
+      "loss": 0.7496,
+      "step": 610
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.17174670100212097,
+      "learning_rate": 3.065e-05,
+      "loss": 0.7219,
+      "step": 620
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.22892743349075317,
+      "learning_rate": 3.115e-05,
+      "loss": 0.7111,
+      "step": 630
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.5913789868354797,
+      "learning_rate": 3.1650000000000004e-05,
+      "loss": 0.7097,
+      "step": 640
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.25620004534721375,
+      "learning_rate": 3.215e-05,
+      "loss": 0.7306,
+      "step": 650
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.3311476409435272,
+      "learning_rate": 3.265e-05,
+      "loss": 0.7406,
+      "step": 660
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.4708893597126007,
+      "learning_rate": 3.3150000000000006e-05,
+      "loss": 0.7071,
+      "step": 670
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.22469288110733032,
+      "learning_rate": 3.3650000000000005e-05,
+      "loss": 0.7116,
+      "step": 680
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.2909330725669861,
+      "learning_rate": 3.415e-05,
+      "loss": 0.7592,
+      "step": 690
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.21274766325950623,
+      "learning_rate": 3.465e-05,
+      "loss": 0.7144,
+      "step": 700
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.23929230868816376,
+      "learning_rate": 3.515e-05,
+      "loss": 0.7382,
+      "step": 710
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.3021218180656433,
+      "learning_rate": 3.565e-05,
+      "loss": 0.7287,
+      "step": 720
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.23557321727275848,
+      "learning_rate": 3.615e-05,
+      "loss": 0.7475,
+      "step": 730
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.22648084163665771,
+      "learning_rate": 3.665e-05,
+      "loss": 0.7457,
+      "step": 740
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.2707761824131012,
+      "learning_rate": 3.715e-05,
+      "loss": 0.7322,
+      "step": 750
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.2799661457538605,
+      "learning_rate": 3.765e-05,
+      "loss": 0.7682,
+      "step": 760
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.326861172914505,
+      "learning_rate": 3.8150000000000006e-05,
+      "loss": 0.73,
+      "step": 770
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.31985723972320557,
+      "learning_rate": 3.8650000000000004e-05,
+      "loss": 0.744,
+      "step": 780
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.21637533605098724,
+      "learning_rate": 3.915e-05,
+      "loss": 0.7161,
+      "step": 790
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.26102888584136963,
+      "learning_rate": 3.965e-05,
+      "loss": 0.7553,
+      "step": 800
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.2469174712896347,
+      "learning_rate": 4.015000000000001e-05,
+      "loss": 0.7444,
+      "step": 810
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.2022310495376587,
+      "learning_rate": 4.065e-05,
+      "loss": 0.7089,
+      "step": 820
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.1729898750782013,
+      "learning_rate": 4.115e-05,
+      "loss": 0.7122,
+      "step": 830
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.22450707852840424,
+      "learning_rate": 4.165e-05,
+      "loss": 0.7099,
+      "step": 840
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.22389698028564453,
+      "learning_rate": 4.215e-05,
+      "loss": 0.7445,
+      "step": 850
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.24962696433067322,
+      "learning_rate": 4.265e-05,
+      "loss": 0.7052,
+      "step": 860
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.23780404031276703,
+      "learning_rate": 4.315e-05,
+      "loss": 0.7423,
+      "step": 870
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.2434270679950714,
+      "learning_rate": 4.3650000000000004e-05,
+      "loss": 0.7291,
+      "step": 880
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.3332350552082062,
+      "learning_rate": 4.415e-05,
+      "loss": 0.7306,
+      "step": 890
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.5084495544433594,
+      "learning_rate": 4.465e-05,
+      "loss": 0.735,
+      "step": 900
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.27552103996276855,
+      "learning_rate": 4.5150000000000006e-05,
+      "loss": 0.7082,
+      "step": 910
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.2184416800737381,
+      "learning_rate": 4.5650000000000005e-05,
+      "loss": 0.7066,
+      "step": 920
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.21738290786743164,
+      "learning_rate": 4.6150000000000004e-05,
+      "loss": 0.6977,
+      "step": 930
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.21888123452663422,
+      "learning_rate": 4.665e-05,
+      "loss": 0.7083,
+      "step": 940
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.2516074776649475,
+      "learning_rate": 4.715e-05,
+      "loss": 0.7451,
+      "step": 950
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.39361557364463806,
+      "learning_rate": 4.765e-05,
+      "loss": 0.7013,
+      "step": 960
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.2461744099855423,
+      "learning_rate": 4.815e-05,
+      "loss": 0.7325,
+      "step": 970
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.2554089426994324,
+      "learning_rate": 4.8650000000000003e-05,
+      "loss": 0.7384,
+      "step": 980
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.2717374861240387,
+      "learning_rate": 4.915e-05,
+      "loss": 0.7172,
+      "step": 990
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.35919347405433655,
+      "learning_rate": 4.965e-05,
+      "loss": 0.6851,
+      "step": 1000
+    },
+    {
+      "epoch": 0.24,
+      "eval_loss": 0.7169972658157349,
+      "eval_runtime": 67.4497,
+      "eval_samples_per_second": 29.652,
+      "eval_steps_per_second": 0.934,
+      "step": 1000
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.19895273447036743,
+      "learning_rate": 4.999999713608037e-05,
+      "loss": 0.7104,
+      "step": 1010
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.21741873025894165,
+      "learning_rate": 4.999994622197174e-05,
+      "loss": 0.7217,
+      "step": 1020
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.3415576219558716,
+      "learning_rate": 4.999983166535371e-05,
+      "loss": 0.716,
+      "step": 1030
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.22503221035003662,
+      "learning_rate": 4.99996534665179e-05,
+      "loss": 0.7166,
+      "step": 1040
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.22469323873519897,
+      "learning_rate": 4.999941162591795e-05,
+      "loss": 0.7173,
+      "step": 1050
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.24848666787147522,
+      "learning_rate": 4.999910614416952e-05,
+      "loss": 0.7442,
+      "step": 1060
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.23361551761627197,
+      "learning_rate": 4.999873702205027e-05,
+      "loss": 0.7142,
+      "step": 1070
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.2513304054737091,
+      "learning_rate": 4.999830426049987e-05,
+      "loss": 0.6998,
+      "step": 1080
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.2846020758152008,
+      "learning_rate": 4.999780786062003e-05,
+      "loss": 0.6967,
+      "step": 1090
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.2192545235157013,
+      "learning_rate": 4.999724782367441e-05,
+      "loss": 0.7259,
+      "step": 1100
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.37123605608940125,
+      "learning_rate": 4.999662415108872e-05,
+      "loss": 0.7228,
+      "step": 1110
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.3963713049888611,
+      "learning_rate": 4.999593684445063e-05,
+      "loss": 0.6848,
+      "step": 1120
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.24653227627277374,
+      "learning_rate": 4.9995185905509836e-05,
+      "loss": 0.7181,
+      "step": 1130
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.25218069553375244,
+      "learning_rate": 4.999437133617799e-05,
+      "loss": 0.7263,
+      "step": 1140
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.20412948727607727,
+      "learning_rate": 4.9993493138528765e-05,
+      "loss": 0.6965,
+      "step": 1150
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.21635311841964722,
+      "learning_rate": 4.9992551314797775e-05,
+      "loss": 0.7042,
+      "step": 1160
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.21901027858257294,
+      "learning_rate": 4.999154586738264e-05,
+      "loss": 0.6967,
+      "step": 1170
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.24104245007038116,
+      "learning_rate": 4.9990476798842935e-05,
+      "loss": 0.7009,
+      "step": 1180
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.2433982491493225,
+      "learning_rate": 4.998934411190018e-05,
+      "loss": 0.7052,
+      "step": 1190
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.5261262655258179,
+      "learning_rate": 4.9988147809437876e-05,
+      "loss": 0.7103,
+      "step": 1200
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.36266064643859863,
+      "learning_rate": 4.998688789450146e-05,
+      "loss": 0.7028,
+      "step": 1210
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.2509474754333496,
+      "learning_rate": 4.9985564370298274e-05,
+      "loss": 0.7047,
+      "step": 1220
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.1925598382949829,
+      "learning_rate": 4.9984177240197665e-05,
+      "loss": 0.6943,
+      "step": 1230
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.22494031488895416,
+      "learning_rate": 4.998272650773083e-05,
+      "loss": 0.6953,
+      "step": 1240
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.24152888357639313,
+      "learning_rate": 4.998121217659092e-05,
+      "loss": 0.7181,
+      "step": 1250
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.24296504259109497,
+      "learning_rate": 4.997963425063297e-05,
+      "loss": 0.6934,
+      "step": 1260
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.2438274323940277,
+      "learning_rate": 4.9977992733873906e-05,
+      "loss": 0.7088,
+      "step": 1270
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.21081580221652985,
+      "learning_rate": 4.997628763049257e-05,
+      "loss": 0.6833,
+      "step": 1280
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.2111068218946457,
+      "learning_rate": 4.9974518944829626e-05,
+      "loss": 0.6922,
+      "step": 1290
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.309772253036499,
+      "learning_rate": 4.997268668138766e-05,
+      "loss": 0.7195,
+      "step": 1300
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.2199670672416687,
+      "learning_rate": 4.997079084483105e-05,
+      "loss": 0.6924,
+      "step": 1310
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.21781674027442932,
+      "learning_rate": 4.996883143998605e-05,
+      "loss": 0.6873,
+      "step": 1320
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.2969052791595459,
+      "learning_rate": 4.996680847184072e-05,
+      "loss": 0.7135,
+      "step": 1330
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.23092682659626007,
+      "learning_rate": 4.996472194554495e-05,
+      "loss": 0.6827,
+      "step": 1340
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.29615628719329834,
+      "learning_rate": 4.996257186641042e-05,
+      "loss": 0.6738,
+      "step": 1350
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.23491479456424713,
+      "learning_rate": 4.99603582399106e-05,
+      "loss": 0.7396,
+      "step": 1360
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.1986425668001175,
+      "learning_rate": 4.9958081071680726e-05,
+      "loss": 0.6787,
+      "step": 1370
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.19658410549163818,
+      "learning_rate": 4.99557403675178e-05,
+      "loss": 0.66,
+      "step": 1380
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.29404616355895996,
+      "learning_rate": 4.995333613338057e-05,
+      "loss": 0.6824,
+      "step": 1390
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.23689547181129456,
+      "learning_rate": 4.9950868375389514e-05,
+      "loss": 0.7008,
+      "step": 1400
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.22758638858795166,
+      "learning_rate": 4.99483370998268e-05,
+      "loss": 0.6946,
+      "step": 1410
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.20259805023670197,
+      "learning_rate": 4.994574231313634e-05,
+      "loss": 0.6734,
+      "step": 1420
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.21935436129570007,
+      "learning_rate": 4.994308402192366e-05,
+      "loss": 0.7094,
+      "step": 1430
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.23414015769958496,
+      "learning_rate": 4.9940362232956026e-05,
+      "loss": 0.7019,
+      "step": 1440
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.2623121440410614,
+      "learning_rate": 4.993757695316228e-05,
+      "loss": 0.6793,
+      "step": 1450
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.31137487292289734,
+      "learning_rate": 4.993472818963295e-05,
+      "loss": 0.6902,
+      "step": 1460
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.22092680633068085,
+      "learning_rate": 4.993181594962013e-05,
+      "loss": 0.6912,
+      "step": 1470
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.23723623156547546,
+      "learning_rate": 4.992884024053754e-05,
+      "loss": 0.652,
+      "step": 1480
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.26028233766555786,
+      "learning_rate": 4.9925801069960454e-05,
+      "loss": 0.6969,
+      "step": 1490
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.23856200277805328,
+      "learning_rate": 4.992269844562572e-05,
+      "loss": 0.6723,
+      "step": 1500
+    },
+    {
+      "epoch": 0.36,
+      "eval_loss": 0.6912096738815308,
+      "eval_runtime": 67.244,
+      "eval_samples_per_second": 29.742,
+      "eval_steps_per_second": 0.937,
+      "step": 1500
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.167326420545578,
+      "learning_rate": 4.9919532375431677e-05,
+      "loss": 0.6897,
+      "step": 1510
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.26183101534843445,
+      "learning_rate": 4.991630286743823e-05,
+      "loss": 0.6844,
+      "step": 1520
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.25918036699295044,
+      "learning_rate": 4.991300992986676e-05,
+      "loss": 0.6645,
+      "step": 1530
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.2354690134525299,
+      "learning_rate": 4.99096535711001e-05,
+      "loss": 0.6672,
+      "step": 1540
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.24444998800754547,
+      "learning_rate": 4.990623379968257e-05,
+      "loss": 0.6749,
+      "step": 1550
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.2562585771083832,
+      "learning_rate": 4.990275062431989e-05,
+      "loss": 0.6652,
+      "step": 1560
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.22618699073791504,
+      "learning_rate": 4.98992040538792e-05,
+      "loss": 0.7163,
+      "step": 1570
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.20957091450691223,
+      "learning_rate": 4.9895594097389044e-05,
+      "loss": 0.7053,
+      "step": 1580
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.26973310112953186,
+      "learning_rate": 4.989192076403928e-05,
+      "loss": 0.6716,
+      "step": 1590
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.24841250479221344,
+      "learning_rate": 4.9888184063181154e-05,
+      "loss": 0.6884,
+      "step": 1600
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.25283706188201904,
+      "learning_rate": 4.98843840043272e-05,
+      "loss": 0.6632,
+      "step": 1610
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.196313738822937,
+      "learning_rate": 4.988052059715126e-05,
+      "loss": 0.712,
+      "step": 1620
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.21621187031269073,
+      "learning_rate": 4.987659385148842e-05,
+      "loss": 0.6732,
+      "step": 1630
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.2180820256471634,
+      "learning_rate": 4.987260377733502e-05,
+      "loss": 0.7045,
+      "step": 1640
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.20110243558883667,
+      "learning_rate": 4.986855038484862e-05,
+      "loss": 0.6604,
+      "step": 1650
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.20813095569610596,
+      "learning_rate": 4.9864433684347964e-05,
+      "loss": 0.7051,
+      "step": 1660
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.22026021778583527,
+      "learning_rate": 4.9860253686312964e-05,
+      "loss": 0.6752,
+      "step": 1670
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.23409809172153473,
+      "learning_rate": 4.9856010401384654e-05,
+      "loss": 0.6805,
+      "step": 1680
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.1982721984386444,
+      "learning_rate": 4.985170384036521e-05,
+      "loss": 0.6888,
+      "step": 1690
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.24984286725521088,
+      "learning_rate": 4.984733401421785e-05,
+      "loss": 0.6741,
+      "step": 1700
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.24465042352676392,
+      "learning_rate": 4.9842900934066874e-05,
+      "loss": 0.6781,
+      "step": 1710
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.26656293869018555,
+      "learning_rate": 4.98384046111976e-05,
+      "loss": 0.7086,
+      "step": 1720
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.23073184490203857,
+      "learning_rate": 4.9833845057056336e-05,
+      "loss": 0.6966,
+      "step": 1730
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.23296333849430084,
+      "learning_rate": 4.982922228325037e-05,
+      "loss": 0.6769,
+      "step": 1740
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.22282017767429352,
+      "learning_rate": 4.982453630154794e-05,
+      "loss": 0.6581,
+      "step": 1750
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.22943562269210815,
+      "learning_rate": 4.981978712387815e-05,
+      "loss": 0.6803,
+      "step": 1760
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.29295608401298523,
+      "learning_rate": 4.9814974762331034e-05,
+      "loss": 0.6514,
+      "step": 1770
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.21845972537994385,
+      "learning_rate": 4.981009922915743e-05,
+      "loss": 0.691,
+      "step": 1780
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.24616730213165283,
+      "learning_rate": 4.980516053676903e-05,
+      "loss": 0.6948,
+      "step": 1790
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.2354935109615326,
+      "learning_rate": 4.9800158697738264e-05,
+      "loss": 0.6778,
+      "step": 1800
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.3475668430328369,
+      "learning_rate": 4.979509372479837e-05,
+      "loss": 0.6626,
+      "step": 1810
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.2817700207233429,
+      "learning_rate": 4.9789965630843265e-05,
+      "loss": 0.6724,
+      "step": 1820
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.28306934237480164,
+      "learning_rate": 4.978477442892758e-05,
+      "loss": 0.6673,
+      "step": 1830
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.2593139111995697,
+      "learning_rate": 4.9779520132266575e-05,
+      "loss": 0.6866,
+      "step": 1840
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.2618165612220764,
+      "learning_rate": 4.9774202754236145e-05,
+      "loss": 0.6671,
+      "step": 1850
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.22875259816646576,
+      "learning_rate": 4.9768822308372784e-05,
+      "loss": 0.6743,
+      "step": 1860
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.20143887400627136,
+      "learning_rate": 4.976337880837351e-05,
+      "loss": 0.6866,
+      "step": 1870
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.23163306713104248,
+      "learning_rate": 4.975787226809587e-05,
+      "loss": 0.6772,
+      "step": 1880
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.21628862619400024,
+      "learning_rate": 4.975230270155791e-05,
+      "loss": 0.6706,
+      "step": 1890
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.21055322885513306,
+      "learning_rate": 4.9746670122938105e-05,
+      "loss": 0.6536,
+      "step": 1900
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.20045976340770721,
+      "learning_rate": 4.974097454657534e-05,
+      "loss": 0.6477,
+      "step": 1910
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.20996223390102386,
+      "learning_rate": 4.9735215986968874e-05,
+      "loss": 0.6526,
+      "step": 1920
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.1778746396303177,
+      "learning_rate": 4.972939445877831e-05,
+      "loss": 0.6951,
+      "step": 1930
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.2416403591632843,
+      "learning_rate": 4.972350997682354e-05,
+      "loss": 0.6703,
+      "step": 1940
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.23652783036231995,
+      "learning_rate": 4.9717562556084735e-05,
+      "loss": 0.6401,
+      "step": 1950
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.23598232865333557,
+      "learning_rate": 4.9711552211702274e-05,
+      "loss": 0.6668,
+      "step": 1960
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.22741574048995972,
+      "learning_rate": 4.970547895897672e-05,
+      "loss": 0.6765,
+      "step": 1970
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.20418991148471832,
+      "learning_rate": 4.96993428133688e-05,
+      "loss": 0.6771,
+      "step": 1980
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.24983160197734833,
+      "learning_rate": 4.969314379049932e-05,
+      "loss": 0.6656,
+      "step": 1990
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.2511340379714966,
+      "learning_rate": 4.968688190614919e-05,
+      "loss": 0.6605,
+      "step": 2000
+    },
+    {
+      "epoch": 0.48,
+      "eval_loss": 0.6730201244354248,
+      "eval_runtime": 67.2807,
+      "eval_samples_per_second": 29.726,
+      "eval_steps_per_second": 0.936,
+      "step": 2000
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.22646427154541016,
+      "learning_rate": 4.96805571762593e-05,
+      "loss": 0.6729,
+      "step": 2010
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.2343759387731552,
+      "learning_rate": 4.9674169616930574e-05,
+      "loss": 0.6635,
+      "step": 2020
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.23573775589466095,
+      "learning_rate": 4.966771924442385e-05,
+      "loss": 0.654,
+      "step": 2030
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.2696162760257721,
+      "learning_rate": 4.966120607515987e-05,
+      "loss": 0.6728,
+      "step": 2040
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.25032511353492737,
+      "learning_rate": 4.965463012571927e-05,
+      "loss": 0.6743,
+      "step": 2050
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.24621817469596863,
+      "learning_rate": 4.964799141284247e-05,
+      "loss": 0.6596,
+      "step": 2060
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.23296469449996948,
+      "learning_rate": 4.964128995342966e-05,
+      "loss": 0.6471,
+      "step": 2070
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.22769325971603394,
+      "learning_rate": 4.963452576454082e-05,
+      "loss": 0.6499,
+      "step": 2080
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.2572191655635834,
+      "learning_rate": 4.9627698863395564e-05,
+      "loss": 0.6499,
+      "step": 2090
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.21782545745372772,
+      "learning_rate": 4.962080926737319e-05,
+      "loss": 0.6641,
+      "step": 2100
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.22925855219364166,
+      "learning_rate": 4.9613856994012567e-05,
+      "loss": 0.6618,
+      "step": 2110
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.2127605825662613,
+      "learning_rate": 4.960684206101214e-05,
+      "loss": 0.6821,
+      "step": 2120
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.2520267069339752,
+      "learning_rate": 4.9599764486229865e-05,
+      "loss": 0.6748,
+      "step": 2130
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.20282019674777985,
+      "learning_rate": 4.9592624287683176e-05,
+      "loss": 0.6674,
+      "step": 2140
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.46226009726524353,
+      "learning_rate": 4.958542148354891e-05,
+      "loss": 0.6422,
+      "step": 2150
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.2241724133491516,
+      "learning_rate": 4.957815609216329e-05,
+      "loss": 0.6612,
+      "step": 2160
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.23772084712982178,
+      "learning_rate": 4.957082813202186e-05,
+      "loss": 0.6614,
+      "step": 2170
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.2049807757139206,
+      "learning_rate": 4.9563437621779465e-05,
+      "loss": 0.6827,
+      "step": 2180
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.22666208446025848,
+      "learning_rate": 4.955598458025015e-05,
+      "loss": 0.6522,
+      "step": 2190
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.26957106590270996,
+      "learning_rate": 4.954846902640718e-05,
+      "loss": 0.6817,
+      "step": 2200
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.22548554837703705,
+      "learning_rate": 4.954089097938294e-05,
+      "loss": 0.6444,
+      "step": 2210
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.22197945415973663,
+      "learning_rate": 4.9533250458468914e-05,
+      "loss": 0.6779,
+      "step": 2220
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.22711919248104095,
+      "learning_rate": 4.9525547483115617e-05,
+      "loss": 0.6416,
+      "step": 2230
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.22576060891151428,
+      "learning_rate": 4.951778207293255e-05,
+      "loss": 0.6598,
+      "step": 2240
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.22785042226314545,
+      "learning_rate": 4.950995424768818e-05,
+      "loss": 0.6499,
+      "step": 2250
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.20107567310333252,
+      "learning_rate": 4.9502064027309836e-05,
+      "loss": 0.6509,
+      "step": 2260
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.23379752039909363,
+      "learning_rate": 4.94941114318837e-05,
+      "loss": 0.6551,
+      "step": 2270
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.22154928743839264,
+      "learning_rate": 4.948609648165475e-05,
+      "loss": 0.6428,
+      "step": 2280
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.23805582523345947,
+      "learning_rate": 4.947801919702667e-05,
+      "loss": 0.6633,
+      "step": 2290
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.24756357073783875,
+      "learning_rate": 4.946987959856188e-05,
+      "loss": 0.6561,
+      "step": 2300
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.23767632246017456,
+      "learning_rate": 4.9461677706981374e-05,
+      "loss": 0.6552,
+      "step": 2310
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.32177987694740295,
+      "learning_rate": 4.9453413543164775e-05,
+      "loss": 0.6705,
+      "step": 2320
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.29961881041526794,
+      "learning_rate": 4.94450871281502e-05,
+      "loss": 0.6551,
+      "step": 2330
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.2768830955028534,
+      "learning_rate": 4.943669848313427e-05,
+      "loss": 0.6468,
+      "step": 2340
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.24728989601135254,
+      "learning_rate": 4.9428247629472e-05,
+      "loss": 0.6506,
+      "step": 2350
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.2641989290714264,
+      "learning_rate": 4.941973458867677e-05,
+      "loss": 0.6517,
+      "step": 2360
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.18787774443626404,
+      "learning_rate": 4.941115938242028e-05,
+      "loss": 0.6355,
+      "step": 2370
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.26642099022865295,
+      "learning_rate": 4.940252203253248e-05,
+      "loss": 0.6811,
+      "step": 2380
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.24736198782920837,
+      "learning_rate": 4.939382256100154e-05,
+      "loss": 0.6493,
+      "step": 2390
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.3060832619667053,
+      "learning_rate": 4.938506098997374e-05,
+      "loss": 0.6707,
+      "step": 2400
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.24164025485515594,
+      "learning_rate": 4.937623734175346e-05,
+      "loss": 0.6508,
+      "step": 2410
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.26685982942581177,
+      "learning_rate": 4.936735163880313e-05,
+      "loss": 0.6441,
+      "step": 2420
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.17739726603031158,
+      "learning_rate": 4.9358403903743124e-05,
+      "loss": 0.6499,
+      "step": 2430
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.23560577630996704,
+      "learning_rate": 4.9349394159351735e-05,
+      "loss": 0.645,
+      "step": 2440
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.23137669265270233,
+      "learning_rate": 4.9340322428565135e-05,
+      "loss": 0.6615,
+      "step": 2450
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.3485463857650757,
+      "learning_rate": 4.933118873447728e-05,
+      "loss": 0.6283,
+      "step": 2460
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.20602945983409882,
+      "learning_rate": 4.932199310033987e-05,
+      "loss": 0.6831,
+      "step": 2470
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.21541930735111237,
+      "learning_rate": 4.931273554956227e-05,
+      "loss": 0.6722,
+      "step": 2480
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.2092689573764801,
+      "learning_rate": 4.930341610571151e-05,
+      "loss": 0.6609,
+      "step": 2490
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.23792122304439545,
+      "learning_rate": 4.9294034792512126e-05,
+      "loss": 0.6475,
+      "step": 2500
+    },
+    {
+      "epoch": 0.6,
+      "eval_loss": 0.6643247008323669,
+      "eval_runtime": 67.0911,
+      "eval_samples_per_second": 29.81,
+      "eval_steps_per_second": 0.939,
+      "step": 2500
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.2101823091506958,
+      "learning_rate": 4.928459163384619e-05,
+      "loss": 0.6604,
+      "step": 2510
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.1859540194272995,
+      "learning_rate": 4.927508665375321e-05,
+      "loss": 0.6252,
+      "step": 2520
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.2371070235967636,
+      "learning_rate": 4.926551987643007e-05,
+      "loss": 0.6413,
+      "step": 2530
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.2149161994457245,
+      "learning_rate": 4.9255891326230964e-05,
+      "loss": 0.6558,
+      "step": 2540
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.21805475652217865,
+      "learning_rate": 4.9246201027667354e-05,
+      "loss": 0.631,
+      "step": 2550
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.22705689072608948,
+      "learning_rate": 4.9236449005407895e-05,
+      "loss": 0.6785,
+      "step": 2560
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.21902668476104736,
+      "learning_rate": 4.9226635284278355e-05,
+      "loss": 0.657,
+      "step": 2570
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.21482336521148682,
+      "learning_rate": 4.9216759889261586e-05,
+      "loss": 0.6514,
+      "step": 2580
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.20008622109889984,
+      "learning_rate": 4.9206822845497444e-05,
+      "loss": 0.6516,
+      "step": 2590
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.205527663230896,
+      "learning_rate": 4.919682417828271e-05,
+      "loss": 0.6336,
+      "step": 2600
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.263959139585495,
+      "learning_rate": 4.9186763913071065e-05,
+      "loss": 0.6449,
+      "step": 2610
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.21950668096542358,
+      "learning_rate": 4.917664207547297e-05,
+      "loss": 0.6916,
+      "step": 2620
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.215659499168396,
+      "learning_rate": 4.916645869125564e-05,
+      "loss": 0.6441,
+      "step": 2630
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.22789780795574188,
+      "learning_rate": 4.915621378634301e-05,
+      "loss": 0.6576,
+      "step": 2640
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.27431604266166687,
+      "learning_rate": 4.914590738681555e-05,
+      "loss": 0.6487,
+      "step": 2650
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.21077977120876312,
+      "learning_rate": 4.913553951891036e-05,
+      "loss": 0.6688,
+      "step": 2660
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.2798519730567932,
+      "learning_rate": 4.9125110209020954e-05,
+      "loss": 0.6618,
+      "step": 2670
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.24790987372398376,
+      "learning_rate": 4.911461948369731e-05,
+      "loss": 0.6598,
+      "step": 2680
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.2008027732372284,
+      "learning_rate": 4.91040673696457e-05,
+      "loss": 0.6388,
+      "step": 2690
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.23889446258544922,
+      "learning_rate": 4.9093453893728733e-05,
+      "loss": 0.6526,
+      "step": 2700
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.1997964233160019,
+      "learning_rate": 4.908277908296518e-05,
+      "loss": 0.6521,
+      "step": 2710
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.24335375428199768,
+      "learning_rate": 4.907204296452997e-05,
+      "loss": 0.6636,
+      "step": 2720
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.23431982100009918,
+      "learning_rate": 4.906124556575411e-05,
+      "loss": 0.6557,
+      "step": 2730
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.21532365679740906,
+      "learning_rate": 4.90503869141246e-05,
+      "loss": 0.628,
+      "step": 2740
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.22962962090969086,
+      "learning_rate": 4.903946703728436e-05,
+      "loss": 0.6422,
+      "step": 2750
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.22807002067565918,
+      "learning_rate": 4.90284859630322e-05,
+      "loss": 0.6631,
+      "step": 2760
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.27668192982673645,
+      "learning_rate": 4.90174437193227e-05,
+      "loss": 0.6778,
+      "step": 2770
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.2503020763397217,
+      "learning_rate": 4.900634033426616e-05,
+      "loss": 0.6483,
+      "step": 2780
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.3033277094364166,
+      "learning_rate": 4.8995175836128536e-05,
+      "loss": 0.6572,
+      "step": 2790
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.2776722311973572,
+      "learning_rate": 4.898395025333136e-05,
+      "loss": 0.662,
+      "step": 2800
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.31251537799835205,
+      "learning_rate": 4.897266361445165e-05,
+      "loss": 0.6617,
+      "step": 2810
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.38570669293403625,
+      "learning_rate": 4.8961315948221884e-05,
+      "loss": 0.6419,
+      "step": 2820
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.23277431726455688,
+      "learning_rate": 4.894990728352988e-05,
+      "loss": 0.6373,
+      "step": 2830
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.2519891858100891,
+      "learning_rate": 4.893843764941874e-05,
+      "loss": 0.6759,
+      "step": 2840
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.2358052134513855,
+      "learning_rate": 4.892690707508677e-05,
+      "loss": 0.6826,
+      "step": 2850
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.23103272914886475,
+      "learning_rate": 4.8915315589887436e-05,
+      "loss": 0.624,
+      "step": 2860
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.2128864973783493,
+      "learning_rate": 4.8903663223329245e-05,
+      "loss": 0.654,
+      "step": 2870
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.22060807049274445,
+      "learning_rate": 4.889195000507568e-05,
+      "loss": 0.6376,
+      "step": 2880
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.24493062496185303,
+      "learning_rate": 4.888017596494517e-05,
+      "loss": 0.6648,
+      "step": 2890
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.2580796480178833,
+      "learning_rate": 4.886834113291094e-05,
+      "loss": 0.6607,
+      "step": 2900
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.26906147599220276,
+      "learning_rate": 4.8856445539101016e-05,
+      "loss": 0.6441,
+      "step": 2910
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.2145451009273529,
+      "learning_rate": 4.884448921379805e-05,
+      "loss": 0.6543,
+      "step": 2920
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.2156788855791092,
+      "learning_rate": 4.8832472187439345e-05,
+      "loss": 0.6185,
+      "step": 2930
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.25632360577583313,
+      "learning_rate": 4.882039449061673e-05,
+      "loss": 0.639,
+      "step": 2940
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.2496800720691681,
+      "learning_rate": 4.8808256154076436e-05,
+      "loss": 0.6377,
+      "step": 2950
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.39942485094070435,
+      "learning_rate": 4.8796057208719124e-05,
+      "loss": 0.6083,
+      "step": 2960
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.27986374497413635,
+      "learning_rate": 4.8783797685599706e-05,
+      "loss": 0.6528,
+      "step": 2970
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.22164517641067505,
+      "learning_rate": 4.877147761592733e-05,
+      "loss": 0.6456,
+      "step": 2980
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.29858505725860596,
+      "learning_rate": 4.875909703106527e-05,
+      "loss": 0.643,
+      "step": 2990
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.22595979273319244,
+      "learning_rate": 4.874665596253084e-05,
+      "loss": 0.6419,
+      "step": 3000
+    },
+    {
+      "epoch": 0.72,
+      "eval_loss": 0.6583871841430664,
+      "eval_runtime": 67.1961,
+      "eval_samples_per_second": 29.764,
+      "eval_steps_per_second": 0.938,
+      "step": 3000
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.19871610403060913,
+      "learning_rate": 4.8734154441995364e-05,
+      "loss": 0.6556,
+      "step": 3010
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.2072724997997284,
+      "learning_rate": 4.872159250128401e-05,
+      "loss": 0.6764,
+      "step": 3020
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.22515198588371277,
+      "learning_rate": 4.87089701723758e-05,
+      "loss": 0.6703,
+      "step": 3030
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.2292303442955017,
+      "learning_rate": 4.869628748740347e-05,
+      "loss": 0.6288,
+      "step": 3040
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.25133493542671204,
+      "learning_rate": 4.8683544478653395e-05,
+      "loss": 0.6652,
+      "step": 3050
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.29863354563713074,
+      "learning_rate": 4.867074117856555e-05,
+      "loss": 0.6352,
+      "step": 3060
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.2168973684310913,
+      "learning_rate": 4.865787761973334e-05,
+      "loss": 0.6562,
+      "step": 3070
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.2312694936990738,
+      "learning_rate": 4.864495383490363e-05,
+      "loss": 0.6463,
+      "step": 3080
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.2333671748638153,
+      "learning_rate": 4.863196985697655e-05,
+      "loss": 0.6267,
+      "step": 3090
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.23559433221817017,
+      "learning_rate": 4.86189257190055e-05,
+      "loss": 0.623,
+      "step": 3100
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.22314316034317017,
+      "learning_rate": 4.860582145419703e-05,
+      "loss": 0.6298,
+      "step": 3110
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.22701682150363922,
+      "learning_rate": 4.859265709591073e-05,
+      "loss": 0.6359,
+      "step": 3120
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.21683742105960846,
+      "learning_rate": 4.857943267765919e-05,
+      "loss": 0.6666,
+      "step": 3130
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.2519407272338867,
+      "learning_rate": 4.856614823310788e-05,
+      "loss": 0.6731,
+      "step": 3140
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.23536355793476105,
+      "learning_rate": 4.855280379607509e-05,
+      "loss": 0.629,
+      "step": 3150
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.19676193594932556,
+      "learning_rate": 4.853939940053184e-05,
+      "loss": 0.6329,
+      "step": 3160
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.23497731983661652,
+      "learning_rate": 4.852593508060177e-05,
+      "loss": 0.6584,
+      "step": 3170
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.2515724003314972,
+      "learning_rate": 4.8512410870561084e-05,
+      "loss": 0.6177,
+      "step": 3180
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.20130635797977448,
+      "learning_rate": 4.8498826804838436e-05,
+      "loss": 0.6645,
+      "step": 3190
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.23828361928462982,
+      "learning_rate": 4.8485182918014876e-05,
+      "loss": 0.6401,
+      "step": 3200
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.2598424255847931,
+      "learning_rate": 4.847147924482371e-05,
+      "loss": 0.6193,
+      "step": 3210
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.20697768032550812,
+      "learning_rate": 4.845771582015046e-05,
+      "loss": 0.626,
+      "step": 3220
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.19661648571491241,
+      "learning_rate": 4.8443892679032775e-05,
+      "loss": 0.6283,
+      "step": 3230
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.22154510021209717,
+      "learning_rate": 4.843000985666028e-05,
+      "loss": 0.6571,
+      "step": 3240
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.3334987759590149,
+      "learning_rate": 4.841606738837458e-05,
+      "loss": 0.6553,
+      "step": 3250
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.6604078412055969,
+      "learning_rate": 4.8402065309669085e-05,
+      "loss": 0.6239,
+      "step": 3260
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.22238047420978546,
+      "learning_rate": 4.838800365618898e-05,
+      "loss": 0.6725,
+      "step": 3270
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.21566419303417206,
+      "learning_rate": 4.837388246373108e-05,
+      "loss": 0.6597,
+      "step": 3280
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.9337890148162842,
+      "learning_rate": 4.83597017682438e-05,
+      "loss": 0.6454,
+      "step": 3290
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.2716673016548157,
+      "learning_rate": 4.8345461605827014e-05,
+      "loss": 0.6526,
+      "step": 3300
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.23388756811618805,
+      "learning_rate": 4.8331162012732e-05,
+      "loss": 0.6642,
+      "step": 3310
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.21243353188037872,
+      "learning_rate": 4.8316803025361304e-05,
+      "loss": 0.6233,
+      "step": 3320
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.28600814938545227,
+      "learning_rate": 4.8302384680268684e-05,
+      "loss": 0.6559,
+      "step": 3330
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.19718654453754425,
+      "learning_rate": 4.8287907014159004e-05,
+      "loss": 0.6357,
+      "step": 3340
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.2330627292394638,
+      "learning_rate": 4.827337006388816e-05,
+      "loss": 0.6692,
+      "step": 3350
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.18776822090148926,
+      "learning_rate": 4.8258773866462927e-05,
+      "loss": 0.6804,
+      "step": 3360
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.24929189682006836,
+      "learning_rate": 4.8244118459040944e-05,
+      "loss": 0.6455,
+      "step": 3370
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.23884710669517517,
+      "learning_rate": 4.8229403878930566e-05,
+      "loss": 0.6345,
+      "step": 3380
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.2579725682735443,
+      "learning_rate": 4.821463016359078e-05,
+      "loss": 0.6584,
+      "step": 3390
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.2741701900959015,
+      "learning_rate": 4.8199797350631136e-05,
+      "loss": 0.6463,
+      "step": 3400
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.2254386991262436,
+      "learning_rate": 4.818490547781159e-05,
+      "loss": 0.6399,
+      "step": 3410
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.2172977328300476,
+      "learning_rate": 4.816995458304249e-05,
+      "loss": 0.661,
+      "step": 3420
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.23464016616344452,
+      "learning_rate": 4.815494470438441e-05,
+      "loss": 0.6516,
+      "step": 3430
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.27163979411125183,
+      "learning_rate": 4.813987588004807e-05,
+      "loss": 0.6451,
+      "step": 3440
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.20852161943912506,
+      "learning_rate": 4.8124748148394285e-05,
+      "loss": 0.6394,
+      "step": 3450
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.2224467694759369,
+      "learning_rate": 4.81095615479338e-05,
+      "loss": 0.6427,
+      "step": 3460
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.2205616980791092,
+      "learning_rate": 4.8094316117327245e-05,
+      "loss": 0.6491,
+      "step": 3470
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.21984492242336273,
+      "learning_rate": 4.8079011895384985e-05,
+      "loss": 0.6271,
+      "step": 3480
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.29770565032958984,
+      "learning_rate": 4.806364892106707e-05,
+      "loss": 0.6048,
+      "step": 3490
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.2440354973077774,
+      "learning_rate": 4.8048227233483127e-05,
+      "loss": 0.6307,
+      "step": 3500
+    },
+    {
+      "epoch": 0.85,
+      "eval_loss": 0.6532164216041565,
+      "eval_runtime": 67.3224,
+      "eval_samples_per_second": 29.708,
+      "eval_steps_per_second": 0.936,
+      "step": 3500
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.25720447301864624,
+      "learning_rate": 4.803274687189222e-05,
+      "loss": 0.6337,
+      "step": 3510
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.21051880717277527,
+      "learning_rate": 4.8017207875702814e-05,
+      "loss": 0.6398,
+      "step": 3520
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.2977493107318878,
+      "learning_rate": 4.800161028447261e-05,
+      "loss": 0.6346,
+      "step": 3530
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.19749422371387482,
+      "learning_rate": 4.798595413790848e-05,
+      "loss": 0.6325,
+      "step": 3540
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.21522672474384308,
+      "learning_rate": 4.7970239475866386e-05,
+      "loss": 0.6244,
+      "step": 3550
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.24835069477558136,
+      "learning_rate": 4.7954466338351224e-05,
+      "loss": 0.612,
+      "step": 3560
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.25459495186805725,
+      "learning_rate": 4.793863476551677e-05,
+      "loss": 0.6317,
+      "step": 3570
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.28858670592308044,
+      "learning_rate": 4.7922744797665544e-05,
+      "loss": 0.6485,
+      "step": 3580
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.30466488003730774,
+      "learning_rate": 4.790679647524873e-05,
+      "loss": 0.6307,
+      "step": 3590
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.36087316274642944,
+      "learning_rate": 4.789078983886607e-05,
+      "loss": 0.6749,
+      "step": 3600
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.2436164766550064,
+      "learning_rate": 4.787472492926575e-05,
+      "loss": 0.6239,
+      "step": 3610
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.2527044415473938,
+      "learning_rate": 4.78586017873443e-05,
+      "loss": 0.6584,
+      "step": 3620
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.25492456555366516,
+      "learning_rate": 4.784242045414651e-05,
+      "loss": 0.6487,
+      "step": 3630
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.2293190211057663,
+      "learning_rate": 4.782618097086528e-05,
+      "loss": 0.646,
+      "step": 3640
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.7955949902534485,
+      "learning_rate": 4.780988337884157e-05,
+      "loss": 0.6391,
+      "step": 3650
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.2331811785697937,
+      "learning_rate": 4.779352771956425e-05,
+      "loss": 0.6267,
+      "step": 3660
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.23316140472888947,
+      "learning_rate": 4.7777114034670006e-05,
+      "loss": 0.6405,
+      "step": 3670
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.2253628969192505,
+      "learning_rate": 4.776064236594327e-05,
+      "loss": 0.6415,
+      "step": 3680
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.2378271222114563,
+      "learning_rate": 4.774411275531606e-05,
+      "loss": 0.6362,
+      "step": 3690
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.22514688968658447,
+      "learning_rate": 4.7727525244867896e-05,
+      "loss": 0.6375,
+      "step": 3700
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.20843249559402466,
+      "learning_rate": 4.771087987682571e-05,
+      "loss": 0.6166,
+      "step": 3710
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.2776215076446533,
+      "learning_rate": 4.7694176693563705e-05,
+      "loss": 0.6441,
+      "step": 3720
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.21339628100395203,
+      "learning_rate": 4.767741573760327e-05,
+      "loss": 0.6388,
+      "step": 3730
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.24429263174533844,
+      "learning_rate": 4.766059705161288e-05,
+      "loss": 0.6263,
+      "step": 3740
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.2099929004907608,
+      "learning_rate": 4.764372067840795e-05,
+      "loss": 0.6383,
+      "step": 3750
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.22066868841648102,
+      "learning_rate": 4.7626786660950784e-05,
+      "loss": 0.6313,
+      "step": 3760
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.22953028976917267,
+      "learning_rate": 4.760979504235038e-05,
+      "loss": 0.6264,
+      "step": 3770
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.3549204170703888,
+      "learning_rate": 4.759274586586242e-05,
+      "loss": 0.6368,
+      "step": 3780
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.24681401252746582,
+      "learning_rate": 4.757563917488909e-05,
+      "loss": 0.6422,
+      "step": 3790
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.2417946755886078,
+      "learning_rate": 4.755847501297898e-05,
+      "loss": 0.6438,
+      "step": 3800
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.2778252959251404,
+      "learning_rate": 4.7541253423827006e-05,
+      "loss": 0.6312,
+      "step": 3810
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.20280030369758606,
+      "learning_rate": 4.7523974451274275e-05,
+      "loss": 0.6349,
+      "step": 3820
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.24433667957782745,
+      "learning_rate": 4.7506638139307966e-05,
+      "loss": 0.6648,
+      "step": 3830
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.3273336887359619,
+      "learning_rate": 4.7489244532061225e-05,
+      "loss": 0.6498,
+      "step": 3840
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.23943009972572327,
+      "learning_rate": 4.747179367381307e-05,
+      "loss": 0.6358,
+      "step": 3850
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.26670822501182556,
+      "learning_rate": 4.745428560898824e-05,
+      "loss": 0.6563,
+      "step": 3860
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.18882083892822266,
+      "learning_rate": 4.7436720382157116e-05,
+      "loss": 0.6358,
+      "step": 3870
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.23588411509990692,
+      "learning_rate": 4.741909803803562e-05,
+      "loss": 0.6396,
+      "step": 3880
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.24063749611377716,
+      "learning_rate": 4.740141862148503e-05,
+      "loss": 0.6412,
+      "step": 3890
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.19155845046043396,
+      "learning_rate": 4.738368217751196e-05,
+      "loss": 0.62,
+      "step": 3900
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.20259159803390503,
+      "learning_rate": 4.736588875126816e-05,
+      "loss": 0.6172,
+      "step": 3910
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.22824090719223022,
+      "learning_rate": 4.734803838805048e-05,
+      "loss": 0.6286,
+      "step": 3920
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.2356242537498474,
+      "learning_rate": 4.7330131133300686e-05,
+      "loss": 0.6121,
+      "step": 3930
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.22052697837352753,
+      "learning_rate": 4.731216703260538e-05,
+      "loss": 0.6402,
+      "step": 3940
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.2817922532558441,
+      "learning_rate": 4.7294146131695874e-05,
+      "loss": 0.6256,
+      "step": 3950
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.23476284742355347,
+      "learning_rate": 4.7276068476448097e-05,
+      "loss": 0.644,
+      "step": 3960
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.25303345918655396,
+      "learning_rate": 4.725793411288242e-05,
+      "loss": 0.649,
+      "step": 3970
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.2615244686603546,
+      "learning_rate": 4.723974308716361e-05,
+      "loss": 0.6158,
+      "step": 3980
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.2798680365085602,
+      "learning_rate": 4.722149544560067e-05,
+      "loss": 0.6149,
+      "step": 3990
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.19780561327934265,
+      "learning_rate": 4.720319123464672e-05,
+      "loss": 0.6167,
+      "step": 4000
+    },
+    {
+      "epoch": 0.97,
+      "eval_loss": 0.6494756937026978,
+      "eval_runtime": 67.2363,
+      "eval_samples_per_second": 29.746,
+      "eval_steps_per_second": 0.937,
+      "step": 4000
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.22284342348575592,
+      "learning_rate": 4.718483050089891e-05,
+      "loss": 0.6282,
+      "step": 4010
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.242294043302536,
+      "learning_rate": 4.7166413291098246e-05,
+      "loss": 0.6418,
+      "step": 4020
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.19954660534858704,
+      "learning_rate": 4.714793965212955e-05,
+      "loss": 0.6293,
+      "step": 4030
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.2239541858434677,
+      "learning_rate": 4.712940963102126e-05,
+      "loss": 0.6441,
+      "step": 4040
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.2324070930480957,
+      "learning_rate": 4.711082327494536e-05,
+      "loss": 0.6355,
+      "step": 4050
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.27540045976638794,
+      "learning_rate": 4.709218063121725e-05,
+      "loss": 0.6476,
+      "step": 4060
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.37015092372894287,
+      "learning_rate": 4.7073481747295614e-05,
+      "loss": 0.6161,
+      "step": 4070
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.2678948938846588,
+      "learning_rate": 4.7054726670782304e-05,
+      "loss": 0.6404,
+      "step": 4080
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.2190527468919754,
+      "learning_rate": 4.703591544942224e-05,
+      "loss": 0.6619,
+      "step": 4090
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.23498305678367615,
+      "learning_rate": 4.701704813110325e-05,
+      "loss": 0.6297,
+      "step": 4100
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.2132258415222168,
+      "learning_rate": 4.6998124763855984e-05,
+      "loss": 0.6462,
+      "step": 4110
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.2967916429042816,
+      "learning_rate": 4.697914539585376e-05,
+      "loss": 0.6347,
+      "step": 4120
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.2503927946090698,
+      "learning_rate": 4.6960110075412473e-05,
+      "loss": 0.6264,
+      "step": 4130
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.2261209934949875,
+      "learning_rate": 4.694101885099045e-05,
+      "loss": 0.6351,
+      "step": 4140
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.2164410799741745,
+      "learning_rate": 4.692187177118832e-05,
+      "loss": 0.626,
+      "step": 4150
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.2934332489967346,
+      "learning_rate": 4.690266888474893e-05,
+      "loss": 0.6243,
+      "step": 4160
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.23883183300495148,
+      "learning_rate": 4.688341024055718e-05,
+      "loss": 0.624,
+      "step": 4170
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.1730484664440155,
+      "learning_rate": 4.686409588763991e-05,
+      "loss": 0.6473,
+      "step": 4180
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.27762770652770996,
+      "learning_rate": 4.6844725875165775e-05,
+      "loss": 0.6508,
+      "step": 4190
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.243989497423172,
+      "learning_rate": 4.682530025244514e-05,
+      "loss": 0.6305,
+      "step": 4200
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.21361881494522095,
+      "learning_rate": 4.6805819068929925e-05,
+      "loss": 0.5969,
+      "step": 4210
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.22126366198062897,
+      "learning_rate": 4.678628237421348e-05,
+      "loss": 0.6192,
+      "step": 4220
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.2569611370563507,
+      "learning_rate": 4.6766690218030495e-05,
+      "loss": 0.6235,
+      "step": 4230
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.26861709356307983,
+      "learning_rate": 4.674704265025683e-05,
+      "loss": 0.6413,
+      "step": 4240
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.19330435991287231,
+      "learning_rate": 4.672733972090943e-05,
+      "loss": 0.6413,
+      "step": 4250
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.25647568702697754,
+      "learning_rate": 4.6707581480146136e-05,
+      "loss": 0.6411,
+      "step": 4260
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.23879855871200562,
+      "learning_rate": 4.6687767978265625e-05,
+      "loss": 0.6188,
+      "step": 4270
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.25180870294570923,
+      "learning_rate": 4.666789926570725e-05,
+      "loss": 0.6388,
+      "step": 4280
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.26070743799209595,
+      "learning_rate": 4.6647975393050904e-05,
+      "loss": 0.6254,
+      "step": 4290
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.2934954762458801,
+      "learning_rate": 4.662799641101691e-05,
+      "loss": 0.6089,
+      "step": 4300
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.24827881157398224,
+      "learning_rate": 4.6607962370465866e-05,
+      "loss": 0.6233,
+      "step": 4310
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.7880712151527405,
+      "learning_rate": 4.658787332239856e-05,
+      "loss": 0.6305,
+      "step": 4320
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.23941326141357422,
+      "learning_rate": 4.6567729317955796e-05,
+      "loss": 0.6449,
+      "step": 4330
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.21050909161567688,
+      "learning_rate": 4.654753040841829e-05,
+      "loss": 0.6324,
+      "step": 4340
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.2372525930404663,
+      "learning_rate": 4.6527276645206516e-05,
+      "loss": 0.6098,
+      "step": 4350
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.23655132949352264,
+      "learning_rate": 4.65069680798806e-05,
+      "loss": 0.6457,
+      "step": 4360
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.26810187101364136,
+      "learning_rate": 4.648660476414017e-05,
+      "loss": 0.6327,
+      "step": 4370
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.2635047137737274,
+      "learning_rate": 4.6466186749824235e-05,
+      "loss": 0.6359,
+      "step": 4380
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.2955858111381531,
+      "learning_rate": 4.6445714088911076e-05,
+      "loss": 0.6412,
+      "step": 4390
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.2971380949020386,
+      "learning_rate": 4.6425186833518054e-05,
+      "loss": 0.61,
+      "step": 4400
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.22041000425815582,
+      "learning_rate": 4.6404605035901505e-05,
+      "loss": 0.6494,
+      "step": 4410
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.2463519424200058,
+      "learning_rate": 4.638396874845666e-05,
+      "loss": 0.6223,
+      "step": 4420
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.2519693374633789,
+      "learning_rate": 4.636327802371742e-05,
+      "loss": 0.6399,
+      "step": 4430
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.23022256791591644,
+      "learning_rate": 4.6342532914356284e-05,
+      "loss": 0.6106,
+      "step": 4440
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.21000300347805023,
+      "learning_rate": 4.632173347318421e-05,
+      "loss": 0.6463,
+      "step": 4450
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.2383776307106018,
+      "learning_rate": 4.630087975315045e-05,
+      "loss": 0.6202,
+      "step": 4460
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.2528972327709198,
+      "learning_rate": 4.627997180734244e-05,
+      "loss": 0.6401,
+      "step": 4470
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.20749187469482422,
+      "learning_rate": 4.625900968898565e-05,
+      "loss": 0.6541,
+      "step": 4480
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.2283320277929306,
+      "learning_rate": 4.623799345144348e-05,
+      "loss": 0.6311,
+      "step": 4490
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.21833132207393646,
+      "learning_rate": 4.6216923148217096e-05,
+      "loss": 0.6272,
+      "step": 4500
+    },
+    {
+      "epoch": 1.09,
+      "eval_loss": 0.6476519107818604,
+      "eval_runtime": 67.279,
+      "eval_samples_per_second": 29.727,
+      "eval_steps_per_second": 0.936,
+      "step": 4500
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.2190130203962326,
+      "learning_rate": 4.619579883294528e-05,
+      "loss": 0.6405,
+      "step": 4510
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.32068029046058655,
+      "learning_rate": 4.617462055940433e-05,
+      "loss": 0.6101,
+      "step": 4520
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.2526598572731018,
+      "learning_rate": 4.6153388381507886e-05,
+      "loss": 0.6252,
+      "step": 4530
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.19768927991390228,
+      "learning_rate": 4.613210235330686e-05,
+      "loss": 0.6138,
+      "step": 4540
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.24341870844364166,
+      "learning_rate": 4.611076252898919e-05,
+      "loss": 0.6072,
+      "step": 4550
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.2283903807401657,
+      "learning_rate": 4.60893689628798e-05,
+      "loss": 0.6396,
+      "step": 4560
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.21138328313827515,
+      "learning_rate": 4.606792170944041e-05,
+      "loss": 0.6306,
+      "step": 4570
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.4939797818660736,
+      "learning_rate": 4.604642082326944e-05,
+      "loss": 0.6259,
+      "step": 4580
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.22861245274543762,
+      "learning_rate": 4.60248663591018e-05,
+      "loss": 0.6274,
+      "step": 4590
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.35914289951324463,
+      "learning_rate": 4.6003258371808825e-05,
+      "loss": 0.6357,
+      "step": 4600
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.2684130072593689,
+      "learning_rate": 4.598159691639809e-05,
+      "loss": 0.6299,
+      "step": 4610
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.2696840763092041,
+      "learning_rate": 4.5959882048013294e-05,
+      "loss": 0.6224,
+      "step": 4620
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.21357202529907227,
+      "learning_rate": 4.59381138219341e-05,
+      "loss": 0.6119,
+      "step": 4630
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.20233683288097382,
+      "learning_rate": 4.591629229357601e-05,
+      "loss": 0.5956,
+      "step": 4640
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.2905120849609375,
+      "learning_rate": 4.5894417518490225e-05,
+      "loss": 0.6106,
+      "step": 4650
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.2116033285856247,
+      "learning_rate": 4.5872489552363475e-05,
+      "loss": 0.6218,
+      "step": 4660
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.22969362139701843,
+      "learning_rate": 4.585050845101791e-05,
+      "loss": 0.6419,
+      "step": 4670
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.23150619864463806,
+      "learning_rate": 4.582847427041097e-05,
+      "loss": 0.6128,
+      "step": 4680
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.23322267830371857,
+      "learning_rate": 4.580638706663517e-05,
+      "loss": 0.6114,
+      "step": 4690
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.23113666474819183,
+      "learning_rate": 4.578424689591805e-05,
+      "loss": 0.6173,
+      "step": 4700
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.2632519602775574,
+      "learning_rate": 4.5762053814621975e-05,
+      "loss": 0.643,
+      "step": 4710
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.19099248945713043,
+      "learning_rate": 4.573980787924399e-05,
+      "loss": 0.6201,
+      "step": 4720
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.23567742109298706,
+      "learning_rate": 4.5717509146415705e-05,
+      "loss": 0.6163,
+      "step": 4730
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.21497173607349396,
+      "learning_rate": 4.5695157672903144e-05,
+      "loss": 0.6226,
+      "step": 4740
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.22423885762691498,
+      "learning_rate": 4.567275351560658e-05,
+      "loss": 0.6052,
+      "step": 4750
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.2486392855644226,
+      "learning_rate": 4.5650296731560396e-05,
+      "loss": 0.6102,
+      "step": 4760
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.19872063398361206,
+      "learning_rate": 4.562778737793298e-05,
+      "loss": 0.633,
+      "step": 4770
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.19403629004955292,
+      "learning_rate": 4.560522551202651e-05,
+      "loss": 0.6074,
+      "step": 4780
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.2101258486509323,
+      "learning_rate": 4.558261119127686e-05,
+      "loss": 0.6019,
+      "step": 4790
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.2554053068161011,
+      "learning_rate": 4.555994447325344e-05,
+      "loss": 0.6095,
+      "step": 4800
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.3209736943244934,
+      "learning_rate": 4.5537225415659054e-05,
+      "loss": 0.6095,
+      "step": 4810
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.1882716864347458,
+      "learning_rate": 4.551445407632973e-05,
+      "loss": 0.624,
+      "step": 4820
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.219781294465065,
+      "learning_rate": 4.549163051323461e-05,
+      "loss": 0.6212,
+      "step": 4830
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.1968742161989212,
+      "learning_rate": 4.5468754784475764e-05,
+      "loss": 0.6088,
+      "step": 4840
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.23859383165836334,
+      "learning_rate": 4.5445826948288074e-05,
+      "loss": 0.6429,
+      "step": 4850
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.21909688413143158,
+      "learning_rate": 4.542284706303906e-05,
+      "loss": 0.6163,
+      "step": 4860
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.23673193156719208,
+      "learning_rate": 4.539981518722876e-05,
+      "loss": 0.6261,
+      "step": 4870
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.23555058240890503,
+      "learning_rate": 4.537673137948954e-05,
+      "loss": 0.598,
+      "step": 4880
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.3052713871002197,
+      "learning_rate": 4.5353595698586e-05,
+      "loss": 0.6485,
+      "step": 4890
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.2587834298610687,
+      "learning_rate": 4.533040820341477e-05,
+      "loss": 0.6104,
+      "step": 4900
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.2710438370704651,
+      "learning_rate": 4.53071689530044e-05,
+      "loss": 0.6286,
+      "step": 4910
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.2709255516529083,
+      "learning_rate": 4.528387800651517e-05,
+      "loss": 0.6199,
+      "step": 4920
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.3711196780204773,
+      "learning_rate": 4.5260535423239e-05,
+      "loss": 0.6303,
+      "step": 4930
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.2359853833913803,
+      "learning_rate": 4.523714126259923e-05,
+      "loss": 0.6232,
+      "step": 4940
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.2610563039779663,
+      "learning_rate": 4.5213695584150495e-05,
+      "loss": 0.6266,
+      "step": 4950
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.23091551661491394,
+      "learning_rate": 4.519019844757863e-05,
+      "loss": 0.6264,
+      "step": 4960
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.20323573052883148,
+      "learning_rate": 4.516664991270041e-05,
+      "loss": 0.6137,
+      "step": 4970
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.2040981501340866,
+      "learning_rate": 4.5143050039463476e-05,
+      "loss": 0.6433,
+      "step": 4980
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.24421969056129456,
+      "learning_rate": 4.511939888794617e-05,
+      "loss": 0.6221,
+      "step": 4990
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.22230570018291473,
+      "learning_rate": 4.5095696518357375e-05,
+      "loss": 0.6002,
+      "step": 5000
+    },
+    {
+      "epoch": 1.21,
+      "eval_loss": 0.6445114016532898,
+      "eval_runtime": 67.1925,
+      "eval_samples_per_second": 29.765,
+      "eval_steps_per_second": 0.938,
+      "step": 5000
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.2543829381465912,
+      "learning_rate": 4.5071942991036354e-05,
+      "loss": 0.6598,
+      "step": 5010
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.29779505729675293,
+      "learning_rate": 4.504813836645258e-05,
+      "loss": 0.6196,
+      "step": 5020
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.2295481115579605,
+      "learning_rate": 4.5024282705205635e-05,
+      "loss": 0.6227,
+      "step": 5030
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.234407439827919,
+      "learning_rate": 4.500037606802502e-05,
+      "loss": 0.6363,
+      "step": 5040
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.2559088468551636,
+      "learning_rate": 4.497641851577e-05,
+      "loss": 0.6471,
+      "step": 5050
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.21824946999549866,
+      "learning_rate": 4.495241010942945e-05,
+      "loss": 0.6362,
+      "step": 5060
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.2934998571872711,
+      "learning_rate": 4.4928350910121714e-05,
+      "loss": 0.6285,
+      "step": 5070
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.2297659069299698,
+      "learning_rate": 4.4904240979094436e-05,
+      "loss": 0.6102,
+      "step": 5080
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.2708946466445923,
+      "learning_rate": 4.488008037772441e-05,
+      "loss": 0.6186,
+      "step": 5090
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.23141387104988098,
+      "learning_rate": 4.485586916751743e-05,
+      "loss": 0.6314,
+      "step": 5100
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.2494589388370514,
+      "learning_rate": 4.483160741010809e-05,
+      "loss": 0.6413,
+      "step": 5110
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.2377118617296219,
+      "learning_rate": 4.480729516725971e-05,
+      "loss": 0.617,
+      "step": 5120
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.2568589746952057,
+      "learning_rate": 4.47829325008641e-05,
+      "loss": 0.6143,
+      "step": 5130
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.23955343663692474,
+      "learning_rate": 4.475851947294145e-05,
+      "loss": 0.6257,
+      "step": 5140
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.2029058337211609,
+      "learning_rate": 4.4734056145640135e-05,
+      "loss": 0.6074,
+      "step": 5150
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.3426890969276428,
+      "learning_rate": 4.470954258123661e-05,
+      "loss": 0.6511,
+      "step": 5160
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.28450432419776917,
+      "learning_rate": 4.4684978842135175e-05,
+      "loss": 0.6227,
+      "step": 5170
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.2370108813047409,
+      "learning_rate": 4.4660364990867895e-05,
+      "loss": 0.6198,
+      "step": 5180
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.33506277203559875,
+      "learning_rate": 4.463570109009441e-05,
+      "loss": 0.6112,
+      "step": 5190
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.21070849895477295,
+      "learning_rate": 4.461098720260173e-05,
+      "loss": 0.606,
+      "step": 5200
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.3238009214401245,
+      "learning_rate": 4.458622339130416e-05,
+      "loss": 0.6413,
+      "step": 5210
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.23879234492778778,
+      "learning_rate": 4.456140971924309e-05,
+      "loss": 0.6229,
+      "step": 5220
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.24163304269313812,
+      "learning_rate": 4.45365462495868e-05,
+      "loss": 0.6268,
+      "step": 5230
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.22378313541412354,
+      "learning_rate": 4.451163304563038e-05,
+      "loss": 0.6218,
+      "step": 5240
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.21948544681072235,
+      "learning_rate": 4.448667017079554e-05,
+      "loss": 0.5945,
+      "step": 5250
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.2834113836288452,
+      "learning_rate": 4.44616576886304e-05,
+      "loss": 0.6449,
+      "step": 5260
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.2503283619880676,
+      "learning_rate": 4.443659566280938e-05,
+      "loss": 0.6243,
+      "step": 5270
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.35525229573249817,
+      "learning_rate": 4.441148415713303e-05,
+      "loss": 0.6283,
+      "step": 5280
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.2495538890361786,
+      "learning_rate": 4.438632323552786e-05,
+      "loss": 0.5959,
+      "step": 5290
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.24201400578022003,
+      "learning_rate": 4.436111296204617e-05,
+      "loss": 0.6242,
+      "step": 5300
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.24746832251548767,
+      "learning_rate": 4.4335853400865906e-05,
+      "loss": 0.5989,
+      "step": 5310
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.2049688845872879,
+      "learning_rate": 4.4310544616290475e-05,
+      "loss": 0.6375,
+      "step": 5320
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.2205064594745636,
+      "learning_rate": 4.42851866727486e-05,
+      "loss": 0.6253,
+      "step": 5330
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.2159041166305542,
+      "learning_rate": 4.425977963479415e-05,
+      "loss": 0.6027,
+      "step": 5340
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.2615651488304138,
+      "learning_rate": 4.423432356710597e-05,
+      "loss": 0.6134,
+      "step": 5350
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.23875150084495544,
+      "learning_rate": 4.420881853448771e-05,
+      "loss": 0.5926,
+      "step": 5360
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.25981611013412476,
+      "learning_rate": 4.418326460186769e-05,
+      "loss": 0.623,
+      "step": 5370
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.5433134436607361,
+      "learning_rate": 4.415766183429871e-05,
+      "loss": 0.6143,
+      "step": 5380
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.23302482068538666,
+      "learning_rate": 4.413201029695788e-05,
+      "loss": 0.625,
+      "step": 5390
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.2442236691713333,
+      "learning_rate": 4.410631005514646e-05,
+      "loss": 0.623,
+      "step": 5400
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.25277549028396606,
+      "learning_rate": 4.4080561174289713e-05,
+      "loss": 0.6298,
+      "step": 5410
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.26706215739250183,
+      "learning_rate": 4.405476371993673e-05,
+      "loss": 0.5957,
+      "step": 5420
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.19037441909313202,
+      "learning_rate": 4.4028917757760216e-05,
+      "loss": 0.6228,
+      "step": 5430
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.21772147715091705,
+      "learning_rate": 4.4003023353556406e-05,
+      "loss": 0.6402,
+      "step": 5440
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.19529011845588684,
+      "learning_rate": 4.3977080573244836e-05,
+      "loss": 0.5973,
+      "step": 5450
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.22065111994743347,
+      "learning_rate": 4.395108948286818e-05,
+      "loss": 0.6181,
+      "step": 5460
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.3971559405326843,
+      "learning_rate": 4.3925050148592124e-05,
+      "loss": 0.6217,
+      "step": 5470
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.2587713897228241,
+      "learning_rate": 4.389896263670514e-05,
+      "loss": 0.5931,
+      "step": 5480
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.23226861655712128,
+      "learning_rate": 4.387282701361837e-05,
+      "loss": 0.6207,
+      "step": 5490
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.2062036246061325,
+      "learning_rate": 4.3846643345865425e-05,
+      "loss": 0.6303,
+      "step": 5500
+    },
+    {
+      "epoch": 1.33,
+      "eval_loss": 0.6429007649421692,
+      "eval_runtime": 67.2472,
+      "eval_samples_per_second": 29.741,
+      "eval_steps_per_second": 0.937,
+      "step": 5500
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.2225368171930313,
+      "learning_rate": 4.3820411700102216e-05,
+      "loss": 0.6123,
+      "step": 5510
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.2737579941749573,
+      "learning_rate": 4.3794132143106795e-05,
+      "loss": 0.6255,
+      "step": 5520
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.2531569004058838,
+      "learning_rate": 4.376780474177918e-05,
+      "loss": 0.5988,
+      "step": 5530
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.27578508853912354,
+      "learning_rate": 4.37414295631412e-05,
+      "loss": 0.6075,
+      "step": 5540
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.2573574185371399,
+      "learning_rate": 4.37150066743363e-05,
+      "loss": 0.6217,
+      "step": 5550
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.28215906023979187,
+      "learning_rate": 4.3688536142629376e-05,
+      "loss": 0.6202,
+      "step": 5560
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.23517820239067078,
+      "learning_rate": 4.366201803540663e-05,
+      "loss": 0.6056,
+      "step": 5570
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.26566624641418457,
+      "learning_rate": 4.363545242017534e-05,
+      "loss": 0.6046,
+      "step": 5580
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.20942912995815277,
+      "learning_rate": 4.360883936456377e-05,
+      "loss": 0.5939,
+      "step": 5590
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.24037909507751465,
+      "learning_rate": 4.358217893632093e-05,
+      "loss": 0.6218,
+      "step": 5600
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.22285695374011993,
+      "learning_rate": 4.355547120331641e-05,
+      "loss": 0.6332,
+      "step": 5610
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.2597566545009613,
+      "learning_rate": 4.352871623354025e-05,
+      "loss": 0.6626,
+      "step": 5620
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.37357744574546814,
+      "learning_rate": 4.350191409510275e-05,
+      "loss": 0.6143,
+      "step": 5630
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.24381165206432343,
+      "learning_rate": 4.347506485623424e-05,
+      "loss": 0.6369,
+      "step": 5640
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.27438944578170776,
+      "learning_rate": 4.344816858528501e-05,
+      "loss": 0.6535,
+      "step": 5650
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.20593568682670593,
+      "learning_rate": 4.342122535072505e-05,
+      "loss": 0.6251,
+      "step": 5660
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.25618818402290344,
+      "learning_rate": 4.33942352211439e-05,
+      "loss": 0.6182,
+      "step": 5670
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.20692601799964905,
+      "learning_rate": 4.33671982652505e-05,
+      "loss": 0.6261,
+      "step": 5680
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.2003064602613449,
+      "learning_rate": 4.334011455187298e-05,
+      "loss": 0.6331,
+      "step": 5690
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.22887033224105835,
+      "learning_rate": 4.3312984149958516e-05,
+      "loss": 0.6292,
+      "step": 5700
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.21389944851398468,
+      "learning_rate": 4.328580712857313e-05,
+      "loss": 0.6094,
+      "step": 5710
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.26887232065200806,
+      "learning_rate": 4.32585835569015e-05,
+      "loss": 0.5942,
+      "step": 5720
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.30249106884002686,
+      "learning_rate": 4.3231313504246875e-05,
+      "loss": 0.6473,
+      "step": 5730
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.2598324418067932,
+      "learning_rate": 4.320399704003075e-05,
+      "loss": 0.6198,
+      "step": 5740
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.27700358629226685,
+      "learning_rate": 4.317663423379283e-05,
+      "loss": 0.6378,
+      "step": 5750
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.22245779633522034,
+      "learning_rate": 4.3149225155190766e-05,
+      "loss": 0.6458,
+      "step": 5760
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.27054449915885925,
+      "learning_rate": 4.3121769874e-05,
+      "loss": 0.6347,
+      "step": 5770
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.22811339795589447,
+      "learning_rate": 4.309426846011362e-05,
+      "loss": 0.6382,
+      "step": 5780
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.27327001094818115,
+      "learning_rate": 4.306672098354211e-05,
+      "loss": 0.6114,
+      "step": 5790
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.24687005579471588,
+      "learning_rate": 4.3039127514413266e-05,
+      "loss": 0.5949,
+      "step": 5800
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.39029061794281006,
+      "learning_rate": 4.3011488122971945e-05,
+      "loss": 0.6281,
+      "step": 5810
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.2590855360031128,
+      "learning_rate": 4.298380287957989e-05,
+      "loss": 0.6068,
+      "step": 5820
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.18194353580474854,
+      "learning_rate": 4.29560718547156e-05,
+      "loss": 0.6114,
+      "step": 5830
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.8544417023658752,
+      "learning_rate": 4.292829511897409e-05,
+      "loss": 0.6336,
+      "step": 5840
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.23495988547801971,
+      "learning_rate": 4.290047274306678e-05,
+      "loss": 0.6185,
+      "step": 5850
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.20022150874137878,
+      "learning_rate": 4.2872604797821246e-05,
+      "loss": 0.6001,
+      "step": 5860
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.40812554955482483,
+      "learning_rate": 4.284469135418108e-05,
+      "loss": 0.6365,
+      "step": 5870
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.23241877555847168,
+      "learning_rate": 4.28167324832057e-05,
+      "loss": 0.6046,
+      "step": 5880
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.26784956455230713,
+      "learning_rate": 4.2788728256070165e-05,
+      "loss": 0.604,
+      "step": 5890
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.25370949506759644,
+      "learning_rate": 4.2760678744065005e-05,
+      "loss": 0.6368,
+      "step": 5900
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.2346085160970688,
+      "learning_rate": 4.273258401859602e-05,
+      "loss": 0.5944,
+      "step": 5910
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.32089290022850037,
+      "learning_rate": 4.2704444151184117e-05,
+      "loss": 0.6124,
+      "step": 5920
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.19013577699661255,
+      "learning_rate": 4.267625921346513e-05,
+      "loss": 0.6167,
+      "step": 5930
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.27597373723983765,
+      "learning_rate": 4.2648029277189616e-05,
+      "loss": 0.6002,
+      "step": 5940
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.2857840657234192,
+      "learning_rate": 4.261975441422268e-05,
+      "loss": 0.5885,
+      "step": 5950
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.22939664125442505,
+      "learning_rate": 4.259143469654382e-05,
+      "loss": 0.6086,
+      "step": 5960
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.24333885312080383,
+      "learning_rate": 4.25630701962467e-05,
+      "loss": 0.6066,
+      "step": 5970
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.2737683057785034,
+      "learning_rate": 4.253466098553899e-05,
+      "loss": 0.6338,
+      "step": 5980
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.23578789830207825,
+      "learning_rate": 4.2506207136742196e-05,
+      "loss": 0.6088,
+      "step": 5990
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.3092876076698303,
+      "learning_rate": 4.247770872229143e-05,
+      "loss": 0.6405,
+      "step": 6000
+    },
+    {
+      "epoch": 1.45,
+      "eval_loss": 0.6421223282814026,
+      "eval_runtime": 67.2062,
+      "eval_samples_per_second": 29.759,
+      "eval_steps_per_second": 0.937,
+      "step": 6000
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.24720308184623718,
+      "learning_rate": 4.2449165814735294e-05,
+      "loss": 0.6489,
+      "step": 6010
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.2425301969051361,
+      "learning_rate": 4.242057848673561e-05,
+      "loss": 0.6237,
+      "step": 6020
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.1967163234949112,
+      "learning_rate": 4.239194681106733e-05,
+      "loss": 0.6256,
+      "step": 6030
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.24658535420894623,
+      "learning_rate": 4.2363270860618264e-05,
+      "loss": 0.5891,
+      "step": 6040
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.22012916207313538,
+      "learning_rate": 4.233455070838895e-05,
+      "loss": 0.6072,
+      "step": 6050
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.32060471177101135,
+      "learning_rate": 4.230578642749245e-05,
+      "loss": 0.62,
+      "step": 6060
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.2524150013923645,
+      "learning_rate": 4.2276978091154164e-05,
+      "loss": 0.6002,
+      "step": 6070
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.22831355035305023,
+      "learning_rate": 4.2248125772711635e-05,
+      "loss": 0.6245,
+      "step": 6080
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.3438965678215027,
+      "learning_rate": 4.221922954561439e-05,
+      "loss": 0.6127,
+      "step": 6090
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.20607294142246246,
+      "learning_rate": 4.2190289483423704e-05,
+      "loss": 0.6074,
+      "step": 6100
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.3653343915939331,
+      "learning_rate": 4.216130565981249e-05,
+      "loss": 0.6156,
+      "step": 6110
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.19855797290802002,
+      "learning_rate": 4.213227814856501e-05,
+      "loss": 0.6107,
+      "step": 6120
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.2605169415473938,
+      "learning_rate": 4.210320702357678e-05,
+      "loss": 0.6246,
+      "step": 6130
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.2300351858139038,
+      "learning_rate": 4.207409235885435e-05,
+      "loss": 0.6347,
+      "step": 6140
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.9489219188690186,
+      "learning_rate": 4.2044934228515063e-05,
+      "loss": 0.6217,
+      "step": 6150
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.2525247037410736,
+      "learning_rate": 4.201573270678696e-05,
+      "loss": 0.6115,
+      "step": 6160
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.2521913945674896,
+      "learning_rate": 4.1986487868008525e-05,
+      "loss": 0.5965,
+      "step": 6170
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.24830523133277893,
+      "learning_rate": 4.195719978662851e-05,
+      "loss": 0.618,
+      "step": 6180
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.23050466179847717,
+      "learning_rate": 4.192786853720576e-05,
+      "loss": 0.6135,
+      "step": 6190
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.21613763272762299,
+      "learning_rate": 4.1898494194409e-05,
+      "loss": 0.6155,
+      "step": 6200
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.22277826070785522,
+      "learning_rate": 4.1869076833016676e-05,
+      "loss": 0.611,
+      "step": 6210
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.2443763017654419,
+      "learning_rate": 4.1839616527916736e-05,
+      "loss": 0.5888,
+      "step": 6220
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.513171374797821,
+      "learning_rate": 4.181011335410645e-05,
+      "loss": 0.6096,
+      "step": 6230
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.21167334914207458,
+      "learning_rate": 4.178056738669221e-05,
+      "loss": 0.6108,
+      "step": 6240
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.22910219430923462,
+      "learning_rate": 4.175097870088937e-05,
+      "loss": 0.6172,
+      "step": 6250
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.2219991236925125,
+      "learning_rate": 4.172134737202202e-05,
+      "loss": 0.6188,
+      "step": 6260
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.266703337430954,
+      "learning_rate": 4.16916734755228e-05,
+      "loss": 0.6331,
+      "step": 6270
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.23374725878238678,
+      "learning_rate": 4.166195708693273e-05,
+      "loss": 0.6173,
+      "step": 6280
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.2778140902519226,
+      "learning_rate": 4.163219828190099e-05,
+      "loss": 0.6345,
+      "step": 6290
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.23954761028289795,
+      "learning_rate": 4.160239713618474e-05,
+      "loss": 0.6205,
+      "step": 6300
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.23126688599586487,
+      "learning_rate": 4.1572553725648945e-05,
+      "loss": 0.6136,
+      "step": 6310
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.209469735622406,
+      "learning_rate": 4.1542668126266135e-05,
+      "loss": 0.6181,
+      "step": 6320
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.24480338394641876,
+      "learning_rate": 4.151274041411626e-05,
+      "loss": 0.627,
+      "step": 6330
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.27108994126319885,
+      "learning_rate": 4.148277066538648e-05,
+      "loss": 0.6169,
+      "step": 6340
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.2842053472995758,
+      "learning_rate": 4.145275895637095e-05,
+      "loss": 0.6347,
+      "step": 6350
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.23858144879341125,
+      "learning_rate": 4.142270536347066e-05,
+      "loss": 0.6172,
+      "step": 6360
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.18353480100631714,
+      "learning_rate": 4.139260996319321e-05,
+      "loss": 0.5939,
+      "step": 6370
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.24537691473960876,
+      "learning_rate": 4.136247283215263e-05,
+      "loss": 0.6219,
+      "step": 6380
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.30072882771492004,
+      "learning_rate": 4.1332294047069215e-05,
+      "loss": 0.6368,
+      "step": 6390
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.2931046485900879,
+      "learning_rate": 4.1302073684769256e-05,
+      "loss": 0.6108,
+      "step": 6400
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.23465721309185028,
+      "learning_rate": 4.1271811822184914e-05,
+      "loss": 0.6262,
+      "step": 6410
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.21646186709403992,
+      "learning_rate": 4.124150853635398e-05,
+      "loss": 0.6221,
+      "step": 6420
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.19293636083602905,
+      "learning_rate": 4.1211163904419716e-05,
+      "loss": 0.6147,
+      "step": 6430
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.2320639044046402,
+      "learning_rate": 4.1180778003630624e-05,
+      "loss": 0.5902,
+      "step": 6440
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.24089358747005463,
+      "learning_rate": 4.115035091134027e-05,
+      "loss": 0.6307,
+      "step": 6450
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.2302493005990982,
+      "learning_rate": 4.1119882705007094e-05,
+      "loss": 0.6283,
+      "step": 6460
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.23400545120239258,
+      "learning_rate": 4.108937346219417e-05,
+      "loss": 0.612,
+      "step": 6470
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.21376444399356842,
+      "learning_rate": 4.1058823260569056e-05,
+      "loss": 0.6185,
+      "step": 6480
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.23962020874023438,
+      "learning_rate": 4.1028232177903586e-05,
+      "loss": 0.599,
+      "step": 6490
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.20366187393665314,
+      "learning_rate": 4.099760029207366e-05,
+      "loss": 0.6041,
+      "step": 6500
+    },
+    {
+      "epoch": 1.57,
+      "eval_loss": 0.6386786103248596,
+      "eval_runtime": 67.1728,
+      "eval_samples_per_second": 29.774,
+      "eval_steps_per_second": 0.938,
+      "step": 6500
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.19989852607250214,
+      "learning_rate": 4.096692768105904e-05,
+      "loss": 0.6166,
+      "step": 6510
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.2329440414905548,
+      "learning_rate": 4.093621442294318e-05,
+      "loss": 0.6111,
+      "step": 6520
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.3589790165424347,
+      "learning_rate": 4.0905460595912995e-05,
+      "loss": 0.5906,
+      "step": 6530
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.21661555767059326,
+      "learning_rate": 4.0874666278258696e-05,
+      "loss": 0.6191,
+      "step": 6540
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.21175633370876312,
+      "learning_rate": 4.0843831548373536e-05,
+      "loss": 0.6092,
+      "step": 6550
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.2863156199455261,
+      "learning_rate": 4.0812956484753686e-05,
+      "loss": 0.615,
+      "step": 6560
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.2971532344818115,
+      "learning_rate": 4.078204116599796e-05,
+      "loss": 0.6144,
+      "step": 6570
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.3574163019657135,
+      "learning_rate": 4.0751085670807684e-05,
+      "loss": 0.6297,
+      "step": 6580
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.23878632485866547,
+      "learning_rate": 4.072009007798643e-05,
+      "loss": 0.6164,
+      "step": 6590
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.19213823974132538,
+      "learning_rate": 4.0689054466439865e-05,
+      "loss": 0.6161,
+      "step": 6600
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.23007014393806458,
+      "learning_rate": 4.0657978915175524e-05,
+      "loss": 0.5919,
+      "step": 6610
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.6034734845161438,
+      "learning_rate": 4.062686350330262e-05,
+      "loss": 0.601,
+      "step": 6620
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.234625443816185,
+      "learning_rate": 4.059570831003184e-05,
+      "loss": 0.6298,
+      "step": 6630
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.2770217955112457,
+      "learning_rate": 4.056451341467514e-05,
+      "loss": 0.6066,
+      "step": 6640
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.3099740147590637,
+      "learning_rate": 4.053327889664556e-05,
+      "loss": 0.6296,
+      "step": 6650
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.2302497923374176,
+      "learning_rate": 4.050200483545697e-05,
+      "loss": 0.5892,
+      "step": 6660
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.24834463000297546,
+      "learning_rate": 4.047069131072396e-05,
+      "loss": 0.6186,
+      "step": 6670
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.4216982126235962,
+      "learning_rate": 4.043933840216154e-05,
+      "loss": 0.6059,
+      "step": 6680
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.19061365723609924,
+      "learning_rate": 4.040794618958499e-05,
+      "loss": 0.5777,
+      "step": 6690
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.27531304955482483,
+      "learning_rate": 4.0376514752909644e-05,
+      "loss": 0.6293,
+      "step": 6700
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.29000842571258545,
+      "learning_rate": 4.034504417215071e-05,
+      "loss": 0.6534,
+      "step": 6710
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.2087034285068512,
+      "learning_rate": 4.0313534527423014e-05,
+      "loss": 0.6039,
+      "step": 6720
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.23176586627960205,
+      "learning_rate": 4.028198589894086e-05,
+      "loss": 0.6382,
+      "step": 6730
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.23297011852264404,
+      "learning_rate": 4.0250398367017754e-05,
+      "loss": 0.6058,
+      "step": 6740
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.27167224884033203,
+      "learning_rate": 4.021877201206628e-05,
+      "loss": 0.6133,
+      "step": 6750
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.2260424941778183,
+      "learning_rate": 4.018710691459783e-05,
+      "loss": 0.6564,
+      "step": 6760
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.20277553796768188,
+      "learning_rate": 4.015540315522242e-05,
+      "loss": 0.6115,
+      "step": 6770
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.23194430768489838,
+      "learning_rate": 4.012366081464851e-05,
+      "loss": 0.5972,
+      "step": 6780
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.4496138095855713,
+      "learning_rate": 4.009187997368275e-05,
+      "loss": 0.601,
+      "step": 6790
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.2352319061756134,
+      "learning_rate": 4.0060060713229806e-05,
+      "loss": 0.6126,
+      "step": 6800
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.2365293949842453,
+      "learning_rate": 4.002820311429218e-05,
+      "loss": 0.5925,
+      "step": 6810
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.27040597796440125,
+      "learning_rate": 3.9996307257969936e-05,
+      "loss": 0.6373,
+      "step": 6820
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.29712024331092834,
+      "learning_rate": 3.996437322546053e-05,
+      "loss": 0.6174,
+      "step": 6830
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.2615072727203369,
+      "learning_rate": 3.993240109805863e-05,
+      "loss": 0.6315,
+      "step": 6840
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.19362391531467438,
+      "learning_rate": 3.990039095715587e-05,
+      "loss": 0.6553,
+      "step": 6850
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.27068230509757996,
+      "learning_rate": 3.986834288424064e-05,
+      "loss": 0.5997,
+      "step": 6860
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.24307383596897125,
+      "learning_rate": 3.9836256960897916e-05,
+      "loss": 0.6098,
+      "step": 6870
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.2581513226032257,
+      "learning_rate": 3.980413326880902e-05,
+      "loss": 0.618,
+      "step": 6880
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.2150333821773529,
+      "learning_rate": 3.9771971889751426e-05,
+      "loss": 0.6039,
+      "step": 6890
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.28339913487434387,
+      "learning_rate": 3.9739772905598525e-05,
+      "loss": 0.6122,
+      "step": 6900
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.5444418787956238,
+      "learning_rate": 3.970753639831949e-05,
+      "loss": 0.6242,
+      "step": 6910
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.2532576322555542,
+      "learning_rate": 3.967526244997896e-05,
+      "loss": 0.6069,
+      "step": 6920
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.2285206913948059,
+      "learning_rate": 3.9642951142736936e-05,
+      "loss": 0.6039,
+      "step": 6930
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.2928161919116974,
+      "learning_rate": 3.961060255884849e-05,
+      "loss": 0.6098,
+      "step": 6940
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.3057214617729187,
+      "learning_rate": 3.957821678066359e-05,
+      "loss": 0.6299,
+      "step": 6950
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.2918532192707062,
+      "learning_rate": 3.9545793890626925e-05,
+      "loss": 0.6069,
+      "step": 6960
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.20853939652442932,
+      "learning_rate": 3.951333397127762e-05,
+      "loss": 0.6228,
+      "step": 6970
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.2272331267595291,
+      "learning_rate": 3.948083710524909e-05,
+      "loss": 0.6051,
+      "step": 6980
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.22748872637748718,
+      "learning_rate": 3.9448303375268794e-05,
+      "loss": 0.6259,
+      "step": 6990
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.23295633494853973,
+      "learning_rate": 3.9415732864158026e-05,
+      "loss": 0.5912,
+      "step": 7000
+    },
+    {
+      "epoch": 1.69,
+      "eval_loss": 0.6369640827178955,
+      "eval_runtime": 67.3439,
+      "eval_samples_per_second": 29.698,
+      "eval_steps_per_second": 0.935,
+      "step": 7000
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.25066569447517395,
+      "learning_rate": 3.9383125654831747e-05,
+      "loss": 0.594,
+      "step": 7010
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.1922658085823059,
+      "learning_rate": 3.935048183029831e-05,
+      "loss": 0.6048,
+      "step": 7020
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.23781736195087433,
+      "learning_rate": 3.93178014736593e-05,
+      "loss": 0.6002,
+      "step": 7030
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.25118449330329895,
+      "learning_rate": 3.9285084668109276e-05,
+      "loss": 0.6013,
+      "step": 7040
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.2312859296798706,
+      "learning_rate": 3.925233149693561e-05,
+      "loss": 0.6118,
+      "step": 7050
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.25878220796585083,
+      "learning_rate": 3.921954204351826e-05,
+      "loss": 0.6091,
+      "step": 7060
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.25731998682022095,
+      "learning_rate": 3.9186716391329485e-05,
+      "loss": 0.6221,
+      "step": 7070
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.5314686894416809,
+      "learning_rate": 3.9153854623933785e-05,
+      "loss": 0.6488,
+      "step": 7080
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.2735700309276581,
+      "learning_rate": 3.912095682498752e-05,
+      "loss": 0.5929,
+      "step": 7090
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.24641837179660797,
+      "learning_rate": 3.908802307823883e-05,
+      "loss": 0.6052,
+      "step": 7100
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.2285420298576355,
+      "learning_rate": 3.905505346752733e-05,
+      "loss": 0.6117,
+      "step": 7110
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.33809909224510193,
+      "learning_rate": 3.902204807678398e-05,
+      "loss": 0.6045,
+      "step": 7120
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.2631922662258148,
+      "learning_rate": 3.8989006990030776e-05,
+      "loss": 0.5957,
+      "step": 7130
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.2324107438325882,
+      "learning_rate": 3.89559302913806e-05,
+      "loss": 0.6154,
+      "step": 7140
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.28898167610168457,
+      "learning_rate": 3.8922818065037e-05,
+      "loss": 0.607,
+      "step": 7150
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.26003435254096985,
+      "learning_rate": 3.888967039529398e-05,
+      "loss": 0.6149,
+      "step": 7160
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.203714519739151,
+      "learning_rate": 3.885648736653574e-05,
+      "loss": 0.6087,
+      "step": 7170
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.24271652102470398,
+      "learning_rate": 3.8823269063236525e-05,
+      "loss": 0.6175,
+      "step": 7180
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.32109883427619934,
+      "learning_rate": 3.8790015569960346e-05,
+      "loss": 0.5913,
+      "step": 7190
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.23378537595272064,
+      "learning_rate": 3.8756726971360835e-05,
+      "loss": 0.6011,
+      "step": 7200
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.30031922459602356,
+      "learning_rate": 3.872340335218096e-05,
+      "loss": 0.6384,
+      "step": 7210
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.18759281933307648,
+      "learning_rate": 3.869004479725285e-05,
+      "loss": 0.6099,
+      "step": 7220
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.18216826021671295,
+      "learning_rate": 3.865665139149759e-05,
+      "loss": 0.6137,
+      "step": 7230
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.215232253074646,
+      "learning_rate": 3.862322321992495e-05,
+      "loss": 0.6235,
+      "step": 7240
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.28549501299858093,
+      "learning_rate": 3.858976036763323e-05,
+      "loss": 0.6003,
+      "step": 7250
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.2499440610408783,
+      "learning_rate": 3.855626291980901e-05,
+      "loss": 0.6031,
+      "step": 7260
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.22503146529197693,
+      "learning_rate": 3.852273096172694e-05,
+      "loss": 0.6188,
+      "step": 7270
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.2220824956893921,
+      "learning_rate": 3.848916457874952e-05,
+      "loss": 0.6354,
+      "step": 7280
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.20054738223552704,
+      "learning_rate": 3.845556385632688e-05,
+      "loss": 0.6124,
+      "step": 7290
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.28167736530303955,
+      "learning_rate": 3.842192887999659e-05,
+      "loss": 0.6066,
+      "step": 7300
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.23211318254470825,
+      "learning_rate": 3.83882597353834e-05,
+      "loss": 0.6366,
+      "step": 7310
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.24710510671138763,
+      "learning_rate": 3.835455650819905e-05,
+      "loss": 0.5974,
+      "step": 7320
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.20109450817108154,
+      "learning_rate": 3.832081928424205e-05,
+      "loss": 0.6194,
+      "step": 7330
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.228953555226326,
+      "learning_rate": 3.828704814939742e-05,
+      "loss": 0.6056,
+      "step": 7340
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.24477165937423706,
+      "learning_rate": 3.8253243189636554e-05,
+      "loss": 0.6425,
+      "step": 7350
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.2542911767959595,
+      "learning_rate": 3.8219404491016956e-05,
+      "loss": 0.6043,
+      "step": 7360
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.22980964183807373,
+      "learning_rate": 3.8185532139681964e-05,
+      "loss": 0.5909,
+      "step": 7370
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.2753266394138336,
+      "learning_rate": 3.8151626221860636e-05,
+      "loss": 0.6307,
+      "step": 7380
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.2397356778383255,
+      "learning_rate": 3.811768682386747e-05,
+      "loss": 0.6254,
+      "step": 7390
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.2399650514125824,
+      "learning_rate": 3.808371403210219e-05,
+      "loss": 0.6151,
+      "step": 7400
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.26702845096588135,
+      "learning_rate": 3.8049707933049524e-05,
+      "loss": 0.652,
+      "step": 7410
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.2288055717945099,
+      "learning_rate": 3.8015668613279e-05,
+      "loss": 0.5837,
+      "step": 7420
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.25483304262161255,
+      "learning_rate": 3.798159615944471e-05,
+      "loss": 0.6001,
+      "step": 7430
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.21478036046028137,
+      "learning_rate": 3.79474906582851e-05,
+      "loss": 0.6437,
+      "step": 7440
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.24260424077510834,
+      "learning_rate": 3.7913352196622765e-05,
+      "loss": 0.5938,
+      "step": 7450
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.2208358496427536,
+      "learning_rate": 3.7879180861364155e-05,
+      "loss": 0.6158,
+      "step": 7460
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.19732904434204102,
+      "learning_rate": 3.7844976739499454e-05,
+      "loss": 0.5967,
+      "step": 7470
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.2552759349346161,
+      "learning_rate": 3.781073991810229e-05,
+      "loss": 0.6201,
+      "step": 7480
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.22178280353546143,
+      "learning_rate": 3.7776470484329554e-05,
+      "loss": 0.5918,
+      "step": 7490
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.23287595808506012,
+      "learning_rate": 3.7742168525421115e-05,
+      "loss": 0.6121,
+      "step": 7500
+    },
+    {
+      "epoch": 1.81,
+      "eval_loss": 0.6359540820121765,
+      "eval_runtime": 67.1833,
+      "eval_samples_per_second": 29.769,
+      "eval_steps_per_second": 0.938,
+      "step": 7500
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 20690,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 5,
+  "save_steps": 500,
+  "total_flos": 1.621757966538716e+19,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}